AN ANALYSIS OF THE RELIABILITY OF MANAGEMENT ~ EARNINGS FORECASTS PUBLISHED IN ALTERNATIVE FORMATS AND INVESTIGATION OF SELECTED MANAGEMENT FORECAST DISCLOSURE PRACTICES Dissertation for the Degree Of Ph. D. MICHIGAN STATE UNIVERSITY WILLIAM CHARLES BOYNTON 1976 This is to certify that the thesis entitled AN ANALYSIS OF THE RELIABILITY OF MANAGEMENT EARNINGS FORECASTS PUBLISHED IN ALTERNATIVE FORMATS AND INVESTIGATION OF SELECTED MANAGEMENT FORECAST DISCLOSURE PRACTICES presented by William Charles Boynton has been accepted towards fulfillment of the requirements for Ph.D. d . Business - Accounting egree 1n Major professor Date g‘q' 7é 0-7639 ABSTRACT AN ANALYSIS OF THE RELIABILITY OF MANAGEMENT EARNINGS FORECASTS PUBLISHED IN ALTERNATIVE FORMATS AND INVESTIGATION OF SELECTED MANAGEMENT FORECAST DISCLOSURE PRACTICES by William.Charles Boynton A diversity of views on the reliability of management fore- casts continues to exist and important disclosure policy issues remain unresolved. As a result, neither the American Institute of Certified Public Accountants, the Financial Accounting Standards Board, nor the Securities and Exchange Commission have taken positions either encour— aging or discouraging the disclosure of forecasts. Yet, in the view of many, forecast disclosures are the most significant financial infor— mation not now regulated. Suggested alternatives to the current status of unregulated voluntary forecast disclosures have included prohibiting or mandating forecast disclosures or regulating voluntary disclosures. It was the purpose of this study to provide empirical data relevant to evaluating these alternative policies. In particular, representing an extension of prior research, data were obtained on.the frequency and reliability of past voluntary disclosures issued in alternative formats. Additionally, exploratory research was carried out to obtain data on factors associated with selected management forecast disclosure prac— tices including the decision to disclose or not disclose a forecast, William Charles Boynton and the format and timing of disclosures made. In this part of the study, emphasis was placed on determining whether comparable treatment has been given to the disclosure of favorable and unfavorable expecta- tions. The data base for the study consisted of selected forecasts of earnings per share (EPS) for an annual period issued during the period 1969 through 1972 by the managements of firms on the Compustat Primary Industrial File. The source of the forecasts was the wall Street Journal. A total of 163 forecasts in point format, 70 in open-range format (minimum estimate stated), and 150 in closed-range format (both ‘minimum.and maximum estimates stated) were included, indicating that significant numbers of forecasts have been issued in each format. It was found that approximately 26 percent of the Primary File firms were represented in the data base. A common shortcoming of the forecast disclosures observed was failure to specify the precise earnings variable being forecasted (i.e., simple, primary, or fully diluted EPS before or after extraordinary items). Based on this finding, it was recommended that, as a minimum, future standards for improving disclosures require that the variable forecasted be described fully and that it be one for which actual results will be published in the financial statements. The reliability of forecasts issued in each of the formats was assessed in terms of bias and objectivity. Bias refers to the conserva- tive or Optimistic character of forecasts and was assessed by computing proportions of over and underpredictions and by computing the means of distributions of relative forecast errors. Objectivity refers to the William Charles Boynton variability of the relative errors associated with forecasts in a given format. Comparisons among proportion, mean, and variance statistics computed on the frequency distributions of relative forecast errors for forecasts in each format were used to test hypotheses about differences in the bias and objectivity of forecasts in the different formats. Such comparisons were also made to test the validity of inferences which might be drawn by users based on the format of a forecast. Based on the results of hypothesis tests, no evidence of either a conservative or optimistic bias was found for point forecasts. But, results indicating that forecasts labeled as minimum estimates in open and closed-range forecasts are not conservatively stated relative to point forecasts, that closed-range forecasts tend to be stated in arbitrarily narrow ranges, and that point forecasts appear to be no more objective than range forecasts suggest that forecasts like some of those studied may be misleading. Because significant proportions of small relative errors were found for forecasts in each format, while at the same time substantial numbers of large relative errors were found, it was recommended that the disclosure of forecasts be neither prohibited nor mandated at the present time. But, the potential for misleading inferences to be drawn by users based on the format of forecasts like those studied was cited as evi- dence supporting recommendations that forecasts be accompanied by probabilistic or other statements about the certainty of the forecasts to facilitate users in determining the degree of reliability to attach to them. The final part of the study dealt with the association between William Charles Boynton selected independent variables and management forecast disclosure prac- tice variables. No significant difference was found in firms' deci- sions to disclose or not disclose forecasts based on the accuracy of the firms' forecasts for the prior year. But firms issuing forecasts in two consecutive years tended to use a range format in the second year if the prior year's forecast was judged inaccurate. No association was found between the horizon of a forecast and its format. Several hypotheses about the association between the favorable versus unfavorable nature of a firm's earnings expectations and disclo- sure practices were tested. No significant associations were found between the direction of change in expected earnings and the decision to disclose or not disclose initial or revised forecasts, or the time of issuance of initial or revised forecasts. These findings suggest that regulations aimed at ensuring comparable treatment of favorable and unfavorable expectations may not be necessary. AN ANALYSIS OF THE RELIABILITY OF MANAGEMENT EARNINGS FORECASTS PUBLISHED IN ALTERNATIVE FORMATS AND INVESTIGATION OF SELECTED MANAGEMENT FORECAST DISCLOSURE PRACTICES by William Charles Boynton A DISSERTATION Submitted to Michigan State University in partial fulfillment of the requirements ’ for the degree of DOCTOR OF PHILOSOPHY Department of Accounting and Financial Administration 1976 ACKNOWLEDGMENTS For their guidance, patience, and encouragement, I wish to express my sincerest appreciation to the members of my dissertation committee, Drs. Naryellen McSweeney, Daniel Collins, and Roland F. Salmonson, Chairman. I also wish to express my gratitude to the Department of Accounting and Financial Administration, Michigan State University, and to the American Institute of Certified Public Accountants for financial assistance during the completion of the thesis. Finally, to my wife, Rosalie, and son, Trevor, for the sacri- fices they have made and the love and encouragement they have provided, I will be forever thankful. ii TABLE OF CONTENTS Page ACKNOWLEDGMENTS. o o o o o o o o o o o o o o o o o o o o o o o o 11 LIST OF TABLES o o o o o o o o o o o o o o o o o o o o o o o o 0 v1 LIST OF FIwRESQ o o o o o o o o o o o o o o o o o o o o o o o o Viii Chapter I. ECONOMIC DECISION MAKING, ACCOUNTING, AND THE DISCLOSURE OF MANAGEMENT FORECASTS . . . . . . . . . . 1 Introduction and Purpose of the Research . . . . . . l The Function of Accounting . . . . . . . . . . . . . 4 Arguments For and Against the Disclosure of Management Forecasts . . . . . . . . . . . . . . 8 Position of the Accounting Profession. . . . . . . . 11 Position of the Securities and Exchange Commission . . . . . . . . . . . . . . . . 15 Summary and Overview . . . . . . . . . . . . . . . . 19 II. REVIEW OF PRIOR EMPIRICAL RESEARCH ON MANAGEMENT FORECASTS . . . . . . . . . . . . . . . . . 23 Major Empirical Research Studies on Manage- ment Forecasts. . . . . . . . . . . . . . . . . . . 24 Summary of Findings on Frequency of Forecast Disclosures . . . . . . . . . . . . . . . . . . . . 42 Summary of Findings on Accuracy of Management Forecasts . . . . . . . . . . . . . . . . . . . . . 43 Summary of Findings on Selected Management Forecast Disclosure Practices . . . . . . . . . . . 46 III. RESEARCH QUESTIONS, DATA COLLECTION PROCEDURES, - AND PROFILE OF THE MANAGEMENT EARNINGS FORECAST DATA USED IN THE STUDY . . . . . . . . . . . . . . . . 47 Research Questions . . . . . . . . . . . . . . . . . 47 Frequency of management earnings forecast disclosures in point, open, and closed- range formats . . . . . . . . . . . . . . . . . 48 Reliability of forecasts published in different formats . . . . . . . . . . . . . . . 48 iii Chapter Variables associated with management forecast disclosure practices . . . . . . Data Collection Procedures . . . . . . . . . . Management forecast disclosure variables . Actual earnings data . . . . . . . . . . . Implications of the Data Collection Procedures Implications of using published management forecast data . . . . . . . . . . . . . . Implications of restricting sample to fore- casts published by firms on Compustat's Primary Industrial File . . . . . . . . . Implications of the study period . . . . . Statistical implications . . . . . . . . . Profile of the Published Management Earnings Forecast Data Used in the Study . . . . . . . Frequency of forecasts by firm . . . . . . Frequency of forecasts by format and year. Frequency of forecasts by horizon and year Mean horizons of forecasts classified by format and year. . . . . . . . . . . . . . Frequency of forecasts by industry grouping and year. . . . . . . . . . . . . s wry O O O O O C O O O O O O O O O O O O O 0 IV. DATA ANALYSIS AND RESEARCH FINDINGS. . . . . . . . Analysis of Reliability . . . . . . . . . . . . Reliability measured as proportion of fore- casts which are right versus wrong . . . . Reliability measured in terms of degree of closeness to being right. . . . . . . . Measurement of relative forecast errors . . . Materiality and the evaluation of reliability Elimination of non-independent observations for purposes of statistical tests. . . . . . Analysis of point forecasts . . . . . . . . Analysis of open-range forecasts. . . . . . Analysis of closed-range forecasts. . . . . Comparative analysis of forecasts published in point, open, and closed-range formats . Summary of analysis of reliability. . . . . Analysis of Association Between Independent Variables and Selected Management Forecast Disclosure Practices . . . . . . . . . . . . . Association between prior forecast accuracy and current disclosure . . . . . . . . . . iv Page 49 50 50 56 56 56 57 58 59 6O 60 61 64 65 66 66 68 68 113 115 Chapter Association between horizon and disclosure format. . . . . . . . . . . . . . . . . . . Association between actual earnings trend and current disclosure. . . . . . . . . . . Associationlbetween forecasted earnings trend and horizon of initial forecasts. . . Association between direction of error in initial forecast and disclosure of revision Association between direction of revision and timing of revision. . . . . . . . . . . Summary of association analyses. . . . . . . V. SUMMARY, CONCLUSIONS, AND RECOMMENDATIONS . . . . . Forecast Disclosure - Developments, and Issues Examined in This Thesis. . . . . . . . . Frequency of Forecasts Issued in Point, Open- range, and Closed-range Formats . . . . . . . . Summary. . . . . . . . . . . . . . . . . . . Conclusions and Recommendations. . . . . . . Analysis of Reliability. . . . . . . . . . . . . smry O O O O O O O I O O I O O O O I O O 0 Conclusions and Recommendations. . . . . . . Analysis of Association Between Independent Variables and Disclosure Practice Variables . . Summary. . . . . . . . . . . . . . . . . . . Conclusions and Recommendations. . . . . . . Recommendations for Further Research . . . . . . BIBL 10M 0 O O O O O O O O O O O O O O O O O O O O O O O Page 119 121 125 127 130 131 134 134 138 138 140 141 141 145 150 150 153 154 156 10. 11. 12. 13. 14. 15. LIST OF TABLES Analysts' and Portfolio Managers' Evaluations of Accuracy of Forecasts . . . . . . . . . . . . . Accuracy of Internal Quarterly and Yearly Forecasts. O O O 0 ~ 0 O O O O O O O O O O O O O I 0 Summary of Prior Empirical Studies on Accuracy of Management Earnings Forecasts . . . . . . . . . Frequency of Firms for Which One, Two, Three, or Four Forecasts are Included in Given Fiscal Years. Frequency of Firms for Which at Least One Forecast is Included in One, Two, Three, or Four Fiscal Years 0 C O O I O O O O O O O C O O O O O O O O O 0 Frequency of Forecasts by Format and Year. . . . . Frequency of Forecasts by Horizon and Year . . . . Mean Horizon in Days by Format and Year. . . . . . Frequency of Forecasts by Industry Grouping and Year 0 O I I O O O O O O O I C O I O O O O O 0 Sample Sizes for Original and Reduced (Independent) Samples. . . . . . . . . . . . . . . Summary of Descriptive Statistics and Hypothesis Tests for Point Forecasts. . . . . . . . . . . . . Summary of Descriptive Statistics and Hypothesis Tests for Openerange Forecasts . . . . . . . . . . Frequency Distribution of Widths of Closed-range Forecasts. . . . . . . . . . . . . . . . . . . . . Summary of Descriptive Statistics and Hypothesis Tests for Closed-range Forecasts . . . . . . . . . Results of Tests of Proportions Using Independent smles O O I O O O O O O O O O O O O O O O O I I 0 vi Page 32 34 44 62 62 63 65 66 67 78 84 90 93 99 103 Table Page 16. Results of Tests of Variances Using Independent smles. O O I O O O O O O O O 0 O O O O O O O O O O O 105 17. Results of Tests of Means of Signed Relative Differences Using Independent Samples. . . . . . . . . 108 18. Results of Tests of Absolute Means Using Independent Samples. . . . . . . . . . . . . . . . . . 109 19. Summary of Descriptive Statistics Based on Independent Samples Used in Comparative Analysis of Point, Open-range, and Closed-range Forecasts . . . 111 20. Results of Tests of Association Between Earnings Trend and Current Disclosure . . . . . . . . . . . . . 124 21. Results of Tests of Association Between Forecasted Earnings Trend and Horizon of Initial Forecasts. . . . 127 '22. Criteria for Classifying Revisions as Downward andupwardooooooooooo'ooooooooooo 128 23. Observed and Expected Frequencies of Downward and Upward Revisions . . . . . . . . . . . . . . . . . 130 24. Summary of Hypothesis Tests of Associations Between Independent Variables and Disclosure Practice Variables . . . . . . . . . . . . . . . . . . 132 vii Figure 1. 9. 10. Objectivity Measured in LIST OF FIGURES and Bias Components of Reliability Terms of Degree of Closeness to Be ins Right 0 O I O O O O O O O I O O O O O O O 0 Frequency Distribution of Relative Errors in Paint Forecasts O O O O O O O O O O O I O O O O 0 Frequency Distribution of Relative Differences in Openrrange Forecasts Measured from Minimum Estimates. . Frequency Distribution of Relative Differences Classified by Interval in Closed-range Forecasts Frequency Distribution of Relative Differences in Closed-range Forecasts Measured from Midpoints. . Frequency Distribution of Relative Differences in Closed-range Forecasts Measured from Minimum Estimates of Ranges. . . . . . . . . . . Contingency Accuracy of of Forecast Contingency Accuracy of Format Used Contingency Table Showing Relationship Between Forecast for Year n-1 and Existence for Year n . . . . . . . . . . . . . Table Showing Relationship Between Forecast for Year n-1 and Forecast inYear n O O O O O O I O O O O O I Table Showing Relationship Between Forecast Status and Trend in Actual EPS. . . . . Contingency Table Showing Relationship Between Forecast Status and Trend in Actual EPS Growth Rate. 0 O O O O O O C I O O O O O O O O 0 viii Page 72 79 86 91 94 97 116 118 123 123 CHAPTER I ECONOMIC DECISION MAKING, ACCOUNTING, AND THE DISCLOSURE OF MANAGEMENT FORECASTS Introduction and Purpose of the Research The challenge to incorporate forecasted information, particu- larly expected income, into the financial reporting framework was 1 The challenge issued to the accounting profession early in the 19608. was kept alive for a decade as the profession searched for ways to develop more relevant financial reporting techniques. Yet aside from the research efforts of a few interested individuals, little was done by the profession to meet the challenge or even to evaluate its merits. By the early 19703, however, it was apparent that forecasts were increasingly being disseminated by corporate managements through various media including interviews, the financial press, and occasion- ally annual reports. In the view of some, forecasts had become unques- tionably the most significant financial information left largely unregu— lated. Concerns arose over the fact that there were no standards or guidelines that the issuer, financial analyst, or the investor could rely on in issuing or interpreting a forecast. Moreover, the Securities and Exchange Commission, recognizing that forecasts had become wide- spread in the securities markets and believing them to be relied upon 1Rudy Schattke, "Expected Income-A Reporting Challenge," Accounting Review 37 (October 1962): 670-76. 1 2 in the investment process, became concerned that all investors did not have equal access to this material information. As a result of this situation, in November 1972, the Securities and Exchange Commission announced public hearings would be held relating to the disclosure, both in filings with the SEC and otherwise, of esti- mates, forecasts, or projections of economic performance by issuers whose securities are publicly traded.1 This action signaled the need for an urgent and comprehensive review of all aspects of the disclosure of forecasted information. Less than a year later, further impetus to renewing and intensifying interest in disclosing forecasted information was provided by the publication of the long awaited report of the AICPA Study Group on the Objectives of Financial Statements. In this docu- ment the Study Group formally and publicly renewed the challenge to the accounting profession to incorporate forecasted information into the financial reporting framework under certain conditions.2 That these events did indeed stimulate further action in this area is apparent from a review of the literature of the ensuing period. Both the SEC and the AICPA have published major documents on forecasting. These documents, which are reviewed briefly in later sections of this chapter, were aimed at improving standards for the preparation and dis- closure of financial forecasts. But the latest publications of both bodies state positions neither encouraging nor discouraging the disclo- sure of forecasts. And examination of the documents and comments on 1Securities and Exchange Commission, "Securities Exchange Act Release No. 9844." 2Study Group on the Objectives of Financial Statements, Objec- tives of Financial Statements (New'York: American Institute of Certified Public Accountants, Inc., 1973). 3 them by others reveals that considerable controversy continues to sur- round the issue of forecast disclosure.1 The central issue at hand is whether the status quo on fore— cast disclosure should be changed, and if so, how. Alternatives to the status quo include prohibiting forecast disclosures, regulating voluntary disclosures, and mandating forecast disclosures. Any evaluation of proposals for change should involve a compari- son of circumstances before and after the change. Unfortunately, at present too little is known even of the nature of.past and present un- regulated voluntary disclosures. For example, questions have been raised regarding the frequency with which forecasts are disseminated through various media, the reliability of the forecasts, whether favor- able and unfavorable expectations are given comparable disclosure treat- ment, and what policies managements follow with respect to revisions. The lack of substantive evidence on which to formulate policy is apparent in the cautious approaches taken by the AICPA and the SEC in their recent publications. Unfortunately, the lack of evidence makes it difficult even to evaluate the wisdom of certain aspects of the pro- posals made therein. Accordingly, it was the purpose of this research to accumulate further evidence which would facilitate the evaluation of current and future proposals for change in the area of forecast disclosure. In particular, the research was designed to bear evidence on several questions concerning forecast disclosure practices prior to the proposed 1For example, see "SEC Disclosure Plan on Profit Forecasts Challenged as Hindrance to Predictions," Wall Street Journal, May 20, 1975, p. 12. 4 regulation thereof. First, data were obtained on the frequency of forecasts published in point, open-range, and closed-range formats. Then the reliability of the forecasts issued in each format was analyzed. Finally, exploratory research on factors associated with management forecast disclosure practices was conducted. Disclosure practice variables examined were the decision to disclose or not dis- close a forecast, and the format and timing of forecast disclosures. Emphasis in this part of the study was on ascertaining the compara- bility of disclosure practices relative to favorable and unfavorable expectations. The remainder of this chapter presents a discussion of the function of accounting, the arguments for and against the disclosure of forecasts, and a more complete review of the positions of the accounting profession and the SEC on forecast disclosure. The final section provides a summary of this chapter and an overview of the organization and structure of the remaining chapters. The Function of Accounting Because of the traditional historical orientation of accounting, some accountants and others believe that forecasted information lies outside the purview of accounting. This section reviews the basis for incorporating the formal disclosure of forecasted information into the financial reporting framework. The Committee to Prepare a Statement of Basic Accounting Theory defined accounting as "the process of identifying, measuring, and come municating economic information to permit informed decisions by users 5 of the information."1 The facilitation of decision making is also embodied in the definition of accounting provided in APB Statement No. 4: Accounting is a service activity. Its function is to provide quantitative information, primarily financial in nature, about economic entities thatzis intended to be useful in making economic decisions. ' But while accounting produces primarily historical data, a great deal of decision making is based on expectational data. Schattke notes: . . . much must necessarily be done in our modern busi- ness world on the basis of expectations. Production must be planned, channels selected, volume of production set and labor hired, all in advance of the sale of product. . . . Thus plans and commitments are made and our economy moves on the basis of expectations, . . .3 Schattke further quotes economist John R. Hicks on the relevance of cer- tain accounting data for decision making:, Income ex post calculations (looking back) are objective; they have their place in economic and statistical history, they are a measuring rod of economic pzogress; but . . . they have no significance for conduct. Accountants have responded to the need for expectational data for internal decision making purposes primarily through the development of comprehensive budget systems. Less tangible response has been made to the need for expectational data for external decision making although 1Committee to Prepare a Statement of Basic Accounting Theory, A Statement of Basic Accounting_Theory (Evanston: American Accounting Association, 1966), p. 1. 2Accounting Principles Board, APB Statement No. 4: Basic Cog: cepts and Accounting Principles UnderlyingpFinancial Statements of Business Enterprises (New York: American Institute of Certified Public Accountants, Inc., 1970), par. 9. 3Schattke, "Expected Income-A Reporting Challenge," p. 670. 4Ibid., p. 671, citing John R. Hicks, Value and Capital, 2d ed. (London: Oxford University Press, 1946), p. 179. 6 the need for such data particularly expected income, is widely recog- nized. For example, Hendriksen notes: . . . most of the decisions of creditors and investors, including the stockholders of large corporations, require a prediction of the future distributions by the firm.1 The Committee for ASOBAT also observed: Almost all external users of financial information reported by a profit-oriented firm are involved in efforts to re- dict the earnings of the firm for some future period. The relevance of expectational data for users' decision models has been emphasized repeatedly in the finance literature.3 More recently, the importance of expectational data was offi- cially recognized by the practicing arm of the accounting profession. APB Statement No. 4 states under "Objectives of Financial Accounting and Financial Statements:" A related general objective is to provide financial infor- mation that assists in estimating the earning potential of the enterprise.4 The issue is emphasized again in the report of the AICPA Study Group on the Objectives of Financial Statements which states: 1Eldon S. Hendriksen, Accounting Theory (Homewood, Illinois: Richard D. Irwin, Inc., 1970), pp. 128-29. ZASOBAT, p. 23. 3For example, for a discussion of the relationship between investment value and earning power, see William.S. Gray III, "Proposal for Systematic Disclosure of Corporate Forecasts," Financial Analysts Journal 29 (January—February 1973):64; and Henry A. Latane and Donald L. Tuttle, Security Analysis and Portfolio Management (New York: The Ronald Press Company, 1970), pp. 277-78 and 385—93. The importance of earning power in credit analysis is discussed in Robert W. Johnson, Financial Management, 4th ed (Boston: Allyn and Bacon, Inc., 1971), p. 316. 1'Accounting Principles Board, par. 79. 7 The basic objective of financial statements is to pro- vide information useful for making economic decisions. . . . All economic decisions look to the future. . . . An objective of financial statements is to provide information useful for the predictive process. Financial forecasts should be provided when they will enhance the reliability of users' predictions. Thus, subject to the ability of expectational data in the form of financial forecasts to enhance the reliability of users' predictions, there appears to be theoretical support in the economics, finance, and accounting literature for incorporating forecasted information into the financial reporting framework. Whether providing external users with managements' forecasts does enhance the reliability of users' predictions has yet to be tested. The conduct of such a test is compli- cated by the necessity to contemplate multiple user decision models, the precise form of which may not be publicly known. But it seems logical to conjecture that to the extent management forecasts are relied upon in the investment process, their ability to enhance the reliability of users' predictions would be directly related to the reliability or accuracy of the management forecasts themselves. Thus the reliability of managements' forecasts is an issue in the decision to incorporate such forecasts into the financial reporting framework. Accordingly, a summary of the previous research of others into the reliability of management forecasts is reported in Chapter 2 of this study, and an extension of this research constituted a major part of this study. But a decision to extend the formal financial reporting frame- work to include financial forecasts involves consideration of factors beyond the theoretical basis therefor and the reliability of forecasts. 1Objectives of Financial Statements, pp. 61-65 passim. 8 Some of these factors are discussed in the following section. ‘Agguments For and'Against the Disclosure of Management Forecasts The principal argument in favor of the disclosure of management forecasts is their relevance for economic decision making. As implied in the quotation above from Objectives of Financial Statements, this relevance does not result from the forecasts as ends in themselves, but rather from their use as means to the enhancement of users' own forecasts of a company's financial prospects. Opponents of disclosing management forecasts contend that users have other sources of information from which to formulate their own forecasts, namely historical accounting data and forecasts prepared by investment analysts. But, the Accountants International Study Group had this to say about the usefulness of these sources relative to the disclosure of management forecasts: While these sources provide some information, the directors [management] may be expected to possess more knowledge of the internal workings of their company and at least a com- parable understanding of the factors external to the company. It is therefore probable that the best source of a forecast about a company lies within the company itself. Using the same argument, the AICPA's Management Advisory Services Execu- tive Committee concluded: The management of a company, through the use of its fore- casting system, is in the best position to determine the single most probable forecasted financial result; . . . IAccountants International Study Group, Published Profit Fore- casts (n.p.: Accountants International Study Group, 1974), par. 16. 2Management Advisory Services Executive Committee, Guidelines for Systems for the Preparation of Financial Forecasts, Management Advisory Services Guideline Series Number 3 (New York: American Institute of Certified Public Accountants, Inc., 1975), p. 8. 9 A considerable amount of empirical data has been collected on the accuracy of forecasts available from alternative sources.1 The Basi,et. a1.,study found management forecasts to be slightly more accu- rate than analysts' forecasts. On the other hand, studies by Green and Segall, Copeland and Marioni, and Lorek, et. a1.. produced conflicting evidence regarding the relative superiority of management forecasts versus forecasts of time-series models based on historical accounting data. Further details of these and other studies related to management fore- casts are presented in Chapter 2 of this thesis. A second major argument favoring the prompt and routine dis— closing of management forecasts through the financial reporting frame- work is that forecasted information would thereby be made available equitably to all interested parties. It is possible that in the absence of the formal public reporting of forecasts, selected persons or groups 1For examples of studies examining the predictive ability of historical accounting data, see Philip Brown and Victor Niederhoffer, "The Predictive Content of Quarterly Earnings," Journal of Business 41 (October 1968):488-97, and Werner Frank, "A Study of the Predictive Significance of Two Income Measures," Journal of Accounting Research 7 (Spring 1969):123—36. Other researchers have attempted to evaluate the relative superiority of managements' or analysts' forecasts versus forecasts of naive models. For example, see David Green, Jr., and Joel Segall, "The Predictive Power of First-Quarter Earnings Reports," Journal of Business 40 (January l967):44—55, and "The Predictive Power of First-Quarter Earnings Reports: A Replication," Journal of Accounting Research 4 (suppl. l966):21-36; R. M. Copeland and R. J. Marioni, "Executives Forecasts of Earnings Per Share Versus Forecasts of Naive Models," Journal of Buginess 45 (October l972):497-512; Edwin J. Elton and Martin J. Gruber, "Earnings Estimates and the Accuracy of Expec- tational Data," Management Science 18 (April 1972):409-24; and Kenneth S. Lorek, Charles L. McDonald, and Dennis H. Patz, "A Comparative Analysis of Management Forecasts and Box-Jenkins Forecasts of Earnings," Accounting Review 51 (April 1976):321-30. For a comparative study of the accuracy of corporate and security analysts' forecasts, see Bart A. Basi, Kenneth J. Carey, and Richard D. Twark, "A Comparison of the Accuracy of Corporate and Security Analysts' Forecasts of Earnings," Accounting;Review 51 (April 1976):244-54. 10 may have attained an unfair advantage by gaining private access to fore- casted information. For example, many corporate managements have long maintained a practice of revealing corporate expectations at meetings held for investment analysts and institutional investors. This infor- mation may ultimately have reached a wider audience through news stories in the financial press, but often only after some delay if at all. Numerous arguments have been presented against the disclosure of forecasts. Corporate managements have expressed concerns about resulting damages sustained in relation to competitors and about possible loss of credibility if forecasts are not achieved. Accountants have ex- pressed concern that if forecasts are incorporated into the financial reporting framework and not achieved, the credibility of all financial reporting will be diminished. Accountants and others have also expressed concern about the lack of standards for the preparation and dissemination of forecasts. Both managements and accountants have been very much con- cerned about the legal liability associated with forecast disclosures. Other questions have been raised by various parties about the problem of forecasts rapidly becoming outdated and the possibility that manage- ment would disclose forecasts only when it would be advantageous. Con- cern has also been expressed that once short-range forecasts were issued managements might make decisions aimed at achieving those forecasts to the detriment of attaining long-term objectives. Finally, formal reporting, especially if it involved certification, might impair the timeliness and therefore the usefulness of forecasts. Most of these arguments against disclosing forecasts can be overcome or mitigated to varying degrees by establishing proper standards and exerting judicious regulatory authority. The attempts ll of the accounting profession and the Securities and Exchange Commission to deal with some of these arguments will be discussed in the following two sections. Position of the Accounting Profession The position of the accounting profession on forecasting at the time of the previously mentioned SEC sponsored public hearings in late 1972 was expressed by representatives from the AICPA.and the NAA in testimony given at those hearings. Wellace Olson, speaking for the AICPA, noted that the senior technical committees of the Institute had not reached definite conclusions on the subject of forecasts and stated: we believe that after establishing suitable guidelines, the Commission should permit publication of forecasts for a trial period during which time it could encourage companies to disclose forecasts. This should provide the experience necessary to form a sound basis for reaching a decision as to whether prohibition or permissive or mandatory publica- tion would best serve the public interest in the long run. The NAA Committee on Management Accounting Practices - Subcommittee on Forecasts testified: . . . the publication of foreward estimates of material aspects of the company with statements of the basic under- lying assumptions is highly desirable, but at the discretion of management. However, before a mandatory requirement should be imposed, significantly more study and research work needs to be done. Thus, the AICPA and the NAA advocated pursuing the study of the viability of disclosing forecasts, withholding final conclusions pending the results of further study. 1"Trial Period Suggested for Publication of Forecasts (News Report)," Journal of Accountanny 135 (January 1973):10. 2"NAATestifies on Forecasts," MaQESPment Accounting 54 (February l973):53—54. 12 Testimony by others at the public hearings influenced the accounting profession's further development of a position on forecasting. Most influential was the testimony from corporate executives, financial analysts, lawyers, and academicians which revealed widespread dissatis- faction with the lack of guidelines or standards for the preparation and dissemination of forecasts. The accounting profession's response to this testimony has been primarily through the work of three divisions of the AICPA - the Management Advisory Services Division, the Accounting Standards Division, and the Auditing Standards Division. The Financial Accounting Standards Board has not as yet involved itself in the matter of forecast disclosure. Responding to the need for guidelines or standards for the preparation of forecasts, early in 1975 the AICPA's Management Advisory Services Division published Guidelines for Systems for the Preparation of Financial Forecasts. Significant in relation to the research ques- tions studied in this thesis is the following conclusion quoted from Guideline No. 1: Because forecasts are not exact and are subject to varying .degrees of inaccuracy, preparing a forecast in a manner that conveys the degree of uncertainty associated with it is very useful and should be encouraged. This guideline is intended to encourage the development of ranges, probabilistic statements, or estimates of error as supplements to the single most prob- able forecasted result. Such information is useful to der- score the essentially uncertain nature of all forecasts. Among other topics covered in the guidelines are accounting principles to be used, sources of information relevant to a forecast, identifica- tion of assumptions, and documentation and review of forecasts. 1Guidelines for Systems for the Prennration of Financial Forecasts, p. 8. 13 The position of the AICPA as of early 1975 on the preparation of forecasts, and the motivation for the "Guidelines" document, are summarized in the following quotation: The publication of financial forecasts is neither advocated nor discouraged. This document has been prepared because financial forecasts are being disseminated and accordingly, there is a need for authoritative guidelines for their preparation.1 The need for guidelines or standards for the dissemination of forecasts was addressed by the Accounting Standards Division of the AICPA. In 1975, it issued a document titled "Statement of Position on Presentation and Disclosure of Financial Forecasts." The recommendations in this document parallel the guidelines in the MAS document for systems for the preparation of forecasts. For example, regarding format for forecast presentation and dissemination, the Statement of Position states: Financial forecasts should be expressed in specific mone- tary amounts representing the single most probable forecasted result. The tentative nature of a financial forecast would be emphasized if the single most probable result for key mea- sures (e.g., sales and net income) was supplemented by ranges or probabilistic statements, and the presentation of such is encouraged. While a range informs the user of the probabilistic nature of the forecast, expressing a financial forecast solely in terms of ranges could result in the user's attributing an un- warranted degree of reliability to the forecast ranges, because many users might assume (a) that a range represented the spread between the best possible result and the worst possible result, or (b) that the range was based on a scientifically determined interval. Management should be in the best position to deter- mine the single most probable result and this burden should not be placed on outsiders. Also, single point estimates are neces- sary to aggregate the forecasts of an enterprise's individual operations, as well as to facilitate comparison between the forecast and later historical results. 11bid., p. 2. 2Accounting Standards Executive Committee, "Statement of Posi- tion on the Presentation and Disclosure of Financial Forecasts," (New York: American Institute of Certified Public Accountants, Inc., 1975), p. 4. 14 Other significant recommendations pertain to the disclosure of assump- tions deemed necessary for forecasts to be understood and properly evaluated, and disclosure of an issuer's intentions regarding the updating of forecasts. The position of the AICPA as of 1975 regarding the dissemination of forecasts is summarized in the following quotation: This Statement provides guidance as to presentation and disclosure for those who choose to issue information about the future described as financial forecasts. Nothing herein should be interpreted to mean that the publication of finan- cial forecasts is recommended or that a financial forecas is deemed to be a part of the basic financial statements. Finally, concurrent with the efforts of the MAS and Accounting Standards Divisions of the Institute, the Auditing Standards Division has been studying matters related to the CPA's involvement with finan- cial forecasts. It has been argued that the credibility and utility of forecasted information would be diminished if it is not subjected to independent review. Consequently, the Auditing Standards Division is studying the development of auditing and reporting standards for the review and attestation of forecasts. Pending the publication of such standards, the current position of the profession on the CPA's involve- ment with the dissemination of forecasts is expressed in the AICPA's Code of Professional Ethics. Rule 204 of the Code prohibits a member from permitting his name to be used in conjunction with any forecast in a manner which may lead to the belief that the member vouches for the achievability of the forecast.2 1Ibid., p. 3. 2American Institute of Certified Public Accountants, Code of Professional Ethics (New York: American Institute of Certified Public Accountants, Inc., 1972), p. 22. 15 Thus, the accounting profession has been actively involved in formulating positions on the preparation, dissemination, and independent review and attestation of forecasts. Concurrently, the SEC has been formulating its own position on forecasting. Position of the Securities and Exchange Commission As noted previously, in late 1972 the SEC ordered public hearings for the purpose of gathering information relevant to a reassessment of the Commission's policies relating to the disclosure of forecasts1 of economic performance. In February 1973, the Commission released a state- ment which included the following general conclusions: Information gathered at the hearings reinforced the Come mission's own observation that management's assessment of a company's future performance is information of significant importance to the investor, that such assessment should be able to be understood in light of the assumptions made, and that such information should be available, if at all, on an equitable basis to all investors.2 Consistent with these conclusions, the Commission announced plans to abandon its long standing policy of generally prohibiting the disclo- sure of forecasts in SEC filings. This policy change was to be imple- mented through the future issuance of forecast disclosure guidelines and changes in the securities laws. The SEC released the first set of proposed implementing guidelines 1The SEC uses the term "projection" throughout its releases to refer to estimates of most probable results. But the term "forecast" is more commonly used in this context in business literature in general and specifically in AICPA publications. Due to its wider acceptance, the term forecast is used throughout this thesis. 2"Statement by the Commission on the Disclosure of Projections of Future Economic Performance," Securities Act of 1933: Release No. 5362/February 2, 1973, and Securities Exchange Act of 1934: Release No. 9984/February 2, 1973, reprinted in SEC Docket 1 (February 13, 1973):4-5. 16 in April 1975.1 That release included proposals which would have imposed a complex reporting system under the federal securities laws to be followed whenever a registrant publicly disclosed a forecast. Under the proposals, registrants could voluntarily make initial forecast dis- closures in registration statements or on Forms lO-K or 8-K. But firms disclosing forecasts through any media other than SEC filings would have become subject to mandatory SEC reporting requirements. Specifically, the details and circumstances of all such forecasts were to be reported on Form 8-K. Additionally, comparisons of such forecasts with actual results were to be provided in subsequent registration statements and 10—K reports. All forecast information contained in an issuer's lo-K report was to have been included in the issuer's annual report to secu- rity holders. Finally, continuation of regular public forecasting would have been required or notification provided to the SEC of the reasons for no longer making public forecasts. In spite of the inclusion of so-called "safe-harbor" rules which were intended to limit legal liability for inaccurate forecasts, strong apposition to the proposals was expressed by executives, accountants, and lawyers.2 Moreover, there were indications that rather than comply with the complex forecast reporting framework proposed, many firms would cease disclosing forecasts altogether.3 1"Notice of Proposed Rule[s]. . . and Proposed Amendments . . . to Implement the 'Statement by the Commission on the Disclosure of Projections of Future Economic Performance' . . .," Securities Act of 1933: Release No. 5581/April 28, 1975, and Securities Exchange Act of 1934: Release No. ll374/April 28, 1975, reprinted in the SEC Docket 6 (May 13, 1975):746-61. 2See "SEC Disclosure Plan on Profit Forecasts Challenged as Hindrance to Predictions," Wall Street Journal, May 20, 1975, p. 12. 3Ibid. 17 In the face of this opposition, the Commission issued a new release in April 1976 which included the following statement: Due to the important legal, disclosure policy and tech- nical issues raised by the commentators with respect to the [1975] projection prOposals . . . the Commission has deter- mined that all of these proposals should be withdrawn, except for the amendment to Rule 14a-9 which is adopted as proposed. However, the Commission is also of the view that the question of inclusion of projections in Commission filings is an important one which should be addressed at this time. The extensive public record in this matter, supplemented by [the] staff's experience in processing filings that have included projections, even though limited, provides adequate bases for the publication for public com- ment of a new approach to this question. The amendment to Rule 14a-9 implements the Commission's original prOposal to cease prohibiting the inclusion of forecasts in SEC filings. The proposed new approach is to make disclosure of forecasts in Commis- sion filings entirely voluntary subject only to general disclosure guides. Pertinent to the study of past forecast disclosure practices in this research are the following excerpts from the prOposed guides regarding the format of forecast disclosures: . . . Traditionally, projections have been given for three financial items generally considered to be of primary importance to investors: revenues, net income and earnings per share. These three items usually are presented together in order to avoid any misleading inferences that may arise when the indi- vidual items reflect contradictory trends. There may be instances, however, when it is appropriate to present earnings from continuing operations, or income before extraordinary items in addition to or in lieu of net income. . . . 1"Notice of Adaption of an Amendment to Rule 14a-9 Under the 1934 Act and Withdrawal of the Other Proposals Contained in Release No. 33-5581 . . . and Notice of Publication for Comment of Proposed Guide 62 and 4, "Disclosure of Projections of Future Economic Performr ance" . . .," Securities Act of 1933: Release No. 5699/April 23, 1976, and Securities Exchange Act of 1934: Release No. 12371/April 23, 1976, reprinted in SEC Docket 9 (May 11, l976):472-75. 18 . . . management must disclose what in its opinion is the most probable specific amount or the most reasonable range for each financial item projected. Ranges should not, however, be so wide as to make the disclosures meaningless. Moreover, several projections based on varying assumptions may be judged by management to be more meaningful than a single number or range and would be permitted. While not specifying that forecasts be accompanied by probability state- ments, the proposed guides would require the following additional disclo— sures: "Investors should be cautioned against attributing undue certainty to management's assessment and should be informed of management's inten— tions with respect to furnishing updated projections."2 Also pertinent to this research is the following statement from the April 1976 release: . . . the Commission wishes to remind issuers of their respon- sibility to make full and prompt disclosure of material facts, both favorable and unfavorable, regarding their financial con— dition, and that this responsibility may extend to situations where management knows its previously disclosed assessments no longer have a reasonable basis.3 Bearing on this concern, this research includes an analysis of the come parability of past disclosures of favorable and unfavorable expectations with respect to the issuance of both initial and revised forecasts. The release does not include safe-harbor rules. But the Come mission included the statement that it "is of the view that reasonably based and adequately presented projections should not subject issuers to liability under the federal securities laws, even if the projections lIbid., p. 475. 2Ibid. The guides would also require the disclosure of key assumptions, and suggest that management give consideration to dis- closing the accuracy of its previous forecasts. 31bid., p. 474. l9 "1 prove to be in error. Thus, the latest SEC release appears to be aimed at setting broad standards for the voluntary disclosure of forecasts in Commission filings without imposing a reporting system so complex as to discourage firms from making any such disclosures. The Commission summarized its position as follows: It should be noted . . . that the Commission is neither encouraging or discouraging the making and filing of pro- jections because of the diversity of views on the importance and reliability of projections. This issue, along with the question of the need for a safe-harbor rule for projections, may be among those appropriately considered by the Advisory Committee on Corporate Disclosure. In the interim, however, the Commission believes that it should not stand in the way of companies choosing to project in filings, subject to the general disclosure guidelines contained in the . . . [proposed] guide[s]. . .2 Summarygand Overview This chapter has presented the theoretical foundation for incor- porating management forecast information into the formal financial reporting framework, other arguments for and against the disclosure of such forecasts, and the evolution of the positions of the accounting profession and the SEC on forecasting. Regarding the latter, it was observed that while both the AICPA and the SEC have been actively involved in the development of standards and guidelines for forecast preparation and disclosure, in both cases the actions taken to date have been primarily in recognition of the fact that forecasts have been and are being prepared and disseminated. Both bodies neither encourage llbid., p. 473. 2Ibid. 20 nor discourage the dissemination of forecasts. This situation can be attributed in part at least to failure to obtain sufficient evidence as a basis upon which to formulate firmer convictions. The current AICPA and SEC positions appear to represent a moder- ate approach toward establishing standards for forecast disclosure. But, compliance with the AICPA standards for the preparation and dis- semination of forecasts is voluntary except for the involvement of Institute members. Thus, the impact that the Institute's efforts will have on actual disclosures is uncertain. Moreover, the SEC's guides, if adapted, while being enforceable with respect to forecasts disclosed in SEC filings, will have an unknown effect on the frequency and nature of disclosures outside SEC filings. Unfortunately, at present it is difficult to evaluate the wisdom of the current approaches of the AICPA and the SEC. Too little is known of the quality of past and present forecast disclosure practices, and it is difficult to estimate the impact on future voluntary disclosures of the standards and guidelines being developed. Conceptually, the volume tary framework being developed should be evaluated based on a comparison of the quantity and quality of forecast disclosures under the new stane dards and guidelines with the quantity and quality of such disclosures in the past. Appropriate consideration must also be given to the differ- ential costs involved. Both the AICPA and the SEC have acknowledged a need for futher experience with and analysis of forecasts as a basis for further action. A study of forecasts issued during a trial period under the new guide- lines is one approach to gaining such information. While such a study will be of value, it does have a number of limitations. First, 21 substantial delay will be entailed in obtaining sufficient experience to permit meaningful analysis. Second, depending on the degree of success in clarifying some of the ambiguities in the recent pronounce- ments, the number of firms disclosing forecasts may decrease substan- tially. Finally, this approach.will provide no further information about past forecast disclosure practices, information that is needed if an appropriate appraisal of change-producing actions is to be made. Information collected to date on past forecast disclosure prac- tices is incomplete and, on some points, contradictory. Accordingly, this researcher feels that the accounting profession should inquire further into past experience with disclosure of forecasted information as a potentially valuable source of additional evidence bearing on some of the major issues surrounding forecasting. For the purposes of this thesis, it was not practical to attempt to study all forms of past management forecast data. Therefore, the analysis was restricted to forecasts of earnings per share, a form of disclosure which has been both prevalent and relevant to user decision models. Nbr was it practical to attempt to study all the major issues surrounding forecasts. Therefore, emphasis was placed on obtaining additional data on three aspects of past forecast disclosures. First, data on the frequency of past public forecast disclosures in point, open-range, and closed-range formats was obtained. The data collection procedures used and a profile of the data base obtained are presented in Chapter 3. Second, in view of the continued diversity of views about the reliability of forecasts, measurements of the reliability of forecasts issued in each of the three formats were made. Specifics of the 22 methodology for measuring reliability and the research findings on reliability are presented in the first part of Chapter 4. The final aspect of forecast disclosures examined in this research was the analysis of factors associated with the decision to disclose or not disclose a forecast, and the format and timing of fore- casts issued. Specifics of the methodology for this analysis of dis- closure practices, and the research findings, are presented in the second part of Chapter 4. Chapter 2 of this study presents a review of the literature con- cerning prior empirical research on management earnings forecasts. As indicated above, Chapter 3 contains a description of the data collection procedures and a profile of the data base used in the study, and Chapter 4 contains the research methodology and findings. Finally, Chapter 5 contains a summary of the results of the study and the conclusions and recommendations based thereon. CHAPTER II REVIEW OF PRIOR EMPIRICAL RESEARCH ON MANAGEMENT FORECASTS It was asserted in Chapter 1 that the accounting profession and the SEC lack sufficient evidence upon which to formulate and evaluate policy decisions on incorporating management forecasts into the formal financial reporting framework. Nonetheless a considerable amount of related empirical research has been published. Among the forecasting issues investigated in prior empirical research are the frequency of internal and publicly disclosed forecasts, forecast accuracy, factors associated with accuracy, beliefs by outsiders about management earnings forecast disclosure practices, stock market reaction to earnings fore- casts, and management behavioral implications of forecast disclosure. This chapter provides a review of the most significant aspects of this body of literature. Emphasis is placed on review of the prior research on issues further investigated in this thesis. The chapter is organized as follows. First, the major studies are reviewed individually. Following the individual reviews, compara- tive summaries of prior findings on frequency of forecast disclosures, accuracy, and selected disclosure practices are presented. Shortcomings of the prior research are identified as a basis for formulating the specific research questions investigated in this study. 23 24 Major Empirical Research Studies on Management Forecasts Green and Segall studies.1 In 1966 and 1967, the results of a pair of studies by these researchers dealing primarily with the predic— tive power of first-quarter earnings reports were published. The objec- tive of these studies was to determine the forecasting value of interim reports. The methodology involved the development and testing of a num- ber of "naive models" which were used to extrapolate forecasts from his- torical data (e.g., multiplying the first-quarter's earnings per share (EPS) by four to arrive at a forecast of annual EPS). In addition to comparing the relative predictive accuracy of annual-based versus interim- based time-series models, the researchers located and studied a limited number of actual earnings forecasts publicly disclosed by executives of the companies in their original sample. The management forecasts were investigated to provide a benchmark for the evaluation of the various time-series models. The original study involved the analysis of naive forecasts of annual earnings for 1964 for 46 companies listed on the New York Stock Exchange (NYSE) and 12 actual (management) forecasts for those companies found in the Wall Street Journal Index for 1963 and 1964. The replica- tion involved the analysis of naive forecasts of annual earnings for 1965 for 43 of the companies in the original sample plus a new sample _ of 44 additional NYSE firms, and 15 1965 management earnings forecasts found in the wall Street Journal Index for 1964 and 1965 for companies 1David Green, Jr., and Joel Segall, "The Predictive Power of First-Quarter Earnings Reports," Journal of Business 40 (January 1967): 44-55; and "The Predictive Power of First-Quarter Earnings Reports: A Replication," Journal of Accounting;Research 4 (Suppl. l966):21—36. 25 in the new sample. Interestingly, on the basis of both the original study and the replication, the researchers concluded that first-quarter reports, as then prepared, were of little help in forecasting annual EPS. Further, the researchers were "not impressed" with the management forecasts, and concluded that the naive forecasts were "not inferior" to the presumably more sophisticated management forecasts. Because the Green and Segall total sample of 27 published fore- casts located in the wall Street Journal Index was comprised of all forms of verbal and quantitative forecasts, only 11 of which were specific, the researchers could not compute an overall measure of accu- racy. About the most that can be said beyond the previously stated con- clusion about the relative accuracy of the naive forecasts versus the management forecasts is that in 18 out of the 27 cases the management forecasts indicated the correct direction of change in net income from the prior year. Copeland and Marioni study.1 A number of readers found Green and Segall's conclusions incredible. Copeland and Marioni decided to replicate the part of Green and Segall's study involving the comparison of management forecasts with those produced with naive models using later data. The researchers studied 50 management earnings forecasts published in the wall Street Journal in 1968 and 25 published in each of the years 1964 and 1965. The sample was obtained by scanning issues of the Wall Street Journal until the desired numbers of forecasts were 1R. M. Copeland and R. J. Marioni, "Executives' Forecasts of Earnings per Share versus Forecasts of Naive Models," Journal of Business 45 (October l972)=497-512. 26 located which were in specific, numerical point or range format and for which the required quarterly EPS data were available. In cases where the forecasts stated that EPS would "approximate," "exceed," or be "at least" a specific figure, the amount declared was used as the forecast. An arithmetic mean was computed and used as the forecast ‘ for cases where a closed-range estimate had been published. The accu- racy of the forecasts was then calculated in terms of both absolute dollar and relative (percentage) errors, the latter computed using the formula Forecast - Actual Actual Of the fifty 1968 management forecasts which had an average horizon of 7.5 months, the researchers found that forecasts which turned out to be overestimates outnumbered the underestimates by only two fore- casts, with two forecasts being precise. The average relative error, including sign, was found to be +15.8 percent and ignoring the sign, +20.1 percent. On this basis the researchers concluded that firms which overestimated earnings did so with a much higher degree of inaccuracy than those that underestimated earnings. It should be noted, however, that the "average" statistics are heavily biased by the inclusion of just three extreme values, all of which were in excess of +100 percent. Excluding the three extreme values, the average including sign would have been +5.58 percent, and ignoring sign, +10.13 percent. For the samples of 25 published forecasts found in each of the years 1964 and 1965, only the absolute errors were calculated in order to rank the accuracy of the management forecasts with those produced from naive models. The size of the errors was not reported. Contrary 27 to Green and Segall's conclusion, Copeland and Marioni concluded that the management forecasts were "substantially better" than those produced from the naive models in each of the three years studied. A Daily study.1 Daily conducted a limited empirical investiga- tion of forecasting accuracy by obtaining earnings and sales forecasts directly from cooperating companies. While more than 50 firms were requested to participate, only 12 firms cooperated, providing 66 earnings (net income) and 65 sales forecasts constituting from three to seven years of data for individual firms. An assumption made by Daily was that these management-oriented (internal) forecasts would represent a reasonable surrogate for the type of forecast that might be publicly reported. Daily defined accuracy as follows: Accuracy 3 Actual Results x 100 Forecasted Amount Based on this measure, be determined that of the 65 revenue and 66 net income forecasts made by the 12 firms providing data, 90 percent of the revenue forecasts and 47 percent of the net income forecasts fell within plus or minus 10 percent of actual results. Differences exceeding 15 percent between forecasted and actual net income were present in one- third of the observations. Based on the analysis of the net income data, he concluded "a reasonable doubt should exist regarding the ability of firms to forecast operating results with the degree of accuracy . . . necessary to satisfy the requirements of investors;' and based on the analysis of the revenue data he concluded that "if forecasts would be 1R. Austin Daily, "The Feasibility of Reporting Forecasted Information," Accounting Review 46 (October 197l):686-92. 28 deemed relevant information to investors, it may be possible at the present time to report such information."1 Daily also attempted to identify factors associated with fore- cast accuracy. Independent variables examined were size of firm as represented by (1) annual net income and (2) annual revenues, and accuracy of a firm's forecast of revenue. The three variables together resulted in a coefficient of determination of only .194, indicating there was no strong association between the variables examined and the accuracy of earnings forecasts. Accuracy was found to vary across industry classification with banks forecasting all categories of oper- ations more accurately than any other group of firms represented in the data. McDonald study.2 Like Daily, McDonald investigated the accuracy of management forecasts and the association between several variables and the occurrence of forecast errors. However, McDonald studied pub- lished forecasts of net earnings per share in point format found in the January through April issues of the Wall Street Journal for the five year period 1966 through 1970. Only forecasts issued within the first 120 days of the fiscal year forecasted were included in his sample. The sample contained 201 EPS forecasts representing 152 firms for which one forecast was located, 23 firms for which two forecasts were located, and one firm for which three forecasts were located. McDonald measured the accuracy of published management forecasts of net earnings per share using the following calculation: llbid., p. 692. 2Charles LeRoy McDonald, "An Empirical Examination of Published Predictions of Future Earnings" (Ph.D. dissertation, Michigan State University, 1972). 29 Relative Pre- = Actual Earnings - Predicted Earningg diction Errors Predicted Earnings His analysis of 201 forecasts published in point format and with minimum horizons of 245 days revealed a tendency toward overprediction, 63.7 percent of the observations being overpredictions, 33.8 percent under- predictions, and 2.5 percent exact predictions. The relative prediction errors ranged from -395.6 percent to +108.5 percent with the mean rela- tive error for the five-year period covered by the study being -13.6 percent. The removal of four extreme overpredictions, arbitrarily defined as observations lying outside two standard deviations from the mean, reduced the average relative prediction error to -lO.2 percent. McDonald reported that 35.3 percent of the forecasts studied fell within five percent of actual earnings and 48.8 percent within ten percent of actual earnings, while 39.8 percent were more than 15 percent from actual earnings. On this basis he concluded: ". . . some of the pre- dictions £332 to be reliable enough to be useful,"1 and ". . ., pub- lished annual financial statements should include predictions of earnings for the forthcoming year."2 MeDonald found that of the general industry groupings in his data, utilities were the best predictors. Using correlation analysis he also investigated the association between prediction errors and the following variables: (1) change in aggregate corporate profits, (2) change in industry profits, (3) fluctuation in past operating earnings, (4) relative extraordinary gains and losses, and (5) size of firm. 11bid., p. 110. 21bid., p. 113. 30 None of the variables proved to be significant at the .05 level. McDonald also used multiple regression analysis to examine factors associated with prediction errors. After removing four extreme values from his sample, he found that four independent variables remain- ing in his regression model explained 63.34 percent of the variation of prediction errors. Those variables were fluctuation in past operating earnings, relative extraordinary gains and losses, size of firm, and change in operating earnings. Most of the explained variation was attributed to the variable representing extraordinary items. Two endogenous variables, change in industry profits and change in aggre- gate corporate profits, were not significant at the .05 level and were deleted from the model. Financial Analysts' Federation research project on corporate forecasts.1 During 1972, the Financial Analysts Federation sponsored a research project to examine the desirability and possible content of a formal system.of forecast disclosure. In addition to a comprehensive search of the published literature relating to corporate forecasting, the project included a selective survey of corporate forecasts published in the Wall Street Journal between October 1971 and September 1972. Based on the examination of eighty-nine forecasts classified by firm size and industry grouping, it was concluded that forecasting is "pervasive" and that there is "no firm size or industry group for which forecasting is impossible."2 The nature of the forecasts examined was 1For a summary of the findings of the project, see Samuel S. Stewart, "Research Report on Corporate Forecasts," Financial Analysts Journal 29 (January-February 1973):77-85. 2Ibid., p. 79. 31 not further described. The accuracy of the management forecasts apparently was not examined. Instead, the project included a study of the accuracy of financial analysts' forecasts on the basis that "it might be assumed that the accuracy of analysts' forecasts is closely related to the accuracy of management forecasts."l Based on a limited examination of the forecasting records of only a few large institutional investors, the following "tentative" conclusions were stated: (a) there seemed to be a slightly Optimistic bias to most forecasts; (b) the relative accuracy of analysts' forecasts is often not much better than the accu- racy of forecasts based on simple, extrapolative models; however, analysts are consistently superior to models at turning points and in difficult-to-forecast industries; (c) the shorter the forecasting hori- zon, the more accurate the forecast; and (d) the accuracy of forecasting is strongly influenced by the nature of the industry. Finally, the FAF research project included a survey to obtain data about FAF members' (professional analysts' and portfolio managers') experiences with forecasting, including their impressions as to the current extent and accuracy of forecasting. Regarding extent, more than 40 percent of the respondents indicated that they receive some type of forecast from.more than half of the companies they follow. Data on the respondents' impressions about accuracy are shown in Table 1. Interestingly, portfolio managers rated management forecasts slightly more accurate than did the analysts, and the portfolio managers cone sidered management forecasts to be slightly more accurate than analysts' forecasts. 11bid. 32 TABLE 1 ANALYSTS' AND PORTFOLIO MANAGERS' EVALUATIONS 0F ACCURACY OF FORECASTS Ratingfiby All Portfolio Accuracy Analysts Managers One Year Management Forecasts Perfect 02 0% £102 382 502 1202 532 422 Warse 92 82 One Year Analysts' Forecasts Perfect - 02 £102 - 412 1202 - 522 worse - . 72 SOURCE: Samuel S. Stewart, "Research Report on Corpo- rate Forecasts," Financial Analysts Journal 29 (January-February 1973):82. Among other responses, the survey revealed that most responr dents felt that there was a wide gap in the availability of forecasts to professional versus other investors. FAF members also viewed man— agement forecasts as significant information that plays an important role in investment decision making.1 Financial Executives Research Foundation study.2 This was a two-part study to examine (l) the many questions relating to manage- ment's use of internal forecasts and (2) to determine management's attitude toward public disclosure of business forecasts. Questionnaire 11bid., p. 83. 2See, "How'Accurate Are Forecasts?" Financial Executive 41 (March 1973):26-32. 33 responses were received from 338 companies. Questions dealt with the extent to which managements prepared forecasts for internal use, the accuracy of those forecasts, factors associated with accuracy, manage- ments' opinions about the job the financial community is doing in fore- casting corporate earnings per share, and communications between man- agement and financial analysts. Regarding the extent to which corporate managements prepare forecasts, 95 percent of the 338 companies responding to the question- naire reportedly prepared internal forecasts of corporate sales, ex- penses, and earnings. The findings of the study regarding the accu- racy of both quarterly and annual internal forecasts of various finan- cial variables are summarized in Table 2. It is not clear from the report whether forecasted or actual results were used in the denominator of the percent variance calculations. The results of the quarterly variance analysis were summarized as follows: Corporate expense is the most reliable forecast, with 97 percent.of.the responding companies having a variance less than plus-or-minus 10 percent. Corporate sales are easily predicted, as 94 percent of the companies had a variance of plus-or-minus 10 percent. Expenses by division and by corporate chart of accounts were also highly reliable with 93 percent and 92 percent of the companies having a variance of less than 10 percent. Earnings by division expe— rienced the greatest variance from the expected results: almost one quarter of the companies responding indicated that their results typically differ from the forecasts by more than 10 percent.1 Regarding the analysis of the yearly variances it was stated: The fact that the corporate and division earnings fore- cast ranked eighth and ninth in terms of accuracy are (sic) 1"How Accurate Are Forecasts?", p. 27. ACCURACY OF INTERNAL QUARTERLY 34 TABLE 2 AND YEARLY FORECASTS Cumulative Percent Variance2 Rank Type of Forecast 52 102 152 202 21+2 QUARTERLY VARIANCE 1 Corporate expenses 802 972 992 1002 02 2 Corporate sales 722 942 962 982 22 3 Expenses by division 742 932 972 992 12 4 Expenses by corporate chart of accounts 722 922 952 992 12 5 Changes in capital structure 712 902 912 962 42 6 Changes in productivity 692 872 922 982 22 7 Sales by division 552 872 952 982 22 8 Corporate earnings 582 852 902 932 72 9 Earnings by division 432 762 872 922 82 YEARLY VARIANCE 1 Corporate expense 652 902 972 982 22 2 Expenses by corporate chart of accounts 552 842 922 982 22 3 Corporate sales 532 842 932 952 52 4 Changes in capital structure 532 842 922 952 52 5 Expenses by division 532 822 952 972 22 6 Changes in productivity 472 772 922 972 32 7 Sales by division 362 742 882 942 62 8 Corporate earnings 372 702 802 872 132 9 Earnings by division 222 582 732 822 182 1Based upon cumulative variance at 102. Percent variances respresent plus or minus differences. SOURCE: "How Accurate Are Forecasts?" Financial Executive 41 (March 1973):27. 35 indicative of the difficulty in forecasting. For corporate earnings, 13 percent of the companies responding do not come within plus-or-minus 20 percent of their expectations over the period of a year. Eighteen percent do not come within plus-or-minus 20 percent of their projected division earn- ings. This experience clearly shows that internal forecasts are not precise estimates of a company's earnings. In fact, Table [2] shows that for a great many companies they are not even reasonable estimates of earnings. The report concludes: The public disclosure of internal forecasts would seem to have limited usefulness to the investor because of the inaccuracy of the forecasts.2 In contrast to the Daily and McDonald findings, the Financial Executives Research Foundation study reported no association between forecast accuracy and the industry classification of the respondent. Also, in contrast to Daily's results, it was reported that smaller companies experienced a greater variance in their yearly sales fore- casts than did larger companies. The study also reported no signifi- cant association between accuracy and the existence or length of written assumptions. As for the job the financial community is doing in forecasting company earnings, over sixty-five percent of the responding companies indicated a belief that a good job was being done. However, it was reported that some companies considered a ten percent variance from internal forecasts to be a good estimate while others were willing to accept a variance of as much as seventy percent. Regarding communica- tions with financial analysts, when a company's internal forecast differs "substantially" from one published by the financial community, 11bid., pp. 27—28. 2Ibid., p. 32. 36 half the respondents reported that they inform the analyst of the dif- ference in expectations. Most respondents stated they inform an analyst if he is beyond the range of "reasonableness." AICPA research project on accountants' reports on forecasts. Two articles have been published reporting the results of parts of a larger AICPA research project on accountants' reports on forecasts.1 The research reported in the first of these articles represented an attempt to capitalize on past experience with forecasting in the United Kingdom. There, forecasts are sometimes included in prospectuses issued in takeovers or mergers and are required to be included in pro- spectuses for companies quoted or seeking quotation on the London Stock Exchange. The study dealt in part with the accuracy of these forecasts but primarily with the independent accountants' involvement with such forecasts. The latter topic is outside the purview of this thesis. However, two observations made by the English accountants based on their experi- ence with forecasting are of interest. First, the English accountants indicated they are not in favor of presenting range forecasts on the grounds that this format is not an adequate means of communicating the probabilistic nature of forecasts. Second, the English accountants expressed a great reluctance to be associated with forecasts with hori- zons exceeding 18 months. Because of differences in the British and American economies and legal environments, it is hazardous to draw inferences relative to 1See, D. R. Carmichael, "Reporting on Forecasts: A U. K. Per- spective," Journal of Accountancy 135 (January 1973):36-47, and Richard J. Asebrook and D. R. Carmichael, "Reporting on Forecasts: A Survey of Attitudes," Journal of Accountanny 136 (August 1973):38-48. 37 forecast disclosure in this country based on the accuracy of British forecasts or the experiences of the British accountants with forecasts. Accordingly, for purposes of this literature review, the review of the United Kingdom study is limited to the foregoing comments. The second article concerning the AICPA research project on forecasts reported the results of a survey of attitudes existing in the United States concerning proposals to expand the disclosure of fore- casts. A questionnaire was mailed to large samples of CPAs, chartered financial analysts (CFAs), and financial executives. The questionnaire covered numerous forecasting issues including equity in the dissemination of forecasts, auditors' reports on forecasts, and behavioral aspects of forecasting. Regarding the latter, interestingly approximately 40 per- cent of both the CPA and CPA respondents expressed beliefs that corpo- rations would generally tend to understate forecasts. A majority of both groups felt that fear of losing public confidence would deter cor- porations from purposely overstating forecasts. In general, the survey results indicated considerable support for broader disclosure of forecasts on a voluntary basis, but no sup- port for a mandatory disclosure requirement. Also, a consensus against CPAs reporting on forecasts was revealed. Foster study.1 This study attempted to assess users' reactions to earnings forecasts by measuring the trading volume and price reaction of the stock market to the release of estimated earnings per share. Using the Wall Street Journal Annual Index, Foster selected a sample 1George Foster, "Stock Market Reaction to Estimates of Earnings per Share by Company Officials," Journal of AccountingyResearch 11 (Spring 1973): 25-37. 38 of 68 estimates of EPS. These estimates were published before the release of a preliminary earnings report or audited financial statements, but 2:53; the end of the fiscal period forecasted. The mean time-lapse between the release of the EPS estimate and the preliminary earnings report was just 18 trading days. On the basis of both the volume and price studies, Foster concluded that both individual investors and the aggregate market perceive the estimates of EPS to have informational content and do react to that source of annual earnings rather than waiting for the release of preliminary earnings reports on complete audited annual report data. Ferris study.2 This study also pertains to forecasting experi- ence in the United Kingdom. But the study is of interest in terms of its potential implications for similar phenomena in the United States. Through The Financial Times, 70 firms were identified which had issued prospectuses containing forecasts during the period December 1972 through December 1973. Of those, useable responses to a questionnaire ‘were received from individuals extensively involved in the forecasting process in 31 firms. The questionnaire was developed towards two basic objectives: (1) to ascertain whether management did or did not inten- tionally overestimate the profit forecast, and (2) to determine whether management intentionally utilized internal behavioral responses in an effort to reduce forecast deviations to an acceptable level. Regarding the first objective, the researcher found that 21 of the 31 firms studied, or approximately 68 percent, did intentionally 1Kenneth R. Ferris, "Profit Forecast Disclosure: The Effect on Managerial Behavior," Accounting and Business Research 5 (Spring 1975): 133-39. 39 manipulate (underestimate) the forecast that was published in the pro- spectus. Ferris further concluded that the intentional underestimation was primarily a function of two factors: (1) a learned attitude of con- servatism, and (2) a concern for the reaction of the business community to forecast failure. Regarding the second objective, respondents for 13 of the 31 firms (42 percent) admitted to utilizing accounting adjustments, defined as the use of alternative accounting methods or the adjustment of accounting records, for the purpose of reducing expected deviations between actual and forecasted results. Moreover, 26 (71 percent) indi- cated that their operating decisions had been consciously influenced by their concern for achieving the published forecasts. Basi, Carey, and Twark study.1 This study focused on the rela- tive forecasting ability of managements and financial analysts by com- paring the accuracy of forecasts issued by both groups for the same firms and the same time periods. Management forecasts for 88 firms which were referenced in the Wall Street Journal Annual Index for 1970 and 1971 were examined together with analysts' forecasts for the firms obtained from Standard and Poor's Earnings Forecaster. The sample included point and closed-range forecasts of EPS expressed in dollars or as a percentage increase or decrease from the previous year's EPS. Closed-range forecasts were converted to point forecasts by using the midpoint of the range. Open-range ("at least" type) forecasts were excluded. 1BartA. Basi, Kenneth J. Carey, and Richard D. Twark, "A Com- parison of the Accuracy of Corporate and Security Analysts' Forecasts of Earnings," Accounting Review 51 (April 1976):244-54. 40 For the samples of forecasts examined, on average the analysts forecasts overestimated EPS by nearly nine percent while the management forecasts averaged a six percent overestimate. The corresponding mean absolute percentage estimates were 14 and 10 percent, respectively. While acknowledging the occurrence of several large errors in excess of 100 percent, the researchers did not report the effect of such extreme values on the mean error statistics. More than 70 percent of the fore- casts by both the analysts and the executives were within :10 percent of actual EPS. Based on the entire sample, the cumulative absolute percentage error distribution for the management forecasts dominated the corre- sponding distribution for the analysts' forecasts. That is, the propor- tion of management forecasts which fell at or below a given absolute per- centage error level was always greater than the corresponding proportion of analysts forecasts. However, the first degree stochastic dominance of the management forecasts was not statistically significant. On other matters studied, the researchers found both management and analysts' forecasts to be more accurate for utilities than non- utilities. Forecasts for firms on the New York Stock Exchange were generally more accurate than those for firms on the American Stock Exchange. Lorek, McDonald, and Patz study.1 This study represents an extension of the prior research done by Green and Segall and Copeland and Marioni on the relative accuracy of management forecasts versus 1Kenneth S. Lorek, Charles L. McDonald, and Dennis H. Patz, "A Comparative Examination of Management Forecasts and Box-Jenkins Forecasts of Earnings," Accounting;Review 51 (April l976):321-30. 41 forecasts produced by extrapolative models relying solely upon past earnings data. But rather than using simple (naive) models, Lorek, et.a1., thought it would be more appropriate to use more sophisticated time-series models to test the hypothesis of the superiority of fore- casts issued by informed management. Rather than imposing a single model or set of models to be used to extrapolate future earnings for all firms in their sample, the researchers used a procedure known as the Box-Jenkins methodology to determine the most appropriate time-series model for each firm. Fore- casts were then generated with these firm specific models utilizing from 32 to 52 observations of past quarterly earnings data depending on their availability for specific firms. The sample consisted of 40 firms randomly selected from among those represented in the sample of management forecasts used in the McDonald study reviewed earlier in this chapter. It may be recalled that the sample for that study consisted of point forecasts disclosed in.the Wall Street Journal during the period 1966 through 1970. Only forecasts issued within the first 120 days of the fiscal year forecasted were included. Based on the comparison of the accuracy of the management forecasts with the time-series forecasts for the 40 firms, the researchers rejected the null hypothesis of no difference between the accuracy of the two sources of forecasts. They accepted the alternate hypothesis that the time-series forecasts were more accurate than the management forecasts. This concludes the identification and review of prior empirical studies on management forecasts. The next three sections of this chapter present comparative summaries of the findings of prior research on 42 aspects of forecasting further investigated in this thesis - specifi- cally, the frequency of forecast disclosures, forecast accuracy, and selected management forecast disclosure practices. Summary of Findings on Frequency of Forecast Disclosures Data on the frequency with which firms both prepare and voluntar- ily disclose forecasts is relevant to evaluating alternative policies on future forecast disclosure. The Financial Executives Research Foundation study reviewed above showed that the preparation of internal forecasts by firms is widespread, 95 percent of the 338 companies responding to a sur- vey indicating they prepare forecasts of sales, expenses, and earnings. The extent of voluntary disclosure of such forecasts is not ' known. Prior studies have provided some data on frequency of disclosure but varying sampling objectives and sample selection criteria make it difficult to generalize from the findings. For example, Copeland and Marioni scanned the Wall Street Journal for forecasts quantified in point or range format but stapped after locating arbitrary predetermined sample sizes. McDonald scanned the January through April issues of the Wall Street Journal for the five year period 1966 through 1970 but limited his sample to point forecasts with minimum horizons of 245 days. A further indication of the state of knowledge about the fre- quency of management forecast disclosures is provided by the following statement from a Wall Street Journal article: The SEC doesn't know how many companies make projections in an average year, but estimates run into the thousands. Several years ago the agency said that in November 1972 alone, The Wall Street Journal carried 153 reports of forecasts by corporate managers.17 1"SEC Proposes Firms Report to Agency Profit Forecasts Given to Analysts, Press," Wall Street Journal, April 29, 1975, p. 2. 43 The article did not say how the term "forecasts" was defined. Alterna- tive definitions and the numerous types of media through which forecasts may be disclosed add to the difficulty of determining the frequency of forecast disclosures. This study adds to the data available on frequency of disclo- sures by determining the proportion of firms from a defined universe of firms which have disclosed forecasts through one medium (the Wall Street Journal). To obtain a more complete picture of forecast disclosures, liberal limits on horizon were used. Also, since the studies reviewed above have provided little data on the frequency with which forecasts have been disclosed in alternative formats, an objective of this study was to provide data on the frequency of use of point, open-range, and closed-range formats. Details of the data collection procedures are presented in Chapter 3. Summary of Findings on Accuracy of Management Forecasts In Chapter 1, it was noted that the SEC cited continuing diver- sity of views on the reliability of forecasts as one reason for its decision expressed in the April 1976 release to neither encourage nor discourage the disclosure of forecasts. While this diversity of views may relate in part to lack of data on how reliable forecasts must be to be useful, it may also relate in part to the diversity of findings in prior studies on the accuracy of forecasts. The results of eight of the studies dealing with the accuracy of management forecasts reviewed previously in this chapter are summarized in Table 3. In addition to the results, Table 3 provides a brief description of the samples studied based on the specificity of details 44 .uuoooouOu osolouosol on newness. cocoon agovOl nouns-Iol«u concouu-«aaoo no sou-ounce .uoaooo no Nana casual also ounloooou no nos u~.o~+ nachos o>wuo~ou unsucce- coo: uo+ .uouuo o>uuo~oo sound. coo: .«onuoo no name agenda oloo nooooooOu no «em .unauoo uo uc~« nueuwo also coo-oOuOu an new .uoauoo no acne casual also ouoooouOu no no: no.n~I .uouoo o>uuo~ou wanna. coo: .uoauoo no no—« casual lloo ou-ooouOu no use .luouol sauna no loo-oouou ou nauseous. acumen ouoooooou uooloulcl: u~.c~+ «nouns cauosuou nus—ones coo: no.n~+ «nouns o>uuonou scene. and: .nusloouOu anal Indusol co nodulunu no: woman“ casual ladle uo lulmoouou .nuousulo nu onslao no and» Ioluuv uooquo slalomvsu luoooooOu nu no on uoaoouoh ooooouOthosuo< aaauue mummmaummummmmm 63:8... 32 .vo«u«ooao no: alloouou HOIUOHOEIdQDHQ< can u unooouou ACQUU‘ dosoouusooxu nouooosuu .lucolouocsl ha Iona cu vouaaean couloUuOu unsusuoo no unequoa->o .ouoallll cuuouuuon no hooks. I soda-novoh ouohuoa< «laud-au- .-hov new no sensuous lsldnul .Aoolouu uuua< swoops» scape-n. am: ass .o .o:..« chad-opus as woo-oon .ouoooouou geese was «an I unoconoz .ucolouocul loam suuoouuv condense clause use no nu-ooouou on I sauna III.oeosol n.s no condos: uncoop- .nm3 ecu no nos-cu coon ow vouooOA .ounluOu sun-uIvoao«o as. .clno .unuom an ou-ooOuou man an I «sown-x as. used-coo .noonnnoau onus» «on Iowan «saso< an: an voosououou .o>«uoo«uaoao and mono.» unquauosu ounluou cannon: nu nusooouou an I museum use noose suns-o. qua-so: oouuu announc- soaoauuolon oualoa as. mason uhmcufllch auluzl¢fl hllxuu loom osooooaaoooaa poo .Hoowvoa .oooasaouuoooo .o>«uoaou=u .maaweoa an I cm s m n ma 5 m m oomomo ammo and ooomofim so I co m m NH 0 mN n . NH m N ovmoo Haooou moo mammoaonz on I on a «H a n mm as «H a m poua>upp massage» was .mow .oaooooao .oONo Iooaosaaoo .oowooouoomsoua me I cc mm an ac cm ANN an em Ne «o wouusooousnoz an I ma Nnma Hume chad mead asouu Nhad Heed chad aoma unannouu ouoo vonmNHosm mom voomooouom huonsooH UHm ufimwnIN Hoow Howsoaoo Hooch Moor Hoouwm Mdmw 92¢ UZHmbomu NMBmDQZH Mm mam202 __l_ 100 . 0 Total Cases i—é—B Table 13 shows that indeed very narrow ranges were often used with over half of the ranges encompassing less than :3 percent around the mid- point. Ninety-eight percent of the ranges encompassed less than or equal to :10 percent around the midpoints. Further study of the range widths indicated that they tended to be expressed in discreet intervals of $.05, $.10, and $.25 rather than some common percentage of the midpoints. 94 In view of the nonrstandardized and frequently narrow ranges characterized in the sample, further analysis of the reliability of range forecasts in terms of degree of closeness to being right seemed appropriate. For this analysis, first a frequency distribution of relar tive differences was compiled by substituting the range midpoints for forecasted EPS in the same relative error or difference formula used for the analyses of point and openrrange forecasts. Since the identity of the end points of the ranges is lost in this analysis, reference can be made only to "relative differences" as opposed to "relative errors." The frequency distribution prepared in this manner is presented in Figure 5. The bar at the extreme left end of the distribution includes three cases of positive forecasts associated with negative actual EPS. ..£L, 16 247—3-3'L‘LTJ-r _z_‘ 2 3‘ 51;o -,99 -,50 -,‘o -,30..,15 .,20 -,15 -,10 -.05-.001F001 .05 .10 .15 .20 .25 .30 .40 .50 .99 31.0 0 Figure 5. Frequency Distribution of Relative Differences in Closed-range Forecasts Measured froe Midpoints. 95 The variance of this distribution is .075. Again the statistic is of little significance by itself but is examined further in the next section of this chapter on the comparative analysis of forecasts issued in different formats. An alternative feel for the dispersion in the relative differences can be obtained from further inspection of the distribution in Figure 5 which shows that 83 or 55 percent of the mid- points fell within 15 percent of actual EPS, 107 or 71 percent fell within :10 percent and 114 or 76 percent fell within :15 percent of actual. The signed mean of the distribution in Figure 5 is -.092. This indicates an optimistic bias and is consistent with the finding that there was a slightly higher frequency of forecasts with midpoints exceeding actual EPS than falling below actual EPS. But since the former exceeded the latter by only 4 cases, the negative bias component is more a reflection of the greater proportion of extreme negative observations relative to extreme positive observations than of any consistent tendency toward optimistically biased midpoints. To Obtain a better indication of the degree of bias in the majority of forecasts, the mean of the relative differences was recomr puted after removing outliers defined again as relative differences with absolute values greater than or equal to 100 percent. There were 9 out- liers in the original sample of 150 closed-range forecasts including three cases of positive forecasts associated with negative actual earn- ings. The mean after removing outliers was -.034 compared to -.O92 before their removal. As a further test of bias in the distribution of relative dif- ferences for closed-range forecasts, the mean was subjected to the 96 following hypothesis test: Hos: The mean signed relative difference based on mid- points of closed-range forecasts (excluding out- ;liers) is equal to zero. The test was based on the mean of the reduced sample of 91 independent closed-range forecasts less 8 outliers. The values of the mean and variance of the sample were -.039 and .026, respectively, resulting in a 2 value for the one-sample test of means of -2.l87 which is signifi- cant at the .05 level for the two-tailed test (.025 z - £1.96). Since the sign of the mean was negative, the results of this test indicate that there is a statistically significant optimistic bias relative to the midpoint of closed-range forecasts. The means of the absolute relative differences based on the range midpoints were also computed for both the original and independent samples, both with and without outliers. The mean absolute relative difference was .144 based on the entire sample of 150 closed-range forecasts and .089 after removing nine outliers. The mean absolute relative difference based on the independent sample of 91 forecasts was .175, and after removing eight outliers it was .096. These and other descriptive statistics, and the results of the hypotheses tested, based on the midpoints of closed-range forecasts, are summarized in Table 14 on page 99. To provide a further basis for comparing the closed-range fore- casts with those issued in point and Openrrange formats an additional frequency distribution of relative differences was compiled by substi- tuting the minimum estimates of the ranges in the relative error or difference formula. That frequency distribution is presented in Figure 6. 97 22 12 ri-_:_.-’-,_3_ 3 3 3 "'1.0 O.” ..50 O.“ -.30"025 -020 -015 ‘010 -005 ‘0q0m1 .05 01° .15 020 025 .30 0‘0 050 0” 11.0 Figure 6. Frequency Distribution of Relative Differences in Closed-range Forecasts Measured froe Minimum satin-ates of Ranges. The analysis of this distribution proceeded in a manner similar to that reported above for the distribution of relative differences based on the range midpoints. Key statistics from this analysis are summarized in Table 14 below together with the statistics based on the previous analysis by range midpoints for comparison purposes. Null hypothesis H06 was formulated analogous to 303 for open? range forecasts as follows: B06: The proportion of closed-range forecasts for which the minimum EPS estimates exceed actual EPS is greater than or equal to .5. The computed 2 value for the onersample test of proportions was -2.l94 which is significant at the .05 level for the one-tailed test (.05 z - -l.645). Thus the null hypothesis was rejected in favor of the alternate hypothesis that the proportion of closed-range forecasts for which the mflnimum.estimates exceed actual EPS is significantly less than .5. 98 Finally, null hypothesis H07 was formulated analogous to H05 in the analysis of closed-range forecasts based on midpoints as follows: H07: The mean signed relative difference based on minimum estimates of closed-range forecasts (excluding outliers) is equal to zero. The computed 2 value for the one-sample test of the mean was -.3502 which is not significant at the .05 level for the two-tailed test (.025 Z - 11.96). Thus, the bias component for the distribution of rela- tive differences based on the minimum estimates of closed-range forecasts was not significantly different from zero. The results of the above hypothesis tests are also summarized in Table 14. Further analysis of the reliability of closed-range forecasts relative to point and open-range forecasts is reported in the next section. Comparative Analysis of Forecasts Published in Point, Open, and Closed-ragga Formats It seems reasonable to assume that the selection of the format for publishing a forecast is related to management's preference with regard to conveying different, although incomplete, information about the unspecified probability distributions underlying the forecasts. For example, conceivably the choice of disclosure format might be related to the degree of management's uncertainty about future earnings. In that event, one would expect forecasts issued in point format to represent "best" or "most probable" estimates in which managements have a high degree of confidence. On the other hand, one would expect forecasts to be disclosed in one of the range formats when managements are less certain about future earnings - that is when the underlying probability distributions are flatter. In turn, one would expect to find less ST9 330?: <2 <2 <2 <2 <2 <2 <2 6...: so: on assoc ea Aeueuuuso usuoouuuov eueeuouou euoequoeoqu no seas-«use Islands so posse eoeououuuo 0>uue~0u secede sees say "No: <2 <38?— <2 <2 <2 <2 <2 <2 .n. 3 3.3 no sea» houses» ed man mesons venous sou-sauce emu assuage eau sous: sou sueeoouou enseuloeeodo no couuuonoue oak .oo: <2 <2 <2 <2 <38?— <2 <2 <2 .83 3 <3... 3 2.0226 9.2.388. eueeoouou ensequeeouu no eunuoneus so see-a eoeeueuuqe sequeueu oeouqe sees 05h «no: «Hana mumuzuomrm cod. and. mac. non. mac. and. mac. «em. eeuoeueuuqo 0>uueneu eueuoeae no see: coo.I noo.I noo.I aeo.I anc.I n-.I uueu0u ooeuue no one: cue. «on. due. one. owe. goo. nae. m~o. eeuoeueuuue saga-nee assua- uo «unease» I «on. I «no. I I I I dengue no useuuen one nuauus eseo eeueluuee Islands an I I I I I «no. I nun. .qesuoe mo aeoouee Qua saga“) eleo sueuoeoas as I mom. I son. I non. I son. menace oeooeune euouonoas Aw I fion. I new. I son. I new. «ensue sods: dueu eeueluuee ensures Au I man. I sen. I men. I sen. dengue eeeeeune sous-«nee Islands as I won. I con. I won. I con. Aeueeueuou uauuuv aesuue vesseAIouse sense» As “sous: sou eueeueuou no seenuquoum mm ~e ~<~ and no as ac“ can onus one-em monumuhdhu euewuuao ences» euewauso odes-m eueunusc euaeem eueauuso nausea need sumac» seed euuuou 000a euuuen seen undue» [$me «and» an Seawall-.. useoeeeeoou aesuuquo uoeoeeeeosu «esuuuuo eouesuuee loaded: lufld at“: In“; so oases eqehqeo< mkm¢0fluch wondllnflmoau 80h aHmNH mummzhcara 92¢ unufla~huhhuuuafln ho rddxxfim on ”Adda 100 dispersion in the relative differences between estimated EPS and actual EPS associated with point forecasts than with range forecasts. Conceivably, the choice of format could also be related to the desire of management to focus attention on particular segments of the unspecified probability distributions underlying the forecasts. For example, management might express a forecast in such terms as "fully diluted net earnings are expected to be 'at least' $3.00 per share" with the intention of focusing attention on the upside potential of earnings. In connection with this line of reasoning, it is imperative from the standpoint of users that the choice of disclosure format not be used as a device for misleading users. For example, it further seems reasonable to infer that the probability of future EPS exceeding an "at least" or minimum EPS estimate is greater than the probability of earnings falling below that amount. The finding in the previous section on open? range forecasts that the proportion of such forecasts for which actual EPS fell below the minimum.estimates was significantly less than .5 supports this expectation. But, if the selection of disclosure format is related to differences in the probability distributions underlying the forecasts, one would expect the proportion of minimum estimates which exceed actual to be significantly smaller than the proportion of point;forecasts which exceed actual. At the same time, if the minimum EPS estimates in openvrange forecasts are more conservative than point forecasts as one would expect, the bias component or mean of the open? range forecasts would likely be significantly more positive than that for the point forecasts. This line of reasoning suggests that comparisons among desig- nated proportions, variances, and means of the relative errors or 101 differences for forecasts published in different formats may be useful. In particular, such comparisons may be used to examine the extent to which forecast disclosure practices conform to inferences which might be drawn based on disclosure format. Accordingly, several hypotheses were formulated and tested as described in the following paragraphs. All tests were based on the relevant statistics for the reduced samples of independent observations. Because the occurrence of extreme relative differences does not bias the proportion statistics, tests of propor- tions are based on the independent samples including outliers. But because the extreme values were found to distort the variances and means of the relative differences, tests of those statistics are based on the independent samples excluding outliers. For convenient reference, the relevant statistics reported in the previous sections are summarized in Table 19 on page 111. Comparative analysis of proportions. Because the proportions of forecasts issued in different formats which are classified as "right" in this thesis are not measured on the same basis, comparisons among such proportions are not meaningful. The comparative analysis of pro- portions here focuses on proportions of forecasts for which actual EPS fall below the point estimates in point forecasts, minimum estimates in open-range forecasts, and the midpoints or minimum estimates in closed—range forecasts. To determine whether there are significant differences among these proportions for forecasts issued in different formats, the following null hypotheses were tested: H08: The proportion of openrrange forecasts for which actual EPS fall below the minimum EPS estimates is greater than or equal to the proportion of point forecasts for which actual EPS fall below the point estimates. Ho 10' Ho P. H°12 102 The proportion of closed-range and point fore- casts for which actual EPS fall below the range midpoints and point estimates, respectively, are equal. The proportion of closed-range forecasts for which actual EPS fall below the minimum estimates is greater than or equal to the proportion of point forecasts for which actual EPS fall below the point estimates. The proportion of openrrange forecasts for which actual EPS fall below the minimum EPS estimates is greater than or equal to the proportion of closed-range forecasts for which actual EPS fall below the range midpoints. The proportions of closed-range and openrrange forecasts for which actual EPS fall below the minimum estimates are equal. The two independent samples test of proportions was applied for each null hypothesis using the following test statistic: Xé—i )C'MDC”) where for each null hypothesis: P and P - the first and second proportions identified, 1 f and f 1 2 - the numbers of forecasts possessing the characteristic of interest in the samples from the first and second populations identified, and 2 n1 and n - the size of the samples from the first and second populations identified. The tests were conducted at the .05 level of significance. The results of the tests are presented in Table 15. In Table 15 it can be seen that while the minimum estimates in openerange forecasts exceeded actual EPS less often than the point estimates in point forecasts as would be expected (H08), the difference 103 TABLE 15 RESULTS OF TESTS OF PROPORTIONS USING INDEPENDENT SAMPLES Null Computed Critical Hypothesis P1 P2 2 2 Decision H08: popen g ppoint .244 .371 -l.496 -1.645 Not rejected 309: Pclosed (MID) - point .505 .371 1.851 £1.96 Not rejected H010: pclosed(MIN) 2 ppoint .385 .371 .197 -1.645 Not rejected H011: popen Z pclosed(MID) .244 .505 -2.903 -1.645 Rejected 3°12: popen - pclosed _ (MIN) .244 .385 -l.636 ‘ $1.96 Not rejected in proportions was not significant at the .05 level. But the direction of the difference in the same proportions for closed-range and point forecasts was opposite that expected (H010) - that is, the minimum esti- mates in closed-range forecasts exceeded actual EPS more often than did the point forecasts. While the difference is not statistically signifi- cant at the .05 level, the implication is that the closed-range minimum estimates are no more conservative than point forecasts and therefore are not very reliable as indicators of "minimumfi earnings. Ho9 provides a test of the a priori reasoning that point fore- casts and the midpoints of closed-range forecasts are equivalent as EPS estimators. Table 15 shows that the two-tailed test was not significant. But further inspection indicates that a one directional test would lead to the conclusion that the midpoints of closed-range forecasts exceed 104 actual EPS significantly more often than do point forecasts. Thus, the midpoints of closed-range forecasts may be said to be significantly optimistically biased relative to point estimates. Comparing the open and closed-range proportions, Table 15 shows that as would be expected the minimum estimates in open-range forecasts exceed actual EPS significantly less often than do the closed-range mid- points (Ho Also, it may be noted that the minimum estimates in the ll)' sample of openrrange forecasts tended to be more conservative than the mdnimum.estimates in the sample of closed-range forecasts but the differ- ence was not significant at the .05 level (H012). In summary, based on the above analysis, it can be concluded that the open—range and point forecasts are significantly more conserva- tive than the midpoints of closed-range forecasts like those studied. Comparative analysis of dispersion. Next, to determine whether there are significant differences among formats in the objectivity or dispersion of the relative error or differencecmeasures, the following null hypotheses were tested: Ho : The variances of the relative differences for 13 point and openerange forecasts are equal. Ho ° The variances of the relative differences for point and closed-range forecasts (based on mid- points) are equal. Ho ° The variances of the relative differences for 15 point and closed-range forecasts (based on minimum estimates) are equal. H016: The variances of the relative differences of open- range and closed-range forecasts (based on mid- points) are equal. Ho ' The variances of the relative differences for open- range and closed-range forecasts (based on minimum estimates) are equal. 105 The significance test used for each hypothesis was the Bartlett-Box F l test. The results of the tests are presented in Table 16. TABLE 16 RESULTS OF TESTS OF VARIANCES USING INDEPENDENT SAMPLES . . 2 2 Bartlett- Signifi- .N Hypothisis s1 82 Box canoe Decision F Level H013: 02 point - 62 open .024 .013 5.510 .019 Rejected H014: 02 point I 02 closed(MID) .024 .026 .199 .656 Not rejected H015: a2 point - 02 closed(MIN) .024 .024 .006 .938 Not rejected H016: a2 open - 62 closed(MID) .013 .026 6.973 .008 Rejected H017: 02 open - o2 closed(MIN) .013 Rejected .024 5.632 .018 Table 16 reveals significant differences at less than the .05 level in all pairs of variances involving openrrange forecasts. Specifically, the variance of the relative differences for openrrange forecasts was significantly less than the variance of the relative 1The standard F test for equality of variances is known to be sensitive to violations of the assumption of normality. Since non- ' normality in the populations of relative differences of interest here is suspected, a modified F test known as the Bartlett-Box F test was employed. This test is less sensitive to nonrnormality. The Bartlett- Box F statistics and significance levels reported in Table 16 were obtained as output of the "Oneway" (analysis of variance) procedure of the "Statistical Package for the Social Sciences." For an explanr ation of the modified F test, see C. E. P. Box and S. L. Andersen, "Permutation Theory in the Derivation of Robust Criteria and the Study of Departures from.Assumption," Journal of the Royal Statistical Society (Series B) 17 (No. 1, 1955):16:22. 106 differences for point forecasts (3013) and less than the variances of both distributions for closed-range forecasts (Ho and Ho There 16 l7)° is no significant difference in the variance of the relative differ- ences for point forecasts and the variances for either closed-range distribution (Ho14 and HolS)° A further indication of the comparative dispersion of the rela- tive differences for forecasts issued in different formats can be ob- tained by comparing the proportions of relative differences for each format which fell within :10 percent of zero. As reported previously, these proportions were 74, 78, 68, and 59 percent for the point, open- range, closed-range (based on midpoints), and closed-range (based on minimum estimates) relative differences, respectively. Both analyses suggest the openrrange forecasts possess greater objectivity. Comparative analysis of means. Finally, to determine whether there are significant differences among formats in the bias components or means of the relative differences, the following null hypotheses were tested: H018: The mean signed relative difference for open-range forecasts is less than or equal to the mean signed relative difference for point forecasts. 3019 The means of the signed relative differences for closed-range forecasts (based on midpoints) and point forecasts are equal. Ho The mean signed relative difference for closed-range forecasts (based on minimum.estimates) is less than or equal to the mean signed relative difference for point forecasts. Ho The mean signed relative difference for openrrange forecasts is less than or equal to the mean signed difference for closed-range forecasts (based on ‘midpoints). 21‘ H022: The means of the signed relative differences for 101 open-range forecasts and closed-range forecasts (based on minimum estimates) are equal. The two independent samples test of means was applied for each null hypothesis using the following test statistic: J-‘l ' E2 S1 +82 1“1 n2 where for each null hypothesis: £1 and E2 I the first and second means identified, 512 and S 2 - the variance of the samples from the first and 2 second populations identified, and n and n2 - the size of the samples from the first and 1 and second populations identified. The results of the tests which were based on the independent samples excluding outliers, and which were conducted at the .05 level of sig- nificance, are presented in Table 17. Table 17 reveals that the mean for closed-range forecasts based on midpoints is significantly smaller than the mean for point forecasts (Ho This indicates that the range midpoints are significantly 19’ ' optimistically biased relative to the point forecasts. Additionally, Table 17 shows the mean of the closed-range forecasts based on midpoints to be significantly smaller than the mean of the open-range relative differences as would be expected (Ho The means of the relative 21)' differences for minimum.estimates were not significantly different from the mean for point estimates, thus failing to demonstrate any significant 1The arguments presented in footnote 1 on page 83 regarding robustness of the one sample test of means to nonrnormality also apply to the two independent samples test of means. For each of the hypotheses tested in this section, the sample size for each sample mean was considerably greater than 30. 108 TABLE 17 RESULTS OF TESTS OF MEANS OF SIGNED RELATIVE DIFFERENCES USING INDEPENDENT SAMPLES Null 2 i Computed Critical Hypothesis 1 2' z 2 Decision Hols: u open 5 u point .028 .018 .425 1.645 Not rejected H019: u closed(MID) - u point -.039 .018 -2.385 £1.96 Rejected H020: u closed(MIN)-s u point -.006 .018 -l.026 1.645 Not rejected H021: u open 5 u closed(MID) .028 -.039 2.716 1.645 Rejected 3022: u open - u closed(MIN) .028 -.006 ‘1.406 $1.96 Not rejected conservative bias in the minimum estimates (Bo and Ho 18 20)' To provide an additional comparison among formats of the magnitude of the relative differences but not of their direction, the absolute means were examined. Hypothesis tests of the differences between pairs of the absolute means provide a method of testing the validity of the inference that point forecasts represent best estimates which generally have a higher probability of occurring than do specified common points in range forecasts such as the minimum estimates in either open or closed- range forecasts or the midpoints of closed-range forecasts. If this inference were valid, one would expect to find larger absolute mean rela- tive differences for the range forecasts than for point forecasts. Thus, hypotheses H023 through H027 were formulated similar to Ho18 through . Ho 22 except for the substitution of absolute for signed means and changes in the directionality of some of the tests. Null hypotheses Ho23 through 109 8027 are summarized in Table 18 together with the results of the tests which again were based on the Z test statistic for two independent sample means.1 No significant differences were found at the .05 level, failing to conform to the a priori reasoning that the selection of disclosure format is related to the degree of uncertainty about the forecast. TABLE 18 RESULTS OF TESTS OF ABSOLUTE MEANS USING INDEPENDENT SAMPLES Null - - Computed Critical Hypothesis :1 1‘2 z z ”“19““ 3023: u open 5 u point .073 .081 -.412 1.645 Not rejected 8024: u closed(MID) 5 u point .096 .081 .730 1.645 Not rejected H025: u closed(MIN) s u point .100 .081 .998 1.645 Not rejected H026: u open - - u closed(MID) .073 .096 -l.l35 $1.96 Not rejected 8027: u open - u closed(MIN) .073 .100 -1.442 $1.96 Not rejected Summary of Analysis of Reliability The foregoing analysis suggests that neither proportions of right versus wrong forecasts nor measures of the bias and objectivity components of the degree of closeness to being right in the form of means and variances are adequate by themselves for assessing the 1While the population of absolute relative differences is highly skewed, the sampling distribution of i will still be approxi- mately normal for large samples like those used in these tests. 110 reliability of published forecasts. But a composite of these statistics does provide information about the reliability of past forecasts and the kinds of inferences that can be made about like forecast disclosures. A summary of the key statistics from the preceding analyses is presented in Table 19. In the separate analyses of format, both the analysis of propor- tions and the analysis of the meansshowed the point forecasts to be un- biased. The separate analysis of open-range forecasts found the minimum estimates to be conservatively biased in the sense that significantly less than half of the minimum.estimates turned out to be overpredictions and the mean or bias component of the distribution of relative differ- ences was positive and significantly different from zero. However, the proportion and mean statistics for the open-range forecasts were not significantly different from those for the point forecasts, leading to the conclusion that the minimum estimates in open-range forecasts are not significantly more conservative than point forecasts. Therefore, it cannot be inferred that the probability of actual EPS falling below the minimum estimates in open-range forecasts is less than that of actual EPS falling below forecasts issued in point format. The same conclusion applies to the minimum.estimates in closed-range forecasts based on the finding that none of the statistics related thereto was significantly different from those for the point forecasts. Nor was there any signifi- cant difference found in the degree of conservatism or bias between the minimum.estimates in apenrrange and closed-range formats. The separate analysis of closed-range forecasts based on mid- points revealed them to be optimistically biased based on the mean of the relative differences which was significantly less than zero. 111 TABLE 19 SUMMARY OF DESCRIPTIVE STATISTICS BASED ON INDEPENDENT SAMPLES USED IN COMPARATIVE ANALYSIS OF POINT, OPEN‘RANGE, AND CLOSED-RANGE FORECASTS Samples (Closed-Range Minimum Opens Midpoint Estimate Point Range Analysis Analysis Statistics based on independent samples including outliers: Sample size 97 45 91 91 Proportions of forecasts for which actual EPS fell below: a) point estimates in point forecasts .371 b) minimum estimates in open- range forecasts .244 c) midpoints of closed-range forecasts .505 d) minimum estimates in closed- _ range forecasts .385 Statistics based on independent samples excluding outliers: Sample size 93 44 83 83 Variance of signed relative differences .024 .013 .026 .024 Mean of signed relative differences .018 .028 -.039 -.006 Mean of absolute relative differences .081 .073 .096 .100 112 Moreover, the midpoints were found to be significantly optimistically biased relative to point forecasts and the minimum estimates in open! range forecasts. In terms of objectivity, each of the separate analyses by format revealed that the absolute values of relative differences ranged from zero to greater than 100 percent. The comparison of variances among formats revealed that the openrrange forecasts had a significantly smaller variance than either point forecasts or closed-range forecasts. There was no significant difference in the variance of the point and closed-range forecasts. But comparisons among the absolute mean rela- tive differences revealed no significant differences among forecasts, failing to support the a priori reasoning that the selection of disclo- sure format is related to the degree of uncertainty about future earn- ings. Thus, no inference should be drawn about the dispersion in the probability distribution of future earnings underlying a forecast based on the disclosure format. Finally, fairly large proportions, 74, 78, and 68 percent, respectively, of the point estimates, minimum estimates in openprange forecasts, and midpoints of the closed-range forecasts were found to come within :10 percent of actual EPS. This concludes the analysis of the reliability of management forecasts of earnings per share published in point, openrrange, and closed-range formats. The remainder of this chapter reports on the investigation of the association between designated independent vari- ables and selected management forecast disclosure practice variables. 113 ANALYSIS OF ASSOCIATION BETWEEN INDEPENDENT VARIABLES AND SELECTED MANAGEMENT FORECAST DISCLOSURE PRACTICES While the research reported in the first part of this chapter represents a refinement and extension of prior empirical research on the reliability of published earnings forecasts, the research reported in the remainder of this chapter is exploratory in nature, dealing with an issue not examined in prior empirical studies. Specifically, this exploratory research addresses the third research question posed in Chapter III - what variables are associated with management forecast disclosure practices? . The variables examined in this part of the study were: Dependent (Disclosure Practice) Variables: 1. Decision to disclosure or not disclose: a. An.initial forecast. b. A revised forecast. 2. Format (point, openrrange, or closed-range) of a forecast. 3. Timing of a forecast disclosure: a. Horizon - measured as the number of days between the issuance of a forecast and the end of the forecasted period. b. Time lapse between the issuance of an initial fore- cast and a subsequent revision measured in days. Independent Variables: 1. Accuracy of a firm's prior forecast: a. Accuracy of prior year forecast classified as accurate or inaccurate. b. Accuracy of initial forecast for year classified as overprediction or underprediction. 2. Horizon of a forecast (measured as described above). 3. Actual earnings trends: a. Change in actual EPS for year n relative to year n91. b. Change in actual EPS growth rate for year n relative to year n91. 4. Forecasted earnings trends: a. Change in forecasted EPS for year n relative to actual EPS for year n91. b. Change in implicit forecasted EPS growth rate for year n relative to actual EPS growth rate for year n91. 114 5. Issuance of a revised forecast disclosure classified as upward or downward. Specific relationships examined were the association between: 1. Prior year forecast accuracy and current year disclo- sure format. 2. Horizon and disclosure format. 3. Actual earnings trend and the decision to disclose or not disclose a forecast. 4. Forecasted earnings trend and the horizon of a fore- cast disclosure. 5. Direction of error in initial forecast and decision to disclosure or not disclose a revision. 6. Direction of revision and timing of revision. Relationships three through six listed above deal with an examination of the existence and timing of initial and revised forecast disclosures to determine whether favorable and unfavorable earnings expectations are given comparable treatment. In addition to providing users with information about the possible motives behind management forecast disclosure practices, the analysis of association between the designated independent variables and disclosure practice variables should be helpful in evaluating the need for, and desirability of, specific proposals for the regulation of forecast dis- closures. For example, evidence that managements tend to disseminate favorable expectations on a timely basis but to suppress unfavorable expectations would lend support to proposals that managements be re- quired to issue revisions on a timely basis-and/or disclose reasons for withdrawing a forecast or withdrawing from the forecast disclosure system altogether. The analyses reported in the following sections are based on the disclosure practices of the 233 firms represented in the data base used in the analysis of reliability, or subsamples thereof, as explained in each section. A.limitation.which should be acknowledged at the outset 115 is that certain analyses utilize the classification of a firm as not having issued a forecast when no forecast by the firm was included in the data base for a given year. In some cases, conceivably the firm could have issued a forecast through a medium other than the wall Street Journal or in a form which in.some other way failed to meet the selec- tion criteria for inclusion of a forecast in this study. This limits? tion is mitigated, however, by two factors. First, the wall Street- Journal undoubtedly has been, and continues to be, the major communica- tion channel for the wide dissemination of management earnings forecasts. Second, forecasts not meeting the other selection criteria for this study may be ambiguous or lack comparability with those forecasts meeting the selection criteria. Association Between Prior Forecast Accuracygand Current Disclosure The purpose of this analysis was to determine whether the accu- racy of a firm's prior forecast disclosure has any effect on (1) its decision to issue or not issue a forecast in the current period or (2) the format of a current forecast if issued. The sample for the analysis of association between prior accuracy and the decision to issue or not issue a forecast consisted of 180 firms for which one or more forecasts issued during the first three years of the four-year study period were included in the data base. For firms represented by more than one forecast_during the three years, the year included (year n91) was randomly selected. The initial forecast was used whenever more than one forecast occurred for the selected year. For each of these firms, the format of the first forecast included in the data base in the following (current) year (year n), if any, was 116 then determined. The contingency table shown in Figure 7 was then constructed by classifying each firm by the accuracy of its forecast for year n-1 and the presence or absence in the data base of a forecast by the firm for year n. .A dichotomous measure of accuracy was used, year n-l forecasts ' not meeting the following criteria for "accurate" being classified as "inaccurate": Format of Forecast For Year n Accurate when Point -.10 5 Relative Difference s .10 Open-range Relative Difference z 0 Closed-range ‘Minimum estimate 5 Actual EPS 5 Maximum estimate The 10 percent criterion for point forecasts is arbitrary, but a level frequently cited in the literature as representing a reasonable toler- ance for error.1 The criteria for open and closed-range forecasts are based on literal evaluations of the forecasts. Year n Year n-l Forecast No Forecast Total Accurate 27 74 101 Inaccurate .18 61 79 Total 45 135 180 Figure 7. Contingency Table Showing Relationship Between Accuracy of Forecast for Year nrl and Exis- tence of Forecast for Year n. 1For example, the SEC's April 1975 proposals on forecasts adopted a 10 percent criterion for reasonable error. 117 To determine whether the variables represented in the contine gency table in Figure 7 are independent, the X2 test of independence was used to test the following null hypothesis: H028: There is no association between the accuracy of a firm's forecast for year n91 and its deci- sion to issue or not issue a forecast in year n. The test statistic used was: J £ -£ 2 x2 2 § (ojk .ejk) 3-1 k-l £31k where: f - the number of observations in the (jk)th cell of the contingency table, fe - the expected number of observations in the (jk)th cell of the contingency table based on the sample marginal proportions. The computed X2 for the contingency table was .481 which is not significant at the .05 level (critical X? with one degree of freedom - 3.84l). Thus, the null hypothesis of no association between forecast accuracy in year n91 and the decision to issue or not issue a forecast in year n was not rejected. It may be noted that Figure 7 shows that the majority of firms were not represented in the data base in year n regardless of the accuracy of their forecasts in year nrl. A further analysis was made to determine whether for firms issuing forecasts in two consecutive years there is any association between forecast accuracy in year n91 and the forecast format used in year n. The sample for this analysis consisted of one pair of fore- casts for each of the 50 firms in the data base for which there were forecasts for two consecutive years. For firms with more than one pair of consecutive forecasts in the data base, one pair was randomly 118 selected for inclusion in this sample.1 Each firm was classified by the accuracy of its forecast for the first year (year n91) using the same criteria used in the preceding analysis. Each firm was also classified by the format (point or range) of its forecast for year n. The range category includes both open and closed-range formats. The resulting contingency table is shown in Figure 8. Year n Format . Year n91 Point Range Total Accurate l4 13 27 Inaccurate _5 18 g_3_ Total 19 31 SO Figure 8. Contingency Table Showing Relationship Between Accuracy of Forecast for Year n-1 and Forecast Format Used in Year n. To determine whether the variables represented in the contingency table in Figure 8 are independent, the X2 test of independence was used to test the following null hypothesis: Hozg: There is no association between the accuracy of a firm's forecast for year n91 and the format of its forecast for year n. ’ The computed X2 for the contingency table was 5.461 which is sig- nificant at the .05 level (critical X? with one degree of freedom - 3.841). 1A3 shown in Figure 7, the sample of 180 firms used in the analysis of prior accuracy and the decision to issue or not issue a fore- cast included 45 firms issuing forecasts in both years nel and n. The five additional firms in the present sample were among the 180 firms in the previous sample. But the pairs of consecutive forecasts for these firms did not commence with the randomly selected base year included in the previous sample. 119 Therefore, Ho29 was rejected. Inspection of the contingency table reveals the nature of the relationship. Firms with forecasts classified as inaccurate for year n91 used one of the range formats in year n sig- nificantly more often, relative to the use of the point format, than did firms with forecasts for year n91 that.were classified as accurate. This is in accord with what one would expect. Association Between Horizon and Disclosure Format Since point forecasts may be perceived by some users as more precise or more certain than range forecasts, it might be hypothesized that forecasts with shorter horizons and therefore presumably less unr certainty would more often be disclosed in point format, while fore- casts with longer horizons would more often be disclosed in either open or closed-range format. To examine this relationship, the point-biserial correlationcoefficient which is used to measure the association between one nominal-dichotomous variable and one ratio variable was computed.1 Format was treated as the nominal-dichotomous variable with value 0 for point format and value 1 for range format. No distinction was made between open and closed-range format (for which it happens a supple- mentary analysis showed the mean horizons to be equal at 141 days). Horizon, defined as the number of days between the issuance of a fore- cast and the end of the forecasted period, was treated as the ratio variable. The following null hypothesis about the correlation coeffi— cient was then tested: 1For an explanation of the point-biserial correlation coeffi- cient and a related significance test, see Gene V. Glass and Julian C. Stanley, Statistical Methods in Education and Psychologz_(Englewood Cliffs: Prentice-Hall, Inc., 1970), pp. 163-5, 318. 120 Ho3o: The population point-biserial correlation coeffi- cient for the relationship between horizon and format is less than or equal to zero. The formula used to compute the point-biserial correlation coefficient was: r b I x1 - xo . 1 11o p Sx n(nrl) where: E1 I the mean horizon of those forecasts in range format, Eh I the mean horizon of those forecasts in point format, Sx I the standard deviation of all n horizons, n1 I the number of range forecasts, the number of point forecasts, and D I n I n1 +'nb. As is evident from the formula, rpb is a measure of the difference between the average horizons for the range and point forecasts. The coefficient can assume values of from.-l to +1, the extremes representing a large difference between means and a value of zero representing no difference. H030 was then tested by computing the following test statistic for the computed value of rpb: a rpb /(1 - 1:31,)! (n-z) t If rpb is zero in the population sampled, than t is approximately distri- buted as Student's t-distribution with n-2 degrees of freedom. Note that a one-tailed test of rpb is closely equivalent to testing the null hypothesis that in the population, the mean horizon for range forecasts is less than or equal to the mean horizon for point forecasts. 121 The point-biserial correlation coefficient was computed for the sample of 233 independent forecasts used in the analysis of reliability, 97 of those being in point format and 136 in range format (45 in open? range format plus 91 in closed-range format). The mean horizons of the point and range forecasts were 146 and 141 days, respectively. The camp puted value of rpb was -.023 indicating virtually no association between horizon and format in the sample forecasts. As would be expected given the small difference between the sample mean horizons, the null hypothesis about rPb was not rejected (computed t I -.350; critical t (i.975 t with 231 degrees of freedom I 1.98 (approximately))). It may be noted, that in the sample forecasts, the mean horizon for point forecasts was actually slightly (not significantly) larger than the mean horizon for range forecasts. Association Between Actual EarningggTrend and Current Disclosure One might hypothesize that management would be more reluctant to issue forecasts when they expect their firms' economic performance to decline than when it is expected to rise. To examine this relations ship, the earnings performance for each firm for year n in a sample of firms which issued forecasts in two consecutive years (years n21 and n) was determined. The earnings performance of these firms was then com? pared to the earnings performance of a sample of firms for which no forecasts were included in the data base for years n but for which forecasts were included for the prior years (years n91). Actual earnings performance in year n was used as a surrogate for management's expectations for year n for both groups of firms. This was done to avoid any bias that might result from using the 122 forecasts for year n to represent the expectations of one group while using actual earnings in year n to represent the expectations of then: other group. The objective was to determine whether there is an asso- ciation between the favorable or unfavorable character of a firm's earnings expectations and the issuance of a forecast. If the firms for which no forecast was included for year n were found to have a signifi- cantly larger frequency of unfavorable earnings trends than the firms for which forecasts were included in the data base for year n, an association, though not causality, would be demonstrated. The require- ment that each firm had issued a forecast for year n91 addresses the examination to factors associated with the decision to withdraw from the forecast disclosure system.once having entered it. Two measures of a fimm's actual earnings trend for year n were used as surrogates for the firm's expectations for that year. The first was the change in the firm's actual EPS in year n relative to year nrl. The second was the change in the firm’s actual EPS growth rate for year n relative to year n91. This second variable is based on the idea that management might still be reluctant to disclose a fore- cast of increasing EPS if the rate of growth is less than that expected by the market based on past performance. Earnings per share for years nrl and n92 were restated on the basis of common equivalent shares outstanding at the end of year n prior to making the trend analysis. Both trend variables were converted to dichotomous measures classifying trend asjfavorable or unfavorable based on the direction of the change. This eliminates the problem of nonrmeaningful percentages which result when the signs of the earnings variables change. The sample for this analysis consisted of all firms (50) with 123 forecasts for two consecutive years in the data base (repeating firms) and a matching randomly selected sample of 50 firms for which forecasts were included in the data base for year n-l but not year :1 (withdrawing firms). One pair of forecasts was randomly selected for firms with more than one set of two consecutive forecasts in the data base. Two contingency tables were then constructed based on the classification of each firm by the favorable versus unfavorable status of each of the two surrogate earnings expectation variables for year n, and the firm's status as a repeat forecaster or withdrawer in year n. These tables are shown in Figures 9 and 10. Trend in Actual EPS Forecast Status in Year n in Year n Favorable Unfavorable Total Repeater 39 ll 50 Withdrawer _3_§ _ll 50 ' Total 72 28 100 Figure 9. Contingency Table Showing Relationship Between Fore- cast Status and Trend in Actual EPS. Trend in Actual EPS Growth Forecast Status Rate in Year n in Year 11 . Favorable Unfavorable Total ' Repeater 23 27 50 Withdrawer _l_9_ _31 50 Total 42 58 100 Figure 10. Contingency Table Showing Relationship Between Fore- cast Status and Trend in Actual EPS Growth Rate. 124 To determine whether the variables represented in the contingency tables are independent, the X2 test of independence was used to test the following null hypotheses: H031: There is no association between a firm's forecast status in year n (repeat forecaster versus withdrawer) and the trend in its actual EPS in year n. Ho32: There is no association between a firm's forecast status in year n (repeat forecaster versus withdrawer) and the trend in its actual EPS growth rate in year n. The results of the tests are summarized in Table 20. TABLE 20 RESULTS OF TESTS OF ASSOCIATION BETWEEN EARNINGS TREND AND CURRENT DISCLOSURE Null Corrected Significance Decision Hypothesis Chi-square Level Ho31 1.24 .266 Not rejected Ho32 .37 .543 Not rejected The frequency of withdrawer firms which experienced unfavorable earnings trends in year n on both trend variables was larger than the corresponding frequencies for repeater firms as can be seen in Figures 9 and 10. But Table 20 shows that the differences in frequencies were not significant in either table at the .05 level. Thus, the null hypotheses of no association were not rejected. The sample data did not support the hypothesis that unfavorable earnings trends are associated with firms' decisions to withdraw from the forecast disclosure system. Alternatively, it might be hypothesized that, while not leading to the complete suppression of forecast disclosures, 125 unfavorable expectations might be delayed or published on a less timely basis than favorable expectations. This reasoning led to the analysis reported in the next section. Association Between Forecasted EPS Trend and Horizon of Initial Forecasts The sample of 50 firms used in the preceding analysis for which forecasts were included in the data base for two consecutive years was also used in the analysis reported in this section. For each of the sample firms, the horizon of the initial forecast issued by the firm in year n (the second consecutive year for which a forecast for the firm was included in the data base) was correlated with two earnings expecta- tions trend variables. For this analysis, actual rather than surrogate earnings expec- tations were available. The first variable was the forecasted change in EPS for year n relative to year n91. The second variable was the implicit forecasted EPS growth rate for year :1 relative to year n91. Forecasted EPS for year n and actual EPS for years n-1 and n-2 were restated on the basis of common equivalent shares outstanding at the end of year n. Both of the continuous earnings expectations trend variables were converted to dichotomous measures with the categories favorable and unfavorable representing the direction of change in the forecasted variable relative to the prior year. To examine the relationship between the forecasted EPS trend variables and the horizon of initial forecasts, biserial correlation coefficients which were used to measure association between one dichotomous measure with an underlying normal distribution (the fore- casted EPS variables) and one ratio measure (horizon) were computed. 126 The computational formula used for the biserial correlation coeffi- cients was: his -—--—- - where: X1 and in I the mean horizons of initial forecasts classi- fied as favorable and unfavorable, respectively, 3x I the standard deviation of all n horizons, nl and no I the numbers of initial forecasts classified as favorable and unfavorable, respectively, n I n ‘+ no, and l u I the ordinate of the unit normal distribution at the point above which lies 100(n1/n) percent of the area under the curve. The biserial correlation coefficients were then tested for sig- nificance to provide a test of the following null hypotheses: H033: The population biserial correlation coefficient for the relationship between horizon and forecasted trend in EPS is equal to zero. H034: The population biserial correlation coefficient for the relationship between horizon and forecasted trend in the EPS growth rate is equal to zero. The test statistic used was: rbis orbis z a which, if the population value of r s is zero, is approximately normally bi distributed with mean 0 and standard deviation 1. The results of this analysis are summarized in Table 21 which shows that there was no sig- nificant association between either forecasted EPS trend variable and the horizon of initial forecasts. 127 TABLE 21 RESULTS OF TESTS OF ASSOCIATION BETWEEN FORECASTED EARNINGS TREND AND HORIZON OF INITIAL FORECASTS Null - E r Computed Critical Hypothesis x1 0 bis 2 2 Decision 3°33zrbis I 0 187 175 .046 .202 £1.96 Not rejected Ho34:rbis I 0 155 215 -.279 -l.577 £1.96 Not rejected Association Between Direction of Error in Initial Forecast and Disclosure of Revision Further insight into managements' forecast disclosure practices can be obtained by studying past practices with respect to the issuance of revisions. A question of interest is whether comparable treatment is given to favorable versus unfavorable changes in expectations. This section reports the results of an examination into the frequency with 'which upward (favorable) versus downward (unfavorable) revisions occur compared to the relative frequencies with which we would expect them to occur. The next section reports the results of an investigation into the comparability of the timeliness of the issuance of upward and down- ward revisions. The analysis of the frequency of revisions is based on the notion that if comparable treatment is afforded both upward and downward revi- sions, the proportion of upward revisions to all revisions and the pro- portion of downward revisions to all revisions should correspond to the proportions of underpredictions and overpredictions, respectively, in the population of initial forecasts. To determine whether the disclo- sure practices of the firms represented by revisions in the data base 128 conform to this expectation, a sample was compiled consisting of one set of initial and revised forecasts for each firm which issued a revi- sion. If the data base included multiple revisions for a given firm, the first revision and associated initial forecast for a year randomly selected from.among those in which the firm issued revisions was included in the sample. This resulted in a sample of 32 sets of initial and revised forecasts. Each revision was then classified as either upward or downward based on the criteria displayed in Table 22. The minimum estimates in openerange and closed-range forecasts and the point estimates in point forecasts were treated as common reference points based on the finding in the analysis of reliability that there is no significant difference among formats in the bias or degree of conservatism reflected in these estimates. Format combinations not represented in the table did not occur in the sample. Based on these criteria, 16 of the 32 revisions were classified as downward and 16 as upward. TABLE 22 CRITERIA FOR CLASSIFYING REVISIONS AS DOWNWARD 0R UPWARD Revised Forecast (F2) intizit Minimumestimates (Fe; Point Open-range Closed-range l Upward Downward Upward Downward Upward Downward Point F2>F1 F2F1 szFl F2F1 F2F1 F2