'h"; n‘ v» 4 ‘ weak A ~. ., ”.1 J: . «1' flux: $3314!- .3. ‘ . 25¢“; “ 5-.-. .. . ... v.“ - . . . a,\afi‘ 335.! ~ f In in ~ “‘3‘“: y . . Luv"? A H: 4-‘4114 3‘. > -.. A urn ‘. 3a. a' n.‘ ’ 1 13213; W " .13; “f“ . . 4' _; g-‘afi :~ ~. :5" . 4’51”»:‘1‘ ' W32“;- ' ""1549 .( -'-'» u, ‘ " i‘v -‘. . W x . l " ‘ "'- Q 0‘ H u‘ r r0 Jun-1‘ n. m am"? ‘3'}: -v } ’a .'L.‘.‘, ~ ’.VII., ‘ '1 u-« , ‘., w- 3‘?‘ ' {1‘- THESIS tlfliiflifliliiliifilfiflifliiiiil liifll’ifll’iifiil 3 1293 01037 9935 This is to certify that the dissertation entitled Motivations For and Reactions to Voluntary Management Disclosures presented by Susan Gill has been accepted towards fulfillment of the requirements for Doctor QLEhilosnnhylegree in W Qrug/fi/QH 0 a Major professor ( Date {II/(Q01 /(73 MSU Lt an Affirmative Action/Equal Opportunity Institution 0- 12771 LIBRARY Michigan State University PLACE ll RETURN BOX to remove this checkout from your record. TO AVOID FINES return on or botoro duo duo. DATE DUE DATE DUE DATE DUE Mm MSU loAn Atflrmotlvo ActioNEquol Opponunity inotltwon Wm: MOTIVATIONS FOR AND REACTIONS TO VOLUNTARY MANAGEMENT DISCLOSURES BY Susan Gill A DISSERTATION Submitted to Michigan State University in partial fulfillment of the requirements for the degree of DOCTOR OF PHILOSOPHY Department of Accounting 1994 ABSTRACT MOTIVATIONS FOR AND REACTIONS TO VOLUNTARY MANAGEMENT DISCLOSURES BY Susan Gill This study examines stock market and analyst reactions to a number of commonly issued voluntary management disclosures. Potential motivations for managers to issue the announcements are also tested. An understanding of management strategies in issuing announcements can assist policy makers in disclosure regulation, and investors in interpretation of the disclosures. An understanding of market and analyst reactions to management announcements can assist managers in making efficient decisions regarding optimal disclosure format. A number of frequently issued management announcements were identified, including qualitative earnings comments, planned cost cuts, planned changes in capital expenditures and product price changes. Market reactions to the announcements are tested by examining the variance of stock returns, and analyst reactions are tested by examining both the percentage of analysts revising their forecasts and the magnitude of forecast revisions. Potential motivations for the issuance of the management announcements are also investigated, by comparing both analyst forecast errors and the divergence of beliefs among analysts, between announcing firms and matched non-announcers. In addition, the propensity for firms to issue management announcements as preemptors of unexpectedly poor quarterly earnings is also examined. Results indicate that firms issue qualitative earnings comments at times when analyst forecast errors are relatively large, and both analysts and the market appear to react to the announcements. It appears that firms also issue cost cutting announcements to signal erroneous earnings expectations, however neither analysts nor the market react to the announcements. Results are inconsistent for both capital expenditures and price changes, which may, at least partially be due to data constraints. The results suggest that if firms issue various types announcements for the purpose of correcting erroneous earnings expectations, only announcements that specifically mention earnings are effective. ACKNOWLEDGEMENTS There are a number of people to whom I would like to express my thanks. First, I would like to thank my dissertation committee, Joe Anthony, Siva Nathan, Sue Haka, and Kathy Petroni. Not only were they easily accessible for purposes of guidance and assistance, but they made my dissertation a priority when time was of the essence. I would also like to acknowledge the contribution of I/B/E/S Inc. for providing the analyst forecast information, which was essential for this study. I would like to acknowledge the support of a number of my fellow doctoral students. My thanks to Robin Clement who was a major source of much needed computer assistance. My thanks also go to the other members of the 'Thursday Night Group' - Pat Essex, Julie David, Kim Galligan and Barb Esteves. They provided necessary comic relief, allowing me to retain my sanity and keep life in the doctoral program in perspective. Outside of the doctoral program, support came from my friend Ken Olson, who always assumed that I would succeed. Last, but definitely not least, my accomplishments were augmented by my very good friends Nancy Mullett and Lori Baukus. For many years we have shared hopes, dreams and laughter. During my doctoral studies they provided much needed emotional support, and will always have a special place in my life. TABLE OF CONTENTS LISTOFTABLES...’ ...... OOOOOOOOOOOOOOOOOOOOOO0.0.0.... ....... Vii LISTOFFIGURESOO0..OOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOO ix CHAPTERI INTRODUflION ...... OOOOOOOOOOOOOOO0.0.0.000...COO... 1 CHAPTER II LITERATURE REVIEW AND HYPOTHESIS DEVELOPMENT. . . . . . 7 EMPIRICAL TESTS OF EARNINGS FORECASTS.................... 7 THEORETICAL EXPLANATIONS OF VOLUNTARY DISCLOSURE. . . . . . . . . 12 HYPOTHESIS DEVELOPMENT................................... 15 Information Content of Management Announcements. . . . . 18 Motivations for Management Announcement Issuance. . . . 19 To Adjust Existing Earnings Expectations. . . . . . . 19 To Avoid Legal Liability....................... 23 ANALYST REACTION TO MANAGEMENT ANNOUNCEMENTS. . . . . . . . . . . . . 24 CWTERIII RESMCHDESIGNOOOOOOOOOOOOOOOOOOOOOOOOOOOOO...O. 27 SAMPLESELECTION......................................... 27 Qualitative Officer Comments........................ 28 Officer Earnings Comments (Pre-period end). . . . . 29 Officer Earnings Comments (Post-period end). . . . 29 General Company Outlook........................ 29 Cost CuttingMeasures............................... 29 PlantClosings................................. 29 Production..................................... 30 Work Force Changes............................. 30 PriceChanges....................................... 30 Capital Expenditures................................ 31 BudgetChanges................................. 31 Capital Projects............................... 32 Miscellaneous....................................... 32 THESAMPLE000000000 ....... 0.0.0.0000... 00000000 0 00000 0000 34 TESTSOFHYPOTHESE80.000.000.000.000.0.0..00.00.00...0.0. 42 Information Content of Management Announcements. . . . . 42 Analyst Forecast Errors as Announcement Motivators. . 48 Divergence of Beliefs as Announcement Motivators. . . . 54 Preemption of Bad News as Announcement Motivators. . . 56 Increased Percentage of Analysts Revising Forecasts. 58 Magnitude of Analyst Forecast Revisions. . . . . . . . . . . . . 61 CHAPTERIV RESULTS00000.000000.000.000.0000000000000000...... 64 MARKETmmION0000000000000.0000000000000000.00000 000000 64 POTENTIAL MOTIVATIONS FOR ISSUING ANNOUNCEMENTS. . . . . . . . . . 75 Analyst Error and Uncertainty as Motivators. . . . . . . . . 75 Bad News Quarterly Earnings as Motivators. . . . . . . . . . . 83 ANALYSTRECTION800000.0000000.000000.0.000..0.00000...0. 96 Number of Analysts Revising Forecasts. . . . . . . . . . . . . . . 104 Magnitude of Analyst Forecast Revisions. . . . . . . . . . . . . 111 CHAPTERV CONCLUSIONS...00.000000000000000...000000000000000.114 GENERALCONCLUSIONS.00000000000000.000000000.00.00.00.00. 114 IMPLICATIONS AND CONTRIBUTIONS........................... 117 LIMITATION80000000.0.00.00.00.00000000000000.000.000.0000119 WREsmCHOO00.00000......000000000000.0.000.000000. 121 APPENDIX000000000000.000000.00000000000000000000 0000000 0000000123 LISTOFREFERENCESOOOOOOO0.000.000.00000000000.000000000000000 129 vi LIST OF TABLES 1321: IILLE. EASE 1 Sample Selection...................... ...... 35 2 Number of Announcements Issued by Individual Firms (Full Sample)......... 37 3 Timing of Announcements (Full Sample)....... 38 4 Announcements by Industry (Full Sample)..... 40 5 Classification of Announcements............. 41 6 Classification of Announcements by Issuer... 43 7 Final Sample with Portfolio Matches......... 51 8A Market Reactions-Main Classifications....... 66 BB Market Reactions-Officer Earnings Comments.. 68 8C Market Reactions-Cost Cutting Announcements. 69 80 Market Reactions-Capital Expenditures....... 70 9A Analyst Forecast Errors-Forecast............ 77 98 Analyst Forecast Errors-Actual.............. 78 10 Analyst Divergence of Beliefs............... 81 11A Management Announcements as Preemptors of Bad News Earnings-All Announcements......... 85 118 Management Announcements as Preemptors of Bad News Earnings-Officer Earnings Comments. 86 11C Management Announcements as Preemptors of Bad News Earnings-Non-Earnings Comments..... 87 110 Management Announcements as Preemptors of Bad News Earnings-Cost Cutting Comments..... 88 vii 11E 11F 11G 12 13A 13B 13C 13D 13E 13F 13G 14 15 16 Management Announcements as Preemptors of Bad News Earnings-Capital Expenditures...... 89 Management Announcements as Preemptors of Bad News Earnings-Price Changes............. 90 Management Announcements as Preemptors of Bad News Earnings-Miscellaneous Comments.... 91 Actual and Implicit Fourth Quarter Forecasts 95 Management Announcements as Preemptors of Bad News Earnings-A11 Announcements......... 97 Management Announcements as Preemptors of Bad News Earnings-Officer Earnings Comments. 98 Management Announcements as Preemptors of Bad News Earnings-Non-Earnings Comments..... 99 Management Announcements as Preemptors of Bad News Earnings-Cost Cutting Comments.....loo Management Announcements as Preemptors of Bad News Earnings-Capital Expenditures......101 Management Announcements as Preemptors of Bad News Earnings-Price Changes.............102 Management Announcements as Preemptors of Bad News Earnings-Miscellaneous Comments....103 Analysts Issuing Revisions..................105 Analysts Issuing Year Ahead Revisions.......108 Magnitude of Analyst Forecast Revisions.....113 LI ST OF FIGURES Mantle Rage 1 Event Period for Market Reaction Tests........... 45 2 Timing of Forecast Error Computation ............. 49 ix CHAPTER I INTRODUCTION Since the seminal Ball and Brown [1968] study that used abnormal stock price reactions as evidence that accounting earnings convey information to the market, considerable accounting research has focused on the information content of earnings related disclosures such as earnings components [Lipe, 1986], earnings reporting timeliness [Chambers and Penman, 1984], and management forecasts of earnings [Patell, 1976; Penman, 1980; Waymire, 1984]. The most common empirical studies of voluntary disclosure have been in the area of management earnings forecasts, with the primary focus being the information content of the various disclosures. Although there is general agreement that management earnings forecasts are informative [Pate11, 1976; Ajinkya and Gift, 1984; Waymire, 1984], little is known regarding the motivations behind their disclosure. What is known, is that earnings forecasts are released infrequently. Lees [1981] documented that although most large companies routinely prepare earnings forecasts for internal use, only about 10% of the firms released forecasts to the public. Further, firms that 2 release earnings forecasts do not do so consistently. Ajinkya and Gift [1984] found that 143 of the 191 firms in their sample released only one earnings forecast during the eight year period under examination, and only three firms issued earnings forecasts in four or more of the years. The scarcity of publicly released earnings forecasts has been cited as a failure of the ability of voluntary market mechanisms to produce desirable quantities of public information about firms [Patell, 1976; Penman, 1980]. However, Dye [1986] demonstrates that when a manager's private information contains both proprietary and non- proprietary information, non-disclosure or partial disclosure of information may be optimal. This suggests that after considering costs, a rational manager may choose to disclose information other than an earnings forecast as a signal to the market regarding earnings expectations. The existence of accounting rules and regulations that do not allow all aspects of firm value to be effectively communicated through earnings provides further support for non-earnings information as an optimal form of disclosure [Healy and Palepu, 1993]. According to Healy and Palepu, a consequence of ineffective communication of managers' superior information can result in the misvaluation of firms by the market. As the undervaluation of firms can result in difficulties raising capital and increased expense in public capital markets, and overvaluation can result in legal 3 liabilities for failure to disclose relevant information, managers have incentives to correct erroneous market valuations at either end of the spectrum. One means of signalling firm misvaluation is through a change in financial policy, such as increasing or omitting dividends [Healy and Palepu, 1988]. Another, perhaps less costly signal, may be through additional voluntary disclosure. Gibbins, Richardson, and Waterhouse [1991] discuss the importance of understanding underlying motivations behind voluntary disclosure to improve both the quality of disclosure and the quality and functioning of disclosure systems. The latter is of concern to both standard setters and practitioners, as an understanding of the management of financial disclosure can assist in the analysis of potential consequences of disclosure regulation and deregulation. It can also serve to foster an understanding of the relation of accounting information to the total economic system of which it is a part. In a prior study of voluntary financial disclosure, Gibbins, Richardson, and Waterhouse [1990] proposed that some firms develop an opportunistic preference for disclosure management. They posit that there exists a spectrum of voluntary disclosure policies followed by firms, ranging from routine disclosures of mandated information to the use of financial disclosure as a strategic firm advantage.‘ At this time, little is known regarding the specifics of voluntary firm disclosure, such as the type of information announced, the motivation behind the announcement, and the subsequent reaction to the announcement. The purpose of this study is to address these issues. The main contribution of the study is an increased understanding of voluntary management disclosures, which is expected to benefit both the issuers and the potential users of the information. A better understanding of the motivations behind management disclosures can allow the market to better interpret the disclosure, and thus make more efficient investment decisions. An understanding of market reactions to management announcements can allow managers to make more efficient decisions as to optimal disclosure formats. Finally, information regarding market and analyst reactions to voluntary disclosure can assist policy makers in disclosure regulation. This study identifies a number of frequently issued management announcements, including qualitative comments on earnings, planned cost cuts, changes in planned capital expenditures, and product price changes. Market reactions to the announcements are tested by examining the variance of 1 Gibbins et. al. considered all financial disclosure to be voluntary, as even mandated disclosures have voluntary components such as timing, presentation and interpretation. 5 stock returns, and analyst reactions are tested by examining both the percentage of analysts revising their forecasts and the magnitude of forecast revisions. Potential motivations for the issuance of management announcements are also investigated, by comparing both analyst forecast errors and the divergence of beliefs among analysts, between announcing firms and matched non-announcers. In addition, the propensity for firms to issue management announcements as preemptors of unexpectedly poor quarterly earnings is also examined. In general, qualitative comments on earnings are found to be very significant, in both motivation and reaction tests. Cost cutting announcements appear to be issued by firms to signal analyst forecast errors and poor quarterly earnings, however neither analysts nor the market demonstrate significant reactions to the announcements. Neither announcements of planned capital expenditures nor announcements of price changes attain significance in either motivation or reaction tests, with the exception of a significant increase in analyst year-ahead forecast revisions following the capital expenditure announcements. The remainder of this paper is arranged as follows: Chapter 2 reviews existing literature relevant to this study and develops the hypotheses to be tested. Chapter 3 presents the research design and sample description, and Chapter 4 discusses test results. Conclusions, limitations 6 and suggested directions for future research are discussed in Chapter 5. Finally, the appendix provides examples of the various types of announcements included in the sample. CHAPTER II LITERATURE REVIEW AND HYPOTHESIS DEVELOPMENT This section begins with a review of the existing empirical studies of the accuracy and information content of management earnings forecasts, discusses the frequency of publicly released earnings forecasts, and concludes with a review of the theoretical literature regarding optimal voluntary disclosure. EMPIRICAL TESTS OF EARNINGS FORECASTS Voluntary management disclosures, in the context of management earnings forecasts, have been extensively studied by accounting researchers. Earnings forecasts were a topic of intense debate in the early 19708 when the Securities and Exchange Commission expressed an interest in developing a framework for the orderly disclosure of earnings forecasts [Patell, 1976]. The SEC believed that projections of earnings could be informative to the market, and thus suggested that this information should be made available to all investors. Therefore, motivated in part by a desire to empirically address some of the SEC's concerns, accounting researchers concentrated their research efforts on the 8 accuracy of management earnings forecasts. The inherent truthfulness of the forecasts was not a major concern of the researchers, as the market's ability to compare the earnings forecasts with reported earnings provides an incentive for managers to disclose earnings forecasts honestly. This incentive for honest disclosure is expected to hold regardless of whether the managers' compensation is directly tied to ending firm value, as long as managers assume that past performance will somehow influence future wages [Fama, 1980]. An incentive for honesty does not guarantee forecast accuracy, and since accuracy was assumed necessary in order for earnings forecasts to be informative, tests of forecast accuracy appeared to be warranted. In addition, researchers assumed that managers who voluntarily chose to forecast earnings were more confident in their forecasting ability than those who did not. Therefore, if management earnings forecasts were shown to be inaccurate, it would suggest that mandated earnings forecasts could be an inefficient method of supplying information to the market. The results of the early studies were largely inconclusive, primarily because of the lack of an accepted standard with which to compare relative forecast accuracy. Both McDonald [1973] and Basi, Carey and Twark [1976] found that approximately 70% of management earnings forecasts were upwardly biased (within approximately 10% of the actual 9 reported earnings). The remaining firms varied widely (with 5% of the Basi et. al. firms having forecast errors greater than 50%). Similar inaccuracies were found by Ruland [1978] in his comparison of management and analyst forecasts, although both were more accurate than predictions made using a naive model of past earnings. After considering the inaccuracy of management forecasts, and the likelihood that managers issuing voluntary forecasts have better forecasting abilities than those who did not, the case for mandating earnings forecasts was not supported. By 1976, the SEC had withdrawn their original proposal requiring mandatory disclosure of earnings forecasts, and instead suggested guidelines for companies wishing to release forecasts voluntarily. Accounting researchers remained interested in the subject of management earnings forecasts, although the focus of the research shifted from tests of accuracy to tests of the information content of management earnings forecasts. As stated by Penman [1980]: ”....while forecast 'accuracy' may be necessary for information content, the issue is whether investors can use forecasts to obtain predictions of factors relevant to firms' values which could not be made in the absence of forecasts.” Patell [1976] examined the relation between abnormal stock returns during a 17 week earnings forecast period as compared to that of the surrounding four year non-forecast 10 period. He found a significant and generally positive reaction to the forecasts regardless of the direction of the information 'surprise' and concluded that the act of voluntary financial disclosure rather than its content drove the market reaction. Conversely, Waymire [1984] found that the market reaction to earnings forecasts was positively associated with both the direction and magnitude of the information surprise. He explained the difference between his results and prior research as being due to improved methodology and the use of analyst forecasts as proxies for earnings expectations. Waymire concluded that his results supported the content rather than the act of disclosure as the force driving the market reaction. He did not however discount the importance of the act of disclosure, as he suggested that the act is possibly related to the magnitude of the surprise. Penman [1980] also found a significant and generally positive stock price reaction around the time of management earnings forecasts, but suggested that the driving force was the general tendency of firms to disclose good rather than bad news, as the majority of the forecasters in his sample were firms reporting greater than expected earnings. This finding was later confirmed by Lev and Penman [1990], although both studies found that at least a few firms released forecasts of bad news. Other studies [Ajinkya and Gift, 1984; Ruland, et. al., 1990] confirmed the positive ll associations between market reaction and earnings surprise, but were unable to document the good news bias. Study of the good news bias was extended through tests of industry information transfer. Penman [1980] posited that if firms signal good news through the release of earnings forecasts, then non-forecasting firms indirectly signal bad news. This suggests a negative market reaction to non-forecasting firms at times when earnings forecasts are disclosed by other firms in the industry. Lev and Penman [1990] tested this screening hypothesis and found that in the year of the earnings forecast, forecasting firms generally had higher earnings per share increases than either the market in general or matched control groups of non-forecasters. As expected, the forecasting firms' stock returns were significantly positive, however the non- forecasters did not have negative reactions, but instead had positive (but insignificant) residuals. One explanation for these results is that the forecasting firms were signalling above average industry earnings, which resulted in a positive information transfer to the non-forecasting firms. Early empirical studies of management earnings forecasts contained samples confined to specific point or range forecasts of earnings. This allowed an easy comparison of the earnings forecasts with existing market expectations of earnings and/or actual reported earnings. Recently, however, studies by Ruland et. al. [1990] and 12 Skinner [1992] used an expanded definition of earnings forecasts, specifically including qualitative earnings comments. One interesting finding by Skinner was evidence that suggested managers are more likely to use qualitative rather than quantitative earnings forecasts to signal bad news. In summary, prior research has consistently demonstrated that management earnings forecasts convey information to the market. However, disagreement exists as to whether the act of disclosure or the content of disclosure drives the market reaction and whether managers tend to release primarily good news forecasts and thus non- disclosers indirectly signal bad news. In addition, recent studies indicate that hard earnings forecasts are not the only vehicle used by management to signal changes in earnings expectations. THEORETICAL EXPLANATIONS OF VOLUNTARY DISCLOSURE Observed discrepancies in the empirical earnings forecast literature can possibly be attributed to the scarcity of publicly released management earnings forecasts. The principal reasons cited by managers for not releasing earnings forecasts are lack of confidence in their abilities to predict earnings [Lees, 1981; Ajinkya and Gift, 1984], potential liability for inaccurate earnings forecasts [Trueman, 1986; Lees, 1981; Ajinkya and Gift, 1984], and the 13 potential for competitors and others to adversely use the proprietary earnings forecast information [Lees, 1981; Gray, Radebaugh, and Roberts, 1990]. The various direct and indirect costs of financial disclosure have been the basis for theoretical analyses of optimal financial disclosure. Verrecchia [1983] posits that management will have an incentive to disclose earnings forecasts only after some 'threshold' level of benefit has been reached (i.e. when the expected market reaction to the disclosure is at least as great as the cost of disclosure). Verrecchia proposed that managers may either disclose or withhold a signal (earnings forecast). However, choosing to disclose will decrease the value of the firm by the cost of the disclosure. Therefore once cost is considered, the market cannot be sure that a lack of disclosure is indicative of bad news, or of good news with high disclosure costs. Thus, the market is expected to react less negatively to the withholding of information if the cost of disclosure is perceived to be high. Trueman [1986], proposes that the act of disclosure rather than the content of the disclosure conveys the information to the market. Trueman's hypothesis is based on both signalling and principal-agent theories. He posits that investors, being unable to directly observe a manager's ability, will revise their assessment of his/her ability based on the speed with which the manager recognizes and l4 adapts to changing environmental conditions. The ability to recognize and adapt to change is evaluated by the signals sent to the market in the form of voluntary disclosure, an example of which may be an earnings forecast. If there were no costs associated with the release of an earnings forecast the timing of the forecast would not affect the period's earnings, as optimal production decisions are independent of whether or not a forecast is released. Therefore, managers could increase investor's expectations of future earnings by issuing a revised earnings forecast as soon as new information regarding expected earnings is known. This theory can be used to explain the existence of bad news earnings forecasts; as earnings will eventually be disclosed to the market, the signalling properties of an early forecast release may partially offset the negative effects of the news itself. Both Verrecchia's and Trueman's theories indicate that costs play a major role in management's willingness to disclose financial information. It has been shown that managers' perceptions of the costs of disclosing earnings forecasts are significant. Gray et. al. [1990] surveyed Chief Financial Officers of United States and United Kingdom firms as to the perceived costs of various financial disclosures and found that the most costly disclosures were quantitative forecasts of sales and profits (total, line-of- business, geographical). The indirect costs of disclosure 15 (competitive disadvantage of releasing proprietary information, potential lawsuits for incorrect forecasts, trade union or employee claims) were considered much more significant than the direct costs (collection, preparation and dissemination). Managers may still achieve the positive effects of earnings forecasts and avoid some of the costs, if the disclosure of other, less costly information can be used to indirectly signal expectations of earnings. The possibility that non-forecasting firms indirectly signal earnings expectations through other forms of disclosure was suggested by Penman [1980] and Dye [1986]. Whether signalling or direct disclosure of information is optimal depends on both the efficiency of the signal in communicating the information and the cost of the information disclosure. This study investigates whether management announcements appear to be used as an indirect signal of management's earnings expectations. HYPOTHESIS DEVELOPMENT Hoskin, Hughes and Ricks [1986] found significant incremental information content in both prospective operating data and prospective qualitative comments made by managers when released concurrently with annual earnings. The implication that the market found these management announcements informative was considered surprising by the 16 authors, as the comments are neither audited, nor can they be directly compared to actual results. In addition, the 'expectations hypothesis' proposes that the market will discount less precise, qualitative types of earnings forecasts [King et. al., 1990]. However, after considering the Verrecchia theory of the cost/benefit considerations surrounding voluntary disclosure, and the Trueman hypothesis regarding the information content of the act of voluntary disclosure, I posit that these announcements are used by management and interpreted by the market as credible, alternate forms of the potentially more costly quantitative earnings forecasts. A study of Chief Financial Officer's perceived costs of various financial disclosures [Gray et. al., 1990] supports the hypothesis that firms view the release of officer comments and prospective operating data as less costly than the release of earnings forecasts. In their survey of 195 United Kingdom and 220 United States multinationals, Gray et. al. examined the perceived net costs and benefits of various types of disclosure, and their perceived importance and impact. In general, they found that most voluntary disclosure items are perceived to result in a net cost, with major costs assigned to quantified forecasts, inflation adjusted profits and narrowly defined segment information. Items considered fairly neutral included broadly defined segment information and non-quantified descriptions of 17 company activities. In a ranking of the 33 disclosure items tested (with number one being the most costly) quantified earnings and sales forecasts were ranked number 3 (7) by the U.K. (U.S.) CFO's. Disclosure of objectives, plans and policies was ranked 20 (22), qualitative company prospects were ranked 28 (25) and capital expenditure plans were ranked 19 (23).2 Gray et. al. provide evidence that qualitative announcements are perceived by managers to be less costly than hard earnings forecasts. As such, if managers are able to use these announcements to signal the market as to changing expectations, qualitative management announcements could be a strategic, cost effective means of disclosure. This study attempts to discover whether managers do appear to use these announcements as strategic signals, and whether analysts and the market appear to react to the announcements. The first step in this study is to test the information content of management announcements when issued in isolation. Although Hoskin et. al. [1986] demonstrated that 2 The two disclosures perceived most costly to U. K. firms were narrowly defined line-of-business earnings disclosures and descriptions of major legal proceedings. The disclosures perceived most costly to U. S. firms were the value added statement, various segment information (line-of-business, geographical), inflation adjustments and quarterly interim statements. 18 prospective operating data and officer comments contain significant incremental information when released concurrently with earnings, it has not been shown empirically that these announcements contain significant information when released alone. Since the existing research regarding voluntary disclosure has not conclusively identified the motivational factors affecting release, it is possible that management announcements released concurrently with annual earnings and those released at other times are motivated by different factors or interpreted by the market differently. Therefore the information content of the announcements warrants testing. If management announcements convey useful information to the market it is expected that stock returns will vary significantly around management announcement dates. Unlike earnings forecasts, which can be compared to an expectations model of earnings to determine the direction of the information 'surprise', qualitative comments cannot be as easily assessed. As attempts to classify the direction of the news are largely subjective (e.g., Does an announcement of a plant closing indicate new, negative information or information regarding a positive solution to a known problem?), testing the variance of the stock return is considered more appropriate than the use of a directional test [Patell, 1976]. This leads to Hypothesis 1 (in null form): 19 H10: There is no increase in the variance of firms' stock returns immediately following the release of management announcements by the firms. EWMW This section examines two non-mutually exclusive motivations for managers to release announcements; the desire to adjust prevailing market expectations of earnings (as measured by analyst forecasts), and the desire to avoid the legal liability associated with 'bad news' earnings announcements. ec According to Lees [1981], although only 10% of NYSE firms release direct earnings forecasts to the public, 65% informally assist analysts. This assistance consists of projections, statistics and background information on company operations. The use of this indirect channel of disclosure can be of mutual benefit to the firms and the analysts by reducing disclosure costs and improving analysts' forecasts and thus their professional reputations. A problem arises when analysts do not agree with management's estimations. Disagreement can occur when analysts do not agree with management's justifications for, or interpretations of, information provided, or when 20 analysts have knowledge of management's prior inaccuracies. Although management can have an advantage in interpreting firm specific data, analysts may be better able to interpret general economic and industry data [Jennings, 1987]. Management can make their disagreement with analysts known by publicly issuing their own earnings forecasts, but in doing so will incur proprietary costs. In addition to the proprietary costs already discussed, it has been suggested that there are analyst imposed costs to firms that issue earnings forecasts that directly contradict existing analyst forecasts [Ajinkya & Gift, 1984]. These costs include potential sanctions against the firms such as future unfavorable coverage by analysts, or negative advice to clients regarding the investment potential of the firms. Empirical evidence indicating that management earnings forecasts tend to confirm rather than contradict existing analyst forecasts [Ruland, et.al. 1990; Ajinkya and Gift, 1984] supports this conjecture. Although managers have an incentive to avoid the costs associated with contradicting analysts, managers also have an incentive to correct earnings expectations that are significantly inaccurate. Without providing an explanation as to why, Ajinkya & Gift [1984] hypothesized that large earnings surprises, resulting in large stock price fluctuations are costly to firms. King et. al. [1990] further develop this hypothesis, explaining that managers 21 have incentives to align investor expectations with their own, in order to reduce transaction costs in the market. Voluntary disclosure decreases the information asymmetry between managers and investors, which preempts the value of private information acquisition. This in turn decreases trading gains from privately held information, resulting in increased shareholder welfare. Trueman [1986] has also shown that the earlier a manager releases information demonstrating his/her knowledge of and ability to react to changing economic conditions, the earlier the market will anticipate the manager will do so in the future. Therefore, timely correction of inaccurate earnings expectations should result in an upward revision in market assessment of a manager's ability. Since the incentive to disclose information must be weighed against the potential costs of disclosure, it is expected that managers will wait to issue corrective signals until existing analyst earnings forecasts are largely different from managers' expectations. In addition, it is expected that managers will choose to send the signal in the least costly manner possible. The use of a management announcement as an indirect corrective signal provides information in a manner that allows analysts to revise their forecasts without incurring negative reputational factors, and thus could be perceived as the most efficient means of communicating management's expectations. Therefore, it is 22 expected that immediately prior to the issuance of a management announcement, analyst forecast errors for the firms issuing announcements are greater than analyst forecast errors for similar (industry-based) non-announcing firms at the same point in time. This leads to Hypothesis 2 (in null form): H20: Absolute analyst forecast errors prior to the issuance of manage- ment announcements are no greater for firms issuing the announcements than for similar non-announcers at the same point in time. At any point in time there exist numerous analyst earnings forecasts for a given firm. Due to various levels of analyst abilities, available information, and forecast age, some earnings forecasts are more accurate than others. In addition, firm specific characteristics such as diversification or earnings variability affect analysts' abilities to forecast earnings. This can result in a significant divergence of beliefs among analysts as to earnings expectations. To attain greater efficiency in the capital markets it is expected that managers have an incentive to provide additional information to reduce uncertainty when disagreement between analysts is large. Therefore, it is expected that the divergence of beliefs (as measured by the standard deviation of existing analyst forecasts) among analysts following firms issuing management 23 announcements, is greater than the divergence of beliefs among analysts following non-announcers. This leads to Hypothesis 3 (in null form): H30: The standard deviation of analyst forecasts immediately prior to the issuance of management announce- ments is no greater for firms issuing the announcements than for similar non-announcers at the same point in time. I E 'l I 1 1' l'Ji! Skinner [1992] examines the desire to avoid legal liability as a motivation for managers to issue earnings forecasts. The basic premise of his hypothesis is that the U. 8. security laws provide managers with incentives to disclose bad news prior to the issuance of mandated earnings announcements.3 Skinner looks at both quantitative (point and range estimates) and qualitative forecasts of earnings as preemptors of quarterly earnings announcements, and finds that large negative earnings surprises are preempted much more frequently than other earnings announcements. He also finds that 'bad news' preemptors are more likely to be disclosed in a qualitative, rather than quantitative manner. Although Skinner's hypotheses are supported, his 3 See Skinner [1992] for a discussion of evidence that earnings announcements followed by large stock price declines trigger class action lawsuits, regardless of merit. 24 results indicate that only 25% of 'very bad' earnings surprises are preempted. This suggests that if managers do have an incentive to preempt bad news, they may be using other, non-earnings announcements as preemptors also. My study replicates Skinner's hypothesis,and expands the types of disclosure to include non-earnings management announcements. This leads to Hypothesis 4 (in null form): H40: Firms are no more likely to issue management announcements in periods preceding disclosure of large negative quarterly earnings surprises, relative to other times of the year. ANALYST REACTION TO MANAGEMENT ANNOUNCEMENTS Studies have shown that analysts react to interim earnings announcements and to management earnings forecasts (and the ensuing market reactions) through the issuance of forecast revisions [Stickel, 1989; Baginski and Hassell, 1990]. If management announcements are released to signal management earnings expectations, it is expected that analysts will react to these announcements through forecast revisions also. Therefore, it is expected that the percentage of analysts revising their forecasts for a firm following the issuance of a management announcement will be greater than the average percentage of analysts revising their forecasts for the firm in non-announcement periods. 25 This leads to Hypothesis 5 (in null form): H50: There is no increase in the percentage of analysts revising their forecasts for a firm in the period immediately following the issuance of management announcements, relative to non- announcement periods. Little is known regarding the mechanics of forecasting by analysts. Analysts have many sources of information available, both economy wide and firm specific, which are continually being updated. Exactly what information is considered, how it is weighted, and the time table used to act on the information through forecast revision is not known. Therefore, it is not unlikely that routine forecast revisions occur at times other than immediately following firm disclosures. However, if management announcements are released to signal management's disagreement with existing earnings expectations, or management's belief that changing economic conditions will have pervasive effects on the firm, then it is expected that the average magnitude of analyst forecast revisions are greater for firms issuing the announcements than for similar non-announcing firms at the same point in time. This leads to Hypothesis 6 (in null form): H60: 26 The average magnitude of analyst forecast revisions immediately following the issuance of management announcements is no greater for firms issuing the announcements than for similar non-announcers at the same point in time. CHAPTER III RESEARCH DESIGN The hypotheses are empirically tested using the Wall Street Journal, and the CRSP, Compustat and IBES data tapes as data sources. Hypotheses 1, 4 and 5 use the sample firms as their own control, examining differences between time periods when management announcements are issued and time periods when they are not. Hypotheses 2, 3 and 6 use portfolios of firms matched as to industry, size and fiscal year end as controls. Tests of these hypotheses examine differences between sample firms and control portfolios at the same point in time. SAMPLE SELECTION The population from which the sample is chosen consists of the 1,413 firms identified as being included on CRSP, Compustat and IBES data tapes as of 1990. Potential management announcements are initially identified by examining the Wall Street Journal Index for the years 1986 - 1990.‘ These years are the most recent with complete data ‘ Although it has been shown that the Wall Street Journal does not contain an exhaustive collection of announcements released by firms [Wright and Groff, 1986], it 27 28 available for both test and estimation periods for all six hypotheses. The use of a five year period is to insure a sufficiently large number of observations for statistical validity. To be included in the sample an announcement must be specifically attributed to a firm, and appear to be both prospective in nature and voluntarily announced. The specific categories of announcements collected are described below (with examples of actual announcements provided in the Appendix). WW These announcements are included because of their similarity to earnings forecasts. Although early empirical studies of management earnings forecasts used only hard, point or range estimates, current studies have included these softer 'forecasts' [Ruland et. a1. 1990, Skinner 1992] with significant results. In addition, qualitative officer comments issued in conjunction with earnings were found to contain significant incremental information in Hoskin et. al. [1986]. This category is further subdivided into three types of announcements as discussed below. contains many announcements of the type that are used in this study [Thompson, Olsen and Dietrich, 1987]. As the content of the general announcements is expected to vary widely, a keyword search of the Broad Tape is not considered feasible. Therefore, the Wall Street Journal is selected as a cost effective single source of announcements. 29 O ' e ' s e - e ' nd These are comments concerning expectations of future earnings. They are similar to management earnings forecasts, except that they are qualitative rather than quantitative in nature. 0 t - e The content of these announcements is identical to that of the previous category, however the announcements are issued after the end of the fiscal period but prior to the announcement of actual earnings. W These refer to general company expectations, but do not specifically discuss earnings. They include comments on general growth prospects, expected changes in product demand, market share and competitiveness. 0 ° es These announcements are included due to their potential as signals of market demand or firm reaction to economic events. This category is also further subdivided into three types of announcements. ElanLQlesings Both temporary and permanent plant closings are included, however sales and spin-offs of entire business segments are not. The latter are expected to have pervasive economic effects of their own and therefore, the 30 reactions to these announcements are not expected to be motivated by the same factors as reactions to the types of disclosures included in this study. In addition, as sales and spin-offs involve the participation of one or more other parties, the timing of the announcements is not entirely under the control of the issuing firm, suggesting the announcements may not be purely voluntary. mun These are announcements of future changes in production due to expectations of increasing or decreasing demand. Both temporary and permanent production changes (except those directly associated with plant closings) are included. Announcements of production changes, when issued in conjunction with annual earnings announcements were found to contain significant incremental information in the Hoskin et. al. [1986] study. Wanna: This category includes announcements of future layoffs, hiring freezes, salary cuts and early retirements (again, not including those directly associated with plant closings or production changes). W In addition to the expected income effects of changing prices, announcements of product price changes can also signal market demand or market position. Shepard [1970] 31 states that initial announcements of price changes, if followed by corresponding changes from other firms in the industry are interpreted as signals of market dominance. Therefore it is reasonable to assume that firms announce price changes as signals of changing earnings expectations. W Announcements of planned changes in capital expenditures may also be used as signals of changing economic conditions or market position. Although tests of the incremental information content of capital expenditures did not reach conventional levels of significance in Hoskin et. al., these announcements may be more informative when released alone. It is also possible that capital expenditure announcements may signal changes in long range, rather than short range expectations. If so, these announcements may be found to affect long term analyst earnings forecasts, without having an immediate impact on current stock returns, due to uncertainty surrounding the interpretation of the information. This category is further sub-divided into two types. l W These are announcements regarding planned changes in general capital spending for the year, generally announced as a percentage or dollar change from the previous year. 32 W This category consists of announcements of dollar amounts to be spent on specific expansion or upgrading projects, undertaken independently of other business entities. We These announcements, specifically attributed to high ranking company officials, are comprehensive statements of company goals and expectations, often including two or more of the previously discussed categories. This study does not include announcements of retrospective events, as they are not expected to provide new information, and therefore are not necessarily indicative of changes in expectations. Announcements of contract awards and joint ventures are not included for reasons similar to those discussed regarding sales and spin- offs of business segments; although they may convey information to the market, the ability of the announcing firm to manage the information disclosure is reduced, and thus these announcements are not necessarily motivated by the same factors as the announcements under study. The announcements are also required to meet the following criteria: 1) availability of CRSP daily returns, 33 2) availability of Compustat total asset and quarterly earnings information, 3) availability of IBES analyst forecasts, 4) for tests of market reactions and potential motivations for announcement issuance the absence of other potentially confounding announcements within the two weeks surrounding the management announcement is required (one week on either side of the announcement), 5) for tests of analyst revisions the sample is further restricted to require the absence of other potentially confounding announcements within the six weeks surrounding the management announcement (three weeks on either side of the announcement). The absence of potentially confounding announcements is required to reduce the likelihood that observed market reactions and analyst revisions are driven by something other than the management announcements. The six week restriction for tests of analyst forecast revisions was chosen somewhat arbitrarily. There is little theoretical or empirical evidence regarding the speed with which analysts formally react to information through the release of a revised forecast. Jennings [1987] found most reaction to management earnings forecasts took place within the first week, with further revisions occurring in the following two weeks, and very little revision in the fourth or later weeks. Therefore, requiring three weeks on either side of the announcement is expected to include the majority of the revisions, while still controlling for confounding events. The CRSP data tapes provide stock returns used to compute the abnormal stock price reaction to the management 34 announcements (Hypothesis 1). The Compustat data provides quarterly earnings announcements for tests of Hypothesis 4, and data used to form control groups for tests of Hypotheses 2, 3, and 6. The IBES data provides analysts forecasts for tests of Hypotheses 2 through 5. THE SAMPLE An examination of the Wall Street Journal Index resulted in the identification of 1,882 management announcements from firms originally identified as meeting the data base requirements. Of these, 722 were eliminated due to their issuance in conjunction with or within one week of, other potentially confounding announcements (usually earnings announcements). The content of the remaining 1,160 announcements was verified by reading the Wall Street Journal articles cited. This resulted in a further deletion of 400 announcements, as shown in Table 1. The 133 announcements classified as inappropriate were found to have content incompatible with one of the categories previously defined. The most common examples of these were announcements originally assumed to be qualitative earnings comments that actually contained quantitative earnings forecasts. Another common example included announcements originally assumed to be capital expenditure announcements that were actually joint ventures with other companies. The 65 announcements classified as not specifically 35 Table 1 Sample Selection m 1 Initial Announcements 1,882 Issued in conjunction with potentially confounding announcements ( 314) Issued within one week of potentially confounding announcements ( 408) Remaining announcements with content verified in Wall Street Journal 1,160 Announcements Deleted Due to: Inappropriate Content ( 133) Not specifically announced by firm ( 65) Potentially confounded by stock splits or takeovers ( 142) Not prospective in nature ( 53) Articles not found in Wall Street Journal ( 7) Full Sample‘ 760 Issued within 1-3 weeks of potentially confounding announcements ( 319) Reduced Sampleb 441 a. The full sample is the base sample for tests of market reactions (Hypothesis 1) and tests of potential motivations for firms to issue management announcements (Hypotheses 2, 3 and 4). b. The reduced sample is the base sample used for tests of analyst reactions to management announcements (Hypotheses 5 and 6). ' 36 announced by a firm, most often consisted of comments taken from analyst or Wall Street Journal company profiles. These articles cited either the company's annual report or previously issued management announcements as their source. Although analyst interpretations of prior comments could logically be expected to have an impact on the market, they are not appropriate for purposes of this study. An additional 142 announcements were deleted due to the announcing firms involvement in stock splits or on-going takeover activity in proximity to the announcement release. The 53 announcements considered not prospective were released at the time of, or after, the announced event. This was often the case with announcements of both plant closings and price increases. Finally, seven announcements could not be found, apparently due to typographical errors in the Wall Street Journal Index. The base sample, used for tests of Hypotheses 1 through 4, consisted of 760 announcements, issued by 378 firms. For tests of Hypotheses 5 and 6, the sample was further reduced by 319 announcements issued within three weeks of other potentially confounding announcements. The announcements were issued by a number of firms in a variety of industries and occur fairly uniformly throughout the year, as shown in Tables 2, 3 and 4. Table 2 provides the breakdown of the number of announcements issued per firm. Table 3 shows the timing of the announcements, and 37 Table 4 provides the breakdown of firms by industry classification. As shown in Table 2, the majority of the firms in the sample issued three or fewer announcements that met the criteria for inclusion during the five year period. Some of the largest firms actually issued a greater number of announcements than indicated in the table. However these firms, with almost daily news coverage, also had a large number of announcements eliminated due to proximity to potential confounders. The deletion of many of the multiple announcements minimizes statistical problems due to lack of Table 2 Number of Announcements Issued by Individual Firms (Full Sample) iii _1 Number of Announcements Number of Total Number of . Issued“ Firmsb Announcementsc ; 1 235 235 i 2 55 110 f 3 42 126 4 13 52 5 12 60 > 5 21 177 Total 378 760 a. Refers to the number of announcements issued by individual firms during the five years studied. b. Refers to the total number of firms issuing the corresponding number of announcements during the five year period. c. Refers to the total number of announcements issued during the five year period. 38 independence, however the results of this study may not be generalizable to those very large firms. As shown in Table 3, the announcements were issued fairly uniformly throughout the year. The months with the fewest announcements can be explained by the fact that the majority of the sample firms have December fiscal year-ends. The subsequent clustering of quarterly earnings announcements results in a larger number of announcement deletions due to potential confounders in these months. Table 3 Timing of Announcements (Full Sample) Month 1986 1987 1988 1989 1990 Total January 14 15 15 13 14 71 February 17 9 13 8 6 53 March 15 8 19 16 21 79 April 16 6 8 4 10 44 May 24 15 15 7 10 71 June 19 18 19 13 16 85 July 9 8 4 7 10 38 August 10 3 13 8 13 47 September 14 19 18 10 20 81 October 13 7 11 14 12 57 ‘November 7 9 14 15 13 58 December 12 16 15 14 19 76 Total 170 133 164 129 164J 760 39 Table 4 shows a breakdown by industry, with similar two-digit industry classifications combined for convenience. The large number of industries represented indicates that the issuance of management announcements is not unique to any specific industry. Table 5 provides the number of announcements in each of the main classifications and sub-categories for both the full and reduced samples. The majority of announcements are classified as Officer Earnings Comments, making up 42.89% (44.22%) of the full (reduced) sample. The additional time restriction, imposed to eliminate potentially confounding announcements during periods of analyst forecast revisions, does not significantly change the distribution of the reduced sample as to announcement classification. Finally, the announcements were classified as to whether they were specifically attributed to a company official or spokesperson, or were a miscellaneous reference to the firm. This was done for two reasons; 1) under the hypothesis that announcements are issued to signal the market of changing economic conditions or earnings expectations, announcements specifically originating from company officials may have different signalling properties than those that do not, and 2) the origin of the announcements was not always obvious even after reading the referenced articles. The latter concern arose because of the number of capital expenditure announcements that cited a 40 Table 4 Announcements by Industry (Full Sample) .= Number of Firms Industry Cosmetics[Pharmaceuticals 34 Machinery 33 Household Appliances 31 Financial 28 Motor Vehicles/Aircraft/Ships 21 Natural Gas Transmission 21 paper 19 Petroleum Refining 18 Steel/Non-Ferrous Metals 18 Food/Tobacco 17 Wholesale 16 Retail Stores 14 Communications 13 Precision Machinery 13 Transportation 12 Crude Oil 9 Publishing/Printing 9 Computers 8 Construction 8 Plastics 8 Mining Metals 7 Textiles 5 Travel/Leisure 5 Metal Work 4 Lumber 3 Concrete 2 Professional Services 1 I Total 378 , 41 Table 5 Classification of Announcements Full Per- Reduced Per- Announcement Type Sample' cent? Samplec centb Officer Comments Pre-f iscal-year-end 2 27 157 J Post-f iscal-year-end 58 16 General Outlook 41 22 Total 326 42.89% 195 44.22% Cost Cuts Plant Closings 45 17 Layoffs 97 62 Production Changes 31 18 Total 173 22.76% 97 21.99% Capital Expenditures Budget Changes 53 31 Specific Projects 71 36 ‘ Total 124 16.32% 67 15.19% Price Changes 54 7.11% 32 7.26% Miscellaneous 83 10.92% 50 11.34% Total 760 441 The full sample refers to announcements that are free of potential confounders for one week on either side of their issuance date, and is the basis for tests of market reactions (Hypothesis 1) and potential motivations for firms to issue announcements (Hypotheses 2, 3 and 4). This column provides the percentage of the total sample represented by the corresponding main category of announcement. The reduced sample refers to announcements that are free of potential confounders for three weeks on either side of their issuance date, and is the basis for tests of analyst reactions (Hypotheses 5 and 6). 42 companys' annual report as the source of information (and were thus deleted from the sample). In the event that some of the unattributed announcements also originated from annual reports, it is expected that these announcements would not have the same impact as announcements currently issued by the firms. Table 6 provides the breakdown of announcements with respect to officer attribution. In summary, the announcements are made by a large number of companies, in a wide variety of industries, and are spread fairly evenly throughout the year. The sample is dominated by qualitative comments on earnings, but includes a sufficient number of announcements, at least in the main categories, for purposes of statistical testing. TESTS OF HYPOTHESES Six hypotheses are tested in this study. Hypothesis 1 tests the market reaction to the issuance of management announcements. Hypotheses 2, 3, and 4 examine potential motivations for firms to issue management announcements. Hypotheses 5 and 6 test analyst reactions to management announcements. Hypothesis 1 tests market reactions to management announcements by examining the variance of stock returns immediately following the issuance of management 43 Table 6 Classification of Announcements by Issuer a firm, but not necessarily to a named officer or spokesperson. Officer/ Spokes- Announcement All person Type Announcers' Announcedb Percentc Officer Comments Pre-fiscal-year-end 227 154 Post-fiscal-year-end 58 27 General Outlook 41 36 Total 326 217 (66.56%) Cost Cuts Plant Closings 45 14 Layoffs 97 62 Production Changes 31 15 Total 173 91 (52.60%) Capital Expenditures Budget Changes 53 19 Specific Projects 71 14 Total 124 33 (26.61%) Price Changes 54 19 (35.19%) Miscellaneous 83 58 (69.88%) Total _ 760 k 418 (55.00%) a. Number of management announcements directly attributed to Number of announcements directly attributed to a named officer or spokesperson of the firm. Percentage of each main category of announcement that is directly attributed to a named officer or spokesperson. 44 announcements. If the market reacts to the announcements it is expected that the variances of firms' stock returns are abnormally large immediately following the issuance of announcements. To test abnormal return reaction around the time of the management announcement, an estimate of the expected return is required. This is obtained using Ordinary Least Squares (OLS) regression on the following market model: Ric-“TBRntTeit (1) where Rit = the return for firm i at time t, obtained from the CRSP data tapes, IR“ = the value-weighted return on the market at time t, also obtained from the CRSP data tapes, eit a an independent error term, assumed normally distributed, *9 II the estimated intercept for firm i, B, = the estimated slope coefficient for firm i, H. II 1,...,N firm observations, t: = 1,...,T days in the non-forecast estimation period. The regression serves to remove the market-wide elements of price change from the individual firm return. The regression is run over a 200 day estimation period surrounding a 12 day event period. The estimations of the parameters a and B, a and b respectively, obtained during 45 the estimation period are used to compute the expected Rit during the event period. The event period consists of a two day event window (the day the information is reported in the Wall Street Journal and the previous day) surrounded by five days on either side as shown in Figure 1. estimation event estimation period peTiod period t ' ill! ' H ' i (~106, -7) [(-6,(-l, 0), +5)] (+6, +106) event window Figure 1 Event Period for Market Reaction Tests The two day event window controls for the possibility of a reporting delay by the Wall Street Journal due to the time of day that the announcement is released. The surrounding days enable an analysis of the stock returns during a period for which other potentially confounding events have been eliminated. Non-significant return variances surrounding the announcements provide additional support that significant deviations during the event window 46 are due to the release of the management announcements and not some other, unrelated economic event. Abnormal returns are computed individually for the cumulative two day event window, and for each of the surrounding ten days, as shown below: "it'Rit' (“u-+513“) (2) where t = -6,.....,+5 days Patell's U-statistic is used to test the significance of the abnormal return variance. The U-statistic is computed as: ”It (3 1) ”it ' 3 . 31 C1: where u"? = the squared abnormal return for firm i at time t, 83 = the estimated variance of the residuals during the estimation period; C = an adjustment to the variance reflecting the expected increase due to prediction outside the estimation period, computed as; it 1+_]_'.+ Lac-EL T 1' 23(anp-RD‘ t-I (3.2) 47 where T = the number of days in the estimation period, and E'Ji-iR-t (303) 3.1 The expected value of the Uit is greater than 1, however the Uit can be standardized to yield an expected value of 1 as follows; 2 ”1: T1 - 4) The above Uit has an expected value of 1 and a variance of 2(T,'- 3)/(Ti«- 6). Therefore the‘U" for the event period serves as a ratio of the information conveyed by the management announcement, to the average information conveyed over the estimation period. A Uit greater than 1 would imply a greater than average amount of information conveyed by the management announcement. The significance of the deviation of the Uit from 1 is tested using one-sided 2- statistics. 48 11:, - o ‘ e o - a- i!!°-l ‘u‘n ,. 'vat- Hypothesis 2 is a test of a potential motivation for firms to issue management announcements. The hypothesis tests whether firms issuing management announcements are different in some way from firms that do not. The specific difference analyzed is the magnitude of mean analyst forecast errors immediately prior to the sample firms release of management announcements. The mean analyst forecast errors of the sample firms are compared to an average of the mean forecast errors of portfolios of similar, non-announcing firms, at the same point in calendar time, as shown in Figure 2. Each of the firms in a portfolio of non-announcers must meet the following criteria: 1) belong to the same SIC industry classification as the corresponding sample firm, 2) be of similar size (as measured by total assets) as the corresponding sample firm. 3) have the same fiscal year end as the corresponding sample firm. Initially, firms are matched by industry using 4-digit SIC codes. For any firm with less than three control matches at the 4-digit level, the criterion is expanded to include 3-digit, and if necessary, 2-digit codes. The purpose of the industry match is to control for the effects of general economic conditions on analyst forecast errors. The use of portfolios is to reduce the problems associated 49 firmi I l l -1 O portfolio,| ! -1 where firmi = the firm issuing the management announcement. portfolioi = a group of non-announcing firms, similar in nature to firm i, day 0 = the date the management announcement is issued by firm i, day -1 - the date the mean analyst forecast error for firm i is compared to the average consensus analyst forecast error for portfolio i. Figure 2 Timing of Forecast Error Computation with the dissimilarities among firms within industry classifications, such as varying levels of diversification and business risk [Baginski, 1987]. Firms are considered a size match if total assets are within 15% of the total assets of the corresponding sample firm.5 The size match is required as previous studies [Cox, 1985; Thompson, Olsen and Dietrich, 1987] have found 5 A 15% match was selected after attempting matches at 10%, 15% and 20% of the sample firms' assets. The 10% match resulted in a considerably smaller final sample than the 15% match. The 20% match increased the size of some of the portfolios, but did not significantly increase total sample 3 ze. 50 that larger firms receive greater news coverage than smaller firms, which is expected to impact analyst forecast accuracy. Fiscal year end is considered a match if the control firms' fiscal years end in the same month as the corresponding sample firms. The fiscal year end requirement is included as both analyst forecast accuracy [Elton, Gruber and Gultekin, 1984] and analyst propensity to revise forecasts [Baginski and Hassell 1990; Stickel, 1989] have been shown to increase with the time of fiscal year. The control firms are also required to meet the same data availability and absence of confounding events as the original sample. After potential control firms are identified, the Wall Street Journal Index is read to verify that the selected firms have not issued management announcements, earnings forecasts or announcements of actual earnings within one month prior to the sample firms issuance of a management announcement. Firms that have issued any of these types of disclosures are deleted from the control portfolios. Analyst forecasts made within one month prior to the sample firms' management announcement date are obtained from the IBES data tapes for all remaining firms. The one month period is chosen to obtain the most recent estimates (assumed to most closely approximate current market expectations) while still controlling for confounding 51 events. Firms with no IBES information, or with less than two forecast revisions during the six week period are deleted. The final sample consisted of 211 firm announcements, as shown in Table 7. As the magnitude, rather than the direction of the forecast error is of concern, a mathematical average of absolute forecast errors is used. The test statistic, similar to that in Hassell and Jennings [1986], is computed Table 7 Sample Selection Announcements Matched with Control Portfolios Initial number of announcements with 392 available control firm matches Announcements deleted due to: Control firms having management announcements issued within one month prior to sample firms issuance of a management announcement (41) Sample firms with less than 2 analyst forecasts issued within one month prior to date of management announcement (42) Control firms missing IBES data, or with less than 2 analyst forecasts within one month of corresponding sample firms' issuance of a management announcement (98) Final Sample of Announcements 211 52 as follows for the sample firms: g! (5.1) where Ei== actual reported earnings per share for sample firm i, as of the end of the forecast year, AI} = the average of all analyst forecasts for sample firm i, that were released within one month prior to the issuance of a management announcement. Absolute analyst forecast errors for the control portfolios are computed as; (5.2) where Epi = actual reported earnings per share for control firm p, as of the end of the forecast year, AFM - the average of all analyst forecasts for control firm p, released within one month prior to firm i's issuance of a management announcement, i = the sample firm for which firm p is a control, p = 1,...,N firms in the control portfolio. The use of actual earnings to standardize the forecast 53 errors controls for the variability in the size of earnings between firms. The statistic is also tested using forecasted earnings to standardize. Differences between the two groups are computed as: In“: - ‘mumlTlt (503) The absolute values of the differences are ranked according to size, with the sums of both the ranks of the positive differences and the ranks of the negative differences computed. If firms issue management announcements to signal errors in existing earnings expectations, absolute forecast errors for sample firms are expected to be greater than absolute forecast errors for control portfolios. Therefore, the sum of positive ranks is expected to be significantly larger than the sum of negative ranks. The significance of the differences between the groups is tested using a one-sided Wilcoxon matched-pairs signed- rank test. H12o is rejected for small negative rank sums. The critical value for rejection [Bain, 1987] is approximated as: . n n+ .n+ (n)(n+1) 5 t”’ z” 24 + 4 ( ) 54 It is possible that results could be confounded by the number of analysts following each firm. As the firms are matched on industry, size and fiscal year end, there is no reason to assume that the number of analysts following the announcing firms is systematically different than the number of analysts following the non-announcers, however this issue is addressed in sensitivity tests. fli E E 1' E E ! lll' ! Hypothesis 3 compares the divergence of beliefs among analysts following firms issuing management announcements to the average divergence of beliefs among analysts following non-announcers. The test statistic is similar to that used in tests of Hypothesis 2, with the standard deviation of firms' outstanding forecasts used as measures of divergence of beliefs among analysts. The deviation is then standardized by the mean of the firms' outstanding forecasts to control for the variability in the size of earnings between firms. The test statistic is computed as follows: sosm,,-[ (3"?) 1100 (7 . 1) I t 55 where ofit a the standard deviation of the analysts forecasts for sample firm i, at time t, AF = the average of all individual analyst forecasts for sample firm i, issued within one month prior to the firms' issuance of a management announcement. oit The one month period is chosen because it has been shown that wide fluctuations in the ages of outstanding forecasts introduce measurement error into the computation of consensus among analysts [Brown and Han, 1991]. In addition, more recently issued (within 30 days) analyst forecasts have been shown to be more accurate than a consensus of all forecasts outstanding [Brown, 1991]. Control portfolios are formed using the same criteria as in tests of Hypothesis 2. The divergence of beliefs for the control portfolios is computed as a mathematical average of the divergence of beliefs of each of the firms within a portfolio, as shown below; (”ar. c)J 702 smru- “[fi—L 91: 100 ( ) P1 = the standard deviation of all analysts forecasts for control firm p, issued up to one month prior to the release of a management announcement by the corresponding sample firm i. where UAW,“ 56 AF“pit = the average of all analyst forecasts for control firm p, issued up to one month prior to the release of a management announcement by the corresponding sample firm i, the number of firms in the control portfolio for which firm i is the announcing sample firm. :0 II The difference between the standard deviations of the sample firms and those of the control firms are computed as: SDSMIc " 3M1}: (7 o 3) The differences are ranked according to their absolute values and the sums of the positive and negative ranks are computed. A one-sided Wilcoxon matched-pairs signed-rank test is used to assess significance. As it is expected that sample firms' standard deviations are larger than those of than control portfolios, Hi3o is rejected for very small values of negative rank sums. -1ig _o. o '._.g -‘: 1,3 x-Jlo_, -!.-! i. v- o - To test Hypothesis 4, quarterly and annual earnings announcement dates are collected for the sample firms over the five year test period. Earnings surprise is computed in two ways. To be consistent with Skinner [1991], surprise is computed as follows: 57 Price,_ 1 (3.1) where EPSI == actual earnings per share for the current quarter, EPS.,12 = actual earnings per share for the corresponding quarter of the previous year, Pricers = firm stock price, as of 1 month prior to the quarterly earnings announcement date. It was necessary for Skinner to use a naive model of market earnings expectations because the firms in his sample were relatively small, and therefore generally not followed by analysts. An advantage of this study is that the sample firms are required to be followed by IBES analysts. Therefore a better (more recent) approximation of earnings expectations, based on analyst forecasts, can be determined. Earnings surprise is computed as: Ems,-iumws, Price”1 (8 ° 2) where EPSI = actual earnings per share for the current quarter, AFEPSI = analyst forecasted earnings per share for the current quarter, PriceM = firm stock price, 1 month prior to the quarterly earnings announcement date. 58 Analyst forecasted earnings per share consists of an average of all analyst forecasts issued up to one month following the prior quarterly earnings announcement. This allows analysts to act on the information contained in the earnings announcement, while insuring the revisions occur prior to a firm's issuance of a management announcement. Firms are placed into deciles based on the magnitude and direction of quarterly earnings surprise. Each firm quarter is coded as to whether a management announcement was issued prior to the announcement of quarterly earnings. The percentage of firms having issued management announcements is then computed for each decile. Chi-square tests are used to determine the significance of the percentage differences between deciles. The use of the two methods to compute earnings surprise will provide evidence as to whether any significant findings are due to the inclusion of additional types of announcements, or the use of the more current measure of earnings expectations. 1 ‘2 '0. " ‘1 ao‘ - no. .‘ ..‘ i‘ _Q' O ‘ ‘ The test of the increase in the percentage of analysts revising their forecasts is similar to that used in Stickel [1989]. The number of analysts following each of the sample firms is identified from the IBES data base. As some of the management announcements considered in this study may have 59 long range income effects, the percentage of analysts revising their current and those revising their year-ahead forecasts are calculated. Management announcements made in the last month of a firm's fiscal year (and post fiscal year end comments on earnings) are not used in tests of forecast revisions for the current year, due to inadequate time for analyst revisions prior to the end of the fiscal year. The percentage of analysts revising their forecasts for a particular firm is computed as; NR m],- A: (100) (9) = the number of analysts revising forecasts for firm 1, during week t, where NRit An = the total number of analysts following firm 1, during week t, A three week event period is used based on the speed of analyst revisions following management earnings forecasts in Jennings [1987]. The abnormal percentage of analysts revising their forecasts is computed for each of the weeks 0 to +2 (with week 0 being the week of the management announcement) as follows: 1m“- PRn-‘fléj (10.1) 60 where PR1== average weekly percentage of analysts revising their forecasts over the 49 weeks surrounding the event period (weeks -26 to -1 and +4 to +26), t = week 0,...,week +3. An average abnormal weekly percentage of analyst revisions is also computed for both the first two weeks and the entire three week period as: E (PIER-PR?) (10.2) where n = 2 to 3 weeks included in the average. The mean abnormal percentage of analyst revisions (MARn) for each of the three weeks, and mean average abnormal percentage of analyst revisions (MAARR) over the two and three week periods, are computed as: (11.1) MAR“ ' 1N: I and l! mu (11.2) m], - '1 61 where N = the number of firms in the sample. To test the null hypothesis that the mean abnormal revisions equal 0, one sided t-statistics are calculated as; t - ”it, (12.1) omit and .MAAR t: - 1‘ (12.2) am], where o n is the estimated standard deviation of the mean percentage revisions over the weeks -26 to -1 and +4 to +26. H .! :e E E J ! E ! E I . Hypothesis 6 tests whether the magnitude of analyst forecast revisions immediately following the issuance of a management announcement is greater for announcing firms than non-announcing firms. To test Hypothesis 6, the average magnitude of analyst forecast revisions is computed for the sample firms and for control portfolios of similar non-announcing firms. The control portfolios are formed using the same criteria as in tests of Hypotheses 2 and 3. The average absolute forecast revisions for each of the announcing firms, over the three 62 week period‘ are computed as; It 3 [2“ F41: ' Fur-1 2 a-l Fat-1 N mwmwu- ”1 t“ f] (13.1) where Fur the revised forecast by analyst a, for firm i, issued up to three weeks following firm i's release of a management announcement, “t4 = the forecast by analyst a, for firm i, existing as of the release date of a management announcement by firm i, Nu the number of analysts revising their forecasts for firm i. t = the three week period following the release of a management announcement by firm i. The average absolute forecast revisions for the control portfolios over the three week period are computed as: (13.2) ‘5 As many firms had a small number of analysts revising their forecasts, tests of individual weeks were not feasible. 63 revised forecast by analyst a, for control firm p, issued up to three weeks following the release of a management announcement by the corresponding sample firm i, where FM,it “fib1 = forecast by analyst a, for control firm p, existing as of the release date of a management announcement by the corresponding sample firm i, Nu = the number of analysts revising their forecasts for control firm p. m = the number of firms in the control portfolio for which sample firm i is the announcing firm, t = the three week period following the release of a management announcement by sample firm i. The differences between the revision magnitudes of the sample and control firms are ranked according to their absolute values, and the sums of the positive and negative ranks are computed. A one-sided Wilcoxon matched-pairs signed-rank test is used to assess significance. As it is expected that sample firms' revision magnitudes will be greater than those of control portfolios, H60 is rejected for very small values of negative rank sums. CHAPTER IV RESULTS Results for each of the six hypotheses are presented in order, categorized as follows; tests of capital market reaction to the issuance of management announcements (H1), tests of potential motivations for firms to issue management announcements (H2, H3, H4) and analyst reactions to management announcements (H5, H6). MARKET REACTION Market reaction to the issuance of management announcements is tested in Hypothesis 1. As Hypothesis 1 does not make directional predictions regarding the market reaction, Patell's U-statistic (standardized stock return variances) is the chosen test statistic. If the market reacts to management announcements, the variance in stock returns is expected to be abnormally large immediately. following the issuance of an announcement. As the expected value of the U-statistic is 1, H11.o is rejected for U- statistics significantly greater than 1. One tailed z-tests are used to measure significance. Tests are conducted on the sample as a whole, the main categories and the sub-categories of announcements. The main 64 65 categories consist of officer earnings comments (sub-divided into categories of pre- and post-fiscal year end earnings comments, and general comments on overall company expectations), planned cost cuts (sub-divided into production changes, plant closings and work force cuts), capital expenditures (sub-divided into changes in planned capital budgets and expenditures for specific projects), product price changes, and miscellaneous announcements (generally a combination of two or more of the prior categories). Tests are conducted on the full sample of announcements, and also on sub-samples consisting of only those announcements specifically attributed to company officers or spokespersons. Firms with more than 25 missing returns during the testing and estimation periods are eliminated, resulting in the deletion of 43 announcements by 19 firms. Results are presented in Tables 8A - 8D. The number of announcements in each category, the computed u-statistic and the probabilities for the related z-statistics are shown. Panels A1 (A2) present results for the entire sample and the main categories for all announcers (officer/spokesperson announced). Panels El (82) present the same information for the subdivision of officer earnings comments, Panels C1 (C2) for cost cuts and Panels D1 (DZ) for capital expenditures. The announcements in their entirety are highly significant (p < .001), although the results appear to be 66 Table 8A Market Reactions - Main Classifications Panel A1 -- Full Sample, All Announcers' I0»P|-3|-2|-1I0T—1llz 3 All Mme-ante (nu-716) Uetetc 0.312 0.903 1.025 2.331 1.003 0.933 1.003 Prob(zu)d .999 .931 .323 .000** .440 .723 .453 Officer Eerninoe Cements (r1603) mm“ 0.323 0.952 1.195 3.937 0.399 0.929 0.957 Preb(zu)d .932 .719 .009** .000*9 .391 .305 .399 Cost Cute (ft-169) ustat° 0.390 0.750 0.993 1.114 1.035 0.949 1.119 Probtzu)d .993 .999 .512 .095 .221 .377 .140 Cepitel Expenditures (M121) umt‘ 0.991 0.750 0.374 1.043 1.123 0.321 1.204 91-33(20)“ .533 .997 .915 .302 .090 .974 .013* Price Ci'ienoee («an Uetetc 0.935 0.533 0.912 1.240 0.355 2.115 1.115 Prob(zu)d .533 .993 .719 .053 .332 .000** .224 Iiiecelleneoue (rt-78) mm" 0.740 1.434 1.231 1.333 0.992 0.332 0.334 ane announcements, regardless of attribution. b. Days are given in relation to the two day event window, with day zero being the average abnormal return over two days (the day the management announcement is issued and the previous day). c. U-stats are the squared abnormal stock returns for the event days, standardized by the estimated variance of the return over the 200 day estimation period. d. Prob(zu) is the probability associated with a one-sided z-test of observing values greater than the reported U- stat by chance (* indicates p <.05, ** indicates p <.01). 67 Table 8A (cont'd) Panel A2 -- Full Sample, Officer/Spokesperson Attribution' f q I -1 I 0 I 1 I 2 3 All Announcements (mfln) Ustetc 0.739 0.923 1.252 1.933 1.055 0.399 1.053 Prob(2u)d .999 .355 .000** .000** .224 .913 .221 Officer Earnings Consents (1:199) mm‘ 0.344 0.337 1.331 2.323 0.352 0.904 1.022 Probau)d .933 .905 .00033 .0003* .923 .329 .417 unto“: an») 03:31‘ 0.535 0.340 0.993 1.035 1.324 0.949 1.535 Prob(zu)d .999 .354 .503 .291 .000** .329 .000** Capital Expenditures ("QM 03:31“ 0.324 0.797 1.205 1.732 1.554 1.230 1.737 prob4zu)‘ .749 .732 .215 .001** .017- .142 .002** Price Changes (n-12) 03:31“ 0.333 0.410 1.123 0.755 0.702 0.723 0.309 Prob(2u)d .322 .924 .373 .722 .734 .745 .377 1IflumUuuas , ("an 1 Ustetc 0.357 1.333 1.535 1.204 0.930 0.373 0.739 1 prawn)" .930 .000“ .003" .149 .579 .950 .903 * repor 5 es s using on y those announcements specifically attributed to a named company official. Days are given in relation to the two day event window, with day zero being the average abnormal return over two days (the day the management announcement is issued and the previous day). U-stats are the squared abnormal stock returns for the event days, standardized by the estimated variance of the return over the 200 day estimation period. Prob(Z¢) is the probability associated with a one-sided z-test of observing values greater than the reported U- stat by chance (* indicates p <.05, ** indicates p <.01). 68 Table 88 Market Reactions - Officer Earnings Comments Panel Bl -- All Announcers' I - 03yb I -3 I -2 -1 I 0 I 1 I 2 I 3 Pre-Fiacal Year End (71-212) mm" 0.309 0.937 1.052 4.407 0.390 0.939 1.047 , "33153)" .9744 .3293 .2931 .0000» .3333 .7324 .3153 Peat-Fiacal Year End (n-55) Ustatc 0.714 0.973 1.913 4.940 1.153 1.122 0.771 Prob(zg)d .9292 .5517 .0000** .0000** .7323 .7324 .3310 General Cements (71-37) unu‘ 1.052 0.729 0.395 0.973 0.579 0.374 0.773 Prob(al)d .6129 .8769 .6293 .5398 .9633 .9177 .8266 ‘ Panel 82 -- Officer/Spokesperson Announced' anb -3 -2 -1 I 0 I 1 I 2 I 3 Pre-Fiacal Vear End (Ii-138) Ustatc 0.733 0.773 1.093 3.005 0.330 0.933 1.034 , "3345.1“ .9744 .9371 .2233 .0000” .9192 .5433 .3397 Peat-Fiscal Year End (71-27) Ubtatc 0.512 1.493 2.305 4.557 1.144 0.733 1.133 91-3315,)“ .9313 .0351- .0000“ .0000“ .3015 .3340 .2709 General Cements (n-SIo) » Uatatc 1.024 0.393 0.359 1.042 0.317 0.370 0.773 announcements, while Panel 82 uses only announcements specifically attributed to a named company official. b. Days are given in relation to the two day event window, with day zero being the average abnormal return over two days (the day the management announcement is issued and the previous day). c. U-stats are the squared abnormal stock returns for the event days, standardized by the estimated variance of the return over the 200 day estimation period. d. Prob(z¢) is the probability associated with a one-sided z-test of observing values greater than the reported U- stat by chance (* indicates p <.05, ** indicates p <.01). 69 Table 8C Market Reactions - Cost Cutting Announcements Panel C1 -- All Announcers' 03yb -3 -2 -1 0 1 2 3 I Production Consents ‘ (n83?) mm“ 0.472 0.373 0.541 0.352 0.337 1.133 0.435 wax.g' .mn .m3 3n. .n3 .mm 2&1 .mn Plant Closings new) ‘ Ustatc 0.790 0.544 0.733 1.323 0.493 0.573 0.431 Probt )d .324 .973 .323 .075 .937 .970 .939 F Berk Force Cuts * emu) i Ustatc 0.730 0.331 0.343 1.139 1.434 1.033 1.342 . m 1“ _ .935 .324 , 353 .129 .001 _ .323 .000** Panel C2 ff Officer/SpokespersonAnnounced‘ 03yb -3 -2 -1 0 1 - 2 3 Production Coolants (nsik) u.:.:° 0.721 1.010 1.001 1.117 0.970 1.110 0.333 1 mm id .737 .503 .500 .332 .523 .390 .303 n I Plant Closings (nsik) * Ustatc 0.399 0.709 0.933 1.333 0.433 0.330 0.433 ‘ mm 1“ .933 .737 .512 .201 .912 .955 .910 ‘ Bork Force Cuts (n359) Ustatc 0.523 0.342 0.330 1.033 2.015 1.011 1.929 (Prob! 1‘ .995 .302 .773 .421 .00033 _:f§9_u_-___:9°9f:_. a. 'ane repor s a e resu s o mar e es s using a announcements, while Panel C2 uses only announcements specifically attributed to a named company official. b. Days are given in relation to the two day event window, with day zero being the average abnormal return over two days (the day the management announcement is issued and the previous day). c. U-stats are the squared abnormal stock returns for the event days, standardized by the estimated variance of the return over the 200 day estimation period. d. Prob(Zd) is the probability associated with a one-sided Z-test of observing values greater than the reported U- stat by chance (* indicates p <.05, 3* indicates p <.01). 70 Table 8D Market Reactions - Capital Expenditures Panel D1 -- All Announcers' 2 l 3 Capital Budget Cements (men Ustatc 0.314 0.923 0.777 1.510 1.273 0.933 1.373 Probl id .323 .341 .339 .00533 .031 .523 .0273 Capital Project Cmnts on») Ustatc 1.033 0.329 0.947 0.712 0.973 0.395 1.107 Prebl )‘ .313 .935 .313 .953 .543 .932 .233 Panel D2 -- Officer/Spokesperson Announced“ ‘ I l. Oeyb I -3 I -2 I -1 I 0 I 1 I 2 2] 3 Capital Budget Cements Unn5 03:33“ 0.357 0.700 0.333 0.317 1.013 1.031 0.377 Problzu)d .339 .305 .313 .399 .433 .433 .323 3 Capital Projects (ti-1‘) Ustetc 1.032 0.913 1.319 1.104 0.334 1.454 0.332 Prob( )‘ . . .0133 a. Panel D1 reports the results of market tests using all announcements, while Panel D2 uses only announcements specifically attributed to a named company official. b. Days are given in relation to the two day event window, with day zero being the average abnormal return over two days (the day the management announcement is issued and the previous day). c. U-stats are the squared abnormal stock returns for the event days, standardized by the estimated variance of the return over the 200 day estimation period. d. Prob(z¢) is the probability associated with a one-sided z-test of observing values greater than the reported U- stat by chance (* indicates p < .05, *3 indicates p < .01). 71 driven by the officer earnings comments. When this category is further sub-divided, it appears that both pre- and post- fiscal year end earnings comments convey information, while the general outlook comments do not. It is possible that either the latter category does not provide new information to the market, or the announcements are too general to allow efficient interpretation. The cost cutting category exhibits marginal significance (p < .095) during the two-day event period. This appears to be a function of announcements of plant closings (p < .075), as neither of the other two categories approach significance. Due to the nature of these announcements it is possible that the information was initially released somewhere other than to the Wall Street Journal (perhaps locally). It is also possible that information leakage occurred prior to the official announcement, possibly even motivating the release of the announcement. An additional explanation is that, based on general economic or firm specific conditions, the market already had some expectation that cost cutting moves were likely. Therefore, unless the actual cuts are larger or smaller than anticipated, no new information is contained in the management announcement. A measure of market expectations of cost cuts, which is beyond the scope of this study, is necessary to statistically test this explanation. However anecdotal support is provided by occasional analyst 72 comments relating to announced work force cuts. These comments generally categorize the announced lay-offs as greater or less than expected, with occasional speculation as to whether further cuts would be forthcoming. The analyst interpretations might also explain the significance of the abnormal returns in days following announced work force cuts; although the market might not have been able to interpret (and thus did not react to) the original firm announcement, they might instead react to the analyst interpretations. As a whole, reactions following the capital expenditure comments are not significant, although the sub-category of planned changes in capital budgets attains significance in instances when the announcements are attributed to specific company officials. There are a number of likely explanations for these results. First, as previously discussed in Chapter 3, the information might not originate from a currently issued announcement, but instead consist of a summary of information contained in the company's annual report. This is particularly likely in the case of announcements not specifically attributed to a company official. The brevity of many of these announcements (combined with the fact that some similar announcements were specifically attributed to annual reports) indicate the possibility of their use as standard 'filler' by the Wall Street Journal. Another explanation is that these 73 announcements delineate firms responses to existing economic conditions. As in the case of cost cuts, the expenditures may have been anticipated, and thus already factored into current expectations. Finally, the announcements of specific projects might also be affected by prior information leakage, especially if the formal announcement is made late in the planning stage. Reactions to product price changes are marginally significant (p < .056). However, since price change announcements are issued infrequently and by a less diverse group of firms than those issuing other types of announcements, it is difficult to interpret these results. Specifically, firms in both the steel and publishing industries issued multiple price change announcements. Therefore, although the market appears to react to the announcements, their issuance might be an artifact of standard industry practice rather than a strategic use of voluntary disclosure. The category of miscellaneous announcements attained significance, not only immediately following the announcement, but also for the preceding two days. This suggests that the announcements were possibly issued in reaction to some existing market speculations affecting stock returns. It is also possible that the combination of- numerous types of announcements into a single miscellaneous category results in too much noise to allow meaningful 74 interpretation. Interestingly (except for the capital expenditure announcements, as previously discussed), the announcements do not appear to provide greater information when specifically attributed to a company official than when attributed to the firm in general. Categories attaining significance using all announcements also attained significance using the officer-spokesperson sub-category, while those that initially did not attain significance were also found insignificant when testing the respective sub- category. Although these results were initially considered somewhat surprising, it is possible that the market considers the Wall Street Journal a credible source of information, and/or interprets the absence of specific officer attribution to be a result of Journal editing rather than company strategy. In either case, attribution would not necessarily be an important factor in the interpretation of the announcement. For purposes of completeness, results are presented for all categories of announcements, although it is doubtful whether those categories reduced to a very few observations (specifically officer attributed announcements of production, plant closings, price changes and capital expenditures) provide generalizable information. As indicated, the restriction of announcements to only those specifically attributed to company officers, 75 drastically reduces the sample. As neither overall sample content (as previously discussed and shown in Table 5) nor market reaction appears to be significantly affected by officer attribution, the remaining hypotheses do not test these announcements as a separate category. POTENTIAL MOTIVATIONS FOR ISSUING ANNOUNCEMENTS Hypotheses 2, 3 and 4 test potential motivations for firms to issue management announcements. Hypothesis 2 tests whether firms issue the announcements as a signal that existing market expectations are in error, while Hypothesis 3 tests whether firms issue announcements to decrease market uncertainty regarding earnings. Hypothesis 4 tests whether firms release qualitative management comments as preemptors of large negative quarterly earnings surprises. WW Hypotheses 2 and 3 compare sample firms with matched control portfolios of non-issuers, in an attempt to determine motivations for the release of management announcements. The control firms are matched with the sample firms on the basis of industry, size and fiscal year end. Hypothesis 2 is tested by comparing the magnitude of analyst forecast errors between the two groups of firms. Hypothesis 3 compares sample and control firms as to the dispersion within existing analyst forecasts. 76 The method of matching the sample and control firms was previously discussed in Chapter 3, with the resulting sample detailed in Table 7. The final sample used in tests of Hypothesis 2 is further reduced by four announcements issued by firms with reported earnings of $0, which did not allow mathematical computation of the test statistic. If firms issue management announcements as a signal that existing market expectations of earnings are in error, analyst forecast errors immediately prior to the issuance of management announcements are expected to be greater for sample firms than for control portfolios of non-announcers at the same point in time. No predictions are made regarding the direction of the forecast errors, as firms have incentives to correct both over- and under-estimations of earnings. Tests are conducted on the entire sample and the main sub-categories of announcements. The errors are standardized using both actual earnings and forecasted earnings, with essentially the same results. Tables 9A and 98, present the respective results. While almost all of the standardized errors are less than 4 (with most of these less than 1), a few outliers were very large (ranging from 10 - 82). The existence of outliers is not a problem, as the significance of the differences between groups is measured using the non-parametric Wilcoxon matched-pairs signed-rank test. However, for purposes of table presentation, the 77 Table 9A Analyst Forecast Errors-Forecast“ Variable wean Median Range Std. Oev. 411 Announcers (n . 207) 433411b .3315 .1331 0.0014 - 5.000d 1.1343 431141b .3277 .0937 0.0000 - 5.000d .7349 Z-stat (prob): 3.303 (.0013) Officer Earnings Consents (n a 69) 433411b 0.7349 0.2309 0.0103 - 5.000d 1.2101 431141b 0.3337 0.0945 0.0023 - 3.3333 0.7721 Z-stet (prob)c 1.333 (0.0343) Cost Cutting Announcements (n I 61) 433411b 0.3140 0.1254 0.0014 - 5.000d 1.0324 433415 0.3133 0.0937 0.0000 - 5.000d 0.7339 Z-stat (prob): 2.739 1.00033) F Capital Expenditure Announcements (n - ££) 433411b 0.3433 0.1731 0.0024 - 5.000d 1.5134 431141b 0.3324 0.1113 0.0004 - 5.000d 0.7702 Z-stet (prob)° 1.902 (.0293) Price Changes (n I 18) 435411b 0.1323 0.0351 0.0053 - 1.1403 0.2333 431141b 0.3333 0.1111 0.0172 - 4.0333 0.9353 Z-stet (prob): -0.501 (.392) Miscellaneous (n I 15) 41131111b 0.3395 0.1331 0.0123 431141b 0.0373 0.0173 Z-stat (prob)c 0.454 (.3231 1,- '3 uses orecas e- earnings analyst forecast errors. _ b. ABSAM refers to forecast errors for sample firms, while ABMAT refers to forecast errors for the matched control portfolios. c. The z-statistic measures the significance of the differences in the forecast errors between the two groups (ABSAM - ABMAT). H2 (ABSAM not greater than ABMAT) is rejected for sma l values of negative rank sums (* indicates p < .05). d. Very large error outliers ( > 10) are truncated to a value of 5.0. Truncation had no effect on either rankings or significance, however truncated values report more meaningful mean and standard deviation values. 78 Table 98 Analyst Forecast Errors Actuala Variable Mean Median Range Std. Dev. All Annomcers (n a 207) 433411" 0.3315 0.1331 0.0014 - 5.000“ 1.1343 431141b 0.4515 0.0933 0.0000 - 5.000d 0.9545 Z-stet (prob)c 2.921 (.0023) Officer Earnings Cements (n a 69) 435411b 0.7349 0.2309 0.0103 - 5.000d 1.2101 431141b 0.4350 0.0395 0.0023 - 5.000d 0.9340 Z-stet (prob): 1.133 (.123) Cost Cutting Announcements (n I 61) 43S411b 0.3140 0.1452 0.0014 - 5.000d 1.0324 433413 0.4592 0.1095 0.0000 - 5.000d 0.9794 Z-stat (prob)° 2.345 (.0103) Capital Expenditure Annomcements (n a 44) 433411b 0.3433 0.1731 0.0024 - 5.000d 1.5134 431141b 0.4970 0.1331 0.0004 - 5.000d 0.9311 Z-stet (prob): 2.035 (n . .0193) Price Changes (n I 18) I 435411b 0.1323 0.0351 0.0053 - 1.1403 0.2333 431141b 0.2314 0.1111 0.0175 - 1.3243 0.3131 Z-stat (prob)c -1.24 (.393) Miscellaneous ( n I 15) 433411” 0.3395 0.1331 0.0127 - 1.9739 0.5375 431141b 0.3993 0.0739 0.0172 - 5.000d 1.2301 Z-set 10) are truncated to a value of 5.0. Truncation had no effect on either rankings or significance, however truncated values report more meaningful mean and standard deviation values. 79 inclusion of the outliers results in mean and standard deviations that are not meaningful. Therefore, tests were run after both eliminating the outlying observations, and truncating the outliers to a value of 5. Neither manipulation significantly changed the results, with the latter having no effect at all. As such, tables 9A and 98 are presented using truncated values for outliers. H2o is rejected for the sample as a whole, and for three of the sub-categories. Absolute forecast errors for firms issuing cost cutting and capital expenditure announcements are significantly greater for announcing than non-announcing firms under both computations of forecast errors. Officer earnings comments approach significance when errors are standardized by actual earnings and attain significance when standardized by forecasted earnings. Neither price changes nor miscellaneous announcements are significant under either computation. However as discussed before, the categories of price changes and miscellaneous announcements consist of a small number of announcements, perhaps not providing generalizable results. Overall, these results are somewhat inconsistent with the results of the market tests. Tests of analyst forecast errors suggest that managers issue various types of announcements to signal errors in prevailing market expectations, however it appears that market reaction to the announcements is primarily confined to those 80 announcements containing references to earnings. This suggests that either the market does not interpret the other types of announcements as signals of erroneous earnings expectations, or the content of the announcements is not specific enough to result in uniform changes in expectations. Hypothesis 3 tests whether managers issue announcements when analyst uncertainty regarding earnings is high. This is accomplished by comparing the divergence of beliefs (measured by the standard deviation of recent analysts forecasts) among analysts following the sample firms with the divergence of beliefs among analysts following control firms. If managers are motivated to issue announcements to reduce existing uncertainty regarding earnings, the standard deviation of analyst forecasts for the sample firms is expected to be greater than the standard deviation of the forecasts for the corresponding control portfolios. Results for the entire sample and main categories are presented in Table 10. As discussed earlier in regards to table presentation of the results of Hypothesis 2, Table 10 is also shown using truncated values for outliers. H3o is rejected for the sample as a whole and for several sub-categories, specifically capital expenditures and price changes. A comparison of the results of Hypotheses 2 and 3 shows that, except for the category of 81 Table 10 Analyst Divergence of Beliefs' Variable Mean Median Range Std. Oev. All Annemcers (n I 210) 305411b 0.1303 0.0744 0.0000 - 3.000d 0.3024 301141b 0.1411 0.0593 0.0000 - 3.000d 0.3004 , z-stat (0133)“ 1.374 (.035) , Officer Earnings Cements (Ii-69) 305411b 0.1323 0.0525 0.0000 - 1.5155 0.2242 501141b 0.1233 0.0433 0.0000 - 1.4533 0.2233 Z-stet (prob)c 0.986 (0.166) Cost Cuttim Amomcements (n I 61) 305411b 0.1910 0.0737 0.0000 - 3.2342 0.4403 304413 0.1514 0.0353 0.0032 1.4142 0.2371 Z-stat gprobi° 0.315 (.373) Capital Expenditure Amemcmnts (n I U.) 303411b 0.1707 0.0392 0.0000 - 1.7321 0.2733 301141b 0.0933 0.0303 0.0000 - 0.3315 0.1232 puuggwf La7nmv) Price Changes (11 I 18) 303411b 0.2397 0.1139 0.0092 - 1.2745 0.3254 501141b 0.7233 0.0574 0.0000 - 3.000d 0.7443 Z-stet (prob): 1.323 (.093) Miscellaneous (11 I 16) 30344” 0.0390 0.0539 0.0043 - 0.2772 0.0722 301141b 0.0342 0.0753 0.0294 - 0.2033 0.0525 - a 2'3” 333*" '2-068020" . a. e s an-ar- 3eV1a on 0 ex s ng irm orecas s (standardized by the firm's mean analyst forecast) is used as the measure of analyst divergence of beliefs. SDSAM is the divergence measure for sample firms, while SDMAT is the average divergence for control portfolios. The z-statistic measures the significance of the differences of the measures between the two groups. H3o (SDSAM not greater than SDMAT) is rejected for small values of negative rank sums (* indicates p < .05). Very large standard deviation outliers ( > 10) are truncated to a value of 3.0. Tests performed using actual and truncated values provide the same results, however truncated values report more meaningful mean and standard deviation values. 82 planned capital expenditures, announcements found significant when testing forecast errors were not significant when testing divergence of beliefs and vice versa. The discrepancy between the hypotheses in regards to the results of the officer earnings comments and the cost cuts is noteworthy. Since the issuance of a management announcement imposes costs (direct, proprietary or both) on a firm, it is reasonable to assume that firms will issue announcements only when the perceived benefits of issuance are greater than the perceived costs. The results of Hypotheses 2 and 3 suggest that firms perceive greater benefit in correcting erroneous (but on average fairly consistent) forecasts, than divergent (but on average fairly correct) forecasts. It is also possible that the issuance of an announcement is motivated by a combination of accuracy and uncertainty. As such, using a regression that includes an interaction between the two measures might clarify the results. To further explain the results, the two groups of firms were examined to determine whether significant differences existed in either the average number of analysts following the firms, or the average age of the outstanding forecasts. Systematic differences in the average number of analysts following the firms could affect the results, as larger numbers of forecasts mitigate the effects of individual analyst error. A comparison of the number of analysts 83 following each of the two groups did not indicate significant differences between the two groups (an average of 6.21 analysts per firm following the sample firms, and 5.17 analysts per firm following the control portfolios). Systematic differences between the groups as to the age of the outstanding forecasts could also be expected to affect results, as older forecasts are likely to introduce measurement error into the test statistics. However by design, only those forecasts issued within one month prior to the issuance of a management announcement are included, and so forecast age is not expected to influence the results. WWW Hypothesis 4, a replication and extension of Skinner [1991], tests whether managers are more likely to issue announcements to preempt large negative earnings surprises than otherwise. To be consistent with Skinner, a naive model of earnings expectations is used, consisting of actual earnings from the corresponding quarter of the previous year. Again for consistency, the firms are arranged into deciles based on the magnitude of earnings surprise (from largest negative earnings surprise to largest positive earnings surprise). The firms are also placed into three categories; very bad, intermediate and very good (with very 84 (bad) good defined as (-) + 5% earnings surprise). If firms issue management announcements as preemptors of large negative earnings surprises, the announcements are expected to be issued more frequently in quarters with large negative earnings surprises than at other times. The significance of the differences between deciles is tested using Chi-square statistics. The results are very similar to Skinners. Significantly more management announcements occur in periods of large negative earnings surprises than do otherwise, as shown in Table 11A. Management announcements are issued as preemptors of 13.67% percent of the largest negative earnings surprises. These results appear to be driven by officer earnings comments and cost cutting announcements. As there is some overlap in Skinner's announcements and those used in this study (both studies include qualitative earnings comments, which account for 60% of the preemptors of the largest negative earnings surprises), the tests are also run on a separate category consisting of all non- earnings comments, and on the main categories of the other types of announcements. When testing sub-samples, only those firms issuing an announcement of the type in the indicated sub-sample are retained. If firms issue more than one type of announcement during the 20 quarters in the test. period, the quarters containing the irrelevant announcements are deleted. This explains the reductions in firm quarters 85 Table 11A Management Announcements as Preemptors of Bad News Earnings8 All Announcements Panel 1 - Very Bad, Intermediate, and Very Good News Iotel b Range of c Firm Dimer (X) d Level Earnings Surprise Ouerters Preeapted Ouerters 1 < -.05 135 57 (13.67%) -.05 to .05 2,078 665 ( 9.12%) > .05 62 32 ( 9.17!) Chi Squeree Probability - .009 Panel 2 - Bad News Partitioned into Deciles i Total b Range of Firm Mater (1) d Oecile Earnings Surprisec Ouerters Preegted Ouarters 1 < -.03 595 86 (16.65%) 2 3.0360 to -.0160 593 73 (12.31%) 3 3.0139 to -.0060 639 63 ( 9.86%) 6 -.0059 to -.0010 668 75 (11.23%) 5 3.0009 to .0008 666 36 ( 7.33%) 6 .0009 to .0025 570 66 ( 7.721) 7 .0026 to .0050 606 36 ( 5.962) 8 .0051 to .0098 569 67 ( 8.26%) 9 .0098 to .0261 586 66 ( 7.51%) > 53 l 9.03%) «a O Chi Square. probability - 0.000 Actual earnings for the corresponding cpsrter of the previous year are used as the estimte of expected earnings. Firm qaerters are placed into levels (deciles) based on the magnitude of the qaerterly earnings surprise unexpected earnings standardized by stock price). Quarters containing the largest negative earnings surprise are assigned to level 1 (decile 1) and qierters containing the largest positive earnings surprise are assigned to level 3 (docile 10). Range of earnings surprise reports the wper and lower bouids of earnings surprise within the corresponding level (decile). Preemted (porters ere warters in idiich management amomcments are issued prior to the arsiomcment of actual (parterly earnings. Chi Square tests are used to measure significance of the differences between the levels (deciles), as to the proportion of cpsrters in idiich management amomcements are issued. 86 Table 118 Management Announcements as Preemptors of Bad News Earnings8 Officer Earnings Comments Panel 1 -Very Bad, Intermediate, and Very Good News Total b Range of Firm Mlmer (2) d Level Earnings Surprisec Ouarters Preeapted Ouarters 1 < -.05 236 36 (16.61%) -.05 to .05 2,793 201 ( 7.20%) .05 197 16 ( 7.11%) > Chi squaree Probability - .000 anl 2 , “3 Newwaiieint9“115,- Total Range of Firm Mlmer (%) Earnings Surprisec Ouarters Preemtod Ouarters < -.0370 325 63 (13.23%) -.O360 to -.0160 331 36 (10.88%) -.0130 to 323 27 ( 8.36%) -.0050 to 36 ( 9.07%) .0000 to 10 ( 6.03%) .0009 to 17 ( 5.28%) .0026 to 16 ( 5.08%) .0069 to 26 ( 7.36%) .0095 to 20 ( 6.31%) > ,, 20 ( 6.23%) 1 2 3 6 5 6 7 8 9 d O Chi squaree probability . 0.000 e. Actlml earnings for the corresponding glerter of the previous year are used as the estimate of expected earnings. ‘ b. Firm glarters are placed into levels (deciles) based on the magnitude of the glarterly earnings surprise (mexpected earnings standardized by stock price). Ouerters containing the largest negative earnings surprise are assimed to level 1 (docile 1) and glarters containing the largest positive earnings surprise are assigned to level 3 (docile 10). c. Range of earnings surprise reports the taper and lower bomds of earnings surprise within the corresponding level (docile). d. Preemted glerters ere giarters in ahich management moments are issued prior to the annomument of actual glorterly earnings. e. Chi Sglere tests are used to measure simificance of the differences between the levels (deciles), as to the proportion of giarters in which mnegement arviolaicements are issued. 87 Table 11C Management Announcements as Preemptors of Bad News Earningsa Non-Earnings Comments PMI 1 ' ”EYE“ Wtedat' and “Y G?” N“? Total In Range of c Firm I”? (%) d Level Earnings Surprise Quarters Preegpted Quarters 1 < -.05 226 23 (10.18%) -.05 to .05 2,860 263 ( 9.20%) > .05 18 ( 8.16%) Chi Square. Probability - .758 Panel 2 - Bad News PartitionedintoDecilesA Total Range of Firm liner of Earnings liurprise‘= Quarters Preemted Quarters < -.0350 336 60 (11.98%) -.0360 to -.0160 363 60 (11.66%) -.0130 to -.0060 366 36 ( 9.89%) -.0050 to -.0010 356 39 (11.02%) 0.0000 to .0007 255 22 ( 8.63%) .0008 to .0027 362 31 ( 9.06%) .0028 to 320 18 ( 5.63%) .0056 to 332 23 ( 6.93%) .0112 to 336 26 ( 7.19%) > 329 d 1 2 3 6 5 6 7 8 9 i 1‘ 9 Chi Square. probability . 0.069 a. Actual earnings for the corresponding marter of the previous year are used as the estimate of expected earnings. b. Firm mortars are placed into levels (deciles) based on the magnitude of the marterly earnings surprise unexpected earnings standardized by stock price). Quarters containing the largest negative earnings surprise are assigned to level 1 (decile 1) and qsarters containing the largest positive earnings surprise are assimed to level 3 (decile 10). c. Range of earnings surprise reports the taper and lower beads of earnings surprise within the corresponding level (decile). d. Preewted marters are cpsrters in idiich management amomcements are issued prior to the moment of actual (parterly earnings. e. Chi Smare tests are used to measure significance of the differences between the levels (deciles), as to the proportion of quarters in idiich management annomcements are issued. 88 Table 110 Management Announcements as Preemptors of Bad News Earnings3 Cost Cutting Comments Panel 1 - Very Bad, Intermediate, and Very Good News total 8 Range of c Fire Iunber (%) d Level Earningsg§urprise Quarters Preempted Quarters 1 < -.05 111 16 (12.61%) -.05 to .05 1,697 98 ( 6.55%) > .05 115 5 ( 6.35%) 41..=...; Chi squaree probability . .02: Panel 2 - Bad News Partitioned into Deciles _fi { total b Range of c Firs Mulber of d Decile Earnings Surprise Quarters Preenpted Quarters 1 < -.0370 173 22 (12.72%) 2 -.0360 to -.0150 175 12 ( 6.86%) 3 -.0160 to -.0070 173 13 ( 7.51%) 6 -.0060 to -.0010 207 15 ( 7.25%) 5 .0000 to .0009 132 12 ( 8.33%) 6 .0010 to .0030 173 8 ( 5.20%) 7 .0031 to .0057 171 8 ( 6.68%) 8 .0058 to .0106 171 11 ( 6.29%) 9 .0105 to .0310 175 5 l 2.91%) 10 > .0312 172 11 ( 6.60%) Chi Square. probability . 0.059 a. Actual earnings for the corresponding quarter of the previous year are used as the estimate of expected earnings. b. Firs quarters are placed into levels (deciles) based on the magnitude of the quarterly earnings surprise (unexpected earnings standardized by stock price). Quarters containing the largest negative earnings surprise are assigned to level 1 (decile 1) and quarters containing the largest positive earnings surprise are assigned to level 3 (decile 10). c. Range of earnings surprise reports the upper and lower bounds of earnings surprise within the corresponding level (decile). d. Preeapted quarters are quarters in uhich aenagesent announce-ants are issued prior to the announcesent of actual quarterly earnings. e. Chi Square tests are used to neasure significance of the differences between the levels (deciles), as to the proportion of cparters in idiich nenegeeent «momma are issued. £P9 Table 11E Management Announcements as Preemptors of Bad News Earningsa Capital Expenditures Panel 1 - Very Bad, Intermediate,and VeryGoodNews_ Total b Range of Fire lumber (%) d Level EarningsSurprisec Quarters Preempted Quarters I 1 < -.05 77 3 ( 3.90:) i -.05 to .05 1,136 02 ( 7.22%) ‘ 3 > .05 90 6 < 6.67%) j Chi Squaree Probability I .538 Panel 2- Bad News Partitioned intoDeciles Total Range of Fire Earnings Surprisec Quarters Preeapted Quarters < -.0320 132 9 ( 6.82%) -.0310 to -.0150 12 ( 8.57%) -.0160 to -.0060 12 ( 8.82%) -.0050 to -.0020 9 ( 7.20%) -.0010 to .0007 9 ( 7.69%) .0008 to .0029 12 ( 8.63%) .0030 to .0066 2 ( 1.65%) .0065 to 8 ( 6.11%) .0117 to .0269 9 ( 6.77%) > 9 ( 6.98%) 1 2 3 b 5 6 7 8 9 1 ..a O Chi Square. Probability . 0.596 a. Actual earnings for the corresponding quarter of the previous year are used as the estimate of expected earnings. b. Firs quarters are placed into levels (deciles) based on the aagnitude of the quarterly earnings surprise (unexpected earnings standardized by stock price). Quarters containing the largest negative earnings surprise are assigned to level 1 (decile 1) and quarters containing the largest positive earnings surprise are assigned to level 3 (decile 10). c. Range of earnings surprise reports the upper and lower bounds of earnings surprise within the corresponding level (decile). d. Preeapted quarters are quarters in which manage-ant announce-ants are issued prior to the sma-lent of actual cpsrterly earnings. e. Chi Square tests are used to measure significance of the differences between the levels (deciles), as to the proportion of qasrters in which aanagenent manna are issued. 90 Table 11? Management Announcements as Preemptors of Bad News Earnings8 Price Changes Panel 1 - Very Bad, Intermediate, and Very GoodNews_ total b Range of Fin I”? (%) d Level Earnings Surprisec Quarters Preemted Quarters < -.05 60 2 ( 5.00%) -.05 to .05 323 26 ( 8.05%) 3 > .05 25 6 (26.00%) Chi Squaree probability - .017f Panel 2 - Bad NewsPartitioned into Deciles total Range of Fire I”? of Earnings Surprisec Quarters Preemted Quarters < -.0330 39 2 ( 5.13%) -.0310 to -.0130 60 7 (17.50%) -.0120 to -.0070 39 1 ( 2.56%) -.0060 to -.0020 60 6 (10.00%) -.0010 to .0006 36 2 ( 5.56%) to 60 3 ( 7.50%) to 38 2 ( 5.26%) to 38 1 ( 2.63%) 39 3 ( 7.69%) 39 9 (23.08%) 1 2 3 6 5 6 7 8 9 d O miamn°wwwnnysmuf a. Actual earnings for the corresponding garter of the previous year are used as the estimate of expected earnings. b. Fire glarters are placed into levels (deciles) based on the mitude of the glarterly earnings surprise (uiexpectsd earnings standardized by stock price). Quarters containing the largest negative earnings surprise are assimed to level 1 (decile 1) and glarters containing the largest positive earnings surprise are assigned to level 3 (decile 10). c. Range of earnings surprise reports the mper and lower bonds of earnings surprise within the corresponding level (decile). d. Preamtad garters are garters in diich nanagaaant alwiolnceaents are issued prior to the amolncaaent of actual giarterly earnings. e. Chi Sglare tests are used to neasure significance of the differences between the levels (deciles) as to preeption by amolnceaents. f. Due to the saall tamer of observations, Chi Sgiare results are probably not valid, this table is presented for comleteness only. 511 Table 11G Management Announcements as Preemptors of Bad News Earningsa Miscellaneous Comments Panel 1 -Very Bad, Intermediate, and Very Good News_ Total b Range of Fire lumber (%) Level EarningsSurprisec Quarters Preempted Quarters 1 < -.05 56 6 ( 7.61%) -.05 to .05 819 57 ( 6.96%) > .05 57 1 ( 1.75%) Chi squarae probability - .306 5 Bad NewsPartitioned into Deciles Panel 2 Total Range of Firm Huber of d Earnings Surprisec Quarters Preemted Quarters < -.0260 95 9 ( 9.67%) -.0250 to -.0110 97 11 (11.36%) -.0100 to -.0050 87 6 ( 6.60%) -.0060 to .0000 10 ( 7.96%) .0001 to 5 ( 7.81%) .0013 to .0026 6 ( 6.59%) .0027 to .0065 6 ( 6.17%) .0067 to .0086 6 ( 6.55%) .0089 to 7 ( 7.65%) > , 2 ( 2.17%) 1 2 3 I. 5 6 7 8 9 d 0 Chi Sglare' Probability I 0.323 a. Actual earnings for the corresponding giarter of the previous year are used as the estimate of expected earnings. b. Firm gaarters are placed into levels (deciles) based on the mamitude of the gasrterly earnings surprise (mexpected earnings standardized by stock price). Quarters containing the largest negative earnings surprise are assimed to level 1 (decile 1) and giarters containing the largest positive earnings surprise are assigned to level 3 (decile 10). c. Range of earnings surprise reports the lpper and lower bolnds of earnings surprise within the corresponding level (decile). d. Preamted glarters are glarters in mich nnagement amomcmnts are issued prior to the arrlolncemant of actual gnrterly earnings. e. Chi Sgaare tests are used to measure significance of the differences between the levels (deciles), as to the proportion of glarters in diich management armomcements are issued. 92 for the sub-samples, as shown in Tables 11B - llG. Non-earnings announcements, when divided into deciles, are marginally significant, but are not significant when divided into the very bad/good news categories. It appears as though in the extreme deciles, the non-earnings announcements are more prevalent at the lower end of the decile (The decile containing the most negative earnings surprise is comprised of 40 announcements with negative surprise of at least 3.5%. However, when the most negative surprise is defined to be at least 5%, the number of announcements decreases to 23). Therefore, significance is not attained when the number of categories is reduced. Officer earnings comments and cost cutting announcements are highly significant, both when measured in deciles and when separated into very bad/good news categories, whereas capital expenditures and miscellaneous announcements do not approach significance in either presentation. This is consistent with the results of Hypothesis 2, in which managers appear to issue comments on earnings and cost cutting announcements when the market expectations of earnings is in error. It also appears that firms are more likely to issue the announcements when earnings are overestimated than when they are underestimated. Price changes appear to attain significance (although in the opposite direction than expected), however due to the 93 small number of observations in a majority of the cells, Chi-square results are not valid for this sub-sample. The results are presented for purposes of completeness only. The initial test of Hypothesis 4 extended Skinner's study by expanding the scope of the announcements tested. An additional planned extension was the use of IBES quarterly analyst forecasts as a more precise measure of earnings expectations. Due to problems with data collection, the results of this extension might not be meaningful, and are presented for completeness only. Contrary to the results obtained when using the naive measure of earnings expectations, significance is obtained only for the complete sample, and only when divided into deciles. No sub-categories attain significance under either partitioning. There are at least two explanations for these findings, the first being that when using a more recent measure of earnings expectations, H4o can not be rejected. The second explanation is that missing and inaccurate IBES quarterly data result in noisy earnings surprise measures, and thus obscure actual results. The problem of missing data, a result of the fact that fewer analysts issue forecasts of quarterly earnings than of annual earnings, resulted in the deletion of some of the original firms. Firms were deleted if either data in all quarters corresponding to the issuance of management announcements were missing, or if data for more than 10 of 94 the 20 firm quarters was missing. An exception was made for missing fourth quarter data. By design, fourth quarter forecasts are collected one month after the issuance of actual third quarter results. Therefore, it is possible that analyst fourth quarter forecasts are implicitly rather than explicitly issued (through forecasts of annual earnings). Implicit fourth quarter forecasts are thus estimated and substituted for the missing data. Implicit forecasts are computed as follows: 3 AFR-AFm-E A0, 0.1 where All”E = the average annual earnings forecast outstanding for firm i, existing one month after announcement of third quarter earnings, AQi = actual earnings announced for quarter i. To investigate the validity of this assumption, the implicit forecasts are compared with the actual fourth quarter forecasts for firms with available fourth quarter data. The significance of the differences between the two groups is measured using t-tests, and as shown in Table 12, the differences between the actual and implicit forecasts are not significant at conventional levels (p < .05). A further deletion of firms was required due to discrepancies in IBES and Compustat as to the reported firms' fiscal year-ends. A comparison of IBES and Compustat actual earnings was done in an attempt to discover whether 95 Table 12 Actual“ and Implicitb Fourth Quarter Forecasts Dimer of Actual Fourth Ouarter Mean Difference Earnings Between d Year Forecasts Forecastsc T-statistic (Prob) 1986 66 .2266 1.2530 (.2168) 1987 65 .1831 0.8056 (.6268) 1988 60 -.1083 -0.6355 (.5275) 1989 86 .2306 1.8126 (.0735) 63 .0533 0.5236 Actual earnings forecasts are the average of all outstanding fourth quarter earnings forecasts (issued up to one month after the announcement of third quarter earnings) for a particular firm. Implicit earnings forecasts are the outstanding annual earnings forecasts (issued up to one month after the announcement of actual third quarter earnings) minus the sum of the actual earnings for the first three:quarters of the year. This column reports the mean of the differences of (implicit - actual) earnings forecasts, for all firms with available fourth quarter earnings forecasts. T-statistics are used to measure the significance of the mean differences from zero. 96 the discrepancy was due to an error in reporting the date of the fiscal quarter, or reporting the earnings information. However, as earnings reported by IBES do not always correspond to Compustat earnings (or GAAP in general) the cause of the discrepancy could not be determined. Similar discrepancies in reported IBES quarterly data are reported in Philbrick and Ricks [1991]. Although IBES reports that this problem has since been corrected, the years included in this study are of an early enough time period so as to possibly provide unreliable quarterly data. Therefore, the absence of significance in tests of Hypothesis 4 is likely due to data problems rather than improper specification of the hypothesis. For purposes of completeness, results of the tests are reported in Tables 13A - 13G. ANALYST REACTIONS Hypotheses 5 and 6 test analyst reactions to management announcements. Hypothesis 5 tests whether analysts are more likely to revise their forecasts in the period following a management announcement than at other times throughout the year. Hypothesis 6 tests whether the magnitude of analyst forecast revisions following the issuance of a management announcement is greater for announcing firms than for matched control portfolios of non-announcers. Both hypotheses use the restricted sample of 319 announcements for test purposes. 97 Table 13A Management Announcements as Preemptors of Bad News Earnings8 All Announcements Panel 1- Very Bad ,‘Intermediate, and VeryGood News Total Firm Dimer (%) Quarters Preggpted Quarters Earnings Surprisec < - .05 135 18 (13.33%) -.05 to .05 2,078 257 (12.37%) > .05 62 6 ( 9.52%) Chi Square. probability . .629 Panelz - BadNewsPartitionedinto Deciles Total . b Range of c Firm Number (%) d Decile Earnings Surprise Quarters Preempted Quarters 1 < -.0265 227 27 (11.89%) i 2 -.0261 to -.0111 225 62 (18.67%) i 3 -.0110 to -.0062 223 19 ( 8.52%) 6 -.0061 to -.0036 233 37 (15.88%) 5 -.0033 to -.0016 229 17 ( 7.62%) 6 -.0013 to .0000 256 30 (11.72%) 7 .0001 to .0013 185 22 (11.89%) 8 .0016 to .0035 231 32 (13.85%) 9 .0036 to .0107 227 29 (12.78%) 10 > .0108 219 26 (10. 96%) Chi Square. probability . 0.017 a. TEES analyst forecasts are used as the estimate of expected earnings. b. Firm quarters are placed into levels (deciles) based on the magnitude of the quarterly earnings surprise (unexpected earnings standardized by stock price). Quarters containing the largest negative earnings surprise are assigned to level 1 (decile 1) and quarters containing the largest positive earnings surprise are assigned to level 3 (decile 10). c. Range of earnings surprise reports the upper and lower bounds of earnings surprise within the corresponding level (decile). d. Preempted quarters are quarters in which management announcements are issued prior to the announcement of actual quarterly earnings. e. Chi Square tests are used to measure significance of the differences between the levels (deciles), as to the proportion of quarters in which management announcements are issued. 98 Table 138 Management Announcements as Preemptors of Bad News Earnings8 Officer Earnings Comments Panel 1 - Very Bad,Intermediate, andVery GoodNews Total 1 b Range of c Firm Rmr (%) d \ Level Earnings Surprise Quarters Preeaptad Quarters i 1 < -.05 80 11 (13.75%) 2 -.05 to .05 1,- 179 (16.52%) 1 3 > .05 29 6 (13.79%) _ ’ Chi Square. probability . .977 Panel 2 - Bad News Partitioned into Deciles Total 1 b Range of c Firm liner (%) d l Decile Earnings Surprise Ouarters Preemted Quarters 1 < -.0250 136 18 (13.63%) 2 -.0269 to -.0110 168 30 (20.27%) 3 -.0100 to -.0061 126 9 (7.26%) 6 -.0058 to -.0030 161 26 (16.91%) 5 -.0029 to -.0010 131 15 (11.65%) 6 .0000 to .0000 136 21 (15.67%) 7 .0001 to .0013 106 20 (18.87%) 8 .0016 to .0036 135 20 (16.82%) 9 .0037 to .0116 133 18 (13.53%) 10 > .0118 136 19 (13.97%) Chi Square. probability - 0.197 a. TEES analyst forecasts are used as the estimate of expected earnings. b. Firm garters are placed into levels (deciles) based on the witude of the garterly earnings surprise (mexpected earnings standardized by stock price). Ouarters containing the largest negative earnings surprise are assigned to level 1 (decile 1) and garters containing the largest positive earnings surprise are assimed to level 3 (decile 10). c. Range of earnings surprise reports the wiper and lower bunch of earnings surprise within the corresponding level (decile). d. Preanted garters are garters in leiich mt manta are issued prior to the want of actlal garterly earnings. e. Chi Sgare tests are used to measure significance of the differences between the levels (deciles), as to the proportion of garters in idiich aanagemant manta are issued. 99 Table 13C Management Announcements as Preemptors of Bad News Earnings8 Panel 1 - Very Bad, Intermediate, and Very Good News d. Panel 2 -Ead News Partitioned into Deciles Non-Earnings Comments 7 _ ‘. Total Range of Firm Inter (%) Earnings Surprisec Quarters Preempted Quarters < -.05 92 10 (10.87%) -.05 to .05 1,353 151 (11.16%) > .05 22 1 ( 6.55%) d Chi Squaree probability - .617 Total Range of Firm I”? of Earnings Surprisec Quarters Preempted Quarters < .0260 150 15 (10.00%) -.0230 to -.0100 158 23 (16.56%) -.0090 to -.0052 175 17 ( 9.71%) -.0066 to -.0031 16 (11.97%) -.0029 to -.0010 156 15 ( 9.76%) .0000 to 18 (13.16%) .0001 to 11 ( 7.80%) .0017 to 17 (11.72%) .0061 to 15 (10.07%) > 17 (12.06%) 1 2 3 6 5 6 7 8 9 d 1 ° Chi Square. probability . 0.806 IBES analyst forecasts are used as the estimate of expected earnings. Firm quarters are placed into levels (deciles) based on the magnitude of the quarterly earnings surprise (unexpected earnings standardized by stock price). Quarters containing the largest negative earnings surprise are assigned to level 1 (decile 1) and quarters containing the largest positive earnings surprise are assigned to level 3 (decile 10). Range of earnings surprise reports the upper and lower bounds of earnings surprise within the corresponding level (decile). Preampted quarters are quarters in which management announcements are issued prior to the announcement of actual quarterly earnings. Chi Square tests are used to measure significance of the differences between the levels (deciles), as to the proportion of garters in idiich unagement moments are issued. 100 Table 13D Management Announcements as Preemptors of Bad News Earningsa Cost Cutting Comments Panel lp- Very Bad, Intermediate, and Very Good News_ Total Range of Firm lumber (%) Earnings Surprisec Quarters Preempted Quarters < -.05 62 7 (11.29%) -.05 to .05 766 60 ( 7.13%) > .05 13 0 (0.00%) Chi squaree probability - .356 d Panel 2- BadNews Partitionedinto Deciles_ Total Range of Fine Number of Earnings Surprisec Quarters Preempted Quarters < -.0290 86 9 (10.67%) -.0230 to -.0120 88 8 ( 9.09%) -.0090 to -.0060 96 6 ( 6.25%) -.0057 to -.0030 86 5 ( 5.95%) -.0028 to -.0010 85 7 ( 8.26%) .0000 to .0000 72 8 (11.11%) .0001 to .0017 81 2 ( 2.67%) .0018 to .0039 9 (10.71%) .0060 to .0122 7 ( 8.75%) > .0126 6 ( 7.06%) _ d 1 2 3 6 5 6 7 8 9 ‘ d 0 Chi Square. probability - 0.629 a. TEES analyst forecasts are used as the estimate of expected earnings. b. Firm quarters are placed into levels (deciles) based on the magnitude of the quarterly earnings surprise (unexpected earnings standardized by stock price). Quarters containing the largest negative earnings surprise are assigned to level 1 (decile 1) and quarters containing the largest positive earnings surprise are assigned to level 3 (decile 10). c. Range of earnings surprise reports the upper and lower bounds of earnings surprise within the corresponding level (decile). d. Preempted quarters are quarters in which management announcements are issued prior to the announcement of actual quarterly earnings. e. Chi Square tests are used to measure significance of the differences between the levels (deciles), as to the proportion of quarters in which management announcements are issued. f. Due to the small number of observations in a majority of the cells, Chi Square results are probably not valid -- this table is presented for completeness only. 101 Table 13E Management Announcements as Preemptors of Bad News Earnings8 Capital Expenditures Panel 1 - Very Bad, Intermediate, and Very Good News Total b Range of Firm liner (%) d Level Earnings Surprisec Quarters Prethed Quarters 1 < -.05 25 0 ( 0.00%) -.05 to .05 581 66 ( 7.57%) 2 I 3 > .05 11 0 ( 0.00%) 1 — Chi Square. probability . .230 Panel 2 - Bad News Partitioned into Deciles Total Range of Firm liner of Earnings Surprisec Quarters Preempted Quarters d < -.0100 68 2 c 2.96%) -.0170 to -.0090 57 5 < 8.77%) -.0080 to . 60 5 ( 8.33%) -.0060 to . 57 6 (10.53:) -.0020 to . 73 6 c 5.68%) .0000 to 62 2 ( 3.23:) .0006 to 61 5 ( 3.20:) .0021 to 60 5 ( 3.33:) .0069 to 61 6 ( 9.86%) > so 6_( 6.90%) f‘_ Chi squaree probability . 0.760 OOVOUIbHN-D d O a. TDES analyst forecasts are used as the estimate of expected earnings. b. Firm garters are placed into levels (deciles) based on the magnitude of the garterly earnings surprise unexpected earnings standardized by stock price). Olarters containing the largest negative earnings surprise are «aimed to level 1 (decile 1) and garters containim the largest positive earnings surprise are assigned to level 3 (decile 10). c. Range of earnings surprise reports the laper and lower bouids of earnings surprise within the corresponding level (decile). d. Preespted garters are garters in idiich management amolncements are issued prior to the amomcament of actual garterly earnings. e. Chi Sgare tests are lead to measure simificance of the differences between the levels (deciles), as to the proportion of garters in idiich management araiomcements are issued. f. Due to the small nldaer of observations in a majority of the cells, Chi Sgare results are probably not valid -- this table is presented for completeness only. 102 Table 13? Management Announcements as Preemptors of Bad News Earningsa Price Changes Panel 1 - Very Bad, Intermediate, and Very Good News Total b Range of Firm lumber (%) 8 Level Earnings Surprise‘= Dlarters Prethed Quarters 1 < -.05 7 1 (16.29%) 2 -.05 to .05 231 19 ( 8.23%) .05 0 ( 0.00%) Chi Square. probability - .775 > Panelz-Bad NewsPartitionedintoDecilespfi # "7' ‘7 Total I b Range of c Firm Number of d Decile EarningsiSugprise Quarters Preempted Quarters 1 < -.0160 25 3 (12.00%) 2 -.0130 to -.0070 26 2 ( 8.33%) 3 -.0060 to -.0060 25 2 ( 8.00%) 6 -.0039 to -.0021 23 3 (13.06%) 5 -.0018 to -.0010 15 0 ( 0.00%) 6 -.0001 to .0000 32 3 ( 9.38%) 7 .0001 to .0021 25 3 (12.00%) 8 .0022 to .0066 26 0 ( 0.00%) 9 .0067 to .0123 23 1 ( 6.35%) 10 > .0126 26 3 (12.50%) Chi Square. probability . 0.723 a. IBES analyst forecasts are used as the estimate of expected earnings. b. Firm quarters are placed into levels (deciles) based on the magnitude of the quarterly earnings surprise (unexpected earnings standardized by stock price). Quarters containing the largest negative earnings surprise are assigned to level 1 (decile 1) and quarters containing the largest positive earnings surprise are assigned to level 3 (decile 10). c. Range of earnings surprise reports the upper and lower bounds of earnings surprise within the corresponding level (decile). d. Preempted quarters are quarters in which management announcements are issued prior to the announcement of actual quarterly earnings. e. Chi Square tests are used to measure significance of the differences between the levels (deciles), as to the proportion of quarters in which management announcements are issued. f. Due to the small number of observations in a majority of the cells, Chi Square results are probably not valid -- this table is presented for completeness only. 103 Table 136 Management Announcements as Preemptors of Bad News Earningsa Miscellaneous Comments Panel 1-Very Bad,Intermediate, and Very GoodNews A i ‘. Total b Range of Firm lumber (%) d Level Earnings Surprisec Quarters Preempted Quarters 1 < -.05 26 2 ( 8.33%) -.05 to .05 381 26 ( 6.82%) ms , t Chi Square. probability . .360 ) Panel 2A- Bad News Partitionedinto Deciles Total Range of Firm d EarningsSurprisec Quarters Preegpted Quarters < -.0190 62 ( 7.16%) -.0100 to -.0090 60 ( 7.50:) -.0000 to -.0050 65 ( 6.67%) -.0069 to -.0030 32 (12.50:) -.0029 to -.0010 66 ( 6.35%) .0000 to .0000 69 (12.26%) .0001 to .0015 33 ( 9.09:) .0016 to .0067 63 ( 6.65%) .0069 to .0115 39 ( 2.56%) > .0118 60 ( 5.006) 1 Chi squarae probability s0.739 if) (if ODVOU‘bMNd NdNUO~O~UUU _a 1 O a. TEES analyst forecasts are used as the estimate of expected earnings. b. Firm quarters are placed into levels (deciles) based on the magnitude of the quarterly earnings surprise (unexpected earnings standardized by stock price). Duarters containing the largest negative earnings surprise are assigned to level 1 (decile 1) and quarters containing the largest positive earnings surprise are assigned to level 3 (decile 10). c. Range of earnings surprise reports the upper and lower bounds of earnings surprise within the corresponding level (decile). d. Preempted quarters are quarters in which management announcements are issued prior to the announcement of actual quarterly earnings. e. Chi Square tests are used to measure significance of the differences between the levels (deciles), as to the proportion of quarters in which management announcements are issued. f. Due to the small nulaer of observations in a majority of the cells, Chi Sgare results are probably not valid -- this table is presented for completeness only. 104 W Hypothesis 5 tests whether analysts appear to use either the issuance of, or the information contained in, management announcements to revise their existing forecasts. If management announcements contain information used by analysts, the abnormal percentage of analysts revising their forecasts during each of the three weeks immediately following issuance of a management announcement is expected to be greater than the average weekly percentage of analyst revisions occurring throughout the year. H25o is rejected for abnormal percentage revisions greater than zero, with significance measured by one-sided t-tests. Results for the entire sample and for the main categories of announcements are presented in Tables 14A - 14F. Results are presented individually for week 0 to week +2, along with cumulative two and three week averages. For comparison purposes, the abnormal percentage revisions for week -1, the week prior to the management announcement, are also reported. As expected, none of the week -1 abnormal revisions are significant. H50 is rejected for the sample as a whole for week +1 and week +2, and for the cumulative two week average. The only sub-sample for which HISo is rejected is officer earnings comments, with significance attained in week +1 and week +2 and for both cumulative two and three week averages. In no case is H50 rejected in week 0. This could be explained by a lag in the time it takes analysts to 105 Table 14 Analysts Issuing Revisions Panel A -- All Announcements (n=285) Time Period. Abnormal Revisions Prob.‘ Heek (-1) 0.00627 .9212 Beak ( 0) -0.01039 .8702 Beak (+1) 0.23739 .0015** Heek (+2) 0.09606 .1683 Cumulative Average Beaks (0,1) 0.10816 .0286* Cumulative Average Heeks (0,1,2) 0.10026 .0198* Panel B -- Officer Earnings Comments (n=118) ‘ Time Period. Abnormal Revisions' Prob.c week (-1) -0.08656 .3163 week ( 0) -0.10569 .2263 week (+1) 0.63859 .0001** week (+2) 0.25681 .0127' Cumulative Average weeks (0,1) 0.25278 .0039'* I Cumulative Average weeks (0,1,2) 0.25358 .0008** ‘ Panel C -- Cost Cuts (n=66) I Time Period. Abnormal Revisions Prob.c Heck (-1) -0.03616 .7667 H week ( 0) 0.03985 .8068 week (+1) -0.02106 .8626 Beck (+2) -0.11567 .3362 Cumulative Average weeks (0,1) 0.00960 .9197 Cumulative Average weeks (0,1,2) -0.03166 .6592 Panel D -- Capital Expenditures (n=43) Time Period. Abnormal Revisions Prob.c Beak (-1) 0.20570 .2965 week ( 0) 0.20616 .1821 Beak (+1) -0.08530 .5039 Heek (+2) 0.15666 .6785 Clmlative Average weeks (0,1) 0.06062 .5823 Cumulative Average Heats (0,1,2) 0.09183 .6182 106 Table 14 (cont'd) Panel E -- Price Changes (n=27) Time period' Abnormal Revisions week (-1) 0.12611 .6686 week ( 0) 0.16529 .6050 week (+1) -0.16236 .1126 week (+2) -0.16225 .6678 mlative Average Hacks (0,1) 0.00166 .9899 " Cwlative Average weeks (0,1,2) 0.05507 .6102 Panel F -- Miscellaneous (n=31) Time Period' Abnormal Revisions Prob.c I week (~11 0.06125 .6269 week ( 0) -0.22307 .2205 week (+1) 0.06656 .6578 week (+2) -0.26817 .1027 Cumulative Average weeks (0,1) -0.06825 .6559 Clmlative Average weeks (0,1,2) -0.15933 .2130 II a. Time periods are given in relation to the issuance of a management announcement, with week 0 being the week in which the announcement is issued. The cumulative average refers to a mathematical average of the revisions over the weeks indicated. b. Abnormal revisions are the percentage of analysts revising their forecasts during the corresponding week minus the average weekly percentage of analysts revising forecasts for the 48 weeks surrounding the event period (weeks 0 - (+3)). c. T-statistics are used to measure the significance of the abnormal revisions from zero, with the corresponding probabilities presented here (* indicates significance at < .05, ** indicates significance at < .01). 107 react to the management announcements through forecast revisions, or to a lag in recording the revision in the IBES data base. 7 Some sub-samples of management announcements are assumed to contain information that has greater long term than short term impact, specifically cost cuts and planned capital expenditures. Therefore, tests of Hypothesis 5 are also run using analyst year-ahead forecast revisions. The scarcity of analysts providing long range forecasts combined with the requirement that firms be followed by at least three analysts, reduces the sample considerably. The reduction occurs in both the number of announcements available for analysis, and the number of weeks in the estimation periods. Therefore, after running the test as originally planned, the restriction on the number of analysts following the firms was relaxed so as to include firms followed by two analysts. The tests were also run after eliminating firm announcements for which the number of weeks in the estimation period was less than 20, less than 30, and less than 40.. As the results are essentially the same in all cases, Tables ISA-15F present results for the original manipulation only. 7 According to Brown and Han [199], IBES personnel indicate that the reporting lag problem has been trivial since 1982 when the data collection process became centralized. However, it is still possible that some of the revisions are actually made prior to the recorded 'input date', which would bias against results. 108 Table 15 Analysts Issuing Year-Ahead Revisions Panel A -- All Announcements (n=198) Time Period. Abnormal Revisions Prob.° Beek (-1) -0.03763 .5800 Beak ( 0) -0.03586 .6109 Beak (+1) 0.19719 .0212‘ Beak (+2) 0.10821 .2012 Cwlative Average Beaks (0,1) 0.08068 .1636 leative Average Beaks (0,1,2) 0.08985 _ .0638* Panel B -- Officer Earnin 3 Comments (n =83) Time Period. Alnornl Revisions Prob.c Beek (-1) -0.06939 .3163 Beak ( 0) -0.05938 .2263 week (+1) 0.67867 .0051++ ; Beak (+2) 0.39899 .0153* Clmlative Average Beaks (0,1) 0.20965 .0256" leative Average Beaks (0,1,2) 0.27276 .0019“ J Panel C -- Cost Cuts (n=49) Time Period. Abnormal Revisions Prob.c ‘ Beek ('1) -0.13678 .2879 l Beak ( 0) 0.06590 .7855 Beak (+1) 0.03221 .8013 Beak (+2) -0.06006 .7669 ‘ leative Average Beaks (0,1) 0.03905 .7261 ‘ Clmlativa Average Beaks (0,1,2) 0.01269 .8906 Panel D -- Ca ital E enditures (n=24) I Time Period. Abnoml Revisionsg Prob.c Beak ('1) 0.20570 .9618 Beak ( 0) 0.20616 .7813 Beak (+1) -0.08530 .0615* Beak (+2) 0.15666 .6509 Cwlative Average Beaks (0,1) 0.06062 .1027 leativa Average Beaks (0,1,2) 0.09183 109 Table 15 (cont'd) Panel E -+ Price Changes (n=18) Time Period. Abnornl Revisions Prob.‘ ‘ Beek (-1) 0.69677 .1195 \ Beak ( 0) 0.15525 .5829 t Beak (+1) 0.18267 .3986 : Beak (+2) -0.06653 .6263 ! Clmlative Average Beaks (0,1) 0.16896 .6169 Cwlative Average Beaks (0,1,2) 0.09713 .6811 Panel P -- Miscellaneous (n=24) Time Period. Beek (-1) Beak ( 0) Beak (+1) Beak (+2) Cwlative Average Beaks (0,1) leative Average Beaks (0,1,2) a. Time periods are given in relation to the issuance of a management announcement, with week 0 being the week in which the announcement is issued. The cumulative average refers to a mathematical average of the revisions over the weeks indicated. b. Abnormal revisions are the percentage of analysts revising their forecasts during the corresponding week minus the average weekly percentage of analysts revising forecasts for the 48 weeks surrounding the event period (weeks 0 - (+3)). c. T-statistics are used to measure the significance of the abnormal revisions from zero, with the corresponding probabilities presented here (* indicates significance at < .05, ** indicates significance at < .01). 110 The results for the long range forecast revisions are similar to those for current year revisions. H5o is rejected for the entire sample of announcements, but this again appears to be driven by officer earnings comments. The only significant difference between the long and short- term revisions occurs in the sub-sample of capital expenditure announcements. Although capital expenditure announcements did not appear to motivate analysts to revise their current earnings forecasts, a significantly greater than normal number of analysts revised their long-term forecasts. Although the abnormal number of analyst forecast revisions indicate the announcements contain information relevant to firm valuation, the absence of significant market reactions (Hypothesis 1) suggests that the market might not be able to interpret the information. As such, if the announcements are issued to correct existing market expectations, an alternate form of communication would be more effective. Overall, the results of Hypothesis 5, with respect to short term revisions, are consistent with Hypothesis 1, in which significant market reaction was observed for officer comments only. As it has been shown that analyst forecasts incorporate information in stock prices [Abarbanell, 1991] it is possible that the analyst revison is driven by the market reaction to the announcement, rather than the announcement itself. 111 WWW Hypothesis 6 is also a test of whether analysts appear to use management announcements when revising their forecasts. Analyst forecast revisions immediately following the issuance of management announcements by sample firms are compared to analyst forecast revisions of matched non- announcers at the same point in time. If management announcements include information relevant to analysts, or if the issuance of a management announcement signals that existing forecasts are in error, it is expected that the magnitude of forecast revisions are greater for firms issuing management announcements than for non-announcers. The sample used in tests of Hypothesis 6 is a sub- sample of that used in tests of Hypotheses 2 and 3. In order to reduce the effects of individual analyst forecast errors, both sample and control firms were required to be followed by at least two analysts. After eliminating firms that did not meet this criterion, the final sample was reduced to 74 firm announcements, with control portfolios containing an average of 2.6 forecast revisions per firm, and sample firms an average of 7.1 forecast revisions. The disparity in the number of revisions is not unexpected, as the results for Hypothesis 5 indicated a larger number of analysts revise their forecasts following a management announcement. A one-sided Wilcoxon matched pairs signed rank test is 112 used to test significance between the two groups, with H6o rejected for very small values of negative rank sums. As shown in Table 16, H60 is rejected for the sample as a whole, primarily due to the sub-category officer earnings comments. The small number of firm announcements retained for this sample indicates that a separate analysis for each of the categories may not be meaningful. However, for purposes of completeness the analysis was performed, with these results also presented in Table 16. In addition, tests were run on the combined sample of non-earnings comments. None of the sub-categories attained significance. It is possible that the magnitude of the forecast revisions could be driven by a systematic bias in the age of the outstanding forecasts for either group. Although this is not expected to be the case, the average forecast age was computed and found not significantly different between the two groups. The sample firms had a mean forecast age of 89 days (standard deviation 73.527 days) while the control portfolios had a mean forecast age of 88 days (standard deviation of 71.39 days). Overall, it appears that analysts react to management earnings comments both through the number and magnitude of forecast revisions. They do not however, appear to react to other types of management announcements. 113 Table 16 Magnitude of Analyst Forecast Revisions“ Variable Mean Median Rangg Std. Dev. All Announcers (n = 76) ssvsawb 0.1709 0.1002 0.0107 - 1.1221 0.1601 REVMATb 0.1611 0.0772 0.0066 - 1.7955 0.2676 Z-stat (prob)c 2.166 (.015*) Officer Earnings Cements (nI30) nevsanb 0.1996 0.1002 0.0162 - 1.1221 0.2369 nevw11b 0.1019 0.0666 0.0160 - 0.6092 0.1061 Z-stat (prob)c 2.666 (.006") Mon-Earnings Comments (n I 66) pevsanb 0.1522 0.0976 0.0107 - 0.6675 0.1312 REVMATb 0.1992 0.0612 0.0066 - 1.7955 0.3556 Z-stat (prob)c 0.976 (.166) Cost Cutting Announcements (n I 18) REVSAMb 0.1176 0.0933 0.0107 - 0.6875 0.1312 REVMATb 0.1808 0.0967 0.0269 - 1.5062 0.3351 Z-stat (prob)c 0.605 (.365) Capital Expenditures (n I 16) pevsaub 0.1919 0.1159 0.0206 - 0.6675 0.1661 REVMATb 0.2715 0.0662 0.0086 - 1.7955 0.6765 Z-stat (prob)c 0.620 (.268) Price Changes (n I 5) REVSAMb 0.1139 0.0670 0.0670 - 0.2660 0.0916 REVMATb 0.0626 0.0506 0.0262 - 0.1996 0.0721 Z-stat (prob)c 0.676 (.251) L Miscellaneous (n I 5) asvsanb 0.1626 0.0976 0.0705 - 0.3605 0.1313 REVMATb 0.1591 0.0926 0.0665 - 0.3192 0.1166 mm (min) «251-» a. Revision magnitudes are computed as the absolute value of (revised forecast - previous forecast) standardized by previous forecast. b. REVSAM is the measure of the test statistic for the sample firms, while REVMAT is the average statistic for the control firms. c. The Z-statistic measures the significance of the differences of the deviations between the two groups. N60 (REVSAM no greater than REVMAT) is rejected for small values of negative rank sums (** p < .01, * p < .05). a ‘1..- CHAPTER V CONCLUSIONS This chapter presents the conclusions and implications of the study. The general conclusions are discussed first, followed by the implications and contributions of the research. A discussion of both the limitations of the study, and suggestions for future research are also presented. GENERAL CONCLUSIONS The purpose of this study was to investigate potential strategic uses of information disclosure by firms. Four specific types of commonly issued management announcements were identified and tested; qualitative officer earnings comments, announcements of planned cost cuts, changes in planned capital expenditures and product price changes. In addition, a miscellaneous category was used for announcements that contained elements of more than one of the other categories. Potential motivations for the issuance of the announcements, and subsequent market and analyst reactions to the disclosures were examined. Qualitative officer earnings comments were found 114 115 significant in both motivation and reaction tests. The comments were issued at times when analysts annual earnings forecast errors were significantly greater for sample firms than for matched control portfolios. The announcements were also significantly more likely to be issued prior to 'bad news' quarterly earnings announcements than to either 'good' or 'neutral' news announcements. Significant abnormal stock returns, abnormal percentages of analysts revising their forecasts, and abnormal magnitudes of analyst forecast revisions immediately following the issuance of the announcements suggest that both the market and analysts react to the issuance of qualitative earnings comments. Cost cutting announcements, such as layoffs, plant closings and production changes were found significant in motivation tests but not in reaction tests. The announcements of planned cost cuts were issued at times when analyst forecast errors for annual earnings were significantly greater than forecast errors for matched control portfolios. The announcements were also more likely to be issued prior to quarterly 'bad news' earnings announcements than at other times. However, the issuance of the announcements was not followed by a significant analyst reaction, and the reaction by the market was only marginally significant. As discussed previously, this may have been due to information leakage prior to formal disclosure, or official disclosure somewhere other than the Wall Street 116 Journal. It is also possible that these announcements are interpreted as signals of long-term rather than short-term earnings changes. Therefore, the absence of a subsequent market reaction could be explained as market uncertainty in interpreting the information. The remaining announcements do not reach significance in either the motivation or the reaction tests, with the exception of a significant abnormal percentage of analysts revising their year-ahead annual earnings forecasts following announcements of changes in planned capital expenditures. One explanation for the results is that these types of announcements may be routinely rather than strategically disclosed. Additionally, the sources of the capital expenditure and price change announcements are not always clear. If the sample includes announcements that are not actually current firm releases, the variables would contain measurement error, thus confounding the results. The results may also have been affected by the constraints placed on the data collection. In general, capital expenditure and price change announcements are not issued as frequently as either qualitative earnings comments or announcements of planned cost cuts. After deleting announcements to control for potential confounders, the sample sizes of these types of announcements were sometimes too small to allow meaningful statistical tests. Finally, the miscellaneous announcements might have 117 been more informative if a hierarchy of disclosure had been hypothesized so as to enable inclusion of these announcements into the appropriate main categories. IMPLICATIONS AND CONTRIBUTIONS The main contribution of this study is the insight it provides into the use of voluntary disclosure by firms, with implications for policy makers, firms and investors. As stated in Pownall et. al. [1993], policy debates on mandatory disclosure of qualitative information suggest a need for investigation into the information content of alternative forms of disclosure. An analysis of market and analyst reactions to qualitative disclosures could provide insight as to the most cost effective means for firms to provide information to the public. Finally, information regarding potential motivations for firms to issue qualitative disclosures may allow investors to better interpret the disclosures. In this study, types of disclosures commonly made by firms were identified, potential motivations behind their release, and market reactions to the disclosures were analyzed. Two categories of disclosure, qualitative earnings comments and planned cost cuts, appear to be issued by firms as signals of errors in prevailing market expectations of earnings. However, only the earnings comments are reacted upon by either analysts or the stock 118 market. Although qualitative earnings comments are not as specific as point or range estimates of earnings, the reactions by both analysts and the market following their issuance, suggest that the announcements are considered both credible and informative. Therefore, if qualitative earnings comments are less costly to issue than hard earnings forecasts, they could be used by firms as an efficient means of conveying earnings expectations to the market. The results of this study also provide an explanation for the reported scarcity of publicly announced management earnings forecasts. Early management earnings forecast studies typically constrained the definition of earnings forecasts to include only hard point or range estimates of earnings. Therefore, the reported scarcity of forecasts may not indicate the existence of less than optimal information disclosure, but instead suggest that early researchers were overly restrictive in their definition of earnings forecasts. The implications of the results for the cost cutting announcements are not as obvious. Although firms have a propensity to issue the announcements when prevailing earnings expectations are in error, neither analysts nor the market appear to react to the announcements. Whether or not these types of disclosures provide effective communication 119 requires further study. If firms issue the announcements to provide the market with information to facilitate accurate earnings predictions, this type of announcement does not appear to be an effective means of communication. If, on the other hand, firms are issuing the announcements as evidence that they have met legal requirements requiring disclosure of all relevant firm information, they may serve their purpose. The announcements may also be issued to explain future bad news earnings announcements in an attempt to mitigate the expected negative market reaction. In either case, further research is needed to determine whether the announcements serve as effective communication. Neither price changes nor changes in planned capital expenditures appear to be used as strategic information disclosures. This may be a function of these types of disclosure or of the limitations of this study. LIMITATIONS The main limitation of the study relates to the generalizability of the results. The initial problem with generalizability is one found in any study of voluntary announcements; the inherent self selection bias of the sample. Since all firms in the sample were selected based on their voluntary choice to issue a management announcement, the results may not be generalizable to firms that do not choose to do so. 120 Another limitation is due to the data availability requirements, particularly the requirement that all sample firms be followed by analysts. This resulted in the elimination of very small firms. The matching constraints used to identify control firms (size, industry, and absence of management announcements during the event period) resulted in the elimination of some very large firms. Therefore the results may not be generalizable to firms at either end of the size spectrum. Although no other single source of announcements is expected to be systematically better than the Wall Street Journal, the use of a single source to obtain sample announcements is expected to introduce editorial bias into the sample. This could affect the study through either the omission of sample firms whose announcements were not reported in the Wall Street Journal, or the inclusion within the control portfolios of firms that had actually issued management announcements during the event period. As the latter would be expected to bias against results, it is expected to be less of a problem than the omission of potential sample announcements. The use of SIC codes to assemble control portfolios is also expected to introduce error in to the matching process, due to the divergence of firms in the same SIC industry codes [Amit & Livnat, 1990]. This is also expected to bias 121 against results, and therefore is not considered a major problem. FUTURE RESEARCH The results of this study suggest a number of ideas for future research in the area of management disclosure. During the data collection stage of this study, it became apparent that there exist firms that issue qualitative earnings comments in some years, and issue point or range forecasts in others. Therefore, an extension of this study would be to determine the motivations behind, and the variation in the reactions to the format of forecast disclosure. Specifically, are firms more likely to issue qualitative forecasts when subsequently announced earnings are 'bad' than at other times, and is the magnitude of the market reaction to qualitative forecasts smaller than the reaction to quantitative forecasts? Another extension is an examination of disclosures between competing firms. Using a more refined methodology for matching firms, incorporating risk and leverage in addition to size and industry, an examination of firms responses to competitors disclosures would be interesting. For example, a qualitative earnings announcement may be a timely strategy used to offset negative market reactions following announcements issued by competitors. Extensions investigating firms in periods subsequent to 122 the issuance of cost cutting announcements as 'bad news' earnings preemptors would also be interesting. Specifically, a study examining the likelihood of shareholder litigation following 'bad news' earnings announcements preempted by cost cutting announcements as compared with 'bad news' earnings not preempted. Finally, a market study of responses to preempted and non-preempted 'bad news' earnings would be interesting. APPENDIX 123 The Appendix presents examples of announcements from each of the categories tested. As many of the announcements were included in lengthy articles, only a summary of the main point is shown here. Company names have been omitted, and comments in brackets refer to Wall Street Journal editorial comments regarding the announcement. QUALITATIVE OFFICER COMMENTS Qualitative officer comments are sub-divided into specific earnings comments, and comments on the general company outlook, as follows: Earnings Comments ...scaled back earnings projections. Said 1989 will be 'moderately lower' than projected, and 1990 won't be as 'robust' as expected..."we are making adjustments based on what we have learned in 1989, and expect that growth in sales and earnings will continue next year and that profits in 1991 will be substantially greater than in 1990.” ...expects 'lackluster' earnings in the fourth quarter [declined to be more specific] as company braces for a downturn in the steel industry and an economic slowdown... ‘ ...gross margins will not fulfill expectations for the balance of 1989, although third quarter earnings will be greater than last year due to a one time gain from the sale of a facility... ...expects loss for the third quarter and for all of 1990 as the company refocuses on marketing and gets out of manufacturing. Fiscal 1991 is targeted as the 'turnaround year'... 124 General Comments ...expects net income to grow an average of 35% per year through 1992, assuming an average gold price of about $400 an ounce [would not disclose predictions of current earnings]... ...expects U.S. sales to remain steady, but expects growth in Europe and the Far East... ...estimates a 3% market share in the U.S. in 1990... ...business is growing at a 'startlingly high' 40%..."the market in general is doing well, and we're doing at least as well as the market." : COST CUTTING COMMENTS The category of cost cutting comments is also sub divided, as shown below: Plant Closings ...plans to close 1 or 2 plants because of industry slow down. [Declined to specify which plants]..."We have too many plants in our system, so the older or most inefficient capacity has got to go"... ...closing all mini-branches in discount stores... "It proved to be a tough market. It's simply very difficult to convince people to do business in a different way"... ...plans to temporarily close plants to decrease excess production capacity...plans to keep inventories in line with orders... ...plans to close an automotive coatings plant and a paint thinner plant by early 1991, to reduce excess capacity... 125 Work Force Cuts ...will slash work force by 8% over the next six months "to respond to anticipated cuts in defense spending, and to become more competitive in 1990... ... will cut U.S. exploration and production staff by about 10%...the moves were prompted by an internal forecast that prices, though up recently, will be relatively depressed over the next 5 to 10 years... ...plans to decrease work force a further 2,000 [which is less than expected]...”at this time we don't have plans for more... ... will restructure domestic businesses, eliminating jobs and saving money...expected to be completed by early next year... Production ...citing a drop in gas prices, is cutting back on natural gas production by 90%...said selling gas at prices less than economic levels is not consistent with maximizing shareholder value... ...plans to stop production for about a week at four newsprint mills, and expects to schedule further production cuts because of soft demand... ...plans to halt production of airborne warning and control systems aircraft for an indefinite period in 1991...plans will change if it gets more customers for the surveillance plane... ...plans to stop production by June 30th, of its lithium battery, which has been plagued with problems since introduction in 1986...despite the move, the company still maintains the battery is a technological breakthrough... 126 CAPITAL EXPENDITURES The category of capital expenditures is sub-divided into planned expenditures for special projects, and planned changes in overall capital budget spending. loqhuhh' Special Projects ...directors approved a 3 year, $260 million program to expand and modernize copper production facilities...the capital spending will complete the 1 transition from custom smelting to fully integrated b production... ...will spend about $1.1 billion Canadian over the next 3 years to upgrade and expand cable TV and cellular phone interests...”the spending program is aimed at insuring the survival of the company's cable TV business in the face of possible long—term competition from phone companies, as well as expansion of fast-growing cellular phone network"... ...will pump $105 million into building a casino complex and adding rooms to existing hotels...hasn't yet worked out the financing, but will use both internal and external sources... ...plans to spend $19 million to double the capacity of one of their plastics plants... 127 Capital Budget Comments ...expects 1989 capital and exploration spending to increase to about $4.2 billion...”sees plenty of good investment opportunities that could keep spending at or above 1989 levels in real terms for the next several years"... ..."as we're estimating oil prices of $19 per barrel capital spending is therefore budgeted to be about the same in 1990 as in 1989"... ...plans to spend about $320 million for oil and gas development this year, up from $200 million in 1989... ...plans to spend $540 million on capital projects this year, up 48% from last year...of the total, about $242 million will be spent on oil and gas, $175 million on chemical projects and $127 million on marketing... Product Price Changes ...raising prices 12-15%...justified based on company's absorption of several large price increases over the past year... ...plans to cut prices and eliminate the practice of negotiating deals with customers... ...raising prices effective 9/18...explained increase by citing rising production and distribution costs... ...plans to increase advertising rates..."given the publication's efficiency in reaching a uniquely influential audience we feel this increase is both fair and reasonable"... 128 Miscellaneous Comments ...will reduce work force by about 17%...also will post a wider than expected operating loss for the period due to price cutting in the industry... ...will reduce work force 15% to bolster productivity...research and development spending will also be trimmed to bring it in line with the industry...expected revenue for 1988 to increase 7% [declined to comment on profit]... ...lowering expectations for fiscal 1989 revenue...is considering consolidating operations and will probably out work force in the near future...takeover appears less likely than last year, consequently plans to revitalize growth strategy before making acquisitions... ...hopes to expand in Asia and Eastern Europe in order to shrink its heavy reliance on the United States, although the U. S. will still probably represent 40% of total business... LI ST OF REFERENCES LIST OF REFERENCES Abarbanell, J.S. "Do Analysts' Earnings Forecasts Incorporate Information in Prior Stock Price Changes?" Journal of Accounting and Economics, 14:1991, 147-165. Ajinkya, B., and M. Gift. "Corporate Managers' Earnings Forecasts and Symmetrical Adjustments of Market Expectations.” Journal of Accounting Research, Autumn 1984, 425-444. Amit, R. and J. Livnat. ”Grouping of Conglomerates by Their Segments' Economic Attributes: Towards a More Meaningful Ratio Analysis.” Journal of Business Finance 8 Accounting, Spring 1990, 85-100. Baginski, S. and J. Hassell. "The Market Interpretation of Management Earnings Forecasts as a Predictor of Subsequent Financial Analyst Forecast Revision.” Accounting Review, January 1990, 175-190. Bain, L.J., and M. Engelhardt, 1987. Introduction to Probability and Mathematical Statistics. Boston: PWS Publishers. Ball, R.J., and P. Brown. "An Empirical Examination of Accounting Income Numbers". Journal of Accounting Research, Autumn 1968, 158-179. Basi, B. A., K.J. Carey, and R.D. Twark. "A Comparison of the Accuracy of Corporate and Security Analysts' Forecasts of Earnings”. Accounting Review, April 1976, 244-254. Brown, L.D. "Forecast Selection When All Forecasts Are Not Equally Recent." International Journal of Forecasting, 1991, 349-356. , and J.C.Y. Han. ”The Impact of Earnings Announcements on Convergence of Beliefs.” Unpublished working paper, School of Management, State University of New York at Buffalo, October 1991. Chambers, A., and S. Penman. ”Timeliness of Reporting and Stock Price Reaction to Earnings Announcements". Journal of Accounting Research, Spring 1984, 21-47. Cox, C. ”Further Evidence on the Representativeness of Management Earnings Forecasts.” Accounting Review, 692-702. Dye, R.A. ”Proprietary and Nonproprietary Disclosures". Journal of Business, 1986, 331-366. 129 l. 130 "Disclosure of Nonproprietary Information". Journal of Accounting Research, Spring 1985, 123-145. Elton, E.J., M.J. Gruber, and M.N. Gultekin. "Professional Expectations: Accuracy and Diagnosis of Errors." Journal of Financial and Quantitative Analysis, December 1984, 351-363. Fama, E. "Agency Problems and the Theory of the Firm". Journal of Political Economy, April 1980, 288-307. Gibbins, M., A. Richardson, J. Waterhouse. "Management of Corporate Financial Disclosures: Opportunism, Ritualism, Policies, Processes”. Journal of Accounting Research, Spring 1990, 121-143. . ”The Management of Financial Disclosure: Theory and Perspectives”. Unpublished monograph, 1991. E Gray, 8., L. Radebaugh, and C. Roberts. "International Perceptions of Cost Constraints on Voluntary Information Disclosures: A Comparative Study of U.K. and U.S. Multinationals”. Journal of International Business Studies, 1990, 597-622. Hassell, J.M. and R.H. Jennings. ”Relative Forecast Accuracy and the Timing of Earnings Forecast Announcements". Accounting Review, January 1986, 58-75. Healy, P. and K. Palepu. ”Earnings Information Conveyed by Dividend Omissions". Journal of Financial Economics, 21:2, 1988, 149-176. . ”The Effect of Firms' Financial Disclosure Strategies on Stock Prices". Accounting Horizons, March 1993, 1-11. Hoskin, R., J. Hughes, and W. Ricks. "Evidence on the Incremental Information Content of Additional Firm Disclosures Made Concurrently with Earnings”. Journal of Accounting Research, Supplement 1986, 1-32. Jennings, R. ”Unsystematic Security Price Movements, Management Earnings Forecasts, and Revisions in Consensus Analyst Earnings Forecasts". Journal of Accounting Research, Spring 1987, 90-110. King, R., G. Pownall, and G. Waymire. ”Expectations Adjustment via Timely Management Forecasts: Review, Synthesis, and Suggestions for Future Research." Journal of Accounting Literature, Vol. 9, 1990, 113-144. 131 Lees, F. ’-3 I 8 cs- - . . 9° 1 - 1 no . - .-ts. The Conference Board, Report #804, 1981. Lev, B., S. Penman. "Voluntary Forecast Disclosure, Non- Disclosure, and Stock Prices". Journal of Accounting Research, Spring 1990, 49-76. Lipe, R.C. ”The Information Contained in the Components of Earnings". Journal of Accounting Research, Supplement 1986, 37-64. A McDonald, C.L. "An Empirical Examination of the Reliability of Published Predictions of Future Earnings". Accounting Review, July 1973, 502-510. Patell, J. ”Corporate Forecasts of Earnings per Share and Stock Price Behavior: Empirical Tests". Journal of Accounting Research, Autumn 1976, 246-276. Penman, S. "An Empirical Investigation of the Voluntary Disclosure of Corporate Earnings Forecasts.” Journal of Accounting Research, Spring 1980, 132-160. Philbrick, D.R. and W.E. Ricks. "Using Value Line and IBES Analyst Forecasts in Accounting Research." Journal of Accounting Research, Autumn 1991, 397-417. Pownall, G., C. Wesley, and G. Waymire. "The Stock Price Effects of Alternate Types of Management Earnings Forecasts." The Accounting Review, October 1993, 896-912. Ruland, W. "The Accuracy of Forecasts by Management and by Financial Analysts”. Accounting Review, April 1978, 439- 447. , S. Tung, and N. George. "Factors Associated with the Disclosure of Managers' Forecasts." Accounting Review, July 1990, 710-721. Sheperd, W. 1970. Market Power and Economic Welfare. New York, Random House. Skinner, D. J. ”Why Firms Voluntarily Disclose Bad News". Working Paper, University of Michigan, 1992. Stickel, S. "The Timing of and Incentives for Annual Earnings Forecasts Near Interim Earnings Announcements". Journal of Accounting and Economics, 1989, 275-292. 132 Thompson, R., C. Olsen, J.R. Dietrich. "Attributes of News About Firms: An Analysis of Firm-Specific News Reported in the Wall Street Journal Index". Journal of Accounting Research, Autumn 1987, 245-274. Trueman, B. ”Why Do Managers Voluntarily Release Earnings Forecasts?" Journal of Accounting and Economics, 1986, 53- 71. Verrecchia, R. "Discretionary Disclosure”. Journal of Accounting and Economics, 1983, 179-194. Waymire, G. "Additional Evidence on the Information Content of Management Earnings Forecasts". Journal of Accounting Research, Autumn 1984, 703-718. Wright, C., J. Groff. "Uses of Indexes and Data Bases for Information Release Analysis". Accounting Review, January 1986, 91-100.