Earnings Management: Do Firms Play Follow the Leader?

Transcription

1 Earnings Management: Do Firms Play Follow the Leader? Brian Bratten Von Allmen School of Accountancy Gatton College of Business and Economics University of Kentucky Jeff L. Payne Von Allmen School of Accountancy Gatton College of Business and Economics University of Kentucky Wayne B. Thomas John T. Steed School of Accounting Michael F. Price College of Business University of Oklahoma January 4, 2011

2 Earnings Management: Do Firms Play Follow the Leader? Abstract: Prior research provides evidence that investors respond to the information released by other firms in the same industry (intra-industry information transfers). We examine whether managers also respond to this information by strategically managing reported earnings and timing earnings announcements based on the earnings news of the industry leader. We find that when industry leaders earnings announcements indicate bad news (i.e., missed analyst forecasts), subsequent firms report reduced levels of income-increasing discretionary accruals, are more likely to manage earnings down, and use fewer income-decreasing special items. These results are consistent with follower firms facing a decrease in pressure to meet earnings expectations. We also find that when leader firms miss expectations, follower firms that miss expectations are more likely to accelerate the timing of their earnings announcement to herd with the leaders bad news. Our study contributes to the academic literature on intra-industry information transfers and suggests that management s incentives to manipulate earnings should be considered in future extensions of this research. Keywords: Information transfer; earnings management; herding.

3 Earnings Management: Do Firms Play Follow the Leader? 1. Introduction In this study we investigate earnings management and announcement timing in a setting where managers have reduced incentives to meet or beat analysts forecasts. Varian [2002] suggests that managers are more willing to disclose disappointing news in settings where the market is expecting it, such as during economic downturns where many companies record large write-downs of balance sheet accounts. He suggests that managers perceive that market participants evaluate earnings performance on a relative basis, or grade on a curve. Accordingly, managers may expect a lower penalty for their own bad news when bad news is commonly reported among peer firms. We extend this idea to an intra-industry setting where industry leaders (e.g., Alcoa) announce earnings that disappoint financial markets (i.e., report earnings that miss analysts forecast). When industry leaders report earnings that fail to meet analysts forecast, we expect that managers of firms that subsequently report earnings (follower firms) in the same industry have reduced incentives to meet or beat analysts forecasts, as investors already anticipate disappointing performance. In fact, in such a setting follower firms may take the opportunity to clean their balance sheets in the current reporting period and reserve additional accruals to increase reported earnings in future periods. When follower firms have bad news, they may also take the opportunity to blend in with the bad news of leader firms by accelerating the timing of their earnings announcements. Prior research on the correlation of earnings news between firms in the same industry focuses primarily on the extent to which investors of one firm react to the earnings information of another firm (i.e., intra-industry information transfers). Initially, Freeman and Tse [1992] find - 1 -

4 that investors of follower firms react significantly to information provided by leader firms in the same industry. Subsequent studies provide mixed evidence about the efficiency of this investor reaction. Ramnath [2002] and Easton, Gao, and Gao [2010] document an apparent underreaction by investors to intra-industry earnings news. In other words, based on the past relation between firms earnings news, investors fail to fully impound the implications of early announcers earnings news for late announcers earnings news. However, Thomas and Zhang [2008] document an over-reaction by investors of late announcers firms on the early announcers earnings announcement date. By design, these studies do not control for management s incentives to manage earnings toward analysts forecasts, thereby assuming comparable levels of potential manipulation in reported earnings between early and late announcers. Our investigation differs from these prior studies. We focus on managers reactions to intra-industry earnings information. Clearly, prior research demonstrates that the market penalizes firms when announced earnings miss analysts forecasts (Kinney, Burgstahler, and Martin [2002], Brown and Caylor [2005], Payne and Thomas [2011]). Managers recognize this and manage earnings to avoid missing expectations based on analysts forecast (e.g., Healy and Whalen [1999], Payne and Robb [2000]). However, the severity of such penalties to missing expectations could be influenced by other economic events, such as the earnings performance of related firms (Farrell and Whidbee [2003]). In addition, the impact of bad news in earnings on managers compensation and employment may be directly affected by the relative performance of other firms in the industry (Morck, Shleifer, and Vishny [1989], Murphy [1999], Mergenthaler, Rajgopal, and Srinivasan [2009]). To the extent that other firms, especially industry leaders, are experiencing unexpected bad performance, a follower manager could - 2 -

5 perceive that she will be held less accountable for reporting similar unexpected bad performance. We predict that these managers are more willing (or consider it less costly) to report earnings that miss expectations when prior earnings announcements of leader firms in their industry miss expectations. In addition to examining managers behavior over discretionary accounting choices in reporting earnings, we also seek evidence about whether follower managers alter the timing of their earnings announcements, or herd, depending on the news of the industry leader. In an earnings warning setting, recent research finds that managers cluster their bad news with other firms (Tse and Tucker [2010]). The idea is that managers have an incentive to time their bad news to occur soon after their industry peers bad news, to better signal industry-wide or marketwide factors (as opposed to firm-specific or manager-specific factors) as being responsible. As a result, we expect that when follower firms have bad earnings news, their managers accelerate the earnings announcement to blend in when a leader disappoints the market. To perform our analyses, we identify Leaders in each industry. Leaders are identified as the first firm to report earnings that is also in the top quartile of market capitalization within the industry. In any given quarter, each industry has one firm designated as a Leader based on the timing of its earnings announcement. Followers include all firms that announce earnings at least five days subsequent to the Leader. To capture the earnings surprise for the Leader, we use analysts forecasts as a benchmark. Our interest lies in examining Followers earnings management behavior and earnings announcement timing when the Leader reports earnings that miss analysts expectations, creating a negative earnings surprise. For comparison, we also examine Followers earnings management and earnings announcement timing behavior when the Leader meets expectations

6 If bad news reported by Leaders is regarded as an opportunity by Followers to reserve accruals with reduced consequences, we expect a higher (lower) amount of income-decreasing (income-increasing) earnings management when Leaders announce negative earnings surprises. As proxies for accrual management, we consider signed quarterly discretionary accruals and the incidence of negative discretionary accruals. We also compare for Followers the signed forecast error and the incidence of missing the analyst forecast when the Leader meets versus misses. While tests of forecast error do not necessarily provide direct evidence of accrual manipulation, they provide additional evidence of the propensity of firms to meet expectations perhaps through accruals not captured by our discretionary accruals model or other activities such as real earnings management (e.g., Cohen, Dey, and Lys [2008]). We also examine the extent to which negative special items are reported. Managers may reclassify core expenses to special items to meet analysts forecasts (McVay [2006], Fan et al. [2010]) or increase the pricing multiple on earnings (Lipe [1986], Kinney and Trezevant [1997], Bradshaw and Sloan [2002]). When the Leader misses analysts forecast, the need to engage in such reclassification is reduced, likely resulting in fewer negative special items being reported. Finally, we examine if the timing of the followers earnings announcement is influenced by the Leader announcing earnings that misses analysts forecasts. We find evidence that when the Leader in an industry misses analysts forecasts, Followers reported earnings reflect a decrease in pressure to meet expectations. When Leaders miss, Followers are more likely to miss and the average earnings surprise is smaller. In addition, when Leaders miss, Followers have reduced levels of discretionary accruals and are more likely to manage earnings downward using negative discretionary accruals. Finally, Followers report fewer income-decreasing special items when the Leader misses. All of these results indicate that - 4 -

7 Followers are less likely to manage earnings upward (or more likely to manage earnings downward) when the Leaders misses analysts expectations. Such downward manipulation has the potential to create reserves, offering the manager more degrees of freedom for manipulating earnings upward in the future. We also examine Leader behavior and find that the extent of discretionary accruals reported by Leaders does not differ depending on whether they meet versus miss expectations. Differences occur only for Followers depending on whether the Leader meets versus misses. These results suggests that results of prior literature investigating management s use of their discretion over financial statement balances to meet or beat analysts forecast are likely influenced by the reporting behavior of Followers. This indicates an improved understanding of Follower behavior is important for furthering our understanding of management s reporting choices. With respect to announcement timing, we report evidence that Followers who report bad news often report the news earlier when the Leader misses analysts forecasts, which we attribute to a desire to blend the Followers news with the Leader s news. Interestingly, when Followers report good news, they report the news later when the Leader misses, perhaps in an attempt to stand out. Our study provides an important extension to the literature on information transfer, earnings management, and earnings announcement timing (herding). It is well known that industry earnings are correlated, but the academic literature provides mixed evidence on how investors of firms announcing later respond to the earnings reports of earlier announcers. Most importantly, existing literature provides little evidence of what causes the correlation of earnings within an industry beyond what might be expected from general external macroeconomic events

8 We find that the earnings content of Followers (late announcers) is influenced by the earnings news of Leaders (early announcers), indicating management uses their discretion less to modify reported financial statement balances in settings where the market penalty for earnings disappointment is reduced. This suggests earnings management should be an important consideration when studying the market s response to prior and contemporaneous earnings announcements. While most of the earnings management literature focuses on incomeincreasing discretion, we identify and provide evidence related to a setting in which firms may have incentives to engage in discretion that decreases reported earnings. Finally, this study provides evidence generally consistent with herding behavior for follower firms with bad news when the Leader reports bad news. However, interestingly we find strong evidence of antiherding behavior for follower firms with good news, conditioned on the Leader firm reporting bad news. We suggest in this setting Followers delay the earnings announcement so that their earnings is considered separately from prior earnings announcements in their industry. Section 2 discusses related literature and derives our two hypotheses. Section 3 describes our data and methodology. Section 4 presents our main results and supplemental analyses, and section 5 concludes. 2. Prior Literature and Hypothesis Development 2.1. Earnings management Financial reporting standards are designed to permit management to faithfully represent the operations of their firm. Financial statement preparation requires these standards be applied to a broad range of accounting issues. Many of these issues require, or at least allow, significant judgment based on management s understanding of their individual firm s business operations. Examples of these judgments include the valuation of receivables, inventory, and pension - 6 -

9 obligations. However, this same process provides the prospect for opportunistic financial statement reporting. Research indicates that firms manage earnings to improve financial statement inferences in a number of settings, such as before public securities offerings, to comply with loan covenants, and to reward managers (Healy and Whalen [1999]). 1 Far fewer studies have attempted to measure earnings management in settings where firms have incentives to reduce reported earnings. Our study provides one such setting Information transfer Several studies investigate the correlation of earnings between firms within an industry. Because firms announce earnings at different times, this correlation implies that the earnings surprises of firms that announce earliest each quarter have implications for the (yet undisclosed) earnings surprises of firms that will announce later. Most of the studies examining intra-industry earnings correlations are primarily interested in whether this information transfers to investors. This is done by examining the stock price response of firms that have not yet announced ( late announcers ) to the announcements of firms that announce before them ( early announcers ). Publicly available information is priced before earnings are announced (Ball and Brown [1968], Collins, Kothari and Rayburn [1987], Freeman [1987]); this suggests that investors can use announcements by other firms to estimate firm earnings. Freeman and Tse [1992] estimate the relation between the change in quarterly earnings of firms within an industry to the change in quarterly earnings and sales of firms announcing earlier than those firms, and report that this relation varies across industry and is generally high. They 1 Academic research provides significant evidence consistent with this type of managerial behavioral. See, for example, Burgstahler and Dichev [1997], Degeorge, Patel, and Zeckhauser [1999], Payne and Robb [2000], Bartov, Givoly, and Hayn [2002], Beatty, Ke, and Petroni [2002], Moehrle [2002], Burgstahler and Eames [2003], Das and Zhang [2003], Brown and Caylor [2005], Graham, Harvey, and Rajgopal [2005], and Payne and Thomas [2011]

10 also show that late announcers returns around early announcers announcement dates are positively associated with the earnings changes of early announcers, and that the magnitude of this association is increasing in the historical relation between the earnings within an industry. 2 Subsequent research extends Freeman and Tse [1992] by investigating whether the magnitude of the price response for late announcers to the earnings of early announcers is efficient. Ramnath [2002] provides the earliest evidence that such a response is inefficient using a sample of quarterly earnings announcements between 1995 and He finds that analysts and investors of subsequent announcers respond to the earnings surprise of first announcers. However, revised analysts forecast errors and returns around the subsequent announcers earnings announcements are positively correlated with the predictable forecast error, suggesting initial under-reaction by both analysts and investors. 3 Thomas and Zhang [2008] find a similarly inefficient but directionally inconsistent result, using a larger sample from 1973 to They find that investors in late-announcing firms overreact to the information provided by firms announcing earlier within an industry. In their study, the average stock return reaction for late-announcing firms to early-announcing firms warnings is too strong and reverses when the late announcers report earnings. While not directly trying to reconcile the differences between Thomas and Zhang [2008] and Ramnath [2002], Easton, Gao, and Gao [2010] find evidence more consistent with the Ramnath [2002]. They form a trading strategy based on the implied return of late announcers based on the abnormal return of early announcers and the historical relation between abnormal earnings of early and late firms. They find that this strategy generates positive returns, indicating 2 Freeman and Tse [1992] also show that most of the relation between the returns of late announcers and the news of early announcers (or, information transfer ) is related to the first quarterly earnings announcement in an industry. 3 Ramnath [2002] estimates the predictable forecast error for the late-announcing firm by regressing the forecast error in the prior period of the later announcing firm on the forecast error of the first announcing firm, and applying the coefficients from this regression to the forecast error of the earlier announcer in the current period

11 initial under-reaction or drift for investors of late announcers to the news provided by earlier announcers. Beyond market penalties associated with reporting bad news, managers discretionary accrual decisions are also likely to be affected by compensation and employment decisions. The literature on the sensitivity of earnings performance to compensation and employment (Matsunaga and Park [2001], Matsumura, Shin, Wu [2009]) also investigates how these sensitivities relate to peer performance (Farrell and Whidbee [2003], Mergenthaler, Rajgopal, and Srinivasan [2009]). Specifically, Mergenthaler, Rajgopal, and Srinivasan [2009] and Morck, Shleifer, and Vishny [1989] find that relative industry performance is a significant determinant of CEO dismissal and compensation. Aggarwal and Samwick [1999] derive optimal contracts of strategic competition in a framework where managers can be compensated on both their own profits and their rival s (or industry) profits. They find in some settings that consideration of industry performance improves employment contracting. This literature indicates that information on industry profitability provided by early announcers is useful and effective in evaluating management, and may provide incentives for late announcers to adjust reported performance. One stream of these studies provides strong (although conflicting) evidence that market inefficiencies exist related to the information transfer among firms in an industry. However, another stream shows that relative peer performance is important to managers employment and compensation, which suggests that managers may have incentives to report differently depending on how the Leader within an industry performs. The focus of our study is to extend this research by considering management s ability to influence the content of earnings announcements. We - 9 -

12 consider the influence on subsequent announcers (Followers) of an industry Leader, the firm in the top quartile of market capitalization that is the first to announce. If managers believe the penalty is disproportionately high for missing an earnings target, they may have incentives to manipulate earnings to achieve earnings thresholds (e.g., Burgstahler and Dichev [1997], DeGeorge, Patel, and Zeckhauser [1999]). Managerial behavior to avoid missing earnings targets may also be driven by other incentives, such as career concerns (bonuses, promotion, and retention), contractual obligations (Graham, Harvey, and Rajgopal [2005]), and penalties for missing earnings expectations by a material amount. In these settings, failure to achieve market expectations can lead to reduced market value for the firm as well as damaging the overall reputation of management (Fama [1980], Trueman [1986]). Earnings management behavior around earnings thresholds is therefore influenced by factors in addition to market-based incentives. Within this setting, managers face an asymmetric loss function. Significant costs can be incurred when financial statement users, including investors, management, and creditors, learn of unexpected negative earnings surprises (Fama [1980], Trueman [1986], Skinner [1994]), but little penalty is incurred when neutral or positive information is released (Skinner [1994]). This encourages management to minimize negative earnings news. However, in settings where negative earnings news is expected because prior earnings announcements have disappointed investors, managers may be less inclined to use earnings management to meet or beat earnings expectations. Further, managers could purposely understate performance to allow flexibility in future reporting decisions (Levitt [1998]). If manipulation within GAAP allows firms to use discretion to manage some amount of earnings, this calls into question why any firms would report earnings that barely miss an earnings threshold. We suggest that when an industry Leader reports earnings that miss analysts

13 forecast, managers of follower firms may believe their own expectations have been reduced that allows managers the option of barely missing analysts forecast as the perceived potentially penalty for doing so is smaller. 4 Prior research investigates information transfer among investors from prior earnings announcements without considering management s ability to alter reported earnings. We predict that the content of the information released in the earnings announcement of Followers is influenced by the earnings news of Leaders. Stated formally: H1: When Leaders report earnings that miss expectations, Followers are less likely to manage earnings upward (or more likely to manage earnings downward). We examine H1 using common proxies for earnings management from prior literature. This includes being more likely to miss expectations, reporting less positive discretionary accruals, and being more likely to report negative discretionary accruals. In addition to accrual manipulation activity, H1 also considers that other mechanisms by which managers attempt to avoid missing thresholds, such as income classification shifting (McVay [2006]; Fan et al. [2010]). In an attempt to meet expectations, managers may reclassify (or shift) certain core expenses to negative special items. Such behavior has the effect of increasing core operating profitability, which is more closely tied to analysts expectations. We predict that Followers have reduced incentives to engage in this type of classification shifting when the Leader misses. 4 Other studies examine earnings management by late announcers. Chai and Tung [2001] are motivated by market anticipation of bad news when news is delayed, and find evidence consistent with late announcers taking advantage of this perception by using income decreasing discretionary accruals. Park and Ro [2004] find that a firm s relative earnings performance affects the level of discretionary accruals reported. Note, however, that neither of these studies considers the influence of information provided by prior earnings announcements, which is the focus of our study

14 2.3. Herding Principal-agent concerns can motivate herding behavior as managers performance evaluations are often based on relative not absolute performance (Devenow and Welch [1996]). Literature discussed previously (e.g., Farrell and Whidbee [2003]) documents this relationship between industry performance and the evaluation of management. This motivates managers to monitor and reflect on other firms actions before making important decisions. Herding may occur among individuals whose behavior is correlated, often based on some particular event such as what clothes to wear, what stocks to purchase, and in our study, when to report earnings (Devenow and Welch [1996]). Trueman [1990] shows that management might purposely delay earnings information until other industry-wide bad news is released. In cases where bad news is to be reported this might allow a justification for the news, mitigating potential reputational, litigation, and equity costs. A recent stream of literature investigates how management responds to bad news released by peer firms within an industry by examining how the timing of announcements is influenced by management (Arya and Mittendorf [2005], Acharya, DeMarzo, and Kremer [2008], Tse and Tucker [2010]). For example, Tse and Tucker [2010] examine whether firms cluster bad news contained in earnings warnings. Using a hazard model, they find that firms react when other firms in their industry issue a negative earnings warning. For each warning issued in the preceding five days, a firm is 7.4% more likely to issue their own warning. This suggests that firms prefer to cluster their bad news with the bad news of their industry peers. Earnings announcements are also important events in the information discourse between management and financial statement users. Although earnings announcements are more common than earnings warnings, their timing is not entirely predictable and both news and

15 return volatility vary with the timeliness (Chambers and Penman [1984]). By reporting with the herd, managers may attempt to mask earnings disappointments assuming that investors evaluate earnings announcements as a group and reduce their reliance on individual earnings reports. This leads to our prediction that Followers with bad news are more likely to cluster their earnings announcements with the Leader when the Leader misses. Stated formally: H2: When Leaders report earnings that miss expectations, Followers who report bad news in earnings will report earlier. 3. Data and Method To perform our analysis, we use all firm-quarters from 1993 to 2008 with sufficient data on I/B/E/S and COMPUSTAT to compute a quarterly earnings surprise based on analysts forecasts and other financial measures. We retain all four quarters, but limit our sample to December year-end firms to ensure that fiscal firm-quarters are comparable within our sample (Freeman and Tse [1992]; Thomas and Zhang [2008]). Firms are assigned to industries based on the 48 classifications used in Fama and French [1997], and we exclude financial services firms (SIC codes between 6000 and 6700, Fama-French classification industry #22). We identify an earnings announcement Leader for each industry-quarter as the first large firm in each industry to announce earnings, where a large firm is defined as a firm with market capitalization at the end of the previous fiscal year in the top quartile of the firm s industry. When more than one large firm announces on the same day at the same time (or on the same day and the announcement time is not available), the larger firm is considered the Leader. Followers are then identified as all firms within the Leader s industry whose earnings announcement is made at least five days following the earnings announcement of the Leader. For both Leaders and Followers, the benchmark for earnings is the median earnings estimate

16 based on all analysts forecasts reviewed within 90 days prior but no sooner than one day before the Leader s earnings announcement. The firm s Earnings Surprise is calculated as actual reported earnings minus the median analyst forecast, divided by the firm s stock price at the end of the fiscal quarter. 5 Miss is an indicator variable for firms with a negative Earnings Surprise. Our analysis also requires estimation of discretionary accruals. We calculate Discretionary Accruals as the residual from the following regression estimated with quarterly data for all COMPUSTAT firms in each industry-year (industry and firm subscripts omitted): TA q / AvgAssets q = α + β 1 (1 / AvgAssets q ) + β 2[ (ΔSALE q ΔREC q ) / AvgAssets q ] + β 3 (PPE q / AvgAssets q ) + ε q (1) where (COMPUSTAT mnemonics in parenthesis): TA q = total accruals for quarter q, defined as the income before extraordinary items (IBQ) minus the cash flows from operations (OANCFY), 6 AvgAssets q = average of total assets (ATQ) for quarters q and q 1, ΔSALE q = difference between sales (SALEQ) in quarter q and sales in quarter q 1, ΔREC q = change in accounts receivable (RECCHY) from quarter q 1 to quarter q, 7 and PPE q = net amount of property, plant, and equipment (PPENTQ) at the end of quarter q. This discretionary accruals model generally follows Brown and Pinello [2007], although we include a constant term as an additional control for heteroskedasticity (Kothari, Leone, and Wasley [2005]). All variables used to estimate this modified Jones model are winsorized at the 5 We use the unadjusted detail file to calculate the surprise (Payne and Thomas [2003]), and adjustments are made for any stock splits occurring between the time of the median estimate and the earnings announcement date. 6 OANCFY is a year-to-date variable. For equation (1), we use this variable in fiscal quarter one and the difference from the previous quarter in fiscal quarters two through four. 7 RECCHY is a year-to-date variable. For equation (1), we use this variable in fiscal quarter one and the difference from the previous quarter in fiscal quarters two through four

17 1 st and 99 th percentile annually, and we require at least 15 observations to estimate each regression. In addition to the signed Discretionary Accruals measure, we are interested in the incidence of downward management. 8 Accordingly, we define Manage Down as an indicator variable set to one when Discretionary Accruals is less than zero, and zero otherwise. Our last proxy for earnings management captures the extent to which income-decreasing special items (Special Items) are reported. Following Fan et al. [2010], this variable is defined as negative one times special items (SPIQ) divided by sales (SALEQ), after setting incomeincreasing special items to zero. If income-decreasing special items are greater than sales, we set this variable equal to one. To investigate H2, we measure and report the timeliness of the earnings announcements for Leaders and Followers. We compute the Days to Announce as the length of time in days between the fiscal quarter end and the date of the earnings announcement. For our analysis, we eliminate Followers who announce less than five days after the Leader (approximately 5.6% of observations), to ensure our sample retains firms that have adequate time to respond to the Leaders earnings announcement. In Table 1 we present descriptive statistics for our variables of interest as well as general firm characteristics. As shown in Panel A, our final sample consists of 1,947 firm-quarters of 380 unique industry Leaders from 1993 to As shown in Panel B, in these same quarters, we have 7,120 unique Followers and 77,273 Followers-quarter observations. By design, Leaders announce earlier in the quarter (mean = 17.6 days after the quarter-end) than followers do (mean 8 An alternative approach would be to investigate whether managers use income-increasing discretionary accruals to meet or beat analysts forecasts in settings where Leaders have reported earnings that meet or exceed analysts expectations. As noted in Table 1, only 23% of Leaders miss analysts forecasts, creating a more unexpected situation for financial markets than just meeting or beating. Therefore, we focus our investigation in this setting where financial markets are more likely to have been surprised by the Leaders earnings announcement, creating ambiguity in the expectations set for Followers

18 = 33.0 days). Leaders have greater Earnings Surprise (0.037% vs %) and lower incidence of Miss (23.2% vs. 33.4%). Discretionary Accruals are similar between Leaders and Followers, and so is the incidence of Manage Down. 9 Leaders have fewer income-decreasing Special Items than do Followers (0.013 and 0.028, respectively). As one may expect, Leaders are larger, have lower book-to-market ratios, have higher analyst following, and lower sales growth. 4. Results 4.1. Univariate Results Our first main analysis compares the earnings management of firms conditional on the sign of the earnings surprise of the Leader. Table 2 shows the results of this analysis. In Panel A, we partition our sample into groups of Leader-firm observations based on whether the Leader meets analysts earnings expectations (Earnings Surprise 0) or misses (Earnings Surprise < 0). By design, the earnings surprise is positive (first two columns) when the Leader meets and negative when the Leader misses (second two columns). Our primary interest is in the pattern of results for our three proxies of the extent of earnings management, Discretionary Accruals, Manage Down, and Special Items. The average Discretionary Accruals for Leaders when they meet analysts expectations is , which is insignificantly different from the average level of discretionary accruals when Leaders miss expectations (0.0115). The difference in medians is also not significant. Similarly, the incidence of negative discretionary accruals (Mange Down) and the extent of the use of income-decreasing Special Items are not significantly different for Leaders in periods when they meet versus miss analysts expectations. We conclude from this 9 Averages for Discretionary Accruals are above zero for Leaders and Followers (0.012 and 0.013) because the discretionary accruals model is estimated across all COMPUSTAT firms and our final sample consists of I/B/E/Sonly firms which have slightly higher discretionary accruals on average. Our main results are not affected if we estimate discretionary accruals using only firms in our final sample (which causes the average discretionary accrual to be much closer to zero)

19 analysis that Leaders in an industry do not undertake a differential level of earnings management, on average, depending on whether they meet or miss analysts expectations. Prior literature provides evidence on management s use of their discretion over reported financial statement balances to meet or beat analysts forecast (e.g., Payne and Robb [2000], Bartov, Givoly, and Hayn [2002], Matsumoto [2002], Brown and Pinello [2007]). The lack of significant differences in earnings management metrics between Leaders that meet versus miss suggests that prior results are not present among industry Leaders. Instead, as we show below, this result is driven by Followers. In Panel B of Table 2 we examine the same set of variables for Followers. That is, we examine whether the mean and median proxies for earnings management of the Follower differ depending on whether the Leader meets or misses analysts expectations. We find that the average Earnings Surprise for Followers when is more negative the Leader misses than when the Leader meets. The average difference is , or 0.03% of the Follower s market value. For a $40 stock, this implies that the average earnings surprise is approximately $0.01 lower. While such an amount may seem economically insignificant, the results for Miss suggest that much of this effect is occurring at the zero earnings surprise threshold, where earnings management should most reasonably occur. Followers miss 36.76% of the time when the Leader misses expectations, but only 32.47% of the time when the Leader meets expectations. This difference is easily significant. We next turn to measures of discretionary reporting. First, the average level of Discretionary Accruals for Followers is when the Leader misses, compared to when the Leader meets. This difference is significant (t-statistic = 7.24). Thus, followers use accruals for income-increasing earnings management significantly less when the Leader misses

20 its earnings expectation, likely in response to the reduced pressure to meet expectations in this environment. Second, while 33.49% of Followers report negative discretionary accruals when the Leader meets, 35.49% report negative discretionary accruals when the Leader misses, and this difference is also significant (t-statistic = 4.85). When the Leader provides no pressure to present an impressive earnings report, Followers are less likely to manage earnings upward, or perhaps more likely to manage earnings downward to clean their balance sheets and increase reserves for future periods. Third, income-decreasing Special Items are 2.31% of sales for Followers when Leaders miss expectations, compared to 2.91% when the Leader does not miss. This difference is significant (t-statistic = 6.63). Followers less often classify expenses as special when Leaders miss, likely because pressure to report higher core earnings is reduced. Prior research shows that managers opportunistically shift core expenses to special items to increase core earnings because special items are considered transitory and are often excluded from analysts earnings forecasts (Gu and Chen [2004], McVay [2006]). However, when Leaders miss, managers may believe that investors expectations for the Follower are also lower, or that they will receive a lower penalty for reporting lower core earnings. Further, managers may believe that by abstaining from classification shifting when the Leaders miss, in future periods (when Followers have more pressure to report a more favorable earnings report), such classifications will be more believable to investors. These results are consistent with H1. The results presented in Table 2 related to H2, Days to Announce, are inconclusive. For Followers, the average number of days between the fiscal quarter end and the date of the earnings announcement is when Leaders meet, and virtually the same (33.02) when Leaders miss. However, because H2 predicts herding behavior only among Followers with bad

21 news (not all followers), we are cautious to make inferences without further analysis. Next, we examine our discretionary reporting metrics and Days to Announce according to level of earnings news being reported by the Follower. A substantial number of Followers meet analysts expectations even when Leaders miss analysts expectations. Considering this, we perform an additional analysis, presented in Table 3, to determine whether the extent of the response to Leaders surprise varies depending on the sign and magnitude of the earnings news of the Follower. For this analysis we divide our sample into seven groups based on the difference between the firm s reported earnings and the median analyst forecast (Forecast Error). These groups include Followers that easily miss (Forecast Error from $0.10 to $0.06), moderately miss (Forecast Error from $0.05 to $0.02), barely miss (Forecast Error = $0.01), just meet (Forecast Error = $0.00), just beat (Forecast Error = $0.01), moderately beat (Forecast Error from $0.02 to $0.05), or easily beat (Forecast Error from $0.06 to $0.10). We anticipate these groups may face different levels of incentives to manage earnings based on the earnings news of the Leader. 10 While discretion within GAAP may allow for some manipulation of earnings, it is not likely to help firms that will easily miss earnings targets. Instead, only those firms whose true earnings will barely miss the earnings threshold are able to use discretion within GAAP to manage reported earnings to meet the threshold, and thereby avoid a potentially large market penalty. We expect to observe, across the distribution of reported earnings, that there will be an unusually low (high) number of observations that barely miss (meet) the earnings threshold (DeGeorge, Patel, and Zeckhauser [1999], Payne and Robb [2000]) and that reporting behavior 10 We exclude cases of extreme meeting or missing (Forecast Error < $0.10 or > $0.10). This represents 16% of the observations

22 may differ by the magnitude of this difference. Based on the number of observations classified within each Forecast Error group, this distribution is present for our sample. Results from this analysis are presented in Table 3. In addition to providing some evidence that the results presented in Table 2 that support H1 are generally robust, this analysis provides several additional insights. First, compared to the discretionary accruals reported when the Leaders meet expectations, the discretionary accruals of Followers when the Leader misses expectations are consistently lower for all levels of forecast error except the most extreme disappointments (Forecast Error = $0.10 to $0.06). Similar results obtain for the use of special items to alter reported earnings. The barely miss category (Forecast Error = $0.01), has the largest decline in average discretionary accruals for any of the subgroups. Accordingly, Followers that barely miss appear to face the largest extent of reduced pressure to perform favorably, and respond by taking less aggressive positions with their accruals. As expected if firms manage earnings to meet thresholds, Followers who meet by a small amount (Forecast Error = $0.00 or $0.01) exhibit the highest level of discretionary accruals. Second, only Followers that just beat or moderately beat are more likely to manage down when Leaders miss. Although Followers that miss are less aggressive at taking income increasing accruals when Leaders miss, they do not appear to be managing earnings downward more often when the Leader misses. Third, Followers who barely miss analysts earnings expectations by a small amount (Forecast Error = $0.01) show evidence of a timeliness effect by announcing earnings earlier when the Leader misses than when the Leader meets (Days to Announce = 31.6 vs days, t- statistic = 4.66), and this results is also significant for followers with large disappointments (Forecast Error = $0.06 to $0.10). The herding literature finds that firms with bad news

23 attempt to release this news to blend in with the bad news of others, and our result is consistent with this. We find that Follower earnings are released sooner in all settings when Leaders fail to meet analysts forecast. For firms reporting earnings that barely miss, or easily miss, analysts expectations, the difference is significant in support of H2. This is consistent with managers herding in settings where the market has been disappointed by Leader performance. H2 makes no prediction regarding manager behavior in settings where Followers earnings beat analysts forecast, as this would not represent negative information. We find that Followers that easily beat analysts forecast (Forecast Error =$0.06 to $0.10) actually report earnings later when the leader misses Multivariate Results To improve the robustness of our results we also perform multivariate analyses on our variables of interest. This allows us to control for other factors that potentially influence their values. For the sake of parsimony we include the same set of control variables in each separate regression while acknowledging that every control variable shown in prior literature to affect our variables of interest are not included. We use the following model to further investigate our expectations for the earnings management response by followers (firm subscripts omitted): DV q = γ 0 + γ 1 LeaderMiss q + γ 2 ln(market Capitalization q ) + γ 3 Book/Market q + γ 4 Analyst Following q + γ 5 Dispersion q + γ 6 Sales Growth q + γ 7 Fourth (2) Quarter q + γ 8 Days to Announce q + ν q where: DV = Dependent Variable, using Miss, Discretionary Accruals, Manage Down, Special Items,

24 LeaderMiss = 1 if Earnings Surprise is less than zero for the industry Leader, 0 otherwise, Market Capitalization = number of shared outstanding (in millions) times the price at the end of the quarter, Book/Market Analyst Following Dispersion Sales Growth Fourth Quarter Days to Announce = book value of common equity divided by Market capitalization, = number of analysts with an estimate for the quarter s earnings, = standard deviation of analyst forecasts divided by the firm s stock price at the end of the fiscal quarter, = quarterly seasonal percentage change in sales (due to data restrictions, Sales Growth is only available for 1,943 Leaders and 76,381 Followers), = 1 for observations in the fourth fiscal quarter, 0 otherwise, and = number of days between the fiscal quarter end and the earnings announcement. Continuous variables are winsorized at 1% and 99% across the full sample of firms. We include Market Capitalization, Book/Market, and Analyst Following to capture the effect of size and information environment on our variables of interest. Large firms and firms with lower book-to-market ratios may have less incentive to engage in earnings management because there is less uncertainty surrounding their operations. The same may be true for firms with greater analyst following. Inclusion of Dispersion is motivated based on findings in Payne and Robb [2000] that managers alter their use of discretionary accruals to meet or beat analysts forecast based on the dispersion of analysts forecast. As dispersion grows, we expect that earnings will more likely miss analysts forecast, and include fewer positive discretionary accruals. The uncertainty created by the information asymmetry in analysts forecast will also allow managers to make use of income decreasing special items. Sales Growth is included to control for potential differences in management s incentives that are affected by how rapidly their firm is growing

25 Fourth Quarter is an indicator variable that controls for differences in earnings management across quarters (Brown and Pinello [2007] and Fan et al. [2010]). The results of this analysis are presented in Table 4. For all models, LeaderMiss is significant in the direction predicted by H1 and provides results consistent with our prior analyses. When the Leader misses, Followers are more likely to miss, have lower discretionary accruals, more likely to report negative discretionary accruals, and have fewer incomedecreasing special items. In most cases, the control variables are significant in a direction consistent with prior research. Overall, this analysis provides additional evidence supporting H1 and indicates that other factors known to influence our variables of interest are not potential explanations for our results. We perform a similar but not identical analysis for announcement timing to further investigate H2. Our results in Table 3 suggest that the timing of the earnings announcement is affected both by whether the Leader meets or misses and by the news being reported by the Follower. Accordingly, we consider this in the following model, which we estimate separately for followers who miss expectations and for those who meet expectations (firm subscripts omitted): 11 Days to Announce q = γ 0 + γ 1 Days to Announce q 1 + γ 2 LeaderMiss q + γ 3 Earnings Surprise q + γ 4 LeaderMiss q *Earnings Surprise q + γ 5 ln(market Capitalization q ) + γ 6 Book/Market q + γ 5 Analyst Following q + γ 6 Dispersion q (3) + γ 7 Sales Growth q + γ 7 Fourth Quarter q + ν q Variables are as defined in model (2), except that Days to Announce q 1 is the number of Days between the prior fiscal quarter end and the prior earnings announcement, and Earnings 11 As in Table 3, we ignore extreme earnings news (Forecast Error < $0.10 or > $0.10)

26 Surprise is the actual reporting earnings minus the median earnings estimate, divided by the firm s stock price at the end of the fiscal quarter. Results from estimation of model (3) are presented in Table 5. The first two columns present results for followers that miss analysts expectations. These firms announce earnings sooner when the industry Leader disappoints (LeaderMiss = , t-statistic = 2.91), consistent with H2. However, as presented in the second two columns, Followers who meet or exceed analysts expectations actually delay their good news when the industry Leader disappoints (LeaderMiss = , t-statistic = 3.07). This suggests that when followers have good news, they accent this by announcing later Supplemental Analyses In our final analysis we investigate whether our inferences vary based on firms fiscal quarter within the year. Brown and Pinello [2007] report evidence suggesting that because annual earnings are subject to an independent audit, income-increasing earnings management is less likely in the fourth quarter relative to other quarters. Accordingly, our results might be driven by a response to auditors requiring adjustments in the fourth quarter to ensure annual earnings are more representationally faithful. Recall that we limit our sample to December yearend firms to ensure that the fiscal quarter of the Leader and the Follower are the same. In addition to providing a clean setting for our main analysis, this design choice also allows us to examine the results separately for each fiscal quarter without concern for fiscal-quarter differences influencing the results. We report these quarterly results in Table 6. Once again we find that the results from Table 2 are generally robust. We notice consistent evidence of a response by the Follower to the