Financial Accounting Information, Organizational Complexity and Corporate Governance Systems *

Transcription

1 Financial Accounting Information, Organizational Complexity and Corporate Governance Systems * Robert Bushman ** Kenan-Flagler Business School, University of North Carolina - Chapel Hill Qi Chen Fuqua School of Business, Duke University Ellen Engel Abbie Smith Graduate School of Business, The University of Chicago July 2002 Abstract We investigate whether firms exhibit relatively more costly, delegated monitoring when corporate transparency is relatively low. We study how board structure, directors equity incentives, ownership concentration and executive compensation vary with two aspects of corporate transparency: firms financial accounting systems and organizational complexity. We proxy for accounting s governance usefulness with earnings timeliness, the extent to which current earnings incorporate current value-relevant information, and for organization complexity with industry and geographic diversification. We predict that firms substitute costly governance mechanisms to compensate for low earnings timeliness and high levels of organizational complexity. We document evidence generally consistent with these hypotheses. JEL classification: G30, M41, J33 Keywords: Corporate governance, corporate transparency, earnings timeliness, organizational complexity, diversification * Formerly titled The Sensitivity of Corporate Governance Systems to The Timeliness of Accounting Earnings. We have benefited from discussions with and comments from Ray Ball, Sudipta Basu, Bill Beaver, Judy Chevalier, Thomas Hemmer, Bob Kaplan, Randy Kroszner, Darius Palia, Canice Prendergast, Cathy Schrand, Ross Watts, Jerry Zimmerman and accounting workshop participants at CUNY-Baruch, UC-Berkeley, University of Chicago, Harvard Business School, University of Illinois-Chicago, London Business School, University of Minnesota, University of Rochester, University of Texas- Austin, the 1999 Big Ten Faculty Consortium, 1999 Stanford Summer Camp, 1999 Burton Summer Workshop at Columbia, 2000 European Finance Association Annual Meetings and the 2000 AAA Annual Meetings. We are grateful to Hewitt Associates for providing ProxyBase data and to the Graduate School of Business at the University of Chicago for financial support. Finally, we appreciate the research assistance of Xia Chen, Kathleen Fitzgerald, Rebecca Glenn and the data assistance of Darin Clay. ** Corresponding author. Campus Box 3490, McColl Building, Chapel Hill, NC , Phone (919) , address: bushmanr@bschool.unc.edu. 1

2 1. Introduction Public companies in the US employ intricate corporate governance systems to limit selfinterested managerial behavior at the expense of outside investors. This paper investigates how board structure, equity incentives of directors, ownership concentration and executive compensation plans vary with properties of firms information systems and with firms organizational complexity. The theory underlying our study is that firms substitute towards relatively higher use of costly, delegated monitoring mechanisms when corporate transparency is relatively low. While corporate transparency is a broad concept, we focus our analysis on firms primary financial information system, the financial accounting system, and two fundamental aspects of organizational complexity, industry and geographic diversification. 1 Our focus on relations between an expanded set of governance mechanisms and both information systems per se and organizational complexity extends existing research on the benefits of costly monitoring activities. One influential research strand investigates the premise that the scope for moral hazard derives from characteristics of a firm s operating environment and investment opportunity set. The scope for moral hazard presumably increases as firm performance becomes more sensitive to managerial behavior or when it becomes more difficult to monitor such behavior. In a seminal paper, Demsetz and Lehn (1985) conjecture that the scope for moral hazard is greater for managers of firms with volatile operating environments. They document a significant positive cross-sectional relation between stock return volatility and ownership concentration suggesting that ownership concentration increases with the benefits of costly shareholder monitoring. Himmelberg, Hubbard, and Palia (1999) extend Demsetz and Lehn (1985) to include additional firm attributes such as R&D, advertising, and intangible asset intensities. In related research, Smith and Watts (1992) document that firms investment 1 Bushman et al. (2002) develop a framework for corporate transparency that classifies information systems contributing to corporate transparency into three categories corporate reporting, private information acquisition, and information dissemination. Bushman and Smith (2001) posit firm complexity as an important element of transparency. 2

3 opportunity sets, as measured by market-to-book ratios, are associated with benefits to imposing risk on managers via bonus plans and stock option grants. Following this literature, we also take an equilibrium perspective that observed differences in corporate governance structures across firms represent optimal contracting arrangements endogenously determined by firms contracting environments. Our analysis flows naturally from Himmelberg et al. (1999) who document that a significant fraction of the cross-sectional variation in managerial equity ownership is explained by unobserved firm heterogeneity. We posit monitoring technology and organizational complexity as two important characteristics that can be extracted from their unobserved firm heterogeneity and studied independently, while controlling for governance determinants considered in previous research. First, regarding monitoring technology, we explore the possibility that inherent limitations of information systems in generating information relevant for monitoring managerial behavior directly influence governance structure formation. That is, the ability of an information system to measure managerial behavior is an important governance determinant distinct from others, such as the sensitivity of firm performance to managerial behavior. It is conceivable that, holding constant investment opportunities, operating strategies and organizational complexity, inherent differences in the usefulness of a firm s information system would impact the cost-benefit trade-off underlying the existing governance mechanism configuration. Financial accounting systems are a logical starting point for investigating properties of information systems important for addressing moral hazard problems. Audited financial statements produced under Generally Accepted Accounting Principles (GAAP) produce extensive, credible, low cost information that forms the foundation of the firm-specific information set available for addressing agency problems. We theorize that benefits from costly, delegated monitoring activities increase as the ability of a firm s current accounting numbers under GAAP to capture current value relevant information declines. We predict that where the usefulness of financial accounting information is relatively low, firms will substitute towards relatively more costly governance mechanisms to at least partially compensate for 3

4 poor accounting information. This prediction parallels well known results in the capital markets theory literature demonstrating that traders increase costly, private information gathering and processing activities as the precision of accounting disclosures shrinks (e.g., Verrecchia, 1982). It is also consistent with evidence that managerial incentive plans rely relatively more on non-accounting performance measures where financial accounting information is more limited (Bushman, et al., 1996, Ittner, et al., 1997 and Hayes and Schaeffer, 2000), and that ownership concentration across countries is inversely related to the quality of a country s accounting disclosures (LaPorta, et al., 1998). We proxy for the intrinsic governance usefulness of accounting information with earnings timeliness, which, paralleling Ball, Kothari, and Robin (2000) we define as the extent to which current GAAP earnings incorporate current economic income or value-relevant information. While firms can internally choose accounting methods that differ from GAAP, our premise is that our earnings timeliness metric proxies for inherent limitations of accounting measures to capture value relevant information in a timely fashion. We develop several metrics for earnings timeliness based on traditional and reverse regressions of stock prices and changes in earnings. We conjecture that when current accounting numbers fail to capture current value relevant information they are less effective in satisfying governance demands of directors and shareholders. We take extensive efforts to address alternative explanations in an effort to drive the earnings timeliness effect away. In the end, we cannot. We also investigate the relation between organizational complexity and governance structures. While the construct organizational complexity could encompass a wide range of organizational design features, we operationalize it along two dimensions with measures of diversification using segment revenue information to compute Hirfindahl-Hirschman indices measuring with-in firm industry and geographic concentration. Our premise is that organizations structured to compete in multiple industries and/or multiple geographic regions are relatively less transparent than focused firms, and face a range of operational and informational complexities that increase the benefits to costly monitoring activities relative to firms with tighter industry and geographic focus. We predict that as organizational 4

5 complexity increases firms will substitute towards relatively more costly governance mechanism choices. We explore cross-sectional relations between corporate governance systems and both earnings timeliness and organizational complexity of 784 firms in the Fortune Corporate governance choices considered are: 1) aspects of board structure expected to facilitate costly oversight activities 2 ; 2) stockholdings of inside and outside directors; 3) ownership concentration of outside investors; and 4) the proportions of executives contingent pay that are long-term and equity-based. We find that firms with low earnings timeliness have board structures facilitating costly monitoring, higher ownership concentration and greater proportions of contingent pay that are long-term and equity-based, while the equity holdings of directors are not associated with earnings timeliness. Our results also indicate that the benefits to costly monitoring increase with organizational complexity as captured by either industry or geographic diversification. In particular, greater industry and geographic diversification are associated with board structures facilitating costly monitoring and higher ownership concentration. Also, equity incentives of directors are positively associated with industry diversification, and the proportions of contingent pay that are long-term and equity increase in geographic diversification and decrease with industry diversification. These results are conditional on a large set of control variables capturing investment opportunities, operating cycle length, CEO characteristics and regulatory considerations. The remainder of this paper is organized as follows. Section 2 further discusses the role of earnings timeliness and organizational complexity in influencing governance mechanism choices. Section 3 describes our governance variables and develops hypotheses concerning their sensitivity to the timeliness of accounting numbers and organizational complexity. Section 4 describes control variables, sample, and data. Section 5 describes our empirical design and results. Sensitivity analyses are presented in section 6 and section 7 presents a summary and discussion of implications of the paper. 5

6 2. Measuring governance value of accounting numbers and organizational complexity We attempt to directly measure inter-firm differences in both governance value of accounting numbers and organizational complexity. The joint hypotheses motivating our empirical design are that 1) the demand for costly information collection and processing by shareholders and directors is an inverse function of the firms monitoring technology and organizational complexity key elements of corporate transparency, 2) our measures of earnings timeliness and industry and geographic concentration capture key elements of the governance usefulness of accounting systems and organizational complexity and 3) specified governance structures are a response to the strength of the demand for costly information collection and processing by shareholders and directors. Section 2.1 discusses the timeliness of earnings as a measure of accounting s governance value and describes its construction. Section 2.2 further develops our approach to organizational complexity. 2.1 Earnings timeliness as a measure of the governance usefulness of accounting information In this subsection we first motivate our use of earnings timeliness and provide theoretical underpinnings, and then discuss details of the timeliness metric itself. We conjecture that the extent to which current accounting numbers capture the information set underlying current changes in value (i.e., earnings timeliness, as defined in our study) is a fundamental determinant of their governance value to directors and investors. Directors monitor managerial and firm performance, advise and ratify managerial decisions, and provide managerial incentives, and so demand information to help them understand both how and why equity values are changing. Outside investors who monitor firm and managerial performance also demand such information. Stock prices provide information about overall changes in equity value. Accounting systems, by collecting and summarizing the financial effects of 2 Characteristics of board structure that we consider include the percentage of directors that are insiders, board size, the percentage of outside directors who have had executive experience in the same Fama and French (1997) industry grouping as the sample firm, and the average number of other boards on which outside directors serve. 6

7 firms investment, operating, and financing activities, convey information about the underlying sources of changes in equity value. Earnings timeliness measures the extent to which current earnings capture the information set underlying contemporaneous changes in stock price. If stock prices efficiently reflect all information available to market participants, firms with low earnings timeliness by definition have information reflected in price that is not reflected in current earnings. This raises important questions that go to the heart of our hypothesis that earnings timeliness is a determinant of governance choices. If information is impounded in price beyond earnings by informed arbitrage activities, isn t this information also available to the board? And if so, why does the board need costly governance mechanisms to substitute for low earnings timeliness? Why not just use stock price or simply extract the information included in stock price? We draw on economic theory to address these questions in support of our hypotheses. Is the detailed information set reflected in stock price freely available to the board and other outside investors? Grossman and Stiglitz (1980) demonstrate that fully revealing stock prices are incompatible with costly information collection activities of investors. In an equilibrium where costly private information collection and processing activities exist, prices cannot be fully revealing. This implies that boards and others cannot extract the underlying information from price alone. However, one could argue that even if prices are not invertible back to the market s underlying information set, that the board already has direct access to all this information. In this case, the market is simply expending energy to collect and trade on information that is already known by the board, and so earnings timeliness would have no impact on governance as it is irrelevant as a measure of the board s information set. Is this likely to be true on average? The extent to which stock price impounds important information beyond what is known inside the firm is an open question. Research documents a significant relation between changes in stock price and subsequent investment decisions (e.g., Morck. et al., 1990 and Baker, et al., 2001). One potential explanation for this is that stock price communicates new information to a firm s managers that is then incorporated into investment behavior (see Morck, et al., 1990) for a discussion of competing hypotheses). A number of recent theory papers formally derive 7

8 equilibria supporting a strategy directing role of stock price (see Dow and Gorton, 1997, Subramanyan and Titman, 1999, and Dye and Sridhar, 2001). Finally, let s allow that when earnings have low timeliness, stock price is a more important element of boards information sets. Stock price formation is a complex process that has been the subject of countless research studies and speculations by investment professionals. Interpreting the meaning behind stock price movements is not a pedestrian skill. The aggregated nature of information impounded in price potentially limits its governance usefulness (e.g., Paul, 1992). As a result, utilizing stock price as a substitute information source for poor accounting numbers is likely to involve substantial error and to require extensive sophistication and knowledge on the part of board members. Thus, consistent with our hypotheses, costly governance mechanisms supporting greater sophistication in understanding stock price changes may be demanded when earnings timeliness is low. Our timeliness metric is an aggregate of three-firm-specific metrics. The first two metrics are based on firm-specific reverse regressions between annual earnings and contemporaneous stock returns over a period of at least eight years ending in 1994 as follows: EARN t = a 0 + a 1 NEG t + b 1 RET t + b 2 NEG t * RET t + e t (1) EARN t is core earnings of a given firm in year t, defined as earnings before extraordinary items, discontinued operations, and special items, deflated by the beginning of year market value of equity. 3 RET is the 15-month stock return ending 3 months after the end of fiscal year t. NEG is a dummy variable equal to 1 if RET is negative, and 0 otherwise. 4 This specification allows b 1 to capture the speed with which good news reflected in a firm s stock returns is reflected in earnings, while b 1 +b 2 captures the speed with which bad news is reflected in earnings. 3 There are likely relatively high levels of managerial discretion in the timing of recognizing special items relative to the discretion in the timing of core earnings. To the extent that managers ability and incentives to manipulate bottom line earnings vary with corporate governance structures, the use of core earnings is less likely to lead to a violation of our assumption that the timeliness of earnings is exogenous under GAAP. Section 6 further discusses and empirically examines the potential impact of earnings management activities on our earnings timeliness measures. 4 Allowing the intercept and slope to vary with the sign of stock returns is patterned after Basu (1997) and Ball, Kothari, and Robin (2000). For sample firms that do not have any negative stock returns during the estimation 8

9 Our first metric is b 1, which reflects the relative speed (across the sample) with which firms good news is reflected in firms earnings. 5 We expect firms with severe timing problems that delay the recognition within earnings of value-enhancing activities and outcomes to be identified by low values of b 1. Our second metric of the timeliness of earnings is the R 2 from equation (1) which, as observed by Ball, Kothari and Robin (2000), is a decreasing function of the lag with which earnings capture the news reflected in stock returns. Our third metric is the R 2 from equation (2): RET t = a 0 + b 1 EARN t + b 2 EARN t + e t (2) where RET and EARN are defined as before, and EARN t is the change in core earnings from year t-1 to year t, deflated by the market value of equity at the beginning of year t. Equation (2) allows stock prices to vary with both levels and changes in earnings. We expect this measure of the market share of all value relevant information released during the year that is captured by the level and change in annual earnings to be a decreasing function of the lag with which earnings captures changes in equity value. We refer to the R 2 from equation (1) as REV_R2 and to the R 2 from equation (2) as ERC_ R2. Equation (1) allows the estimation of a slope that is expected to be an increasing function of the timeliness of earnings. We refer to this slope (i.e., b 1 in equation (1)) as REV_SLOPE. 6 We develop a composite index for these three individual metrics (REV_SLOPE, REV_R2, and ERC_R2) as our primary metric of the timeliness of earnings. We calculate the percentile rank for each firm in the sample for each of the three metrics. Thus each firm has three percentile values corresponding to its relative value period for model (1) (i.e. NEG =0 for all observations), we drop NEG and NEG*RET from the specification of equation (1). 5 Although it also might be interesting to consider b 2 as a metric, it is not practical to do so because it is not estimatable for a large number of sample firms that have no negative stock returns during the estimation period. 6 In contrast, we do not use the slope from equation (2) because we expect different timing problems to have opposing effects on the slope from equation (2). For example, we expect the smoothing of the recognition of holding gains on assets in place over the lives of the assets to increase the slope in model 2 (analogous to positive effects of the persistence in earnings on ERCs documented in the prior literature). In contrast, we expect distortions in earnings resulting from mismatching of revenues and expenses to decrease the slope in equation (2). 9

10 of each of the three metrics. The composite timeliness metric for a given firm, EARN_TIMELY, is computed for a firm as the average of all three percentile rank values. The use of a composite involving percentile ranks can mitigate potential measurement error in the timeliness metrics (Greene, 2000). 7 We also conduct sensitivity analyses using the first principal component of the three individual timeliness metrics as an aggregate measure of earnings timeliness. We use a summary earnings measure to calculate earnings timeliness as our empirical design makes it practically difficult to utilize a large vector of disaggregated accounting information in our estimations. However, accounting information provides boards and other market participants with a range of disaggregated information with which to monitor the performance of managers. Thus, even if disaggregated accounting information explained 100 percent of the change in equity value, it would not be superfluous to governance as a single stock price measure is not a sufficient statistic for the detailed operational information available from disaggregated accounting information. Finally, our metric for earnings timeliness may also capture information asymmetries between managers and investors which in turn should influence costly monitoring choices. Frankel and Li (2001) measure information asymmetry by the ability of insider trading activity to predict future stock returns, and find that information asymmetry so defined is negatively related to the R 2 from a price on earnings and book value regression, after controlling for analyst following and the extent of voluntary disclosure. Managers having private information does not imply that boards have it, allowing for the possibility of a governance response to low earnings timeliness. 2.2 Measures of organizational complexity To examine links between organizational complexity and firms governance choices, we focus on two aspects of complexity, industry and geographic diversification. Existing research primarily studies industry and geographic diversification separately, rarely incorporating them together as part of 7 The use of the ranks of the timeliness metrics will mitigate measurement error in the metrics only if the rank is determined by 'timeliness' rather than the measurement error in metrics. 10

11 the same research design (exceptions include Bushman et al. (1995), Bodnar et al. (1998) and Duru and Reeb, 2002). We reason that while industry and geographic diversification differ substantially along many dimensions, they share the common characteristic that both organizational features impose significant operational and informational complexity on firms potentially necessitating a governance response. In this section, we first describe the nature of the complexities imposed by these organizational designs, and then discuss existing research that bears on the interpretation of our research design. Multi-industry firms confront the possibilities that internal capital markets are inefficient, that managers less effectively manage diverse lines of business, and that unrelated segments can have conflicting operational styles or corporate cultures. Managers of individual business segments are also shielded from takeover pressure (Cusatis, et al., 1993) and cannot be given powerful equity incentives (Schipper and Smith, 1983 and 1986). Finally, combining diverse operations creates severe information aggregation problems which can result in substantial information asymmetries (e.g., Habib, et al., 1997). While diversified firms in the U.S. must disclose segment data, this information can suffer from problems associated with segment identification, cost allocations, and transfer pricing schemes (e.g., Givoly, et al., 1999). Combining firms together may also induce information asymmetry by suppressing the activities of information intermediaries (Gilson, et al., 2001). Multinational firms face a complex managerial decision-making environment which generates a range of monitoring difficulties. Such firms face cultural and legal diversity across markets and must develop, coordinate, and maintain organizations that span international boundaries. Information complexities arise due to geographic dispersion, cultural differences, multiple currencies, higher auditing costs, differing legal systems, and language differences (Reeb, et al., 1998 and Duru and Reeb, 2002). Such complexities can arise as firms act to arbitrage institutional restrictions such as tax codes and financial restrictions (Bodnar, et al., 1998). For example, firms may employ complex transfer pricing schemes to shift profits to low tax jurisdictions that can complicate efforts by shareholders and board members to understand a firm s foreign operations. 11

12 While we hypothesize that diversification causes costly governance responses, diversified firms may suffer from severe agency problems, implying that perhaps weak governance structures cause diversification (see Denis, et al., 1997 and Anderson, et al., 1998). A recent series of papers document that firms diversified across industry segments tend to have lower values than portfolios of similar focused firms (e.g., Berger and Ofek, 1995, Lamont and Polk, 2002, Lang and Stulz, 1994 and Servaes, 1996). 8 A related literature documents increases in performance following restructuring events such as spinoffs and carve-outs (e.g., Schipper and Smith, 1983 and 1986, Comment and Jarrell, 1995, John and Ofek, 1995, and Daley, et al., 1997). However, the jury is still out on whether diversification causes value losses. Rose and Shepard (1997) document that CEOs of diversified firms are paid more than CEOs of focused firms, and that this wage premium is more consistent with an ability matching story than an entrenchment story. 9 Campa and Kedia (1999) and Graham et al. (2000) argue that the method used to estimate diversification discounts may produce spurious discounts, and Denis et al (1997) and Anderson et al. (1998) find no evidence that existing governance structures are correlated with estimated diversification discounts, and Anderson et al. (1998) find no evidence that failures of internal governance mechanisms are associated with the decision to diversify. Ultimately, if agency problems as manifested in weak governance structures cause diversification, it should work against us finding results consistent with our hypothesis that more extensive costly monitoring is associated with more diversified firms. We use the Compustat Business Industry Segment File to compute revenue-based Hirfindahl- Hirschman indices (HHI) measuring within-firm concentration by industry (IND_CONCENTRATION) and geographic segment (GEOG_CONCENTRATION), computed as the sum of the square of a firms 8 In contrast, existing evidence shows that geographic diversification is associated with higher values (Errunza and Senbet, 1984). 9 Contrary to CEO entrenchment explanations for industry diversification, Rose and Shepard (1997) find that the compensation premium for diversification is invariant to CEO tenure and that incumbents who diversify their firms earn less than newly hired CEOs at already diversified firms. 12

13 sales in a particular segment as a percentage of total firm sales. 10 Higher values of GEOG_CONCENTRATION and IND_CONCENTRATION indicate more geographic and industry concentration, respectively. These measures decrease (nonlinearly) with the number of segments, holding constant variance of segment size, and increase with variance of segment size, holding number of segments constant. Thus a two segment firm with equal segment sales is less concentrated than a two segment firm with unequal segment sales. These measures have an upper bound of 1 (a single segment) and a lower bound of.10 (for a ten segment firm with equal revenue in each segment). While earnings timeliness and industry and geographic concentration are independently associated with corporate governance systems, these transparency proxies are also significantly associated with each other with correlations of 0.15 and 0.11 between earnings timeliness and the industry and geographic concentration measures, respectively. Boards and other market participants likely use disaggregated accounting information in the governance process, suggesting that our earnings timeliness measure may better capture timeliness of the broader accounting information set when firm s operations are more homogeneous in product line and location. The positive correlations between timeliness and concentration are consistent with information loss from aggregation, implying that the concentration measures may also mitigate measurement error in our earnings timeliness metrics. 3. Predictions and description of governance variables This section develops hypotheses concerning the sensitivity of governance structures to the timeliness of earnings and organizational complexity, and describes our governance variables Board structure Fama and Jensen (1983) argue that board effectiveness in monitoring managers depends on the mix of inside and outside directors. They suggest that an optimal board consists of both inside and outside directors: insiders for in-depth knowledge of firm specific activities and competitive 10 Prior research uses various metrics to capture diversification including number of reported segments (Denis, et al., 1997), dummy variables for multi-segments (Anderson, et al., 1998), factor analysis of multiple measures (Duru and Reeb, 2002), entropy measures (Bushman, et al., 1995) and variations of HHI (Rose and Shepard, 1997). 13

14 environment, and outsiders for independence and monitoring skills. They also suggest that effectiveness is enhanced by outsiders with labor market incentives to develop reputations as experts in decision control. Beyond these insights, there exists little formal theory concerning determinants of optimal board composition (see Hermalin and Weisbach (2002) for a recent review of the literature). We consider four aspects of board structure associated with costly monitoring: 1) The percentage of directors who are insiders (%INDIR). 12 A higher proportion of insiders, while potentially undermining board independence, bring insiders in-depth knowledge of firm-specific issues. 13 Insiders also facilitate succession planning by allowing outside directors more opportunity to observe top managers; 2) Board size measured as the total number of directors on the board (#_DIR). Smaller board size may make it easier to coordinate efforts and reduce free-riding problems; 3) Outside director industry expertise computed as the percentage of outsiders with executive experience in the same Fama and French (1997) industry grouping as the sample firm (%_EXPERT). Such directors bring in-depth knowledge about growth opportunities, production functions, risk factors etc. faced by firms in the industry; and 4) Quality of outside directors (i.e. reputation) which, following Shivdisani (1993), we measure with the average number of other boards on which outside directors serve (#OTH_BOARD). 14 We predict that firms with low earnings timeliness will have a higher percentage of insiders on the board, smaller boards, a higher percentage of outside directors with more expertise and higher quality outside directors. While we generally believe that the same directional predictions are valid for firms with higher organizational complexity (i.e., low values for IND_CONCENTRATION and 11 Details concerning data sources and computation of all governance variables are provided in the appendix. 12 Inside directors are officers, retired officers, or relatives of officers of the sample firm. Outside directors are directors who are not officers, retired officers, or relatives of officers of the sample firm. Managers can be required to attend board meetings without being members of the board. To the extent that this occurs, it may weaken the hypothesized relation between the percentage of directors who are insiders and both earnings timeliness and organizational complexity. 13 In an attempt to isolate the positive role of insiders in-depth knowledge of firm specific activities and competitive environment, we control low independence with the number of years the CEO has been on the board and whether the CEO is a founder of the firm. As documented below, both of these variables are positively and significantly related to the percentage of inside directors on the board. 14 A large number of other boards may also indicate that directors are too busy and perhaps of low quality. We rerun all of our analysis eliminating firms where the average number of other boards is greater than 4 and obtain qualitatively similar results. 14

15 GEOG_CONCENTRATION), there is more ambiguity. For example, diversified firms may need bigger boards due to demand for a wider range of industry and international expertise. Also, because our measure of director expertise is computed as the percentage of outsiders with executive experience in the same Fama and French (1997) industry grouping (based on primary SIC code), our measure of board member expertise may be noisy for multi-industry segment firms. We compute a composite board structure variable for each firm (BOARD_STRUCTURE) as the average of the within-sample percentiles of each of the four board characteristics. %INDIR, %_EXPERT, and #OTH_BOARD are sorted in ascending order, while #_DIR is sorted in descending order before computing percentiles because we expect board monitoring to decrease with #_DIR and to increase with the other board characteristics. Hence, we interpret high values of BOARD_STRUCTURE as indicative of board structures that facilitate costly monitoring. 3.2 Equity-based incentives of inside and outside directors Holdings of stock directly link directors financial incentives with those of outside shareholders. We predict that the stockholdings of inside directors will be relatively large in firms with limited accounting information and organizational complexity due to greater benefits of providing inside directors with equity-based incentives to act in the shareholders interests. We also predict that the stockholdings of outside directors will be large to provide the outside directors with powerful financial incentives to engage in costly policing and advising activities for the benefit of the shareholders. To capture directors equity-based incentives we include the average value and percentage of outstanding shares of common stock held by individual inside directors (STKVAL_INDIR and STK%_INDIR) and by individual outside directors (STKVAL_OUTDIR and STK%_OUTDIR). We compute composite variables to capture the strength of equity-based incentives of inside (INDIR_INCENTIVES) and outside directors (OUTDIR_INCENTIVES). Each composite variable represents the average within-sample percentiles of the percentage and value of shares held by the 15

16 corresponding group. In all cases, high values of the composite variables represent relatively large shareholdings. 3.3 Equity-based monitoring incentives of outside shareholders In the spirit of Demsetz and Lehn (1985), we predict that the lower the timeliness of earnings and higher organizational complexity, the higher the concentration of stock ownership by outside shareholders in response to higher benefits of costly monitoring. Our measures of ownership concentration are: 1) The average value of stock held by individual outside investors (STKVAL_SHLDR), computed as total market value of outstanding common stock at the end of fiscal year 1994 minus the value of shares held by officers and directors as a group, divided by the number of common shareholders; 2) The average percentage of stock held by individual outside investors (STK%_SHLDR), computed as the reciprocal of the number of common stockholders as of fiscal year end 1994; 15 3) The percentage of stock held by institutions (%INST); and 3) The percentage of stock held by blockholders owning 5% or more of the firms shares (%BLOCK). We compute a composite variable (SHLDR_CONC) representing the average within-sample percentile of STKVAL_SHLDR, STK%_SHLDR, %INST and %BLOCK. Each of the four individual metrics is sorted in ascending order before computing the percentile values. Hence, high values of SHLDR_CONC represent relatively large average shareholdings by individual outside shareholders. 3.4 Executive compensation We predict that as the timeliness of accounting earnings declines and organizational complexity increases, executive compensation packages will include both a higher proportion of equity-based incentives and a higher proportion of long-term incentives relative to total incentives to better align incentives. A similar prediction is made by Duru and Reeb (2002) relative to industry and geographic diversification. 15 Ideally we would exclude the number of officers and directors from the denominators of STKVAL_SHLDR and STK%_SHLDR. However, the measurement error from not doing so is likely to be small because the number of officers and directors is small relative to the total number of common stockholders. 16

17 To capture the structure of incentive packages of the top five officers we include the percentage of the value of all their incentive plans represented by equity-based plans (EQINC_TOT ) and by longterm plans (LTINC_TOT ). We compute the value of all incentive plans during a year as the combined value of options and restricted stock granted that year as well as long-term performance plan payouts and annual bonus payments during the year. EQINC_TOT is the percentage of total incentives represented by grants of stock options and restricted stock, and LTINC_TOT is the percent of managers total incentive plans represented by grants of options and restricted stock plus any payouts from long-term performance plans (excludes annual bonus payments). For each measure we use the average of the ratio of across the top five officers of a firm in each year , and then average over the four years to minimize distortions associated with non-annual option grant frequencies. We compute a composite variable (EXEC_INCENTIVES) as the average of the percentiles of EQINC_TOT and LTINC_TOT. Both EQINC_TOT and LTINC_TOT are sorted in ascending order before computing percentiles so that high values of EXEC_INCENTIVES represent relatively high importance of long-term and equity-based incentive plans and relatively low importance of short-term bonus plans. 4. Control variables, sample, and data 4.1 Control variables We include a number of variables in our cross-sectional regression models in an effort to control for alternative explanations for variation in governance structures. As discussed in the introduction, it is possible that firms investment opportunity sets or operating cycle lengths actually cause cross-sectional differences in governance configurations, and that our measure of accounting quality is simply correlated with unmeasured aspects of these true causal characteristics. We proxy for investment opportunities and length of operating cycle with the ratio of the market value of equity to the book value of equity (MTB), sales growth (GROWTH_SALE) and the number of years a firm has been public (FIRM_AGE). In sensitivity analysis, we also consider the ratios of R&D to sales, advertising to sales and tangible assets to total assets as additional proxies for the firms investment opportunity set. 17

18 We include two variables motivated by Demsetz and Lehn (1985) - the standard deviation of stock returns (STD_RET) and the market value of equity (MV). Demsetz and Lehn (1985) document a significant relation between these factors and ownership concentration. We expect a positive association between our governance structure variables and STD_RET consistent with greater volatility increasing demand for costly monitoring. We include MV to control for the impact of various unspecified differences relating to size (e.g., information environment, marginal product of manager effort). Prior research documents that board composition and managerial ownership depend on past performance (see Hermalin and Weisbach, 2002, Himmelberg, eta l., 1999 and Kole, 1996). We use prior period return on equity (ROE) to proxy for performance history. Finally, we consider two variables relating the influence of the CEO the number of years the CEO has been a director of the firm (CEO_TENURE) and an indicator variable for whether the CEO is a founder of the firm (see Finkelstein and Hambrick, 1989). With the exception of STD_RET, FIRM_AGE and FOUNDER, all of the control variables are measured as the average over the period of 1989 to FIRM_AGE is measured by the number of years a firm has been on CRSP as of year-end 1994 and STD_RET is estimated over the same period used to estimate models (1) and (2) in section 2.2. FOUNDER is a dummy variable that takes a value of one if the CEO or Chairman of the Board in place during 1994 is also a founder of the firm and zero otherwise. Finally, we follow the related governance literature (e.g., Demsetz and Lehn, 1985, Denis, et al., 1997, Anderson, et al., 1998) and include dummy variable controls for utilities and banks to consider the role of regulation in the formation of governance systems. Regulation may impact the governance structure of a firm due to additional third-party monitoring like regulatory bank audits which may systematically influence optimal governance structures for these firms. It is possible that a low relative earnings timeliness value could simply capture situations where alternative sources of public information vary across firms. As a sensitivity check, we thus consider two variables to capture alternative sources of public information: #ANALF, the number of analyst long-term earnings growth rate forecasts for the firm from the Zacks database, and #MGTF, the number of management forecasts for the firm from the First Call database. 18

19 4.2 Sample and data Our sample is selected from the Fortune 1000 firms included on ProxyBase, a database maintained by Hewitt Associates of compensation and board of directors information from annual proxy statements of firms. We obtain data for fiscal year 1994 from ProxyBase on the number of directors, the number of inside and outside directors, and directors' stock ownership. We also obtain from ProxyBase data for 1994 through 1997 on the structure of the compensation packages of the top five officers. We obtain the percentage of stock held by officers and directors as a group (as reported in proxy statements for fiscal year 1994) from Compact Disclosure. We collect data from proxy statements for fiscal 1994 concerning the backgrounds of outside directors, including the number of other boards on which they currently serve, and current and prior employers. We get data from Compustat on 4-digit SIC codes of sample firms and current / prior employers of outside directors to assess whether outside directors have industry-specific expertise from serving as an executive of another firm in a given sample firm's industry. For this analysis, we assign firms to industries on the basis of the classification scheme reported in Fama and French (1997). 16 We require sample firms to have data on annual earnings and the market value of equity on Compustat and monthly stock return data on the CRSP database for at least eight years during the period 1985 through 1994 to allow computation of the firm-specific timeliness variables. All other financial data are from Compustat. Table 1 describes the industry membership of the 784 sample firms with complete data assigned on the basis of 4-digit SIC codes and the Fama and French (1997) industry classification scheme. 45 industries are represented by sample firms, with 26 industries represented by at least ten firms. Utilities (97 firms), banks (81 firms), insurance (45 firms), retail (42 firms), chemicals (33 firms), wholesale (33 firms), computers (31 firms), machinery (30 firms), petroleum and natural gas (29 firms), business supplies (28 firms), and business services (25 firms) are each represented by at least 25 firms. 16 We make one change to the Fama and French (1997) industry classification scheme; we classify SIC code 7372 (Prepackaged Software) as Computers rather than Business Services. 19

20 Table 2 includes a summary of the sample distribution of our metrics for the timeliness of earnings and organizational complexity. The mean and median levels of the ERC_R2 are 0.37 and 0.35, respectively, while the mean and median levels of REV_R2 are 0.33 and 0.29, respectively. REV_SLOPE, the slope on positive returns in equation (2), has mean (median) value of 0.04 (0.03). Mean values of IND_CONCENTRATION and GEOG_CONCENTRATION, are 0.81 and 0.65, respectively, suggesting greater geographic concentration than industry concentration. Table 2 also summarizes the sample distribution of other firm characteristics considered in our governance estimations, including the market value of equity (MV), the standard deviation of 15-month stock returns (STD_RET), firm age (FIRM_AGE), the market-to-book ratio (MTB), sales growth (GROWTH_SALE), return on equity of prior years (ROE), CEO tenure and CEO founder information (FOUNDER). Finally, Table 2 summarizes the sample distribution of all our governance variables. Table 2 reveals that all model variables display a fair amount of variation across sample firms. 5. Empirical design and results This section describes the design of our empirical tests and the results. We first examine correlations between our composite metric of earnings timeliness with its underlying variables to validate that the composite is capturing the data of interest. We then present our empirical model of the relation between governance variables and the metrics for the timeliness of earnings and organizational complexity. 5.1 Correlations of individual and composite timeliness variables We first examine Pearson and Spearman rank correlations between pairs of individual earnings timeliness metrics, and between individual metrics and the composite timeliness metric (EARN_TIMELY) constructed from the sample percentiles of the individual metrics. Correlations between all three individual timeliness metrics (not tabulated) are positive and highly significant. The Spearman correlation between the ERC_R 2 and REV_R 2 is the highest of the three at 0.41, and the 20

21 correlation between REV_R 2 and REV_SLOPE is the lowest of the three at The composite metric, EARN_TIMELY, is highly positively correlated with each of the three individual metrics ranging from 0.72 with REV_SLOPE to 0.80 with ERC_R 2. Pearson correlations are generally similar to the Spearman correlations in both magnitude and significance. Taken together, these correlations suggest that our individual metrics for the timeliness of the information in the accounting system are linked. Further, the significant correlations between the composite and individual metrics suggest that our use of sample percentiles of individual metrics to compute the composite metric appears to be capturing the essence of the individual metrics. 5.2 Primary model and regression results We estimate the following cross-sectional regression model: DEP_VAR =α + β 1 EARN_TIMELY +β 2 GEOG_CONCENTRATION+β 3 IND_CONCENTRATION + δ 1 MV + δ 2 STD_RET + δ 3 FIRM_AGE +δ 4 MTB + δ 5 GROWTH_SALE + δ 6 ROE + δ 7 CEO_TENURE +δ 8 FOUNDER +δ 9 UTILITY +δ 10 FINANCIAL + ε. (3) DEP_VAR represents a composite variable for the board structure, equity-based incentives of inside or outside directors, equity-based incentives of outside shareholders, or the composition of executive compensation plans. For each DEP_VAR variable whose value is bounded between zero and one, we apply a logistic transformation before estimating the model. 18 The appendix includes detailed descriptions of all variables. Our main interest is in the coefficients on the earnings timeliness and organizational complexity metrics. The composite governance variables are constructed so that high values represent governance systems that facilitate costly monitoring activities by directors and investors, and that rely more heavily 17 We expect the measures, ERC_R 2 and REV_R 2, to be correlated since the regressions they relate to involve similar variables. The R 2 values are not identical however, since an additional independent variable is include in each equation (i.e., the change in earnings is included in the ERC_R 2 model and a dummy variable to distinguish negative returns is included in the REV_R 2 model). 18 DEP _ VAR Specifically, we use the formula log to transform each composite metric. The transformation is (1 DEP _ VAR) designed to convert the bounded distribution into an unbounded one. The transformation is similar to that conducted by Demsetz and Lehn (1985) and Himmelberg, Hubbard, and Palia (1999) in a similar setting. The reported results for the transformed variables are similar to results using the variables before conducting the transformation. 21