THE IMPACT OF GOVERNANCE MECHANISMS ON INSIDERS GUIDANCE OF ANALYSTS AND MANAGEMENT OF EARNINGS SURPRISES

Transcription

1 THE IMPACT OF GOVERNANCE MECHANISMS ON INSIDERS GUIDANCE OF ANALYSTS AND MANAGEMENT OF EARNINGS SURPRISES Lawrence D. Brown J. Mack Robinson Distinguished Professor of Accountancy Georgia State University J. Mack Robinson College of Business School of Accountancy Atlanta, GA Tel: (404) Fax: (404) Huong N. Higgins Assistant Professor Worcester Polytechnic Institute Department of Management 100 Institute Road Worcester, MA Tel: (508) Fax: (508) October 30, 2002 We have benefited from the comments of Sudipta Basu, George Benston, Jennifer Francis, Don Herrmann, Christian Leuz, Stanislav Markov, Gary Meek, Grace Pownall, Wayne Thomas, the participants of the 2002 Southeastern Accounting Summer Research Colloquium, and those of a workshop at Oklahoma State University. We gratefully acknowledge Thomson Financial I/B/E/S for providing earnings per share forecast data, which is part of its broad academic program to encourage earnings expectations research.

2 THE IMPACT OF GOVERNANCE MECHANISMS ON INSIDERS GUIDANCE OF ANALYSTS AND MANAGEMENT OF EARNINGS SURPRISES Abstract We investigate the relation between governance mechanisms and insiders guidance of analysts and management of earnings surprises. Corporate insiders have incentives to manipulate short-term stock prices by creating positive earnings surprises (Levitt 1998). Insiders have two ways to create positive earnings surprises: (1) manage reported earnings upwards or (2) guide analyst forecasts downwards (Matsumoto 2002). Recent studies have shown that stronger governance suppresses insiders management of reported earnings (Leuz et al. 2002; Bhattacharya et al. 2002), suggesting that it is likely to mitigate management of earnings surprises. However, nothing is known about the impact of governance mechanisms either on insiders guidance of analysts or on their management of earnings surprises. Using data from 21countries, we show that managers of firms facing stronger governance are relatively more likely to both guide analysts forecasts downward and create positive earnings surprises. We highlight an important unaddressed agency problem in the financial reporting process by showing that institutional governance does not monitor insiders in either their guidance of analysts or their management of earnings surprises. 2

3 THE IMPACT OF GOVERNANCE MECHANISMS ON INSIDERS GUIDANCE OF ANALYSTS AND MANAGEMENT OF EARNINGS SURPRISES 1. Introduction Avoiding negative earnings surprises is entrenched in today s corporate culture. Insiders manage earnings surprises upwards [earnings surprise management] to avoid litigation, maximize the value of their compensation, and boost credibility (Bartov et al., 2002). They report earnings that meet or beat analysts estimates either by managing reported earnings upwards [earnings management] or by guiding analysts earnings estimates downwards [forecast guidance] (Matsumoto, 2002). 1 The academic literature generally focuses on earnings management as the way to manage earnings surprises (Healy and Wahlen, 1999), but the business press considers forecast guidance as crucial to the earnings surprise game (Barsky, 2002; Bleakley, 1997; Ip, 1997; McGee, 1997; Talley and Craig, 2002; Vickers, 1999). We study the impact of institutional governance on forecast guidance and earnings surprise management. The aspect of governance we consider is that instituted by the legal environments, which best explains structures of capital markets (La Porta et al., 2000a). We focus on three institutional governance mechanisms: rules of laws, system of legal enforcement, and system of accounting standards and disclosure practices (La Porta et al., 1998). We measure these mechanisms as in La Porta et al. (1998) and Leuz et al. (2002), and we add an auditor index to capture the governance role of accounting standards and disclosure practices at the firm level (Ashbaugh and Warfield, 2002; Barton and Simko, 2002; Becker et al. 1998; Francis et al., 1999; Sloan, 2001). Governance reduces the propensity of insiders to engage in earnings management by raising its relative cost (Bhattacharya et al., 2002; Leuz et al., 2002) so, ceteris paribus, it should reduce the propensity of insiders to create 1 Guidance does not suggest that analysts are duped by insiders. In order for positive surprises to recur for the same firm over time, compliance between insiders and analysts must exist. Insiders who own stock and options benefit from positive surprises via higher stock prices. Analysts who advise clients to buy stocks benefit from positive surprises because it helps them to justify their recommendations. Guidance need not be private. It can be public, such as a management forecast. 3

4 positive earnings surprises. 2 However, no research has examined the relation between governance and either forecast guidance [the other mechanism for managing earnings surprises] or earnings surprise management itself, so stronger governance may not possess this important monitoring effect on curbing the propensity of insiders to play earnings surprise games, an activity that is harmful to investors (Levitt 1988; Collingwood 2001; Eccles et al. 2001). We show that firms facing stronger governance mechanisms engage in relatively more downward forecast guidance and in upward management of earnings surprises. We contribute to the literature in several ways. We add to the governance literature by highlighting an important unaddressed agency problem in the financial reporting process. Evidence that stronger governance impedes insiders in their attempts to manipulate earnings (Bhattacharya et al. 2002; Leuz et al. 2002) suggests that, ceteris paribus, stronger governance impedes managers in their attempts to manipulate earnings surprises. Insiders use earnings surprises to reduce outsider interference and protect their private control benefits, foregoing long-term value enhancing projects (Eccles et al. 2001), such as research and development (Lev 2001). Our results reveal that stronger governance fails to provide a monitoring role for mitigating these agency problems. We reconcile diverse international evidence on management of reported earnings versus management of earnings surprises. Insiders in the U.S. are relatively more likely than those in other countries to report earnings that meet or beat analysts estimates (Brown and Higgins 2001), but reported earnings in the U.S. are relatively less likely to be managed (Leuz et al. 2002) and they are relatively more transparent (Bhattacharya et al. 2002) than reported earnings in other countries. By showing that forecast guidance is more prevalent in the U.S. than in any of 20 other countries, we reconcile these diverse findings. Our study is the first to examine forecast guidance in an international context. We show that the guidance metric proposed by Matsumoto (2002) is valid with international data, and that a simplified metric omitting past daily 2 Ball et al. (2000), Land and Lang (2002) and Fulkerson et al. (2002) show that earnings are more conservative and of higher quality in four common-law countries than they are in three code-law countries. They do not examine the link between governance and common/code law country dichotomy directly. However, the four common law countries they study have weaker governance than do the three code law countries they study based on four governance dimensions considered by Gul et al. (2000), suggesting that stronger governance impedes earnings management. 4

5 returns is adequate for detecting forecast guidance. Past daily returns are intuitively appealing for use in modeling analyst revisions because they help update stale analysts forecasts (Stickel 1989), but past research using this updating mechanism (ibid) have found little advantage to doing so. Our findings suggest that future studies on forecast guidance can safely drop the past returns component. Because this simplification of the Matsumoto (2002) metric does not require stock price returns data, it increases sample size, hence power of statistical tests. We add to the earnings surprise, earnings management, and forecast guidance literatures by showing that there exists a mechanism, namely governance, which appears to have opposing effects on the two aspects of earnings surprise management: it impedes earnings management but not forecast guidance. The extant literature has assumed that factors impacting earnings surprise management have the same directional effects both on forecast guidance and earnings management, thus on earnings surprise management (Matsumoto 2002). Our results suggest that future research should not assume that all factors have the same directional effects on both downward forecast guidance of analysts and upward management of reported earnings. We proceed as follows. Section 2 develops hypotheses. Section 3 describes our sample and methodology. Section 4 discusses results of preliminary analyses. Section 5 presents results of our main analyses. Section 6 provides sensitivity test results and section 7 concludes. 2. Hypotheses Development We consider institutional mechanisms following the legal approach, which provides the best explanation for both the structures of capital markets (La Porta et al. 2000), and the economic effects of financial accounting information (Bushman and Smith 2001). Institutional governance mechanisms are important for studying forecast guidance and earnings surprise management for two reasons. First, they help develop financial markets (La Porta et al.1997; Ball et al. 2000; Morck et al. 2000), which favorably impacts financial processes, such as financial forecasting and reporting. Specifically, a richer financial market increases incentives for insiders and analysts to cooperate to create positive earnings surprises. Second, the role of institutional governance mechanisms in investor protection and in mitigating agency problems has been highlighted in prior research (La Porta et al., 1998; 2000; 5

6 Ashbaugh and Warfield, 2002; Sloan, 2001). Worldwide, governance mechanisms indicate the extent that legal systems protect outside investors from expropriation by controlling insiders (La Porta et al., 2000b). Numerous studies have used these mechanisms to proxy for the stance of the law towards investor protection in different countries (La Porta et al., 2000; Bushman and Smith, 2001; Hung, 2001; Gul et al., 2002). Insiders have two ways to create positive earnings surprises: guide analyst forecasts downwards and manage reported earnings upwards (Matsumoto, 2002). Prior research has examined the monitoring role of governance on earnings management, but it has not examined its role on forecast guidance or on earnings surprise management. Insiders use forecast guidance to steer analysts towards a lower earnings number so that they can report earnings that beat analyst expectations (Burgstahler and Eames 2002; Matsumoto, 2002). Forecast guidance should increase with stronger institutional governance for two reasons. First, insiders are more likely to resort to guidance when the relative cost of managing surprises via management of reported earnings increases. Second, insiders are more likely to guide analysts and analysts are more willing to be guided when it is more profitable for parties to play earnings surprise games. The first reason is based on our belief that the system of law and enforcement impedes insiders ability to manage earnings more than it impedes their ability to guide analysts. 3 Because accounting reports are more structured than communications to analyst societies, misleading earnings reports are easier to monitor and prosecute than are misleading communications to analysts. The second reason is based on the economic incentives of analysts and corporate insiders. In the more developed financial markets arising from stronger institutional governance, there is more need for disclosures and developed analyst societies, making it both easier and more profitable for insiders and analysts to create positive earnings surprises. 3 The prevalent legal avenue to monitor the reporting of earnings and the application of accounting standards in the U.S. is based on Rule 10b-5 of the Securities and Exchange Act of 1934 (Sloan, 2001). U.S. legislation governing insiders communication with analysts based on the Private Securities Litigation Reform Act of 1995 and the Securities Litigation Uniform Standards Act of 1998 is only recent. More importantly, this legislation is more permissive than restrictive because they aim to make it difficult to prosecute corporations and their insiders (Casey, 1998), resulting in much fewer litigation cases based on forecast guidance (according to the Stanford Law School s Securities Clearinghouse). Furthermore, because financial accounting information is more structured than other communications to analysts (especially prior to Regulation Full Disclosure, which became effective at the end of our study), the monitoring of earnings reports should be easier relative to that of other communications to analysts. Prosecution of forecast guidance is difficult, because communications constituting cautionary 6

7 More formally, our first hypothesis (in alternative form) is: H1: Managers facing stronger institutional governance mechanisms are more likely to guide analyst forecasts downwards. Recent studies show that stronger governance decreases earnings management, suggesting that, ceteris paribus, stronger governance impedes earnings surprise management. Our first hypothesis suggests that ceteris paribus, stronger governance increases earnings surprise management, exactly the opposite that suggested by recent studies. Thus, at first glance, the a priori relation between stronger governance and earnings surprise management may seem ambiguous. However, we contend that stronger governance is likely to increase earnings surprise management. By making it more profitable for managers and analysts to cooperate with each other to create positive earnings surprises, stronger governance has the direct effect of increasing the propensity of insiders and analysts to play earnings surprise games. We believe that this direct effect, when combined with the diverse indirect effects, is likely to increase the propensity of managers to play earnings surprise games. More formally, our second hypothesis (in alternative form) is: H2: Managers facing stronger institutional governance mechanisms are more likely to manage earnings surprises upwards. 3. Methodology 3.1. Models Model to Measure Forecast Guidance For each firm i, in industry j, in country k, in year t, guidance is the unexpected portion of the earnings forecast (UEF), measured as the difference between the consensus analyst forecast (CF) and the expected analyst forecast (E [F]) for the period: UEF ijkt = CF ijkt E [F ijkt ] (1) languages to guide analysts expectations downwards are less likely to constitute scienter, a necessary cause to successful suits. 7

8 In equation (1), the expected analyst forecast (E [F]) is modeled using a random walk model (EPS from the previous period) with drift (E [D]): E [F ijkt ] = EPS ijkt-1 + E [D ijkt] (2) In equation (2), expected drift (E [D]) is estimated based on prior earnings changes, where drift is the earnings change from the previous year (equation 3), and?expected drift uses drift (D) and ending price (PRICE) for all firms in industry j, country k and year t (equation 4): D ijkt = EPS ijkt - EPS ijkt-1 (3), and E [D ijkt] = (α jkt + β jkt * D ijkt-1 / PRICE ijkt-2 ) * PRICE ijkt-1 (4) Equation (4) is estimated only when there are ten or more firms in the same country-industry-year year (two-digit SIC code for industry). All measurements are based on U.S. dollars using the exchange rate on the first day of the month after the fiscal year end. 4 The translation to a single currency for all firms within each country-industry-year is necessary for estimating regression equation (4). For convenience, we translate all firms to U.S. dollars. Our model is similar to Matsumoto (2002), except we use annual data, examine international firms, account for currency differences, and, because of the limited data availability of international stock returns data, exclude the portion of expected forecast reflected in cumulative daily excess returns after the previous earnings announcement Model to Measure Earnings Surprise Management We estimate earnings surprise management in the traditional way by subtracting the consensus analyst expectation from reported earnings and examining its sign (Bartov et al. 2002; Lopez and Rees 2002; Matsumoto 2002). If reported earnings are equal to or greater than the consensus estimate, unexpected earnings are nonnegative, so reported earnings meet (or beat) analyst estimates. Alternatively, if reported earnings are less than the consensus analyst estimate, unexpected earnings are negative, so reported earnings miss analyst estimates. More specifically, we calculate: UEPS ijkt = EPS ijkt - CF ijkt (5) 4 We use exchange rates at fiscal year end to mitigate effects of currency movements. 8

9 Where UEPS ijkt, EPS ijkt, and CF ijkt are unexpected earnings, reported earnings, and consensus earnings forecasts for firm i, in industry j, in country k, in year t Dependent and Independent Variables Forecast Guidance and Earnings Surprise Management Forecast Guidance (H1): DOWN is coded 1 if UEF from equation (1) is negative and 0 otherwise. Earnings Surprise Management (H2): MEET is coded 1 if UEPS from equation (5) is zero or positive and 0 otherwise Institutional Governance Mechanisms We use three country-specific indices based on La Porta et al. (1998) and add a firm-specific auditor index. The first country-specific index represents laws, denoted LAWS, ranges from 0 to 5 with equal weights from the six anti-director rights as in La Porta et al. (1998) and Leuz et al. (2002) to proxy for the governance role provided by the rules of law. 5 The second country-specific index represents enforcement, denoted ENF, ranges from 0 to 10 with equal weights from the three enforcement factors as in La Porta et al. (1998) and Leuz et al. (2002) to proxy for the governance role provided by legal enforcement. 6 The third country-specific index is an accounting index, denoted ACC, based on the CIFAR rating to capture the governance role of the system of accounting standards and disclosure practices. 7 We perform a factor analysis of the three country-specific factors to obtain a principal factor, denoted FAC, and present results using either one of the three individual country-specific factors or FAC. The fourth 5 Laws protecting outside investors consist of legal rules such as voting power, ease of participation in corporate voting, and legal protection against expropriation by management. Anti-director laws often proxy for rules of laws because they protect shareholders by allowing them to exercise their rights, i.e., vote for directors with ease (La Porta et al., 1998). The six antidirector rights discussed by La Porta et al. (1998) are: 1) vote by mail, 2) sell shares around date of shareholders meetings, 3) have cumulative voting for directors and/or proportional representation on the board, 4) challenge directors decisions in court and/or to force companies to repurchase shares of minority shareholders who object to certain management decisions, 5) have preemptive rights to buy new issues of stock, and 6) call an extraordinary shareholders meeting. 6 Enforcement relates to intermediary functions such as a judicial system empowering courts in law enforcement, and effective accounting systems making company disclosures interpretable and verifiable (La Porta et al., 1998). A system of legal enforcement substitutes for weak laws, for example by allowing active and well-functioning courts to step in and rescue investors abused by insiders. The three enforcement factors discussed by La Porta et al. (1998) are: 1) judicial system efficiency, 2) assessment of rule of law, and 3) corruption index. 7 CIFAR stands for Center for International Financial Analysis & Research. These data have been used in recent accounting, finance and economics studies (La Porta et al. 1998; Bushman and Smith 2002; Hope 2002). 9

10 governance variable, BIG5, is a firm-specific variable, based on whether a firm is audited by a big auditor (1) or not (0). We expect that stronger governance increases both forecast guidance and earnings surprise management Control Variables We include six control variables: (1) LCAP is the logarithm of firm size, with firm size computed as market value of total common shares in millions of U.S. dollars in the forecast month (source: I/B/E/S), (2) GROW is a firm growth index, computed by I/B/E/S as the average annualized earnings per share growth over the past five years expressed as a percent, (3) VOL is a firm volatility index, computed by I/B/E/S as the mean absolute percentage difference between actual reported earnings per share and a five-year historical EPS growth trend line, expressed as a percent of trend line earnings per share, (4) PROF is a profit index, coded 1 if earnings are positive and 0 otherwise, (5) US is an index denoting U.S. firms, coded 1 if the firm is domiciled in the U.S. and 0 otherwise, and (6) YR has the values of 1 to 10 for the calendar year of annual earnings, We include firm size because we expect insiders of large firms to be more likely to guide analysts and to manage earnings surprises as these firms typically have larger investor relations departments (facilitating guidance) and a larger portion of their compensation in the form of stock and stock options (increasing their desire to avoid negative earnings surprises). We include a growth index because growth managers have stronger incentives than value managers to manage earnings surprises (Brown 2001; Skinner and Sloan 2002), thus they are more likely to resort to both forecast guidance and earnings management to achieve their goals. 8 We include a volatility index to capture insiders ability to guide analysts and to avoid negative surprises. When earnings are volatile, it is more difficult for insiders to steer analysts towards targeted numbers because both they and analysts possess noisy 10

11 information regarding upcoming earnings. We include a profit/loss dummy because firms reporting profits are more likely to avoid negative earnings surprises (Degeorge et al. 1999; Brown 2001) so insiders are relatively more likely to resort to forecast guidance to achieve their goals. The US/non US dummy variable is motivated by findings that there is more management of earnings surprises in the U.S. than in other countries (Brown and Higgins 2001), but less management of earnings in the U.S. than in other countries (Leuz et al. 2002). Assuming the validity of these diverse findings, we expect that there is more guidance of analysts in the U.S. than in other countries. The year variable allows for the possibility that guidance and earnings surprise management has changed over our sample period. Indeed, Brown and Higgins (2001) document temporal increases in earnings surprise management in an international context so we expect YR is positive when testing our second hypothesis. Lang et al. (2002) document temporal decreases in earnings management. Assuming the validity of the diverse findings regarding temporal changes in earnings management and in earnings surprise management, we expect that YR is positive when testing our first hypothesis Data Sources and Sample Selection DOWN and MEET are based on data from I/B/E/S and Datastream. The Thomson Financial I/B/E/S International Summary File is our source of consensus forecasts, earnings and stock prices, and Datastream is our source of exchange rates. DOWN and MEET are computed using the last consensus (mean) forecast prior to the annual earnings announcements of firms with at least two preceding years of data. 9 To calculate expected drift for our measure of guidance, we restrict our sample to firms with at least ten observations in a given country-industryyear. 10 Finally, we exclude countries with less than 100 firm-year observations of computable guidance in the tenyear period, We use this rather than the analysts consensus expectation of long-term growth used by Matsumoto (2002) with U.S. data because using it with international annual data would severely restrict our sample. 9 As shown in equations (1)-(4), computation of DOWN requires earnings per share of years t and t-1, and price per share of year t-1 and t Expected drift is obtained from the regressions shown in equation (4). Following Matsumoto (2002), we retain observations with sufficient data to estimate the regressions. 11

12 LAWS, ENF and ACC are based on data from La Porta et al. (1998). The data for BIG5 are obtained from Worldscope. 11 LCAP, GROW and VOL are obtained from I/B/E/S. DOWN is computed for 43,360 observations. We use this sample for our preliminary analyses. Our logit model requires data availability for all variables so our main analyses use 35,785 observations. 4. Preliminary Analyses We conduct analyses to validate DOWN as a measure of forecast guidance. These analyses are important for several reasons. First, we do not strictly adhere to Matsumoto s procedures so we wish to validate the reliability of our proxy. Second, our study is the first to examine forecast guidance using international earnings data so we wish to assess the validity of our metric for this purpose. Third, the construct validity of the Matsumoto (2002) metric has only been examined in a single context (see her table 7), so we wish to enhance validity by examining its performance in numerous contexts. Fourth, if the diverse findings by the extant literature that U.S. managers are relatively more likely than non-u.s. managers to manage earnings surprises upwards but less likely to manage reported earnings upwards are valid, and if DOWN is a valid proxy for measuring guidance, we should find more downward forecast guidance in the U.S. than in other countries. If we obtain such a finding, the validity of the metrics used by both us and the extant literature to measure earnings surprise management, earnings management and forecast guidance is increased Meet or Beat Estimates Matsumoto (2002) examines whether guidance facilitates meeting or beating analyst consensus earnings estimates. She finds that guidance exists 54.12% of the time when U.S. managers report quarterly earnings that meet or beat analyst earnings estimates, but only 49.23% of the time when reporting quarterly earnings falling short of estimates. Her chi-square statistic of is significant at the 0.001significance level. We conduct similar analysis in panel A of Table 1. When international managers report annual earnings meeting or beating analyst estimates, forecast guidance exists 58.29% of the time, but when they report annual earnings falling short of 11 Worldscope names the individual accounting firms. Firms whose auditors constitute the big five accounting firms today are 12

13 analysts estimates, forecast guidance exists only 49.52% of the time. Our chi-square statistic of is significant at p-value <0.0001, indicating that forecast guidance facilitates meeting or beating analyst estimates in an international context. Insert Table 1 about here 4.2 Profits versus Losses The impact of earnings announcements on valuation is more pronounced for profitable firms (Hayn, 1995), so these firms managers have more incentives to manage earnings surprises (Degeorge et al. 1999; Brown, 2001). If managers do use forecast guidance to manage earnings surprises, forecast guidance should be more evident for profitable firms. Panel B of Table 1 shows the results of a contingency analysis of DOWN partitioned into firms whose earnings are positive (profits) versus those whose earnings are not positive (losses). DOWN is coded 1 in 55.50% of profit versus 49.27% of loss firms. The chi-square statistic of is significant at p-value <0.0001, revealing that forecast guidance applies relatively more to profitable firms Profit Firms: Small Positive versus Other Surprises The patterns of earnings surprise management differ between profit and loss firms (Degeorge et al. 1999). Managers attempt to create small positive surprises for profit firms and avoid extreme negative surprises for loss firms (Brown, 2001). We focus on profit firms in this subsection and loss firms in the next one. If managers do guide analyst forecasts downwards to create small profit surprises (Burgstahler and Eames 2002), DOWN is more likely to equal one for small positive profit surprises than for other profit surprises. Panel C of Table 1 provides results of a contingency analysis of DOWN for firms partitioned into small positive versus other profit surprises. Similar to Brown and Higgins (2001), we define earnings surprises as actual minus predicted earnings, divided by the absolute value of actual earnings, and we define surprises within 5% of reported profits as small. 12 DOWN is coded 1 in 60.34% of small positive versus 53.21% of other profit surprises. The chi-square statistic of is coded Big5 =1. All other accounting firms (including missing observations) are coded Big5 = Panels B to D omit earnings equal to zero. Thus, panel B has 105 fewer observations than Panel A. All other tables include earnings equal to zero. 13

14 significant at p-value <0.0001, revealing that forecast guidance is more probable for profitable firms reporting small positive surprises than for profitable firms reporting other surprises Loss Firms: Extreme Negative versus Other Surprises If managers do use forecast guidance to manage loss surprises, DOWN is less likely to equal one when negative loss surprises are extreme than for other loss surprises. Panel D shows findings for a contingency analysis of DOWN for loss firms partitioned into extreme negative versus other surprises. Similar to Brown and Higgins (2001), we consider surprises greater than 100% of reported losses to be extreme negative surprises. DOWN is equal to 1 in 38.71% of extreme negative surprises versus 52.80% of other surprises. The chi-square statistic of is significant at p-value <0.0001, showing less forecast guidance when negative loss surprises are extreme Forecast Guidance in the U.S. versus 20 Other Countries Table 2 shows mean guidance (DOWN ratio) in the U.S. and in 20 other countries. 13 We compute DOWN for 43,360 forecasts of 10,659 firms from the U.S. and another 20 countries (in alphabetical order): Australia, Canada, France, Germany, Greece, Hong Kong, India, Italy, Japan, Korea, Malaysia, Norway, South Africa, Spain, Sweden, Switzerland, Taiwan, Thailand, Turkey, and the U.K. The U.S. has the largest number of firms (5,283) and observations (21,799), over 50 percent of the total observations. Seven other countries have at least 200 firms: Japan (1,646), the U.K. (1,244), Korea (584), Taiwan (259), Thailand (245), Germany (216), and Canada (200). Eight other countries have 500 observations or more: the U.K. (6,556), Japan (6,140), Korea (2,429), Taiwan (927), Thailand (818), Germany (643), Canada (727), and Australia (576). The U.S. DOWN ratio (60.54%) is larger than in all 20 other countries. We undertake three statistical tests to examine if this is a significant result: a binomial test, a means test, and a chi-square test. All three tests reveal significance at the p = level (one-tailed test). In sum, the evidence in Tables 1 and 2 suggests that DOWN is a valid proxy for forecast guidance in an international context. 13 The four non-u.s. countries most likely to guide forecasts downwards are the U.K. (58.92%), Sweden (57.42%), Australia (56.42%), and Canada (56.40%). Five countries are least likely to guide forecasts downwards: Korea (31.87%), Germany (36.08%), France (42.78%), Turkey (43.27%), and Thailand (43.89%). 14

15 Insert Table 2 about here 5. Main Analyses and Results Table 3 shows the Pearson and Spearman correlations amongst forecast guidance, earnings surprise management and our independent variables, with the Pearson (Spearman) correlations presented above (below) the diagonal. Consistent with the notion that the probability of guidance increases with stronger governance, DOWN is positively related to LAWS, ENF, ACC, FAC and BIG5 (p-value <0.0001). Consistent with the notion that guidance is more likely for larger firms and for growth firms, DOWN is positively related to LCAP and GROW (p-value <0.0001). Consistent with our expectation that guidance is less likely to pertain to firms with volatile earnings, DOWN is negatively related to VOL (p-value <0.0001). Also as expected, DOWN is positively related to both PROF and US (p-value <0.0001). Unexpectedly, DOWN is negatively associated with YR using a Spearman correlation (p-value <. 0001), but the negative relation is insignificant using a Pearson correlation. Consistent with our expectation that the propensity to avoid negative earnings surprises increases with stronger governance, MEET is positively related to LAWS, ENF, ACC and FAC (p-value <0.0001). As expected, the relation between MEET and BIG5 is positive, albeit insignificant. Consistent with the notion that larger firms and growth firms are relatively more likely to avoid negative earnings surprises, MEET is positively related to LCAP and GROW (p-value <0.0001). Consistent with our expectation that firms with volatile earnings are less likely to avoid negative surprises, MEET is negatively related to VOL (p-value <0.0001). As expected, MEET is positively related to PROF and US (p-value <0.0001). Finally, as expected, MEET is positively related to YR (Pearson correlation p- value <0. 02; Spearman correlation p-value < 0. 08). Insert Table 3 about here The positive correlations among the three country-specific governance variables, LAWS, ENF and ACC, are high, as they all capture related aspects of governance at the country level. Due to these high positive correlations, we use only one of them at a time in our logit analyses. We also use FAC, the principal factor in a factor analysis of 15

16 the three country-specific variables, in lieu of the individual country-specific factors, in our analyses. Each of our analyses has a country-specific governance variable, and BIG5, a firm-specific variable Hypothesis 1 Table 4 contains results of the following logistic regression: DOWN = a 0 + a 1 GM + a 2 BIG5 + a 3 LCAP + a 4 GROW + a 5 VOL + a 6 PROF + a 7 US+ a 8 YR + ε (6) where: 14 DOWN is forecast guidance, coded 1 when UEF from equation (1) is negative and 0 otherwise. GM is country-specific governance mechanism, which is, alternatively, LAWS, ENF, ACC or FAC. We expect each variable to be positively associated with DOWN. BIG5, a firm-specific governance mechanism via auditors scrutiny, is coded 1 if the firm s auditors are among the big 5 auditors, and 0 otherwise. We expect BIG5 to be to be positively associated with DOWN. LCAP is the logarithm of firm size, with firm size measured as market values of total common shares in millions U.S. dollars. We expect it to have a positive sign. GROW is earnings growth, measured by the I/B/E/S growth index. We expect it to be positively associated with DOWN. VOL is earnings volatility, measured by the I/B/E/S volatility index. We expect it to be negatively related to DOWN. PROF is coded 1 if earnings are positive (a profit), and 0 otherwise. We expect it to have a positive sign. US is coded 1 if U.S., and 0 otherwise. We expect it to have a positive sign. YR is 1 to 10, for the calendar year of the earnings date, We expect it to have a positive sign. Insert Table 4 about here Consistent with hypothesis 1, both GM and BIG5 are positive (p-value <0.0001). This finding prevails regardless of whether LAWS, ENF, ACC or FAC is used as the country-specific governance mechanism. With the 16

17 exception of LCAP, which always has its expected sign but is insignificant in one case, all of the control variables have their expected signs and are significant regardless of whether LAWS, ENF, ACC or FAC is used. More specifically, GROW is positive (p-value <0.0005), VOL is negative (p-value <0.0001), and PROF, US and YR are positive (p-value <0.0001). The guidance model is significant regardless of whether LAWS, ENF, ACC or FAC is used as the country-specific governance mechanism (p-value <0.0001) Hypothesis 2 Table 5 contains results of the following logistic regression: MEET = a 0 + a 1 GM + a 2 BIG5 + a 3 LCAP + a 4 GROW + a 5 VOL + a 6 PROF + a 7 US+ a 8 YR + ε (7) Consistent with hypothesis 2, both GM and BIG5 are significantly positive (p <.0001 for GM; p <.04 for BIG5). This finding prevails regardless of whether LAWS, ENF, ACC or FAC is used as the country-specific governance mechanism. All of the control variables have their expected signs and are significant regardless of whether LAWS, ENF, ACC or FAC is used as the country-specific governance mechanism. More specifically, LCAP and GROW are positive (p-value <0.0001), VOL is negative (p-value <0.005), and PROF, US and YR are positive (p-value <.0001). The earnings surprise management model is significant regardless of whether LAWS, ENF, ACC or FAC is used as the country-specific governance mechanism (p-value <0.0001). Insert Table 5 about here 6. Sensitivity Analyses 6.1. Analyses by Year and Country Our tests assume that our pooled temporal, cross-sectional observations are independent. We chose this procedure because it maximizes sample size, but it overstates our degrees of freedom. 15 To mitigate the overstated degrees of freedom problem, we examine the frequency of correct signs of our governance variables after using the Fama-MacBeth (1973) method of running our logit models in each of 10 years and for each of 21 countries. More 14 The reported results are based on winsorizing continuous firm-specific variables at their 1 st percentiles (min) and 99 th percentiles (max) to mitigate undue influence of extreme observations. 17

18 specifically, we run the models in Tables 4 and 5 for each year from 1991 to 2000 after removing the variable YR, and we determine in how many years FAC and BIG5 have their expected coefficients. We also run the models in Tables 4 and 5 for each of 21 countries after removing the variable FAC, and we determine in how many countries the BIG5 variable has its expected coefficient. An analysis of the forecast guidance model reveals that FAC has its expected positive coefficient in 7 of 10 years, and that BIG5 has its expected positive coefficient both in 5 of 10 years and in 13 of 21 countries. An analysis of the managing earnings surprises model reveals that FAC has its expected positive sign in all 10 years, and that the BIG5 variable has its expected positive sign in 8 of 10 years and in 11 of 21 countries. Overall, our hypotheses are validated in the majority of years and countries we examine. 6.2.U.S. versus non-u.s. Firms U.S. observations constitute nearly half of our data. To assess if U.S. data drive our main results, we replicate our main analyses after omitting U.S observations. Panels A and B of Table 6 replicate Tables 4 and 5 results for non-u.s. firms using FAC as the country-specific governance variable. Panel A shows that both FAC and BIG5 are significantly positive in the forecast guidance model (p-value < and p-value = respectively). All control variables have their expected signs; with the exception of LCAP, they all are significant and at the 0.11 level or better. The forecast guidance model for non-u.s. firms is significant at p-value < Panel B shows that FAC and BIG5 are significantly positive for the non-u.s. domiciled firms for the earnings surprise management model (p-value < and p-value = respectively). All the control variables have their expected signs, and with the exception of VOL, they all are significant and at < level. The earnings surprise management model for non-u.s. firms is significant at p-value < In sum, our results pertain to firms domiciled outside the U.S. Insert Table 6 about here 6.3.The impact of changes in BIG5 on changes in forecast guidance and earnings surprise management 15 Nevertheless, since our p-values are less than.0001 for most variables, our findings are significant even if our test statistics 18

19 Having shown that stronger governance increases forecast guidance and earnings surprise management in a cross-sectional analysis, we now examine if stronger governance increases both forecast guidance and earnings surprise management in a temporal analysis. If firm-specific governance becomes stronger over time, both forecast guidance and earnings surprise management should increase. We examine this issue by focusing on changes in BIG5, our only firm-specific governance variable. Governance strengthens over time when firms switch from a small auditor to a BIG5 auditor (coded 1); it weakens when firms switch from a BIG5 auditor to a small auditor (coded -1), and it remains the same for firms that do not switch (coded 0). Panels A and B of table 7 respectively present results for year to year changes in forecast guidance and earnings surprise management regressed on the independent variables used in Tables 4 and 5 in change form. 16 The dependent variables are coded +1, 0, or -1 based upon the year-to-year differences. DOWN = 1 if analyst forecasts are guided downwards in the current year but not the previous year; DOWN = 0 if analyst forecasts are guided downwards in both years or in neither year; DOWN = -1 if analyst forecasts are guided downwards in the previous but not the current year. The logit model has two intercepts because the dependent variable can take on one of three variables. Panel A shows that BIG5 is significantly positive in the DOWN model (p-value < ). All of the control variables have their expected signs, and they all are significant and at the 0.03 level or better. The DOWN model is significant at p-value < Panel B shows that BIG5 is significantly positive in the MEET model (pvalue = ). All of the control variables have their expected signs and are significant and at the 0.03 level or better. The MEET model is significant at p-value < In sum, the evidence is consistent with our two hypotheses using either cross-sectional or temporal analyses. Insert Table 7 about here 7. Conclusions are inflated a thousand-fold. 16 FAC and YR are omitted because there are no temporal changes in FAC and all temporal changes in YR equal zero. Results are similar when YR (rather than change in YR) is included. 19

20 We investigate the impact of governance mechanisms on corporate insiders propensity to guide analysts and manage earnings surprises. Managing reported earnings and guiding analysts are the two ways that insiders use to manage earnings surprises (Matsumoto, 2002). Prior studies show that stronger governance decreases earnings management (Bhattacharya et al. 2002; Leuz et al., 2002). In contrast, we show that stronger governance increases both downward guidance of analysts and the propensity of insiders to report positive earnings surprises. Agency conflicts arise when insiders create positive earnings surprises in order to profit by selling overvalued shares (Healy and Wahlen, 1999). They also arise when managers focus on meeting short-term earnings targets and forego long-term value-creating activities (Eccles et al. 2001), such as research and development (Lev 2001). We show that governance mechanisms do not mitigate this agency problem. We document that firms facing strong institutional governance mechanisms tend to resort relatively more to forecast guidance and earnings surprise management. We show that both forecast guidance and earnings surprise management increases in firm size, growth, stability, profitability, and that they pertain more to firms domiciled in the U.S., and more in recent years. We pattern our forecast guidance and earnings surprise management models after Matsumoto (2002). We validate our modified version of the Matsumoto forecast guidance model in five different contexts, including showing that forecast guidance is more prevalent in the U.S. than in other countries. The latter allows us to reconcile two sets of diverse results in the international literature: U.S. insiders manage earnings surprises more than their international counterparts (Brown and Higgins, 2001) and U.S. insiders manage reported earnings less than their international counterparts (Bhattacharya et al. 2002; Leuz et al. 2002). We add to the literature in other ways. We show that a Matsumoto (2002)-type forecast guidance metric pertains to international data, enhancing its external validity beyond U.S. data. We find that a simplified version of the metric is sufficiently powerful to detect guidance, allowing researchers to increase sample sizes (hence power of their tests) by not requiring stock returns data to implement the metric. Matsumoto (2002) assumes that earnings management, forecast guidance, and earnings surprise management are related to the same set of variables in the same directional manner. Our findings suggest that not all variables have the same directional effects of earnings 20

21 management, forecast guidance, and earnings surprise management, which should help researchers to specify better their test designs and expectations. 21

22 References Ashbaugh, H. and T. Warfield Audits as a corporate governance mechanism in debt-oriented countries. Working paper, University of Wisconsin Madison. Ball, R., S.P. Kothari, and A. Robin The effect of international institutional factors on properties of accounting earnings. Journal of Accounting and Economics 29 (February): Barton, J., and Simko, P The balance sheet as an earnings management constraint. Accounting Review (forthcoming). Barsky, N The market game. Wall Street Journal. May 8: A18. Bartov, E., D.Givoly, and C. Hayn The rewards to meeting or beating earnings expectations. Journal of Accounting and Economics (June): Beaver, W.H Financial reporting: An accounting revolution. Prentice-Hall, Upper Saddle River, NJ. Becker, C.L., M.L. Defond, J. Jiambalvo, and K.R. Subramanyam The effect of audit quality on earnings management. Contemporary Accounting Research 15 (Spring): Bhattacharya, U., H. Daouk, and M. Welker The world price of earnings opacity. Working paper, Indiana University. Bhushan, R Firm characteristics and analyst following. Journal of Accounting and Economics 11 (July), Bleakley, F. R Again looks like a gangbuster quarter: Despite odds, strong earnings expected in 2 nd period. Wall Street Journal, June 30: A2. Brown, L.D A temporal analysis of earnings surprises: Profits versus losses. Journal of Accounting Research 39 (September): Brown, L. D., and H. Higgins Managing earnings surprises in the US versus 12 other countries. Journal of Accounting and Public Policy 20 (Winter): Brown, L. D., Richardson, G. D., Schwager, S. J An information interpretation of financial analyst superiority in forecasting earnings. Journal of Accounting Research 25 (Spring): Burgstahler, D. and M. Eames Management of earnings and analysts forecasts to achieve zero and small positive earnings surprises. Working paper, University of Washington. Bushman, R. and A. Smith Financial accounting information and corporate governance. Journal of Accounting and Economics 32 (December): Casey, L.L Shutting the doors to state court: The Securities Litigation Uniform Standards Act of Securities Regulation Law Journal 27 (141):

23 Center for International Financial Analysis & Research (CIFAR) International accounting and auditing trends. Volumes I and II. Princeton, NJ. CIFAR Publications, Inc. Collingwood, H The earnings game: Everybody plays, nobody wins. Harvard Business Review 79 (6): Degeorge, F., J. Patel and R. Zeckhauser Earnings management to exceed thresholds. Journal of Business 72 (January): Eccles, R. G., Herz, R. H., Keegan, E. M., and D. M. H. Phillips The value reporting revolution: Moving beyond the earnings game. John Wiley, New York. Economist General electric: The jack and jeff show loses its luster. (May 4): Fama, E.F., and J.D. MacBeth Risk, return, and equilibrium: Empirical tests. Journal of Political Economy 81 (May/June): Francis, J.R., E.L. Maydew, and H.C. Sparks The role of big 6 auditors in the credible reporting of accruals. Auditing: A Journal of Practice and Theory 18 (Fall): Fulkerson, C.L., S.B. Jackson and G.K. Meek The effect of international accounting diversity on earnings management and earnings quality. Working paper, University of Texas at San Antonio. Gul, F. A., J. S. L. Tsui, X. Su, and M. Rong Legal protection, enforceability and tests of the debt hypothesis: An international study. Working paper, City University of Hong Kong. Hayn, C The information content of losses. Journal of Accounting and Economics 20 (September): Healy, P.M., and J. M. Wahlen A review of the earnings management literature and its implications for standard setting. Accountings Horizons 13 (December): Hung, M Accounting standards and value relevance of financial statements: An international analysis. Journal of Accounting and Economics 30 (December): Ip, G. 1997a. Traders laugh off the official estimate on earnings, act on whispered numbers. Wall Street Journal, January 16: C1, C3. Ip, G. 1997b. Rise in profit guidance dilutes positive surprises. Wall Street Journal, June 23: C1, C3. Kasznik, R. and M. McNichols Does meeting expectations matter? Evidence from analyst forecast revisions and share prices. Journal of Accounting Research (June): La Porta, R., Lopez-de-Silanes, F., Shleifer, A., and R. Vishny Legal determinants of external finance. Journal of Finance 52 (July): La Porta, R., Lopez-de-Silanes, F., Shleifer, A., and R. Vishny Law and finance. Journal of Political Economy 106 (December):