Boardroom Centrality and Stock Returns

Transcription

1 Boardroom Centrality and Stock Returns David F. Larcker * Eric C. So Charles C. Y. Wang Draft: May 27, 2010 Abstract: Firms with central boards of directors earn superior risk-adjusted stock returns. We measure centrality as the extent to which a firm s board shares network links to the boards of other companies. Initiating a long position in central firms and a short position in non-central firms earns an average return of 4.68% per year. Consistent with centrality conveying economic benefits, a firm s centrality is positively related to future growth in return-on-assets (ROA) and the extent to which realized earnings exceed analysts earnings forecasts. Return prediction and growth in ROA are most pronounced among financially distressed firms as well as young firms and those with high growth potential. We corroborate and extend these findings by examining the impact of network links to financial institutions. Together, our results suggest that board of director networks signal economic benefits that are not immediately reflected in stock prices. * All three authors are at Stanford University. Larcker (Larcker_David@gsb.stanford.edu) is the James Irvin Miller Professor of Accounting at the Stanford Graduate School of Business (GSB); So (eso@stanford.edu) is a Doctoral Candidate in Accounting at the Stanford GSB; Wang (charles.cy.wang@stanford.edu) is a Doctoral Candidate in the Department of Economics. 1

2 Social and economic networks are a central feature of virtually all economic activities. These networks serve as a conduit for interpersonal and inter-organizational support, influence, and information flow. The links between individuals in these networks are the channels by which information is communicated and this exchange of information produces an environment that provides opportunities for or constraints on individual actions (Wasserman and Faust, 1994, p. 4). Economists and sociologists have studied the influence of social networks on labor markets, political outcomes, information diffusion, and many others. One important network in corporate finance is the social network formed by firms boards of directors. While there has been a large body of work studying the structure of boardroom social networks, why they form, and their theoretical impact on firm performance, questions regarding the economic benefits of these networks have been relatively unstudied empirically. This paper examines the empirical link between boardroom social networks and firm performance. There are at least five ways that director networks can positively affect the economic performance of companies: 1. Strategic Information: Directors possess a wealth of information, on industry trends, market conditions, regulatory changes, and other key market data, which can flow across the boardroom network. Information on the actions of customers, suppliers, and competitors can also flow through the network. Firms centrally positioned in the boardroom network may have better access to this information and a comparative advantage in making strategic decisions. 2. Contracting: Social relationships among directors may improve terms of contracts. This can happen in at least two ways. First, social capital can be leveraged to negotiate better contract terms. For example, two firms may engage in quid-pro-quo behavior reinforced by social relationships that exist between directors. Second, social network among directors can reduce asymmetric information in contracting. If a company s directors are also on the board of suppliers of factor inputs (or closely linked to the directors of these suppliers), they can reduce the asymmetry of information that exists when these firms engage in contracts with each other, which can, for example, reduce input costs. For instance, directors may provide credible information about the riskiness or quality of the firm to directors of a financial institution, thus allowing firms to reduce their costs of obtaining capital. 3. Shared Contacts: Directors possess important and useful business contacts accessible through the boardroom network. These contacts can be sources of useful business relationships (e.g., clients, suppliers) or sources of other economic benefits (e.g., political favors) that can help to improve a firm s overall economic performance. For example, Faccio (2006) and Hillman, Zardkoohi, and Bierman 2

3 (1999) find that investors react positively to news that a company CEO or board member receives a political appointment. 4. Best Management Practices: The boardroom network may be a mechanism through which value-improving business innovations can spread. The success and failure of business practices at one firm may provide valuable information to firms contemplating the adoption of similar measures. For example, firms may learn about effective corporate governance mechanisms, efficiency enhancing technology, and innovative compensation structures through the boardroom network. 5. Collusion: The boardroom network represents a channel of communication between companies and can facilitate collusive competitive behavior and yield economic benefits for a set of closely linked firms. These arguments suggest that a firm s connectedness within the corporate boardroom network can have a substantial positive economic impact on organizational performance. consequences: There are also reasons why director networks may lead to negative economic 1. Value-Decreasing Management Practices: The boardroom network may propagate value-decreasing management practices. For example, the boardroom network has been found as an important explanation for the spread of options backdating (Bizjak, Lemmon, and Whitby (2009), Armstrong and Larcker (2009)), practice whose association with shareholder value is large and negative (Bernile and Jarrell (2009)). 2. Board of Director Effort Tradeoff: The connections among board members require that directors serve on the board of more than a single company. Since more board membership reduces the amount of time that the director can use for monitoring each company, there is a trade off between effort and well connectedness. For example, Fich and Shivdasani (2006) find directors who hold many board positions to be associated with poorer monitoring efforts and a decrease in shareholder wealth. 3. Spread of Bad Information: Networks serve as a conduit for providing all types of information. Although this information can be valuable for decision making, it is equally plausible that the information in the network is incorrect or misleading. At the extreme, misleading information can be put into the network to adversely affect firms decision-making. 4. Collusion: The boardroom network can facilitate collusive behavior. This network effect can produce positive results if the cooperating firms are not discovered by regulators. However, price fixing and other similar behavior can result in substantial regulatory discipline, litigation, and loss of reputation. The net impact of collusive behavior can be value decreasing for shareholders. 3

4 In this paper, we examine the impact of a firm s position in the corporate boardroom network, formed by interlocking boards, on firm operating and stock price performance. 1 Using a comprehensive sample of directors over the time period from 2000 to 2007, we compute the connectedness of a firm s board based its position in the aggregate boardroom network using four standard measures of centrality: degree, closeness, betweenness, and eigenvector centrality. We find that central firms earn substantially higher future excess returns compared to non-central firms. The most central firms (i.e. in the highest centrality quintile) outperform the least central firms (i.e. in the lowest centrality quintile) by an average of 4.68% per year following the portfolio formation; this association holds after controlling for the influence of industry membership, size, book-to-market, and momentum. Thus, the position in the corporate boardroom network has a substantive association with a firm s economic performance. This return difference between central versus non-central firms is most pronounced for young firms and firms with high growth potential (measured using the book-to-market ratio). The average difference in returns across the fifth and first quintiles of board centrality is 6.7% for low book-to-market firms and over 9% for young firms. We also find a pronounced return difference among financially distressed firms. The concentration of returns among young firms, distressed firms, and those with high growth potential are consistent with board networks mattering most for firms with large future growth opportunities or firms confronting serious adverse circumstances. Central firms also outperform non-central firms in terms of growth in operating performance. Specifically, return-on-assets (ROA) of central firms increases by 2% more than those of non-central firms in the year following portfolio formation. In addition, this ROA effect persists into the second and third years following portfolio formation. Consistent with the stock return results, improvements in operating profitability are most pronounced among distressed firms, young firms, and those with high potential for growth. We also document a statistically significant relationship between board centrality and the extent to which the firm s realized earnings exceed the consensus analyst forecast. These results are consistent with the hypothesis that analysts and the market fail to fully appreciate 1 As discussed in Section II, we define a board interlock to exist when two companies share at least one board member. We emphasize also that these types of interlocks are not the same as those that form when CEOs of two companies sit on the boards of their respective companies. 4

5 the economic benefits associated with the centrality of a company s position in the boardroom network. Similar to most empirical corporate finance studies, it is difficult to establish whether the centrality-return relation is causal or reverse-causal. In our setting, high performing companies may become central in a network as a result of their good performance, for example if firms naturally want to link to high performing firms, resulting in a positive association between centrality and firm performance. We conduct two analyses to provide insight into whether our results are consistent with a causal hypothesis. First, we examine the effect of changes in boards centrality on future stock returns. We find that changes in board centrality significantly predict risk-adjusted returns, and these returns are concentrated in firms with the largest increases in board centrality. Second, we find that changes in centrality are positively associated with future stock returns even among firms whose board composition is held constant over time. 2 When a firm s board is unchanged, increases in centrality necessarily reflect changes in other parts of the network, for example the centrality of other firms, and are less likely to support a reversal-causal association between firm centrality and future stock returns. These results are consistent with the interpretation that board centrality leads to higher stock returns. Finally, we provide a case study to provide insight into the one possible sources of higher performance by firms with central boards. Specifically, we examine the possibility of economic benefits accruing to firms with board members linked to large capital-providing financial institutions. The benefits from such links may emerge in the form of more favorable lending terms, a reduction in asymmetric information and search costs associated with finding capital suppliers, and more timely access to capital. We hypothesize those firms with links to suppliers of capital exhibit superior performance. Consistent with more favorable capital access, we find that a firm s network proximity to financial institutions is positively associated with changes in operating profitability and future stock returns. Consistent with our results for the overall board network, we find that the superior returns to firms that share a board member with a financial institution are concentrated among distressed firms and those with high growth potential. 2 We consider these changes to be exogenous in the sense that they are driven by external factors that a company cannot control. When board composition does not change, a company s can become more or less central in a network as a result of the changes in other companies boards, which then affect the network structure. 5

6 The remainder of the paper is organized as follows. Section I reviews related literature and Section II describes the data, the construction of the network and network characteristics. Section III discusses our empirical results. Section IV examines the impact of boardroom links to financial institutions. Section V provides the summary and conclusions from our study. I. Related Literature There is an impressive body of existing literature examining the influence of social networks on economic outcomes. This literature covers a wide range of topics including the diffusion and adoption of innovation (Coleman, Katz and Menzel, 1957), coalition formation (Kapferer, 1969), group problem solving (Bavelas, 1950), elite decision making (e.g., Laumann, Marsden, and Galaskiewicz, 1977), information diffusion in labor markets (Granovetter, 1974), and decisions regarding the level of education that an individual pursues and whether or not to undertake criminal activity (Jackson, 2007). The types of networks examined in these papers include social communities, powerful families and political and economic systems (e.g., Padgett and Ansell (1993) examines business interests and marriage patterns in Florentine families in the 1400 s, and Galaskiewicz (1985) analyzes CEOs and social club networks). Most related to our work, there have also been numerous studies on the social structure produced by members of boards of directors. For example, Levine (1972) documents the existence of interlocked directorates between the boards of major banks and the boards of major industrials. Dooley (1969) observes that an industrial company whose board is occupied by a banker can obtain capital at favorable rates. Much of this prior research documents how particular interlocks are created (Pfeffer and Salancik, 1978), how they are maintained (Palmer, Friedland and Singh, 1986), the density or centrality of the network (Davis, Yoo and Baker, 2003), and the stability of the network though time (Beckman, Haunschild and Phillips, 2004). However, relatively little work documents the economic consequences of these board networks on firm performance. 3 3 The notable exceptions to this summary statement include work that documents the impact of the social network on firms decisions to adopt poison pills (Davis, 1991), switch stock exchanges (Rao, Davis and Ward 2000), make political contributions (Mizruchi, 1992), engage in acquisitions (Haunschild, 1993 and 1994, and Beckman and Haunschild, 2002), and strategic choice (Geletkanycz and Hambrick, 1997). 6

7 The association between a firm s centrality, connectedness, or importance in the interlocking boardroom network is ex-ante theoretically ambiguous. There is an abundance of arguments in organizational sociology and economics on why companies with wellconnected boards may benefit from their position in the network. First, interlocking boardroom networks allow firms to improve the terms of contracts between firms (Schoorman, Bazerman and Atkin, 1981). Second, because directors have important knowledge and contacts, being central in the boardroom network gives a firm better access to such useful knowledge, contacts and resources, which can benefit firms and particularly those that operate in uncertain business environments (Mol, 2001; Nicholson, Alexander, and Kiel, 2004). Third, being central in the interlocking boardroom network could also allow for more or better means of information exchange, leading to a reduction in the costs of obtaining information and perhaps improving business decisions (Mizruchi, 1990; Mol, 2001). Fourth, board connections also represent a mechanism through which valueimproving business innovations can spread in this way, being central in a network can add value to a firm (Haunschild and Beckman, 1998). Finally, interlocking boardroom networks may facilitate collusive competitive behavior, which can yield economic benefits. However, there are also reasons why connectedness in the board network can negatively impact a firm. First, board connections can be a mechanism through which valuedestroying business practices can spread. For example, Snyder, Priem, and Levitas (2009) find that the spread of illegal innovations such as backdating of options is spread through the interlocking boardroom network. Second, to the extent that being well-connected in the boardroom network involves having boardroom members that take on many boardroom jobs, a firm may suffer economically from the deteriorating quality in their directors work in the firm (Fich and Shivdasani, 2006; Fich and White, 2001; Loderer and Peyer, 2002). Third, it is possible that misleading or incorrect information is spread though the board network. If this information is used for strategic choices, it can easily result in a decrease in shareholder value. Finally, although collusion can have a positive impact on shareholder value, the resulting regulatory, litigation, and reputation costs can produce substantial losses of shareholder value. Thus, the impact of a company s position in the boardroom network on shareholder value is essentially an empirical question. Some prior research has delved into similar empirical questions. Boyd (1990) finds that among firms facing a more uncertain business environment, those with more 7

8 connections to other companies through interlocks tend to perform better in terms of sales improvements and return on equity. Using a sample of 350 Brazilian firms, Santos, Silveira, and Barros (2009) find that firm value is negatively affected by interlocking boards, particularly boards with busy boards and firms in which CEOs hold directorships in other companies. Similarly, Non and Frances (2007) find a negative relationship between the number of interlocks and future performance for a sample of 101 Dutch firms. Although these prior studies are useful, they are limited in terms of the coverage and choice of measures for their boardroom network. In our study, we include the entire crosssection of publicly traded and large private firms. Moreover, we consider different ways in which, conceptually, a firm can be important in the interlocking boardroom network. Most of the existing papers in this literature focus on a firm s number of interlocks or the number of outside positions held by the firm s directors. In contrast, we take a broader measurement approach for conceptualizing and empirically measuring a firm s importance in the boardroom network. For example, a firm that does not have many boardroom interlocks may still have better access to important resources and information simply by having boardroom connections to one well-connected firm. In the following section we define the different metrics to capture the different dimensions in which a firm can be central or important in the boardroom network. Our paper is also motivated by an emerging literature in finance demonstrating the influence of personal and social connections on various economic outcomes including CEO compensation and retention (Hwang and Kim (2009), Engelberg, Gao, and Parsons (2009)); information transmission from managers to security analysts (Cohen, Frazzini, and Malloy, 2008); mutual fund holdings and performance (Hong, Kubik, and Stein (2005), Kuhnen (2008), Cohen, Frazzini, and Malloy (2010)); and lending and borrowing patterns (Engelberg, Gao, and Parsons, 2010). Our paper contributes to this literature by aggregating network connections of board members at the firm level and examining whether a firm s centrality contributes to, or detracts from, the firm s operating and security price performance. II. Data and Sample Selection A. Boardroom Network Our sample is derived from multiple data sources. We obtain information on companies board of directors from the Board Member Magazine Director Database 8

9 (hereafter referred to as BoardMag) which contains a comprehensive listing of directors on the boards of all publicly traded companies on the NYSE, NASDAQ, and AMEX, as well as private companies with annual sales exceeding $1 billion. This database is updated annually and reflects the most recent information regarding a firm s board of directors as of the publication date. Using the vintages from 2000 to 2007, we construct an undirected and unweighted interlocking boardroom network. Interlocks are defined as follows. Two companies are linked if they share at least one board member (i.e., their boards are interlocked) Two companies are not linked if they share no board member (i.e., their boards are not interlocked) Panel A of Table I reports annual summary statistics on the interlocking boardroom network we construct. On average we have 6,600 companies (which we refer to as nodes ) and 52,000 directors in a year. From 2000 to 2007 we see a 20% drop in the number of nodes in our network from 7,594 to 6,066. This decline is not attributable to errors arising from data collection or network construction, but is consistent with the 28% decline in the total number of publicly traded companies on the three major stock exchanges over this period. The structure of the network can be described by comparing the size of the largest and second largest components of the network.a component is the subset of the network that is totally connected. That is, any node or firm in a component can reach any other node in the component through links. On average, 72% of all companies are connected in a primary central component this ratio is stable from 2000 to In contrast, the second largest components are trivial in size, representing on average 0.001% of the total companies, compared to the central component. We also find that approximately 24% of the companies in a given year are completely isolated. That is, they do not share any board members with any other firms in our dataset. These firms are typically small publicly traded firms or privately held firms. To further characterize the structure of the interlocking boardroom network, Panel B of Table I reports summary statistics on the characteristics of the central component. The average path length of the central component is the average shortest number of steps separating two firms in the component. On average, there is between 5 to 6 degrees of separation between any two firms in the central component of our interlocking boardroom network the 6 degrees of separation is a widely observed phenomenon documented in a variety of social networks. The diameter of the central component is the longest number of steps separating any two firms in the component; in our network we observe an average 9

10 diameter of 16. Thus, we observe a small worlds phenomenon, as documented in many other social networks (see, for example, Milgram (1969)), the phenomenon that large networks, that is networks involving a large number of nodes, exhibit small diameters and average path lengths relative to what is expected to arrive from a random formed network. Additionally, we find the clustering coefficient to be on average 0.19, meaning that 19% of the time two firms interlocked with the same firm are also interlocked with each other. As a benchmark for this measure, we examine the clustering coefficients that would arise from a network formed at random. We simulate 1000 times a network consisting of the same number of nodes and the same expected number of edges as our boardroom network 4, and then compute for each simulated random network its clustering coefficient; these simulations allows us to construct a confidence interval for size of the clustering coefficient that arises from random networks, allowing us to benchmark our empirically observed clustering coefficients. Table I shows that the corporate boardroom network is much more clustered compared to that which would arise from a random network. For example, in 2007 the upper 95 th percentile of the simulated clustering coefficient is , which is approximately 121 times less than the actual clustering coefficient of This observation is consistent with the findings from other social networks (e.g., see Newman (2003), Grossman (2002), Watts (1999), and Adamic (1999)), suggesting that boardroom interlocks are outcomes of strategic decisions. Finally, Panel C summarizes the degree distribution, i.e., the distribution of the number of first-degree links or interlocks to a firm. On average, a firm in our network is interlocked with 5 other firms, where the average remains relatively stable over time, ranging from a high of 5.49 in 2007 to a low of 5.22 in The average degree for a firm in the central component is 6.2, ranging from a low of 5.8 in 2007 to a high of 6.7 in Figure 1 shows boxplots depicting the annual degree distribution of the boardroom network. Consistent with other observed social networks, and consistent with the network arising as a result of strategic decisions, our boardroom network each year exhibits positive skewness in its degree distribution. In summary, the U.S. interlocking boardroom network exhibit many of the same characteristics as observed in other social networks. Our boardroom network consists of a 4 So that the likelihood of a link between any two nodes is the number of edges divided by the number of nodes. 10

11 primary component accounting for more than 70% of all firms each year, and within this primary component we observe low average path length between any two nodes and high clustering, and finally we find a degree distribution that is heavily right-skewed. B. Centrality Measures and Firm Characteristics For each node/firm in the network, we construct four standard measures of centrality used within network theory. The measures capture four separate dimensions in which nodes can be important in a network. The first measure is DEGREE centrality, which measures how connected a node is, and is defined as the number of first degree links to unique outside boards. Note that since our network links are not weighted, we only count the number of unique interlocks. So, letting "(i,j) denote an indicator that nodes i and j are linked, for a given company i in a network, DEGREE i " %#( i, j) (1) where a higher value of DEGREE indicates that the firm shares board members with more firms. Intuitively, more interlocks provide firms with better access to contacts, information, and favors. The second measure is CLOSENESS, which measures how easily a node reaches other nodes in a network, and it is defined as the inverse of the average distance between the node and any other node. Letting l(i,j) be the number of steps in the shortest path between i and j, then for a node i in a network: CLOSENESS i " n #1 % l( i, j) (2) Intuitively, firms with higher closeness centrality can reach any other firm in the network more easily, and therefore may be able to obtain resources, information, or favors with greater facility. Our third measure of centrality is BETWEENNESS (Freeman (1977)), which measures how important a node is in connecting other nodes to each other. BETWEENNESS measures how well-situated a particular node is in terms of the network j $i j $i 11

12 paths that it lies on. Intuitively, if a node is in many paths that connect companies to each other, then such a node may have informational or relational importance in the network since it is vital in connecting nodes to each other. Letting P i (kj) denote the number of shortest paths between nodes k and j that node i lies on and P(kj) denote the total number of shortest paths between nodes k and j, then for a node i in the network: BETWEENNESS i " P i ( kj) /P( kj) & n #1. (3) j $i:i% { k, j} ( )( n # 2) /2 Finally, we also consider EIGENVALUE (Bonacich, (1972)) centrality, which measures a node s importance in terms of the centrality of its neighbors. In particular, this measure of centrality assumes that the centrality of a node is proportional to the centrality of its neighbors, i.e. "# CENTRALITY i $ % j g ij # CENTRALITY j. (4) where λ is the proportionality factor and g ij = 1 if firms i and j are interlocked. Writing (4) in vector form we quickly see that each firm s centrality can be obtained by the EIGENVECTOR of the matrix g 5. That is, " # EIGENVECTOR $ g# EIGENVECTOR (5) This measure of centrality conceptually differs from the previous measures in that it attempts to capture the notions of power and prestige, by virtue of being connected to other important nodes in the network; as such, this definition is self-referential, because how important a firm is depends on the importance the firm s neighbors. Empirically, larger firms tend to have larger boards, which gives rise to a positive correlation between firm size and board centrality. 6 To separate the effects of size and board centrality on firm performance, we create ranked versions of the centrality measures that attempt to purge the centrality measures of its association with size. Specifically, in June of each year, all firms are ranked into quintiles based on market capitalization (denoted as 5 The (i,j) element of this matrix, known as the adjacency matrix, is g ij, and equals 1 whenever firms i and j are linked or 0 otherwise. 6 Table II demonstrates that firm size is approximately 60% correlated with our four raw centrality measures: DEGREE, CLOSENESS, BETWEENNESS, and EIGENVECTOR. In untabulated results, firm size is over 50% correlated with the number of directors on a given board. 12

13 SIZE). Within each SIZE quintile, firms are sorted into quintiles based on the four centrality measures-- DEGREE, CLOSENESS, BETWEENNESS, and EIGENVECTOR, where highest (lowest) values of centrality assume a value of five (one). The use of quintile ranks reduces the influence of extreme values, adds to the ease of interpretation within the regression results, and creates portfolios useful when predicting returns in the cross-section. It should be emphasized that each of the four centrality measures captures a different aspect of a node s importance in a network. However, it is unclear whether one particular network concept may be economically more meaningful than another. In our analysis, we also consider an aggregate centrality measure meant to capture the overall importance of a node or firm in the network. We define our aggregate centrality measure for each node (hereafter, N-SCORE) by taking the average quintile rank in each of the four aforementioned centrality measures. $ N " SCORE # 1 & Quintile DEGREE i % i 4 & +Quintile BETWEENESS ' i ( ) + Quintile ( CLOSENESS i ) ( ) + Quintile ( EIGENVECTOR i ) N-Score is rounded to the nearest integer and, hence, ranges in value from 1 to 5. For each annual volume of the BoardMag data from 2000 to 2007, we construct the entire boardroom network and compute each of the four centrality measures for every firm. We subsequently merge in data on firm characteristics, returns, and analysts consensus forecast, obtained Compustat, CRSP, and IBES, respectively. Table II contains sample statistics of our final data. In Panel A of Table II we report annual summary statistics of firms network characteristics as well as their size (defined as the log of market capitalization of common stock outstanding) and book-to-market ratio. Panel B of Table II contains the firm s characteristics by industry, where industries are categorized in terms of their two-digit GICS sector. Financial firms and those from the information technology sector make up the largest fraction of our sample. Firms in the materials and utilities industries generally display the highest levels of centrality, suggesting that size and maturity, traits associated with materials and utility firms, are closely linked to our centrality measures. Panel C of Table II contains pooled descriptive statistics of the firm characteristics and centrality measures. BETWEENNESS and EIGENVECTOR both display significant skewness indicating that certain firms play a disproportionate role in linking firms within the boardroom network. Panel D of Table II reports the averages of the time-series correlations among the firm ( & ) & * (6) 13

14 characteristics. Pearson (Spearman) correlations are shown above (below) the diagonal. The Spearman correlations document a high level of correlation among the centrality measures, indicating that the raw centrality measures tend to move together. Panel D also documents a high correlation between firm size and the measures of centrality. We adopt a conservative approach to pair network data with security returns. Specifically, measures of a firm s centrality are paired with returns that begin to accumulate at the beginning of July of the year following the measurement of the board s centrality. For example, a firm s most recent board centrality characteristics as of the BoardMag publication date in 2001 are paired to returns that begin to accumulate in July of This approach ensures that the firm s network characteristics are publicly observable prior to the beginning of the return accumulation period. Firm characteristics are obtained from Compustat using the most recent annual financial statements known prior to the accumulation of returns. We assume a four-month lag between a firm s fiscal year end and the release of the firm s annual financial statements. After merging the board network data with CRSP and Compustat, we eliminate firms with stock prices below $1 and firms without at least 6 months of prior return data in CRSP. Our final sample consists of 29,637 firm-years spanning 2000 through III. Empirical Analysis A. Return Prediction Throughout the analysis, we follow Daniel, Grinblatt, Titman, and Wermers (1997) in calculating characteristic-adjusted returns. In each calendar year, characteristic-adjusted returns are calculated as the difference between a firm s cumulative return and the valueweighted average portfolio of firms matched by size, book-to-market, and momentum, where both returns are measured over identical holding periods. Firms are assigned to characteristic-mimicking portfolios via three nested sorts. Firms are first sorted within size deciles, second within book-to-market quintiles, and third within momentum portfolios, resulting in 125 mimicking portfolios. Daniel, Grinblatt, Titman, and Wermers (1997) and Chan, Dimmock, and Lakonishok (2009) demonstrate that dependent sorts display lower tracking error variances than independently sorted portfolios and that characteristic-adjusted returns produce more statistical power than traditional factor models when evaluating fund 14

15 managers ability to produce superior investment returns. 7 Panel A of Table III presents the pooled regression results from regressing one-year ahead characteristic-adjusted returns on the size-adjusted quintile ranks of the four centrality measures and N-Score. t-statistics are reported in parentheses and are based on standard errors clustered by firm and year to account for cross-sectional and time-series dependence in the residuals (Pedersen, 2010; Gow, Ormazabal, and Taylor, 2010). Industry fixed effects are included throughout. The regression results demonstrate that all four centrality measures are significantly related to future returns. For example, the coefficient on Quintile(Degree) is 0.01 with a t-statistic of 2.83, indicating that the highest quintile of degree outperforms the lowest quintile by approximately five (5 x 0.01) % per year, on average, using characteristic-adjusted returns. Given the robust association between the four standard centrality measures and future returns, it is not surprising that our aggregate centrality measure, N-Score, also predicts future returns. Panel B presents the time-series average return to each value of N- Score, where the returns include one- and two-year-ahead cumulative characteristic-adjusted returns, RET1Y, and RET2Y, respectively. The most central firms (fifth quintile) earn an average of 4.86 percent per year more than the least central firms (first quintile) in one-year ahead characteristic-adjusted returns over Return accumulation persists into the second year following portfolio formation: the most central firms outperform the least central firms on average by 9.11 using two-year-ahead characteristic-adjusted returns. 8 Panel C reports the pooled average returns to N-Score by two-digit GICS industries. The high-low returns are most pronounced among firms in the information technology, industrial, and energy sectors, in descending order. To examine whether the centrality return association documented in Table III is persistent across all years in our sample, Table IV reports annual breakdowns of the one-year (Panel A) and two-year (Panel B) ahead characteristic-adjusted returns for equal-weighted portfolios formed on N-Score quintiles. Panel A shows that the most central firms consistently earn higher one-year ahead characteristic-adjusted returns than the least central firms for all years in our sample, with the exception of 2006, mitigating concerns that return predictability is driven by small subset of firm-years. The average difference between high 7 Lower tracking error indicates that characteristic-matched portfolio explains a larger fraction of the portfolios performance, resulting in a lower standard error of the estimate of the portfolio s abnormal performance. 8 In untabulated results, the use of raw, market-adjusted, and size-adjusted returns results in qualitatively similar inferences. Results available upon request. 15

16 and low N-Score quintile portfolios is 4.68% with annual standard deviation of 5.5%, corresponding to an approximated Sharpe ratio of Panel B of Table IV shows that the most central firms earn higher two-year ahead returns than the least central firms for all years in the sample except The use of two-year ahead returns in general increases the magnitude of the return differential as well as the statistical significance, improving the consistency with which N-Score predicts returns. Having observed the pervasiveness of the centrality-return association across time, we next consider whether this relation is also persistent across different types of firms. Specifically, identifying the characteristics of firms for which this association is particularly strong can provide insights into the underlying economic mechanisms driving the returns associated with board networks. One possibility is that highly central firms have superior capital access, improved monitoring, and better information. Under this view, we expect the benefits of boardroom centrality to be most pronounced among firms with higher needs for capital, guidance, and information. Thus, we hypothesize that return prediction is most pronounced among young firms, firms with high growth potential, and distressed firms. Table V contains one-year ahead characteristic-adjusted returns (RET1Y) when firms are independently sorted into portfolios of N-Score and terciles of additional firm characteristics. Panel A demonstrates that the ability of N-Score to predict returns is concentrated within the two lower terciles of BTM. Interpreting lower values of BTM as an indication of expected future growth or higher real options, these results suggest that network centrality matters more for firms with high growth prospects. Panel B presents similar results when sorting firms into terciles of AGE, defined as the number of months since the firm first appeared in CRSP, prior to the June 30 th portfolio formation. We find the centrality-return association is more concentrated around youngest firms: among the youngest tercile of firms, the average difference in returns between the highest and lowest N-Score portfolios is 9.2% per year, compared to 3.3% for the oldest tercile firms. Panel C contains future returns when firms are also sorted based on their Altman Z-score (ALTMAN). Among the most distressed firms (i.e. the lowest tercile of ALTMAN), the differences in returns between highest and lowest tercile centrality firms is over 11% per year; in contrast, among the least distressed firms this return differential is -2.7%. The concentration of returns among firms with low Altman Scores suggests that highly central firms are better able to weather financial distress compared to non-central firms. Together 16

17 the results in Table V are consistent with network links being a proxy for better capital access in response to timely investment opportunities or if circumstances become dire. B. Changes in Centrality and Returns To further establish the predictive relationship between boardroom centrality and future returns, we examine the effect of changes in boards centrality on future stock returns. We calculate ΔDEGREE, ΔCLOSENESS, ΔBETWEENNESS, and ΔEIGENVECTOR as a firm s current quintile rank minus its quintile rank in the prior year. ΔN-Score is defined analogously for changes in N-Score. All of the change measures are winsorized at -2 and 2, where higher values of all five measures correspond to increases in centrality. Panel A of Table VI contains the results from regressing RET1Y on the changes in centrality. The results demonstrate a significant positive relationship between RET1Y and changes in centrality. The coefficient on ΔN-Score is 0.019, indicating that a one unit increase in ΔN- Score increases the firm s characteristic-adjusted return by 1.9% in the year following portfolio formation. The fact that the change in a firm s centrality predicts returns is consistent with the hypothesis that higher board centrality leads to higher stock returns. An additional econometric concern is whether the centrality-return relation is causal or reverse-causal. It may be the case that companies prefer to form interlocks with firms that they anticipate will outperform in the future, which yields the type of empirical association between centrality and return that we observe. To identify a causal relation, we would ideally like to study the association between exogenous changes in board centrality and future returns. Although we do not have a strict exogenous variable, we identify exogenous changes in board centrality by examining the subset of firms whose board members did not change from the prior year to the current year. Changes in centrality for such firms must necessarily arise from changes in the boards of other companies in the network or from their board members taking on different directorship jobs. Changes in centrality for these firms do not result from endogenous choices by the firms in determining or changing their board composition. Panel B of Table VI contains the results from regressions analogous to Panel A but estimated on the sample of firm years for which the board s membership is identical to the prior year. This sample restriction results in a final sample of 7,534 firm years. The coefficients on all of the change variables are positive, with the coefficients on 17

18 ΔBETWEENNESS and ΔN-Score also being statistically significant. The positive coefficient on ΔN-Score is consistent with firms benefiting from changes in centrality that occur without compositional changes to its own board. We note that all of the coefficients attenuate in magnitude. This can be a result of some endogenity bias in the initial results (Panel A) which is mitigated in the more restrictive tests in Panel B. The results in Table VI are consistent with the hypothesis that board centrality leads to, rather than reflects, higher returns. C. Changes in Profitability Panel A of Table VII reports the results of a regression of the year-to-year change in return on assets (ROA), denoted by ΔROA, on firm centrality measures and firm characteristics. ROA equals the firm s realized net income before extraordinary items scaled by beginning of year total assets, and ΔROA is calculated as the firm s realized FY1 ROA minus the firm s current ROA. 9 ΔROA is fitted to quintiles of the 5 centrality measures as well as MOMEN, defined as the firm s cumulative return over the 12 months prior to the June portfolio formation, BTM, and SIZE. The regression results demonstrate that more central firms experience a larger increase in profitability than non-central firms; the coefficient of on N-Score indicates that the highest (fifth) quintile N-Score firms experience an increase in ROA that is 2% (5 x.004) more than the lowest (first) quintile N- Score firms. Panel B demonstrates that the increase in ROA persists for multiple years following the portfolio formation. Specifically, N-Score not only predicts increases in FY1 ROA, it also predicts increases on ROA from FY1 to FY2, and from FY2 to FY3. Our finding that N-Score predicts increases in operating profitability for multiple years following portfolio formation corroborates the Table III results, which demonstrate that N-Score predicts returns in the second year following portfolio formation. Furthermore, we find in Table VIII that the centrality-profitability relation is concentrated among firms that are low book-to-market, young, or financially distressed, corroborating Table V results. These results demonstrate that the return and operating 9 The use of changes in ROA potentially introduce survivorship bias into our analyses by requiring that a firm be present in the Compustat database in concurrent years. To mitigate this concern, we examine the number of firms that are not included in the Table VII analysis due to missing values of FY1 earnings. The results (untabulated) demonstrate no discernable pattern of sample attrition across quintiles of DEGREE. In total, 1,606 firm-years are eliminated from our analyses due to missing FY1 earnings, with the highest N-Score quintile losing the fewest number of firm-years. 18

19 performance predictions are concentrated among the same classes of firms. To the extent that the return prediction results reflect a reaction to expected increases in profitability, we expect to observe that changes in profitability concentrated among the same characteristics of firms where future returns are concentrated. Panel A of Table VIII finds that the difference in ΔROA between the highest versus lowest quintile N-Score firms is largest for the smallest tercile BTM firms at 3.6%. This differential is monotonically decreasing across BTM terciles, where the average difference is not statistically significant for the highest two terciles. Similarly, the difference in ΔROA between the highest versus lowest quintile N- Score firms is only statistically significant (and positive) for firms in the lowest tercile of AGE (3.8%) and for firms in the lowest tercile of the ALTMAN Z-Score (2.3%). The results for centrality-return and centrality-profitability associations are consistent with the hypothesis that more central firms produce better future operating performance, but security prices are slow to incorporate the economic implications of board centrality. To explore the possibility that information on the association between centrality and firm performance are not fully impounded into prices, we analyze in Table IX the association between analyst consensus forecast errors and firm centrality. Under our above hypothesis, we should expect to see greater (and positive) analyst surprises for highly central firms. The sample for Table IX is constructed by merging our base sample with IBES consensus forecasts of FY1 earnings. Consensus forecasts are measured at the conclusion of June to best capture the market expectation of FY1 earnings at the date on which we form our centrality portfolios. The dependent variable used in the analysis is the consensus forecast error, defined as the firm s actual FY1 earnings minus the consensus forecast, and scaled by total assets per share. The consensus forecast error is regressed on quintiles of centrality measure, as well as firm characteristics MOMEN, BTM, and SIZE. We also include two additional control variables. LOSS is a dummy variable that equals one if the firm experienced a loss in the prior quarter. The inclusion of the LOSS control reflects the fact that analysts tend to over-estimate the reversion properties of earnings for loss firms (Brown, 2001). We include AGE quintile as control variable to control for the fact that there is greater information uncertainty surrounding the profitability of young firms (Zhang, 2006). The results in Table IX demonstrate that consensus forecast errors are positively related to board centrality, suggesting that analysts are slow to incorporate the economic implications of board centrality when forming expectations of one-year ahead earnings. The 19

20 positive and significant coefficients on the centrality measures are consistent with board centrality being positively related to the tendency of firms to have realized earnings that exceed the consensus forecast. This finding suggests that at least part of the predictable returns associated with board centrality is related to incorrect expectations regarding the firm s future profitability. IV. Case Study: Links to Financial Institutions The results in Section III are consistent with the notion that companies or individuals gain economic benefits from their social connections. One interpretation of these results is that central firms, by virtue of having more access to other boardrooms and therefore the business landscape, have better information and therefore make better business decisions. Another possibility may be that central firms have better access to capital which enables them to respond to timely investment opportunities and overcome financial downturns. In this section, we provide a case study on the possibility of economic benefits accruing to firms with board members linked to large capital-providing financial institutions. Access to capital is the lifeblood of modern capital structure and essential to the continuity of a firm s day-to-day operations. Within classical corporate finance models, capital is often assumed to flow in a timely fashion and without search costs. In reality, the acquisition of capital is far more complex. In addition to the agency problems highlighted by traditional corporate finance models, firms must seek out lenders, allow for lenders to conduct due diligence, and work toward terms that are agreeable to both parties all of which pose significant costs. Higher capital acquisition costs may severely limit the range of investment opportunities that a firm pursues. Similarly, delays in accessing capital can pose significant costs to the firm including missed investment opportunities and the prevention of default or bankruptcy. Thus, firms that are able reduce the costs of accessing capital may have considerable benefits that are manifested as improved firm performance. To examine the impact of links to financial institutions on firm performance, we create two new variables to measure a firm s boardroom links to financial institutions. First, FINLINK is a dummy variable that equals one for any firm that shares a board member with a firm in the Compustat Bank Annual File with positive total invested capital [Compustat Item ICAPT], and zero otherwise. Second, FINSTEPS equals the minimum 20

21 number of steps in the boardroom network that a firm must travel to reach a board member from a financial institution. We hypothesize that proximity to a financial institution is beneficial to a firm, and firms with lower values of FINSTEPS should enjoy more economic benefits relative to firms that are distant from financial institutions in the boardroom network. The sample used for our analysis is the subset of the sample used in our main analyses after excluding all firms in the financial sector (i.e. firms with two-digit GICS codes equal to 40). The final sample used for Tables X and XI consists of 23,506 firm-years spanning Table X contains the results from a two-way independent sort of one-year characteristic adjusted returns by FINLINK and terciles of additional firm characteristics. The results demonstrate that firms interlocked with likely capital providers have a significantly positive relationship with one-year ahead characteristic-adjusted returns for low book-to-market and high distressed (i.e. low Altman Score) firms. Both findings are consistent with firms benefiting from links to financial institutions when such firms face high growth opportunities or deteriorating financial health (i.e. firms that are most in need of capital). We extend the results of Table X by examining whether closeness to financial institutions, measured by FINSTEPS, are associated with future stock and firm performance. Table XI reports results from regressing RET1Y and ΔROA on quintiles of FINSTEPS and firm characteristics. In both regressions, the coefficient on FINSTEPS is negative and statistically significant, consistent with firms benefiting from being close to financial institutions within the boardroom network. The negative coefficient on FINSTEPS for the dependent variable of RET1Y is consistent with firms close to financial institutions earning superior risk-adjusted returns. Similarly, the negative coefficient FINSTEPS for the dependent variable of ΔROA indicates that firms close to financial institutions experience significantly larger increases in operating performance. This case study provides insight into one way in which a firm s centrality in the corporate boardroom network may yield economic benefits. Specifically, close proximity to the boards of likely capital providers result in higher future stock and operating performance, particularly for those firms most likely to need timely access to capital such as high growth and highly distressed firms. Although there are other explanations for how 21

22 board centrality may yield economic value, preferential access to capital appears to one important explanatory component V. Conclusion Boardroom networks provide a conduit of support, influence, and information flow that can affect the economic performance of firms in the network. In this paper, we build a comprehensive network of firms linked by boardroom ties, and examine whether the position of a firm in this network explains future firm operating and stock price performance. We find that central firms earn significantly higher future returns than noncentral firms; this association holds after controlling for the influence of industry membership, size, book-to-market, and momentum. The most central firms outperform the least central firms by over 4.5% per year following the portfolio formation and exhibit positive hedge returns in all but one year. The return difference is most pronounced among young firms with high potential growth (as measured by the firm s book-to-market ratio) and financially distressed firms, suggesting that board networks may matter most for firms with large future growth opportunities or firms confronting serious adverse circumstances. We also find that central firms experience significantly higher increases in profitability compared to non-central firms. Specifically, central firms have ROA that increases by 2% more than non-central firms in the year following portfolio formation. We also find a statistically significant relationship between board centrality and the extent to which the firm s realized earnings exceed the consensus analyst forecast. The combination of these results suggests that the analysts fail to incorporate the economic implications of boardroom networks into their forecasts in a timely fashion. To the extent that the consensus analyst forecast is a proxy for market expectations, the positive relation between board centrality and future stock returns appears to stem from expectation errors of firm operating profitability. To mitigate concerns that the return relation between board centrality and future stock returns reflects econometric problems such as correlated omitted variables and reverse causality, we explore the performance relation using changes in board centrality. We find that changes in board centrality significantly predict risk-adjusted returns, where the returns are again concentrated in firms with the largest increases in board centrality. Moreover, we 22

23 find evidence that changes in centrality predict future stock returns even among firms whose board composition is constant across adjacent years. This is consistent with firms benefiting from ( exogenous ) increases in centrality that are unrelated to changes of their own board. Finally, we study boardroom links to financial institutions as a possible channel through which firms may alter the costs associated with capital acquisition. This may emerge as favorable lending terms, a reduction in asymmetric information and search costs associated with finding capital suppliers, and more timely access to capital. Network proximity is measured in terms of the degrees of separation between a firm s board members and the board members of a set of likely debt capital suppliers. We find that a firm s network proximity to financial institutions is associated with increases in profitability and higher future returns. We also find that the superior returns to firms that share a board member with a financial institution are concentrated among distressed firms and those with high growth potential. The results of this case study support the idea that one important mechanism for realizing economic benefits from the boardroom network is through ease of capital access, which may be particularly important for firms that are young or distressed. The results in this paper provide consistent evidence that boardroom networks have an important and positive impact on the economic performance of a firm. These results extend the emerging corporate finance literature that highlights the role of networks for setting executive compensation, conveying private information to analysts, and influence firms access to capital. Network effects appear to be important not only in specific settings or decisions, but they have a more general impact on the economic performance of firms. 23

24 References Adamic, L. A. (1999). The Small World Web. Proceedings of the ECDL. Lecture Notes in Computer Science 1696, Berlin: Springer-Verlag. Armstrong and Larcker (2009) Discussion of The Impact of the options backdating scandal on shareholders and Taxes and the backdating of stock option exercise date, Journal of Accounting and Economics, 47, Bavelas, A. (1950). Communication Patterns in Task-Oriented Groups. Journal of the Acoustical Society of America 22, Beckman, C. M., P. R. Haunschild, and D. J. Phillips (2004). Friends or strangers? Firmspecific uncertainty, market uncertainty, and network partner selection. Organization Science 15, Bernile, G. and G. Jarrell (2009). The Impact of the Options Backdating Scandal on Shareholders. Journal of Accounting and Economics :2-26. Bertrand, M, F. Kramarz, A. Schoar and D. Thesmar Politicians, Firms and the Political Business Cycle: Evidence from France. Working Paper. Bizjak, J., M. Lemmon, and R. Whitby (2009). Option Backdating and Board Interlocks. Reveiew of Financial Studies (11): Bonacich, P. (1972). Factoring and Weighting Approaches to Status Scores and Clique Identification. Journal of Mathematical Sociology 2: Bourdieu, P. (1986). Forms of Capital. Handbook of Theory and Research for the Sociology of Education, J.G. Richardson, ed., Westport, Conn.: Greenwood Press. Boyd, B. (1990). Corporate Linkages and Organizational Environment: A Test of the Resource Dependence Model. Strategic Management Journal 11(6), Brown, L., 2001, A Temporal Analysis of Earnings Surprises: Profits versus Losses, Journal of Accounting Research, 39, 2. Chan, L., S. Dimmock, J. Lakonishok, 2010, Benchmarking Money Manager Performance: Issues and Evidence, Review of Financial Studies, 22. Cohen, L., A. Frazzini, and C. Malloy, 2008, The Small World of Investing: Board Connections and Mutual Fund Managers, Journal of Political Economy, 116, Cohen, L., A. Frazzini, and C. Malloy, 2010, Sell-side School Ties, Forthcoming Journal of Finance. Coleman, J. S. (1988). Social Capital in the Creation of Human Capital. American Journal of Sociology 94 (Supplement: Organization and Intuitions: Sociological and Economic Approaches to the Analysis of Social Structure): S95-S

25 Daniel, K., M. Grinblatt, S. Titman, and R. Wermers (1997). Measuring Mutual Fund Performance with Characteristic-Based Benchmarks. Journal of Finance 52, 3. Davis, G. F., M. Yoo and W. E. Baker (2003). "The Small World of the American Corporate Elite, Strategic Organization 3, Dooley, P. C. (1969). The Interlocking Directorate. American Economic Review 59, Engelberg, J., P. Gao, and C. Parsons, 2010, The Value of Rolodex: CEO Pay and Personal Networks, Working Paper, University of North Carolina at Chapel Hill. Engelberg, J., P. Gao, and C. Parsons, 2010, Friends with Money, Working Paper, University of North Carolina at Chapel Hill. Faccio, M. 2006a. Politically Connected Firms. American Economic Review, 96: It is 2006a. Faccio, M., 2006b. The Characteristics of Politically Connected Firms (October 2006). Working Paper. Fich, E. and Shivdasani, A. (2006). Are Busy Boards Effective Monitors? The Journal of Finance 61(2), Fich, E. and L. White (2003). CEO Compensation and Turnover: The Effects of Mutually Interlocked Boards. Wake Forrest Law Review 38, 3. Fisman, R., D. Fisman, J. Galef and R. Khurana Estimating the Value of Connections to Vice-President Cheney. Manuscript. Freeman, L. C. (1977). A Set of Measures of Centrality Based on Betweenness. Sociometry 40, Galaskiewicz, J. (1985). Social Organization of an Urban Grants Economy. Newbury Park, CA: SAGE Publishing. Goldman, E., J. Rocholl and J. So Do Politically Connected Boards Affect Firm Value. AFA 2007 Chicago Meetings Paper. Granovetter, M. (1974). Getting a Job. Cambridge MA, Harvard University Press. Grossman, J. W. (2002). The Evolution of the Mathematical Research Collaboration Graph. Proceedings of the 33 rd Southeastern Conference on Combinatorics (Congressus Numberantium, Vol. 158). Haunschild and Beckman, 1998, When do interlocks matter?: Alternate sources of information and interlock influence. Administrative Science Quarterly, 43:

26 Hillman, A, A. Zardkoohi and L.Bierman Corporate Political Strategies and Firm Performance: Indications of Firm-Specific Benefits from Personal Service in the U.S. Government. Strategic Management Journal, 20: Hong, H., J. Kubik, and J. Stein, 2004, Social Interaction and Stock Market Performance, Journal of Finance, 59, Hong, H., J. Kubik, and J. Stein, 2005, Thy Neighbor s Portfolio: Word-of-Mouth Effects in the Holdings and Trades of Money Managers, Journal of Finance, 60, Hwang, B.H. and S. Kim, 2009, It pays to have friends, Journal of Financial Economics, 93, Jackson, M., 2007, Social and Economic Networks, Princeton University Press. Kapferer, B. (1969). Norms and the Manipulation of Relationships in a Work Context. In Mitchell, J. C. (Ed.) Social Networks in Urban Settings. Manchester University Press. Laumann, E. O., P. V. Marsden and J. Galaskiewicz (1977). Community-elite influence structures: Extensions of a Network Approach. American Journal of Sociology, 83, Levine, J. H. (1972). The Sphere of Influence. American Sociological Review 37, Loderer, C. and U. Peyer (2002). Board overlap, seat accumulation and share prices. European Financial Management 8(2), Myers, C. and G. Schultz (1951). The Dynamics of a Labor Market, New York: Prentice-Hall, Inc. Mizruchi, M. (1990). What Do Interlocks Do? An Analysis, Critique, and Assessment of Research on Interlocking Directorates. Annual Review of Sociology 8, Mol, M. (2001). Creating wealth through working with others: Interorganizational relationships. The Academy of Management Executive 15, 1, 150. Newman, M. E.J. (2003). The Structure and Foundation of Complex Networks. SIAM Review 45: Nicholson, G., M. Alexander, and G. Kiel, 2004, Defining the Social Capital of the Board of Directors: An Exploratory Study, Journal of Management and Organization, 10, Non, M. and P. H. Franses (2007). Interlocking Boards and Firm Performance: Evidence from a New Panel Database. Working Paper. Padgett, J. F., and C. K. Ansell (1993). Robust action and the rise of the Medici, American Journal of Sociology 98,

27 Palmer, D., R. Friedland, and J. Singh (1986). The Ties that Bind: Organizational and Class Bases of Stability in a Corporate Interlock Network. American Sociological Review 51, Pennings, J. (1980). Interlocking Directorates, San Francisco: Josey-Bass Publishers. Pfeffer, J., and G. R. Salancik (1978). The External Control of Organizations: A Resource Dependence Perspective. New York, Harper and Row Publishers. Rees, Albert and G. Shultz, 1970, Workers and Wages in an Urban Labor Union, Chicago: University of Chicago Press. Santos, R. L., A. M. da Silveira, and L. A. B. C. Barros (2009). Board Interlocking in Brazil: Directors Participation in Multiple Companies and its Effect on Firm Value. CEG Working Paper 003/2009. Schoorman, F.D., M. H. Bazerman, and R. S. Atkin (1981). Interlocking Directorates: A Strategy for Reducing Environmental Uncertainty. Academy of Management Review 6, Snyder, Peter J., Richard L. Priem, and Edward Levitas (2009). The Diffusion of Illegal Innovations Among Management Elites. The Academy of Management Proceedings (2009 Volume). Watts, D. J. (1999). Small Worlds: The Dynamics of Networks between Order and Randomness. Princeton, N.J.: Princeton University Press. Zhang, X. F., 2006, Information Uncertainty and Stock Returns, The Journal of Finance, 61, 1. 27

28 Figure 1 Distribution of Degree Centrality of Board Network The figure below displays the annual distribution of degree centrality for the aggregate boardroom network. The box plots provides the min, max, and several percentiles of the distribution. The box represents the interquartile range, where the line in the center of the box represents the median of the distribution. The endpoints of the whiskers around the box represent the 5 th and 95 th percentiles of the distribution. The network is comprised of all firms covered in the Board Member Magazine Director Database. Degree centrality is defined as the number of first degree links that a firm possesses to unique outside boards. 28