Firm network structure and innovation

Firm network structure and innovation

Accepted Manuscript Firm network structure and innovation Tuugi Chuluun, Andrew Prevost, Arun Upadhyay PII: DOI: Reference: S0929-1199(17)30182-7 do...

754KB Sizes 0 Downloads 20 Views

Accepted Manuscript Firm network structure and innovation

Tuugi Chuluun, Andrew Prevost, Arun Upadhyay PII: DOI: Reference:

S0929-1199(17)30182-7 doi: 10.1016/j.jcorpfin.2017.03.009 CORFIN 1174

To appear in:

Journal of Corporate Finance

Received date: Revised date: Accepted date:

9 March 2016 2 February 2017 22 March 2017

Please cite this article as: Tuugi Chuluun, Andrew Prevost, Arun Upadhyay , Firm network structure and innovation. The address for the corresponding author was captured as affiliation for all authors. Please check if appropriate. Corfin(2017), doi: 10.1016/ j.jcorpfin.2017.03.009

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

ACCEPTED MANUSCRIPT

Firm Network Structure and Innovation

IP

T

Tuugi Chuluun* Department of Finance Sellinger School of Business and Management Loyola University Maryland 4501 N. Charles Street, Baltimore, MD 21210 [email protected]

US

CR

Andrew Prevost Grossman School of Business University of Vermont 55 Colchester Avenue, Burlington, VT 05405 [email protected]

M

AN

Arun Upadhyay Department of Finance College of Business Florida International University 11200 S.W. 8th St, RB 247B, Miami, FL 33199 [email protected]

ED

November 2016 Abstract

AC

CE

PT

This study examines how various dimensions of a firm’s network affect innovation and pricing of innovation by market participants. We use board interlocks to build interfirm network and construct different network measures to capture firm centrality in the interfirm network, cohesion within firm network, and diversity, innovativeness, and propinquity of firm network. Our results show that these different characteristics of network connectedness affect firm innovation input and output, particularly for firms in relatively intangible industries. These results are robust to the use of an instrumental variables approach as well as a natural experiment of state-level changes in R&D tax credits. We also find that innovation has a positive (negative) marginal effect on corporate bond yield spreads when firms have lower (higher) connectedness, suggesting that the market perceives innovative activities by more connected firms as less uncertain. Yield spread changes around patent filings further support this finding. Keywords: network; connectedness; innovation; R&D; board interlock; bond yield JEL classification: G30; G31; D80; L14 *Corresponding author. The authors would like to thank the participants of the 4th Annual CIRANO Workshop on Networks in Trade and Finance for comments and suggestions. Tuugi Chuluun acknowledges financial support from summer research grant from Sellinger School of Business and Management of Loyola University Maryland.

ACCEPTED MANUSCRIPT 1. Introduction Innovation has long been considered one of the key determinants of firm value, performance, and survival. Innovation involves the creation of more effective products and processes with the application of new knowledge and information thus making information

T

essential for innovation development (Drucker, 1993; Hall, Jaffe and Trajtenberg, 2005). Social

IP

network theory suggests that a network of connections that a firm maintains can provide

CR

informational advantages and facilitate information diffusion. One such network is the interfirm network created through board interlocks. As firms form and maintain board interlocks, they

US

create a network of direct and indirect ties with each other. The structure of this interfirm

AN

network, in turn, can influence the dynamics of information diffusion among firms and affect various aspects of firm innovation, among others. In this paper, we investigate if characteristics

M

of firm network connectedness impact firm innovation and market participants’ perception of

ED

innovation risk. Specifically, we examine whether the structure of firm network affects firm R&D expenditures and patenting activities and if so, which structural properties enhance firm

PT

innovation. We also examine whether the role of network connectedness varies depending on

CE

firm industry. Finally, we investigate whether network connectedness has a bearing on the pricing of innovation, using the yield spreads of newly-issued corporate debt and secondary-

AC

market corporate bonds.

In the corporate setting, directors serve an important advising function in addition to their monitoring role. Boards advise on many strategic decisions such as the focus and general direction of R&D expenditure, the approach to knowledge and intellectual property management, and the level of investment in innovation. Many practitioner sources emphasize the role of board

1

ACCEPTED MANUSCRIPT in innovation, especially in high-growth or entrepreneurial firms.1 Even in more established firms, boards can help foster innovation. For example, Altria, Ford, Proctor & Gamble, Urban Outfitters, American Express, NorthWestern Corporation among many others have innovation and technology board committees charged with the oversight of, and guidance on, corporate

T

innovation strategy.2 Since knowledge and information are critical inputs to the board’s

IP

innovation advising function, the presence of directors that sit on multiple boards may be

CR

important as they help transmit tacit knowledge and information and expose firms to relevant information.

US

The implications of board connections and interlocks has been the focus of considerable

AN

academic research. One strand of this literature examines this subject from a corporate governance perspective and generally finds that board connectedness reduces the efficacy of

M

corporate governance (Barnea and Guedj, 2009; Liu, 2010; Hwang and Kim, 2012). Another line

ED

of work shows that director networks affect the flow of information and the level of communication between connected firms (Cai and Sevilir, 2012). In line with this latter view,

PT

prior research finds that networks diffuse information and propagate certain corporate practices

CE

such as corporate finance policies (Fracassi, 2015), dividend policy (Bouwman and Xuan, 2010), private equity deal exposure (Stuart and Yim, 2010), and earnings management (Chiu, Teoh, and

AC

Tian, 2013; Bouwman, 2011). Little research, however, has been conducted on how board networks affect innovation. The social network literature suggests that repeated interactions among socially networked individuals lead to elevated levels of mutual trust and trustworthiness (Glaeser et al., 1

Behan, B., “Innovation starts in the boardroom.” Directors and Boards, 39.1, 2014, p. 29; Useem, M., D. Carey and R. Charan, 2014. “How boards can innovate.” Harvard Business Review (Digital Articles), May 21, pp. 2–4. 2 Examples of charters of board innovation committees are available from http://us.pg.com/who-we-are/structuregovernance/corporate-governance/board-committees-charters and http://www.altria.com/About-Altria/Board-ofDirectors-and-Committees/Board-Committees/Innovation/Documents/Innovation-Committee-Charter.pdf.

2

ACCEPTED MANUSCRIPT 2000), which is an important element in information sharing among firms involved in innovations. Helmers, Patnam and Rau (2015) use corporate governance reforms in India as an exogenous shock to examine the effect of board interlocks on patenting and R&D spending among publicly traded companies in India. Dasgupta, Zhang and Zhu (2015) study the effect of

T

prior social connections between managers or board members and supplier and customer firms

IP

on innovation of upstream firms. Dass, Kini, Nanda, Onal and Wang (2013) also focus on

CR

directors from related industries and examine their impact on firm performance. Alternatively, Kang et al. (2014) use CEO-director social connections as a measure of board friendliness to

US

investigate how ‘friendly’ boards affect firm innovation. Oh and Barker (2015) find that when

AN

CEOs serve as independent board members on other firms, they imitate the R&D intensity of firms they are interlocked with in their own firm’s R&D decisions. However, these prior studies

M

do not look at the various dimensions of network connectedness, which goes beyond

ED

documenting the existence of interlocks and describes a firm’s position within the broader interfirm network and the quality and structure of its immediate network.

PT

Further, previous work has not explored if network connectedness affects the pricing of

CE

innovation risk by bond market participants. This is particularly interesting since the extant literature has found mixed results on the costs and benefits of innovation and hence how it is

AC

priced. On one hand, a variety of work (e.g. Lev and Sougiannis, 1996; Eberhart, Maxwell and Siddique, 2004) finds a positive relation between R&D expenditure and stock returns. However, additional work (e.g. Kothari, Laguerre and Leone, 2002; Shi, 2003) shows that R&D expenditure increases cash flow variability suggesting that the benefit of R&D outlays may be offset by the risk. Several recent studies examine how the relation between equity market valuation and innovation may be conditional on factors such as managerial ability (Chen et al.,

3

ACCEPTED MANUSCRIPT 2015), size (Ciftci and Cready, 2011), and corporate governance (Chan et al., 2015) while others point out misvaluation of innovation altogether (Cohen, Diether, and Malloy, 2013; Duqi, Jaafar and Torluccio, 2015). Our use of corporate bond yield spreads allows us to examine if the perceived impact of innovation activity on cash flow risk decreases for firms that are more

T

connected.

IP

To address these questions, we construct a set of network measures that capture a firm’s

CR

centrality in interfirm network (degree, eigenvector, and betweenness), cohesion and diversity within individual firm networks (density, network non-redundancy, and industry diversity), and

propinquity,

and innovative geographic

AN

geographic propinquity, innovative industry

US

innovativeness and propinquity of firm networks (network innovativeness, industry propinquity,

propinquity). Using these network measures, we assess if firm network characteristics impact

M

innovation input and output and we find that different dimensions of firm networks – centrality,

ED

diversity, innovativeness, and propinquity – affect innovation activities of firms while density does not. Moreover, our findings are strongest for firms in less tangible industries and robust to

PT

the use of an instrumental variables approach as well as a natural experiment using state-level

CE

changes in R&D tax credits. We also find that network connectedness affects the pricing of innovation by bond market participants: innovation has a marginal positive (negative) effect on

AC

yield spread when firms have lower (higher) connectedness, suggesting that the market perceives innovative activities by more connected firms as less uncertain. Additional tests based on yield spread changes around patent filing dates provide further evidence of the role of connectedness on the link between innovation and bond pricing. Our work contributes to the board of director literature by showing how boards facilitate information diffusion and, furthermore, contributes to our understanding of how innovation

4

ACCEPTED MANUSCRIPT occurs in firms and how it is priced by the market participants. By documenting how different aspects of firm network affect innovation, our paper also contributes to the growing literature on the consequences of network connectedness on financial decision-making. A well-developed line of research demonstrates that networks and their characteristics affect economic outcomes in

T

various settings (e.g., Kali and Reyes, 2010). In financial research, interest in social networks has

IP

only recently emerged. For instance, Cohen, Frazzini and Malloy (2008) focus on connections

CR

between fund managers and corporate board members via shared education networks. Hochberg, Ljungqvist and Lu (2007, 2010) examine networks in venture capital industry, and other studies

US

focus on the impact of informal networks on borrower terms (Garmaise and Moskowitz, 2003),

AN

mutual fund portfolio decisions (Fu and Gupta-Mukherjee, 2014), stock market participation (Hong, Kubik and Stein, 2004), and portfolio choice (Massa and Simonov, 2012).

M

The remainder of the paper is organized as follows. Section 2 discusses the relevant

ED

literature and describes our hypotheses. Section 3 discusses the various network measures in further detail. After describing the data and the descriptive statistics in Section 4, we

PT

present the results on the relationship between firm networks and innovation input and output in

CE

Section 5. Section 6 presents the results on how firm network connectedness affects pricing of

AC

innovation. Section 7 concludes.

2. Relevant Literature and Hypotheses Network theory states that a firm’s network position determines, in part, the constraints and opportunities that will be encountered by the firm which can affect information gathering, strategic choices, corporate risk-taking, and utilization of scarce corporate resources; therefore, network connectedness is important for predicting firm outcomes and performance. Firm

5

ACCEPTED MANUSCRIPT network itself can vary along different dimensions. These different network characteristics should have implications for the extent of trust, sharing, and cooperation that exist among connected firms which in turn dictate the volume, diversity, and richness of information that travel through network. Hence, firms with certain network capabilities have greater access to

T

valuable information that is relevant for firm innovation. To the extent that connectedness

IP

explains innovation, firm connectedness may also influence the value placed on innovation by

this latter research question is particularly interesting.

CR

market participants. In light of previous mixed findings on the costs and returns to innovation,

US

Networks in general (Garmaise and Moskowitz, 2003) and of directors specifically

AN

(Khwaja and Mian, 2005) impact the availability of credit which is vital for firms engaged in innovative activities. Director networks could also impact the incentive system for top managers.

M

Firms need to have an appropriate level of financial incentives to encourage investments in long-

ED

gestational risky R&D intensive projects that risk-averse managers might not be willing to undertake. Directors coming from successful R&D intensive firms could play an important role

PT

in designing suitable contracts that encourage top managers to invest in innovative projects. Such

CE

directors could also create an environment of risk-tolerance at the top that does not penalize failures related to innovation attempts and help other board members assess the riskiness and

innovation.

AC

overall value of innovative projects. Overall, boards can set the tone and direction at the top for

In this paper, we focus on board-level network centrality, density, diversity, innovativeness, and propinquity. Both the quantity and quality of network ties may matter for information diffusion and thus we expect some or all of the above network aspects to have a potentially significant effect on the intensity of firm innovation. Among these various network

6

ACCEPTED MANUSCRIPT characteristics, perhaps the most commonly studied is centrality which describes the location of a firm in the global network using measures such as degree, eigenvector and betweenness. All else being equal, firms with large networks receive a greater volume of information. Occupying a bridging position and maintaining ties to well-connected partners would also imply a greater

T

access to a wide range of information that can be valuable for firm innovation policies.

IP

In addition to network centrality, cohesive networks of directors characterized by

CR

extensive interconnections may be associated with higher transmission capacity enabling large amounts of information and knowledge to rapidly diffuse and integrate. Dense networks may

US

also make firms more willing to cooperate and share relevant information as such networks give

AN

rise to trust, reciprocity, and shared identity (Coleman, 1988; Granovetter, 1992) particularly for innovation that requires generation of tacit information. Prior organizational research has shown

M

that the extent to which network cohesion affects information diffusion depends on the degree of

ED

tacitness of the information involved. Hansen (1999), for instance, shows that the transfer of tacit complex knowledge relevant to new product development is easier between frequent and close

PT

connections because the motivation to assist a contact is greater than in infrequent and distant

CE

connections. Such cohesive and close relationships can be captured by measures like density. However, a highly cohesive network of directors may also mean that much of the shared

AC

information and knowledge is homogenous and redundant originating from the same sources. Sociologists point out that cohesion can also mean adhering to established norms and standards, which can stifle innovation and creativity. Another aspect of network connectedness that relates to the information diffusion process is diversity. A well-established body of management research and a newly emerging literature in finance examine the impact of diversity on organizational performance with mixed findings on

7

ACCEPTED MANUSCRIPT the nature of the relationship. On one hand, having a diverse network can make firms more innovative because firms with heterogeneous networks may have better access to diverse information pools. For example, in the context of CEO connectedness, Fang, Francis and Hasan (2012) find that heterogeneity in the CEO’s social network is associated with higher market-to-

T

book ratio for firms, suggesting that diversity is considered a component of CEO social capital.

IP

On the other hand, relations based on similarity (or homogeneity) may be stronger due to easier

CR

coordination and communication and therefore affect the quality of information and the amount of cooperative effort shared between firms. Homophily refers to the principle that contact

US

between similar people occurs at a higher rate than among dissimilar people; patterns of

AN

homophily are consistent across different types of relationships such as marriage, friendship, and acquaintance (McPherson et al., 2001).

M

Recent research shows that relationships based on cohesion and diversity may not be

ED

opposites as previously assumed. For instance, Anderson et al. (2011) find that greater heterogeneity does not always improve board efficacy, and whether investors place valuation

PT

premium or discount on board heterogeneity depends on the complexity of the firm.

CE

Furthermore, Reagans and Zuckerman (2001) find that both network diversity and density enhance R&D team productivity, while Phelps (2010) shows that density and diversity can

AC

coexist and their benefits on firm innovation are complementary. Diversity that exists within individual firm networks can be measured by the extent of non-redundant ties (network nonredundancy) or the extent of industry concentration in firm network (industry diversity). Whether a firm is interlocked with innovative firms (network innovativeness) may be relevant for innovation outcomes since several recent papers have shown that firms learn from the experience of their partners and certain corporate practices spread through networks as

8

ACCEPTED MANUSCRIPT evidenced by similarity across interlocked firms. Beckman and Haunschild (2002) show that firms learn from the acquisition experiences of their network partners, and Fracassi (2015) finds that more social connections two firms have, more similar they are in terms of their corporate finance policies. Davis and Greve (1997) document that the practice of poison pill spread

T

through board-to-board diffusion process, and similarly, Bizjak, Lemmon and Whitby (2009)

IP

find that board interlocks helped spread option backdating. Stuart and Yim (2010) also show that

CR

firms with directors, who have experienced a private equity deal at another company, are more likely to become targets of private equity. Faleye et al. (2014) show that firms with CEOs that

US

maintain ties to individuals affiliated with innovative firms have higher R&D investment.

AN

Propinquity based on geography and industry may shed further light on how information relevant to innovation travels between firms. Information is more likely to be shared among

M

firms that are geographically close, and prior research has indeed documented the role of local

ED

networks in information exchange (e.g., Teo, 2009; Butler, 2008). We also surmise that information is more likely to be transferred among those who are in the same industry and that

PT

industry-specific information may also be more relevant for innovation. Hence, we utilize the

CE

measures geographic propinquity and industry propinquity to examine these hypotheses and further interact these measures with network innovativeness.

AC

We hypothesize that the impact of network connectedness on firm innovation is greater for certain types of firms. For example, the role of networks in innovation development may be greater for firms in more intangible industries. We also examine if connectivity influences the perceived value of innovation as reflected by the firm’s debt securities. Prior work (e.g. Lev and Sougiannis, 1996; Eberhart, Maxwell and Siddique, 2004) finds a positive relation between R&D expenditure and stock returns that may vary according to such factors as managerial ability

9

ACCEPTED MANUSCRIPT (Chen et al., 2015), firm size (Ciftci and Cready, 2011), and corporate governance (Chan et al., 2015) while further work demonstrates how innovation may be misvalued by the equity market (Cohen, Diether and Malloy, 2013; Duqi, Jaafar and Torluccio, 2015). Additional work (e.g. Kothari, Laguerre, and Leone, 2002; Shi, 2003) shows that R&D expenditure increases cash flow

T

variability suggesting that the benefit of R&D outlays may be offset by the risk. In particular, Shi

IP

(2003) uses the pricing of debt securities to show that R&D expenditure has a negative impact on

CR

the cost of debt as measured by bond ratings and risk premiums of at-issue bonds. However, Shi (2003) does not distinguish whether R&D expenditure increases firm value or whether the

US

increase in equity value documented in prior work results from a wealth transfer from

AN

bondholder claims. Eberhart, Maxwell and Siddique (2004) show that Shi’s (2003) results are sensitive to the measure of R&D expenditure and demonstrate that R&D has a positive net effect

M

on bondholder wealth. The effect is stronger for firms with greater default risk (i.e., firms with

ED

more to gain from R&D increases) and for firms where bondholders have greater covenant protection. Using patent portfolios, Hsu et al. (2015) find that firms' default probabilities, bond

PT

issuance premiums, and realized excess returns are negatively related to the quantity and quality

CE

of firm patent portfolios. In a similar vein, Francis et al. (2012) show that borrowers with higher innovation capability enjoy lower bank-loan spreads and better nonprice-related loan terms.

AC

We extend this line of work by examining how innovativeness, as measured by innovation input and output, impacts the cost of debt capital. In contrast to prior work, we surmise that the effect is conditional on the extent of network connectedness. Based on the premise that the information diffusion process is linked to the extent of the firm’s network, we surmise that the degree of connectedness may impact the relation between innovation and credit risk. Market participants may perceive networks to be an integral part of innovation and view

10

ACCEPTED MANUSCRIPT R&D expenditure and other forms of innovative activity by firms that are relatively less networked, or with lower network capabilities, as risk-increasing. Thus, innovation may have a marginal positive effect on yield spread when firms are less connected, suggesting that the market associates the impact of innovation as more uncertain when firms have less access to

T

information within the network.

IP

3. Network Measures

CR

We employ a set of network measures that describe the composition of a firm’s network and the firm's position within the broader interfirm network. Specifically, these network

US

measures capture a firm’s centrality in the interfirm network (degree, eigenvector, and

AN

betweenness), cohesion and diversity within the immediate network surrounding a firm (density, network non-redundancy, and industry diversity), and innovativeness and propinquity of firms

M

connected to a focal firm (network innovativeness, industry propinquity, geographic propinquity,

ED

innovative industry propinquity and innovative geographic propinquity). These network measures are computed using undirected binary data over four-year periods, and their

CE

3.1. Centrality measures

PT

descriptions are provided in the Appendix.

Perhaps the most commonly-used network measures gauge the centrality of a firm’s

AC

position within the global network of firms. There are multiple centrality measures that approach centrality from different perspectives. Following prior research, we focus on three normalized measures of network centrality. The simplest measure of centrality is degree, which is the number of relationships a firm has. The higher the number of relationships a firm has, the more access to information it has. We use a normalized degree measure, which refers to the percentage of all other firms a specific firm is connected to during a period.

11

ACCEPTED MANUSCRIPT Our second centrality measure is eigenvector (Bonacich, 1972), and it is similar to an iterated degree measure. Essentially, eigenvector differs from degree in that not only how many relationships a firm has, but also to whom it is connected matters. In other words, the centrality of each firm is determined by the centralities of the firms it is connected to. We normalize this

T

measure by dividing it by the maximum possible eigenvector centrality in the network. Firms

IP

with higher eigenvector centrality are connected to others that are themselves well connected and

CR

become closer to all others in the interfirm network thus occupying more central positions at the center of the information hub.

US

Centrality can also be captured by the extent of bridging positions a firm occupies which

AN

enables a firm to serve as an intermediary, gatekeeper, or broker of valuable resources such as information. We use a betweenness measure proposed by Freeman (1979) that captures how

M

often a firm happens to be positioned on informational linkages between pairs of other firms. A

ED

firm is between two firms if it lies on the shortest possible path (i.e. geodesic distance) between

normalized betweenness.

CE

3.2. Cohesion measure

PT

them. We divide betweenness by the maximum possible betweenness in the network to obtain

Density describes the level of cohesiveness or interconnectedness that exists within an

AC

individual firm’s network. Specifically, density refers to the number of connections that exist among all partners of a firm divided by the maximum number of all connections that can potentially exist among these partners. A densely connected network means that firms that are interlocked to a specific focal firm maintain a large number of ties to each another indicating more trust or coordination which can aid information diffusion. 3.3. Diversity measures

12

ACCEPTED MANUSCRIPT Heterogeneity that exists within individual firm networks may indicate the types of information channels that are present; as a result, heterogeneity can affect innovation either positively or negatively. More heterogeneous networks represent more diverse information, but connections based on similarity may instead be more reliable and informative. Firm network

T

heterogeneity is measured by the extent of non-redundant ties and the extent of industry

IP

concentration in a firm’s network here. The first diversity measure we examine is network non-

CR

redundancy which is based on the idea of structural holes developed by Burt (1992). A structural hole is a gap between two parties with complementary resources or information, and the premise

US

is that unconnected non-redundant ties are more likely to be diverse and offer different points of

AN

view. Network non-redundancy measures the proportion of a firm’s relationships to its partners that are considered non-redundant and referred to as effective size in the social network

M

literature. Specifically, it computes the number of partners minus the average degree of partners

ED

within the firm network, not counting ties to firm itself. If a firm is connected with others, who in turn are connected with the same firms, network non-redundancy measure will be low. Second,

PT

industry diversity measures the industry concentration in a firm’s network using the Herfindahl

CE

index. For each firm, we compute the industry Herfindahl index based on the percentages of all interlocked firms in different Fama-French 30 industries. We use the inverse of the Herfindahl

AC

index so that higher values are associated with greater diversity. 3.4. Innovativeness measure Since extant research documents that learning and information transmission occur in networks (e.g., Beckman and Haunschild, 2002; Fracassi, 2015; Davis and Greve, 1997), having ties to other innovative firms may affect a firm’s innovation policies. Hence, for each firm, we

13

ACCEPTED MANUSCRIPT compute the number of all its interlocked firms that have positive R&D expenditure. We call this measure network innovativeness. 3.5. Propinquity measures Economic geography has long emphasized the importance of co-location and its role in

T

information and resource sharing (Sorenson and Stuart, 2001; Jaffe, Trajtenberg, and Henderson,

IP

1993). Prior finance research also documents the role of local networks in information exchange

CR

(Teo, 2009; Butler, 2008). Therefore, to shed further light on how information relevant to innovation travels between firms, for each firm, we compute the number of all its connected

US

firms that are located in the same metropolitan statistical area (MSA). We call this measure

AN

geographic propinquity. Information is also more likely to transfer among firms in the same industry. This is consistent with the principle of network homophily, which states that

M

information is likely to be shared among those who are similar: Birds of feather flock together

ED

(McPherson et al., 2001). In addition, industry-specific information may indeed be more relevant for innovation. Thus, for each firm, we compute the number of all its connected firms that are in

PT

the same Fama-French 30 industry as the focal firm. We call this measure industry propinquity.3

CE

3.6. Innovative propinquity measures We refine the propinquity measures by interacting them with innovativeness: We capture

AC

the number of all firms that a firm is connected to that are either in the same geographical location or industry as the firm and also innovative (i.e., have positive R&D expenditure). We call these measures innovative industry propinquity and innovative geographic propinquity measures. 3

We also consider indirect connections and compute the number of directly and indirectly connected firms that are either located in the same MSA or in the same industry. Indirect connections are those firms that a firm is connected to through another firm. In other words, we are measuring the number of firms that are in the same geographical location or industry within two degrees of separation. We find qualitatively similar results with these measures that utilize two degrees of separation.

14

ACCEPTED MANUSCRIPT

4. Data We obtain information on the directors of 3,838 unique firms in the S&P 1500 Index from 1996 to 2013 from the RiskMetrics database. To examine how a firm’s position in its

T

interfirm network affects innovation, we map the entire network of these firms based on the

IP

interlocks between their boards. During the sample period, over 10,000 individuals served on the

CR

boards of the sample firms each year, and two firms are considered to have a board interlock whenever a director sits on the boards of both firms. We document the connections among all of

US

the firms over rolling four-year periods and find that all the firms are connected to at least one

AN

other firm during any four-year period. Using these board interlocks among firms, we compute network measures for each firm for each period. Some of the network measures are generated by

M

employing UCINET 6 software (Borgatti, Everett, and Freeman, 2002). For some network

ED

measures, we identify the MSA of a firm’s location using the firm headquarter location. The U.S. Census Bureau lists all the MSAs in the U.S. and their associated zip codes.

PT

We match these network measures from rolling four-year periods to the financial data of

CE

the following year. We obtain various accounting data from the Compustat database for all the firms included in the S&P1500 universe. In line with previous research, we exclude firms in the

AC

financial services and utilities industries. Our control variables include firm size (logged sales), market-book ratio (ratio of market value of assets to book value assets), sales growth, return on assets (ROA), book leverage, cash holdings (scaled by total assets), the unionization rate (described below), capital expenditure (scaled by total assets), and industry competition as measured by the Herfindahl-Hirschman index. These control variables are winsorized at 1 and 99 percentiles to mitigate the effect of outliers and are further described in the Appendix.

15

ACCEPTED MANUSCRIPT Prior research demonstrates that union strength affects risky decisions, including innovation. For example, Chen, Kacperczyk and Ortiz-Molina (2011a) show that strong organized labor curbs managerial incentives to take excessive risk, including risky investment choices. More specifically, Betts, Odgers and Wilson (2001) demonstrate that industry

T

unionization rates and R&D expenditure are negatively correlated. We control for union strength

IP

with the percentage of unionized workers covered by bargaining agreements using industry-level

CR

unionization rates from the Union Membership and Coverage Database (www.unionstats.com) maintained by Barry Hirsch and David Macpherson. Unionization rates are reported at the

US

Census Industry Classification (CIC) level, which we merge to firms in our sample by converting

AN

to the corresponding 4-digit SIC code. A variety of studies (e.g. Klasa, Maxwell and OrtizMolina, 2009; Chen, Kacperczyk and Ortiz-Molina (2011a, 2011b); Chen, Chen and Liao, 2011)

M

provide evidence that industry-level unionization rates effectively proxy for firm-level rates.

ED

With respect to innovation input, our primary variable of interest is R&D intensity, measured as the ratio of R&D expenditure to total assets; following the conventional approach,

PT

we set R&D to zero when it is missing. To measure firm innovation output, we use NBER patent

CE

data that was originally constructed by Hall, Jaffe and Trajtenberg (2001).4 The first measure of innovation output is the number of patents a firm applied for in a given year. Next, we consider

AC

the type of patents. In the innovation literature, patents can be categorized into exploitative vs. exploratory innovation (e.g., Gilsing et al., 2008). A patent is considered exploitative if it is in an area that an organization innovated before and, hence, deepens an organization’s existing knowledge base. Exploratory innovation, on the other hand, broadens an organization’s existing knowledge base and involves a patent in an area that is novel to the organization. It is possible 4

The data is obtained from https://sites.google.com/site/patentdataproject/Home. The patent data is only available until 2006. Thus, our analysis on innovation output is for a shorter time period than that on innovation input which utilizes accounting statement data.

16

ACCEPTED MANUSCRIPT that network plays an even greater role in bringing novel information that results in exploratory innovation rather than exploitative innovation that derives from existing knowledge base. Hence, for each firm, we make a list of all the technology classes in which the firm had filed patents in the previous five years and then compare whether a patent applied in a given year is in a new

T

technology class compared to the previous years to determine whether a patent is exploratory or

IP

not. Then we count the total number of exploratory patents a firm applied for in a specific year.

CR

We also measure patent diversity using the inverse of the Herfindahl Index based on patents a firm filed in a given year in different technology classes. To further quantify the impact and

US

significance of patents, we utilize the total and average number of subsequent citations a firm’s

AN

patents receive. One issue with using citations data is the truncation bias due to the finite length of the sample period. To address this issue, we use the weighting index developed by Hall, Jaffe

M

and Trajtenberg (2005).

ED

We explore if network connectedness influences the risk of innovation perceived by market participants by using the pricing of at-issue corporate bonds obtained from the SDC

PT

Platinum database. We exclude convertibles, bonds for which there is no conventional yield to

CE

maturity (e.g. floating and step-up coupon bonds), and bonds with synthetic features and exotic structures. We obtain at-issue yield to maturity, security type, embedded options, coupon rate

AC

and proceeds amount from SDC. In subsequent analyses, we examine changes in yield spread following patent filing dates obtained from the NBER patent database. We obtain secondary market transactions using corporate bond price data obtained from the TRACE and Mergent FISD databases. We provide further details about these sources below. Table 1 Panel A presents summary statistics of our network measures. Mean degree centrality over all periods is 0.31 percent; eigenvector is 2.12; and betweenness is 0.024. The

17

ACCEPTED MANUSCRIPT mean density of firm networks is 15.33 percent. As for the diversity measures, the mean network non-redundancy is 3.69 and industry diversity measure is 4.39, which is the inverse of Herfindahl index based on the percentages of a firm’s partners in different industries. On average, a given firm is connected to 3.36 other innovative firms with positive R&D spending.

T

As for similarity in location (industry), on average, a firm is directly connected to 1.14 (0.82)

IP

other firms located in the same MSA (industry). On average, a firm is also connected to 0.52

CR

(0.55) other firms with positive R&D spending that are located in the same MSA (industry). Panel B presents the descriptive statistics of the innovation measures and remaining firm-

US

level control variables. R&D intensity has a mean of 0.036 with many firms having no R&D

AN

expenditures. The average number of patents is 16.67. The average number of exploratory patents is 0.71, and the average patent diversity measure is 4.29. On average, patents receive

M

130.88 citations. Firms have mean sales of 5.6 billion dollars. Market-book ratio (defined as the

ED

ratio of market value of assets to book value assets) gauges expected growth and has a mean of 1.92. Sales growth has a mean of 0.096. Book leverage and ROA have a mean of 0.22 and 0.03,

PT

respectively, and the average for cash holdings is 0.11. The unionization rate, on average, is 9.27

CE

percent. Average industry HHI is 0.282. In additional robustness tests, we control for CEO network size, analyst estimates, and institutional ownership. Descriptive statistics for these

AC

variables are presented in Table 1 Panel B.

5. Network Connectedness and Innovation Input and Output 5.1. Single equation least squares estimates We examine if firm network characteristics affect innovation input and output, particularly for firms in industries that rely more on intangible assets. Table 2 presents our results

18

ACCEPTED MANUSCRIPT of firm network structure and R&D intensity. We estimate a regression on R&D intensity using firm network variables along with standard control variables. Since the three centrality measures are highly correlated, we find that the first principal component obtained from principal component analysis explains most of the variation in the three measures. Thus, centrality is a

T

composite measure of three network centrality measures that capture network size (i.e., degree),

IP

closeness (i.e., eigenvector), and extent of bridging intermediary positions (i.e., betweenness)

CR

based on their first principal component. In a similar manner, we report one measure from each category of network variables. For example, since the results of the two diversity measures,

US

network non-redundancy and industry diversity, are qualitatively similar we report results using

AN

only industry diversity. We also report results using only geographic propinquity and innovative geographic propinquity. All the regressions include year and (Fama-French 30) industry fixed

M

effects and the standard errors are clustered at the firm level.

ED

In Table 2 Panel A, the estimated coefficients on all the network variables except for network density are significant in Models (1)-(6). As the coefficient on the centrality measure in

PT

Model (1) shows, the more centrally located firm i is in the interfirm network, the higher the

CE

R&D intensity: this is consistent with the premise that greater network centrality is more conducive for information diffusion. Firms that are connected to other firms in more diverse

AC

industries also tend to spend more on R&D as Model (3) shows. As evidenced by the coefficient estimate of the innovativeness measure in Model (4), firms that maintain ties to innovative firms also spend more on R&D. This is consistent with the previously documented transmission capacity of networks. Certain corporate practices and policies are known to propagate via firm networks (Fracassi, 2015; Bouwman and Xuan, 2010; Bouwman, 2011). The positive and significant coefficient on geographic propinquity in Model (5) shows that firms that maintain a

19

ACCEPTED MANUSCRIPT higher number ties to other firms located in the same geographical area also have higher R&D intensity.5 This suggests that physical proximity promotes information sharing that leads to more R&D activities. Ties to other co-located firms that also happen to be innovative are even more important as evidenced by the significant coefficient in Model (6) that is larger than the

T

coefficient in Model (5). The level of network density, however, is not significant as shown in

IP

Model (2). This suggests that when it comes to innovation input, how cohesive or tight knit a

CR

firm’s network is does not matter. The signs and significance of the remaining control variables are in line with previous findings. Consistent with the results documented by Hirschleifer, Low

US

and Teoh (2012), we find that higher R&D expenditures are associated with smaller firms. R&D

AN

spending is also correlated with higher growth, as evidenced by the positive coefficient estimates for market-book ratio and prior sales growth. Also consistent with Hirschleifer et al. (2012),

M

R&D spending is associated with poorer operating performance, lower book leverage, and higher

ED

cash holdings. Consistent with past findings that demonstrate a negative relation between union presence and R&D expenditure, unionization rate is negative and significant at the 1 percent

PT

level. Firms with higher levels of capital expenditure invest less in R&D, and the negative and

CE

significant coefficient on industry HHI suggests that firms in more competitive industries tend to invest more in R&D.

AC

Next, we re-estimate these regressions in subsamples of firms in tangible and intangible industries as we believe the role of networks in innovation development may be greater for firms in more intangible industries. We define tangibility as the ratio of property, plants, and equipment to lagged total assets and compute industry average tangibility. We further divide industries into those above and below the median industry tangibility. The results in Table 2B are

5

The results remain qualitatively the same when we use propinquity measures based on both direct and indirect connections (i.e., two degrees of separation).

20

ACCEPTED MANUSCRIPT consistent with this premise: The estimated coefficients for firms in intangible industries are typically significant and larger than those obtained for the firms in the tangible industries subsample. This demonstrates that network has a greater impact on innovation input for firms in intangible industries.

T

These results demonstrate that network centrality, diversity, innovativeness, and

IP

propinquity matter for R&D intensity, and the impact of network on R&D intensity depends on

CR

firm characteristics. Our findings are also economically significant. For example, one standard deviation change in the composite measure of centrality (innovativeness) is associated with

US

0.0044 (0.010) increase in R&D intensity in the full sample, where the median R&D intensity is

AN

zero and mean is 0.036.

We now turn to whether network connectedness affects the level and type of innovation

M

output using patent data. The results in Table 3 Panel A show that all network characteristics,

ED

except for density, have statistically significant coefficients. Therefore, higher centrality is associated with higher number of patents which is consistent with more information diffusion

PT

being conducive for innovation. In addition, firms that maintain ties to other firms that are

CE

innovative, geographically close, or in diverse industries file more patents. This is again consistent with the previously documented transmission capacity of networks.

AC

In Panel 3B, we focus on the extent of exploratory patents, which specifically capture innovation in new fields for the filing firm. The results are similar to those obtained from Panel 3A in that all the network variables, except for density, are significant and positive. In Panel 3C, we examine the diversity of patent classes and find that firms that maintain diverse or innovative networks, in turn, innovate in a broad range of technology classes. This suggests that firms use their networks to generate information, and those firms with diverse and innovative networks

21

ACCEPTED MANUSCRIPT receive diverse information leading to the creation of diverse innovation. When we re-estimate these patent regressions in subsamples of firms in tangible and intangible industries in Panels 4D1-4D3, distinct differences emerge in the results across the subsamples. In general, the network variables are significant and have larger estimated coefficients for firms in intangible

T

industries compared to those for firms in tangible industries. Thus, network connectedness and

IP

its information capacity matter more for patents filed by firms in relatively intangible industries.

CR

Next, we examine innovation output in terms of patent impact using number of citations. As the results in Table 4 Panel A show, all the network characteristics, except for density,

US

significantly explain the number of citations that firms’ patents receive. Hence, firms with more

AN

central, diverse, innovative, and geographically proximate networks file more impactful patents that subsequently receive more citations. We also compute the average number of citations by

M

dividing the number of citations by the number of patents and repeat our analysis in Table 4

ED

Panel B, which produces qualitatively similar results. We also test whether these results vary across industry subsamples and find similar results to those in Table 3 subsamples. For the sake

PT

of brevity, these results are not reported in the paper but available from the authors.

CE

To add further insight on the mechanism of networks, in an unreported regression we estimate the likelihood of a firm filing an exploratory patent in a specific technology class given

AC

the number of patents already filed in that class by its connected firms. For each firm, we count the number of patents its connected firms filed in each technology class during the previous five years and estimate a logit model using this measure and additional control variables. Consistent with the view that board connections facilitate information transmission, the (untabulated) results show that a higher number of patents filed in a specific patent class by interlocked firms increases the likelihood of a firm filing an exploratory patent in that patent class; firms are more

22

ACCEPTED MANUSCRIPT likely to innovate in an area, where their interlocked firms have already engaged in innovation activity. Overall, we see that network centrality, diversity, innovativeness, and propinquity affect innovation output as measured by number and type of patents and citations, and the effect is

T

pronounced for intangible asset-intensive firms. Network density, on the other hand, proves to be

IP

largely insignificant in the context of innovation input and output. This shows how the different

CR

dimensions of network connectedness can have different information roles. As for economic significance, for example, a one standard deviation change in the composite measure of

US

centrality is associated with 0.11 increase in logged number of patents in the full sample.

AN

Similarly, one standard deviation changes in network diversity and innovativeness are associated with 0.141 and 0.159 increases in logged number of patents, respectively.

M

5.2. Tests of robustness

ED

We examine the robustness of the results discussed in the previous section first by controlling for additional explanatory variables. The first variable is the extent of the CEO’s

PT

network. Several prior studies have shown how CEO network affects executive compensation

CE

(Engelberg et al., 2013), mergers and acquisitions (El-Khatib et al., 2015; Chikh and Filbien, 2011), corporate fraud (Khanna et al., 2015), country financial development (Ferris et al., 2016),

AC

and firm value (Fang et al., 2012). We view board and CEO networks as complementary information channels, not as substitutes. Hence, we test whether board networks explain innovation independent of the role of CEO network. We use a CEO network measure from the Boardex database that measures the number of individuals a CEO overlapped with through employment, education, and other activities. This variable, therefore, measures the number of external connections that CEO builds over time. It is, however, available only for a subset of the

23

ACCEPTED MANUSCRIPT firms included in our sample. We also control for institutional ownership and analyst coverage and re-estimate the regressions from Tables 2-4 with these three additional variables. We find that the extent of institutional ownership and analyst coverage are largely not significant, but the CEO network measure is significant and positive for R&D intensity. For patent and citation

T

measures, CEO network is occasionally negative, but largely insignificant. In these untabulated

IP

regressions, analyst coverage (institutional ownership) is occasionally positive (negative) and

CR

largely insignificant. However, the results of the board network variables remain significant and qualitatively similar, which shows that board and CEO networks represent two different channels

US

of information. For brevity, we do not report these results as they are based on a much smaller

AN

sample and do not change our main results.6

While the previous section establishes correlation between board network and innovation

M

output, there is also a possibility that potential endogeneity between connectedness and the

ED

dependent variables may be biasing the results. To ensure that our results are robust to the possible simultaneous determination of connectedness and innovation, we re-examine our

PT

primary results in Table 2 using a two-stage least squares (2SLS) approach in Table 5, specifying

CE

the network measure in each respective column as endogenous. Following the approach of Chuluun, Prevost and Puthenpurackal (2014), who identify effective instruments for board

AC

centrality based on distance from other firms and location, the first instrument is the logged number of other public firms that are situated in clustered networks of cities (i.e., megaregions). We define the megaregion of each firm’s headquarters using six U.S. megalopolis regions (Arizona Sun Corridor, California, Cascadia, Great Lakes, Northeast megalopolis, and Piedmont Atlantic Megaregion) and obtain the number of publicly listed companies headquartered in each region using the LexisNexis database. The second instrument is the logged average distance 6

These results are available upon request from the authors.

24

ACCEPTED MANUSCRIPT between the firm’s headquarters and other firms within its Census region, and is constructed using location information from the U.S. Census Gazetteer files and the states belonging to each of the nine regions defined by the U.S. Census.7 In Panel 5A, we report the first-stage coefficient estimates for the two instruments (for

T

brevity we do not tabulate the estimates for the additional second-stage regressors). Logged

IP

public companies is positively related to the network measures as expected and is statistically

CR

significant in five of the six models. Logged distance is negatively, albeit insignificantly, related to connectedness for most of the models. In Panel 5B, we present the second stage estimates. The

US

instrumental variables for the network measures remain positive and statistically significant

AN

except for density in Model (2). The Kleibergen-Paap LM Statistic tests the null hypothesis that the model is underidentified: except for Model (2), the test statistics are significant at the 1

M

percent levels indicating that the models are not underidentified. In the last row, the Hansen J

ED

statistic tests the null hypothesis that the overidentification restrictions are valid. For Models (1)(6), the J statistic p-values are above 0.80, indicating that the null cannot be rejected and that the

PT

instruments are appropriate and well identified in each model.

CE

To further establish that our primary results are free from endogeneity biases, we identify a natural experiment using state-level increases in tax incentives (credits) associated with

AC

corporate R&D expenditure. The question of whether state R&D credits affect private R&D spending is explored by Wilson (2009), who provides state-level R&D tax credit rate changes implemented over the 1992-2006 period. In our context, we use increases in state R&D credits as a natural experiment to test if innovation spending motivated by higher tax credits varies according to the level of firm connectedness. To the extent that network centrality facilitates

7

See Chuluun et al. (2014) for further details about the construction of these two measures.

25

ACCEPTED MANUSCRIPT innovation, we expect to find that the innovation response to higher R&D tax credits should be increasing in the extent of ties to other firms. Using data from Daniel Wilson’s website, there are 14 states that increased R&D tax credits during 2000-2006.8 Table 6 Panel A lists these states and the corresponding percentage

T

increases in the tax credit. Using the firms in our sample that were incorporated in these states as

IP

of the year of the tax credit increase as the treated sample, we identify one matched control firm

CR

from a non-tax credit change state based on industry (Fama-French 30) and size (closest sales within plus / minus 30 percent of the sample firm’s sales) for each sample firm. Similar to

US

Goldman and Peress (2016), we exclude firms incorporated in Delaware from the pool of

AN

potential control firms because many firms locate in Delaware for reasons unrelated to their operations. Using each treated firm and its matched firm counterpart, we construct a three-year

M

panel consisting of one year prior to one year after the credit increase year and create a binary

ED

variable (tax credit increase) equal to one for firms incorporated in states with tax credit increases and zero for matched firms in states with no increase. To test the effect of tax credit

PT

increases on innovation spending, we create a dummy variable equal to one (zero) for the year

CE

following implementation (post) and estimate the following regression model: R & D Intensity  α0  α1Tax credit increase  α2 Post  α3Tax credit increase  Post  Year fixed effects  e j,t

(1)

AC

The coefficient estimate α3 measures the net difference in R&D intensity for firms in states with credit increase in the year following the increase relative to matched firms. In addition, we include year fixed effects in Equation (2). The results of the differences-in differences estimation are presented in Table 6 Panel B. As indicated in Model (1), the primary sample of treated and matched firms is 460 firm-years. Model (1) illustrates that α3 is positive,

8

http://www.frbsf.org/economic-research/economists/daniel-wilson

26

ACCEPTED MANUSCRIPT but not significantly different from zero for the full sample. However, if network connectedness facilitates innovation activities, then more connected firms should have the strongest initial response to an exogenous increase in the incentive to innovate; the effect should be reflected by a positive α3 coefficient estimate. To test if the ability to innovate is conditional on the degree of

T

connectedness, we sort the full sample and create subsets based on centrality terciles and re-

IP

estimate Model (1) for each subset. The results in Models (2)-(3) support this premise: the α3

CR

interaction coefficient estimate is positive and significant at the 5 percent level for the highest

AN

connectedness plays a role in innovation decisions.

US

tercile centrality firms and not significant for the lower two terciles, suggesting that board

6. Innovation, Firm Connectedness, and the Pricing of Corporate Debt Securities

M

Our primary findings connote that information transmission associated with network

ED

connections reduces uncertainty surrounding innovation. Thus, we surmise that innovation has varying effects on the pricing of debt securities according to the level of firm connectedness.

PT

However, an alternative explanation is that more connected boards may be ‘busier’ with less

CE

time to monitor managers (e.g. Faleye, Hoitash and Hoitash, 2011) which, in turn, may result in greater managerial risk taking including innovation activity.9 To distinguish between these views

AC

and to add insight to the prior findings of Shi (2005) and Eberhart, Maxwell and Siddique (2004) who also study the association between innovation and bond pricing, we use a broad sample of at-issue yield spreads obtained from the SDC Platinum database to examine if the extent of network changes the perceived impact of innovation on cash flow risk. Unlike expected stock return, yield-to-maturity is a deterministic measure of return that has well-defined risk premium components related to default, liquidity, and information. Using the at-issue yield-to-maturity 9

We thank our reviewer for suggesting this alternative explanation.

27

ACCEPTED MANUSCRIPT obtained from SDC and constant-maturity Treasury bond indices obtained from the Federal Reserve of St. Louis Economic Data (FRED), we calculate the at-issue credit yield spread for each bond by subtracting the yield-to-maturity for the same time to maturity on the Treasury yield curve using linear interpolation.10

T

We base our analysis on the broad centrality measure described above. We begin by

IP

regressing at-issue yield spreads on the innovation input and output measures (R&D intensity,

CR

logged number of patents, and logged number of citations), the composite board centrality measure, and independent variables that control for additional bond- and firm-level

US

characteristics likely to be related to the risk premium on corporate bonds. The cross-sectional

AN

model is specified as follows:

Yield spread  α0  α1 Innovation measure  α2Centrality  α3 Residual bond rating  α4Callable  α5 Putable  α6 Coupon rate  α7Time - to - maturity   8 Log(Issue amount)  α9 Baa  Aaa spread

(2)

M

 10Log(No. estimates)  α11Tobin' s q  12Sales growth  α13ROA  α14 Std. ROA  α15 Book leverage

ED

 α16 Firm size  Fama - French 30 industry fixed effects  Year fixed effects  e j,t

Credit ratings are determined by variables that are also used to explain yield spreads; thus, to

PT

discern the impact these variables have on yield spreads independent of their effect on credit

CE

ratings, we follow Mansi, Maxwell and Miller (2004) by creating a residual bond rating variable that is purged of the information contained in the bond- and firm-specific control variables.

AC

Residual bond rating is the residual of a regression of Moody’s bond ratings (converted to numerical equivalents ranging from 1 (“C”) to 21 (“Aaa”)) on the right side variables specified in Equation (2) and provides an overall measure of default risk independent of the direct effects the additional control variables may have on bond ratings.

10

FRED provides daily yields to maturity for constant maturity Treasury bond indices for 3-month, 6-month, 1-year, 2-year, 3-year, 5-year, 7-year, 10-year, and 20-year, and 30-year maturities.

28

ACCEPTED MANUSCRIPT In addition to the board centrality measure, the additional control variables are drawn from a large body of work on the determinants of yield spreads (e.g. Mansi, Maxwell and Miller, 2011; Ortiz-Molina, 2006; Klock, Mansi and Maxwell, 2005; Bhojraj and Sengupta, 2003). The binary variables callable and putable control for embedded call and put

T

options, respectively. Time to maturity controls for the effects of bond term on yield spread and

IP

coupon rate controls for positive coupon effects documented in prior work (e.g. Campbell and

CR

Taksler, 2003). Logged issue amount controls for liquidity: Larger issues are associated with economies of scale in underwriting and reduction in liquidity risk (Bhojraj and Sengupta, 2003).

US

Exposure to systematic fluctuations in credit spreads are captured by Baa-Aaa spread; we obtain

AN

monthly values of the seasoned Moody’s Baa and Aaa corporate bond yield indices from the St. Louis Fed economic data repository (FRED). With respect to the firm-level explanatory

M

variables, we control for analyst coverage and the quality of the information environment

ED

surrounding the firm with Logged No. Analyst estimates. Market-book ratio (sales growth) gauge expected (realized) growth, respectively. Profitability is measured with ROA, while Std. (ROA)

PT

measures cash flow risk. Book leverage is an alternative gauge of proximity to default. Finally,

CE

following the analysis above, we use logged sales to control for firm size. We control for unobservable effects related to industry and time by including Fama-French 30 industry and year

AC

indicator variables.

Table 7 provides summary statistics for the at-issue sample and the control variables used in the regression models. There are 2,769 at-issue yield spread observations with non-missing values of the centrality measure and additional control variables used in Equation (2). Approximately two-thirds of the sample bonds contain call provisions and a small minority contain put provisions. About 37 percent of bonds are rated at the Baa level, the most common

29

ACCEPTED MANUSCRIPT rating category within the sample. The typical (median) bond has a yield-to-maturity (yield spread) of 5.37 (1.62) percent, has an original maturity of about 10 years, and has an issue size of approximately $750 million. Descriptive statistics for the innovation measures are provided in Panel C. R&D intensity has a mean of 0.0202. In Panel C, there are 15 analysts following the

T

typical firm. The typical (median) market-book ratio is 1.56 with five-year sales growth of about

IP

6 percent. ROA (Std. ROA) is 6.07 percent (2.16 percent), the leverage ratio of 29 percent, and

CR

sales is 11,067 ($Mil.)

In Table 8, we estimate Equation (2) over centrality subsets, using the first two terciles to

US

represent firms with low to moderate levels and the top tercile to represent high levels of network

AN

connectedness. In Models (1)-(2) using R&D intensity, innovation input is positive albeit not significantly related to yield spread within the lowest two terciles. However, R&D intensity

M

becomes negatively and highly significant (p=0.00) in the top tercile, suggesting that network

ED

centrality reduces bondholder perceptions of risk associated with innovation input. In Models (3)-(4), we estimate Equation (2) using logged number of patents as the measure of

PT

innovativeness. Consistent with the R&D intensity coefficient estimate in Model (1), number of

CE

patents is positive and significant at the 5 percent level in the low-to-mid category of centrality. In contrast, number of patents becomes negative and statistically significant at the 10 percent

AC

level when board centrality is in the top tercile of the sample, indicating that the perception of risk surrounding innovation output decreases when firms have greater centrality. Finally, we repeat these regressions using logged number of citations in Models (5)-(6). Similar to our findings using R&D intensity and number of patents, number of citations is positive and significant at the 5 percent level for centrality terciles 1-2 and significantly negative at the 5 percent level in the top tercile. Viewed collectively, these results support the premise that bond

30

ACCEPTED MANUSCRIPT market participants have sharply different views of the risk consequences of innovation when firms are more (less) central: Lower centrality implies greater risk and more uncertain output to innovative activity, while higher centrality lowers the perceived risk and reduces uncertainty about the effects of innovation.11

T

The results of Table 8 are potentially subject to endogeneity bias due to the joint

IP

determination of yield spread and innovation activity. As a test of robustness, we examine

CR

changes in yield spreads around news of innovation output using secondary bond market transactions. We obtain corporate bond transactions data from TRACE (Trade Reporting and

US

Compliance Engine) and Mergent FISD (Fixed Income Securities Database). The TRACE

AN

database begins in 2002 and includes the time of execution, price, yield to maturity, and trade volume. The Mergent FISD database consists of pricing (buy and sell) information by life

M

insurance companies from the National Association of Insurance Commissioners (NAIC). To

ED

maximize the number of bonds used in the analysis, we combine the two databases and eliminate any duplicate transactions. For each day, we aggregate multiple buy and/or transactions into a

PT

daily yield to maturity using the par amount of each transaction as weights. Following the

CE

process described above, we estimate a yield spread by subtracting the interpolated constant maturity Treasury curve.

AC

To measure the effect of innovation output on bond prices, we examine changes in yield spread around patent filing dates from the NBER patent database. Specifically, we select patents with no other patent grants from the same firm within 90 days before and after. To the extent that the event is unanticipated by the market, the change in the yield spread following the filing

11

Alternatively, we estimated three models using the full sample that include interactions between each respective innovation measure and firm network centrality. The interaction coefficient estimate measures the amount of change in the slope of the regression of yield spread on the innovation measure when centrality changes by one unit. The coefficient estimates for all three interactions are negative and significant at the 5 percent levels or lower.

31

ACCEPTED MANUSCRIPT reflects the response of bond market participants to an exogenous change in innovation output.12 Our basic methodology is to identify the closest transaction date prior to- and following the filing date for bonds that have at least one transaction within the 30-day period before- and following the filing. To estimate the pricing impact of the filing, we subtract the post-filing- from the pre-

T

filing yield spread. We aggregate multiple changes in bond yield spreads for a given issuer to a

IP

weighted average firm basis using the offering amount of each bond as weights. Table 9 Panel A

CR

provides characteristics of the resulting sample, including weighted average yield spreads prior to- and following the filings, total par amounts transacted on the pre- and post-filing dates, and

US

the proximity of the transaction dates to the filing date, and characteristics of the bonds in the

AN

analysis (years to maturity, rating, and years from issuance).

Table 9 Panel B illustrates reactions of bond prices to patent filings conditioned on

M

connectedness. To minimize the influence of outlying yield spread observations, we winsorize

ED

the distribution at the 2.5 percent tails. We follow the analyses above by creating centrality terciles based on the year of the filing. Overall, we find that yield spreads decrease by

PT

approximately 13 basis points following the filing. In support of our earlier cross-sectional

CE

findings using the at-issue sample, Panel B demonstrates that this effect is primarily driven by top-tercile centrality issuers: for firms in this subset, the mean (median) yield spread decreases

AC

by 25 basis points (24 basis points) following the filing, compared to a reduction of 6 basis points (7 basis points) for the lower two terciles. In sum, these results continue to support the interpretation that the advising services and other resources provided by connected directors help reduce the perceived risk of innovation by outside market participants, resulting in a lower risk premium demanded on the firm’s debt securities. 12

Since filing a patent with the U.S. Patent and Trademark Office is public news, we use filing dates as announcement dates. It is possible that a filing of a specific patent is anticipated by the market, but this will only bias our results in the opposite direction.

32

ACCEPTED MANUSCRIPT Prior research provides a theoretical rationale (e.g. Duffie and Lando, 2001) and empirical evidence (e.g. Yu, 2005; Lu, Chen and Liao, 2010) for the premise that yield spread becomes more sensitive to information uncertainty as bond maturity decreases. Accordingly, in Table 9 Panel C we examine if yield spread changes are pronounced for short-maturity bonds by

T

disaggregating the firm-level changes in yield spread to the bond-level; we regress these changes

IP

in bond yield spreads on the centrality measure, a binary variable for short-maturity debt, an

CR

interaction between centrality and the maturity indicator, and additional control variables for credit rating (bond rating), bond liquidity (logged bond age and the combined amount of bonds

US

transacted on the yield spread transaction dates), the closeness in days that the two transaction

AN

dates are to the patent filing date, and firm size. The centrality × maturity less than 5y interaction is significant at the 10 percent level. Furthermore, reducing the maturity to 3 years or less

M

intensifies the magnitude and the significance of the interaction’s coefficient estimate

ED

approximately twice (to 5 percent), respectively. Overall, these results provide further support for the intuition that greater network centrality allows market participants to better discern the

7. Conclusions

CE

PT

impact of innovation output on the risk premia of information-sensitive debt.

AC

In this study, we examine if network connectedness affects firm innovation and the value placed on innovation by market participants. Specifically, we investigate whether the various dimensions of a firm’s network affect firms’ R&D expenditures and patenting activities, and if so, what structural properties enhance firm innovation. Furthermore, we also examine whether the role of network varies depending on firm industry. In addition, we investigate if firm network affects the pricing of innovation by bond market participants.

33

ACCEPTED MANUSCRIPT We use board interlocks to create interfirm networks. As firms form and maintain board interlocks, they create a network of direct and indirect ties with each other. This network, in turn, can influence the dynamics of information diffusion among firms and affect firm innovation since information is integral to the innovation process. To answer these questions, we construct a

T

set of network measures that capture network centrality, density, diversity, innovativeness, and

IP

propinquity. We find that these different dimensions of firm network – centrality, diversity,

CR

innovativeness, and propinquity – indeed affect innovation activities of firms, especially for those firms in less tangible industries. Firms that have more extensive and central networks and

US

connected to other innovative firms, firms in more diverse industries, or local firms are

AN

associated with higher innovation input and output. However, density (i.e., cohesiveness) of a firm’s network does not matter in this context of innovation input and output supporting the view

M

that cohesive networks can have homogenous and redundant information which may not be

ED

relevant for innovation. The above results are robust to the use of an instrumental variables approach and a natural experiment of state-level changes in R&D tax incentives as well as

PT

additional control variables including CEO network.

CE

We also find that firm network affects pricing of innovation by bond market participants: innovation has a marginal positive (negative) effect on yield spread when firms have lower

AC

(higher) connectedness, suggesting that the market perceives innovative activities by less connected firms as more uncertain and hence, risk-increasing. Further tests on bond yield spread changes surrounding patent filings support these results. By documenting how different aspects of firm network affect innovation, our paper contributes to the growing literature on networks in finance. This work is also relevant to the board of director literature by showing how boards aid

34

ACCEPTED MANUSCRIPT information diffusion and contributes to our understanding of how innovation occurs and is

AC

CE

PT

ED

M

AN

US

CR

IP

T

priced by the market participants.

35

ACCEPTED MANUSCRIPT References Anderson, R., David, R., Upadhyay, A., Zhao, W., 2011. The economics of director heterogeneity. Financ. Manage. 40, 5-38. Barnea, A., Guedj, I., 2009. Director networks. Unpublished working paper.

T

Beckman, C., Haunschild, P., 2002. Network learning: The effects of partners' heterogeneity of experience on corporate acquisitions. Admin. Sci. Quart. 47, 92-124.

CR

IP

Betts, J.R., Wilson, C.W., Wilson, M.K., 2001. The effects of unions on research and development: an empirical analysis using multi-year data. Can. J. Economics. 34, 785-806. Bhojraj, S., Sengupta, P., 2003. Effect of corporate governance and bond ratings and yields: the role of institutional investors and outside directors. J. Bus. 76, 455-475.

US

Bizjak, J., Lemmon, M., Whitby, R., 2009. Option backdating and board interlocks. Rev. Financ. Stud. 22, 4822-4847.

AN

Bonacich, P., 1972. Factoring and weighting approaches to status scores and clique identification. J. Math. Sociol. 2, 113-120.

ED

M

Borgatti, S., Everett, M., Freeman, L., 2002. UCINet 6 for Windows: software for social network analysis. Harvard, MA: Analytic Technologies. Bouwman, C., 2011. Corporate governance propagation through overlapping directors. Rev. Financ. Stud. 24, 2358-2394.

PT

Bouwman, C., Xuan, Y., 2010. Director overlap and firm financial policies. Case Western Reserve University and Harvard University Working Paper.

CE

Burt, R., 1992. Structural holes: The social structure of competition. Cambridge, MA: Harvard University Press.

AC

Butler, A.W., 2008. Distance Still Matters: Evidence from municipal bond underwriting. Rev. Financ. Stud. 21, 763-784. Cai, Y., Sevilir, M., 2012. Board Connections and M&A Transactions. J. Financ. Econ. 103.2, 327-49 Campbell, J.Y., Taksler, G.B., 2003. Equity volatility and corporate bond yields. J. Financ. 58, 2321-2350. Chen, T., Chen, Y. Liao, H., 2011. Labor unions, bargaining power and corporate bond yield spreads: Structural credit model perspectives. J. Bank. Financ. 35, 2084-2098.

36

ACCEPTED MANUSCRIPT Chan, K., Chen, H-K, Hong, L-H, Wang, Y., 2015. Stock market valuation of R&D expenditures-the role of corporate governance. Pac-Basin. Financ. J. 31, 78–93. Chen, H., Kacperczyk, M., Ortiz-Molina, H., 2011a. Do nonfinancial stakeholders affect the pricing of risky debt? Evidence from unionized workers. Rev. Financ. 16, 347-383. Chen, H., Kacperczyk, M., Ortiz-Molina, H., 2011b. Labor unions, operating flexibility, and the cost of equity. J. Financ. Quant. Anal. 46, 25-58.

IP

T

Chen, Y., Podolski, E.J., Veeraraghavan, M., 2015. Does managerial ability facilitate corporate innovative success?. J. Empir. Financ. 34, 313-326.

CR

Chikh, S., Filbien, J., 2011. Acquisitions and CEO Power: Evidence from French Networks. J. Corp. Financ. 17(5), 1221–36.

US

Chiu, P., Teoh, S., Tian, F., 2013. Board interlocks and earnings management contagion. Account. Rev. 88 (3), 915-944

AN

Chuluun, T., Prevost, A., Puthenpurackal, J., 2014. Board ties and the cost of corporate debt. Financ. Manage. 43, 533-568.

M

Ciftci, M., Cready, W.M., 2011. Scale effects of R&D as reflected in earnings and returns. J. Account. Econ 52, 62-80.

ED

Cohen, L., Frazzini, A., Malloy, C., 2008. The small world of investing: board connections and mutual fund returns. J. Polit. Econ. 116, 951-979.

PT

Cohen, L., Diether, K., Malloy, C., 2013. Misvaluing innovation. R. Financ. Stud. 26, 635-666. Coleman, J.S., 1988. Social capital in the creation of human capital. Am. J. Sociol. 94, 95-120.

CE

Dasgupta, S., Zhang, K., Zhu. C., 2015. Innovation, social connections, and the boundary of the firm. Working paper, Hong Kong University of Science & Technology.

AC

Dass, N., Kini, O., Nanda, V., Onal, B., Wang, J., 2014. Board expertise: do directors from related industries help bridge the information gap?. Rev. Financ. Stud. 27, 1533-1592. Davis, G., F., Greve, H., R., 1997. Corporate elite networks and governance changes in the 1980s. Am. J. Sociol. 103, 1-37. Drucker, P. F., 1993. Postcapitalist Society. New York: HarperCollins Publishers. Duffie, D., Lando, D., 2001. Term structure of credit spreads with incomplete accounting information. Econometrica. 69, 633–664. Duqi, A., Jaafar, A., Torluccio, G., 2015. Mispricing and risk of R&D investment in European firms. Eur. J. Financ. 21, 444–465. 37

ACCEPTED MANUSCRIPT Eberhart, A. C., Maxwell, W. F., Siddique, A., 2004. An examination of long-term abnormal stock returns and operating performance following R&D increases. J. Financ. 59, 623–50. El-Khatib, Rwan, Kathy Fogel, and Tomas Jandik, 2015. CEO network centrality and merger performance. J. Financ. Econ. 116 (2), 349–82. Engelberg, Joseph, Pengjie Gao, and Christopher A. Parsons, 2013. The price of a CEO’s rolodex. Rev. Financ. Stud. 26 (1), 79–114.

IP

T

Faleye, O., Kovacs, T., Venkateswaran, A., 2014. Do better-connected CEOs innovate more?. J. Financ. Quant. Anal. 49, 1201-1225.

CR

Fang, Y., Francis, B., Hasan, I., 2012. More than connectedness - heterogeneity of CEO social network and firm value. Unpublished working paper.

US

Ferris, S.P., Javakhadze, D., French, D.W., 2016. Managerial social capital and financial development: a cross‐ country analysis. Unpublished working paper.

AN

Fracassi, C., 2015. Corporate finance policies and social networks. Manage. Sci. forthcoming.

M

Francis, B., Hasan, I., Huang, Y., Sharma, Z., 2012. Do banks value innovation? Evidence from US firms. Financ. Manage. 41, 59–85.

ED

Freeman, L., 1979. Centrality in social networks: conceptual clarification. Soc. Networks. 1, 215-239.

PT

Fu., R., Gupta-Mukherjee, S., 2014. Geography, informal information flows and mutual fund portfolios. Financ Manage. 43, 181–214.

CE

Garmaise, M., Moskowitz, T., 2003. Informal financial networks: theory and evidence. Rev. Financ. Stud. 16, 1007-1040.

AC

Gilsing, V., Nooteboom, B., Vanhaverbeke, W., Duysters, G., van den Oord, A., 2008. Network Embeddedness and the Exploration of Novel Technologies: Technological Distance, Betweenness Centrality and Density. Res. Policy. 37, 1717–31. Glaeser, Edward, David Laibson, Jose Scheinkman, and Christine Soutter, 2000. Measuring trust. Q. J. Econ. 115(3), 811-846. Goldman, J. and J. Peress, 2016. Firm innovation and financial analysis: how do they interact? Working paper, London School of Economics. Granovetter, M., 1992. Economic institutions as social constructions: a framework for analysis. Acta Sociologica 35, 3–11.

38

ACCEPTED MANUSCRIPT Hall, B. H., A. B. Jaffe, and M. Trajtenberg, 2005. Market value and patent citations. Rand. J. Econ. 36(1): 16-38 Hall, B. H., Jaffe, A.B., Trajtenberg, M., 2001. The NBER patent citation data file: lessons, insights and methodological tools. NBER Working Paper 8498. Hansen, M.T., 1999. The search-transfer problem: the role of weak ties in sharing knowledge across organization subunits.”. Admin. Sci. Quart. 44, 82–111.

IP

T

Helmers, C., Patnam, M., Rau, R.P., 2015. Do board interlocks increase innovation? Evidence from a corporate governance reform in India. Unpublished working paper.

CR

Hirschleifer, D., Low, A., Teoh, S., 2012. Are overconfident CEOs better innovators?. J. Financ. 67, 1457–1498.

US

Hochberg, Y., Ljungqvist, A., Lu, Y., 2007. Whom you know matters: Venture capital networks and investment performance. J. Financ. 62, 251-302.

AN

Hochberg, Y., Ljungqvist, A., Lu, Y., 2010. Networking as a barrier to entry and the competitive supply of venture capital. J. Financ. 65, 829-859.

M

Hong, H., Kubik, J., Stein, J., 2004. Social interaction and stock-market participation. J. Financ. 59, 137-163.

ED

Hsu, P-H, Lee, H-H, Liu, A-Z, Zhang, Z., 2015. Corporate innovation, default risk, and bond pricing. J. Corp. Financ. 35, 329-344.

PT

Hwang, B.H., Kim, S., 2012. Earnings management and social ties. Unpublished working paper.

CE

Jaffe, A., Trajtenberg, M., Henderson, R., 1993. Geographic Localization of Knowledge Spillovers as Evidenced by Patent Citations. Q. J. Econ. 108 (3), 577-598.

AC

Kali, R. Reyes, J., 2010. Financial contagion on the international trade network. Econ. Inq. 48, 1072-1101. Kang, J-K, Liu, W-L, Low, A., Zhang, L., 2014. Friendly boards and innovation. Working paper, Nanyang Technological University. Khwaja, A., Mian, A., 2005. Do Lenders Favor Politically Connected Firms? Rent provision in an Emerging Financial Market. Q. J. Econ. 120(4), 1371-1411. Khanna, V., Kim, E., Lu, Y., 2015. CEO Connectedness and Corporate Fraud. J. Financ. 70, 1203–1252. Klasa, S., Maxwell, W.F., Ortiz-Molina, H., 2009. The strategic use of corporate cash holdings in collective bargaining with labor unions. J. Financ. Econ. 92 421-442. 39

ACCEPTED MANUSCRIPT Klock, M.S., Mansi, S., Maxwell, W.F., 2005. Does corporate governance matter to bondholders?. J. Financ. Quant. Anal. 40, 693-719. Kothari, S. P., Laguerre, T. E., Leone, A., J., 2002. Capitalization versus expensing: evidence on the uncertainty of future earnings from capital expenditures versus R&D outlays. Rev. Account. Stud. 7, 355-382

T

Lev, B., Sougiannis, T., 1996. The capitalization, amortization, and value-relevance of R&D. J. Account. Econ. 21, 107–138

CR

IP

Liu, Y., 2008. Employment networks and the CEO labor market. Working paper, University of Maryland. Lu, C., Chen, T., Liao, H., 2010. Information uncertainty, information asymmetry and corporate bond yield spreads. J. Bank. Financ. 34, 2265-2279.

US

Mansi, S., Maxwell, W.F., Miller D.P., 2004. Does auditor quality and tenure matter to investors? J. Account. Res. 42, 755-793.

AN

Mansi, S., Maxwell, W.F., Miller D.P., 2011. Analyst forecast characteristics and the cost of debt. Rev. Account. Stud. 16, 116-142.

ED

M

Massa, M., Simonov, A., 2012. The role of college interaction in portfolio choice. Forthcoming, Rev. Financ. McPherson, M., Smith-Lovin, L., Cook, J., 2001. Birds of a feather: homophily in social networks. Annu. Rev. Sociol. 27, 415-444.

PT

Oh, W.Y., Barker, V.L. III., 2015. Not all ties are equal: CEO outside directorships and strategic imitation in R&D investment. Forthcoming, J. Manage.

CE

Ortiz-Molina, H., 2006. Top management incentives and the pricing of corporate public debt. J. Financ. Quant. Anal. 41, 317-340.

AC

Phelps, C.C., 2010. A longitudinal study of the influence of alliance network structure and composition on firm exploratory innovation. Acad. Manage. J. 53, 890-913. Reagans, R., Zuckerman, E.W., 2001. Networks, diversity, and productivity: the social capital of corporate R&D teams. Organ. Sci. 12, 502-517. Shi, C., 2003. On the trade-off between the future benefits and riskiness of R&D: A bondholders’ perspective. J. Account. Econ. 35: 227-54. Sorenson, O., Stuart, T., 2001. Syndication networks and the spatial distribution of venture capital investments. Am. J. Sociol. 106 (6), 1546-1588.

40

ACCEPTED MANUSCRIPT Stuart, T., Yim, S., 2010. Board interlocks and the propensity to be targeted in private equity transactions. J. Financ. Econ. 97, 174-189. Teo, M., 2009. The geography of hedge funds. Rev. Financ. Stud. 22, 3531-3561. Wilson, D., 2009. Beggar Thy Neighbor? The in-state, out-of-state, and aggregate effects of R&D tax credits. Rev. Econ. Stat. 91(2), 431-436.

AC

CE

PT

ED

M

AN

US

CR

IP

T

Yu, F., 2005. Accounting transparency and the term structure of credit spreads. J. Financ. Econ. 75, 53–84.

41

ACCEPTED MANUSCRIPT Appendix 1: Variable description Firm network measures

x

ij

Percent of all other firms a specific firm is connected to. Specifically, Degree i 

Degree

j

n 1

, where xij equals

to one when there is a tie between firms i and j, and n equals to the number of all firms in the network. Eigenvector centrality captures how close a firm is to all other firms by taking into account the eigenvector

Eigenvector

centralities of the interlocked firms. Formally, ei



x e

ij j , where λ is a proportionality constant, and

j

Betweennes si 

Betweenness

b

IP

T

xij equals to one when there is a link between firms i and j, and we normalize this measure by dividing it by the maximum possible eigenvector centrality in the network. It measures how often a firm falls on the shortest possible paths between pairs of other firms. jik , where bjik is the proportion of all paths linking distinct firms j and k that pass

CR

jk

through firm i, and we divide it by the maximum possible betweenness in the network to obtain normalized betweenness. The percent of all possible connections that can exist among the interlocked firms of a firm that are actually

x

present. Densityi 

AN

CE

Ratio of R&D expenditures (XRD) to total assets (AT), and firms with missing R&D information are assigned a R&D value of zero. Source: Compustat Logged number of patents a firm applies for in a year. Source: NBER patent database Logged number of exploratory patents (i.e., patents filed in technology classes that the firm did not file any patents in the previous five years) a firm applies for in a year. Source: NBER patent database. (Inverse of the) Herfindahl index of patents a firm filed in a year in different technology classes. Source: NBER patent database. Logged number of citations a firm’s patents subsequently receive. Source: NBER database Logged (number of citations/ number of patents). Source: NBER patent database Logged number of analyst estimates (NUMEST). Source: I/B/E/S Annual sales (SALE) in millions of USD. Source: Compustat Ratio of market value of assets to book value assets. Market value of assets is defined as total assets plus market equity minus book equity. Market equity is computed by multiplying common shares outstanding by fiscal-year closing price. Book equity refers to stockholders’ equity. Source: Compustat Five year geometric growth in SALE. Source: Compustat Debt in current liabilities plus long-term debt (DLTT) divided by total assets (AT). Source: Compustat Net income (NI) divided by total assets (AT). Source: Compustat Standard deviation of ROA for the prior five years. Source: Compustat Ratio of cash holdings (CH) totota assets (AT). Source: Compustat Property, plant and equipment (PPE) divided by lagged total assets (AT). Source: Compustat Percentage of workers covered by collective bargaining agreements at the CIC industry level. Source: www.unionstats.com

AC

Patent diversity

whom firm i is connected to, and ni equals to the number of all firms interlocked to a firm i (i.e., degree of firm i). The proportion of a firm’s relationships to its partners that are considered non-redundant, calculated as the number of partners minus the average degree of partners within the firm network, not counting ties to the focal firm itself. (Inverse of the) Herfindahl index computed using the percentage of interlocked firms in different Fama2 French 30 industries. 𝐼𝑛𝑑𝑢𝑠𝑡𝑟𝑦. 𝑑𝑖𝑣𝑒𝑟𝑠𝑖𝑡𝑦𝑖 = 1/ ∑𝑁 𝑚=1 𝑠𝑚 where sm refers to the percentage of all partners of firm i in industry m. Logged number of all interlocked firms with positive R&D expenditure. Logged number of all interlocked firms located in the same metropolitan statistical area (MSA) as the firm. Logged number of all interlocked firms that are in the same Fama-French 30 industry as the firm. Logged number of all interlocked firms located in the same MSA as the firm and also have positive R&D expenditure. Logged number of all interlocked firms that are in the same Fama-French 30 industry as the firm and also have positive R&D expenditure.

PT

Network innovativeness Geographic propinquity Industry propinquity Innovative geographic propinquity

Log (No. patents) Log (No. exploratory patents)

, where xjk equals to one when there is a link between firms j and k with

M

Industry diversity

Other firm variables R&D intensity

ni (ni  1)

ED

Network non-redundancy

Innovative industry propinquity

jk

jk

US

Density

Log (No. citations) Log (Average no. citations) Log (No. analysts) Sales (($Mil.)) Market-book ratio

Sales growth Book leverage ROA Std. (ROA) Cash Tangibility Unionization rate Appendix (cont’d)

42

ACCEPTED MANUSCRIPT CEO network

Number of individuals a CEO overlapped with through employment, education, and other activities in the Boardex database. Source: Boardex. Capital expenditure (CAPX) scaled by total assets (AT). Source: Compustat 2 Herfindahl Hirschman Index for the issuer’s 3-digit SIC code, calculated as ∑𝑁 𝑖=1 𝑠𝑖 , where si is the proportion of sales of firm i in the issuer’s 3-digit SIC industry and N is the number of firms in the industry. Source: Compustat Equity ownership by institutions holding at least 5 percent. Source: Thomson Financial.

Capital expenditure Industry HHI

Institutional ownership

Table 1: Summary Statistics The following table provides the descriptive statistics of the various variables used in our study. The definitions of the variables are included in the Appendix. Median

Max

No. Obs.

15.33

13.54

0

4.39 3.69

3.28 2.82

0 1

3.36

3.43

1.14 0.82

PT

0.036 16.67 0.71 4.29 130.88 1.78 5.6 1.92 0.097 0.22 0.03 0.11 9.27 0.048 0.321 628.18 9.75 0.81

IP

15,779 15,779 15,779

10.00

100.00

13,598

3.25 3.00

22.27 21.28

15,779 15,779

0

2.00

22

14,015

1.66 1.21

0 0

1.00 0

16 10

12,902 13,691

0.98 1.04

0 0

0 0

10 10

12,902 13,691

0.058 112.01 2.14 4.75 1,174.50 4.97 12.76 1.11 0.15 0.19 0.13 0.11 9.80 0.047 0.217 733.73 7.18 0.17

0 0 0 1 0 0 0 0.70 -0.26 0 -0.68 0.001 0 0.001 0.037 0 1 0.18

0.005 0 0 2.67 0 0 1.4 1.57 0.072 0.20 0.05 0.08 5.5 0.033 0.262 394 8 0.84

0.29 4108 38 50.93 53,323 101.88 86.85 6.91 1.23 0.87 0.27 0.55 74.1 0.337 1 6,033 49 1.03

15,764 8,557 8,347 2,844 8,524 8,524 15,770 15,725 15,351 15,776 15,767 15,624 14,632 15,487 3,417 5,534 14,837 12,150

43

US

0.042 0.044 0

CR

2.14 4.98 0.36

AN

0.25 0.99 0.038

ED

0.24 2.41 0.002

M

AC

Min

0.31 2.12 0.024

0.52 0.55

CE

Panel 1B: Other Firm Variables R&D intensity Number of patents Number of exploratory patents Patent diversity Number of citations Average number of citations Sales ($ billions) Market-book ratio Sales growth Book leverage ROA Cash Unionization rate (%) Capital expenditure Industry HHI CEO network Analyst estimates Institutional ownership

Std. Dev.

T

Mean Panel 1A: Network Variables Centrality measures (%) Degree Eigenvector Betweenness Cohesion measure (%) Density Diversity measures Industry diversity Network non-redundancy Innovativeness measure Network innovativeness Propinquity measure Geographic propinquity Industry propinquity Innovative propinquity measure Innovative geographic propinquity Innovative industrial propinquity

ACCEPTED MANUSCRIPT Table 2: Firm Network Connectedness and R&D Intensity Panel A provides the estimated coefficients of linear regressions of R&D intensity using the full sample. In Panel B, the regressions are estimated for subsamples of firms in intangible and tangible industries, respectively, and only the estimated coefficients on the network variables are reported. P-values are based on robust cluster-adjusted standard errors and are provided in parentheses. ***, **, and * correspond to significance at the 1, 5, and 10 percent level, respectively.

Model (1) Network Centrality

Model (2)

Model (3)

Model (4)

0.0009** (0.02)

IP

Panel 2A: Full Sample Model (5)

Model (6)

0.0032*** (0.00) 0.0000 (0.91)

Diversity

0.0136*** (0.00)

CR

Innovativeness Geographic propinquity

Debt ratio ROA Cash

ED

Sales growth

-0.0032*** (0.00) 0.0120*** (0.00) -0.0114** (0.03) -0.0790*** (0.00) 0.0846*** (0.00) 0.0164** (0.04) -0.0005*** (0.00) -0.0403*** (0.00) -0.0169*** (0.00) Yes Yes 13,991 0.457 171.4

Unionization rate

PT

Capital expenditure

AC

CE

Industry HHI

-0.0045*** (0.00) 0.0124*** (0.00) -0.0139*** (0.01) -0.0738*** (0.00) 0.0859*** (0.00) 0.0235*** (0.00) -0.0005*** (0.00) -0.0407** (0.01) -0.0136** (0.01) Yes Yes 12,265 0.461 149.6

AN

Market-book ratio

-0.0046*** (0.00) 0.0118*** (0.00) -0.0119** (0.02) -0.0771*** (0.00) 0.0841*** (0.00) 0.0210*** (0.01) -0.0005*** (0.00) -0.0384*** (0.01) -0.0171*** (0.00) Yes Yes 13,991 0.462 172.7

M

Log (Sale)

US

Innovative geographic propinquity

Industry fixed effects Year fixed effects No. Obs. R-squared F-statistic

T

Density

-0.0073*** (0.00) 0.0117*** (0.00) -0.0139*** (0.01) -0.0673*** (0.00) 0.0787*** (0.00) 0.0304*** (0.00) -0.0005*** (0.00) -0.0295* (0.07) -0.0149*** (0.01) Yes Yes 12,415 0.481 157.5

0.0039** (0.01)

-0.0043*** (0.00) 0.0120*** (0.00) -0.0140*** (0.01) -0.0720*** (0.00) 0.0881*** (0.00) 0.0212** (0.01) -0.0004*** (0.00) -0.0390** (0.02) -0.0141** (0.01) Yes Yes 11,592 0.463 143.2

0.0129*** (0.00) -0.0047*** (0.00) 0.0115*** (0.00) -0.0130** (0.01) -0.0707*** (0.00) 0.0856*** (0.00) 0.0218*** (0.01) -0.0004*** (0.00) -0.0380** (0.02) -0.0139** (0.01) Yes Yes 11,592 0.471 148.1

Panel 2B: Sensitivity of R&D Intensity to Network Measures for Intangible and Tangible Industry Subsamples Intangible No. Obs. Tangible No. Obs. Diff. in coefficients χ2 Centrality 0.0045*** 9,454 0.0003 4,537 17.47*** (0.00) (0.27) Density 0.0000 9,454 0.0000 4,537 0.01 (0.75) (0.39) Diversity 0.0008 8,299 0.0003 3,966 0.67 (0.14) (0.16) Innovativeness 0.0162*** 8,391 0.0024*** 4,024 46.62*** (0.00) (0.00) Geographic propinquity 0.0050** 7,823 0.0012 3,769 2.63* (0.02) (0.13) Innovative geo. propinquity 0.0154*** 7,823 0.0034** 3,769 13.51*** (0.00) (0.02)

44

ACCEPTED MANUSCRIPT Table 3: Firm Network Connectedness and Patents Panels A-C provide the estimated coefficients of linear regressions of various patent measures using the full sample. In Panel D, the regressions are estimated for subsamples of firms in intangible and tangible industries, respectively. P-values are based on robust cluster-adjusted standard errors and are provided in parentheses. ***, **, and * correspond to significance at the 1, 5, and 10 percent level, respectively. Model (2)

Model (3)

Model (4)

Diversity

0.0527*** (0.00)

Innovativeness

Model (5)

Model (6)

T

-0.0006 (0.62)

IP

Panel 3A: Dependent Variable = Log (Number of Patents) Full Sample Model (1) Network Centrality 0.0790*** (0.00) Density

0.2079*** (0.00)

CR

Geographic propinquity Innovative geo. propinquity

Debt ratio ROA Cash R&D intensity Sales growth

ED

Unionization rate Capital expenditure

PT

Industry HHI

CE

Industry fixed effects Year fixed effects No. Obs. R-squared F-statistic

0.3383*** (0.00) 0.1341*** (0.00) -0.4315*** (0.00) -0.6056*** (0.00) -0.0642 (0.77) 4.9581*** (0.00) -0.6091*** (0.00) -0.0076* (0.07) 0.4613 (0.32) 0.1796 (0.25) Yes Yes 6,643 0.467 82.37

0.3364*** (0.00) 0.1348*** (0.00) -0.4261*** (0.00) -0.5821*** (0.00) -0.1316 (0.53) 4.4884*** (0.00) -0.6364*** (0.00) -0.0077* (0.07) 0.5085 (0.26) 0.2076 (0.20) Yes Yes 6,659 0.469 82.80

0.3509*** (0.00) 0.1392*** (0.00) -0.4102*** (0.00) -0.6610*** (0.00) -0.0245 (0.91) 4.5910*** (0.00) -0.5855*** (0.00) -0.0090** (0.03) 0.3952 (0.39) 0.1286 (0.40) Yes Yes 6,222 0.440 69.88

Model (3)

Model (4)

Model (5)

Model (6)

US

Market-book ratio

0.3872*** (0.00) 0.1252*** (0.00) -0.3479*** (0.00) -0.6354*** (0.00) -0.0453 (0.81) 4.7588*** (0.00) -0.7262*** (0.00) -0.0065* (0.09) 0.2288 (0.57) 0.2370 (0.12) Yes Yes 7,498 0.449 81.67

0.2462*** (0.00) 0.3453*** (0.00) 0.1355*** (0.00) -0.4045*** (0.00) -0.6553*** (0.00) -0.0576 (0.79) 4.3810*** (0.00) -0.5833*** (0.00) -0.0090** (0.03) 0.3792 (0.42) 0.1374 (0.37) Yes Yes 6,222 0.444 71.24

AN

0.3526*** (0.00) 0.1224*** (0.00) -0.3557*** (0.00) -0.5926*** (0.00) -0.0444 (0.82) 4.5868*** (0.00) -0.6230*** (0.00) -0.0070* (0.07) 0.2787 (0.48) 0.2359 (0.12) Yes Yes 7,498 0.454 83.94

M

Log (Sale)

AC

Panel 3B: Dependent Variable = Log (Number of Exploratory Patents) Model (1) Model (2) Centrality 0.0265*** (0.01) Density -0.0005 (0.26) Diversity

0.1166*** (0.01)

0.0246*** (0.00)

Innovativeness

0.0759*** (0.00)

Geographic propinquity

0.0379** (0.02)

Innovative geo. propinquity Additional control variables Industry fixed effects Year fixed effects No. Obs. R-squared F-statistic

Yes Yes Yes 7,342 0.326 47.30

Yes Yes Yes 7,342 0.323 46.83

Yes Yes Yes 6,506 0.339 46.45

45

Yes Yes Yes 6,520 0.337 45.97

Yes Yes Yes 6,088 0.310 38.75

0.0969*** (0.00) Yes Yes Yes 6,088 0.315 39.30

ACCEPTED MANUSCRIPT Panel 3C: Dependent Variable = Patent Diversity Model (1) Centrality 0.1927 (0.11) Density

Model (2)

Model (3)

Model (4)

Model (5)

Model (6)

-0.0008 (0.90)

Diversity

0.1649*** (0.00)

Innovativeness

0.6759*** (0.00)

Geographic propinquity

0.1071 (0.60)

Yes Yes Yes 2,519 0.352 17.67

Yes Yes Yes 2,363 0.359 17.27

Yes Yes Yes 2,361 0.360 17.25

IP

Yes Yes Yes 2,519 0.355 17.71

CR

Additional control variables Industry fixed effects Year fixed effects No. Obs. R-squared F-statistic

T

Innovative geo. propinquity

AC

CE

PT

ED

M

AN

US

Panel 4D: Sensitivity of Patent Measures to Network Measures for Intangible and Tangible Subsamples Panel 4D1: Dependent variable = Log (Number of Patents) Intangible No. Obs. Tangible Centrality 0.1029*** 5,079 0.0328 (0.00) (0.20) Density -0.0010 5,079 0.0003 (0.51) (0.79) Diversity 0.0728*** 4,499 0.0159 (0.00) (0.30) Innovativeness 0.2926*** 4,515 0.0724 (0.00) (0.11) Geographic propinquity 0.1166** 4,194 0.0919* (0.03) (0.09) Innovative geo. propinquity 0.2549*** 4,194 0.1496 (0.00) (0.11) Panel 4D2: Dependent Variable = Log (Number of Exploratory Patents) Centrality 0.0296** 4,992 0.0167 (0.02) (0.16) Density -0.0006 4,992 0.0001 (0.27) (0.90) Diversity 0.0297*** 4,426 0.0109* (0.00) (0.07) Innovativeness 0.0913*** 4,439 0.0422** (0.00) (0.04) Geographic propinquity 0.0329* 4,122 0.0328 (0.06) (0.19) Innovative geo. propinquity 0.0964*** 4,122 0.0684* (0.00) (0.08) Panel 4D3: Dependent Variable = Patent Diversity Centrality 0.2189 2,055 0.1494 (0.16) (0.11) Density -0.0003 2,055 0.0049 (0.97) (0.70) Diversity 0.2275*** 1,911 0.0399 (0.00) (0.53) Innovativeness 0.8522*** 1,910 0.3446* (0.00) (0.05) Geographic propinquity -0.0149 1,769 0.7579** (0.95) (0.01) Innovative geo. propinquity 0.2337 1,769 0.9907** (0.42) (0.03)

46

No. Obs. 2,419

0.3439 (0.19) Yes Yes Yes 2,177 0.339 13.72

Yes Yes Yes 2,177 0.337 13.72

Diff. in coefficients χ2 3.40*

2,419

0.45

2,144

5.15**

2,144

10.26***

2,028

0.09

2,028

0.67

2,350

0.70

2,350

0.71

2,080

3.93**

2,081

2.95*

1,966

0.01

1,966

0.32

464

0.13

464

0.12

452

3.34*

451

2.02

408

4.34**

408

1.86

ACCEPTED MANUSCRIPT Table 4: Firm Network Connectedness and Citations Panels A-B provide the estimated coefficients of linear regressions of two patent citations measures using the full sample. Pvalues are based on robust cluster-adjusted standard errors and are provided in parentheses. ***, **, and * correspond to significance at the 1, 5, and 10 percent level, respectively. Panel 4A: Dependent Variable = Log (Number of Citations) Model (1) Model (2) Network Centrality 0.0804** (0.03) Density -0.0007 (0.68) Diversity

Model (3)

Model (5)

Model (6)

T

0.0630** (0.01)

0.2378*** (0.00)

IP

Innovativeness

CR

Geographic propinquity Innovative geo. propinquity

0.3835*** (0.00) 0.1909*** (0.00) -0.6098*** (0.00) -0.8317*** (0.00) 0.2087 (0.53) 5.6740*** (0.00) -0.8442*** (0.00) -0.0065 (0.32) 0.7270 (0.16) 0.1845 (0.34) Yes Yes 6,643 0.420 71.05

US

0.4096*** 0.4447*** (0.00) (0.00) Market-book ratio 0.1818*** 0.1847*** (0.00) (0.00) Debt ratio -0.4906*** -0.4828*** (0.00) (0.00) ROA -0.8204*** -0.8634*** (0.00) (0.00) Cash 0.1379 0.1361 (0.62) (0.62) R&D intensity 5.2838*** 5.4583*** (0.00) (0.00) Sales growth -0.8508*** -0.9558*** (0.00) (0.00) Unionization rate -0.0057 -0.0052 (0.34) (0.37) Capital expenditure 0.4631 0.4120 (0.29) (0.37) Industry HHI 0.2750 0.2761 (0.14) (0.14) Industry fixed effects Yes Yes Year fixed effects Yes Yes No. Obs. 7,498 7,498 R-squared 0.405 0.403 F-statistic 71.27 70.39 Panel 4B: Dependent Variable = Log (Average Number of Citations) Model (1) Model (2) Centrality 0.0271* (0.06) Density -0.0003 (0.70) Diversity

AC

CE

PT

ED

M

AN

Log (Sale)

Innovativeness

Model (4)

Model (3)

0.3842*** (0.00) 0.1912*** (0.00) -0.6076*** (0.00) -0.8085*** (0.00) 0.1172 (0.71) 5.1502*** (0.00) -0.8792*** (0.00) -0.0065 (0.33) 0.7847 (0.12) 0.2212 (0.27) Yes Yes 6,659 0.420 70.87 Model (4)

0.1338** (0.03)

0.4062*** (0.00) 0.1995*** (0.00) -0.5825*** (0.00) -0.8834*** (0.00) 0.2472 (0.45) 5.2375*** (0.00) -0.7876*** (0.00) -0.0088 (0.20) 0.6880 (0.20) 0.1596 (0.41) Yes Yes 6,222 0.396 60.10

0.2974*** (0.01) 0.3985*** (0.00) 0.1947*** (0.00) -0.5745*** (0.00) -0.8759*** (0.00) 0.2084 (0.52) 4.9799*** (0.00) -0.7822*** (0.00) -0.0088 (0.19) 0.6730 (0.22) 0.1698 (0.38) Yes Yes 6,222 0.399 61.08

Model (5)

Model (6)

0.0220** (0.03) 0.0900** (0.01)

Geographic propinquity

0.0412 (0.10)

Innovative geo. propinquity Additional control variables Industry fixed effects Year fixed effects No. Obs. R-squared F-statistic

Yes Yes Yes 7,498 0.344 77.80

Yes Yes Yes 7,498 0.343 76.89

Yes Yes Yes 6,643 0.358 77.20

47

Yes Yes Yes 6,659 0.359 77.30

Yes Yes Yes 6,222 0.342 64.10

0.0906** (0.03) Yes Yes Yes 6,222 0.343 64.85

ACCEPTED MANUSCRIPT Table 5: Two-Stage Least Squares Estimates of Firm Network Connectedness and R&D Intensity Table 5 presents two-stage least squares coefficient estimates of R&D intensity regressed on measures of network connectedness. The instruments are the number of logged public companies and logged average distance. Panel 5A provides first-stage coefficient estimates for these instruments, and Panel 5B provides second-stage estimates for the instrumental variables and additional control variables. P-values are provided in parentheses. ***, **, and * correspond to significance at the 1, 5, and 10 percent levels, respectively. Model (3) 0.050** (0.02) -0.045 (0.56) Yes

0.0355*** (0.01)

Density

Model (4) 0.0376*** (0.00) -0.0010 (0.97) Yes

T

Additional control variables Panel 5B: Second Stage Estimates Network Centrality

Model (2) -0.003 (0.97) 0.057 (0.88) Yes

Model (5) 0.018*** (0.00) -0.039 (0.13) Yes

Model (6) 0.022*** (0.00) -0.024 (0.20) Yes

IP

Log (Distance)

Model (1) 0.033*** (0.00) -0.010 (0.82) Yes

CR

Panel 5A: First Stage Estimates Log (Public companies)

-0.0464 (0.86)

0.0196** (0.03)

US

Diversity Innovativeness

AN

Geographic propinquity Innovative geo. propinquity

Market-book ratio

-0.0635 (0.86) 0.0060 (0.87) 0.0377 (0.90) 0.1303 (0.91) -0.2315 (0.90) 0.0904 (0.84) -0.0020 (0.83) -0.0564 (0.76) -0.1890 (0.82) Yes Yes 12,976 -108.787 0.313 0.030 (0.98) 0.061 (0.80)

ED

Debt ratio

-0.0188*** (0.00) 0.0098*** (0.00) -0.0181** (0.01) -0.0557*** (0.00) 0.0775*** (0.00) 0.0656*** (0.00) -0.0006*** (0.00) -0.0267*** (0.00) -0.0201 (0.48) Yes Yes 12,976 0.043 16.49 11.337 (0.00) 0.000 (0.99)

M

Log (Sale)

ROA

PT

Cash Sales growth

CE

Unionization rate HHI

AC

Capital expenditure Industry fixed effects Year fixed effects No. Obs. R-squared F-statistic K-P LM statistic (p-value) Hansen J statistic (p-value)

48

-0.0240*** (0.01) 0.0100*** (0.00) -0.0232*** (0.00) -0.0518*** (0.00) 0.0880*** (0.00) 0.0893*** (0.01) -0.0008*** (0.00) -0.0254*** (0.01) -0.0265 (0.39) Yes Yes 11,557 0.060 14.67 7.353 (0.02) 0.059 (0.81)

0.0283*** (0.00) 0.0502*** (0.01)

-0.0116*** (0.00) 0.0109*** (0.00) -0.0148*** (0.01) -0.0563*** (0.00) 0.0720*** (0.00) 0.0395*** (0.00) -0.0005*** (0.00) -0.0254*** (0.00) -0.0197 (0.39) Yes Yes 11,548 0.458 23.80 37.947 (0.00) 0.018 (0.89)

-0.0093*** (0.00) 0.0098*** (0.00) -0.0076 (0.31) -0.0657*** (0.00) 0.0941*** (0.00) 0.0365*** (0.00) -0.0003* (0.06) -0.0254*** (0.00) -0.0161 (0.56) Yes Yes 10,943 0.261 17.59 19.596 (0.00) 0.133 (0.72)

0.0487*** (0.00) -0.0073*** (0.00) 0.0097*** (0.00) -0.0096 (0.13) -0.0629*** (0.00) 0.0801*** (0.00) 0.0250*** (0.00) -0.0004** (0.01) -0.0242*** (0.00) -0.0296 (0.18) Yes Yes 10,943 0.403 20.70 46.889 (0.00) 0.000 (0.99)

ACCEPTED MANUSCRIPT Table 6: Differences-in-Differences Regressions Using R&D Intensity as the Outcome Variable Panel A provides states that increased R&D tax credits between 2000-2006 using information provided by Daniel Wilson’s website. Panel B provides differences-in-differences estimates using R&D intensity for firms incorporated in these states based on the year of the tax credit increase, matched on Fama-French 30 industry and size (sales). P-values are based on robust cluster-adjusted standard errors and are provided in parentheses. ***, **, and * correspond to significance at the 1, 5, and 10 percent level, respectively.

T

Tax Credit (%) 15.0 20.0 5.0 7.0 10.0 8.0 10.0 3.0 7.0 3.0 5.0 4.0 5.0 10.0

AN

US

CR

IP

Panel A: States that Increased R&D Tax Credits between 2000 and 2006 State Year CA 2000 HI 2000 ID 2001 IL 2004 IN 2003 LA 2003 MD 2000 NE 2006 OH 2004 SC 2001 SC 2002 TX 2001 TX 2002 VT 2003

ED

M

Panel B: R&D Tax Credit Increases and R&D Intensity Model (1) Full Sample Constant

Post

PT

Tax credit increase

Tax credit increase * Post

AC

CE

Year fixed effects No. Obs. R-squared F-statistic

0.0338*** (0.00) -0.0110 (0.26) 0.0549 (0.16) 0.0268 (0.57) Yes 460 0.068 3.459

49

Model (2) Centrality Tercile 3 0.0357*** (0.00) -0.0170 (0.12) 0.0064 (0.63) 0.0504** (0.05) Yes 161 0.115 2.823

Model (3) Centrality Terciles 1-2 0.0337*** (0.00) -0.0100 (0.46) 0.0914 (0.17) 0.0032 (0.97) Yes 299 0.083 2.167

ACCEPTED MANUSCRIPT Table 7: Bond- and Firm-level Summary Statistics Table 7 provides the summary statistics for 2,769 at-issue bond observations in the primary sample based on 549 unique firm issuers over the sample period 2000-2013. The variable descriptions are provided in the Appendix. Panel A: Bond Issue Characteristics No. Issues Embedded Options Callable 1,847 Putable 7 Credit (Moody) rating: Aaa 33 Aa1-Aa3 233 A1-A3 803 Baa1-Baa3 1,034 Ba1-Ba3 341 B1-B3 282 Caa1-Ca 43

T

IP Median 0.0537 0.0162 10.0137 0.0535 747

Max 0.0661 0.0288 10.0630 0.0660 48,932

Std. Dev.

Min

Median

Max

0.0420 4,634 366

0.0000 0 0

0.0005 0 0

0.6255 53,323 4,108

7.6762 0.7960 0.1201 0.0651 0.0458 0.1476 55,288

1 0.6839 -0.2779 -0.5854 0.0005 0.0000 138

14 1.5588 0.0603 0.0607 0.0216 0.2900 11,067

43 5.8425 0.9320 0.3419 0.7818 1.5210 444,948

ED

M

AN

US

Min 0.0084 -0.0091 2.2630 0.0045 3.308

AC

CE

0.0119 0.0841 0.2900 0.3734 0.1231 0.1018 0.0155

St Dev. 0.0218 0.0177 8.1207 0.0218 2,620

PT

Panel C: Pooled Issuer Characteristics Mean Innovation measures RD Intensity 0.0202 No. Cites (No. obs. = 823) 558 No. Patents (No. obs. = 823) 56 Financial control variables No. estimates 15.2929 Market-book ratio 1.8033 Sales growth 0.0811 ROA 0.0607 Std. ROA 0.0350 Book leverage 0.0317 Sales ($mil) 29,558 No. Obs. 2,769

0.6670 0.0025

CR

Panel B: Pooled Bond Characteristics Mean Yield to maturity 0.0531 Yield spread 0.0216 Time to maturity (years) 11.1580 Coupon rate 0.0526 Proceeds amount ($Mil.) 1,337 No. Obs. 2,769

Proportion of Sample

50

ACCEPTED MANUSCRIPT Table 8: Impact of Innovation on Yield Spread Conditional on the Level of Network Centrality Table 8 presents the results of OLS regressions of yield spread regressed on the three innovation measures and additional control variables, estimated over subsets based on Centrality terciles. All the regressions include year and industry fixed effects, and standard errors are clustered by firm. P-values based robust standard errors are reported in parentheses. *, **, and *** denote significance at the 10-, 5-, and 1-percent levels, respectively.

Log (Number of patents)

Model (3) Centrality Tercile 3

Model (4) Centrality Terciles 1-2

-0.0005* (0.09)

0.0005** (0.02)

Log (Number of citations)

Coupon rate Time to maturity Log (Issue amount) Monthly Baa–Aaa spread Log (No. estimates) Market-book ratio

PT

Sales growth ROA

Book leverage Log (Sale)

AC

Year fixed effects Industry fixed effects No. Obs. R-squared F-statistic

CE

Std. ROA

-0.0006** (0.04) -0.0008*** (0.00) -0.0010* (0.08) 0.0056*** (0.00) 0.7807*** (0.00) -0.0001*** (0.00) 0.0006** (0.04) 1.3742*** (0.00) -0.0015*** (0.00) -0.0007 (0.27) 0.0024 (0.32) -0.0059 (0.18) 0.0336*** (0.00) 0.0043** (0.01) -0.0001 (0.75) Yes Yes 548 0.923 124.5

CR

US

Putable

-0.0004 (0.41) -0.0007** (0.04) -0.0012 (0.31) -0.0023 (0.41) 0.6111*** (0.00) -0.0001** (0.03) 0.0002 (0.73) 0.9940*** (0.00) -0.0010 (0.22) -0.0012* (0.07) -0.0034 (0.40) 0.0115 (0.17) 0.0310** (0.02) 0.0017 (0.72) -0.0004 (0.48) Yes Yes 275 0.874 91.63

AN

Callable

-0.0001 (0.53) -0.0008*** (0.00) 0.0003 (0.30) -0.0020 (0.62) 0.7977*** (0.00) -0.0005*** (0.00) 0.0000 (0.88) 0.7145*** (0.00) -0.0010*** (0.00) -0.0008** (0.01) 0.0027* (0.06) -0.0101*** (0.00) 0.0159*** (0.00) 0.0045*** (0.00) -0.0003 (0.11) Yes Yes 1,830 0.930 412.1

M

Residual bond rating

-0.0004 (0.11) -0.0011*** (0.00) 0.0010** (0.02) 0.0011 (0.25) 0.6823*** (0.00) -0.0004*** (0.00) -0.0000 (0.89) 0.7326*** (0.00) -0.0006 (0.25) -0.0006 (0.22) 0.0015 (0.40) -0.0108 (0.18) 0.0265*** (0.00) 0.0038** (0.04) -0.0008*** (0.00) Yes Yes 939 0.905 120.5

ED

Centrality

51

Model (5) Centrality Tercile 3

T

R&D intensity

Model (2) Centrality Terciles 1-2 0.0085 (0.13)

IP

Model (1) Centrality Tercile 3 -0.0170*** (0.00)

-0.0004** (0.05) -0.0004 (0.48) -0.0008** (0.03) -0.0011 (0.32) -0.0029 (0.32) 0.6112*** (0.00) -0.0001** (0.03) 0.0002 (0.77) 1.0044*** (0.00) -0.0008 (0.31) -0.0014** (0.03) -0.0031 (0.42) 0.0103 (0.25) 0.0291** (0.02) 0.0015 (0.74) -0.0005 (0.39) Yes Yes 275 0.875 114.0

Model (6) Centrality Terciles 1-2

0.0004** (0.04) -0.0006** (0.05) -0.0008*** (0.00) -0.0010* (0.07) 0.0057*** (0.00) 0.7790*** (0.00) -0.0001*** (0.00) 0.0007** (0.03) 1.3752*** (0.00) -0.0015*** (0.00) -0.0007 (0.29) 0.0023 (0.35) -0.0059 (0.19) 0.0337*** (0.00) 0.0043** (0.01) -0.0001 (0.78) Yes Yes 548 0.923 123.5

ACCEPTED MANUSCRIPT Table 9: Firm Network and Bond Yield Spread Changes Surrounding Patent Filing Dates Table 9 presents changes in yield spreads surrounding patent filing dates based on the closest transaction days prior to- and following the filing date. The sample is based on 141 issuers with a total of 360 issues trading within 30 days prior to- and following the filing date. In Panel B, multiple bond yield spreads from the same issuer are weighted using the par amounts of each issue as weights. Panel C provides least squares coefficient estimates using bond-level changes in yield spread as the dependent variable. Yield spreads are winsorized at the 2.5 percent tails. *, ** and *** indicate significance at 10 percent, 5 percent and 1 percent levels, respectively.

IP

T

Min -0.0129 5 1 -0.0136 2 1 0.0740 1 (Aaa) 0.000 -2.1870 451

Log (Bond age)

0.0002 (0.300) 0.0010*** (0.007) 0.0000 (0.917) 0.0012** (0.025) 0.0003 (0.565) Yes Yes 360 0.405 6.328

AC

Bond (Moody’s) rating

CE

Maturity less than 3y

PT

ED

M

AN

US

CR

Panel A: Bond- and Issuer Level Descriptive Statistics Mean Std. Dev. Pre-filing firm-level yield spread 0.0236 0.0228 Pre-filing trading volume ($000) 3,701 5,492 Days between transaction and filing date 10.014 7.873 Post-filing firm-level yield spread 0.0224 0.0215 Post-filing trading volume ($000) 5,119 15,400 Days between filing and transaction date 9.565 8.656 Time to maturity (years) 9.267 8.175 Bond (Moody’s) rating 8.244 3.539 Bond age 3.378 2.256 Centrality 0.9262 1.4946 Sales 20,448 53,121 Panel B: Change in Firm-level (Weighted Average) Yield Spread by Centrality Terciles No. Obs. Pre-Filing Full sample 141 0.0236 (0.0165) Centrality tercile 3 47 0.0191 (0.0136) Centrality terciles 1-2 94 0.0259 (0.0202) Panel C: Regressions of Bond-level Changes in Yield Spread Model (1) Centrality 0.0003 (0.204) Centrality × Maturity less than 5y -0.0009* (0.080) Maturity less than 5y 0.0022*** (0.005) Centrality × Maturity less than 3y

Log (Total trading volume)

Log (Total days surrounding filing date) Log (Sale) Year fixed effects Firm level fixed effects No. Obs. R-squared F-statistic

52

Post-Filing 0.0224 (0.0145) 0.0166 (0.0133) 0.0253 (0.0176)

Median 0.0165 1,526 7 0.0145 1,406 7 6.805 8 (Baa1) 2.256 0.8014 5,308 Change in spread -0.0013* (-0.0013)*** -0.0025*** (-0.0024)*** -0.0006 (-0.0007)** Model (2) 0.0003 (0.137)

-0.0020** (0.013) 0.0011 (0.369) 0.0003 (0.129) 0.0010** (0.010) 0.0001 (0.603) 0.0012** (0.027) 0.0003 (0.495) Yes Yes 360 0.394 5.951

Max 0.1128 280,000 29 0.0997 154,000 30 38.9178 21 (Ca) 18.447 4.5993 345,977 P-value 0.062 (0.000) 0.000 (0.000) 0.507 (0.010)

ACCEPTED MANUSCRIPT Highlights

CE

PT

ED

M

AN

US

CR

IP

T

We study how network connectedness affects innovation and pricing of innovation. We use board interlocks to construct various firm network measures. Different network characteristics affect firm innovation input and output. The type of industry has a bearing on the connectedness-innovation relation. The impact of innovation on bond yield spreads depends on network connectedness.

AC

    

53