Modigliani and Miller meet Chandler: Organizational Complexity

Transcription

1 Modigliani and Miller meet Chandler: Organizational Complexy and Capal Structure Alberto Manconi * Massimo Massa * Abstract We study how the degree of organizational complexy of a firm relates to s corporate financial policies. We measure complexy as the number of layers in the firm s subsidiary structure, and focus on a sample of US firms over the period We argue that organizational complexy makes the firm opaque and increases the asymmetry of information between and the market. We show that organizational complexy is strongly related to stock market-based measures of information asymmetry i.e., Amihud s (2002) illiquidy, Llorente et al. s (2002) information asymmetry coefficient, the number of analysts tracking the firm, and the equy bid-ask spread. In line wh the predictions of the pecking order theory, firms characterized by a more complex organizational structure resort less to equy and more to debt financing, have higher leverage, display a higher investment-cash flow sensivy and hold more cash to finance future investment. Complexy, while related to information asymmetry, is not just another proxy for the existing measures of asymmetry of information (e.g., Bharath et al., 2008) and s impact on firm financial policies is stronger than that of other conventional proxies for information asymmetry. Given s high persistence over time, complexy is an ideal candidate for the fixed firm-specific effect that has been identified as one of the main determinants of leverage (Kayhan and Tman, 2007, Lemmon et al., 2008). By making the firm more opaque and increasing information asymmetry, complexy reduces the value of equy (Tobin s Q and Rhodes-Kropf et al. s (2005) overvaluation measures). Moreover, by liming the divisional managers incentives to engage in risk taking behavior, complexy reduces the overall riskiness of the firm and s probabily of distress. This improves the value for the bondholders, reducing the cost of debt: more complex firms have lower bond yield spreads on the secondary market. JEL Classification: G34, G30, L25 Keywords: organizational structure; capal structure; pecking order. * Finance department, INSEAD, Boulevard de Constance Fontainebleau, France. alberto.manconi@insead.edu; massimo.massa@insead.edu. Alberto Manconi gratefully acknowledges financial support from Fondazione IRI and Associazione Borsisti Marco Fanno.

2 Introduction Firms are complex organizations often structured as multi-layer hierarchical ( vertical ) structures. At the top, there is the headquarters ( HQ ) and below the different subuns (divisions or subsidiaries). Firms have varying degrees of verticaly : over our sample period , the complexy of US firms varied from 1 to 7 levels of vertical hierarchy. At any one level, there could be as many as 111 distinct subuns. If we consider a firm wh three levels, HQ and two subsequent levels, each wh 30 subuns, simply the need to get all these enties to communicate wh one another would require about 500 direct phone lines, to say nothing of communications involving more than two nodes in the network. This involves a high degree of organizational complexy. The higher the number of layers of the firmsubsidiaries network, the higher the organizational complexy. Organizational complexy makes the firm opaque and difficult to evaluate for the providers of capal, increasing the asymmetry of information between the firm and the market. At the same time, reduces the incentives of the divisional managers to collect difficult-to-transfer information soft information, i.e. information based on direct personal interaction wh the managers of the firm and to engage in risk taking behavior (Stein, 2002). What are the implications for the firm? The goal of this paper is to test how the type of organizational structure of a firm affects s corporate financial policies. We focus on the degree of complexy of the organizational structure of a firm, controlling for whether the firm is a conglomerate or single-segment firm and for the number of industrial sectors in which operates. We relate the complexy of the organizational structure to the way the firm finances self. We rely on the pecking order theory to draw predictions about the effects of organizational complexy on financing choices. A broad definion of the pecking order theory implies that firms characterized by higher information asymmetry should prefer debt over equy financing, have investment more sensive to their internal financing, and hold greater cash reserves to finance future investment. As a measure of firm opacy and information asymmetry, complexy provides an ideal dimension along which to test the pecking order theory. Moreover, helps to address the standard cricism (e.g., Frank and Goyal, 2003) that the pecking order theory seems to better describe the financing choices of large, established firms the very same firms that one would not expect to be characterized by a high degree of information asymmetry. Large and established firms may be more opaque simply 2

3 because they are more complex organizations, regardless of the stabily of their cash flows and the qualy of information on their technology. These observations point to the existence of a relation between organizational structure and capal structure. We test this hypothesis using information on the organizational structure of the US firms over the period We look at the organizational structure, i.e. the different vertical layers in which the firm is structured (divisions or subsidiaries), and use this information to construct a measure of organizational complexy, which we call Complexy. Complexy is defined in terms of the number of layers, from the top down, in which the organization is structured. For example, a structure wh at the top HQ (e.g., global HQ), a regional HQ, and a subsidiary below the regional HQ, has Complexy equal to 3. We start by showing that Complexy is indeed related to (causes) information asymmetry. We base this result on a number of information asymmetry proxies retrieved from the lerature: Amihud s (2002) illiquidy, Llorente et al. s (2002) information asymmetry coefficient, the (log) number of analysts tracking the firm, and the equy bid-ask spread. We show that all of these measures are related to our measure of organizational complexy. This relationship is not only statistically significant, but also economically relevant. A un increase in Complexy (i.e. one addional layer in the organization structure of the firm) leads to an 8% higher Amihud s (2002) illiquidy, a 7% higher Llorente et al. s (2002) information asymmetry coefficient, 14% fewer analysts tracking the firm, and a 21% higher equy bid-ask spread. We then use Complexy to construct a test of the pecking order theory. We start by relating organizational complexy to capal structure. The pecking order theory poss a relationship between capal structure and information asymmetry. The higher the information asymmetry, the less the firm should resort to equy financing, and the more should eher issue debt or rely on retained earnings. We test this hypothesis by employing the Shyam-Sunder and Myers (1999) methodology, as well as directly focusing on leverage. The findings are strongly in favor of the pecking order. Firms characterized by higher Complexy tend to finance their financial defic wh more debt and less equy. For every dollar of debt financing by a high- Complexy firm (Complexy greater than 1), a low-complexy firm (Complexy equal to 1) issues only 0.67 dollars of debt. Conversely, for every dollar of equy financing by a high-complexy firm, a low-complexy firm issues up to 1.67 dollars of equy. These effects survive after controlling for the level of information asymmetry in the spir of Bharath et al. (2008), corroborating our claim that by examining the effects of Complexy we are not simply looking at yet another proxy 3

4 for information asymmetry, but rather at a determinant of information asymmetry based on organizational structure. Organizational complexy affects leverage ratios in two ways. First, the cumulative impact of organizational complexy on secury issuance decisions is reflected in the cross-section of leverage ratios. Firms characterized by higher Complexy have higher leverage, both in market and book value terms. A un increase in Complexy translates into a 9% (5%) higher market (book) leverage. Second, organizational complexy impacts the convergence to target leverage ratios. We show that, after controlling for convergence to a target leverage based on the standard determinants of leverage (Fama and French, 2002, Flannery and Rangan, 2006), leverage ratios adjust to a separate component of target leverage, related to Complexy only. A second implication of pecking order hypothesis is that the investment of a more opaque firm is more sensive to s internal financing. A number of tests of the pecking order theory are relatively silent on this issue, focusing mainly on issuance decisions (e.g. Shyam-Sunder and Myers, 1999, Fama and French, 2005). We look directly into the role of internal financing in the pecking order theory, by looking at how Complexy impacts the sensivy of investment to cash flow. We find that firms characterized by a higher Complexy display a significantly stronger investment-cash flow sensivy. The investment-cash flow sensivy of a high- Complexy firm (Complexy greater than 1) is about four times as large as that of a low-complexy firm (Complexy equal to 1). A third implication of the pecking order theory is that a more opaque firm should store cash in order to finance future investment. A more opaque firm, however, is also subject to greater agency costs, including the agency costs of cash (Jensen, 1986). In order to migate agency costs, a more vertical firm should then optimally decide to hold less cash. We disentangle these two effects by decomposing cash holdings into a component related to future investment and a residual component, imputed to agency costs. Consistent wh the pecking order argument, a un increase in Complexy increases the cash holdings component related to future investment ( pecking order cash) by 0.13%. Consistent wh the agency costs story, on the other hand, a un increase in Complexy leads to a reduction in the residual cash holdings component ( agency cash) by 12%. Overall, these findings support the pecking order hypothesis. It is also important to note that while complexy is related to information asymmetry, is not just another proxy for the existing measures of asymmetry of information (e.g., Bharath et al., 2008). Indeed, the impact of complexy on financing decisions is stronger than 4

5 that of the other conventional proxies for information asymmetry. We argue that complexy is one of the main root determinants of asymmetry. These findings have also another important insight. Given persistence over time, complexy is an ideal candidate for the fixed firm-specific effect that has been identified as one of the main determinants of leverage and helps to explain s strong stickiness (Leary and Roberts, 2007, Kayhan and Tman, 2007, Lemmon et al., 2008). What are the implications of Complexy for firm value? There are two competing effects. On the one hand, by making the firm more opaque and increasing information asymmetry, Complexy should lower the value of equy. This cost of opacy can be rationalized in terms of an illiquidy premium as well as agency costs of cash. On the other hand, by lowering the incentives of the divisional managers to collect difficult-to-transfer information and to engage in risk taking behavior, complexy should reduce the overall riskiness of the firm. This should improve the value for the bondholders and therefore induce a wealth transfer from equy holders to bondholders (e.g., Mansi and Reeb, 2002). We investigate equy value focusing on both the firm s Tobin s Q, as well as s stock mispricing. For the latter, we use the mispricing measures of Rhodes-Kropf et al. (2005). We find that more complex firms have lower value, as the market discounts the costs of opacy. The valuation impact of organizational complexy is economically meaningful: a un increase in Complexy results in a 4% lower Tobin s Q. The fact that this effect is only present in the subsample of firms wh higher agency cash i.e. cash in excess of the firm s financing needs for future investment suggests that the cost of opacy is related to the agency costs of cash. At the same time, we also provide evidence of a posive effect of complexy on the value of debt. The yield spreads on the firm s bonds in the secondary market are lower for more complex firms. The effect is economically relevant: a un increase in Complexy corresponds to an 11% (15 bpts) lower yield spread. This is due to a lower probabily of default. A un increase in Complexy lowers the expected default frequency (EDF) by 19%. This is not trivially due to higher tangibily of the assets nor to diversification related to the setting up of a conglomerate, as we explicly control for these factors in our regression specifications. Throughout the analysis, we deal wh the potential endogeney of the choice of organizational structure wh an instrumental variable approach. We rely on the economics and strategy lerature on the determinants of organizational structure, and construct a number of instruments related to the firm s competive environment and the complexy of the firm s product. Our results survive, and in some cases are strengthened, under instrumental variable estimation. 5

6 These findings indicate that organizational structure is important in affecting firms behavior and financing decisions. They are broadly consistent wh the view that organizational complexy increases the opacy of the firm, as well as wh Stein s (2002) theory of organizations. Our findings contribute to several strands of lerature. First, they relate to the lerature on the theory of organizations. Starting wh Chandler (1962), there is a consolidated body of lerature on the theory of organization and hierarchies whin the firm, both in strategy and economics. The analysis of the type of organizational structure has focused on the relationship between the central administrative un (HQ) and the subordinated uns divisions or subsidiaries. Firm hierarchy is seen as an instrument to help managers to supervise workers (e.g., Calvo and Wellisz, 1979; Rosen, 1982). Different types of structures represent different reactions to the competive environment of the firm (Roberts, 2004) in terms of incentives and information. In particular, the type of organizational structure helps to define the way information is processed and communicated whin the firm and to the HQ (Radner, 1993, Bolton and Dewatripont, 1994, Garicano, 2000). By delegating decision rights, a less hierarchical structure can help to generate higher qualy information (e.g., Aghion and Tirole, 1997, Stein, 2002) and for to be transmted to HQ (e.g., Williamson, 1967). However, also reduces the HQ s abily of control (Williamson, 1967; Calvo and Wellisz, 1978). The type of organizational structure also determines the incentives whin the organization (e.g. Lazear and Rosen, 1981, Rah, 2003, Cuñat and Guadalupe, 2006). A less hierarchical and more decentralized structure provides stronger incentives for the divisional managers, and allows the firm to better employ pay-for-performance sensivy incentives so as to better align incentives wh the firm s objectives (e.g., Prendergast, 2002, Mookerjee, 2006, Wulf, 2007). Therefore, overall, a more complex (hierarchical) structure helps coordination, at the expense of speed and flexibily in reacting to the market challenges (e.g., Alonso et al., 2008). We contribute to this lerature by providing a first link between organizational structure and corporate financial policies. Moreover, we show that organizational complexy may have implications both in terms of transparency and in terms of lower incentives to take risk. Second, we relate to the capal structure theory by providing a direct test of the pecking order hypothesis. In doing this, we relate to the recent contributions showing stickiness in leverage (Kayhan and Tman, 2007, Lemmon et al., 2008). We contribute to this lerature by providing a candidate explanation for the stickiness of 6

7 leverage: persistence in organizational structure. Moreover, we help to address the apparent puzzle that pecking order theory seems to better describe the financing choices of large, established firms the very same firms that do not look like good candidates for high-information asymmetry firms (Frank and Goyal, 2003). We show that this is no puzzle at all, to the extent that these firms tend to have higher complexy, and a more complex organization creates information asymmetry between the firm and investors. Third, our results are related to the lerature on wealth transfer and value of the firm (Mansi and Reeb, 2002). We show that organizational complexy reduces the value of equy, by increasing the discount due to information asymmetry and the scope for agency costs. On the other hand, we document that organizational complexy increases the value of debt, by reducing the scope for risk taking, hence migating the probabily of default. The remainder of the paper is articulated as follows. Section 2 describes the data and the econometric methodology. Section 3 relates complexy to information asymmetry. Section 4 uses complexy to test the pecking order theory. Section 5 provides some insights on the link between complexy and firm value (both equy and debt). A brief conclusion follows. 2.. Data description 2.1 Sample and main variables Our primary data sources are the annual CRSP-Compustat Merged files (containing firm-level accounting data). We obtain information on the organizational structure of each company through Dun & Bradstreet s (D&B) Million Dollar Database. This data source contains information on the identy of each subsidiary/division, s posion in the corporate structure, s number of employees, s SIC code and s location. 1 We focus on all the nonfinancial, non-public utily firms appearing in the merged CRSP/Compustat dataset in the period We then merge these data wh the D&B dataset and retain the observations for which we are able to retrieve information on the subsidiary structure from the D&B dataset. We exclude firms wh book equy below $250,000 or total assets below $500,000. All variables are censored at their 1 st and 99 th percentiles for robustness. Our main variable of interest Complexy is defined as the number of layers in the organization below the HQ. We compute as follows. Starting from a subsidiary 1 D&B also contains some accounting information on subsidiaries, but is to be too sparse for our purposes. 7

8 firm in the D&B data set, we identify s immediate parent firm, then the parent s parent firm, and so on, until we reach an ultimate parent firm which has no parent. For each link to an immediate parent firm, we add 1 to the complexy measure. The resulting degree of complexy is then attributed to the ultimate parent firm. It ranges between 1 and 7. We consider the number of layers as the measure of complexy as we think is the important factor that increases opacy. 2 We consider a number of proxies for information asymmetry retrieved from the lerature: Amihud s (2002) illiquidy, Llorente et al. s (2002) information asymmetry coefficient, firm s coverage by analysts, and the equy bid-ask spread. Amihud s (2002) measure of stock illiquidy is defined as: Illiquid y = 1000 Rt / $ Vt, where R denotes the stock s return, and $V is the dollar volume over the same period as the return. The measure of information asymmetry of Llorente et al. (2002) is the estimate of the coefficient C 2 from the regression: R C C R C R V ε t = t t 1 t 1 + t 1 where R is the daily stock return, and V the corresponding de-trended trading volume. We estimate the above regression firm by firm, year by year (or month by month in the monthly specifications), thus obtaining for each firm a series of estimates of C 2. The analyst coverage is defined as the natural logarhm of 1 plus the number of analyst forecasts of EPS, firm by firm, year by year (or month by month in the monthly specifications). Finally, the bid-ask spread of a firm s stock is defined as the average ratio of the absolute difference between the average bid and ask prices from TAQ, divided by the midpoint between the average bid and ask prices. We compute this quanty firm by firm, year by year (or month by month in the monthly specifications). We look at a number of variables related to the firm s capal structure and financing decisions. We consider both market and book leverage. Market Leverage is defined as long term debt (Compustat data em 9) plus short term debt (em 34) divided by long term debt plus short term debt plus stockholders equy (em 216). 2 In unreported tests, we also consider a horizontal measure of organizational complexy, defined as the maximum number of subuns across all layers of the organization. Given that subuns of the same firm will share knowledge, possibly assets, and culture, we argue that the important factor that increases opacy is the vertical dimension, i.e. the number of layers of the firm-subsidiaries network. Consistent wh this argument, the horizontal measure of complexy turns out to be a noisier measure than our vertical complexy. Indeed, most of our results go through when we replace Complexy by horizontal complexy; both the statistical significance and the economic magnude of the estimated effects, however, are much smaller. Accordingly, we focus here on the vertical complexy measure. 8

9 Book Leverage is defined as the ratio of total liabilies (em 181) to total assets (em 6). Throughout the analysis, we distinguish the impact of organizational complexy from that of diversification (Berger and Ofek, 1995). We do so by controlling for the number of industrial segments reported by the Compustat Segments file for the firm. The definion of the other variables is provided in the Appendix. We present summary statistics for our variables in Table I. Complexy ranges from 1 to 7. Most firms have a low degree of organizational complexy about 80% of our sample firms have Complexy equal to 1; however, firms wh Complexy equal to or greater than 2 account for 73% of the sample in terms or total assets, or 67% in terms of market value of equy. Unreported statistics show that complexy changes very slowly. The probabily of having the same degree of organizational complexy for two consecutive years is above 70%. 2.2 Endogeney Issues One potential issue is the endogeney of the organizational structure. The complexy of a firm is the outcome of a series of decisions. Mostly, they are rooted back in the past e.g., past mergers and acquisions and therefore they are predetermined if not strictly exogenous. However, there is still potential endogeney if the organizational structure is related to unobservable characteristics that are also driving the corporate financial policies (e.g., leverage) we study. To address this issue, we will resort to an instrumental variable estimation approach, in which Complexy is explained in terms of a number of exogenous instruments. We argue that the choice of the organizational structure is affected by the overall objectives of the firm, and shaped by s environment. We rely on Stein s (2002) model of organizations. The model poss that different structures perform differently in terms of generating information about investment projects and allocating capal to these projects. Vertical hierarchies make more difficult to transfer information, and especially information that is hard to codify. Therefore, hierarchical structures are better in the case of hard information, while flat structures are better in the case of soft information. A decentralized approach wh small, single-manager firms is most likely to be attractive when information about projects is soft and cannot be credibly transmted. In contrast, large hierarchies perform better when information can be costlessly hardened and passed along inside the firm (Stein, 2002). Evidence of this has been found in the banking industry, where small banks are better able to collect and act on soft information than large ones (Berger et al., 2005). 9

10 A more hierarchical/complex structure allows a better control of managerial behavior. By making necessary to codify the information passed-on to the superior, a vertical structure may enhance the effectiveness of risk management. At the same time, however, makes the firm less agile and nimble, and less capable to deal wh innovation and competion. The intuion is grounded in the standard strategy lerature showing that the organization of a firm is geared to deal wh the competive challenges of the industry in which operates, as well as the demand environment (Chandler, 1962, Williamson, 1967, Calvo and Wellisz, 1979, Rosen, 1982, Garicano, 2000, Roberts, 2004). In general, a more complex structure helps coordination at the expense of speed and flexibily in reacting to market challenges (e.g., Alonso et al., 2008). This suggests the existence of an optimal degree of complexy dictated by the trade-off between agily and risk management. While we cannot observe what drives the specific optimum for each firm, we can use some environmental and competion variables to explain (instrument) the degree of complexy of the firm, and then relate complexy to observable firm financial policies. In particular, we will use instruments that are based on the characteristics (average Complexy, R&D, prof concentration) of the industry in which the firm operates, as well as some measure of geographical distribution of the firm across the country. Based on these insights from the lerature, we resort to a number of instrumental variables for Complexy, which are related to the firm s competive environment and the need to monor risk taking by managers, i.e. to the trade-off between agily and risk management. We define Prof concentration as a measure of product market competion, equal to the Herfindahl index of concentration of profabily 3 whin the firm s 3-dig SIC code grouping as the firm in a given year. Industry R&D is equal to the total R&D expense ratio 4 of all firms in the same 3-dig SIC code grouping as the firm in a given year. Subsidiary dispersion is the number of distinct US states in which the firm has a subsidiary. Competors complexy is equal to the average Complexy of all the firms operating in the same 3-dig SIC code grouping as the firm in a given year, excluding the firm self. We also include as instruments the interactions between Subsidiary dispersion and the firm s Size (log of Total Assets, Compustat data em 6), as well as the interaction between Prof concentration and the firm s Size. We document the relationship between these variables and Complexy in Table II. To test for the qualy of the instruments, we report the F test statistic for the 3 Profabily is defined as the ratio of Net Income, Compustat data em 172, to Total Assets, Compustat data em 6). 4 The R&D expense ratio is the ratio of the 3-dig SIC grouping s R&D expenses (Compustat data em 46) to the 3-dig SIC grouping s Total Assets (Compustat data em 6) 10

11 joint significance of the exogenous determinants of Complexy. Moreover, in the subsequent analysis we routinely report the Hansen test of overidentification. The results show that the hypothesis of weak instruments (Staiger and Stock, 1997) is rejected wh F test higher than the conventional threshold value of 10. Moreover, the fact that the Hansen s overidentification tests consistently fail to reject the null suggests that the instruments do not seem to affect the dependent variable(s) through a channel different from their impact on Complexy. 3.. Complexy and Information Asymmetry We start by relating Complexy to information asymmetry. We argued that Complexy renders the firm more opaque, thereby increasing information asymmetry between the firm and investors. We test this claim by estimating the following regression: A i,t = α + β C + γ x + ε, (1) i,t where i denotes the firm, t is the time period, A is a measure of information asymmetry, C is our measure of complexy, and x is a set of standard control variables. We use as proxies of information asymmetry: Amihud s (2002) illiquidy, Llorente et al. s (2002) information asymmetry coefficient, the log-number of analysts tracking the firm, and the average equy bid-ask spread, as defined above. The vector x of control variables includes: Size, Tobin s Q, Leverage, Cash Flow, Cash Holdings, Dividend Payout, Number of Segments. A detailed definion of these variables is provided in the Appendix. We estimate equation (1) using two alternative specifications. The first is a pooled estimation based on firm-year observations, wh the relevant information asymmetry measure computed over the year. All the specifications include industry (Fama-French 12 industries) and year fixed effects. All the explanatory variables are lagged one year. Following Petersen (2009), standard errors are clustered around firms. The second estimation is based on the Fama-MacBeth two-step procedure, performed on monthly observations. In this case, the information asymmetry proxy is based on monthly data. Also in this case, we include industry fixed effects. All the explanatory variables are expressed in their beginning-of-the-year values (i.e. their value at the beginning of the year in which the relevant measure of information asymmetry is measured). We report the results in Table III. Panel A reports the pooled estimates, while Panel B reports the Fama-MacBeth estimates. In both panels, the estimates of columns (1)-(4) are based on simple pooled estimation, while the estimates of 11 i,t i, t

12 columns (5)-(8) are based on instrumental variables estimation, as defined above. In the Fama-MacBeth specifications, we report the sum of the Hansen J-statistics for each of the individual cross-sectional regressions. The corresponding p-value is based on the assumption of independence of the cross-sectional J-stats, under which the test statistic has 2 χ distribution wh degrees of freedom equal to the sum of those of the cross-sectional J-stats. The results show a strong posive correlation between the proxies for information asymmetry and Complexy. This holds across the different specifications, both in the case of pooled estimation and in the case of Fama-MacBeth. It also holds in the instrumental variables specifications. The results are not only statistically significant, but economically very relevant. A un increase in Complexy (i.e., one addional layer in the organizational structure) increases Amihud s (2002) illiquidy by 8%, Llorente et al. s (2002) information asymmetry coefficient by 7%, the bid-ask spread by 21%, and reduce the number of analysts by 14%. These findings hold after controlling for standard determinants of information asymmetry, as well as for the number of industrial segments where the firm operates. These results show a clear link between complexy and information asymmetry. The IV results provide some hints on causaly. They show that exogenously driven cross-sectional variations in Complexy increase information asymmetry between the firm and the market. We now use the effect of organizational complexy on information asymmetry to test the implications the pecking order theory. 4.. Complexy and the Pecking Order The previous results show that Complexy is directly related to asymmetry. We now build on these results to use Complexy to test the pecking order theory. We proceed as follows. First, we relate Complexy to the firm s financing choices and leverage. Then, we relate Complexy to the investment-cash flow sensivy. Finally, we relate to the firm s cash holding policy. 4.1 Complexy and Capal Structure We start by looking at the direct implications of organizational complexy for capal structure. The pecking order theory poss that the higher the information asymmetry, the less the firm should resort to equy to finance investment, and the more should eher issue debt or use s retained earnings. Empirically, this implies that firms characterized by higher Complexy should resort more to debt financing, as opposed to issuing equy to cover the Financial Defic. 12

13 We proceed as follows. First, we relate Complexy to the firm s secury issuance decisions, adopting the Shyam-Sunder and Myers (1999) methodology to gauge whether Complexy affects the choice between equy and debt financing. Next, the cumulative impact of financing decisions driven by Complexy will affect the crosssection of capal structure. We test this hypothesis first by directly relating Complexy to leverage, and second by showing how leverage ratios adjust to a component of target leverage directly related to Complexy, and orthogonal to tradional determinants of target leverage ratios Complexy and the Debt/Equy Choice: Shyam-Sunder and Myers (1999) tests We start by focusing on issuances. The Shyam-Sunder and Myers (1999) methodology implies that firms characterized by higher complexy would finance their Financial Defic (DEF) more wh debt and less wh equy. We implement by testing: Secury issue = α + β DEF + ε (2) where the dependent variable is the net issue of a given type of secury, eher debt or equy eher market value or book value. Following Fama and French (2005), we define Net debt issue as the change in total liabilies (Compustat em 181) scaled by total assets (Compustat em 6). We define Financial Defic (DEF), following Fama and French (2005), as: DEF = ( Total assets Total assets ) ( Retained earnings Retained earnings ) It is the difference between the yearly change in total assets (Compustat data em 6) and the yearly change in retained earnings (em 36), scaled by total assets. Again following Fama and French (2005), we define net debt issues dl as the yearly change in total liabilies (Compustat em 181) divided by total assets. We also define Net Market Equy Issues ds M as the product between (i) the spl-adjusted growth in shares outstanding and (ii) the average of the spl-adjusted stock price at the beginning and the end of the fiscal year, scaled by total assets. Finally, we define Net Book Equy Issues ds B as the difference between the yearly change in stockholders equy (Compustat em 216) and the yearly change in retained earnings (em 36), scaled by total assets. We refer to the Appendix for a more detailed definion. We estimate specification (2) separately for firms characterized by high and low degree of Complexy (i.e., Complexy higher or lower than the sample median of 1). We report the results in Table IV, Panel A. In columns (1)-(2), the dependent variable is Net Debt Issues (dl); in columns (3)-(4), the dependent variable is Net Total assets.

14 Market Equy Issues (ds M ); in columns (5)-(6), the dependent variable is Net Book Equy Issues (ds B ). The results show that firms characterized by higher complexy tend to finance their Financial Defic more wh debt and less wh equy. The last row of Panel A reports, for each model specification, the F test statistics for the difference between the coefficients on DEF for high and low Complexy. They always strongly reject the null hypothesis that Complexy does not affect the way the Financial Defic is financed. The difference is also economically significant: for every dollar of Financial Defic that a high-complexy (Complexy greater than 1) firm covers by issuing debt, a low-complexy firm (Complexy less than 1) issues 0.67 dollars of debt only. Conversely, for every dollar of financial defic that a high- Complexy firm covers by issuing equy, a low-complexy firm issues up to 1.67 dollars of equy. These results show that Complexy does indeed affect the financing choice. However, may be the case that we just pick some spurious correlation wh existing measures of information asymmetry/illiquidy, so that our findings would be consistent wh those of, for example, Bharath et al. (2008). To address this issue, we further spl the sample based on the proxies of information asymmetry we employed earlier on (Amihud s (2002) illiquidy ratio, Llorente et al. s (2002) measure of information asymmetry, the (log) number of analysts covering the firm, the average bid-ask spread from TAQ), as well as the average proximy to equy capal and the average proximy to bond capal. We define the proximy to equy capal as the average proximy of the firm to instutional investors holding equy. 5 We define the proximy to debt capal as the average proximy of the firm to instutional investors holding bonds. 6 A firm is characterized by high (low) information asymmetry if the relevant information asymmetry proxy is higher (lower) than the sample median (for the case of log(analysts), a firm is characterized by high information asymmetry when log(analysts) is below the sample median). The results are reported in Panel B. They show that the impact of Complexy on firm financing survives, regardless of the control for alternative measures of 5 We consider the universe of instutional investors tracked by the Thomson Financial 13F data set. The proximy between a firm and an instutional investor is defined as the inverse of the distance between their headquarters, expressed in thousands of kilometers. We compute a weighted average of the proximy of the firm to all instutional investors in the 13F dataset. Each investor s weight is defined as follows. It is the ratio of the investor s size (total market value of s equy holdings) to the total market value of all firms located whin a radius of 1000 Km from the investor s headquarters. 6 We consider the universe of instutional investors tracked by the Lipper EMAXX bond holdings data set. The proximy between a firm and an instutional investor is defined as the inverse of the distance between their headquarters, expressed in thousands of kilometers. We compute a weighted average of the proximy of the firm to all instutional investors in the 13F dataset. Each investor s weight is defined as follows. It is the ratio of the investor s size (total par value of s bond holdings) to the total market value of all firms located whin a radius of 1000 Km from the investor s headquarters. 14

15 information asymmetry. The rows labeled F-stat report the F test statistic for the difference between the coefficients on DEF in Vertical dimension- information asymmetry and Vertical dimension- information asymmetry subsamples (resp. the Vertical dimension- information asymmetry and Vertical dimension- information asymmetry subsamples). Regardless of the level of the tradional measures of information asymmetry, the coefficient on DEF for firms wh high Complexy is significantly higher than for firms wh low Complexy in the debt issue regressions, and lower in the equy issues regressions, across all specifications of Panel B. These results support a link between organizational complexy and financing choices that is separate from the impact of tradional measures of information asymmetry. We now see whether this cumulates over time generating a relationship between leverage ratios and Complexy Complexy and the Cross-Section of Leverage Ratios We showed that firms wh higher Complexy tend to rely more on debt financing. To the extent that the cumulative impact of issuance decisions is reflected in the current cross-section of leverage ratios, we expect that firms characterized by higher Complexy to have higher leverage. We therefore regress leverage on Complexy, along wh a standard set of control variables. In order to make sure that our results are not mechanically driven by market valuations, we use both Market and Book Leverage (as defined in the Appendix). All the regression specifications include industry (Fama-French 12 industries) and year fixed effects; all explanatory variables are lagged one year. Following Petersen (2009), standard errors are clustered around firms. The results are reported in Table V. In Panel A, the dependent variable is Market Leverage, while in Panel B, the dependent variable is Book Leverage. In both panels, columns (1)-(2) are based on simple OLS estimates, while the estimates of columns (3)-(6) are based on instrumental variables estimation, where the instruments for Complexy are the same as in the previous section. To address the issue of persistence in leverage (e.g., Leary and Roberts, 2005, Lemmon et al., 2008), columns (5)-(6) also control for the lagged values of the dependent variable. Finally, the even-numbered columns extend the baseline specification by controlling for market timing effects, using Baker and Wurgler s (2002) External Finance Weighted Average Q (Q EFWA ). The results show a strong posive correlation between leverage and Complexy, providing us wh a second piece of evidence supporting the pecking order theory and the role of organizational complexy in shaping corporate financial policy. Firms 15

16 characterized by higher Complexy display a higher leverage, both in market and book value terms; the impact of Complexy is, moreover, economically relevant: a un increase in Complexy (i.e., one addional layer in the organizational structure) results in a 9% (5%) higher Market (Book) Leverage Complexy and Adjustment to Target Leverage Ratios Recent lerature (e.g., Kayhan and Tman, 2007) provides cross-sectional evidence on the failure of pecking order, finding that capal structures tend to move towards stable targets in the long run. While our sample does not span a long enough period to be able to exactly replicate these tests (we only have ten years of data at our disposal), we will try to address this issue, by estimating a modified model of adjustment to target leverage (e.g., Fama and French, 2002, Flannery and Rangan, 2006, Kayhan and Tman, 2007). The high persistence in Complexy suggests that will concur to determine the target leverage ratio. At the same time, we want to separate s effect from the other tradional determinants of target leverage. We therefore proceed as follows. First, we estimate two components of the target leverage ratio, in two steps. The first component of target leverage is related to all the classic determinants of leverage, excluding Complexy. It is estimated as the predicted value from a first-step regression wh the same specification as in the previous section, and including Size, Tobin s Q, Div. payout, Cash flow, Cash, No. Segments, Q EFWA, industry and year fixed effects (Target I, Lˆ ). The second component of target leverage is related to Complexy only. This is estimated as the predicted value from a regression of the residuals from the first-step leverage regression on Complexy (Target II, We then estimate our partial adjustment model as: L = α + β = α + β L β Lˆ 1 2 ( L L ) β Complexy 3Lˆ + γ x + γ x + ε + ε = Complexy Lˆ ). where the dependent variable is the yearly change eher in Market Leverage or in Book Leverage. L denotes the target leverage, Lˆ s component related to the classic determinants of leverage excluding complexy, and (3) Complexy Lˆ is the component related to Complexy only. x is a vector of addional controls such as the change in Tobin s Q, the change in the Baker-Wurgler external finance-weighted average Q, the financial defic DEF, and the posive part of the financial defic DEF +. In order to account for the fact that Target I and Target II are generated regressors (Murphy and Topel, 1985), we resort to bootstrap standard errors, clustered around firms following Petersen (2009) and Kayhan and Tman (2007). We expect β 3 > 0. 16

17 The results are reported in Table VI. Consistent wh our working hypothesis, the estimated coefficient on Complexy Lˆ is negative, and statistically significant. Therefore, organizational complexy contributes to the determination of target leverage as a separate factor from other tradional determinants of leverage ratios. Even more crically, this finding helps to provide some insight on the relatively fixed firmspecific effect that has been identified in the lerature as one of the main determinants of leverage (Leary and Roberts, 2005, Kayhan and Tman, 2007, Lemmon et al., 2008). 4.2 Complexy and Sensivy to Internal Financing A second implication of pecking order is that the investment of a more opaque firm is more sensive to s internal financing. To the extent that firms characterized by higher organizational complexy follow more closely follow the pecking order, higher Complexy should impact investment and increase the investment-cash flow sensivy. We test this hypothesis by running: Capex, (4) = α + β Cash Flow 1 + γ C-1 Cash Flow 1 + δ C 1 + µ x 1 + ε where the dependent variable is the ratio of capal expense Capex (equal to the ratio of Capal Expenses (Compustat data em 128) to lagged Total Assets (Compustat data em 6)), C is Complexy, and x is a vector of familiar control variables. We argue that γ > 0, i.e. higher Complexy implies a larger investment-cash flow sensivy. The results are reported in Table VII, columns (1)-(2). The estimates of column (2) are based on instrumental variables estimation, wh the instruments for Complexy defined as above. Consistent wh our hypothesis, the coefficient on the Complexy-cash flow interaction term is posive, and statistically significant. In order to better gauge the impact of Complexy on the investment-cash flow sensivy, we separately regress Capex on cash flow (along wh the other controls in (4)) for firms wh high and low Complexy (i.e. above and below 1). The investment-cash flow sensivy is more than four times larger for high-complexy firms than for low-complexy firm (the F-statistic for the difference is 15.65, wh p- value ). This finding provides further support for the pecking order theory, suggesting that more complex firms i.e., firms wh more information asymmetry tend to finance investment wh their internal financial slack. 4.3 Complexy and Cash Holdings 17

18 A third implication of the pecking order theory is that a more opaque should store cash, in order to build financial slack to finance future investment. On the other hand, a more opaque firm is also subject to greater agency costs, including the agency costs of cash (Jensen, 1986). In order to migate agency costs, should therefore decide to hold less cash. This generates a confounding effect for our testing strategy as more opacy i.e., higher Complexy may be related to higher agency problems and therefore create an incentive to migate them by reducing cash holdings. We disentangle the two effects by splting cash holdings into a component related to future investment and a residual component, imputed to agency costs. In particular, we construct the pecking order-related cash (P.O. Cash) and the agencyrelated cash (Ag. Cash). The former is constructed as the predicted value from a regression: Cash * = α + β Capex + ε where Capex* is the average investment (Capex ratio) over years t, t + 1, and t + 2. Ag. Cash, in turn, is obtained as the residuals from this regression. That is, P.O. Cash is effectively the amount of cash that will be needed to finance future investment, while Ag. Cash corresponds to cash holdings in excess of the amount needed for investment. We regress Cash, P.O. Cash and Ag. Cash on Complexy as well as the set of standard control variables employed throughout the paper. The results are reported in Table VII. In columns (3)-(4), the dependent variable is the Cash holdings ratio (cash to total assets as defined in the Appendix),while in columns (5)-(6), the dependent variables are the two components of Cash, related to the pecking order (P.O. Cash, column (5)) and agency costs (Ag. Cash, column (6)). The estimates of columns (3)-(6) are based on instrumental variables estimation. As before, we estimate a pooled specification including industry and year fixed effects, wh standard errors clustered at the firm level (Petersen, 2009). The results show that overall cash is negatively related to firm complexy. However, when we separately identify the P.O. Cash and the Ag. Cash, we do find a posive relationship between complexy and P.O. Cash. This supports our working hypothesis that firms characterized by a higher degree of asymmetry of information wh the market will resort more to retained earnings to finance their investment and therefore will store more cash. Overall, the impact of agency costs of cash seems to dominate: total cash holdings are negatively related to Complexy, and a un increase in Complexy (i.e. one extra layer in the organizational structure) leads to 18

19 a reduction in agency-related cash holdings Ag. Cash by 12%. The component of cash related to future investment, however, is indeed posively related to Complexy: a un increase in Complexy (i.e. one extra layer in the organizational structure) results in an increase in P.O. Cash by 0.13%. While the impact of Complexy on the pecking order-related component of cash holdings is consistent wh the predictions of the pecking order theory, s economic significance is modest. Taken together, these findings would suggest that the main channel of influence of asymmetry of information as suggested by pecking order is not the build-up of cash reserves, but the preference of debt financing over equy financing. However, the fact that we are able to appraise s economic magnude helps us address concerns raised in the lerature that existing tests of the various theories of capal structure fail to account for the predictions of dynamic versions of the pecking order theory (e.g. Fama and French, 2002), under which the firm stores cash today, in order to have greater financial slack for future investment. 5.. Complexy and Firm Value We now investigate the relationship between Complexy and firm value. We argued that complexy increases information asymmetry. We documented in the previous Section that the increased opacy of the firm creates scope for agency costs. At the same time, however, by reducing the incentive to collect information and act upon, Complexy also lowers the degree of risk taking by the firm (Stein, 2002). What are the implications for firm value? The first effect should have a negative impact on equy value, as increases the illiquidy premium on the stock and allows the managers to expropriate the shareholders. The second effect on risk taking, on the other hand, reduces the Jensen and Meckling (1976) asset substution effect, and should therefore benef the bondholders. We will separately consider the impact of Complexy on equy and bond valuations. 5.1 Complexy and Equy Value We start wh the equy value, and relate to Complexy. We consider two measures of firm value. The first is the standard Tobin s Q. The second is the Rhodes-Kropf et al. (2005) measure of mispricing. Mispricing is defined as the residual from the regression of the log-market values on a series of determinants of valuation such as the book value of assets, net income and leverage, by year and industry. We regress eher Tobin s Q or mispricing on Complexy and a standard set of control variables. 19