Whisper forecasts and earnings management

... summary and conclusions in Section VI II Background and Hypotheses Development II Whisper forecasts Analysts’ Behavior and Earnings Management Academics and financial press writers define whisper forecasts. .. the Whisper Forecast, the Analysts’ Forecast and Reported Earnings for Quarters when both Analysts’ Forecasts and Whisper Forecasts are Available Figure Distribution of Forecast Errors and Whisper. .. value-relevant are analysts’ and whisper earnings forecasts to investors? and 2) Are whisper forecasts relevant enough to investors that managers engage in earnings management to meet or beat this

Whisper Forecasts and Earnings Management A DISSERTATION SUBMITED TO THE FACULTY OF THE GRADUATE SCHOOL OF THE UNIVERSITY OF MINNESOTA BY: Arnoldo Jose Rodriguez IN PARTIAL FULFILMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY Judy A. Rayburn April, 2005 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. UMI Number: 3165898 INFORMATION TO USERS The quality of this reproduction is dependent upon the quality of the copy submitted. Broken or indistinct print, colored or poor quality illustrations and photographs, print bleed-through, substandard margins, and improper alignment can adversely affect reproduction. In the unlikely event that the author did not send a complete manuscript and there are missing pages, these will be noted. Also, if unauthorized copyright material had to be removed, a note will indicate the deletion. ® UMI UMI Microform 3165898 Copyright 2005 by ProQuest Information and Learning Company. All rights reserved. This microform edition is protected against unauthorized copying under Title 17, United States Code. ProQuest Information and Learning Company 300 North Zeeb Road P.O. Box 1346 Ann Arbor, Ml 48106-1346 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. UNIVERSITY OF MINNESOTA This is to certify that I have examined this copy o f a doctoral Dissertation by Arnoldo J. Rodriguez And have found that it is complete and satisfactory in all respects, and that all revisions required by the final examining committee have been made Judy A. Rayburn Signature 4 Y t Date GRADUATE SCHOOL Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Acknowledgments I gratefully acknowledge useful comments by and discussions with John Dickhaut, Judy Rayburn, Pervin Shroff, Susan Watts, Marc Bagnoli, Gerard McCollough, Tong Lu, Jack Stecher, Luis Sanz, Niels Kettelhohn, Nicolas Marin, Esteban Brenes, Carlos Quintanilla and participants at the XI Latin American Congress o f Internal Auditors. i Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Abstract This paper examines the recent market phenomenon of whisper forecasts of earnings and whether their appearance affected investors, managers and market behavior. We explore the relation between official earnings per share analysts’ forecasts, realized earnings, and whisper forecasts. We find that the mean analysts’ forecast error for a sample of growth firms has increased over time and shows a pessimistic bias. When whisper forecasts are used as earnings estimates, no bias is evident and the mean whisper error is significantly lower than the mean analysts’ forecast error. The unexpected component of earnings better explains abnormal returns around the earnings announcement date when whisper forecasts are used as earnings expectations instead of analysts’ forecasts. In view of recent evidence, that managers manipulate earnings to meet or just beat earnings thresholds, we test whether managers regard whisper forecasts as a relevant threshold to meet or beat. We find that firms that were able to meet or just beat the whisper forecast reported higher abnormal accruals when compared with a cross-section of firms in the same industry. This finding is consistent with the hypothesis that whisper forecasts were not only a more accurate predictor of earnings, but also a market relevant threshold that provoked accrual manipulation by managers to reach aggressive unofficial estimates. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table of Contents Page I. Introduction 2 II. Background and Hypotheses Development 6 HI. Sample Selection and Description 16 IV. Research Design and Methods 17 V. Results 21 VI. Concluding Remarks 28 iii Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. List of Tables Page Table 1. Comparative Firm Characteristics 35 Table 2. Separation of the Groups 36 Table 3. Historical Comparison of Earnings Performance 37 Table 4. Results of the Kotmogorov Smirnov Two Sample Test 38 Table 5. Comparison of Analysts’ Forecast Errors for the Sample of Growth Firms 39 Comparison of Analysts’ and Whisper Forecast Errors for the Sample of Firms for which Whisper Forecasts are Available 41 Table 7. Explanatory Power of Forecast Error and Whisper Error 43 Table 8. Mean Cumulative Abnormal Returns by Group 44 Table 9. Earnings Management Discretionary Accruals by Group 45 Table 10. Earnings Management Discretionary Accruals by Group 46 Table 6. iv Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. List of Figures Figure 1. Comparison of the Whisper Forecast, the Analysts’ Forecast and Reported Earnings for Quarters when both Analysts’ Forecasts and Whisper Forecasts are Available Figure 2. Distribution of Forecast Errors and Whisper Errors v Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. I. Introduction “Whisper forecasts of earnings” or “whisper numbers” are unofficial estimates of earnings made available to the general public prior to the release of quarterly earnings. Whisper forecasts first appeared in the mid-nineties. The business press argues that whisper forecasts gained popularity when the official analysts’ forecasts, especially for growth firms, were believed to have downward bias. The hypothesis was that this downward bias was induced by analysts to maintain good client-relations by offering an “easy to beat” earnings benchmark 1 ‘j This hypothesis is consistent with the following anecdotal observation. : “Analysts, investors, and corporate managements observed that companies that exceeded expectations outperformed in the short run, while those that met expectations did not. This led companies to under-promise and over-deliver, and led analysts to be conservative in their published reports in order to be able to write that the company was doing better than expected and therefore its stock should outperform. Another factor in the whisper forecasts game was that prices were so expensive that even the most obtuse analyst could understand that prices could not be justified by any reasonable set of published estimates, but only by results that were better than expected (exactly how much better they had to be to underpin valuations was ignored)." ‘Empirical evidence (for example, Myers and Skinner (1998)) suggests that companies, especially in growth industries, that meet or beat analysts’ expectations are priced at a premium, whereas those that disappoint suffer disproportionately in the market (Skinner and Sloan(2001)). The unprecedented increase in stock-based compensation, especially for growth firms, also provides managers with strong personal incentives to meet or just beat analysts’ expectations (Gaver, Gaver, and Austin (1995)). 2 www.ragingbull.com. Michael Pearson,: “Are whispers for real”, 1999. 2 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Research by Bagnoli, Beneish, and Watts (1999) documents that whisper forecasts are more accurate predictors of quarterly earnings than analysts’ forecasts, even after controlling for the differential timing o f the release of these numbers. They further show that trading strategies based on the relation between analysts’ forecasts and whisper forecasts earn significant excess returns. If whisper forecasts better reflect market expectations, as the evidence seems to suggest, we argue that managers will treat these forecasts as their quarterly earnings target. Previous literature on whisper forecasts focused on the accuracy and pre-eamings announcement returns for firms for which a whisper number was available, but they left unanswered questions, e.g., how the market reacts to whisper forecasts around earnings announcements, whether their presence affects managers’ behavior, and if so, which methods were used by managers to meet earnings targets . This paper tests whether a significant percentage of firms meet or just beat the whisper forecast of quarterly earnings. It also tests whether these firms engage in earnings manipulation to meet or just beat the whisper forecast threshold. In a sample of growth firms, we find that the mean optimistic bias in analysts’ forecasts declined over our sample period, 1990-2000. Consistent with popular press explanations for the appearance of whisper forecasts on the investment scene, the mean and median analysts’ forecast error for this sample in fact shows a pessimistic bias over the later sub-period, 1997-2000. On the other hand, for a sample of 140 growth firms for which whisper forecasts are available during the period 1997-2000, no mean bias in the whisper forecast is evident. Interestingly, we find no significant difference in the mean bias in analysts’ forecasts over the sub-period 1990-96 and the whisper forecast over the Prior research by DeGeorge, Patel, and Zeckhauser (1998) provides evidence consistent with managers attempting to meet or just beat analysts’ quarterly earnings forecasts. 3 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. sub-period 1997-2000. This evidence indicates that whisper forecasts during the late nineties were more in line with analysts’ forecasts during the early nineties (in terms of the systematic forecast bias.) On the other hand, over the sub-period 1997-2000, for quarters for which both analysts’ and whisper forecasts are available, we find significant pessimistic bias in analysts’ forecasts in contrast to the unbiased whisper forecasts. Consistent with previous research, we find that whisper forecasts are on average more accurate than analysts’ forecasts, as indicated by the mean absolute and squared forecast errors. The improved accuracy of whisper forecasts relative to analysts’ forecasts holds even after controlling for the timing advantage of whispers4. Whisper forecasts are more accurate than the mean o f analysts’ forecasts released within a period of thirty days preceding the earnings announcement. We find that for 84.2% of the sample firms the whisper forecast is greater than or equal to the analysts’ forecast. We find that 24% of our sample firms meet or just beat (by one cent) the whisper forecast, while 22% of firms meet or just beat analysts’ forecast5. Thus, it appears that for a significant number o f firms, managers may have been using the whisper forecast as an earnings threshold rather than the analyst forecast. The result is consistent with recent academic findings that managers try to avoid negative news related to earnings announcements6. Additional results indicate that managers of firms that meet or just beat the whisper forecast may engage in earnings manipulation. 4 It was possible to collect and send whisper forecasts up to the day prior to the earnings announcement. We use the mean forecast in the First Call Database as die analysts’ forecast. Therefore, there is a timing advantage for whisper forecasts relative to analysts’ forecasts. 5 The cases when the analysts’ forecast and the whisper forecast were separated by less than 1 cent are excluded from this analysis and represent about 14.5% o f the sample. The reason to do this is that under the 1 cent difference, it is not feasible to distinguish which o f the two earnings forecasts firms were trying to meet or beat. 6 Brown (2001), Burgsthaler and Eames (2000). 4 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. We find that abnormal accruals for firms that meet or just beat the whisper forecast are significantly higher than those for an industry-matched control group. Consistent with previous literature, we also find that firms that meet or just beat the analysts’ forecast, show higher abnormal accruals than the industry matched control group. Consistent with anecdotal evidence, we also find that unexpected earnings based on whisper forecasts are more closely associated with abnormal returns during earnings announcement periods than are unexpected earnings based on analysts’ forecasts. The result is not sensitive to different specifications of unexpected earnings based on analysts’ forecasts and whisper forecasts. Additionally, firms that were able to meet or beat the whisper number show higher positive abnormal returns than the control groups. Consistent with whisper forecasts being a relevant market threshold, firms that were able to beat the analysts’ forecast but not the whisper forecast show negative abnormal returns. Firms that were not able to meet the whisper had negative abnormal returns similar to those of firms that missed both the whisper and analysts’ forecast. The results of this paper provide insights into effects of whisper forecasts on investors, firms, and markets. Healy and Whalen (1999) criticize the lack of academic research to address specific questions about methods, circumstances and opportunities for earnings management. This paper is relevant to accounting literature because it addresses 1) earnings management behavior in the presence of more aggressive earnings targets, 2) the value relevance o f biased versus unbiased earnings forecasts, and 3) the emergence of new thresholds based on unofficial information, and their impact on the behavior of managers and investors. 5 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Section II defines whisper forecasts and motivates the hypotheses. Section III describes the whisper database used to conduct the tests. Section IV discusses the research design. Section V describes the results followed by a summary and conclusions in Section VI. II. Background and Hypotheses Development II. 1 Whisper forecasts. Analysts’ Behavior and Earnings Management Academics and financial press writers define whisper forecasts as: ■ Unofficial estimates of earnings communicated to and among investors before a company releases its quarterly earnings (Thestreet.com) ■ Real-time market estimate of earnings per share (Fool.com) ■ Investors’ assessment of a company’s true earnings potential (whispemumbers.com). Financial market observers report that whisper forecasts have existed since 1990. Before the mid-1990’s, the numbers were generated by sell-side analysts for preferred (wealthy) clients as a value-added service to remove the apparent bias in earnings estimates. Consistent with this statement is the apparent loss of trust that individual investors have in analysts’ opinions and estimates in recent times (attributable to the conflict of interest that is present inside some financial supermarket companies). Mike Thompson (director of research for BullDogResearch.com), refers to analysts’ opinions as “clearly providing a service and a marketing function to the investment-banking side o f their businesses”. 6 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. During the 1995-2001 period, analysts appeared to be unwilling to downgrade or to set aggressive earnings targets for companies that might have turned to their brokerages for future investment banking. In the case of private clients, analysts are motivated to increase clients’ returns. One way of achieving the latter objective is to take a conservative approach to their public earnings expectations (pessimistic bias) since companies they recommend to clients will report earnings that beat the analysts’ estimate. Whisper forecasts may be the reaction by investors who were now able to communicate electronically (‘whisper’) the unbiased earnings potential of firms at a very low cost. The whispers quickly spread across the Internet. On average, the whispers were more accurate than analysts’ estimates (FirstCall, I/B/E/S). The accuracy o f whispers was validated by specialized websites, such as www.whispemumbers.com, as well as academic research (Bagnoli et al. 1999), and consequently, whisper forecasts started to gain credibility among the investment community (CNBC, Bloomberg, CNNFN). By the mid 1990’s, whisper forecasts were commonly used by investors. As the Internet began its global development and expanded around 1997 it was common to observe financial reporters announce whisper forecasts in parallel with the firm’s earnings estimates. Because whisper forecasts are unofficial earnings estimates, their validity and informational properties were questioned. Bagnoli et al. (1999) documented basic characteristics of whisper forecasts and some whisper content not contained in analysts’ estimates (First Call). Specifically, Bagnoli et al. (1999) design tested the information content and the probability of achieving abnormal returns from trading strategies based on whisper forecasts; they measured returns from the date the whisper forecasts were disclosed to the date that analysts’ quarterly earnings were made public. They found that 7 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. whisper forecasts contain some eamings-relevant information not contained in the First Call analysts’ forecasts. They also found that by following a trading strategy based on whisper forecasts, analysts’ estimates, and their pre-announcement locations, an investor could attain additional returns by buying/selling stocks according to whether the whisper was above or below the analysts’ estimate. The authors did not examine how markets reacted to the earnings announcement in the presence of whisper forecasts or if managers engaged in earnings management practices to reach a whisper threshold (Bagnoli et al. 1999). Related research has shown that managers manipulate earnings to exceed thresholds. DeGeorge, Patel and Zeckhauser (1999) found that, for both market and psychological reasons, managers try to beat earnings thresholds. One threshold managers try to reach, with or without earnings management, is the quarterly earnings estimates issued by analysts. DeGeorge et al. (1999), and Burgsthaler and Eames (2001) provide evidence that firms manage earnings upward, particularly to meet analysts’ expectations. Kasznik (1999) found that managers use unexpected accruals to manage earnings upward when firms are in danger of falling short of managers’ earnings forecasts. Prior research assessed the impact of missing earnings expectations. Companies that are able to meet or exceed certain thresholds and avoid earnings disappointments are priced at a premium, while those that disappoint —especially growth firms —suffer disproportionately. (Myers and Skinner (1998), Skinner and Sloan (2001).) Collins and Kothari (1989) show that market reaction to earnings announcements is greater for growth firms (Note that whisper forecasts of earnings were circulated mostly for growth firms.) The increased use of executive options, especially for high-tech growth firms is a 8 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. related issue. The reward to senior executives depends on stock-price performance or earnings, or both (Healy, 1985, and Gaver, Gaver and Austin, 1995). For example, higher earnings imply higher management bonuses because annual bonuses are either a direct function of annual earnings or are a direct function of common stock value, which is related to reported earnings (Gaver, Gaver, and Austin, 1995; Healy, 1985). In summary, there exists a group of growth companies whose managers hold a large amount o f equity linked to stock market performance of their shares and financial markets that react disproportionately to earnings disappointments from these firms. We hypothesize that this observation provides an incentive for managers to engage in earnings management to beat a threshold . We now turn to an analysis of these thresholds. Analysts have incentives to bias forecasts to benefit private clients, maintain good informational relations with companies, and promote investment-banking activities with firms (Francis and Philbrick (1993)). Whisper forecasts may be a consequence of the relationship between companies and the analysts that follow them. Companies try to manage expectations downward and analysts may be rewarded when companies they recommend beat estimates. Thus, whisper forecasts are a reaction of investors to the observed pessimistic bias of analysts’ estimates. This raises two important questions: 1) How value-relevant are analysts’ and whisper earnings forecasts to investors? and 2) Are whisper forecasts relevant enough to investors that managers engage in earnings management to meet or beat this “new” threshold? 9 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. II.2 Accuracy, bias, and relevance of analysts’ forecasts of earnings when compared with whisper forecasts o f earnings II.2a Comparison of analysts’ forecast errors and whisper forecast errors To test the most general hypothesis that whisper forecasts were relevant to investors, we examine the change in analysts’ earnings forecast errors and whisper forecast errors during the sample period (1990-2000). Specifically, we compare the distribution of analysts’ forecast error for the period 1990-1996 (called Period I or pre-whisper period) and the period 1997-2000 (Period II or whisper period) to the distribution of whisper forecast errors for 1997-2000. If whisper forecasts are relevant and accurate there has to be a change in the accuracy of analysts’ estimates that justifies the existence of whisper forecasts. For Period I or the pre-whisper period, analysts’ forecasts for firms included in the sample were neutral or slightly pessimistic. If there was indeed a shift in the analysts’ bias due to the incentives previously discussed, we expect the mean analysts’ forecast error to increase from Period I to Period II, the period where whisper forecasts became available for these firms . The hypothesis is H I (a) Analysts’forecast error increases from Period I to Period II. 10 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. If the whisper number becomes a substitute for the analysts’ forecast, then whisper forecast errors should approximate the characteristics of analysts’ forecast errors before whispers. Thus, HI (b) The characteristics o f whisper forecast errors in Period II are not significantly different from those o f analysts’forecast errors in Period I. II.2b Comparison o f the overall distribution of analysts’ forecast errors and whisper forecast errors The shape of the distribution of analysts’ forecast errors before whisper forecasts has the common characteristics of distributions that reflect earnings management. A whispererror distribution similar to the distribution of analysts’ forecast errors before whispers is consistent with evidence that the firms that are part of that distribution also were involved in eamings-management practices. This test is also a test of the relevance of whisper forecasts; if it is possible to replicate the previous distribution o f analysts’ forecast errors with the distribution o f whisper errors, this is evidence not only that investors were more accurate in setting the whisper number, but also that companies could have moved their attention to focus on the unofficial earnings forecast (whispers), as proposed in the DeGeorge, Patel and Zeckhauser (1999) framework. Thus, H l(ci) The distribution o f analysts’forecast errors in Period I is not different from the distribution o f whisper errors in Period II. 11 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. If whisper forecasts were relevant and if companies were pursuing them, the distribution of analysts’ forecast errors would shift to the right as a consequence of more aggressive whisper earnings estimates. Thus, H l(cii) The distribution o f analysts’forecast errors in Period I is different from the distribution o f analysts’forecast errors in Period II. 11.2c Conflicts of interest, bias, and accuracy o f whisper forecasts If whisper forecasts were relevant to investors as a mechanism to remove the bias in analysts’ estimates, we should expect that whisper forecasts are more accurate and less biased than analysts’ forecasts. The possibility that whispers have a timing advantage over analysts’ estimates may confound the results. Therefore, for tests of this hypothesis, separate results will be presented for a sub-sample of analysts’ estimates. The sub-sample includes only analysts’ estimates that were released within a period of 30 days preceding the earnings announcement. Thus, to test for a shift in bias and accuracy of analysts’ forecasts that occurred during the 1990’s, we hypothesize the following H I (di) The bias and inaccuracy o f analysts ’forecasts are greater in Period II than in Period I. Hl(dii) The bias and inaccuracy o f whisper forecasts are smaller than that o f analysts’ forecasts. II. 3 Value Relevance of Whisper forecasts 12 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. II.3a Abnormal Returns and Unexpected Earnings According to DeGeorge, Patel and Zeckhauser (1999), “executives focus on thresholds, because the parties concerned with the firm do.” One of the relevant parties is the investor, both individual and institutional. If abnormal returns can be better explained by a simple statistic based on whisper forecasts (such as whisper errors) rather than on analysts’ forecast errors, it will be evidence that whisper forecasts were more closely related to true market expectations about earnings than were the analysts’ forecasts. Thus, the test will add validity to the concept of whisper forecasts as a piece of information relevant for markets and for valuation purposes. If whisper forecasts are not able to explain abnormal returns, their relevance will be questionable. Thus, H2 (a) Abnormal returns are better explained by whisper forecast errors than by analysts' forecast errors. II.3b Investors’ Reaction to Earnings Announcements Whisper forecasts are generally speaking more optimistic relative to analysts’ estimates. In order to isolate the effect that the whisper number has on investors’ and managers’ behavior, we divide the sample into nine different groups. The classification of groups will be discussed in detail later. The justification for dividing the sample is to facilitate the analysis of both the returns and the earnings management hypotheses. Based on empirical and anecdotal evidence, if whisper forecasts are relevant thresholds, we 13 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. should expect firms that meet or beat the whisper number to have positive abnormal returns. A confounding factor is that whisper forecasts and analysts’ forecasts are highly correlated. Therefore, for a number of observations, it would be hard to disentangle the effect on abnormal returns of beating the whisper versus beating the estimate. The proposed separation deals with that problem since it splits the sample on earnings performance based on analysts’/whisper estimates conditional on whether the whisper was similar (within +-1 cent), smaller, or larger than the analysts’ estimate. Evidence suggests that managers have strong incentives to meet a threshold in order to shelter their stock price, especially in growth firms. If whisper forecasts represent “true” market expectations, we should expect that even if firms were able to beat the analysts’ estimates but not the whispers, firms would be penalized by the market. If whisper forecasts are relevant to the market, we predict that: H2 (bi) Whisper errors are positively correlated with abnormal returns independently o f the sign o f analysts’forecast errors. H2 (bii) Analysts’forecast errors are positively correlated with abnormal returns independently o f the sign o f whisper errors. H2 (ci) Groups with positive unexpected earnings based on whisper forecasts have higher abnormal returns than groups with positive unexpected earnings based on analysts ’forecasts. H2 (cii) Groups with negative unexpected earnings based on whisper forecasts have more negative abnormal returns than groups with negative unexpected earnings based on analysts’forecasts. II.4 Managers’ reaction to whisper forecasts 14 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Discretionary Accruals DeGeorge, Patel and Zeckhauser (1999) showed that, “executives focus on thresholds, because the parties concerned with the firm do”. We assume that if markets/investors also care about whisper forecasts, it is likely that firms/managers care about whisper forecasts. The DeGeorge et al. (1999) paper also suggested that in the presence of a relevant threshold, companies manage earnings to meet or beat the threshold. To test whether companies manage earnings to beat the whisper estimate we identify companies that we suspect will engage in such practices. As in DeGeorge et al. (1999), we assume that companies that were able to meet or beat the estimate likely managed earnings. To test whether firms were using discretionary accruals to meet or beat whisper earnings forecasts, I use the same grouping of firms that is used for the returns analysis to study earnings management. Firms with a 0 or 1 cent whisper forecast error are suspected of using discretionary accruals to avoid earnings disappointments. Formally, the two hypotheses are: H3 (ai) Abnormal discretionary accruals fo r firms that ju st meet or beat the whisper earnings forecast are higher than those o f their industry peers. H3 (aii) The proportion ofpositive/negative abnormal accruals is higher fo r groups that ju st meet/beat the whisper forecast than fo r other groups. 15 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. III. Sample Selection and Description I manually created the whisper-numbers database from a set of diverse sources. Most important were the web sites that specialize in collecting the whisper forecasts. To create a database, several such web sites were visited and all the whisper forecasts related to the relevant sample were collected for the period between the fourth quarter of 1996 and the fourth quarter of 2000. Several message boards were visited, and with the aid of specialized search engines, nearly 3 million messages were browsed and searched using the keyword “whisper”. When a hit was obtained, the message was then read and if there was a whisper forecast for a specific company, it was included in the database. Also, specialized websites provided an average whisper forecast for companies from the selected sample. For each company quarter, the database contained quarterly whisper forecasts, the analysts’ earnings estimates from Zack’s Investment Services, and the stock-price movements the day after the earnings releases. The search yielded 544 whisper forecasts for about 150 companies. The median whisper number frequency per firm was close to 4. The companies in the sample are above average in terms of size, assets, sales, and market capitalization. The analysts’ estimates of earnings per share were collected from the FirstCall Earnings Database and dates were collected from the I/B/E/S databases. Accounting and financial data was collected from the Compustat Industrial Quarterly file. Data related to returns was collected from the Center for Research on Security Prices (CRSP) daily stock file. 16 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The “selected sample” includes firms that are members of one or more of the following market indices: the Nasdaq 100, the Philadelphia Stock Exchange Technology Index (PSETI) or the ISDEX Internet Index. The name “whisper companies” refers to companies for which a whisper number was found for the 1997-2000 period. “Rest of the companies” refers to the rest of the companies on the FirstCall Database and the Compustat Industrial Quarterly Database. Table 1 lists descriptive characteristics of each of these groups. IV. Research Design and Methods The non-parametric test (Kolmogorov-Smimov two-sample test) is used to test whether or not two samples may reasonably be assumed to come from the same population. The procedure estimates a difference (D): D = max \F(x)-G(x)\ for all x values. The null hypothesis that the two distributions are identical is rejected at the p level of significance if the computed value of D exceeds a certain amount. We compare the ability of the forecast error and the whisper error to explain abnormal returns around earnings announcements. We calculate market adjusted Cumulative Abnormal Returns (CAR) for the 3-day window centered on the date of the earnings announcement. CAR is defined as the difference between the firm’s security return minus the return on a value weighted portfolio. 17 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. For the earnings management hypothesis, we use quarterly data to estimate discretionary accruals. Quarterly data provides a sharper focus on the event when compared with yearly data, increasing the likelihood of detecting earnings management. I use the modified Jones Model to detect earnings management because Jeter and Shivakumar (1999) found the model to be well-specified for both annual and quarterly data. The times-series version of the model estimates firm-specific parameters using data from periods before the event period and the cross-sectional version uses parameters to estimate each period for each firm in the event sample using contemporaneous accounting data of firms in the same industry. Because whisper forecasts were available on a quarterly basis, the detection of earnings management is done with quarterly data. The cross-sectional model is selected, not only because of the evidence that it is wellspecified, but also because the requirement of time-series data would have substantially reduced our sample size. The cross-sectional version of the model was also selected because the abnormal accruals detected should be interpreted as industry-relative abnormal accruals. There is a caveat when dealing with the cross-sectional version of the model. Jeter and Shivakumar (1999) argue that if an industry enjoys favorable economic conditions and if firms enjoy smooth reported earnings, then the actual abnormal accruals for the firms in the industry will be negative (in fact, our results show this characteristic). The cross-sectional model is unlikely to capture all the negative abnormal accruals because earnings management is contemporaneously correlated across firms in the sample. Only those firms whose accruals are negative relative to the industry benchmark are identified as earnings managers. 18 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Each firm’s quarterly accruals relative to the same quarter for the previous year are estimated by Acct = ACA it - ACASH it - A CL it - DEP it Acct = Accruals during quarter t ACASH it = Change in cash over quarter t A CL u Change in current liabilities excluding the current portion of long-term debt over quarter t DEP u = Depreciation during quarter t All changes are calculated relative to the same quarter of the previous year to control for seasonality in accruals. Accruals are calculated for all firms in the sample and for all firms on the CompuStat Full Coverage file. The two-digit SIC code ensures a reasonable match on industry accruals (Jeter and Shivakumar (1999)). The sample contains firms from 18 SIC codes. The number of firms for each SIC group ranged from 30 to 1,028; the mean was 160 firms and the median was 97 for each quarter ranging from the fourth quarter of 1998 to the fourth quarter of 2000. Failure to meet data requirements to run the regression reduced the whisper/quarter observations from 544 to 504. The regression was estimated using the following specification: Accit / ait-i = Pi (1 / ait-i) + P2 (Arevit/ ait.i) + P3 (gppe it / ait-0+ sit Accit / ait-i = Accruals at time t scaled by the beginning of quarter total assets Arevjt / ait-i = Change in revenue with respect to the same quarter o f the previous year scaled by the beginning of quarter total assets 19 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. gppe it / ain = Gross property, plant and equipment at time t, scaled by beginning of quarter total assets a jt-i = Beginning o f quarter total assets The regression coefficients capture normal accruals for each specific industry. Any abnormal or discretionary accruals are estimated for each firm-specific quarterly observation based on the regression coefficient estimates for the industry. The results reported in the paper are the average of the results of the 18 different SIC-Codes that were represented in the selected sample. To improve the accuracy of the analysis, the sample is separated into nine groups. Table 2 illustrates how these groups are formed. Firms that are suspected of engaging in earnings management practices are those that have a forecast error (or whisper error) of 0 or 1 cent. We try to differentiate firms that manage earnings to meet the whisper forecast from firms that manage earnings to meet the analysts’ forecast. First, four preliminary categories were defined: a) Firms suspected of managing earnings to meet/beat the analysts’ earnings estimate (0 and 1 cent analysts’ forecast error, but not 0 or 1 cent whisper forecast error), b) firms suspected of managing earnings to meet/beat the whisper number (0 and 1 cent whisper error but not 0 or 1 cent analysts’ forecast error), c) firms for which it is not possible to isolate and test if they were managing earnings to beat either o f the thresholds (analysts’ forecast error and/or whisper forecast of 0 or 1 cent), and d) firms that were not suspected o f managing earnings to beat a threshold. Each category was further divided based on whether the whisper number was larger or smaller than the analysts’ forecast. Group 1 includes firms that just beat the analysts’ forecast by 0 or 1 cent and beat the whisper forecast by more than 1 cent. Group 2 20 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. includes firms that just beat the analysts’ forecast by 0 or 1 cent, but did not beat the whisper forecast. Group 3 includes firms that just beat the whisper forecast by 0 or 1 cent, but did not beat the analysts’ forecast. Group 4 includes firms that just beat the whisper forecast by 0 or 1 cent and beat the analysts’ forecast by more than 1 cent. Group 5 includes firms that just beat both thresholds and so their earnings management objective is ambiguous. Group 6 contains firms that miss both thresholds by more than one cent (Big Bath), while Group 7 contains firms that beat both thresholds by more than one cent (Big Day). Group 8 includes firms that beat the analysts’ forecast by more than 1 cent but missed the whisper forecast. Group 9 includes firms that beat the whisper forecast by more than 1 cent but missed the analysts’ forecast. The above separation is used to test both the abnormal returns hypotheses and the earnings management hypotheses. V. RESULTS V.l Analysts’ Forecasts. Timing-Adjusted Analysts’ Forecasts, and Whisper Forecasts Figure 1 shows the difference between the analysts’ forecasts, the adjusted analysts’ forecasts (analysts’ forecast2), and the whisper number. The whisper number was a more optimistic estimate of earnings than the analysts’ forecast. The mean whisper estimate is 0.1816 whereas the mean of the analysts’ forecasts is 0.1628. The adjusted analysts’ estimate did not represent a major improvement or change when compared with the analysts’ estimate. The difference between the whisper forecast and the analysts’ 21 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. forecast was consistent over sub periods within the sample. In addition, 87% of the whisper forecasts (477 out of 544) were positive compared with 79.5% of the analysts’ estimates. Reported earnings for the whisper sample were positive in 82.3% of the sample. V.2 Accuracy. Relevance, and Bias Panel A of Table 3 shows that before 1997, close to 64% of the firms in the sample met or beat analysts’ forecasts, and for the period after 1997, 76% did. After 1997, 62% of firms met or beat the whisper estimate. This suggests that the whisper estimate may be as relevant a threshold to managers as the analysts’ forecast. By the 1997-2000 period, 75% o f firms were meeting or beating analysts’ forecasts while the whisper estimate was more difficult to beat. Perhaps the whisper estimate arose to provide a more stringent threshold than the analysts’ forecast for growth firms. Hypothesis 1(a) Shift in bias in analysts’ forecasts and neutrality o f whisper forecasts As reported in Panel B of Table 3, the mean analysts’ forecast error for firms for which whisper forecasts were available increased from 0.00117 (t-value=0.245), or no bias for the period of 1990-1996 (Period I), to 0.010318 (t-value=6.138), a pessimistic bias for the 1997-2000 period (Period II). The difference between the means of the same companies for the two sub-periods was significant at the 1% level with a t-value of -4.33 (untabulated). For the subset of whisper firms, the mean analysts’ forecast error jumps to 22 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 0.01998 (reported in Table 6, Panel A, Column 3). These results suggest that the analysts’ forecast error was increasing for growth firms and increasing even more for the whisper sample. The change to a pessimistic bias was not a widespread phenomenon. For the rest of the companies, the optimistic bias of 1990-1996 persists into the 1997-2000 period. When we compare the analysts’ forecast error o f the growth sample (1990-1996) to the whisper error for the 1997-2000 period, we find no significant difference in the means. The distribution of the earnings forecast error has stimulated much research. DeGeorge, Patel and Zeckhauser (1999) and Burgsthaler and Dichev (1997) used a technique based on the empirical distribution of the forecast error. They showed that a smaller number of observations fall just left of the 0 forecast error. Their argument was that, given the shape of the distribution around 0, the discontinuity between -1 and 0 was consistent with the fact that managers whose firm’s earnings fell slightly below the estimate manage earnings to achieve the analysts’ target. That the mean forecast error is lower than the median forecast error is evidence that managers took an earnings “bath” to save for a better tomorrow (the left tail of the forecast error distribution was long). This condition is also observed in our control sample (rest of the observations). For the period 1990-1996, there is almost a cent difference between the mean (-.009) and median (0) forecast error, consistent with the results of DeGeorge et al. (1999). In contrast, for the whisper sample, the mean whisper error (.0014) is higher than the median (0). The shift and similarities between the means is evident from the graphs o f analysts’ forecast errors and whisper errors (Figure 2). The four quadrants of Figure 2 are labeled as top left - quadrant 1.1, top right - quadrant 1.2, bottom left - quadrant 2.1, bottom 23 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. right - quadrant 2.2. The analysts’ forecast error of the growth sample for the 1990-1996 period (Figure 2, quadrant 1.2) captures a distribution similar in form to DeGeorge et al. (1999). The shape of the distribution is preliminary evidence of earnings management as discussed by the same authors in their 1999 Journal o f Business paper. The graph for the same firms for the 1997 to 2000 period (Figure 2, quadrant 2.1) shows the shift in the bias. Although the 0 cent forecast error remains the mode, the 1 cent and 2 cent bin have an unusually high number of observations (consistent with a pessimistic bias). The distribution of the whisper error for the 1997-2000 period (Figure 2, quadrant 1.1) shows similar characteristics to the distribution of analysts’ forecast errors for the 1990-1996 period (Figure 2, quadrant 1.2). The Kolmogorov-Smimov test, reported in Table 4, does not reject the hypothesis that both distributions, analysts’ forecast errors for Period I and whisper forecast errors for Period II, are identical for the un-scaled results. When the error is scaled by the absolute value of reported earnings the Kolmogorov-Smimov test also cannot reject the null hypothesis of identical distributions (results not tabulated). The results are thus consistent with hypothesis Hl(c). The final graph in Figure 2, quadrant 2.2, shows the analysts’ forecast error for the whisper quarters. The graph shows a bimodal distribution of the analysts’ forecast error at 0 and 2 cents. This result is consistent with the hypothesis that a significant number of firms appear to use the whisper as their earnings threshold rather than the analysts’ earnings forecast for the 1997-2000 period. V.3 Bias and Accuracy Analysis 24 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Hypothesis H l(di) and Hl(dii) Tables 5 and 6 display the accuracy and bias of analysts’ forecasts and whisper estimates. The shift from a no bias to a pessimistic bias of analysts’ forecasts and the overall lack o f bias of whisper estimates are confirmed by the mean and median scaled forecast errors between the relevant periods. Panel A of Table 5 shows that both the mean and median analysts’ forecast error have increased for the growth sample, while Panel A o f Table 6 confirms the lack of bias of whisper estimates; the mean whisper error is 0.00143 (t-value 0.668). The mean percentage whisper error shows a small pessimistic bias (-0.0245), but the test of the mean does not reject forecast neutrality (t-value o f 1.457). Panel B of Table 6 provides additional evidence that whisper forecasts are more accurate predictors of earnings. The mean absolute percentage whisper error of 0.1646 (t= 11.961) is smaller even after controlling for their timing advantage over the analysts’ estimates, which had a mean absolute scaled error of 0.19047 (t= 10.580). The difference in accuracy between whisper estimates and analysts’ forecasts, when whispers are available is significant at the 10% level (t=l .486). Overall, both tables report results that are consistent with hypotheses Hl(di) and Hl(dii). We have presented evidence that suggests that whisper forecasts are more accurate than analysts’ forecasts. If investors found whisper forecasts to be relevant, it could be expected that there was a change in investor behavior when a whisper number was reported around earnings announcements. 25 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. V.4 Relevance o f Whisper forecasts to Market/Investors Hypothesis 2(a) Whisper Errors, Analysts’ Forecast Errors, and Abnormal Returns Table 7 reports results that are consistent with Hypothesis 2(a). The regression of Cumulative Abnormal Returns (CARs) on the percentage whisper error shows an R of .036. This result compares with a regression R of 0.011 for the analysts’ forecast error regression. When both the whisper and analysts’ forecast error are included in the regression, the coefficient of the whisper error is significant at the 1% level, while the analysts’ forecast error coefficient becomes insignificant. This result indicates that the market considers the whisper error to be more value relevant than the analysts’ forecast error. Table 8 reports the cumulative abnormal returns from day t-1 to t+1 centered on the earnings announcement date for the groups described in Table 2. Firms that were able to meet or just beat the whisper number, but not the analysts’ forecast, Groups 3 and 4, had positive abnormal returns. In contrast, groups 2 and 8, which reported earnings that beat analysts’ forecasts but missed the whisper number, had significant negative abnormal returns. This implies that investors may have formed their expectations based on the whisper forecast rather than the analysts’ forecast. Firms that met neither forecast (group 6 or big bath firms) had significant negative abnormal returns, while firms that exceeded both forecasts by more than 1 cent (group 7 or big day) had significant positive abnormal returns. Firms in group 5, that just met both forecasts and firms in group 1, 26 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. that just met analysts’ forecast but beat the whisper forecast by more than 1 cent also have positive CAR as expected. Managers’ reaction to whisper forecasts DeGeorge et al. (1999) state that “executives focus on thresholds, because the parties concerned with the firm do”. The previous results strongly suggest that a very important party -investors- care about the whisper number threshold. Therefore, it is possible that managers would alter discretionary accruals to meet/beat the whisper forecasts. Table 9 reports descriptive statistics from the abnormal accruals computations using the modified Jones model to measure abnormal accruals. Panel A reports the means and medians of coefficient estimates from estimating the prediction model by industry. Panel B reports descriptive statistics for estimated abnormal accruals. Overall, the mean and median abnormal accruals are negative for the sample firms. In Table 10, we focus on abnormal accruals for each of the nine groups described in Table 2. Both the sign and proportion of abnormal accruals are consistent with hypotheses H3(ai) and H3(aii). Groups 3 and 4 (0 and 1 cent whisper error groups, i.e. the groups most likely to manage earnings) have median positive abnormal accruals (scaled by total assets at the beginning of the quarter) of 2.9% and 3.2% respectively. The proportion o f positive to negative abnormal accruals is also higher for these groups (1.44 and 1.56) than for the rest of the groups. In contrast, Groups 2 and 8, which reported earnings that beat analysts’ forecasts but missed the whisper number, had negative abnormal accruals indicating less potential earnings management to increase earnings. 27 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. This is consistent with the ratio of positive to negative abnormal accruals of .67 and .53 for Groups 2 and 8 respectively. Group 5, for which it is unclear if a firm was following the whisper or the analysts’ estimate, abnormal accruals are not significantly different from 0. Group 6, the “bath” group, had negative and significant abnormal accruals, as expected. VI. Concluding Remarks The results of this study imply that during the late 1990s managers were managing earnings to meet or beat the whisper number. We find that whisper forecasts were value relevant even in the presence of analysts’ forecasts and that missing or beating a whisper number was associated with abnormal returns surrounding the earnings announcement. The paper also finds that estimated abnormal accruals are consistent with managers manipulating earnings to meet the whisper forecast. Generally speaking negative surprises are related with negative price revisions (Brown 1987, Skinner and Sloan). Those negative price revisions tend to be steeper for growth firms. Consequently, price and the characteristics of the firm seem to drive earnings management to avoid a whisper number earnings disappointment. Anecdotal and empirical evidence shows many technology firms had significant declines in prices and earnings after 2000. Class action lawsuits against company officials (especially those of technology companies) and also against analysts following the same firms, have increased during the last two years. It has been argued that the main driver for those investor lawsuits has been the selling of shares at “artificially pumped” 28 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. prices as a consequence of “artificially pumped” reported earnings (CNBC). Additionally, earnings restatements have reached record levels for the 2000 to 2002 period (Wu 2002). It is possible the reversal of accruals and the attempts of insiders to beat whisper estimates played a role. 29 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. References 1. Abarbanell, J. and Lehavy, R. 1999, Can Stock Recommendations Predict Earnings Management and Analysts’ Earnings Forecast Errors?, Working Paper, University o f North Carolina and University of California. 2. Ackert, Church and Shehata. 1997, An Experimental Examination of the Effects of Forecast Bias on Individuals’ use of Forecasted Information. Journal o f Accounting Research, Vol 35 No. 1 (Spring) pp 25-41 3. Brown, L. 1999, Managerial Behavior and the Bias in Analysts Earnings Forecasts, Working Paper, Georgia State University. 4. Bagnoli, M., Beneish, M. and Watts, S. 1999, Whisper Forecasts of Quarterly Earnings per Share, Journal o f Accounting and Economics, 28 pp. 27-50 5. Burgstahler D. and Dichev I. 1997, Earnings Management to Avoid Earnings Decreases and Losses, Journal o f Accounting and Economics 24 pp.99-127. 6. Burgstahler, D. and Eames. 2001, Management of Earnings and Analysts’ Forecasts. Working Paper, University of Washington, Seattle and Santa Clara University. 7. Collins, D.W. and Kothari, S.P. 1989, An Analysis of Intertemporal and Cross- sectional Determinants of Earnings Response Coefficients. Journal o f Accounting and Economics 11. pp 143-181. 8. Dechow, P., Sloan, R. and Sweeney, A. 1995, Detecting Earnings Management, The Accounting Review Vol 70. No. 2, pp. 193-225. 9. , Richarson, S. and Tuna, I. 2000, Are Benchmark Beaters Doing Anything Wrong, Working Paper. University of Michigan Business School. 30 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 10. DeGeorge, F, Patel J., Zeckhauser R. 1999, Earnings Management to Exceed Thresholds, The Journal o f Business, January, vol. 72, no. 1, pp. 1-33. 11. Francis, J. and Philbrick, D. 1993, Analysts’ Decisions as Products of Multi-Task Environments, Journal o f Accounting Research 31, 216-230. 12. First Call Earnings and Forecasts Database 19 90-2000. 13. Givoly, D., Bartov, E. and Hayn, C. 2001, The Reward o f Meeting or Beating Earnings Expectations. Working Paper. University of California and New York University. 14. Gaver, J., Gaver, K. and Austin J. 1995, Additional Evidence on Bonus Plans and Income Management, The Journal o f Accounting and Economics (February), pp. 3-28. 15. Healy, P. and Wahlen, J.M. 1999, A Review O f the Earnings Management Literature and Its Implications for Standard Setting, Accounting Horizons Vol 13 No. 4 365-383. 16. Jeter, D. and Shivakumar, L. 1999, Cross-sectional Estimation of Abnormal Accruals Using Quarterly and Annual Data: Effectiveness in Detecting Event-specific Earnings management, Accounting and Business Research, Vol. 29, No.4, pp. 299-319. 17. Kang Sok-Hyon. A Conceptual and Empirical Evaluation of Accrual Prediction Models, Working Paper, Yale School of Management, CT. 18. Kasznick, R. 1999, On the Association Between Voluntary Disclosure and Earnings Management. Journal o f Accounting Research (Spring) 37: 57-82. 19. Matsumoto, D. 2002, Management’s Incentives to Avoid Negative Earnings Surprises. The Accounting Review, Vol. 77 No.3, pp 483-514. 31 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 20. Myers, L. and Skinner D. 1999, Earnings Momentum and Earnings Management, Working Paper, University o f Michigan. 21. Sloan, R. and Skinner D. 2001, Earnings Surprises, Growth Expectations and Stock Returns or Don’t Let an Earnings Torpedo Sink Your Portfolio, Working Paper, University of Michigan. 22. Schiller, R. 2000, Irrational Exuberance, Princeton University Press, New Jersey. 23. Wu, M. 2002, A Review of Earnings Restatements, Working Paper, New York University. 32 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. R eproduced with perm ission Figure 1 Comparison of the whisper forecast, the analysts’ forecast and reported earnings for quarters when both analysts’ forecasts and whisper forecasts are available of the copyright ow ner. Whisper Error (0.0014) Further reproduction Forecast Error2 (0.02253) Forecast Error (0.01998) Errors prohibited without p e rm issio n . Forecasts Analysts’ Forecast (0.1628) Whisper Forecast (0.1816) A nalysts’ Forecast2 (0.169306) Note: Reported Earnings (0.1830) Data inside the parentheses is the mean o f the available observations. Whisper error is defined as (Reported Earnings - Whisper Forecast) Forecast error is defined as (Reported Earnings - Analysts’ Forecast) Forecast error2 is defined as (Reported Earnings - Analysts’ Forecast2) where Analysts’ Forecast2 is the mean o f forecasts issued over a period of 30 davs preceding the earnings announcement. 33 R eproduced with perm ission Figure 2 Distribution of Forecast Errors and Whisper Errors (Un-scaled) of the copyright ow ner. W h is p e r E rro r 97-00 Analysts' Forecast Error Sam ple 90-96 2 0 0 -| 600 150 - o ■O 50 - - _ _ _ _ .oD U flll 0 -2 5 .................................... w 400 1° 0 - -2 0 -15 -10 -5 . ° n UDDdDO - n .. . 5 10 15 0 _ _ . . . . 20 200 25 25 -20 -15 -10 -5 Further reproduction Cents prohibited without p e rm issio n . -5 5 10 15 20 25 Forecast E rror W hisper Qtrs 97-00 400 - -25 -20 -15 -10 0 Cents Analysts' Forecast Error 97-00 300 5 200 ° 100 -! Q o____ ____ ~ ~„___ □n□D[1 || nji 0 lllJJlQQDiiaAa-----------— n. 0 5 10 15 20 Cents 25 140 -I 120 100 5 80 O 60 40 20 0 _ .. . , n -25 -20 -15 -10 .nidi -5 —- 0 5 10 15 Cents Whisper Error is defined as Reported Earnings - Whisper Forecast Forecast Error Sample is defined as (Reported Earnings - Analysts Estimate) for the selected sample (Nasdaq 100, PSE and ISDEX) Forecast Error Whisper Quarters is defined as the Forecast Error ( Earnings-Analysts Estimate) when a whisper number was also available 34 — 20 n 25 Table 1. Comparative firm characteristics Years 1992,1996 and 2000 Selected sample are companies that are members o f one ofthefollowing market indices: Nasdaq 100, Philadelphia Stock Exchange Technology Index or the ISDEXInternet Index. Whisper Firms arefirms for which a whisper number was available at least one quarter of the sample period. Rest of thefirms are companies that are not part of theprevious groups and were included in the FirstCall Database for the 1990-2000period. Whisper Firms 1992 1996 Total Assets (MM$) Mean 3,366.7 3,839 762 Median 351 Sales (MM$) Mean Median 3,209 3,972 444 695 Rest of Firms Selected Sample 2000 1992 1996 2000 9,495 2,990 2,449 370 3,026 727 7,485 2,345 2,191 89 7,344 1,305 2,294 345 3,030 581 5,239 1,179 1,181 87 1,206 92 1,610 117 1992 1996 2000 2,734 4,209 212 140 Price Earnings Ratio Mean 33.2 Median 28.42 127 36.5 74 34.5 34 28.5 40 32 89 35.5 33.1 18 31.22 17 33.4 13.6 Market to Book Ratio Mean 4.57 Median 3.35 6.15 4.91 5.81 4.24 4.41 3.23 5.7 4.43 6.29 3.57 5.31 1.68 3.79 1.89 5.29 1.368 Market Capitalization ($MM) Mean 3,585 9,428 25,157 2,708 7,180 18,967 959 1,327 2,422 948 2,755 8,855 915 2,220 6,622 72 104 87 Year Sales Growth Percentage Mean 39 150 85 177 135 81 9.8 18.7 21 33 47 21 29 42 4 9 10 Median Median 20 35 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. R eproduced with perm ission of the copyright ow ner. Table 2 Separation of the Groups (Basis for Separation is amount of Whisper Error and Forecast Error) 79(14.5%) Further reproduction Oor 1 Forecast Error Group 5 118(90.1%) Group 3 13(9.9%) Group 4 131 (24.8%) Oorl Whisper Number 123(22.6%) prohibited without p e rm issio n . 98(80%) Group 2 25 (20%) Group 1 Forecast Error 48(229%) Group 6 210 (38.6%) Not Oor 1 97(46.2%) Group 7 61 (29%) Group 8 Median Percentage FE Variance N t-value -0.051 0 0.5156 4169 -4.5825 - 5 .0 0 6 5 2 * * 0.0001 -0.11 0 0.6202 2439 -6.908 T-test of Mean Differences(e) Median two-tailed p-value for normal approximation test (1990-96 and 1997-2000) ** Denotes significance at the 1% level (a) Growth firms are firms that are included in one of the following market indices: Nasdaq 100, PSE, ISDEX (b) Reported Earnings - Analysts' Forecast (c) (Reported Earnings - Analysts' Forecast) / Absolute Value of Reported Earnings (d) T-test of mean differences of the forecast error for the period 1990-1996 and 1997-2000 (e) Test of the difference between the mean % FE for the period 1990-1996 and 1997-2000 39 1997-2000 Growth Firms 0.010318 0.01 0.004887 1730 6.1384 0.032517 0.037037 0.3562 1730 2.2659 - 6 .6 4 3 5 9 * * 0.0001 R eproduced with perm ission Table 5 (cont.) Comparison of Analysts Forecast Errors (FE) for the Sample of Growth Firms During 1990-2000 Period of the copyright ow ner. Panel B. Analysis of Forecast Accuracy 1990-2000 Growth Firms (a) Further reproduction Mean Absolute % FE Median Absolute FE Variance N t-test 1990-1996 Growth Firms 0.29257 0.0909 0.4325 4169 28.72** 0.33731 0.1087 0.5186 2439 23.1321 T-Test Difference in Means Median two-tailed p-value for normal approximation test (1990-96 and 1997-2000) prohibited without p e rm issio n . Mean Square % FE Median Square FE Variance N t-test 0.518 0.00826 8.446 4169 11.51 0.2295 0.0769 0.30462 1730 17.295 5.47** (f) 0.0001 0.6322 0.01182 10.67 2439 9.5614 T-test Difference in Means Median two-tailed p-value for normal approximation test (1990-96 and 1997-2000) (f) T-test of mean differences of the absolute forecast error for the period 1990-1996 and 1997-2000 (g) Test of the difference between the MS% FE for the period 1990-1996 and 1997-2000 * Denotes significance at the 5% level ** Denotes significance at the 1% level 40 1997-2000 Growth Firms 0.357124 0.005917 5.2841 1730 6.47** 3.1914** (g) 0.0001 R eproduced with perm ission Table 6 Comparison of Analysts’ and Whisper Forecast Errors for the Sample of Firms for which Whisper Forecasts are Available of the copyright ow ner. Panel A. Analysis of Bias in Whisper Forecasts in Relation to Analysts’ Forecasts Further reproduction Mean Error (unsealed) Median Error (unsealed) Variance n t-value prohibited without p e rm issio n . Whisper Forecast Error 1997-2000 Analysts’ Forecast Error 1990-1996 Analysts’ Forecast Error Whisper Quarters(a) 1997-2000 Analysts’ Forecast Error Whisper Quarters Last 30 D a y s(a) 1997-2000 0.00143925 0 0.002483 544 0.6681 0.005 0 0.01197 1102 2.147* 0.01998 0.015 0.002014 544 10.3845 0.022531 0.015 0.0037544 424 7.5266 -3.92 * (c) -6.45 * (d) 0.0001 -0.71989(e) -0.04815 0 0.345 1102 -2.72 * 0.1411 0.0682 0.2161 544 7.0381 0.142607 0.084657 0.2407 424 5.95624 7.10 * [...]... and whisper earnings forecasts to investors? and 2) Are whisper forecasts relevant enough to investors that managers engage in earnings management to meet or beat this “new” threshold? 9 Reproduced with permission of the copyright owner Further reproduction prohibited without permission II.2 Accuracy, bias, and relevance of analysts’ forecasts of earnings when compared with whisper forecasts o f earnings. .. defines whisper forecasts and motivates the hypotheses Section III describes the whisper database used to conduct the tests Section IV discusses the research design Section V describes the results followed by a summary and conclusions in Section VI II Background and Hypotheses Development II 1 Whisper forecasts Analysts’ Behavior and Earnings Management Academics and financial press writers define whisper. .. 1999), and consequently, whisper forecasts started to gain credibility among the investment community (CNBC, Bloomberg, CNNFN) By the mid 1990’s, whisper forecasts were commonly used by investors As the Internet began its global development and expanded around 1997 it was common to observe financial reporters announce whisper forecasts in parallel with the firm’s earnings estimates Because whisper forecasts. .. previous research, we find that whisper forecasts are on average more accurate than analysts’ forecasts, as indicated by the mean absolute and squared forecast errors The improved accuracy of whisper forecasts relative to analysts’ forecasts holds even after controlling for the timing advantage of whispers4 Whisper forecasts are more accurate than the mean o f analysts’ forecasts released within a period... returns during earnings announcement periods than are unexpected earnings based on analysts’ forecasts The result is not sensitive to different specifications of unexpected earnings based on analysts’ forecasts and whisper forecasts Additionally, firms that were able to meet or beat the whisper number show higher positive abnormal returns than the control groups Consistent with whisper forecasts being... circumstances and opportunities for earnings management This paper is relevant to accounting literature because it addresses 1) earnings management behavior in the presence of more aggressive earnings targets, 2) the value relevance o f biased versus unbiased earnings forecasts, and 3) the emergence of new thresholds based on unofficial information, and their impact on the behavior of managers and investors... preceding the earnings announcement Thus, to test for a shift in bias and accuracy of analysts’ forecasts that occurred during the 1990’s, we hypothesize the following H I (di) The bias and inaccuracy o f analysts forecasts are greater in Period II than in Period I Hl(dii) The bias and inaccuracy o f whisper forecasts are smaller than that o f analysts’ forecasts II 3 Value Relevance of Whisper forecasts. .. returns and the earnings management hypotheses Based on empirical and anecdotal evidence, if whisper forecasts are relevant thresholds, we 13 Reproduced with permission of the copyright owner Further reproduction prohibited without permission should expect firms that meet or beat the whisper number to have positive abnormal returns A confounding factor is that whisper forecasts and analysts’ forecasts. .. f the sign o f whisper errors H2 (ci) Groups with positive unexpected earnings based on whisper forecasts have higher abnormal returns than groups with positive unexpected earnings based on analysts forecasts H2 (cii) Groups with negative unexpected earnings based on whisper forecasts have more negative abnormal returns than groups with negative unexpected earnings based on analysts forecasts II.4... missed the whisper forecast Group 9 includes firms that beat the whisper forecast by more than 1 cent but missed the analysts’ forecast The above separation is used to test both the abnormal returns hypotheses and the earnings management hypotheses V RESULTS V.l Analysts’ Forecasts Timing-Adjusted Analysts’ Forecasts, and Whisper Forecasts Figure 1 shows the difference between the analysts’ forecasts,