Board monitoring and earnings management: Do outside directors influence abnormal accruals? *

Transcription

1 Board monoring and earnings management: Do outside directors influence abnormal accruals? * K.V. Peasnell, P.F. Pope and S. Young Lancaster Universy This version: October 2000 Key Words: Corporate governance; boards of directors; abnormal accruals; earnings management. JEL Classification: M41, G34 * We gratefully acknowledge the helpful comments of S.P. Kothari, Steve Lim, Gilad Livne, Scott Richardson, Lakshmanan Shivakumar, Judy Tsui, Jerald Zimmerman, an anonymous reviewer and seminar participants at Bristol, U.C. Dublin, Dundee, Glasgow, Lancaster, London Business School, Universy of Science and Technology, Hong Kong, Feng Chia Universy, Taiwan, the Scottish Instute for Research in Investment and Finance, the 1998 Annual Meeting of the American Accounting Association, and the 2000 Annual Meeting of the European Finance Association. Financial support was provided by the Research Board of the Instute of Chartered Accountants in England and Wales, The Leverhulme Trust, and the Economic and Social Research Council. Correspondence to: Professor Peter F. Pope, I.C.R.A., The Management School, Lancaster Universy, Lancaster, LA1 4YX, U.K., Tel: (+44) , Fax: (+44) , p.pope@lancaster.ac.uk

2 Board monoring and earnings management: Do outside directors influence abnormal accruals? Abstract This paper examines whether the incidence of earnings management in the UK depends on board monoring. We focus on two aspects of board monoring: the role of outside board members and the aud commtee. Results indicate that the likelihood of managers making income-increasing abnormal accruals to avoid reporting both losses and earnings reductions is negatively related to the proportion of outsiders on the board. Further tests indicate that this association is confined to firms where the separation of corporate ownership and decision control is greatest. In contrast, we find ltle evidence that outside directors influence income-decreasing abnormal accruals when pre-managed earnings are high. While we are unable to find evidence that the existence of an aud commtee directly influences the level of earnings management (eher income-increasing or income-decreasing), tests indicate that the monoring role of outside directors in relation to income-increasing earnings management is more pronounced in cases where an aud commtee exists. Our findings suggest that boards contribute towards the integry of financial statements, as predicted by agency theory.

3 Board monoring and earnings management: Do outside directors influence abnormal accruals? 1. Introduction Boards of directors are widely believed to play an important role in corporate governance, particularly in monoring top management (Fama and Jensen, 1983). Despe a lack of distinction in law between the responsibilies of inside directors and outside directors, the governance debate emphasizes the distinctive contribution that outside directors can make in helping to ensure that managers act in the interests of outside stockholders (Fama, 1980; Fama and Jensen, 1983). Prior research indicates that outside directors influence a wide range of board decisions, including CEO removal (Weisbach, 1988), negotiation of tender offer bids (Byrd and Hickman, 1992), and resistance to greenmail payments (Kosnik, 1987). In this paper, we test whether the influence of outside directors extends to the financial reporting process. Specifically, we examine whether the incidence of earnings management depends on board monoring. Existing studies that examine the relation between board monoring and financial reporting focus on blatant violations of Generally Accepted Accounting Principles (GAAP). For example, Beasley (1996) reports that the incidence of financial statement fraud is lower for firms where the proportion of outside directors is relatively high. Dechow et al. (1996) report similar findings for firms subject to SEC accounting enforcement actions. These studies provide evidence of a link between violations of accounting standards and board structure. However, whether this link extends to more 1

4 subtle accruals-based earnings management permissible whin GAAP remains an unresolved issue, which we address in this paper. 1 Earnings management may take the form of eher income-increasing or incomedecreasing accounting choices. Opportunies for such manipulations arise because of flexibily permted by GAAP, and because is costly to require and enforce less flexible financial reporting rules (Dye, 1988; Evans and Sridhar, 1996). While managers and stockholders may, on occasion, agree that earnings management is desirable (e.g. to avoid polical or regulatory costs), stockholders will face potential agency costs if the interests of managers and stockholders diverge and managers have opportunies for earnings management. Such conflicts of interests will most obviously arise when managers are evaluated on the basis of accounting numbers. The purpose of this paper is to investigate whether boards that are active monors take actions that reduce the incidence of earnings management when the incentives for such manipulations are high. Our tests focus on suations where performance is eher poor (in which case the incentive will be to manage earnings upwards), or exceptionally good (in which case the incentive will be to manage earnings downwards). We use abnormal working capal accruals to proxy for earnings management. We focus on two dimensions of board monoring: the proportion of outside directors and the presence of an aud commtee. Our tests are conducted using UK data. The UK provides 1 We focus exclusively on the subset of earnings management mechanisms related to working capal accruals. Our tests preclude other possible methods of earnings manipulation such as depreciation, asset wre-downs, asset sales, and long term provisioning. Thus, while we use the terms abnormal accruals and earnings management interchangeably throughout the 2

5 an interesting experimental setting for investigating these issues since there is greater variation in outside director representation on boards in the UK than in the US (Peasnell et al., 1999). Furthermore, aud commtees, while commonplace, are not mandatory and do not exist for a significant minory of firms. Our results show that when pre-managed earnings are negative or below last year s reported earnings, abnormal accruals are less posive if the proportion of outsiders on the board is relatively high. Since abnormal accruals are found to be negatively associated wh changes in future accounting performance, we are able to rule out a signaling interpretation of this result. Our findings are consistent wh the prediction that outside directors contribute towards the integry of financial statements. Interpreting the evidence in a slightly different manner, appears that boards wh a relatively high proportion of outside directors have more confidence to tell as is and report losses and earnings reductions. However, boards appear only to intervene in the case of income-increasing earnings management: we find ltle evidence that outside directors influence the amount of income-decreasing earnings management when the incentives to manipulate earnings downwards are strong. Moreover, wh respect to our income-increasing accruals results, addional tests reveal that outsider directors only intervene in the earnings management process when conflicts of interest between stockholders and managers are likely to be high, i.e. when ownership is divorced from control. In contrast to the findings for board composion, we find ltle evidence that the presence of an aud commtee directly influences the level earnings management (eher income-increasing or income-decreasing). However, we do find remainder of the paper, we are mindful that our definion of earnings management is restrictive. We discuss this issue further in section 3. 3

6 evidence that the monoring role of outside directors in relation to income-increasing earnings management is more pronounced in cases where an aud commtee exists. The next section reviews the lerature on the monoring role of boards of directors and develops our hypotheses. Section 3 presents the methods used to identify earnings management and discusses the research design. Section 4 presents the empirical results. Section 5 reports the findings of several sensivy checks. Our conclusions appear in Section Hypothesis Development 2.1. Governance and earnings management Schipper (1989) defines earnings management as purposeful intervention in the external financial reporting process, wh a view to obtaining private gain for stockholders or managers. Stockholders will gain when earnings management is used to signal managers private information (Healy and Palepu, 1995) or to reduce polical costs (Watts and Zimmerman, 1986). 2 However, stockholders may lose when earnings management results in abnormal private gains to managers. Such gains could take the form of increased compensation (Healy, 1985; Holthausen et al., 1995) or reduced likelihood of dismissal when performance is low (Weisbach, 1988). In this paper we 2 Stockholders may also gain when earnings management is used to avoid violation of debt covenants (Watts and Zimmerman, 1986). Bondholders interests are sometimes represented at board level in suations where this problem is most acute. Conflicts between bondholders and stockholders, while important, are not the focus of this study. Henceforth, we intend the term stockholders to be interpreted broadly, to encompass the interests of both stockholders and bondholders. 4

7 focus on the abily of stockholders to constrain costly earnings management through corporate governance mechanisms. Incentives and opportunies for earnings management occur for at least two reasons. They occur because current stockholders engage in costly contracting wh management under condions of uncertainty and asymmetric information. Theory suggests that, in equilibrium, management contracts will perm some accounting discretion (or flexibily) because of the costs and potential adverse risk-sharing consequences of enforcing inflexible accounting rules (Dye, 1988; Evans and Sridhar, 1996). But theory also suggests that, in equilibrium, earnings management opportunies will be subject to constraints (Dye, 1988, p.198). A second motivation for earnings management is the commment to supply information to the capal markets that can create conflicts between managers and (current and potential) investors. Theory suggests that there may be signal jamming in equilibrium, again subject to constraints (Stein, 1989). The main constraint on earnings management is GAAP, supported by the external aud requirement and internal monoring. 3 As the apex of the internal governance system in large corporations, the board of directors is particularly important in the monoring role (Fama and Jensen, 1983). Therefore, condional on earnings management being costly to stockholders, we predict that effective board monoring will be associated wh lower levels of earnings management. Two dimensions of board monoring that have received widespread attention in the lerature are the proportion of board members classified as outside directors (e.g., 3 Addional constraints include income recognion rules and monoring provisions required by specific contracts and regulations. 5

8 Vicknair et al., 1993) and whether the board has an aud commtee (e.g., Pincus et al., 1989; Levt, 1998). The governance lerature emphasizes the role of outside directors in resolving agency problems between managers and stockholders through designing contracts and monoring management behavior. Fama (1980) and Fama and Jensen (1983) argue that outside directors have incentives to be effective monors in order to maintain the value of their reputational capal. Existing empirical evidence generally supports the prediction that board effectiveness in protecting stockholders wealth is a posive function of the proportion of outsiders on the board (Weisbach, 1988; Rosenstein and Wyatt, 1990; Byrd and Hickman, 1992; Brickley et al., 1994). 4 The effectiveness of the board s monoring activies might also depend on how the board is structured and organized. Boards often delegate work on important tasks to standing commtees reporting directly to the main board (Klein, 1998). The board commtee responsible for financial reporting is the aud commtee, if one exists. The aud commtee has specific responsibily for the production of financial statements, and usually for communicating wh the external audor, who also plays a role in monoring financial reporting. Aud commtees are not mandatory in the UK, although listed firms 4 Board effectiveness in monoring of financial reports will depend on the abily of outside directors to understand earnings management methods. Although the level of accounting expertise, and hence monoring effectiveness, will vary across boards, there are at least two reasons for having confidence in outside directors general abily in this regard. First, outside directors often have a financial background. For example, Peasnell et al. (1999) report that over a quarter of all U.K. board members are professionally qualified accountants. Second, outside directors frequently hold senior management posions in other large corporations (Fama and Jensen, 1983). As such, they are likely to be familiar wh financial reporting from a senior management perspective. 6

9 are encouraged to form them. However, the creation of an aud commtee does not absolve the full board from s financial reporting responsibilies. The full board may also have other reasons to consider financial reporting issues because accounting numbers may figure prominently in such matters as management compensation, operating results and investment decisions. The incremental monoring contribution of an aud commtee is an empirical issue on which this study aims to shed light. The preceding analysis takes the perspective that earnings management is undesirable because is costly to stockholders. An alternative signaling scenario needs to be considered. Earnings management will benef stockholders if managers use accounting discretion to signal private information about future performance. Subramanyam (1996) reports evidence that discretionary accruals predict future profabily and dividend changes, suggesting that managers may indeed use accounting discretion to improve the informativeness of earnings. In such circumstances, boards will not be expected to constrain earnings management. Only if a signaling explanation can be rejected, therefore, is appropriate to consider the monoring role of the board of directors Earnings management relative to thresholds Our empirical tests examine the association between board monoring and the incidence of earnings management. Recent evidence suggests that the incidence of earnings management is particularly pronounced when earnings fall below certain thresholds. For example, Burgstahler and Dichev (1997) and Degeorge et al. (1999) find that there is a higher-than-expected frequency of firms wh slightly posive reported 7

10 earnings (and earnings changes) and a lower-than-expected frequency of firms wh slightly negative reported earnings (and earnings changes). The reasons for thresholdbeating earnings management behavior remain unclear. One possibily is that capal market participants implic contracts wh management are defined in terms of these simple thresholds. 5 For example, stockholders may be triggered to invest addional resources in active monoring when losses and earnings declines are reported, wh significant knock-on costs for management in the form of reduced compensation and an increased probabily of dismissal. Consistent wh this view, extant research reveals that firms subject to shareholder activism tend to be characterized by poor (earnings) performance (Karpoff, 1998). Whatever the underlying reason, however, Dechow et al. (2000) report that attempts to avoid reporting losses and earnings declines appear to be driven by managers desires to opportunistically delay reporting poor performance rather than by any attempt to signal that future performance will improve. Thus, if board monoring is associated wh earnings management, we expect that our abily to detect such a relation will be greatest close to threshold points. Consequently, we condion our empirical tests on the proximy of pre-managed earnings to zero earnings and zero change in earnings. We predict that board monoring will constrain income-increasing earnings management when pre-managed earnings undershoot these thresholds. 6 5 Burgstahler and Dichev (1997) suggest that stakeholders may use zero levels and changes in earnings as heuristic cutoffs to economize on the costs of retrieving and processing firm-specific information. 6 The big bath hypothesis predicts income-decreasing earnings management when pre-managed earnings fall well below the threshold. We examine the issue of big bath accrual choices in section

11 The preceding discussion focuses on the link between board monoring and the incidence of income-increasing earnings management. Earnings management, however, is not restricted solely to income-increasing behavior. Prior research indicates that under certain condions, managers may prefer to engage in income-decreasing earnings management. For example, Degeorge et al. (1999) find that managers appear to systematically manipulate reported earnings downwards when pre-managed earnings exceed threshold earnings by a substantial amount. Similarly, Healy (1985), Gaver et al. (1995) and Holthausen et al. (1995) report evidence of income-decreasing accounting choices when managers accounting-based bonuses are at their maximum. Several explanations for these income-decreasing accounting choices have been suggested. One possibily is that managers prefer to shift abnormal posive earnings forward in order to make future thresholds easier to attain. Another possibily is that managers are reluctant to report large gains because of fears that their performance target will be ratcheted up in the future. If downward manipulation imposes significant costs on external parties, boards should be as concerned wh income-decreasing manipulations as they are wh income-increasing earnings management. Therefore, we also test whether board monoring is negatively associated wh income-decreasing earnings management when pre-managed earnings exceed these thresholds by a large margin. While our empirical tests explicly examine both upward and downward manipulations, two factors suggest that the association between board monoring and income-increasing accruals is likely to be more pronounced than the corresponding association between boards and income-decreasing earnings management. The focal points of zero levels and changes in earnings are bright in the sense that they can be 9

12 unambiguously identified by the researcher. In contrast, the focal point triggering downward manipulations is more ambiguous due to cross-sectional variation in firmspecific factors (e.g., the existence and definion of the upper bound in accounting-based bonus contracts). 7 Our inabily to identify the precise points at which downward manipulations will be triggered lowers the power of tests designed to assess the association between board monoring and income-decreasing earnings management. Moreover, we anticipate that boards will have asymmetric loss functions wh respect to earnings management because the penalties associated wh overstating earnings (e.g., ligation risks) are likely to exceed the costs of understating earnings. The incentives for boards to monor income-increasing earnings management may therefore be greater than those to monor income-decreasing earnings management. Whether this is the case remains an empirical issue on which we aim to shed light in this paper The intervening effect of managerial ownership The importance of corporate governance and the need for board monoring depends on the extent to which managers interests diverge from those of stockholders and other investors. We expect stock ownership by managers to lead to a closer alignment of interests. Prior research has found that managerial ownership is associated wh lower levels of earnings management (Dhaliwal et al., 1982). Also, there is evidence of a negative association between managerial stock ownership and both the proportion of 7 UK firms were not required to disclose details of their accounting-based bonus plans before January Even after this date, no explic requirement exists for firms to disclose the actual parameters of a plan (e.g., whether bonuses are subject to an upper bound and if so, how such a bound is defined). 10

13 outside board members (Weisbach, 1988), and the presence of an aud commtee (Pincus et al., 1989). These results suggest that the demand for board monoring declines wh managerial ownership, and that managerial ownership interacts wh board monoring of earnings management. More precisely, we hypothesize that the constraining association between earnings management and (i) the proportion of outside directors and (ii) the existence of an aud commtee will be more pronounced when the level of managerial stock ownership is low. 8 We conjecture that this holds for both income-increasing and income-decreasing earnings management. 3. Methodology 3.1. Measuring earnings management Several earnings management instruments are available to managers, ranging from real decisions such as asset sales (Bartov, 1993) and changes in R&D expendure (Bushee, 1998), to pure financial reporting decisions such as accounting method changes (Watts and Zimmerman, 1986) and accruals choices (Healy, 1985). Manipulation of operating accruals is likely to be a favored instrument for opportunistic earnings management because they have no direct cash flow consequences 9 and are relatively 8 An alternative prediction for below-threshold firms is that high ownership levels could increase the incentives for managers to manipulate earnings upwards, because owner-managers face greater wealth effects if failure to beat the threshold results in a downward revision in the firm s stock price. However, prior research suggests that the incentive-alignment effects of managerial ownership tend to dominate where earnings management behavior is concerned. 9 Subject to any tax effects. These are likely to be of second order importance in the UK because financial reporting and tax accounting are separate. For example, for tax purposes some working capal accruals, such as doubtful debts, are constrained by the tax authories. 11

14 difficult to detect. Consequently, we use abnormal accruals as our proxy for earnings management. 10 Several methods have been proposed in the lerature for decomposing operating accruals into abnormal (or discretionary) and normal (or non-discretionary) components. 11 The most frequently used methods are the Jones (1991) model and the modified-jones model (Dechow et al., 1995). Recent evidence suggests that estimates of managed accruals obtained using these models may reflect confounding effects of factors unrelated to earnings management in the period (Dechow et al., 1995; Guay et al., 1996; Healy, 1996). 12 We therefore follow Healy (1996) and label estimated pre-managed accruals as normal accruals, and estimated managed accruals as abnormal accruals. We estimate abnormal accruals using the modified-jones (m-j) model. Dechow et al. (1995) present evidence that the m-j model is more powerful at detecting sales-based manipulations than the original Jones (1991) model. Consistent wh recent studies (e.g., Becker et al., 1998; DeFond and Subramanyam, 1998), we estimate the m-j model on a cross-sectional basis in order to maximize our sample size and to avoid the survivorship bias problem inherent in the time-series approach. In contrast to some prior studies that 10 Accruals-based measures are theoretically appealing because they aggregate into a single measure the net effect of numerous recognion and measurement decisions, thereby capturing the portfolio nature of income determination (Watts and Zimmerman, 1990). Moreover, to the extent that boards treat earnings from operations and cash from operations as key indicators, a simple comparison of the two numbers will provide them wh direct insight into the aggregate effect of operating accruals. 11 See Dechow et al. (1995) for a review of these models. 12 Confounding factors include exogenous shocks to firm performance, strategic operating decisions, and the reversal of prior-period discretionary accruals. 12

15 model total operating accruals, we focus specifically on the working capal component. Modeling working capal accruals is more likely to improve the power of our tests because systematic earnings management via the depreciation accrual is likely to have limed potential (Beneish, 1999). 13 The m-j model parameters are estimated using the following cross-sectional OLS regression: 14 WC i = ω + ω REV + ν, (1) 0 1 i i where WC i is working capal accruals for firm i in a given year, defined as the change in non-cash current assets minus the change in current liabilies, REV i is the change in revenue, ω 0 and ω 1 are regression coefficients, and ν i is the regression residual. Model estimates are obtained for each industry and year combination. For estimation purposes, we scale all variables by beginning-of-period total assets. Industry-year portfolios wh less than ten observations are excluded from the analysis. Following Dechow et al. (1995), abnormal accruals (AA) are defined as follows: AA i = WC ωˆ + ωˆ ( REV REC )], (2) i [ 0 1 i i 13 In supplementary tests, we repeated all analyses using abnormal accruals derived from a measure of total operating accruals (i.e., working capal accruals minus depreciation). In all cases, our findings are unchanged. 14 In Dechow et al. s (1995) original specification of the m-j model, adjustment of the REV term for the change in receivables ( REC) is done after equation (1) has been fted. In other words, for estimation purposes, the original Jones model and m-j model are equivalent. Recently, however, researchers have estimated cross-sectional versions of the m-j model that adjust REV by REC at the estimation stage (e.g., Rajgopal et al., 1999). We repeated our empirical tests using this alternative specification of the m-j model. In all cases, the findings were consistent wh those based on equations (1) and (2). 13

16 where ˆω 0 and ˆω 1 are the OLS regression estimates of ω 0 and ω 1 respectively, obtained from equation (1), and REC i is the change in receivables during the year in question. This second stage adjustment of REV i by REC i effectively treats ˆω 1 REC i as a component of abnormal accruals Earnings thresholds Our empirical tests seek to isolate the incentives for earnings management behavior by comparing pre-managed earnings wh benchmark or target earnings levels. We define pre-managed earnings (PME t ) as reported earnings (EARN t ) minus estimated abnormal accruals (AA t ). Following the results reported by Burgstahler and Dichev (1997) and Degeorge et al. (1999), we employ two benchmarks for current period earnings: (i) zero and (ii) earnings reported in the previous year (EARN t-1 ). 15 We predict that managers will seek to manipulate earnings upwards when eher PME t < 0 or PME t < EARN t-1. In contrast, we predict that managers will seek to manipulate earnings downwards when PME t exceeds the zero and EARN t-1 thresholds by a large margin, so as to increase the likelihood of meeting the respective threshold next period. We then examine whether the level of board monoring influences the amount of upward 15 Degeorge et al. (1999) also find weaker evidence of a higher (lower)-than-expected frequency of firms wh reported earnings slightly higher (lower) than analysts earnings forecasts, suggesting that managers may also manipulate earnings upwards to meet or exceed such forecasts. We do not consider analysts forecasts as a threshold in this study because the work of Degeorge et al. indicates that the desire to manage earnings to meet or exceed analysts forecasts is dominated by the desire to avoid reporting earnings losses or earnings declines. 14

17 (downward) earnings management when pre-managed earnings undershoot (significantly exceed) target earnings The model We test the monoring role of boards in relation to two forms of earnings management: (i) income-increasing abnormal accruals, when pre-managed earnings are below a specified threshold and (ii) income-decreasing abnormal accruals, when premanaged earnings exceed the threshold by a given margin. We estimate regression (3) separately for each earnings threshold (zero and EARN t-1 ): AA = β + β BELOW β HIGH 2 + β OUT + γ OUT BELOW + λ OUT HIGH + β AC + γ AC + K 4 k= 1 δ Controls k 3 2 k BELOW + λ AC + ε HIGH (3) In equation (3), AA is abnormal accruals estimated using the m-j model. BELOW is an indicator variable taking the value of one if pre-managed earnings are below threshold earnings (PME t < 0; PME t < EARN t-1 ) and zero otherwise. Holding board monoring effects constant, we predict that β 1 will be posive due to income-increasing earnings management behavior. HIGH is an indicator variable taking the value of one if premanaged earnings exceed zero or EARN t-1, as the case may be, by a specified margin and zero otherwise. 16 Again holding board monoring effects constant, we predict that β 2 will 16 For the regressions where PME is benchmarked against zero, we define HIGH as one if PME for firm i in period t exceeds the 75 th percentile of the pre-managed earnings distribution and zero otherwise. For the regressions where PME t is benchmarked against last year s reported earnings, HIGH is set equal to one if PME t minus EARN t-1 for firm i in period t exceeds the 75 th percentile 15

18 be negative due to income-decreasing earnings management behavior. OUT is the proportion of outside board members and AC is an indicator variable taking the value of one if the firm has an aud commtee and zero otherwise. If outside directors and aud commtees fulfil a monoring role and seek to reduce earnings management, we predict that γ 1 and γ 2 will be negative and that λ 1 and λ 2 will be posive Control variables Regression (3) includes proxies for other governance factors, other earnings management stimuli, and firm performance. Controlling for other governance dimensions minimizes the potential correlated omted variables problem that arises when board effectiveness is correlated wh the presence of other monoring mechanisms (Agrawal and Knoeber, 1996). First, we include a measure of board size (BRDSIZE) and an indicator variable (DUAL) taking the value of one if the roles of chairman and chief executive are combined because prior work identifies these variables as reflecting addional board monoring dimensions (e.g., Yermack, 1996; Jensen, 1993). Secondly, we control for three aspects of ownership structure. As discussed earlier, managerial ownership (BRDOWN) acts as a proxy for the extent to which the interests of managers and shareholders diverge. We include a dummy variable (BLOCK) to capture the control potential of large external blockholdings (Dhaliwal et al., 1982). Finally, we also include the total fraction of outstanding shares held by instutional investors (INSTOWN) to control for the special monoring role attributed to instutional stockholders (Rajgopal et of that variable s distribution, and zero otherwise. Alternative cutoff points were also examined wh broadly the same results to those reported in section 4. 16

19 al., 1999). Third, prior research suggests that firms wh Big 5 audors are less likely than other firms to report income-increasing abnormal accruals (Becker et al., 1998) and that they are more likely to have an aud commtee (Pincus et al., 1989). To control for potential audor qualy effects, we include an indicator variable (AUD) defined as one if a firm has a Big 5 audor, and zero otherwise. The set of control variables in regression (3) includes measures of size, leverage and financial performance. We include firm size (SIZE) to control for possible negative correlation between firm size and accounting choice (Christie, 1990). Empirical evidence indicates that income-increasing accruals may be motivated by a desire to delay or avoid costs associated wh debt covenant violation (DeFond and Jiambalvo, 1994). We therefore include leverage (LEV) to capture the likelihood of debt covenant violation. DeFond and Park (1997) present evidence that managers use accruals to smooth earnings to take account of current relative performance, where relative performance is defined as pre-managed earnings less the sample median earnings for the corresponding industry. Accordingly, we include in our empirical model a measure of relative performance (REL), defined as one if pre-managed earnings are below median reported earnings for the industry, and zero otherwise. We also include cash flow from operations (CFO) to control for the association between abnormal accruals and operating cash flow suggested in the results of Dechow et al. (1995) and Guay et al. (1996). Finally, we also include a vector of indicator variables (YEAR) to control for time effects. 17

20 4. Analysis 4.1. Data The empirical tests are conducted using data for U.K. listed firms wh fiscal year ends between June 30, 1993 and May 31, The m-j model is estimated for each industry (Datastream level-6) and year combination, using all firms on the Datastream Active and Research files wh available accruals data. 17 The number of firms used to estimate the models in 620 in 1993, 651 in 1994, and 657 in The maximum number of observations for any given industry-year combination is 56 (general engineering, 1995) while the mean (median) estimation portfolio size is 21 (18) observations. Board data were hand-collected using the following sampling procedure. For each year we selected the largest 1000 listed firms based on market capalization at December 31, as reported in the London Share Price Database. We then excluded all financial firms (SIC codes 60-69) because they have fundamentally different accruals processes that are not captured by the m-j model. We also excluded all regulated utilies (SIC codes 40-44, 46, 48-49) because of differences in their incentives and opportunies to manage earnings. The Price Waterhouse Corporate Register is used as the source of board composion data, and firms annual reports are used to identify the presence or absence of an aud commtee. 18 The Corporate Register is also the main source of managerial stock ownership data, supplemented where necessary by annual reports. The 17 The sample of firm-years used to test the association between earnings management and board effectiveness is a subset of the firm-year observations used to estimate the m-j model. 18 The Corporate Register is published quarterly by Hemmington Scott Ltd. and includes data for all firms listed on the London Stock Exchange. Hemmington Scott constructs their board composion database using information from the London Stock Exchange and Reuters. 18

21 Corporate Register reports the number of shares held by each executive board member, classified into beneficial and non-beneficial categories, together wh the total number of the firm s shares outstanding at the fiscal year-end. It is often impossible to determine the voting and control rights of non-beneficial shareholdings in the U.K. We therefore restrict our definion of ownership to beneficial shareholdings. For the same reason, we also exclude all family and family-related holdings, together wh shares held in employee pension and stock option plans. Data on external share ownership were collected from the Stock Exchange Official Yearbook. 19 Data on audor identy were collected from the Corporate Register. All remaining data, including that used to calculate abnormal accruals, were obtained from the Datastream. The final sample consists of 1,271 firm-years wh the complete data needed to estimate regression (3). The sample comprises 559 firms from 34 Datastream level-6 industrial groups. 20 The distribution of firms across sample years is as follows: 125 firms are included in only one year, 156 firms are included in two years, and 278 firms have data for each of the three sample years. Annual sample sizes are 406, 453 and 412 for 1993, 1994 and 1995, respectively. In contrast to Beasley (1996) and Dechow et al. (1996), none of the firms in our sample have been publicly revealed by the UK 19 The Stock Exchange Official Yearbook records the fraction of shares held by each stockholder for all cases where the holding exceeds 3% of the total outstanding equy. The 3% disclosure rule for external shareholdings is a threshold imposed on all listed firms by the London Stock Exchange. The Yearbook is compiled using data filed wh the London Stock Exchange, supplemented wh data from annual reports. 20 Firms are distributed fairly evenly across the industrial groups. The largest industry is Engineering wh 115 firm-year observations, representing 9% of the final sample. No other industrial group contains more than 6% of the sample. 19

22 accounting standards enforcement agency (the Financial Reporting Review Panel) to have violated GAAP, nor have any been accused of fraudulent financial reporting during the sample period. When pre-managed earnings are benchmarked against zero, 234 firmyear observations (18.4%) are classified as below target, while 260 (20.5%) are classified as well-above target. Alternatively, when pre-managed earnings are benchmarked against EARN t-1, 563 firm-year observations (44.3%) are classified as below target, while 177 (13.9%) are classified as well-above target. The difference in the below-target sample sizes reflects the fact that fewer firms report losses than earnings declines. Descriptive statistics for the explanatory variables are reported in table 1. The average board in our sample contains eight directors, of which approximately 43% are outsiders. Aud commtees are present in 85% of firm-year observations. Addional (untabulated) tests reveal no significant association between the level of pre-managed earnings and eher OUT or AC. In other words, firms whose pre-managed earnings undershoot the threshold have board structures that are indistinguishable from firms wh pre-managed earnings that exceed the threshold. Regarding the addional control variables, the roles of chairman and chief executive are combined in 24% of cases. The typical sample firm has median managerial (instutional) ownership of 2% (19%). These levels are comparable to those reported in previous studies of the UK (e.g., Short and Keasey, 1999). Over half of the sample (53%) has at least one external blockholder whose stake exceeds ten percent. A sizeable majory of firms in the sample (85%) is auded by one of the Big 5 firms. 20

23 Table 2 reports descriptive statistics for abnormal accruals estimated using the m- J model, partioned according to the level of pre-managed earnings. Panel A presents findings when PME is benchmarked against zero, while panel B reports findings when PME is benchmarked against last year s reported earnings. Predictions on the sign and magnude of abnormal accruals are potentially ambiguous when PME undershoots the benchmark by a large amount. On the one hand, potential costs borne by management as a result of low performance, such as addional monoring (DeAngelo et al. 1994) and increased likelihood of dismissal (Weisbach 1988), suggest a preference for incomeincreasing earnings management. On the other hand, the big bath hypothesis predicts that when PME is sufficiently far below benchmark, managers may prefer income-decreasing abnormal accruals, thereby storing up the potential for increasing reported earnings in future periods (Degeorge et al 1999; Healy 1985). Big bath earnings management would potentially confound our empirical tests that pool below-target observations based on the assumption that managers will favor income-increasing abnormal accruals in such cases. In both panels, therefore, we examine abnormal accruals after partioning the belowtarget sub-groups into quintile portfolios based on the extent to which PME undershoots the target. Mean (median) abnormal accruals for the full sample are indistinguishable from zero at conventional levels of significance. For the below-target observations in panel A, mean (median) abnormal accruals are posive and significant in each of the quintile portfolios, suggesting that working capal accruals are being managed upwards in an effort to avoid an earnings loss. Particularly noteworthy are the posive means (medians) for quintiles four and five. These findings are inconsistent wh the big bath hypothesis 21

24 that predicts income-decreasing earnings management when pre-managed earnings undershoot the threshold by a large amount. Similar results are also reported in panel B when pre-managed earnings are benchmarked against last year s reported earnings. In particular, abnormal accruals are posive and significant in each of the below-target quintile portfolios, indicating a tendency for managers to manipulate earnings upwards in an effort to avoid reporting an earnings decline. Overall, these findings suggest that observations in the below-target sub-samples are homogenous wh respect to the direction of earnings management and hence that pooling all below-target observations is justified. Panels A and B of table 2 also report mean and median abnormal accruals for observations where pre-managed earnings exceed the relevant threshold. Consistent wh our predictions of income-decreasing earnings management when pre-managed earnings are high, average abnormal accruals in the well above target partions are negative and significant in both panels. Furthermore, abnormal accrual activy is much less pronounced for firms whose pre-managed earnings are just above the target The rational for earnings management Before considering whether board monoring influences the extent of earnings management, we must first establish that such earnings management is not in stockholders interests. If managers are using their accounting discretion to signal private information about future earnings performance, then is unclear why effective monoring by boards should seek to constrain such behavior. We therefore present evidence on the extent to which the abnormal accruals documented in section 4.1. are 22

25 being driven by signaling considerations. If a signaling rationale explains abnormal accrual activy, then we would expect to observe a posive association between current period abnormal accruals and future earnings changes. We test this using a logistic regression where the dependent variable takes the value of one if one-period-ahead premanaged earnings are higher than those reported in the current period and zero otherwise. In addion to abnormal accruals, the set of explanatory variables also includes measures of operating cash flow and normal accruals. Separate regressions are estimated for the below- and above-threshold sub-samples. The results are presented in table 3. Regardless of whether pre-managed earnings are benchmarked against zero (panel A) or EARN t-1 (panel B), or whether pre-managed earnings are below or above the threshold levels, the estimated coefficients on AA are consistently negative and generally significant. These findings do not support a signaling explanation and as such are consistent wh results reported by Dechow et al. (2000). We therefore conclude that is appropriate to treat abnormal accruals as potentially costly to market participants Board monoring Figure 1 presents plots of earnings and pre-managed earnings, sorted by the level of pre-managed earnings, for sub-samples partioned on the basis of board composion (OUT). Panel A reports the results when pre-managed earnings are benchmarked against zero. The plots are constructed by first ranking all observations where board composion is above (below) the sample median by pre-managed earnings, and then plotting the median values of pre-managed earnings and reported earnings for each percentile of the pre-managed earnings distribution. Similarly, panel B reports results when pre-managed 23

26 earnings are benchmarked against last period s reported earnings. The plots are constructed by ranking all observations where board composion is above (below) the sample median by PME t EARN t-1, and then plotting the median values for PME t EARN t-1 and EARN t EARN t-1 for each percentile of the pre-managed earnings distribution. In both panels, therefore, the difference between the two traces represents abnormal accrual activy. The results in figure 1 confirm the evidence presented by Burgstahler and Dichev (1997) and Degeorge et al. (1999) that firms seek to manage reported earning upwards to exceed benchmarks. In particular, there is evidence of a substantial disconnection between the pre-managed earnings trace and the reported earnings trace when premanaged earnings undershoot the threshold. In such cases, managers appear to be using posive abnormal working capal accruals to help the firm meet or exceed the target. Once pre-managed earnings exceed the target, the level of posive abnormal accruals declines and the pre-managed and reported earnings traces track each other more closely. Note, however, that when pre-managed earnings exceed the target by a large margin (i.e., beyond the 80 th percentile), the earnings trace dips below the pre-managed earnings trace, suggesting that managers are reining in reported earnings through the use of incomedecreasing abnormal accruals. This finding is consistent wh the results reported by Degeorge et al. (1999) for the US. While the patterns described above are evident for both the high and low board composion sub-samples in figure 1, there is some suggestion that the patterns are more pronounced for firms whose proportion of outside directors is less than the sample median. For example, in panel A when pre-managed earnings are benchmarked against 24

27 zero, median reported earnings become posive at the fourth percentile of pre-managed earnings for the low board composion sub-sample, compared wh the eighth percentile for the high composion sub-sample. Similarly in panel B when pre-managed earnings are benchmarked against EARN t-1, the disconnection between the pre-managed earnings trace and the reported earnings trace is more pronounced for the low composion subsample in the region of zero. These results provide preliminary evidence of a link between board monoring and abnormal accrual activy. In addion, there is also evidence that boards wh relatively few outside directors rein in earnings to a larger extent when pre-managed earnings exceed the target by substantial margin. In particular, the disconnection between the pre-managed earnings trace and the reported earnings trace above the 80 th percentile of pre-managed earnings appears much more pronounced in the low board composion sub-sample. To assess the significance of the impact of board monoring on abnormal accrual activy, table 4 presents estimates of regression (3). Columns (i)-(ii) present results for pre-managed earnings benchmarked against zero while columns (iii)-(iv) present findings for pre-managed earnings benchmarked against EARN t-1. In each case, we report two versions of the model: M1 includes only our test variables OUT and AC, together wh the indicator variables BELOW and HIGH and their interactions wh the test variables; M2 is the full model, inclusive of the vector of control variables. We begin our discussion of the results by focusing on columns (i) and (ii) in table 4 when pre-managed earnings are benchmarked against zero. The posive and significant coefficient estimates on BELOW in models M1 and M2 confirm prior findings that managers make posive abnormal accruals when pre-managed earnings are negative. 25

28 Similarly, the negative and significant coefficients on HIGH in M1 and M2 provide evidence that managers seek to manipulate earnings downwards when pre-managed earnings are high. The results also provide evidence that outside directors monor earnings management activy in certain circumstances. In both models, the estimated coefficient on the interaction term OUT BELOW is negative and significant at the 0.01 level. This suggests less income-increasing earnings management behavior to meet or exceed the threshold for firms whose boards comprise a higher proportion of outsiders. Comparison of models M1 and M2 reveals that this finding is insensive to the inclusion of control variables designed to measure other aspects of firms governance structures. In contrast, the estimated coefficients on the OUT HIGH interaction terms are not significant at conventional levels, suggesting that outside directors do not constrain income-decreasing manipulations (but see section 4.4 below). In addion to examining the impact of board composion on earnings management, models M1 and M2 also investigate whether the presence of an aud commtee influences abnormal accrual activy. If aud commtees help to constrain income-increasing earnings management then we would expect the estimated coefficient on the AC BELOW interaction term to be negative and significant. Similarly, if aud commtees help to lim income-decreasing earnings management, then the estimated coefficient on AC HIGH should be posive and significant. Contrary to these predictions, however, the results presented in columns (i) and (ii) of table 4 provide no support for the aud commtee hypothesis: the estimated coefficients typically display the wrong signs and none are significant at conventional levels. Of the control variables included in M2, only BLOCK, SIZE, REL and CFO are significant at the 0.05 level or better. 26