Corporate Governance and Disclosure: The Effect of Institutional Investors and Outsider Directors on the Properties of Management Earnings Forecasts

By Bipin Ajinkya1 Fisher School of Accounting University of Florida Gainesville, FL 32611 [email protected] Sanjeev Bhojraj Johnson Graduate School of Management Cornell University Ithaca, NY 14853 [email protected] Partha Sengupta Robert. H. Smith School of Business University of Maryland College Park, MD 20742 [email protected]

First Draft: May 2002. Current Draft: January 2003

_________________________ 1

Contact Author. We thank Kate Campbell and Charles Lee for their comments. We also thank Brian Bushee for generously providing us his data classifying institutional ownership and Thompson Financial for generously providing us First Call analyst forecast and management forecast data through their Academic program.

Corporate Governance and Disclosure: The Effect of Institutional Investors and Outsider Directors on the Properties of Management Earnings Forecasts

ABSTRACT: This paper investigates the effect of institutional ownership and board of directors’ composition on the properties of management earnings forecasts. Managers acting in their self-interest have incentives to distort actual disclosure relative to optimal disclosure. Governance mechanisms acting in the interests of the providers of capital should mitigate these distortions. We find evidence that institutional ownership and outside directors have a positive effect on the properties of earnings forecasts. Firms with greater institutional ownership and outside directorship are more likely to issue a forecast and consistently do so. Further, these forecasts are likely to be more specific, and accurate. These measures also mitigate the managers’ tendency to issue optimistic forecasts. We also find that institutional ownership creates an environment that enhances the credibility of the forecasts.

1. Introduction The U.S. corporate world is dominated by publicly traded firms with widely dispersed ownership. Typically, shareholders designate firm managers to run the company with the goal of maximizing shareholder wealth. Since shareholders do not participate in day-to-day corporate activities, implicit or explicit governance mechanisms assist in monitoring management actions and protecting shareholders’ interests. One area in which governance mechanisms can act on behalf of shareholders is in ensuring precise, accurate and credible dissemination of information by the firm to its providers of capital. A firm’s optimal disclosure policy is determined by weighing the benefits of the disclosure (especially capital market benefits) against the costs of the disclosure (especially proprietary and litigation costs). However, managers acting in their self-interest may have incentives (including reputation costs and insider trading opportunities) to distort disclosure by issuing fewer, less specific and biased forecasts.

Corporate governance mechanisms,

representing the interests of the shareholders could mitigate these distortions.1 The focus of this study is to determine the relationship between a set of governance mechanisms and the properties of earnings forecasts.

Specifically, we examine the role of institutional investors and outside

directors in facilitating more frequent, specific (i.e., precise), accurate, less optimistic and more credible earnings forecasts from firm managers. Prior work examining the role of institutional investors has focused on whether these shareholders help protect investor interests in various contexts, including mergers and takeovers 1

In “A Survey of Corporate Governance”, Shleifer and Vishny (1997) define corporate governance as the ‘ways in which suppliers of finance to corporations assure themselves of getting a return on their investments’. Corporate governance mechanisms ensure that managers do not steal from the providers of finance. In their view these mechanisms encompass economic and legal institutions. Thus corporate governance goes beyond mechanisms instituted by the organization itself and would include product market competition, monitors like large shareholder and outside directors and laws protecting creditors, minority shareholders etc. Past research has studied large shareholders, institutional ownership, outsider directors and the takeover market as some of the governance mechanisms.

and management turnovers, and whether they help enhance firm performance and value.2 Prior work has also examined the association between analyst disclosure scores and institutional ownership. Healy, Hutton, and Palepu (1999) and Bushee and Noe (2000) find that institutions prefer to buy firms that have sustained disclosure increases. Bushee and Noe (2000) also partition institutions into three groups based on their investment horizon etc. and find that changes in disclosure scores have the greatest impact on ‘transient’ investors. In this study we extend the prior research on institutional ownership and disclosure in a number of ways. First, we examine the effect of institutional ownership on various properties of management forecasts. A key demand of the International Corporate Governance Network, a group representing the interests of major institutional investors, corporations and financial intermediaries is that ‘Corporations should disclose accurate, adequate, and timely information…so as to allow investors to make informed decisions about the acquisition, ownership obligations and rights and sale of shares’. Given that disclosures, especially earnings forecasts are closely watched by institutions and the constant probing of the companies for the short-term earnings outlook, we expect to see a positive impact on a firm’s propensity to issue forecasts, the specificity and accuracy of forecasts issued, and in reducing managerial optimism. Consistent with our hypotheses, we find that firms with higher institutional ownership are more likely to issue a management forecast. They also issue forecasts more frequently (over the four year sample period) and these forecasts tend to be more specific. This is an interesting finding that also contributes to the literature on managerial opportunism in that, unless a policing mechanism is in

2

See for example, Jarrell and Poulsen (1987); Shivdasani (1993); Denis and Serrano (1996); Weisbach (1988); Kang and Shivdasani (1995); Borokhovich, Parrino, and Trapani (1996); Agrawal and Knoeber (1996); Bhagat and Black (1997). Other research explores the effects of governance on CEO compensation (e.g. Core, Holthausen, and Larcker (1999)) and effects of governance on the cost of debt (Bhojraj and Sengupta (2002)). 2

place, a manager is likely to act opportunistically.3 Using a sub-sample of point forecasts we also find that forecast accuracy (absolute forecast error) is positively (negatively) associated with the institutional ownership while forecast optimism is found to be negatively associated with it (i.e., managers are more “conservative” in their forecasts). Finally, our evidence indicates that analysts react more to forecasts issued by firms with higher institutional ownership, suggesting greater credibility.4 Second, we extend prior research by suggesting an endogenous relation between institutional ownership and disclosure.

As discussed above, one of the key demands of

institutions is greater disclosure. This suggests that while disclosure has been found to influence future institutional ownership it is also likely that institutional ownership would affect future disclosure. We carry out a simultaneous regression analysis and a Granger causality lead-lag analysis to examine this issue and consistent with our expectations, we find that institutional ownership does seem to affect future disclosure.5 Finally, we partition institutional ownership based on the investment horizon and ability to generate private benefits and examine their effect on forecast properties. Prior research suggests that institutions are not a homogenous group and their incentives are likely to be determined by their investment horizons as well as their ability to generate private benefits. To test the effect of differential institutional incentives on forecast properties, we carry out further analysis by partitioning the institutional ownership variable, based on prior work. We find that concentrated/dedicated institutional ownership has an adverse effect on forecast properties. 3

See Healy (1985); Jones (1991); Burgstahler and Dichev (1997) etc., for evidence that managers are likely to act opportunistically. 4 Surprisingly, the accounting (and finance) literature contains little empirical evidence on factors influencing information credibility. Williams (1996) finds that analyst reaction to a current management forecast is influenced by the accuracy of past management forecasts 5 Miller and Piotrowski (2000) and Chen (2002) study the effect of institutional ownership on forecast occurrence. However, they do not control or correct for the endogenity between ownership and forecast occurrence. 3

Firms with highly concentrated institutional ownership are less likely to issue earnings forecasts and are less likely to issue them consistently. The earnings forecasts also tend to be less specific. Tests of the effects of concentrated institutional ownership on forecast accuracy and bias however were largely inconclusive. Prior work in governance had examined the role of outside directors and found that outside directors play an active role in monitoring corporate activities such as replacing poorly performing CEOs (Weisbach (1988)) and protecting shareholder interests during takeover fights (Shivdasani (1993)). There is some evidence to suggest that outside directors also monitor the financial information a firm generates (Beasley (1996)).6 To the extent that managers distort actual disclosures away from the optimal disclosure policy (based on cost-benefit tradeoff to the firm) and outside directors play a role in mitigating these distortions, we expect to find that firms with greater outside directors have greater propensity to issue forecasts, issue more specific and accurate forecasts and forecasts that are less optimistic. Consistent with this reasoning we find that the probability of occurrence of management earnings forecasts and the frequency of such forecasts are positively associated with the percentage of the board that consists of outsiders. There is also some evidence to suggest that companies with greater percentage of outside directors make more accurate (and less biased) earnings forecasts although this result does not hold for all sub-samples. Our results, linking corporate governance to disclosure, are interesting given the current scrutiny of corporate governance mechanisms and the state of the financial reporting system. Both governance mechanisms and the financial reporting system have come under siege in the 6

We contacted a few companies/directors to obtain further information and anecdotal evidence on this issue. Our sense is that board meetings occur regularly and earnings releases (including guidance) are discussed. In addition, board members are generally given copies of earnings releases ahead of time. One director we spoke to informed us that the audit committee discussed earnings announcements and guidance with the management. In the event of the guidance being off-mark, the board would hold the management accountable. 4

wake of a series of financial scandals including Enron and WorldCom. These scandals have led to a greater focus on the need for stronger governance and more transparent disclosure. Our results suggest that the two are linked, and that by implementing stronger governance, greater transparency with result. The rest of the paper is organized as follows. Section 2 motivates the paper and develops the hypotheses, while section 3 outlines the method we adopt to test our hypotheses. The results are provided in section 4 and section 5 summarizes and concludes the paper.

2. Motivation and Hypothesis Development Benefits and costs of earnings forecasts: The benefits of issuing management forecasts have been well documented in prior empirical work. Earnings forecasts have been found to reduce the risk of excessive mis-pricing of stock and large stock price volatility (Bushee and Noe (2000)), reduce the likelihood of shareholder litigation (Skinner (1994); Kasznik and Lev (1995)) and reduce bid-ask spreads (Coller and Yohn (1997)). In addition, Frankel, Johnson, and Skinner (1999) find that firms issue more forecasts prior to accessing capital markets.

Prior work has also examined the

benefits of more specific disclosures. Kim and Verrecchia (1991), in analytically examining trading volume and price reactions to public announcements, find that price change at the time of an announcement is increasing in the specificity of the announcement. Baginski, Conrad, and Hassell (1993), using a sample of earnings forecasts of varying specificity provide empirical evidence supporting this theoretical result. Prior evidence also suggests that markets respond favorably to accurate forecasts.

Williams (1996) finds that analyst reaction to a current

management forecast is influenced by the accuracy of past management forecasts. The above

5

discussion suggests that managers acting in the best interests of the firm should enhance transparency by issuing more frequent, specific and accurate forecasts. However, while the financial markets might prefer increased transparency, the optimal disclosure policy is determined by the trade-off between the costs of disclosure (including litigation and proprietary costs) and benefits of the disclosure. Theoretical work (e.g., Darrough and Stoughton (1990); Wagenhofer (1990)) suggests that firms sometimes desist from making voluntary disclosures for fear of revealing proprietary information. This information could be used by competitors to make entry or exit decisions or in determining their response to the firm’s strategy. Firms facing high proprietary costs may choose not to disclose information or disclose information of lower quality. While revelation of proprietary information is an important cost of certain types of voluntary disclosures (e.g., announcing new products, strategic alliances etc.) we believe that it is likely to be less of a factor with earnings forecasts. Earnings forecasts are release of information a short period prior to when it would normally be disclosed. Therefore competitors are unlikely to derive large benefits from an earnings forecast. Competitors are also unlikely to base any strategic decisions (e.g., product positioning, entry or exit decisions etc) on the firm’s earnings forecast. Firms are therefore unlikely to reduce the frequency or quality of their forecasts based on potential proprietary costs. Consistent with this argument, Bamber and Cheon (1998) find mixed results in studying the association between their proxies for proprietary costs (market to book and sales concentration) and forecast specificity. However, to the extent that proprietary costs could influence the optimal forecast decision (and therefore the effect of governance) we include market to book (in keeping with Bamber and Cheon, 1998) in our analysis.7

7

We also carry out the analysis using sales concentration (their other proxy for proprietary information) and our results are unchanged. 6

Prior work suggests that the decision to issue forecasts is also likely to be determined by litigation issues. Francis, Philbrick, and Schipper (1994) find that firms that are prone to litigation are less likely to issue forecasts, while Bamber and Cheon (1998) and Baginski, Hassell, and Kimbrough (2002) find that the forecasts are likely to be less specific in the presence of litigation risk. However, Skinner (1994) and Kasznik and Lev (1995) find that firms are more likely to issue forecasts to communicate bad news. These results suggest that the institutions and outside directors’ incentives to induce more forecasts and forecasts of higher specificity could be tempered for firms in litigation prone industries. We control for the effect of litigation in our study, by including a proxy for firms that are in litigation prone industries (based on Francis et al. (1994)).8

Managerial incentives to distort disclosure Managers acting in their self-interest may have incentives to distort disclosure. They can ‘extract resources from the firm (through perquisite consumption, higher wages, "unwarranted" continued employment, slack, etc.) by maintaining information asymmetry’ (King and Wallin, 1995). Managers may also withhold or issue less specific information for reputation related reasons. Lees (1981) find that one of the principal reasons for managerial reluctance to issue forecasts is the ‘management’s lack of confidence in their ability to predict future trends and events’. Erroneous forecasts could have an adverse impact not just on the manager’s reputation but also on factors like compensation, continued employment etc. In discussing managerial ability, Trueman (1986) expects competent managers to be more likely to forecast and to issue the forecast earlier. Tan, Libby, and Hunton (2002), in debriefing analysts, find that analysts

8

We also run tests using an industry classification based on Kasznik and Lev (1995). Our results remain unchanged. 7

view management forecast accuracy as reflecting managerial competence. Williams (1996) finds that managers who make an erroneous forecast lose their credibility which is reflected when a subsequent forecast is released. To the extent that managers are afraid of making a mistake which would reflect poorly on their competence, they would desist from making a forecast and in the event of a forecast being made they would opt to make a less specific forecast. Another reason for managers to withhold forecasts (especially bad news) or issue forecasts that are less specific is that subsequent events could offset the bad news (Givoly and Palmon, 1982; Patell and Wolfson, 1982; King and Wallin, 1995) or they may be able to manage earnings to offset the bad news. Lobo and Zhou (2001) find that firms that disclose less tend to engage in greater earnings management.

Finally, a large body of literature on insider trading focuses on the

benefits of withholding information. Evidence suggests managers trade on private information relating to upcoming events including corporate sell offs (Hirschey and Zaima, 1989), new issue announcements (Karpoff and Lee, 1991), dividends (John and Lang, 1991) and bond ratings (Elliott, Morse, and Richardson, 1984). Penman (1982) found evidence of greater insider selling before price decreasing forecasts and greater insider buying before price increasing forecasts. However, Noe (1999) finds that managers cluster transactions after an earnings disclosure rather than before them.9 Fried (1998) estimates insider trades at approximately seventy billion dollars each year and insider profits at roughly five billion dollars, most of which are generated through exploiting insider information. The potential reputation costs (and ensuing tangible costs), the incentive to delay disclosure of bad news and potential insider trading gains (if any) would

9

This is consistent with managers being careful not to violate securities laws and incur litigation exposure, by trading immediately prior to a forecast. However, both studies are based on a sample of forecasting firms only. They study trading behavior around these forecasts. However, these studies exclude all the firms that did not make a forecast during a period and the trading patterns of managers in those firms. Thus managers could be trading on information they withheld (i.e., a forecast was never disclosed) and it would not be detected in these studies. 8

suggest that managers acting in their self interest are likely to make fewer forecasts and forecasts of lower specificity. As discussed earlier, the litigation risk is asymmetric (Skinner, 1994) suggesting that an optimal policy would be for managers to be conservative and cautious in the forecasts they issue. In addition to litigation issues, other reasons to issue conservatively biased forecasts include an asymmetric response to negative earnings (Skinner and Sloan, 2001) and the focus of markets on meeting or beating expectations. Soffer, Thiagarajan, and Walther (2000) find that a strategy of forecasts that are conservatively biased (i.e., all bad news and only a portion of good news at the announcement date) results in higher returns after the actual earnings release. Matsumoto (2002) finds evidence suggesting that firms guide analyst forecasts downwards to avoid a negative surprise.

These findings suggest that firms are better off issuing conservative forecasts.

However, managers acting in their self-interest might want to issue more optimistic (or less pessimistic) forecasts in hope that future events or earnings management will ensure their realization. This tendency could stem from their desire to portray themselves as good managers or to trade on the optimistic information disclosed. Prior research has also shown that managers have a predilection to issuing good news and delaying bad news (Givoly and Palmon, 1982 Patell and Wolfson, 1982; Chambers and Penman, 1984); King and Wallin, 1996; Narayanan 2000). 10

10

Early literature studying management forecasts found they were optimistically biased (Patell and Wolfson (1982); Penman (1982); Lev and Penman (1990)). The underlying idea was that managers would issue disclosure only if it entailed providing good news. More recent work however suggests that the bias has shifted, with managers having strong incentive to avoid negative earnings surprises. However, these findings reflect the end product of the effect of governance mechanisms and other factors and do not capture the ex-ante biases of the manager. 9

Role of institutional ownership in mitigating managerial distortions Institutional owners acting in the interests of providers of capital can play a role is mitigating these distortions. Shleifer and Vishny (1986) argue that institutional shareholders, by virtue of their large stockholdings would have incentives to monitor corporate performance since they accrue greater benefits to monitoring and enjoy greater voting power that makes it easier to take corrective actions when deemed necessary. Jarrell and Poulsen (1987) and Brickley, Lease, and Smith (1988) find that institutional shareholders are more likely to vote against harmful amendments that reduce shareholder wealth.11 Institutions seem to desire and demand more disclosure. Disclosures, especially earnings forecasts are closely watched by the market participants.12 This can be evidenced by listening in on conference calls, where institutions consistently probe the company for more specific, unbiased and accurate information about future earnings. Brokerage houses regularly hold conferences where firms make presentations to institutional shareholders about the prospects of the company. Prior work (including Healy et al. (1999); Bushee and Noe (2000)) has shown that institutions prefer to buy firms that have sustained disclosure increases. It seems reasonable that these institutions would continue to demand further disclosure from the firms.

A key principle

of International Corporate Governance Network, a group representing the interests of major institutional investors, corporations and financial intermediaries, is communications and reporting:

11

Agrawal and Mandelker (1990) found a positive relationship between institutional ownership and shareholder wealth effects of various antitakeover charter amendments. McConnell and Servaes (1990) found a positive relationship between institutional ownership and productivity, as measured by Tobin’s q. 12 A sell-side (brokerage-house) analyst we spoke to stated that one of the key pieces of information the buy-side (including pension funds and asset management funds) consistently demands from her and the companies whose stock they own is information about near-term earnings. 10

‘Corporations should disclose accurate, adequate, and timely information…so as to allow investors to make informed decisions about the acquisition, ownership obligations and rights and sale of shares’. Given this disclosure oriented focus of institutions and their constant probing of the companies for the short-term earnings outlook, we expect to see a positive impact on a firm’s propensity to issue forecasts, the specificity and accuracy of forecasts issued, and in reducing managerial optimism. In addition, the presence of a credible monitoring mechanism (and the continuous oversight of the management by institutions) should have a favorable impact on the market’s response to the information (i.e. credibility). However, institutional owners’ incentives to monitor a firm and probe for information are likely to be constrained by the nature of their investment horizons and the private benefits they might derive from weaker public disclosure.

One group that prior research in corporate

governance has focused on is concentrated/dedicated institutional ownership.

These are

institutions that have a large stake in a firm and have low portfolio turnover.

Prior work

suggests that these institutions can have an undue influence over the management and secure benefits that are to the detriment of other providers of capital (other shareholders and bondholders). These benefits could take the form of pecuniary benefits (below market transfer prices, preventing the opening of closed end funds, underwriting or advisory contracts (Barclay, Holderness, and Pontiff, 1993) or non-pecuniary benefits (ability to influence the political, social or environmental policies of the firm etc.).13 These dedicated or blockholder institutions often have better access to private information (Porter (1992)) and therefore need not press the firms for public disclosures. Many may actively prefer fewer forecasts and forecasts of lower quality, 13

The shared benefits hypothesis suggests that blockholding leads to more efficient monitoring of the management and the benefits from such monitoring are shared by all stockholders. See Barclay and Holderness (1992) for a discussion of the private benefits and the shared benefits hypotheses. Other papers examining the benefits of large blockholders include Huddart (1993) and Maug (1998). 11

thereby giving them an advantage over the market. Bushee and Noe (2000) argue that since dedicated institutions are not frequent traders the liquidity benefits of disclosure are not likely to be important to them. Thus we expect these institutions to have either no impact or a negative impact on the frequency and quality of disclosure. In addition to a ‘dedicated institutions’ group, Bushee (1998) and Bushee and Noe (2000) also partition institutions into “quasi-indexers” and “transient” investors. Quasi-indexers are characterized by a passive buy and hold strategy. Their low turnover suggests a long investment horizon, which is similar to the dedicated investors. However, they differ from dedicated investors in that they do not enjoy the same access to private information and private benefits. They are dependent on public information are therefore likely to induce high quality disclosure. Thus, we expect these institutions to have a positive impact on the frequency and quality of disclosure. Transient investors adopt a strategy of high portfolio turnover, thereby having a short investment horizon. Being short-term focused they are not concerned with long-term capital appreciation and dividends (Porter, 1992). These firms do not enjoy the private benefits of dedicated institutions and do not have the same incentives as buy-and-hold institutions to induce disclosure. They are therefore hypothesized to have a favorable impact on disclosure though to a lesser degree than buy and hold institutions. Role of corporate boards in mitigating managerial distortions Corporate boards have the fiduciary duty of monitoring management performance and protecting shareholder interests. While their share of gains/losses will be limited since their stock ownership is usually low, outside directors should possess incentives, by virtue of their position, to monitor management actions more carefully compared to other directors. A number

12

of researchers, including Fama (1980) and Fama and Jensen (1983), have argued that outside directors bear a reputation cost if firm performance is poor. Gilson (1990) finds that directors who leave boards of distressed firms hold fewer directorships subsequently than when they resigned.

Several prior studies have documented the favorable impact (and governance role) of

outside directors on firm decisions aimed at enhancing shareholder wealth.14 However, an alternative view is that outside directors may be ineffective, either because they are appointed by (and have allegiance to) company managers or because the board culture discourages conflict (Mace, 1986; Jensen, 1993).15 Very little has been work has focused on the link between outside directors and information disclosure information.

As discussed earlier managers have an

incentive to distort or selectively release information. Beasley (1996) finds that firms with higher proportion of outside directors are less likely to suffer from financial statement fraud. This suggests that directors do monitor the financial information being generated by the firm (see note 6). However, this does not imply that directors would induce managers to publicly disclose this information (especially if it is detrimental to the interests of the firm). We expect outside directors to act in the interests of the shareholders and induce greater disclosure of financial information only when managers are distorting disclosure policy in their self-interest. To the extent that managers distort disclosure policy and outside directors play a role in mitigating these distortions, we expect to find that firms with greater outside directors have greater propensity to issue forecasts, issue more specific and accurate forecasts and forecasts that are less optimistic. Further, as in the case of institutions the presence of a monitoring mechanism should enhance the 14

Research shows that firms with outsider-dominated boards are more likely to participate in major restructuring events such as mergers, takeovers and tender offers (Lin (1996)) and are more likely to remove poorly performing CEO’s (Weisbach (1988)) and nominate outside CEOs (Borokhovich et al. (1996)). Rosenstein and Wyatt (1990) documented that shareholder wealth increases with the addition of outsiders to the board, while Cotter, Shivdasani, and Zenner (1997) provided evidence to indicate that outside directors enhance shareholder wealth during tender offers. 15 Consistent with these arguments, Yermack (1996) and Bhagat and Black (1997) failed to document an association between the proportion of independent outside directors and firm performance. 13

credibility of the information being disclosed by the firm. However, as discussed earlier the governance role of outside directors and the extent to which they represent shareholder interests could be influence by their own incentives (including personal allegiances, board culture, fear of litigation, and reputation costs).

To the extent that the directors’ own incentives affect their

ability to act on behalf of the shareholders, it would weaken our results.

3. Methodology Sample Selection and Description We obtain management earnings forecast data from the Corporate Investor Guidelines (CIG) database, maintained by First Call, Inc. The CIG database covers the period from mid 1994 to mid 2001 and has over 26,000 observations, including earnings as well as non-earnings forecasts made by companies prior to the official release of reported earnings. It includes point forecasts, range forecasts, open ended forecasts as well as qualitative forecasts (such as “comfortable with analyst expectations”).16

16

We performed a small-sample test of the comprehensiveness of the CIG database. We picked the month of June 2000 for our test. The reason we picked June was that a larger number of forecasts are likely to occur in the last month of a quarter. The CIG database has 211 forecasts made in June 2000. We searched Factiva (formerly known as the Dow Jones News Retrieval) using the keywords "expects earnings", "expects income", "expects losses", "expects profits", "expects results" and three similar lists with first words "forecasts", "predicts" and "sees". These keywords are consistent with those used by Baginski et al. (2002). The search turned up 84 forecasts made in June 2000. Of these 84, 68 appeared in our First Call database while 16 were missing. There appeared to be 143 forecasts that were on the CIG database but did not appear in our search. We then carried out a reverse search using CIG as our source, to identify reasons why our search missed a large number of forecasts. We found that forecast articles included terms like “expects next year’s earnings” or “expects year 2000 income” etc. To correct for these cases we carried out a new search using the keywords "expects w/5 earnings", "expects w/5 income", "expects w/5 losses", "expects w/5 profits", "expects w/5 results" and three similar lists with first words "forecasts", "predicts" and "sees". However, due to the weaker constraint and the resulting large number of hits we restricted our sampling this time to the week of 18th June 2000. CIG has 65 forecasts in the sample for the week. Our Factiva search yielded 53 forecasts. Of these 37 were common to both. 16 forecasts were missing from the CIG database. 28 forecasts that appeared on CIG were missing in the search. Most of the 28 consisted of forecasts where the company made statements like: ‘Staar Surgical Co. (STAA) called current analysts' earnings estimates of 4 cents a share for its fiscal third quarter "accurate."’ or ‘clothing manufacturer Guess? Inc. on Friday said it remains comfortable with securities analysts' current earnings estimates for the second quarter and the balance of 2000.’ 14

Panel A of table 1 provides a summary of the screens applied to identify the primary management forecast sample used in our tests. Initially we selected all earnings forecasts made before the fiscal period end (i.e., we exclude pre-announcements17) for the period 1997-2000. Year 2000 is the last complete year for which management forecast data was available. We ignored forecasts made prior to 1997 for two reasons. First, the number of usable forecasts is substantially lower for years prior to 1997. It is not clear whether this is solely due to fewer forecasts made during this period, or whether the data collection in the earlier years was less comprehensive. Second, for tests of the probability of occurrence we match the management forecast sample with all other (non-forecast) firms (for which the requisite data could be obtained) by fiscal period. The sample was restricted to four years in an effort to reduce problems of interdependence of observations that arise from pooling financial data for the same firm over multiple years. To further reduce problems of interdependence of data, if a company made multiple forecasts during a fiscal period, we retained only the latest forecast. The sample of management earnings forecasts was then matched with First Call data on financial analyst following. Companies that did not have valid ticker symbols and/or were not followed by at least three analysts were dropped. The observations that satisfied the above criteria were then matched with corporate governance data collected from Compact Disclosure. This database provides information on stock ownership collected from Spectrum and information on officers and board of directors collected from proxy statements. The June CDROM for each of the years 1997 to 2000 was examined to obtain institutional ownership and outside directors’ information. The ownership data in these CDROMs represent holdings as of March 1 of each

The only reasonable way we could find them was to reverse search using CIG as our source. This test suggests that CIG is a comprehensive source of forecasts and does better than a normal search. 17 We define pre-announcements as management forecasts issued subsequent to the fiscal period end, but prior to the actual earnings announcements. 15

year.

Every forecast observation was matched with the institutional ownership as of the

immediately preceding March. Thus a forecast made in between March and December 2000 is matched with institutional ownership as of March 2000, while forecasts made in January or February 2000 were matched with institutional ownership as of March 1999. Data on officers and board of directors were based on the latest available proxy statement included in the June CDROMs. Lastly, some observations were lost because of lack of data for control variables (obtained mainly from COMPUSTAT).

The final sample comprised of 1,703 annual

management earnings forecasts (830 point forecasts) and 3,163 quarterly forecasts (908 point forecasts).18 Panel B of table 1 provides a year-by-year analysis of the sample. The table reveals that the number of management forecasts issued has been increasing over time. Analysis of the quarterly forecasts data reveals that in each year the highest number of forecasts occurs in the fourth quarter. [Insert Table 1] Research Design: The specification of the estimation regression we use to evaluate the effect of governance on the forecast occurrence is as follows: OCCURt = f (Corporate governance variables; control variables)

18

Eq. (1)

The actual number of observations used in the regressions varies. For tests of occurrence the management forecast sample was matched with all other firms for which First Call, governance and other financial data were available, resulting in 7,168 observations for tests based on annual forecasts and 24,749 observations for tests based on quarterly forecasts. Tests of forecast specificity were based on 1,689 and 3,011 observations for tests based on annual and quarterly forecasts respectively (some of the original forecast sample firms were missing analyst following information prior to the earnings forecast). For tests of forecast frequency, the number of forecasts made during 1997-2000 was matched with four-year averages of governance and control variables; some observations were lost in the process resulting in 1,098 observations for annual forecasts and 1,256 observations for quarterly forecasts. 16

where, OCCUR = binary variable taking the value “1” if the firm issued an earnings forecast during the fiscal period; “0” otherwise. As discussed earlier, prior work has examined the effect of disclosure on institutional ownership (Healy et al. (1999) and Bushee and Noe (2000). The extant story in the literature is that disclosure practices influences institutional ownership. However, the link between ownership and disclosure could be endogenous. While ownership decisions are influenced by a firm’s disclosure policy, it is also likely that a firm’s disclosure policy is influenced by its institutional ownership. Thus disclosure leads to future institutional ownership, which in turn could lead to future disclosure and so on. We examine this endogenous link between governance and the propensity to disclose by adopting a Granger type lead-lag approach as well as a simultaneous equation analysis. The specification of the lead-lag regression is as follows: OCCURt = f (INST; OUTDIR; OCCURt-1; control variables)

Eq. (2)

where, OCCUR = binary variable taking the value “1” if the firm issued an earnings forecast during the fiscal period; “0” otherwise. INST = percentage of the company’s common stock held by institutions. OUTDIR= percentage of the board of directors that are not also officers of the firm. In this specification INST lags the dependent variable (OCCURt). The variable OCCURt-1 in turn lags INST. The purpose of this specification is to isolate the incremental explanatory power of INST after controlling for the potential effect of prior disclosure on INST and future disclosure. 19 We also carry out a simultaneous equations estimation of the form OCCUR = f (Corporate governance variables; control variables) 19

Eq. (3a)

See Hamilton (1994) pages 304-305 for a description of econometric tests for Granger causality. 17

INST = f (OCCUR; control variables)

Eq. (3b)

If institutional ownership induces disclosure, the coefficient for INST in equation (3a) should be positive. The estimation of the system of equations was performed by first regressing each endogenous variable on all exogenous variables (instruments).

In the second stage,

equations (3a) and (3b) were separately estimated with the right side endogenous variable replaced by its fitted value from the first stage regression.20 In the above specifications, our disclosure variable takes on a value of 1 if the firm issued a forecast in a given period and 0 otherwise. (similar to the 2SLS but not the same since one of the equations require a probit analysis). For tests of forecast occurrence a firm that issues multiple forecasts and one that issues just a single forecast in the period are treated the same. Similarly, a firm might issue just one forecast in our sample period while others might be more consistent in their disclosure policy. Sporadic occurrence of forecasts could be attributed to managerial opportunism, as opposed to a consistent disclosure policy induced by governance. If good governance is inducing disclosure, we should find an association between the number of forecasts that a firm issued in our sample period and the governance variables. This would lend additional support to the results from the occurrence specification. We measure FREQ as the number of forecasts issued by a firm in our sample period. FREQ = f (Corporate governance variables; control variables)

Eq. (4)

To evaluate the effect of governance variables on the quality of earnings forecasts issued by management we focus on the specificity, accuracy and optimism of the forecasts. SPECIFIC = f (Corporate governance variables; control variables)

20

Eq. (5)

See Maddala (1983) page 244 for a discussion of this issue. Since OCCUR is a binary variable, (3a) was estimated using probit both in the first and second stage. 18

where, SPECIFIC = ordinal variable taking the value “3” if the firm issued a point forecast during a fiscal period; “2” if the firm issued an interval forecast; “1” for open-ended forecasts; and “0” for qualitative forecasts. ERROR = f (Corporate governance variables; control variables)

Eq. (6)

where, ERROR = abs (forecasted EPS – actual EPS) / price at the beginning of the fiscal period. BIAS = f (Corporate governance variables; control variables)

Eq. (7)

where, BIAS = (forecasted EPS – actual EPS) / price at the beginning of the fiscal period. In addition, we posit that governance mechanisms can improve the credibility of a firm’s disclosure. Empirical work evaluating the credibility of management forecasts is limited. Prior work has used dispersion or coefficient of variation of analysts’ forecasts as a proxy for lack of consensus (Swaminathan (1991); Lang and Lundholm (1996); Botosan and Harris (2000)). A more credible forecast should result in greater consensus amongst analysts leading to greater convergence and lower dispersion. We therefore use the reduction in standard deviation of analysts’ forecasts as a proxy for credibility. Williams (1996) finds that analysts use their experience with prior management forecasts in evaluating current management forecasts.21 Consistent with Williams (1996) we use the extent to which analysts revise their forecast subsequent to a management forecast as an alternative proxy for credibility. If governance mechanisms do enhance firms’ credibility, then the decrease (increase) in uncertainty (consensus) subsequent to a management forecast should be attributable to governance variables, ceteris paribus. 21

Hirst, Koonce, and Miller (1999) and Tan et al. (2002) in surveying analysts find that the confidence in their forecasts is influenced by the specificity of the management forecasts preceding the forecasts. 19

CREDIBILITY = f (Corporate governance variables; control variables)

Eq. (8)

where, CREDIBILITY is measured one of two ways: REVUNCERT = (standard deviation following a management forecast minus standard deviation prior to the forecast)/ consensus analyst forecast prior to the management forecast. REVCONS = (analyst consensus following a forecast minus analyst consensus prior to the forecast)/price at the beginning of the fiscal period. Measures of Corporate Governance To determine the association between the information disclosure environment and corporate governance, we specify two widely used measures: percentage of institutional ownership, and the proportion of the board consisting of outsiders: INST

= percentage of the company’s common stock held by institutions.

OUTDIR

= percentage of the board of directors that are not also officers of the firm.

In addition, we carry out our analyses by refining the INST variable based on the institutions investment horizon and/or their ability to generate private benefits. We refine our variables using two systems. To evaluate the effect of concentrated ownership on institutional incentives we calculate two measure of blockholding: 22 INST5

= percentage of company’s common stock held by the five largest institutional owners of the firm.

22

Similar variables have been used in the prior literature (e.g., Brickley et al. (1988); Agrawal and Mandelker (1990); Baysinger, Kosnik, and Turk (1991)). 20

INSTCONS by:

= Herfindahl index of institutional ownership concentration, as measured

 shares held by institution i   ∑  i =1 total shares outstanding  N

2

We also partition the INST variable based on Bushee (1998) into dedicated institutions, quasi-indexers and transient investors.23 Control Variables

We selected several additional independent variables to control for other possible determinants of the properties of management forecasts based on prior research. These variables are summarized and described in more detail in Appendix A. LMVAL – log of the market value of a firm’s common equity at the beginning of the fiscal period. We include LMVAL to control for firm size (i.e., inverse of risk). Prior literature has consistently shown evidence supporting the positive association between firm size and management earnings forecasts (e.g., Kasznik and Lev (1995)). AUDIT – “1” if the company is audited by one of the big 5 auditors, “0” otherwise. Auditor reputation could also be a factor in disclosure decisions. Thus, prior research indicates that firms using big five auditors tend to have better disclosure (Lang and Lundholm (1993)) and hence AUDIT is included to control for this. NUMEST – number of analysts following the firm. Prior research (Lang and Lundholm (1993); Lang and Lundholm (1996)) has documented an association between corporate disclosure quality and the number of analysts following a firm. LITIGATE – 1 for all firms in the biotechnology (2833-2836 and 8731-8734), computers (3570-3577 and 7370-7374), electronics (3600-3674) and retailing (5200-5961) industries and 0

23

We are very grateful to Brian Bushee for generously providing us the partitioned institutional ownership dataset. See Bushee (1998) for a description of the partitioning process. 21

otherwise. Francis et al. (1994) argue that firms in certain industries face a greater litigation exposure. We control for litigation exposure by using an indicator variable for firms belonging to those industries.24

MKBK - ratio of market value to book value at the beginning of the fiscal period. We use the variable, MKBK, as a proxy for proprietary costs (Bamber and Cheon (1998)). In keeping with Bamber and Cheon (1998) we also use an alternative proxy (sales concentration) to measure proprietary costs. LOSS - “1” if the firm reported losses in the previous period or median analyst forecast for the current year is negative; “0” otherwise. Prior research suggests that earnings is less value-relevant for loss making firms (Hayn (1995)) and that meeting or beating financial analyst expectation is less important for these firms (Degeorge, Patel, and Zeckhauser (1999)). Matsumoto (2002) finds that firms with losses are less likely to guide analyst forecasts downward. In keeping with Matsumoto (2002) and Choi and Ziebart (2000), we include a LOSS variable as a control variable. HORIZON - number of days between the forecast date and the fiscal period end date. SURPRISE - abs(forecated eps-median analyst forecast)/price at the beginning of the fiscal period DISPFOR - standard deviation of analyst forecasts divided by median forecast. Ajinkya

24

Some studies have used a similar variable TECH which takes the value 1 for all firms in the biotechnology (28332836 and 8731-8734), computers (3570-3577 and 7371-7379) and electronics (3600-3674) industries and 0 otherwise. Kasznik and Lev (1995) argue that high tech companies may be motivated to disclose more than firms in other industries in an effort to reduce litigation risk. O'Brien and Hodges (1991) document that, during the period 1988-91, 30 percent of shareholder lawsuits were filed against high tech firms although these firms represent only about 10 percent of the COMPUSTAT firms. While high tech firms could be more susceptible to litigation risk and may have incentives to disclose more information to avoid this possibility, these firms could also be more likely to have high proprietary costs. We carried out our analysis replacing LITIGATE with TECH and our conclusions relating to the governance variables were unchanged by this alternative specification. 22

and Gift (1984) argue that firms tend to disclose information when there is large information asymmetry in the market. DISPFOR is included to control for the effect of dispersion in analysts’ forecasts on management disclosure choices. NEWS - 1 if the current period EPS is greater than or equal to the previous period EPS; 0 otherwise. Baginski et al. (2002) find that the sign of the random walk differences in earnings are significant determination of forecast occurrence and precision. CHEARN - change in earnings deflated by stock price. Baginski et al. (2002) also find that the magnitude of the random walk differences in earnings are significant determinants of forecast occurrence. DE – long term debt to stockholders equity. We include this variable in the institutional ownership regression to proxy for risk.

Higher levels of leverage have been found to be

associated with higher levels of institutional ownership (Bushee and Noe (2000)). BETA – equity beta for the fiscal period. This variable represents another proxy for risk (Bushee and Noe (2000)) YIELD – dividend yield for the fiscal period. We use this variable in the institutional ownership regression to capture the effect of performance upon which the institutions might make ownership decisions (Bushee and Noe (2000)). DPOS – 1 if the current period EPS is positive and 0 otherwise. We use this as an alternative proxy for firm performance based on which institutions make ownership decisions. LIQUIDITY – log(trading volume/shares outstanding). We include this variable in the institutional ownership regression. It is expected to control for an institutions preference for more liquid stocks (Bushee and Noe (2000)). SNP – 1 if the company for part of the S&P 500, 0 otherwise. This variable is included

23

in the institutional ownership regression to capture preference for S&P 500 stocks (Bushee (2001)) SURP – (management forecast-median analyst forecast)/price at the beginning of the fiscal period.

Williams (1996) finds this variable to be positively associated with analyst

revision. [Insert Table 2] In order to perform the tests we merged our management forecast sample with all other (non-forecast) firms for which we could get requisite financial data. Control firms were matched with the forecasting firms by fiscal period end. The dependent variable, OCCUR, is a binary variable, hence we estimated equation (1) with a probit model. For the dependent variables, SPECIFICITY, which has four ordinal values, we used an “ordered” probit model on equation (5).

OLS was used for estimating equation (4) for frequency, equation (6) for accuracy

(dependent variable ERROR), and for BIAS in equation (7). Finally, the OLS model was used for estimating credibility (dependent variable CREDIBLE in equation (8)).

4. Results Descriptive Statistics

Summary descriptive statistics for selected variables are provided in panels A and B of Table 3. Panel A provides summary statistics for variables based on the annual forecast sample. ERROR, BIAS, HORIZON and SURPRISE are used in regressions based on point forecasts only, so summary statistics for these variables are based on the sample of 1,019 observations. Sample statistics for the other variables are based on the sample of 2,448 observations. Panel B of Table 3 provides summary statistics of variables based on the quarterly forecast sample of

24

4,370 quarterly earnings forecasts (ERROR, BIAS, HORIZON and SURPRISE are based on 1,215 observations). Median market value of equity is about $783 million for the annual forecast sample and about $650 million for the quarterly forecast sample. Median institutional ownership is about 55 (54) percent for the annual (quarterly) forecast sample. The average number of analyst following for the annual forecasters is 7.8 while that for the quarterly forecasters is 7.2. [Insert Table 3]

Variables OCCUR, SPECIFIC and BIAS are not reported in table 3 because, for these variables, only the frequency is meaningful. For OCCUR, the frequency of the management forecast sample is 2,448, while it is 10,251 for the control (no-forecast) sample for the annual forecast data (analogous frequencies are 4,370 and 41,304 for the quarterly data). In case of the variable SPECIFIC, the relative annual frequencies are: 1,097 point forecasts, 657 range forecasts, 530 open ended forecasts, and 49 other forecasts (analogous quarterly frequencies are 1,152 point, 1,318 range, 1,220 open ended, and 179 other). When management forecasts of annual EPS are greater than actual EPS (i.e., BIAS = 1), the related frequency is 439, while for the complementary case (when BIAS = 0) the frequency is 580 (again, the analogous frequencies are 365 and 850 for quarterly forecasts, respectively).

Governance and the Occurrence of Management Forecasts

Table 4 presents results of tests examining the link between governance and forecast occurrence. Columns 3 and 7 which provide results of tests based on annual and quarterly forecasts respectively show that the probability of occurrence of a management earnings forecast is, as expected, positively associated with the two main governance variables, INST and

25

OUTDIR, and the coefficients are statistically significant at the 0.01 level. All the control variables in these regressions are in the expected direction and significant (expect DISFOR and LITIGATE in the quarterly regression). These results are consistent with the argument that active monitoring by institutions and outside directors leads to enhanced disclosure of management forecasts. However, as discussed earlier an alternative explanation of the finding relating to institutions is that they may prefer to invest in companies with transparent disclosure (Bushee and Noe (2000)), which suggests a reverse causality. We therefore carry out further analysis by including a lagged occurrence variable (OCCURL) in the regression as well as a simultaneous regression analysis (using a modified two stage approach). Columns 4 and 8 provide annual and quarterly regression results of governance effects after controlling for past forecast occurrence. This variable should control for determinants of OCCUR, including the extent that past disclosure activity influence future institutional ownership and future disclosure. The coefficient of OCCURL is positive and highly significant suggesting that past disclosure is a good indicator of future disclosure. However, the governance variables continue to provide significant explanatory power suggesting that institutional ownership explains future disclosure after controlling for the correlation between institutional ownership and past disclosure.

The

Granger test comparing the restricted and unrestricted sum of squares residuals rejected the null hypothesis of INST=0 (p-value