When Voluntary Disclosure Isn t Voluntary: Management Forecasts in Japan

Transcription

1 When Voluntary Disclosure Isn t Voluntary: Management Forecasts in Japan Kazuo Kato Osaka University of Economics katou@osaka-ue.ac.jp University of Sydney K.Kato@econ.usyd.edu.au Douglas J. Skinner Graduate School of Business The University of Chicago dskinner@chicagogsb.edu Michio Kunimura Meijo University kunimura@ccmfs.meijo-u.ac.jp March 2006; updated August 2006 Osaka University of Economics Working Paper Series, No Abstract This study provides evidence on management forecasts in Japan, where managers are effectively required to provide sales and earnings forecasts at the beginning of each fiscal year and to update those forecasts regularly. Compared to many other countries where forecasting is voluntary and litigation is an important consideration for management, the Japanese institutional setting provides us with an opportunity to investigate several hypotheses related to managers forecasting incentives. We find that managers initial forecasts in each year are overly-optimistic, especially for firms with poor profitability, and that many managers are consistently overoptimistic from one year to the next. We also find that managers issue downward forecast revisions during the year that largely correct this optimism but that forecasting generally does not display the same asymmetries evident in US data and that are often attributed to litigation. In spite of their systematic over-optimism, management forecasts in Japan are informative, although the stock price reaction to these forecasts is not as large as typically observed in US data. * Kato acknowledges financial support from the Japanese Society for the Promotion of Science, Granted-in-Aid for Scientific Research C Skinner acknowledges financial support from the Neubauer Family Faculty Fellowship at the University of Chicago, Graduate School of Business.

2 1. Introduction There is a large accounting literature on the voluntary release of management earnings forecasts. This literature typically uses data for US firms, and analyzes, among other things, managers incentives to provide earnings forecasts and various properties of these forecasts, including whether they are informative to investors. 1 One general finding in this literature is that managers tend to provide these forecasts when their firms are doing well (e.g., Lev and Penman, 1990; Lang and Lundholm, 1993; Miller, 2002) or when they have adverse information about their firms shortrun earnings prospects (e.g., Skinner, 1994). The literature also finds that management earnings forecasts are relatively infrequent. Pownall, Wasley and Waymire (1993) sample eight weeks chosen at random from each year from 1980 to 1987 and find only 444 management earnings forecasts. 2 Even in more recent years, with an increase in the prevalence of earnings guidance, less than a quarter of listed US firms provide management earnings forecasts, and those that do provide guidance do not do so every quarter (Anilowski, Feng, and Skinner, 2006). There is some evidence of a trend for US firms to reduce the extent to which they provide earnings guidance (e.g., see Graham, Harvey, and Rajgopal, 2005; Chen, Matsumoto, and Rajgopal, 2005; Houston, Lev and Tucker, 2005). 3 1 There is also evidence on management earnings forecasts in countries other than the US (in the UK, Canada, and Australia, for example) but the main features of the institutional environment in these countries are similar to that in the US, in that management earnings forecasts are made voluntarily. 2 This is for US firms based on a search for point, range, minimum or maximum forecasts that uses the Wall Street Journal and the Dow Jones News Retrieval Service. 3 Some well-known firms have recently indicated that they will no longer provide earnings guidance, of which Coca Cola is perhaps the most prominent example. Google has made it clear from the time 1

3 Empirical evidence from the US shows clear evidence of asymmetries in management forecasts and how the market responds to those forecasts. Anilowski, Feng and Skinner (2006) use a large and relatively comprehensive sample to confirm previous evidence that the majority of management earnings forecasts convey negative earnings news, and that the market responds more strongly to negative guidance than to neutral or positive guidance. Given the voluntary nature of these forecasts, these asymmetries are usually interpreted as being due to the strategic nature of management forecasting. 4 This study provides evidence on management forecasting in Japan. Management forecasting is effectively mandated in Japan, which allows us to provide evidence on the properties of forecasting in a setting where forecast disclosure is not a management choice. 5 In the US and other countries where forecasting is voluntary, our interpretation of the forecasts that we observe their properties and informativeness is fundamentally affected by our knowledge that managers choose to disclose this information, and that some other managers choose to withhold similar information (e.g., Verrecchia, 1983; Dye, 1985; Jung and Kwon, 1988). In Japan, however, the fact that all managers are effectively required to provide forecasts is likely to affect the nature of the information they disclose and of its IPO that it will not provide earnings guidance. Proprietary costs have also been offered as a reason that more firms do not provide earnings guidance. 4 Soffer, Thiarajan, and Walther (2000) provide evidence suggesting that management strategically releases earnings guidance that enables them to generate non-negative surprises on earnings announcement dates, while Kothari, Shu and Wysocki (2005) argue that management tends to leak good earnings news ahead of bad earnings news, which helps explain the larger reaction to management earnings forecasts that convey bad news. 5 As far as we are aware, there is little previous research analyzing management forecasts in Japan, apart from an early paper by Darrough and Harris (1991). Conroy, Eades, and Harris (2000) confirm our characterization that management earnings forecasts are essentially mandated in Japan. 2

4 how capital markets respond to that information. This is potentially of interest in the US, where the SEC has, with limited success, encouraged companies to provide more forward looking information to investors. Researchers in the US argue that litigation has first-order effects on managers disclosure practices (e.g., Skinner, 1994, 1997; Johnson, Kasznik, and Nelson, 2001; Brown, Hillegeist, and Lo, 2005; Field, Lowry and Shu, 2005). While previous research makes it clear that litigation affects managers disclosures, it is unclear exactly how important a role litigation plays. 6 There are other reasons managers are likely to face an asymmetric loss function in making forecast decisions, which may be equally or even more important in driving managers forecast disclosure choices. For example, managers credibility/reputation with investors is also likely to affect their disclosure choice and encourage the timely disclosure of adverse earnings news (e.g., see Tucker, 2005). By analyzing the properties of management forecasts in Japan, an environment where the threat of stockholder litigation is much lower than that in the US (and most other countries in the world), we can provide evidence on the extent to which litigation effects explain observed patterns in managers disclosure practices. Finally, because of the mandatory nature of management forecasts in Japan, we can examine time-series properties of management forecasts. This allows us to examine whether managers adopt a consistent disclosure strategy over time, and in particular allows us to gauge the likely importance of credibility and reputation effects on disclosure, especially in the absence of litigation. Another unique feature 6 Consider the fact that there are a total of only securities class action lawsuits filed against listed companies in the US each year, many of which are unrelated to disclosure. Given that there are well over 10,000 listed companies in the US, the unconditional likelihood of being sued is low. 3

5 of the Japanese setting is that managers are required to forecast sales, earnings before extraordinary items and taxes (EBET), and net income (NI); we examine differences between results for the three types of forecast to help validate our inferences regarding managers forecasting incentives. Our evidence shows that the properties of forecasts in Japan are different to what we observe in the US and other countries. First, managers initial forecasts in each fiscal year are overly-optimistic: they generally exceed prior realizations and result in large, negative forecast errors. Moreover, we find that managers of firms with the worst profitability set the most optimistic forecasts for the next year. This is consistent with managers in Japan using their initial forecasts to try and convince investors and other corporate constituents that their firms are performing well, presumably to achieve stock price, performance evaluation, and compensation benefits. Second, we find that managers issue forecast revisions that largely correct their initial optimism, so that earnings surprises are small. Thus, similar to what is observed in US data, investors earnings expectations are set at unduly high levels at the beginning of the year, and then walked down during the year to meet realizations. The difference is that in Japan, where litigation is not a factor, managers do this explicitly using forecasts, likely because the costs of issuing optimistic initial forecasts and subsequently revising those forecasts downward are relatively small. We find little evidence in the Japanese data that managers attempt to preempt earnings news in an asymmetric fashion, a strong tendency in US data (e.g., Skinner, 1994; Soffer et al., 2000). For example, we do not observe Japanese managers systematically pre-announcing adverse earnings news or taking other actions to 4

6 generate non-negative earnings surprises. Given that stockholder litigation is not an important factor in Japan, these differences support the idea that litigation affects disclosure in those countries where we do find asymmetric disclosure. Japanese managers do, however, make forecast revisions that minimize the magnitude of earnings surprises, consistent with Ajinkya and Gift s (1984) expectations adjustment hypothesis. Somewhat surprisingly, we find that managers who release initial earnings forecasts that are overly-optimistic in one year (which happens about 2/3 of the time) are also likely to release overly-optimistic forecasts the next year. This is inconsistent with the idea that consistently biased forecasts reduce the credibility of managers forecasts and/or have adverse reputation effects for managers. However, this may explain why the information content of management forecasts in Japan tends to be smaller than that in the US, although we present evidence that management forecasts in Japan do move firm-level stock prices. Finally, in spite of the fact that managers initial forecasts are systematically overly-optimistic, we also find that these forecasts do help predict future earnings changes, so that they are informative about earnings. Overall, we document a number of differences between managers forecasts in Japan and those in other countries, most notably the US. These differences support the view that litigation has important effects on disclosure in these other countries. We also find that managers in Japan use forecasts strategically to affect perceptions of their performance in the beginning of each fiscal year, but that they subsequently adjust revise their forecasts to avoid earnings surprises. Perhaps surprisingly, we do not find that reputation/credibility effects are sufficient to 5

7 discipline managers tendency to make systematically optimistic forecasts at the beginning of each year. The next section of the paper lays out the institutional setting in Japan in more detail and develops our hypotheses. Section 3 describes our sample and provides initial descriptive statistics. Section 4 reports our main empirical analyses and Section 5 concludes. 2. Institutional Background and Hypothesis Development 2.1 Management Forecasts in Japan The Stock Exchange Act (the Act) in Japan governs disclosure and financial reporting practices for Japanese public companies. In addition, the so-called Timely Disclosure Rules (Kessan-Tannsin or summary of financial statements ) enforced by Japanese stock exchanges impose more stringent requirements on disclosure practices. These rules evolved over time and originated in 1965 from the Kabutoclub, a club of newspapermen at the Tokyo Stock Exchange (TSE), and were eventually incorporated into TSE rules. 7 These rules strongly encourage managers of listed firms in Japan to provide regular forecasts of sales and earnings. Our evidence indicates that the large majority of companies comply with this request and so we argue that disclosure in Japan is effectively mandated. The specific requirements for management forecast disclosures are as follows: 1. Listed companies are expected to release point forecasts of annual earnings at each annual earnings announcement date, as well as revisions of these 7 In their original form, these rules required forecasts of sales, net income, and dividends, along with a summary of financial statement information. The rules are similar on all of the major Japanese stock exchanges, including the TSE, Osaka Stock Exchange, and the JASDAQ, which are the major exchanges in Japan. 6

8 forecasts at interim earnings announcement dates. Thus, forecasts for year t are provided when year t-1 earnings are announced, and revisions (which include confirmations of the previous forecast) are provided when interim earnings are announced. There is no requirement that firms release quarterly forecasts (quarterly reporting has only been required in Japan since March 2003). 2. Managers are expected to forecast sales, earnings before extraordinary items and taxes (EBET), and net income (NI). 3. Forecasts must be updated if there are significant revisions in management estimates, defined as changes in estimated sales of 10% or more and/or changes in either earnings number of 30% or more (hereafter, the Significance Rule ). In contrast to initial forecasts, which are encouraged by stock exchange listing rules, revisions due to significant changes are required under the Act. Although these rules have been in place for some time, our data begins in fiscal (In Japan, most companies have a March 31 year-end, so that fiscal 1997 is the year-ended March 31, 1998.) 2.2 Hypotheses The management forecasting literature has largely developed in the US, where management earnings forecasts are not required and so are characterized as voluntary disclosures. In this setting, analytical papers use adverse selection to argue that, to maximize firm value, managers of firms with good news (above a certain threshold that depends on the cost of disclosure) disclose that news while managers of other firms withhold earnings news (e.g., Verrecchia, 1983; Dye, 1985; 7

9 Jung and Kwon, 1988). Empirical papers confirm that forecasts of annual EPS tend to convey good news and that managers release forecasts when their firms are doing well (Patell, 1976; Penman, 1980; Lev and Penman, 1990; Miller, 2002). On the other hand, there is also evidence that managers of firms with adverse earnings news disclose that news to investors through management forecasts, especially at shorter horizons (e.g., Skinner, 1994; Kasznik and Lev, 1995), suggesting that managers face an asymmetric loss function when reporting earnings surprises. We develop hypotheses regarding several aspects of management forecasting in Japan: (1) properties of management s initial annual earnings forecasts, (2) properties of forecast revisions, (3) the informativeness of management forecasts, (4) managers forecast strategies through time, and (5) differential forecasting incentives for earnings versus sales forecasts. Our first hypothesis concerns how managers set their initial earnings forecasts for each fiscal year. There are three possibilities: (i) managers set initial forecasts in an unbiased, good faith manner (so that forecast errors, on average, are zero), (ii) managers set forecasts conservatively, to be sure realized earnings will meet or beat forecasts, (iii) managers set forecasts aggressively, above their good faith estimates. We argue that (iii) is most likely for several reasons. First, managers are likely to benefit from optimistic forecasts because: (a) if viewed as credible by capital market participants, optimistic forecasts are likely to increase stock prices, (b) providing good news about the firms future earnings prospects helps managers convince the firms constituents (stockholders, the board, lenders, employees, suppliers, etc.) that managers are doing a good job. In Japan, these initial forecasts precede the annual stockholders meeting, when managers are likely 8

10 be especially concerned with how firm performance is perceived. 8 Second, the cost of releasing overly-optimistic initial forecasts is likely to be low because forecasts can subsequently be revised downward at low cost. This is true because managers are required to revise their forecasts at least once during the fiscal period (more often if there are significant changes in expectations) and because litigation costs in Japan are not significant. An alternative possibility is that reputation and credibility concerns cause managers initial forecasts to be unbiased. If managers consistently make overlyoptimistic forecasts, investors and other constituents are likely to factor this into their responses, reducing the benefits of optimistic forecasts. Moreover, managers who consistently release overly-optimistic forecasts are likely to lose credibility with market participants, which reduces the credibility of their future communications and possibly also their overall reputation as effective managers (Hutton and Stocken, 2006). The third possibility is that managers set initial forecasts conservatively, at or below their good faith earnings estimates. This would occur if the costs of subsequently revising forecasts downward are large and managers have incentives to meet or beat expectations. For reasons outlined above, we believe the cost of revising forecasts is likely to be low, which makes this third possibility unlikely. Consequently, we test the following null hypothesis against the alternative that managers initial earnings forecasts are unduly optimistic: H1: Managers initial annual earnings forecasts are unbiased. 8 It may also be the case that these forecasts are linked to the firms internal budgeting process, and so may also be used to help motivate employees perform better, in which case we would also expect optimistic forecasts. 9

11 Our prediction that managers in Japan set initial forecasts that are overlyoptimistic is consistent with previous research in the US. Rogers and Stocken (2005) present evidence that managers are more likely to bias their forecasts upward when (i) they have incentives to do so, and (ii) investors are less likely to detect the over-optimism (both of which are generally true in Japan), while Kothari et al. (2005) provide evidence that managers accelerate the release of good earnings news and delay the release of bad earnings news. Recent papers by Richardson et al. (2004) and Cotter et al. (2006) show that analysts earnings forecasts at the beginning of each fiscal year tend to be too optimistic but that expectations are subsequently revised downward through the release of management earnings guidance, so that earnings surprises are non-negative (see also Soffer et al, 2000). This is consistent with what we expect in Japan where managers can manage expectations directly by releasing forecasts. In the US, where litigation is an important consideration, managers are more likely to manage expectations indirectly by guiding analysts. 9 The benefits of overly-optimistic forecasts are likely to be greater for managers of firms whose current performance is poor and for whom earnings news is likely to relatively more important for stock prices. Managers of firms with poor current earnings will seek to convince stockholders, the board, lenders, and other constituents that their firms performance will improve, giving them stronger 9 Consistent with this, Kothari et al. (2005) point out that evidence on the disclosure of earnings news in the US, both their evidence and that in other research, suggests that US managers strategically leak good news early on in the fiscal period but delay the release of bad news, which tends to be released explicitly through an earnings forecast. In Japan, where litigation is a non-factor, managers can manage expectations directly through the release of earnings forecasts. 10

12 incentives to issue optimistic forecasts. 10 This is likely to be especially important in Japan, where the annual earnings release coincides with the shareholders formal invitation to the annual shareholders meeting, so that the information released on this date is likely to be important when shareholders evaluate managers performance for the year. Managers of growth firms, whose stock prices are relatively more sensitive to investors expectations about future earnings, are also more likely to release optimistic forecasts (e.g., see Lakonishok et al., 1994; Skinner and Sloan, 2002). These arguments lead to H2a and H2b (also stated in null form): H2a: Other factors held constant, there is no relation between the bias in managers initial earnings forecasts and their firms contemporaneous realized earnings performance. H2b: Other factors held constant, there is no relation between the bias in managers initial earnings forecasts and their firms stock price sensitivity to earnings news. We have data on management forecast revisions for the last five years of our sample period, from (fiscal) 2002 through We predict that managers revise their initial forecasts to bring them into line with subsequent realizations because earnings surprises, especially adverse earnings surprises, are likely to be costly for managers. In the US, there is evidence that managers carefully manage expectations to ensure that they can meet or beat analysts earnings forecasts (Skinner, 1994; Soffer et al., 2000; Matsumoto, 2002; Richardson et al, 2004; Cotter et al., 2006), apparently because of the costs of disappointing investors (Skinner and Sloan, 2002; 10 Rogers and Stocken (2005) report a similar result for US firms: that managers of distressed firms are more likely to issue optimistic forecasts than managers of other firms. 11

13 Kasznik and McNichols, 2002). While there is evidence that this is at least partly attributable to litigation (Skinner, 1997), it is also likely that managers bear reputation costs and lose credibility with investors and other constituents if they report adverse earnings surprises (Skinner, 1994; Tucker, 2005; Hutton and Stocken, 2006). Because litigation costs are unlikely to affect disclosure practices in Japan, to the extent that we find evidence that the loss function faced by Japanese managers is asymmetric (i.e., negative earnings surprises are more costly than positive earnings surprises), our evidence supports the importance of reputation/credibility effects in disclosure. More generally, we expect that Japanese managers bear costs when their firms announce earnings surprises, similar to the original expectations adjustment hypothesis of Ajinkya and Gift (1984), which leads to our next two hypotheses: H3: Managers earnings forecast revisions adjust expectations towards subsequent earnings realizations; when initial forecasts are overly-optimistic, forecasts revisions are negative, and vice versa. H4: Earnings surprises (the difference between realized earnings and the most recent management forecast revision) are small in absolute value. It is arguably the case that the reputation costs of releasing large earnings surprises in Japan are larger than those in the US and other countries because Japanese managers are required to release/revise forecasts on a regular basis; that is, based on regulatory requirements, investors and other constituents expect any changes in managers expectations to result in prompt forecast revisions. This means that earnings news (meaning any difference between actual and expected 12

14 earnings) will be especially surprising to investors, and that the reputation costs of earnings surprises may be larger than in the US. 11 Our next hypothesis concerns managers forecasting behavior over time. In the US and most other countries, where forecasting is relatively infrequent, it is difficult for researchers to study managers forecasting practices over time. One advantage of the Japanese setting is that we have data on how managers forecasting decisions in one year are related to those in the following year. We argue above that managers initial earnings forecasts are likely to be systematically optimistic, that this optimism is likely to be more pronounced among firms that are performing poorly, and that forecast revisions issued during the rest of the fiscal year are likely to correct this over-optimism. To the extent that managers forecast decisions in a given fiscal year are consistent with this, we expect that their forecasts in the next fiscal period are more likely to be unbiased. This follows because the benefits managers obtain from overly-optimistic forecasts are likely to dissipate over time as investors and other constituents observe the over-optimism ex-post (i.e., they observe the negative forecast error). If managers obtain a reputation for overlyoptimistic forecasts, the credibility of their forecasts will decline, making it less likely that stock prices respond to the forecasts and/or that they help convince other constituents that managers are doing a good job (see Hutton and Stocken, 2006). Consequently, if managers reputation for making credible forecasts depends on the accuracy of their previous forecasts, we expect that managers who make optimistic 11 In the spring of 2003, Sony announced a large adverse earnings surprise which resulted in a two day decline in its stock price of 27%. This event arguably led to the ouster of Sony s then CEO. (Sony is also listed on the NYSE, which means that it is subject to litigation in the US. However, as a practical matter, foreign firms cross-listed in the US are not subject to nearly the same level of litigation risk as US firms; e.g., Siegel, 2005.) 13

15 forecasts in one year are less likely to issue optimistic forecasts in the next year, and test the following null hypothesis: H5: Other things held constant, there is no relation between the bias in managers initial earnings forecasts for year t and the bias in managers forecasts in year t+1. Our final hypothesis exploits another unique feature of the Japanese setting the fact that Japanese managers are required to forecast sales, EBET, and NI (in the US, managers typically only forecast earnings; sales forecasts are comparatively rare; e.g., Hutton et al, 2003). To the extent that earnings (NI or EBET) is more important to investors and other constituents for valuation and contracting purposes, we argue that managerial optimism will be more pronounced for earnings forecasts than sales forecasts, and so test the following (null) hypothesis: 12 H6: The tendency for managers to bias their initial forecasts and correct that bias in subsequent forecast revisions to avoid reporting forecast surprises on the announcement date is similar for sales and earnings forecasts. 3. Sample and Descriptive Information Our data are from Nikkei Financial Quest, a commercial database provided by the subsidiary of Nikkei, publisher of the main business newspaper in Japan. The initial sample of firm/years with forecast data is shown in Panel A of Table 1 and includes data for 1997 through 2006 (fiscal years). The sample covers almost all listed firms in Japan. We use consolidated financial statements if they are available 12 We do not take a position on whether NI or EBET is likely to be more important. From an equity valuation perspective, EBET (a better measure of ongoing core earnings) is likely to be more important while from a contracting perspective, NI is likely to be more important (since it is more reliably measured and so more likely to be used in contracts). 14

16 and, if not, we use parent company financial statements (in Japan, consolidated financial reporting was required beginning in 2000). As Table 1 indicates, 81% of our firm/year observations are based on consolidated numbers. The column labeled missing indicates the number of firm/years without forecast data on the database in a given year. Missing observations arise either because the firm does not provide a forecast or because no forecast is available on the database. 13 However, the overall number of such missing observations is small around 4% of all firm/years and declines to 2-3% in more recent years. This fraction includes firms in the securities industry, which are exempt from the Timely Disclosure Rule. The fact that such a high proportion of firm/years have available forecast data supports our contention that forecasting is effectively mandated in Japan. To include a given firm/year t in the sample, we require realized and forecast numbers for year t as well as realized numbers for year t-1 (for sales, EBET, and NI). 14 We use these data to compute forecast innovations which we define as the difference between the forecast for year t and the corresponding realization for year t-1. Because these numbers are released simultaneously, innovations indicate how management s expectations for the current period compare to immediate past 13 For the set of firm/years with consolidated financial data available, we have broken down the missing observations into those that represent firms that did not provide forecasts and those that represent firm/years for which the forecast is not available in the database. Only about 62% of these missing observations are firm/years for which the forecast itself is missing (as opposed to the data being missing in the database), reducing the overall fraction of firm/years without forecasts to less than 4%. In addition, a larger fraction of consolidated than parent-only observations are coded as missing (overall, the fractions are 6.4% versus 4.4%), suggesting that firms may be more likely to have provided forecasts at the parent than the consolidated level, which seems plausible given that consolidated reporting was being introduced during this period. 14 A small number of companies did not release the full set of forecast information (Sales, EBET, and NI), which explains why we have a slightly different number of observations for these variables in our tables. 15

17 realized performance, which provides evidence on how management sets their firms earnings forecasts. Panel B of Table 1 reports descriptive statistics for the sample firm/years, presented by year and overall (we present overall numbers, rather than giving separate consolidated and parent numbers). 15 Mean (median) total assets is 403 billion ( 33 billion) while mean (median) market capitalization is 103 billion ( 14 billion). Several of the reported series reflect the relatively weak state of the Japanese economy during much of this period and the improvement that began in (calendar) The overall mean (median) debt-to-equity ratio for these firms is 4.22 (1.44) but falls noticeably after Overall mean (median) profitability for the sample (measured as NI-based ROA) is 1.0% (1.4%) but increases after Market-to-book ratios are low by US standards with an overall mean (median) of 1.75 (1.02). 16 The forecast data to this point describe firm/years for which we have at least the initial annual forecast, meaning the forecast that is usually released at the annual earnings announcement date (i.e., at the beginning of the fiscal year). Beginning in fiscal 2002 Nikkei began collecting forecast revisions as well, which we use to 15 To economize on space, we do not report the industry distribution of sample firms. However, because of the mandatory nature of forecasting, our sample essentially includes all listed Japanese firms and so has the same industry distribution. Details are available upon request. 16 The first set of ROA numbers reported in Panel B are based on NI deflated by lagged total assets. Because managers also forecast EBET, we also compute the ROA numbers using this measure of earnings. Mean (median) ROA based on EBET is 4.0% (3.4%), substantially higher than the NIbased measure. Because extraordinary items in Japan are defined to include gains and losses on the sales of securities and fixed assets (including real estate), the difference between these numbers is often relatively large. In addition, because during this period many Japanese firms were divesting themselves of their relatively large securities portfolios (as part of a gradual unwinding of the keiretsu system) at a time when Japanese equity prices were relatively low, these firms often reported relatively large extraordinary losses during this period, which explains why the ROA numbers based on NI are relatively lower than those based on EBET. 16

18 provide evidence on how managers revise their forecasts during the fiscal period. We summarize these data in Panel A of Table 2 and again separate the observations into those for parent and consolidated numbers. The other columns report the number of forecast revisions. Apart from 2006, for which we do not have a full year of data, the numbers in the 2nd column (which denotes the first revision) are close to those in the initial forecast column, indicating that managers almost always provide at least one revision. This is consistent with the Timely Disclosure Rules because firms are expected to provide forecasts with both annual and interim earnings announcements. The numbers in the subsequent columns indicate that managers of many firms follow up these forecasts with additional revisions, and that in some firm/years these revisions are numerous. We find that managers provide a second revision in 68% of firm/years, a third revision in 51% of cases, a fourth revision in 24% of cases, and a fifth revision forecast in 7% of cases, with managers of a few firms revising more often than this. 17 Panel B of Table 2 provides information on when management provides earnings forecasts. As expected, the large majority (over 90%) of initial forecasts are released at the annual earnings announcement date with most of the rest being released on a stand-alone basis (i.e., not at an earnings announcement date). Most of the first forecast revisions are also released on earnings announcement dates, 17 The number of forecast revisions in Japan is more numerous than in the US. For example, Anilowski et al. (2006) find that for the set of firm/years when annual earnings forecasts are released, managers release a single forecast about half the time (51% of observations), two forecasts 18% of the time, three forecasts 11% of the time, four forecasts 8% of the time, and five or more forecasts in the remaining 12% of cases. Ajinkya, Bhoraj and Sengupta (2005) also provide evidence on the use of multiple forecasts in US data. 17

19 especially after quarterly reporting is introduced. In 2002, 33.9% of these revisions occur on a stand-alone basis while 63.8% were released with the interim earnings announcement. In 2003, however, when quarterly reporting was introduced, 18.3% of revisions occurred on a stand-alone basis while 30.4% were released with interim earnings and 49.9% with first quarter (Q1) earnings. In 2004, only 10.0% of these forecasts revisions were issued on a stand-alone basis while most (70.3%) were released with Q1 earnings and 17.9% with Q2 earnings. The 2005 proportions, at 9.7%, 73.9%, and 15.0%, are similar to those in (We do not discuss the numbers for 2006, since our data are incomplete in this year.) Second revisions display a similar pattern, with most released with Q2 earnings. In 2002 around half of these revisions are stand-alone (48.2%) while the other half (47.2%) are released on the interim earnings announcement date. In 2003, 32.1% are released on a stand-alone basis while 59.0% are with Q2 earnings. These proportions change to 29.3% and 61.6%, respectively, in 2004 and to 28.0% and 63.4% in Finally, third revisions are made mostly on the Q3 earnings announcement date with some made in conjunction with Q2 earnings and fewer again on a stand-alone basis. The proportions are 6.9%, 7.8%, and 85.2%, respectively, in 2002; 53.6%, 24.6%, and 21.8%, respectively in 2003; 61.7%, 27.0%, and 11.2%, respectively, in 2004; and 63.3%, 26.5%, and 10.2%, respectively, in Overall, these results indicate that the large majority of earnings forecasts in Japan fall on earnings announcement dates, a tendency that has become stronger since the advent of quarterly reporting. 18 A large fraction of the 18 This trend is also apparent in US data: Anilowski et al. (2006) find that there is an increasing trend for US firms to issue earnings forecasts on quarterly earnings announcement dates. 18

20 stand-alone forecasts (around 70%) are attributable to the Significance Rule (results available upon request). 4. Hypothesis Tests To provide evidence on whether managers initial forecasts are biased (H1), Table 3 reports forecast innovations and forecast errors. Forecast innovations are forecast earnings for year t minus realized earnings for year t-1, deflated by lagged total assets. Forecast errors are realized earnings for year t minus forecast earnings for year t, also deflated by lagged total assets. We provide results for the sales, EBET, and NI forecasts. To investigate whether managers issue more optimistic forecasts when their firms are performing poorly (H2a), we partition the data into quintiles based on realized profitability (ROA) in year t-1, where quintile 1 contains firm/years with the lowest profitability. 19 The results in Table 3 support the prediction that managers sales and earnings forecasts are systematically optimistic: on average, forecasts innovations are large and positive while forecast errors are large and negative. For the sample overall, sales are forecast to increase by a mean (median) of 6.1% (3.0%) of assets, while EBET is forecast to increase by 1.2% (0.5%) and net income (NI) by 1.7% (0.5%). 20 For all three variables, over three-quarters of the innovations are positive. It is notable in the context of H6 that forecast increases in NI (mean [median] of 1.7% [0.5%]) are about the same as those for EBET (1.2% [0.5%]) even though the levels of the latter variable are generally larger. If it the bottom-line NI number is 19 Results are based on ROA calculated using NI in the numerator. We have also computed ROA using EBET in the numerator, with similar results. 20 All of the means and medians in this table are statistically significantly different from zero except the mean of quintile five. 19

21 more important (to shareholders, lenders, employees, or others), managers have stronger incentives to be optimistic about NI than EBET. 21 Perhaps even more interesting, managers of the most profitable firms (quintile 5) forecast the largest sales increases the mean/median sales innovations are 2-3 times larger than those for the least profitable firms but forecast lower earnings increases. This may be evidence that managers forecast incentives differ for sales and earnings. 22 When we look at how sales innovations vary across the past ROA quintiles, we see that, contrary to H2a, there is an almost monotonic increase in the mean (median) sales innovation as we move from the worst to best ROA quintiles managers of the most profitable firms forecast the largest sales increases. 23 The pattern is different for the earnings innovations. While there is evidence for EBET that managers of firms in the best performing quintile expect better EBET growth than those in quintiles 2-4, the most optimistic managers are those of the worst performing firms, in quintile 1. These managers expect EBET to increase by 3.5% (1.8%) of assets, which is larger than the increase forecast by managers of the best performing firms, of 1.1% (0.7%) [90% of the innovations for quintile 1 are 21 Alternatively, it could simply be that extraordinary items are not as easily forecast as other income statement line items (perhaps because decisions about selling securities, taking write-offs, etc., are not taken until relatively late in the fiscal period) and that these items are initially forecast to be zero but often turn out to be large and negative. 22 We plan to investigate these differences more systematically in the next draft because they suggest an inconsistency across the quintile 1 managers sales and earnings forecast: sales are forecast to increase by 4.3% (2.0%) while NI is forecast to increase by 7.4% (4.0%), which is suggestive of greater inflation for NI. While there are legitimate explanations for this (a large improvement in net margin due to reversal of large one-time expenses) the comparison with quintile 5 (whose managers forecast a sales increase of 11.4% (6.0%) but basically flat earnings (-0.2% (0.4%))) makes the quintile 1 numbers questionable. 23 The exception to this pattern is that mean and median sales innovations in the lowest ROA quintile fall between those of quintiles 2 and 3, perhaps reflecting the fact that managers of the worst performing firms (some of which report losses) expect sales to increase more than their past earnings performance would suggest, consistent with mean reversion. 20

22 positive, compared to 76% in quintile 5]. 24 Managers of firms in the intermediate portfolios expect increases between these extremes with none being as optimistic as the managers in quintile 1. This pattern is even more pronounced for NI: managers of quintile 1 firms expect NI to increase by 7.4% (4.0%) [95% positive] compared to -0.2% (0.4%) [68% positive] for managers of quintile 5 firms. 25 The optimism for managers in quintile 1 could reflect either more pronounced mean reversion for the worst performers (which include loss firm/years), so that managers good faith forecasts are most optimistic for these firms. On the other hand, consistent with our hypotheses, it may be that managers of these firms are systematically optimistic to try and convince investors, the board, and others that their firms performance is improving (thus preserving their jobs) and/or to help motivate their employees to do better. The fact that these results hold for earnings but not sales forecasts is consistent with H6 and managers being opportunistically optimistic. Also consistent with H1, the forecast errors that we report in Table 3 indicate that managers forecasts are systematically optimistic relative to subsequent realizations: all of the means and medians are negative and statistically significant. Looking first at sales forecasts, the overall mean (median) forecast error is -2.8% (- 1.7%) and only 38% are positive. 26 The largest (most negative) forecast errors are evident for the firms with the worst past performance, indicating that managers of these firms are the most optimistic. For these firms the mean (median) sales forecast 24 Differences in both the means and medians of these quintiles are statistically significant. 25 Numbers for the other quintiles are similar to those of the best performing quintile (especially looking at the medians) although the proportion of positive observations declines monotonically across the quintiles suggesting that managerial optimism is inversely related to past ROA. 26 The forecast error means and medians reported in this table are all statistically significantly different from zero. 21

23 error is an economically significant -4.8% (-2.9%); only 32% of these forecast errors are positive. There is also a tendency for the forecast errors to become less negative as we move from quintile 1 to quintile 5. Similar but stronger patterns are evident for the earnings forecasts, supporting the evidence that managers earnings forecasts are overly-optimistic, especially those issued by managers of the poorest performing firms: firms with the worst profitability display both the most positive innovations and the most negative forecast errors. For EBET, the mean (median) forecast error for the quintile 1 firms is -1.8% (-0.6%) [34% are positive] while that for the NI measure is -3.3% (-0.9%) [29% positive] compared to means (medians) for the overall sample of -0.8% (- 0.2%) [42% positive] and -1.4% (-0.4%) [36% positive], respectively. These results are again more pronounced for the NI than the EBET forecasts. To the extent that managers believe that investors focus more attention on bottom-line NI numbers than on EBET, this supports the idea that managers are deliberately overstating the results. It is possible that these results, rather than reflecting systematic managerial optimism, are simply due to unpredictable changes in economic conditions; i.e., that the forecasts were made in good faith but that the negative forecast errors are attributable to one or two years when the Japanese economy unexpectedly changed for the worse after initial forecasts were issued. Table 4 reports forecast innovations and forecast errors for the sample partitioned by fiscal year, from 1997 through Although there is year-to-year variation in both the forecast innovations and the forecast errors, the results are largely consistent across years: mean and median innovations are positive and statistically significant in all years (except for NI in 22

24 2006) while mean and median forecast errors are usually negative and statistically significant. The forecast errors are less negative beginning in 2003 (e.g., the medians in 2003 to 2005 are not reliably negative), which is when the Japanese economy began to improve after its long slump. One interpretation is that managers in Japan expected an economic rebound in every year (which explains their optimistic forecasts) but that these expectations were only realized in 2003 and thereafter when the economy actually improved. 27 Table 5 provides evidence on H2b: whether managers of growth firms (with high M/B ratios) are more likely to provide optimistic forecasts than managers of value firms. To conduct these tests we sort firm/years into deciles based on M/B ratios at the fiscal t-1 year-end. We only report results for the two earnings variables in these tables. The results indicate that managers of high M/B firms tend to forecast larger increases in earnings than managers of other firms although the pattern is somewhat U-shaped, with larger increases being forecast in both of the extreme deciles and a generally monotonic increase from decile 3 through decile 10 (the proportion of positive innovations increases monotonically across all 10 deciles). Unlike Table 3, however, the across-decile variation in forecast innovations does not translate into variation in forecast errors: the median forecast errors are largely flat across deciles while the proportion of positive forecast errors increases monotonically across deciles, indicating that managers optimistic bias, 27 This suggests another explanation for managers systematic over-optimism to help their lenders justify the continued extension of credit to these firms. As discussed by a number of economists (Peek and Rosengren, 2005; Caballero et al., 2006), during the long downturn in Japan many banks engaged in a practice known as evergreening the continued extension of credit to essentially insolvent zombie firms a practice that allowed the banks to continue to postpone the recognition of their bad loan problems. 23

25 which is again pervasive overall, actually decreases with M/B decile. Thus, managers of growth firms forecast and achieve larger earnings increases. To address H3, we next provide evidence on how managers revise their forecasts during the fiscal year. To conduct these tests, we divide the forecasts into six time periods (T) according to when they are released during the fiscal year: (i) T1 denotes forecasts released at the annual earnings announcement, (ii) T2 denotes forecasts released after T1 through the first quarter s earnings announcement date, (iii) T3 denotes forecasts released after T2 through the second quarter s earnings announcement date, (iii) T4 denotes forecast revisions released after T3 through the third quarter s earnings announcement date, (iv) T5 denotes forecast revisions released after T4 through the remainder of the fiscal year, (vi) T6 denotes forecast revisions released after the end of the fiscal year but before the annual earnings announcement date. We report statistics on the forecast errors and forecast revisions for forecasts issued during each of these periods in Table 6. These tests are based data from , the years for which we have data on revisions. The forecast error/revision statistics in Table 6 indicate that managers generally revise their forecasts downward during the fiscal year, as we would expect if managers act to correct their initial forecast optimism. First, consistent with Table 3, we see that for initial forecasts released on the annual earnings announcement date (T1), forecast innovations are systematically positive while forecast errors are systematically negative. These results are not as strong as those in Table 3 because they exclude the earlier years ( ) when managers forecasts were most optimistic (Table 4). Second, the large majority of forecasts issued during T2 (which includes the first quarter earnings announcement) are confirmatory forecasts 24