AN EMPIRICAL EXAMINATION OF PUBLISHED PREDICTIONS 0F FUTURE EARNINGS Thesis for the Degree of Ph. D. MICHIGAN STATE UNIVERSITY CHARLES LeROY MCDONALD 1972 ' LI B R A R Y Michigan State University This is to certify that the thesis entitled AN EMPIRICAL EXAMINATION 0F PUBLISHED PREDICTIONS 0F FUTURE EARNINGS presented by Charles LeRoy McDonald has been accepted towards fulfillment of the requirements for Ph . D Accounting - Bus ines s Jegree in 2mm... Major professor 7/25/72 Date 0-7639 “ amome av '5 IIflAB & SONS' M9! BINDERY IIIII. UP“ amoras 533““ ' " mm “mu ABSTRACT AN EMPIRICAL EXAMINATION OF PUBLISHED PREDICTIONS OF FUTURE EARNINGS by Charles LeRoy McDonald The purpose of this research effort was to examine the reliability of published predictions of future earnings and to examine the statistical association between earnings predictions errors and several variables. These variables include factors that are both exogenous and endogenous to the firms making the earnings predictions. In the literature of accounting it is suggested that financial statements would be more useful to financial state- ment readers if they contained predictions of future earn- ings. If the predictions are to be useful, however, they must be sufficiently reliable to allow decision makers to use the predictions in the decision making processes. The first step of the examination was to acquire published earnings predictions. The wall Street Journal for the years 1966 through 1970 served as the source of the 201 published earnings predictions composing the subpopulation for the study. These earnings predictions were point pre- dictions and were within 120 days after the end of the firm's Charles LeRoy McDonald previous fiscal year. The next step was to examine the reliability of the predictions by comparing predicted earnings to actual earn- ings. The prediction errors were computed relative to pre- dicted earnings to allow comparisons among the predictions. Three findings were reached from the first portion of the research. First, for the five-year period the Relative Pre- diction Errors ranged from an overprediction of 395.6 per- cent to an underprediction of 108.5 percent. The mean pre- diction error was a 13.6 percent overprediction. Of the 201 predictions, 35.3 percent had an error of 5 percent or less, and 48.8 percent had an error of 10 percent or less. The second finding from the first portion of the research was that there was a tendency for overpredictions of earnings. The subp0pu1ation consisted of 63.7 percent overpredictions, 33.8 percent underpredictions, and 2.5 percent exact predictions. This tendency for overpredictions was confirmed by a chi-square one-sample test. The final finding from the first portion of the research was that, as expected, when examined by general industry groupings, the predictions of utilities were the most reliable with a mean Relative Prediction Error of .6 percent overprediction. The second portion of the research consisted of examining the association between Relative Prediction Errors and several variables. The association was examined on a Charles LeRoy McDonald univariate basis using simple correlation analysis, and on a multivariate basis using multiple regression. Three con- clusions were drawn from the association analysis. First, of those variables exogenous and endogenous to the firms making the predictions, only the endogenous variables were significant at the .05 level. The second conclusion resulted from the fact that the second highest association was between Relative Predic- tion Errors and Change in Operating Earnings. This sug- gested that, in general, the earnings predictions of firms consist of small adjustments to the previous year's actual earnings. The final conclusion of the second portion of the research was based on the fact that the highest association was between Relative Prediction Errors and Relative Extra- ordinary Gains and Losses. This fact presented two possi- bilities. First, extraordinary gains and losses were un- expected by the managements of the firms making predictions; or second, the earnings predictions in the suprpulation were actually predictions of Operating earnings and not net earnings. AN EMPIRICAL EXAMINATION OF PUBLISHED PREDICTIONS OF FUTURE EARNINGS BY Charles LeRoy McDonald A THESIS Submitted to Michigan State University in partial fulfillment of the requirements for the degree of DOCTOR OF PHILOSOPHY Department of Accounting and Financial Administration 1972 © Capyright by CHARLES LeROY MCDONALD 1972 ACKNOWLEDGMENTS I wish to express my sincerest appreciation to the members of my dissertation committee, Drs. Richard J. Lewis, Melvin C. O'Connor, and Roland F. Salmonson, Chairman. Their aid and guidance was invaluable in the completion of this dissertation. To James Don Edwards and Gardner Jones, who served as Chairmen of the Department of Accounting and Financial Administration while I was in the doctoral program, I ex- press my gratitude for their guidance and assistance. ‘ Two of my fellow doctoral students I would like to especially thank, William J. Morris for planting in my mind the idea for this dissertation, and Larry R. Lang for inter- minable and invaluable discussions of my ideas. I am extremely grateful for the financial aid that I received throughout the doctoral program. My gratitude is extended to the American Accounting Association, Haskins and Sells Foundation, and the Department of Accounting and Financial Administration, Michigan State University for their aid. A special note of thanks to my typist, Jo McKenzie, not only for her expertise in preparing the manuscript, but in tolerating my anxiety as completion of this dissertation iii approached. Finally, to my wife, Kathy, for her aid, encourage- ment, and love throughout this project, I will be forever thankful. iv TABLE OF CONTENTS Page ACMOWIIEIDGMEN'I‘S O O 0 O O O 0 O O O O O O O O O O O 0 iii LIST OF TABLES O O O O O O O O O O O O O O O O O O 0 Vi i LIST OF P IGURES O O O O O O O O O I O O O O O O O O 0 ix Chapter I 0 INTRODUCTION 0 I O O I O O O O O O O O O O O 1 Purpose of the Research . . . . . . . . . l Informational Needs of Investors. . . . . 3 Content of Financial Statements . . . . . 8 The Approach and Organization Of we Study 0 O O I C O O O O O O O O O 1 5 II. PREVIOUS RESEARCH CONCERNING THE REPORTING OF PREDICTIONS OF FUTURE EARNINGS. . . . . 18 Survey of Literature Calling for the Reporting of Predictions of Future Earnings . . . . . . . . . . l9 Arguments Against Publishing Predictions. . . . . . . . . . . . . . . 21 Arguments For Publishing Predictions. . . 24 The Role of the Public Accountant . . . . 28 Empirical Studies of the Reliability of Predicted Earnings. . . . . . . . . . 32 Summary . . . . . . . . . . . . . . . . . 38 III. RESEARCH QUESTIONS, DATA COMPOSITION AND FINDINGS OF THE DESCRIPTIVE STUDY. . . 40 Research Questions. . . . . . . . . . . . 40 Data Source . . . . . . . 42 Description of the Subp0pulation. . . . . 44 Study Period. . . . . . . . . . . . . . . 48 The Subp0pulation . . . . . . . . . . . . 49 Limitations Introduced by the Data Composition . . . . . . . . . . 53 The Measurement of Prediction Error . . . 54 Findings of the Descriptive Study . . . . 56 V Chapter Page Findings of the Descriptive Study by General Industry Groupings. . . . . . 71 Findings After Removal of Outliers. . . . 71 smry I O O O O O O O O I O O O O 0 O I 74 IV. ANALYSIS OF PUBLISHED PREDICTIONS OF FUTURE WIMS O O O O I O O O O O O O 76 Variables Included in the Analysis. . . . 76 Exogenous Variables. . . . . . . . . . 77 Endogenous Variables . . . . . . 80 Subp0pulation for the Multivariate Analysis . . . . . . . . . . . . . . . . 84 Univariate Analysis . . . . . . . . . . . 85 Multivariate Analysis . . . . . . . . . . 87 Results of the Multivariate Analysis. . . 90 Analysis of Residuals . . . . . . . . . . 92 Reformulation with an Additional Independent Variable . . . . . . . . . . 93 Analysis of Outliers. . . . . . . . . . . 96 Summary . . . . . . . . . . . . . . . . . 100 V. SUMMARY, CONCLUSIONS AND RECOMMENDATIONS. . 102 Summary of Results. . . . . . . . . . . . 102 Conclusions . . . . . . . . . . . . . . . 109 Conclusions from the Descriptive Study . . . . . . . . . . . . . . . . 109 Conclusions from the Association AnaIYSis O O O O O O O O I O O O O O O 111 Recommendations . . . . . . . . . . . . . 113 APPENDIX 0 O O O O O O O O O O O O 0 O O 0 O O O O O 117 SELECTED BIBLIOGRAPHY. . . . . . . . . . . . . . . . 119 vi Table 10 11 12 LIST OF TABLES Results of Predictions of Future Values of Two Income Measures from Past Values of the Same Series--Smcothing Models, 1961-1965 . . . . . . . . . . . . . . . . Results of Predictions of Future Values of Two Income Measures from Past Values of the Same Series--Regression Model, 1961-1965 . . . . . . . . . . . . . . . . Results of Predictions of Future Values of Accounting Income from Past Values of Accounting Income and of Current Income—- Regression Model, 1961-1965 . . . . . . . Earnings Forecasts in wall Street Journal Index, 1963 and 1964. O O O O O O O I O 0 Summary of Correlation Relationships Using Forecasting Accuracy--Net Income as the Dependent Variable . . . . . . . . Number of Predictions of Future Earnings Composing the Subp0pulation . . . . . . . Firms Whose Predictions Compose the Sub- p0pulation Classified into General Industry Groupings. . . . . . . . . . . . Size as Measured by Total Revenue for Firms Whose Predictions Compose the subpopulation O O O O O O O O O O O O O 0 Size as Measured by Total Assets for Firms Whose Predictions Compose the subwpulation O O O O O O O O O O O O O 0 Statistical Results of Five-Year Descriptive Study . . . . . . . . . . . . Distribution of Overpredictions, Under- predictions and Exact Predictions by Year 8 O O O O O O O O O O O O O O O O 0 Results of Chi-Square One-Sample Test of Relative Frequency of Underpredictions and Overpredictions . . . . . . . . . . . vii Page 12 13 13 35 37 49 51 51 52 65 66 68 Table Page 13 Results of One-Sample Runs Test of Randomness of Occurrences of Over- predictions and Underpredictions. . . . . . 70 14 Findings of Descriptive Study by General Industry Groupings for the Five-Years o o o o o o o o o o o o o o o 7 2 15 Findings by Years of Descriptive Study After Removal of Outliers . . . . . . . . . 73 16 Results of Correlation Analysis Between Relative Prediction Errors and Several Variables . . . . . . . . . . . . . 86 17 Level of Significance of Independent Variables in the Initial Model After Deletions I O O I O O O O O I O O O O O O O 92 18 Level of Significance of Independent Variables in the Reformulated Model After Deletions O I O O O O O O O O O O O O 96 19 Correlation Coefficients Between Relative Prediction Errors and Several Variables After Removing Outliers . . . . . . . . . . 97 20 Level of Significance of Independent Variables in the Reformulated Model with Outliers Removed and After Deletion. . . . . . . . . . . . . . . . . . 99 viii Figure LIST OF FIGURES Page Frequency Distribution of Relative Prediction Errors for 1966. . . . . . . . . 59 Frequency Distribution of Relative Prediction Errors for 1967. . . . . . . . . 60 Frequency Distribution of Relative Prediction Errors for 1968. . . . . . . . . 61 Frequency Distribution of Relative Prediction Errors for 1969. . . . . . . . . 62 Frequency Distribution of Relative Prediction Errors for 1970. . . . . . . . . 63 Frequency Distribution of Relative Prediction Errors for Five-year Period, 1966-1970 0 O O O O I O O O O O O O 64 ix CHAPTER I INTRODUCTION Presented in this chapter is a statement of the purpose of the research reported on in this dissertation, an examination of the underlying problem that acted as motivation for the research, and finally the approach and organization of this dissertation. Purpose of the Research A prevailing criterion for judging the standards of accounting is "usefulness" to the financial statement readers. For example, the Committee to Prepare A Statement of Basic Accounting Theory of the American Accounting Asso- ciation said, in setting standards to assess accounting in- formation and to establish guidelines for communicating accounting information, "...the all-inclusive criterion is the usefulness of the information... Usefulness is neces- l sarily determined through the eyes of the user..." The lAmerican Accounting Association, A Statement of Basic Accounting Theory (Evanston, Illinois: American Accounting Association, 1966), p. 3. 1 2 Committee, however, recognized that financial statements currently are not fully meeting the criterion of useful- ness. Evidence of dissatisfaction with extant ac- counting practices abounds. A principal criti- cism relates to the deficiencies of historical cost as a basis of predicting future earnings, solvency, or overall managerial effectiveness. The problem of accounting information being useful for predicting future earnings has also been examined by various authors as reported in the aCcounting literature. Their specific recommendations will be examined in detail in Chapter II. But, in genera1,they recommended that pre- dictions of future earnings be included in published finan- cial statements. The purposes of this dissertation are to examine the reliability of published predictions of future earnings and to try to determine the association between a firm's prediction errors and several exogenous and endoge- nous variables. Following is a discussion of the basic underlying problem that served as motivation for the dissertation. This problem is the apparent divergence between the infor- mational needs of investors and the present content of financial statements. 1Ibid., p. 19. 3 Informational Needs of Investors In the literature of finance, most investor be- havior models are of the following general form.1 n E(DPSi) PV = Z -—-—-——* where: PVO = present value of a share of stock at the beginning of time period 1, DPS- = dividends per share in time period 1, E( ) n expected value, time horizon (infinite in this model), ri = discount rate in time period 1. This model states that the present value of a security is the present value of the expected future dividend stream. An alternative specification of the general model is as follows: E[EPSi(1-bi)] 1 (1+ri)1 5 PV = o i: where: PV0 present value of a security at the beginning of time period 1, EPSi = earnings per share in time period 1, E [] = expected value, l—bi = dividend payout rate in time period 1, r- = discount rate in time period 1, n = time horizon (infinite in this model). 1See, for example, Ezra Solomon, The Theory_of Financial Management (New Ybrk: Columbia University Press, 4 The alternative specification follows from the fact that dividends per share are equal to earnings per share multiplied by the dividend payout rate. It is important to notice that, at least in part, the reliability of the investor behavior model depends upon the reliability of the investor's expectations concerning future earnings. The relationship between future dividends and future earnings is also discussed by Latane and Tuttle.1 To speak quite literally, the only stream of payments actually to flow into the investor's hands while he owns a common stock is the dividend stream. If the investor knew that he had to hold a stock forever, the dividend stream is the only estimate he would want to make before purchasing the stock. Latane and Tuttle next discuss the fact that various companies have different payout rates and different pros- pects for rates of return on earnings retained for reinvest- ment. It then follows that: Since dividend payout rate is so variable and non- comparable, and since dividends seem to be related to the more comparable earnings figures, we shall concentrate on earnings, because a growing earnings trend must eventually result in an in- 2 creased stream of dividend payments to the investor. The above excerpts from the literature of finance indicate that, at least in theory, knowledge of future earnings is . 1Henry A. Latane and Donald L. Tuttle, Securit Analys1s and ortfol1o Management (New YOrk: e ona d Press Company, 1970), pp. 277—278. 2Ibid., p. 278. 5 useful to investors. Even though the only flows realized by the investor are dividends and a terminal cash realiza- tion, it can be assumed that the investor has a transforma- tion function from expected earnings to expected dividends. Therefore, expected earnings acts as a surrogate for the real item of interest, expected cash flows.1 Empirical studies of the information content of earnings figures have yielded results consistent with the theoretical construct that investors desire knowledge of earnings.2 William BeaVer commenting on the results of his study measuring stock price movements for 506 earnings announcements stated: The above normal price activity is what would be expected if changes in equilibrium prices are more likely to occur when earnings reports were See Lawrence Revsine, "Predictive Ability, Market Prices, and Operating Flows," The Accounting Review, Volume XLVI, No. 3 (July, 1971), pp. 482-483. 2See, for example, William H. Beaver, "The Infor- mation Content of Annual Earnings Announcements," Empirical Research in Accounting;, Selected Studies, 1968, a supple— ment to Volume VI of the JOurnal of Accountinngesearch, pp. 67—92; Ray Ball and Philip Brown, "An Empirical Evalu- ation of Accounting Income Numbers," Journal of Accounting Research, Vol. VI, No. 2 (Autumn, 1968), pp. 159-178: Robert G. May, The Influence of Quarterly Earnings Announce- ments on Investor‘ Decisions As Reflected in Common Stock Price Changes (unpublished Ph.D. dissertation, Department of Accounting and Financial Administration, Michigan State University, 1970). 6 released, and hence the evidence is very consistent with earnings reports possessing informational value. In their study, Ball and Brown arrived at a similar conclusion. Of all the information about an individual firm which becomes available during a year, one—half or more is captured in that year's income number. Its content is therefore considerable.2 In his study of the information content of quarterly earnings announcements, Robert G. May stated: A conclusion that there is significant demand for quarterly accounting data to be used by in- vestors in actual decisions seems to be justified..., i.e. that price changes in the weeks of quarterly earnings announcements are significantly greater than average price changes. As the above studies illustrate, investor reaction to earnings announcements can be measured in stock price changes. The figure to which they are reacting, however, 4 Reported earnings is a residual of the is an artifact. recent Operating history of the firm. Since stock market prices do indeed react to earnings announcements, a rational motive for the prediction of future earnings appears. If lBeaver, p. 81. 2Ball and Brown, p. 176. 3May, p. 149. 4Lawrence Revsine, pp. 480—484. 7 an investor is able to predict future earnings, he will be better able to predict future stock prices since stock prices react to earnings that actually materialize. Thus far it has been pointed out that earnings num- bers are useful to investors as described in the literature of finance and in empirical studies Of stock market price reactions to earnings announcements. This idea of earnings figures having utility to investors is also supported by practitioners in the investments field. Rex J. Bates com- menting on Beaver's research stated: 'Reported earnings is the name Of our game!‘ I entered this securities business upon leaving the university twenty years ago, and while I have not verified the observation statistically, I am firmly Of the Opinion that stock prices have become increasingly sensitive to reported earnings.2 This section presented support for the assumption that investors should desire knowledge Of future earnings. This support came from the theoretical construct Of an investor behavior model, from the results of empirical research, and finally from a professional in the securities business. 1Ibid., p. 485. 2Rex J. Bates, "Discussion Of The Information Con— tent Of Annual Earnings Announcements," Empirical Research in Accounting: Selected Studies, 1968, a supplement to Volume VI of the Journal of Accountinngesearch, p. 93. 8 Content Of Financial Statements The previous section established that one group Of financial statement users, investors, should desire know- ledge Of future earnings. An interesting comparison is the content of financial statements with the apparent informa- tional needs of investors. In their current form, most financial statements consist Of only historical data. The balance sheet purports to present the financial position Of the entity at a point in time, which is usually a number of months before the statement is publicly available. Many users Of financial statements dispute the claim that the balance sheet is an adequate representation of the financial position because Of accountants' use Of historical cost in the preparation Of financial statements. The income state- ment is also historical in orientation. It presents the results of Operations for some past time period. Since in- vestors need information concerning future earnings and financial statements typically provide users with historical data, there appears to be a divergence between the two. The accounting profession recognized this diver- gence and addressed the situation by stating that the his- torical earnings presented in the financial statements should serve as a basis for predicting future earnings. The Committee to Prepare A Statement of Basic Accounting 9 Theory Of the American Accounting Association had the following to say about the situation. Almost all external users Of financial infor- mation reported by a profit-oriented firm are in— volved in efforts to predict the earnings of the firm for some future period. Such predictions are most crucial in the case Of present and pro- spective equity investors and their representatives—— considered by many tO be the most important of the user group... The past earnings Of the firm are considered to be the most important single item Of information relevant to the prediction Of future earnings. It follows from this that past earnings should be measured and disclosed in such a manner as to give a user as much aid as practic- able in efforts to make this prediction with a minimum of uncertainty.1 Eldon S. Hendriksen Observed the situation in much the same manner as did the Committee to Prepare a Statement Of Basic Accounting Theory. ...most of the decisions of creditors and investors, including the stockholders of large corporations, require a prediction of the future distributions by the firm... If there is a relationship between reported income and dividend distributions, in- vestors may focus their attention on their expec- tations regarding future incomes of the firm... How can a knowledge of past income aid in the prediction Of future incomes and thus in the cur- rent value Of the firm? ...there is some validity in the Observation that all economic changes, and most physical changes, for that matter, are de endent in large part upon past conditions and events. 1American Accounting Association, pp. 23—24. 2Eldon S. Hendriksen, Accounting Theory (Home- ‘wood, Illinois: Richard D. Irwin, Inc., 1970), pp. 128- 129. 10 The Committee to Prepare a Statement of Basic Accounting Theory in addition to stating that accounting data should serve as the basis for prediction, recommended that multi-valued information be presented in the financial statements. These financial statements would contain infor- mation based on both historical cost and current cost.1 The justification for including the data based on current cost was as follows: Current values, on the other hand, reflect not only the transmfiions of the firm but also the impact Of the environment on the firm beyond the completed transaction. Thus they possess a high degree of relevance for many uses in which prediction is prominent. In an empirical study,3 Werner Frank attempted to examine the contention that historical-cost data can serve as a basis for predicting future earnings. In addition, Frank tested the contention that current-cost data provide more relevant information for purposes Of prediction. Frank's methodology consisted first Of determining the errors resulting from the use Of historical-cost Operating income to predict succeeding years' historical-cost 1American Accounting Association, p. 32. 21bid., p. 30. 3Werner Frank, "A Study of the Predictive Signifi- cance Of Two Income Measures," JOurnal of Accounting :Research, Volume VII, NO. 1 (Spring, 1969), pp. 123—136. 11 Operating income and current—cost Operating income to predict succeeding years' current-cost Operating income. The pre- diction errors were measured as a percentage Of the actual earnings number. Of interest was which Of the two measures better predicted its own future values. Frank's reasoning was as follows: If the predictability Of an income concept is of primary concern, the errors which result from attempting to forecast future values of in- come from past values Of that same income series would be of interest.1 !mm‘ In addition, Frank's methodology used both current- cost Operating income and historical-cost Operating income to predict succeeding years' historical Operating income. Frank's reasoning for the second test was as follows: ...another approach (in addition to the first test) is needed if current-cost income is to be regarded only as possible supplementary information, in- cluded tO enhance the accuracy Of predictions Of the primary accounting income figure. The relative accuracy Of forecasts of accounting income based on past values Of accounting income should then be com- pared with forecasts Of those same values but based on past values Of current-cost income. In order tO test the predictive ability of the two income measures, Frank used two different types Of time series forecasting models, a simple least-squares regression lIbid., p. 131. 21bid., p. 125. 12 model and a series Of moving—average or smoothing models. There probably exist a multitude Of forecasting models em— ployed by investors. Therefore, any conclusions about the predictive ability Of the two income numbers must be limited to those predictions resulting from the forecasting models used. Quite possibly some investors have more reliable, but publicly unknown, forecasting models. The results of Frank's study are presented in the following three tables. TABLE 1--Results of Predictions of Future Values of Two Income Measures from Past Values of the Same Series-~Smothing Models, 1961-1965 Average of percentage Industry Nugber of Number of forecast errors irms forecasts Acctg. Current income income Paper 13 65 9.52 10.4% Chemicals 16 80 9.7 10.2 Drugs 10 50 8.6 12.1 Cement 6 30 9.9 10.9 011 14 70 10.4 13.6 Steel 17 85 21.8 28.1 Source: Werner Frank, "A Study of the Predictive Significance of Two Income Measures," Journal of AccountingiResearch, Volume VII, No. 1 (Spring, 1969). pp. 130-131. 13 TABLE 2--Results of Predictions of Future Values of Two Income Measures from Past Values of the Same Series-- Regression Model, 1961-1965 Averages of percentage Industry Number of Number of forecast errors firms forecasts Acctg. Current income income Paper 13 65 10.2% 11.6% Chemicals 16 80 12.3 11.6 Drugs 10 50 9.5 10.2 Cement 6 30 10.9 11.6 Oil 14 70 16.6 19.4 Steel 17 85 27.9 38.9 Source: Werner Frank, "A Study of the Predictive Significance of Two Income Measures," Journal of AccountingLResearch, Volume VII, No. 1 (Spring 1969). pp. 130-131. TABLE 3—-Results of Predictions of Future Values of Accounting Income from Past Values of Accounting Income and of Current Income—~Regression MOdel, 1961-1965 Averages of percentage Industr Number of Number of forecast errors y firms forecasts Acctg. Current income income Paper 13 65 10.2% 10.4% Chemicals 16 80 12.3 11.1 Drugs 10 50 9.5 9.9 Cement 6 30 10.9 13.6 Oil 14 70 16.6 16.5 Steel 17 85 27.9 37.9 Source: Werner Frank, "A Study of the Predictive Significance of Two Income Measures," Journal of Accountinngesearch, Volume VII, No. 1 (Spring, 1969), pp. 131. 14 Concerning the data presented in Tables 1-3, Frank con- cluded the following: The data summarized in Tables 1—3 indicate that no clear advantage seems to exist for re- porting current-cost income rather than account- ing income in the industries studied, given the hypothesis that better predictions of current income can be made from past values Of current income. This appears to be the case both for forecasts made from smoothing models and from the regression model. Of course, better predic- tions of current income might have resulted had the forecasts been made for a period in which price level fluctuations were greater, or had other industries, other current-cost adjustment procedures, or other income measures been used. Frank's conclusions are not consistent with the contention of the Committee to Prepare a Statement of Basic Accounting Theory that current—cost data should be provided. His re- sults suggested that current-cost data did not improve the forecast errors are also of interest. The smallest error encountered is 8.6 percent and the errors range up to 38.9 percent. Whether these forecasts are reliable enough to be useful is questionable. Whether the forecasts generated in Frank's study are reliable enough to be useful is questionable because of a general lack of knowledge Of investors' materiality func- tion. In addition, each investor probably has a different 1Ibid., p. 133. * 15 idea of what is material and what is immaterial. An empiri- cal study reported by Bernsteinl found that 10 percent to 15 percent of the past five years' average income was the "border zone" dividing material items from immaterial items. This border zone was found in financial statements and re- flects what accountants consider to be material. The in- vestor's concept of materiality is not necessarily the same. Since the suggestion of the Committee to Prepare a Statement of Basic Accounting Theory to include current- cost income for better predictions has not yet been sup— ported by empirical evidence, it seems worthwhile to in— vestigate other suggestions. As previously mentioned, Chapter II presents suggestions by various authors that management's predictions of future earnings be included in the financial statements. This dissertation will attempt to investigate those suggestions. The Approach and Organization Of the Study The approach Of the study was to obtain published predictions Of future earnings and then to compare these predictions with actual earnings results to compute the 1LeOpold A. Bernstein, "The Concept of Materi- ality," The Accounting Review, Volume XLII, No. l 16 prediction errors committed in predicting future earnings. The results of this part of the study are summarized by the presentation of several descriptive statistics. Next, an attempt was made to determine the association between several exoqenous and endogenous variables and the firm's prediction error in both a univariate and a multivariate analysis. As previously mentioned, Chapter II contains a re— view Of the literature concerning the inclusion Of future earnings in the published financial statements. Also considered are the implications for the public accountant and the current situation in England concerning predicted earnings. Finally, included in Chapter II is a re- view Of a previous empirical studies of the reliability of predictions of earnings. Chapter III contains a detailed description of the data selection and a statement of the two research ques- tions. Following is a presentation of the methodology employed for the first research question. Finally the research findings for the first research question are presented. Chapter IV contains a discussion of the methodology for examining the second research question and the research findings. The research findings are in terms Of both univariate and multivariate analyses. 17 Finally, Chapter V contains a summary of the results of the study, the conclusions of the study drawn from those results, and some recommendations implied by the results. CHAPTER II PREVIOUS RESEARCH CONCERNING THE REPORTING OF PREDICTIONS OF FUTURE EARNINGS Research to date concerning the reporting Of pre— dictions of future earnings can be classified into the following general categories: 1. Works calling for the inclusion of predictions of future earnings in the published financial statements; suggested formats for their pre- sentation; and discussion of arguments, both in support Of and against, the reporting of predicted future earnings. Surveys of the situation in England where predictions of future earnings are required in some circumstances. Empirical examinations of the reliability of predictions of future earnings. The works falling into the first two categories are simply summarized. However, those studies falling into the third category are discussed, as well as evaluated, in much 18 pa Su ha I'IE fi ir. 19 greater detail because of their similarity with certain parts Of the research reported in this dissertation. Survey Of Literature Calling for the Reporting of Predictions of Future Earnings Several studies found in the accounting literature have dealt with the divergence between the informational need of investors, i.e. future earnings, and the content of financial statements, i.e. historical data.1 Using the same framework as presented in Chapter I, these authors argue that investors desire knowledge of future earnings and that financial statements should contain managements' predictions of future earnings. By providing these predic— tions, financial statements would become more useful to investors. The authors point out that little additional effort on the part Of the firms would be required to include their predictions in the financial statements. Since most firms currently use budgets for planning and control, predicted 1For example see W. W. COOper, N. DOpuch, and T. F. Keller, "Budgeting Disclosure and Other Suggestions for Improving Accounting Reports," The Accountinngeview, Volume XLIII, NO. 4 (October, 1968), pp. 640—647; Rudy Schattke, "Expected Income--A Reporting Challenge," Th3 Accounting Review, Volume XXXVII, No. 4 (October, 1962); pp. 670-676: and T. R. Wilkinson and L. D. Doney, "Extend— ing Audit and Reporting Boundaries," The Accountinngeview, Volume XL, No. 4 (October, 1965), pp. 753-756. 20 figures already exist internally. It is assumed that managements attempt to forecast as accurately as possible for budgeting purposes since production schedules, size of the labor force, capital asset acquisitions and other in- ternal decisions are based on the expectations contained in the budgets. Since business firms Operate in a world of uncer- tainty, it seems likely that expectations concerning future earnings will not be realized exactly. Therefore, it would be useful if the financial statements contained a compari- son Of predicted earnings and actual earnings for the im- mediately preceding period. In addition,management could explain any differences in textual comments. These explan- ations would allow the reader to evaluate how much of the prediction error is a result of management not adjusting to the unexpected. Furthermore, the statements might con- tain comparisons of the past several periods' predicted earnings and actual results for those same periods. This format would allow the reader to Observe the reliability of past predictions of the firm and decide how much credi- bility to attach to the firm's current prediction. Of course, this assumes a firm that has made reliable predic- tions in the past will continue to make reliable predictions in the future. 21 Finally, the prediction for the next period's earn— ings would be included in the financial statements. Ideally, the period covered by the prediction should be the normal reporting period for the firm. This would allow the previously mentioned comparisons of predicted earnings and actual earnings to be made conveniently. There are two alternatives for the presentation Of the prediction. The prediction could be in the form of a point estimate, or it could take the form Of a range with a probability distribu- tion. Regardless of which form is used, management would be expected to state the assumptions about the economic and political environment upon which the prediction is based. Schattke suggested that possibly multiple predictions could be presented based upon Whether or not a major event, such as a strike, occurs.1 The presentation of the assumptions underlying the prediction would give the statement reader more information to evaluate the credibility of the predic— tion, depending on the degree of agreement with the stated assumptions. Arguments Against Publishing Predictions There have been numerous arguments presented, both for and against, the publication of managements' predictions lSchattke, p. 673. 22 of future earnings. It happens that many of the arguments for publication are in fact refutations of arguments against publication. This section presents the arguments against publication, the next section considers those arguments for publication of predictions of future earnings. 1. The first argument against the publication of predictions of future earnings is that some firms would intentionally overestimate future earnings. Presumably, the purpose for such behavior would be to put management in a favorable position with the stockholders at the time of the publication Of the prediction. It seems that the strongest tendency for such behavior would be when Operations have been going badly and management is being criticized. On the other hand, it can be argued that some firms may intentionally underestimate future earnings. There are two possible explanations for such behavior. One possibility is that businessmen tend to be conservative and they want to avoid the situation of failing to earn as much as had previously been predicted. Such behavior would be the result of management feeling that falling short Of the predicted 23 level of earnings would subject them to more criticism than continually earning more than predicted. The other explanation for the intentional underestimation Of earnings is the hope that investors will credit the greater than expected earnings to managerial ability. Another argument against publishing predicted earnings is that such activity would release too much information to competitors. Firms fear that competitors would learn of plans to enter new markets, major product innovations, or pricing policies, etc., and therefore be able to take competitive action and thwart their planned Operations. The argument against publishing prediction of future earnings that is Offered most frequently is that the predictions would not be reliable. If the predictions are tO be useful to investors, then they must be reliable. Unreliable infor- mation could lead to wrong decisions. Opponents argue that because of the uncertainty and fluc- tuations in the environment in which business firms Operate and the undevelOped techniques of predicting, predictions, that are sufficiently 24 reliable on which to base decisions just cannot be made. For example the Committee to Prepare a Statement of Basic Accounting Theory stated "Accountants generally refrain from reporting budgets relating to future periods of external users, on the ground that the information is not sufficiently verifiable, although it might be highly relevant to the external user's needs."1 It is the argument of unreliability that is the focus of this dissertation and the subject is addressed in greater detail in Chapter III. Arguments For Publishing Predictions 1. Those who argue for publishing predictions of future earnings explain that the prOposed format for presentation Of the predictions would in— hibit a firm from intentionally overestimating 2 future earnings. The inclusion of comparisons of actual earnings with predicted earnings 1American Accounting Association, A Statement of Basic Accounting Theory (Evanston, Illinois: American Accounting Association, 1966), p. 27. 2Wilkinson and Doney, p. 754. 25 would insure that any benefits from over— estimating would be Of short duration. Also, by including comparisons Of several years' figures in the statements, readers would have a chance to notice any tendency for a firm to overestimate earnings in past periods. In addition, if a firm overestimates and fails to reach predicted earnings, this fact is pre- sented in the statements and this comparison Of actual and predicted earnings could cast doubt on the ability Of management to predict or to contend with the unexpected or both. 2. The argument that predictions should not be pub- lished because firms would be intentionally (or even unintentionally) conservative in their estimates can be refuted by considering the alleged Objective of financial management. The literature of finance1 assumes the objective of financial management to be the maximization of shareholder wealth, a measurement which un- doubtedly involves the market price of shares. 1 For example,see Ezra Solomon, The Theory Of Finan- cial Management (New York: Columbia university Press, 1963), pp. 15-26. 26 If a firm underestimates future earnings it will tend to depress the market price of the stock and management would fail in its objec- tive of the wealth maximization of shareholders. In answer to the charge that publication of pre- dicted earnings would release too much informa— tion to competitors, it is doubtful that without the release Of detailed divisional breakdowns and information concerning new products and markets, that competitors would gain enough information to take any competitive actions. Although the publication of predicted earnings would reveal managements' goals, how they are to be accomplished would not be revealed.1 In addition to the above argument,a11 firms would be publishing their predictions and therefore all firms would be on an equal footing with their competitors as far as the releasing of information is concerned. Predictions of future earnings are currently published occasionally in the president's letter contained in the annual report. These 1Schattke, p. 674: and Wilkinson and Doney, p. 755. 27 predictions are presented informally and are not subject to any sort of verification. If the proposed format for presentation of pre- dicted earnings were adOpted, at least the predicted earnings would be presented formally. In addition, the comparisons between previously predicted earnings and actual results would provide much more information than is currently provided since management would be expected to explain differences between predicted and actual earnings. Whether or not the predicted earnings figure would be subject to verification is a matter Of concern to the public accounting pro- fession. This question will be addressed later in this chapter. Another argument for publishing predicted earn- ings arises because Of the previously mentioned fact that for most companies these predictions already exist internally. The possibility exists that influential investors could gain access to this "inside" information and enjoy an advantage in making investment decisions. If these predictions were made public informa- tion, such advantages would diminish and there 28 would be a more efficient allocation Of capital resources. The above list of arguments for and against publishing pre— dicted earnings demonstrates that, with one exception, the arguments against publication can be answered. The one ex- ception is the question of the reliability of predictions Of future earnings. The Role of the Public Accountant If predictions of future earnings are published in the financial statements, it is unknown how much respon- sibility a Certified Public Accountant will assume. The CPA is currently prohibited by rules of ethics from attest- ing to forecasts. The Code of Professional Ethics of the American Institute of Certified Public Accountants has the following to say concerning forecasts. A member or associate shall not permit his name to be used in conjunction with any forecast of the results of future transactions in a manner which may lead to the belief that the member or associate vouches for the accuracy of the fore- cast. If a CPA is to verify in any manner the predictions Of future earnings, there will have to be a change in the 1John L. Carey and William O. Doherty, Ethical Standards Of the Accounting Profession (New York: American Institute of Certified Public Accountants, 1966), p. 73. 29 accounting profession's ethical standards. An opinion from a CPA regarding the accuracy Of the predicted earnings figure is not the only possibility for the CPA. The CPA may only be called upon to give an Opinion as to the accuracy Of the explanations given by management for differences between predicted earnings and actual earn— ings. This way the CPA would continue to work with objec— tive evidence. This procedure would indirectly encourage management to give as accurate a prediction as possible. Although the prediction would not be verified currently, management would be cognizant that at the end of the period an explanation would have to be offered for differences and that explanation would be subject to verification by the CPA. Another possibility for the CPA's responsibility is that of Observing the budgeting process and calculations involved in predicting earnings and rendering an Opinion on these matters. This is currently the situation in England. The following is from the Council of the Institute of Chartered Accountants in England and Wales. In so far as it concerns reporting accountants, the City Code on Take-overs and Mergers requires (a) that where profit forecasts are included in circular, the accounting bases and calculations for them must be examined and reported on by the company's auditors or consultant accountants, and (b) that the circular must contain the accountants' report. The Code also requires the assumptions, including the commercial assumptions, upon which 30 the directors have based their profit forecasts to be stated in the document, and these must be reported on by the company's merchant bank or other adviser, if any. Accountants may be re- tained to report in the latter connection as 'other advisers' under the Code. . .1 In addition to pointing out what the City Code on Take- overs and Mergers (London) required, the Statement listed the main matters to be stated in the accountants' report. The accountants' report under the Code will be addressed to the directors and will normally include statements dealing with the following matters, so far as apprOpriate: (a) the fact that the reporting accountants have carried out a review of the account- ing bases and calculations on which the profit forecasts have been based (b) specific identification of the forecasts and documents to which the report refers (c) if, as will usually be the case, the re— porting accountants have not carried out an audit of estimated results for expired periods, a statement to that effect (d) whether in the opinion of the reporting accountants the forecasts have been prOp- erly compiled on the basis of the assump- tions made by the board of directors, as set consistent with the accounting prac- tices normally adOpted by the company. If the reporting accountants have reason for material reservations about the accounting bases and calculations for the forecasts, or if they have reason to consider them inconsistent with the stated assump- tions, they should qualify their report accordingly. 1Institute of Chartered Accountants in England and Wales, "Accountants' Reports on Profit Forecasts," 815, June 26, 1969, p. l. 21bidoa pp. 4‘5. 31 In addition to the above requirements concerning take-over bids, Hendriksen reported that in connection with the issuance of securities in England, "the auditor's re- port covering a prospectus must include, according to the requirements of the Federation of Stock Exchanges, the in- formation regarding profits for the past ten years or since inception of the company and comment on management's fore- cast Of profits for the next year.1 Whether or not the auditor's responsibility will be extended to the same degree as is currently the case in England is not determinable at the present time. It does seem quite possible that predictions of future earnings will be required in some instances by the Securities and Exchange Commission. In a speech given November 19, 1971, William J. Casey, Chairman Of the S.E.C. reported, "that the S.E.C. was leaning toward an important policy change involving the public forecasting Of corporate financial performance. ...The time has come when we should reexamine the question of projections, forecasts and appraisals in lEldon S. Hendriksen, "Disclosure--Insights Into Requirements in The United Kingdom," The International Journal Of Accounting, Volume IV, No. 2 (Spring, 1969), p. 29.- 2Terry Robards, "S.E.C. Shifts Hinted on Earnings Forecasts," The New York Times, November 19, 1971, pp. 63- 65. 32 our disclosure framework."1 Casey explained that tradition- ally the S.E.C. has not permitted predicted earnings in the financial statements. Concerning predicted earnings in prospectuses issued for new securities offerings, Casey com- mented, "It appears anomalous that projections of sales and income, deemed to be relevant for trading—market purposes, were traditionally not required in prospectuses."2 If CPAS are involved in giving Opinions on financial statements that include predictions of future earnings, it seems likely that there will have to be a reconciliation between full disclosure of relevant information and reasonable verifi- ability and Objectivity.3 Empirical Studies Of the Reliability of Predicted Earnings Three empirical studies of the reliability of pre- dicted earnings have been located. One study reports on the reliability of earnings forecasts in England. The other two studies deal with profit predictions in the United States. The results of these studies are reported in detail. As mentioned earlier, in England predictions of future earnings are required in prospectuses for new issues lIbid., p. 63. 2Ibid. 3Of prime importance, but outside the sc0pe of this dissertation, is the legal liability of CPAs in the area of predicted earnings. 33 Of securities and in proxies for mergers and take-overs. A survey was conducted by the English Panel on Take-Overs and Mergers for the purpose of analyzing the accuracy of the . . . 1 earnings pred1ct10ns. The results Of the survey were sum- marized as follows: Since April 1969, we have examined 210 fore- casts and compared them with actual results. Of these 210 we classified 170 as achieved, by which we mean that the result was within 10 per cent either way of the forecast. This is of course a very arbitrary margin; likewise the classification is unsophisticated because it does not distinguish between forecasts made in the first month and forecasts made in the twelfth month. The results Of the study have few implications for the question of the reliability of predictions included in the annual financial statements. First, the predictions were made by British firms Operating in the British economy and the reliability is not necessarily the same as that of U. S. firms. Secondly, the predictions surveyed ranged from twelve months in advance Of year-end to less than one month in advance of year-end. Certainly those predictions made more than a few months into the year could not be included in the annual report. 1 Ian J. Fraser, "Accountancy and the Merger Move- ment," The Accountant, Volume 165, NO. 5047 (September 9, 1971), pp. 353-355. 2Ibid., p. 355. 34 In trying to determine the reasons for prediction errors, the survey of British firms' predictions concluded: Overall therefore it looks as if about 17 per cent of forecasts are being missed; but half of those, and possibly three quarters, are misses which are either catered for in the stated assump— tions or covered by circumstances which are genu- inely unforeseen.1 2 As part of a larger study, Green and Seagall ex- amined twelve forecasts in the Wall Street Journal Index for 1963 and 1964. Of these twelve predictions, seven could be quantified. The results of their survey is sum- marized in Table 4. Those figures with negative signs in the second column represent predictions made during the year preceding the year for which the prediction was made. The third column explains whether the firm correctly pre- dicted an increase or decrease in earnings. The percentage error in the fourth column is computed as the difference between actual earnings and forecasted earnings as a per- centage of actual earnings. There are two lines to Company 10 because two predictions were found for that particular firm. lIbid. 2David Green, Jr. and Joel Seagall, "The Predictive Power of First Quarter Earnings Reports," Journal of Business, Volume XL, NO. 1 (January, 1967), pp. 44-55. 35 TABLE 4-—Earnings Forecasts in Wall Street Journal Index, 1963 and 1964 Percentage of Fiscal Prediction in Company Year Elapsed When Correct % Error Forecast Reported Direction 1 -17 No ___ 2 -17 Yes --— 3 -5 No 31.6 4 6 No 29.9 5 14 No --- 6 33 Yes --- 7 44 Yes 268.4 8 54 Yes 11.8 9 69 Yes —-- 10 76 Yes 3.0 86 Yes 1.0 11 78 Yes --5 12 95 No 6.6 Source: David Green, Jr. and Joel Seagall, "The Predictive Power of First-Quarter Earnings Reports," Journal of Business, Volume XL, No. 1 (January, 1967), p. 53. Few implications can be drawn from the Green and Seagall study mainly due to the restricted number of obser- vations in the survey. Also, as in the survey of British firms, some predictions were made well into the fiscal year long after the annual report was issued. 36 The final empirical study of the reliability of pre- dicted earnings examined is that of R. Austin Daily.1 Daily contacted 50 firms about participating in his study. Eight— een firms expressed interest and twelve provided useful data. Included in the data were predictions of earnings and actual results for at least five years. Daily treated these data as 64 observations. In summary Daily said, Differences of greater than 10 percent between the forecast and actual results were present for 34 of the 64 (53 percent) observations while 21 Observa- tions (33 percent) had differences exceeding 15 percent. As with the other studies, no absolute statement about whether Daily found the predictions to be reliable or un- reliable can be made without knowledge of the investors' materiality functions. Daily's study also suffers from the small number of firms involved and from the fact that Daily obtained his data directly from the firms involved. Since the predictions were unpublished and Ex pggg, Daily recog- nized the possible bias that firms having a large error in the past may have refused to participate. In his study, Daily also examined the data in an . 1R. Austin Daily, "The Feasibility of Reporting Forecasted Information," The Accountinngeview, Volume XLVI, No. 4 (October, 1971), pp. 686-692. 2Ibid.. p. 690. 37 attempt to find relationships between the prediction errors and some independent variables. Daily used correlation analysis in this part of the study and his results are pre- sented in Table 5. TABLE 5--Summary of Correlation Relationships Using Forecasting Accuracy--Net Income As the Dependent Variable Independent Variable(s) Coefficient 0f Determination 1. Annual Revenues .071 2. Forecasting Accuracy--Revenues .076 3. Annual Net Income .149 4. Annual Revenues, Forecasting Accuracy-- Revenues .116 5. Annual Revenues, Annual Net Income .149 6. Forecasting Accuracy--Revenues, Annual Net Income .192 7. Annual Revenues, Forecasting Accuracy-- Revenues, Annual Net Income .194 Source: R. Austin Daily, "The Feasibility of Reporting Forecasted Information," The Accounting Review, Volume XLVI, No. 4 (October, 1971), p. 692. 38 The independent variables used by Daily in Table 5 are as follows: Annual Revenues the size of the firm as represented by annual revenue Forecasting Accuracy-Revenues the reliability of the firm's prediction of future revenue Annual Net Income . = the size of the firm as represented by annual net income As the information in Table 5 demonstrates, Daily was unsuccessful in finding relationships between predic- tion errors and several independent variables. In the most successful analyses Daily found a coefficient of determina- tion Of only .194. Thus,the ratio of the associated vari- ance to the total variance of the dependent variable asso- ciated with the independent variable is only .194.1 Summary This chapter has presented previous work calling for the inclusion of predictions Of future earnings in the published financial statements. Since published financial statements are accompanied by the Opinion of a Certified 1Charles T. Clark and Lawrence L. Schkade, Statistical Methods for Business Decisions (Cincinnati, Ohio: South-Western Publishing Company, 1969), p. 561. 39 Public Accountant, the possible effects of financial state— ments containing predictions of future earnings on the auditor's responsibility were explored. This exploration included a survey of the current situation in England where such predictions are required in certain circumstances. Finally, the results of previous studies Of the reliability Of predictions Of future earnings were summarized. By illuminating the weaknesses of these studies, the need for the research study to be described in Chapter III hopefully is strengthened. CHAPTER III RESEARCH QUESTIONS, DATA COMPOSITION AND FINDINGS OF THE DESCRIPTIVE STUDY This chapter contains a discussion of the research questions examined by the researcher and reported in this dissertation. In as much as the central theme of the re- search is prediction of earnings, the source Of the data, earnings predictions, as well as the study period are analyzed. Finally, the findings in connection with the first research question are presented. Research Questions The two research questions examined by the researcher and reported in this dissertation are as follows: 1. How reliable are published predictions of future earnings if they are to be included in the published financial statements? 2. What variables are associated with a firm's ability to predict future earnings? As the discussion in Chapter II points out, the prOposals that predictions Of future earnings be included in the 40 41 published financial statements have not been attacked on the grounds of relevance, but their reliability has been questioned. If published earnings predictions are to be useful to financial statement readers, they must be reli- able enough so that the reader will be able tO base deci- sions on them. The requirement of reliability gives moti— vation for the first research question. The first research question suggests a: descriptive examination of earnings predictions contrasted with actual earnings results. The first part of the study consisted of this descriptive study and is described in greater detail in this chapter. After analyzing how reliable predictions Of future earnings are, it seems worthwhile to attempt to identify an association between the reliability of predictions and several variables that are both exogenous and endogenous to 'the firm. The desirability Of such an analysis serves as rmotivation for the second research question. Not only would such an analysis be worthwhile in gaining a better understanding of the prediction process, but also an under— standing Of the forces affecting prediction reliability from an 21‘. ‘Losg examination would help financial statement readers better assess the reliability of predictions from an 25 311322 position. This type of analysis served as moti- \nation for the second part Of the research which is 42 described in detail in Chapter IV. Data Source In order to examine the reliability of predictions of future earnings, it was first necessary to find a source of predictions of future earnings. In Daily's study of the reliability of predictions Of earnings described in Chapter II, the predictions were Obtained through direct correspond- ence with the firms whose predictions were analyzed.1 This source of earnings prediction is subject to bias, a fact which Daily points Out in his article,2 in that firms which experienced large errors in the past in predicting earnings were reluctant to participate. Since these predictions are Obtained SE pgsg, their accuracy is known and inaccuracies can be kept secret by management. Therefore, the sample Obtained may consist of a éfisprOportionate number Of firms that experienced accurate results in predicting earnings. In addition, the purpose of the examination was to assess reliability if the predictions were included in the pub- .lished financial statements, therefore, published predic- ‘tions would be more desirable from two vieWpOints. First, ‘the predictions would be published before actual results 1Daily, The Accounting Review, p. 688. 21bid., p. 691-692. 43 are known, therefore they are 25 EELS predictions and not subject to the previously mentioned bias. Second, it can only be assumed that unpublished predictions would be the same predictions used had management known they would be published. For these reasons it is highly desirable that the source Of predictions for the research be published data. In the study by Green and Seagall, summarized in Chapter II, the source of earnings predictions was pub— lished data.1 The source used was the Wall Street JOurnal 'Igggx. Green and Seagall only analyzed twelve predictions, however, since the published predictions were only a side comment to their analysis of the predictive power of first- quarter earnings reports. Green and Seagall's sample did meet the highly desirable requirement of being published data, but, their study did suffer from the small sample size. For the research in this dissertation, the source of predictions was similar to that used by Green and Seagall. Predictions were obtained by examining the ‘Wall Street Journal. Numerous predictions are announced, usually by company executives, and they are published in ‘Wa11.Street Journal in the form of news stories. Additional 1Green and Seagall, JOurnal of Business, p. 54. 44 requirements, described in the next section, had to be satisfied, however, before the published predictions were considered usable in the research. Description of the Subpopulation Since the research was concerned with the reliability of predictions Of earnings if they are included in the pub- lished annual financial statements, the predictions found in the Wall Street Journal must be usable for such a purpose. To be usable they must be announced early enough in the fiscal year of the firm so that they could possibly be in- cluded in the published annual financial statements. The Securities and Exchange Commission requires annual reports Of firms whose securities are listed to be issued no later than 120 days after the end of the firm's fiscal year.1 Therefore, for a prediction found in the Wall Street JOurnal to be included in the research it had to appear within 120 days after the end of the firm's fiscal year. To reduce the data gathering effort, only the January through April issues of the Wall Street JOurnal were examined to Obtain ;predictions used in the research. The January through .April issues were within 120 days of year-end of all firms 1New York Stock Exchange Guide, Volume III-~Re1ated Laws and Regulations (Commerce Clearing House, Inc., 1962) , p. 6183. 45 who use a calendar year as their fiscal year. Also, by using the January through April issues, the prediction of firms that have a fiscal year ending between September 30 and December 31 could be used if they meet the 120 days requirement. Many of the predictions of earnings found in the Wall Street Journal were released in connection with the annual meeting. Since annual reports are issued to stock- holders before the annual meeting, it was assumed that in the time lag between the issuance Of the annual report and the annual meeting the predictions of earnings did not change significantly. The assumption stated explicitly is that the predictions appearing in the Wall Street Journal are assumed to be the same as the predictions that would be included in the published financial statements. The predictions composing the suprpulation were also required to be worded so that it appeared that net income was being predicted. If the predictions referred to net income before extraordinary items, they were excluded from the suprpulation. It seems reasonable that investors should desire predictions Of net income after extraordinary items to use in their decision making process. It would be small consolation for an investor to base his decision on an increase in net income before extraordinary items and 46 have that prediction come true, only to have the price of the stock not increase because the increase in Operating earnings was offset by an extraordinary loss. To be included in the research, the predictions were also required to be in the form of point estimates. Some predictions are given in the form of ranges, but these are far outnumbered by the point predictions. An additional consideration is that the measurement of errors for range predictions would not be homogeneous with the measurement Of errors for point predictions. The problem is with the measurement Of the prediction error in connection with range predictions. One possibility is to use the differ— ence between the midpoint Of the range and the actual re— sults. This requires the researcher to assume that the probability distribution associated with the range predic- tion is symmetrical. Another possibility is to use the difference between the range and actual results as the prediction error. For example, if the prediction is $1.00-$l.25 per share and actual results are $1.30 per share, the prediction error is $.05 per share. This tech- nique is unacceptable, however, because the various firms Offer predictions whose ranges differ tremendously in relative size. This could lead to the situation of declaring one prediction more accurate than another simply 47 because a much larger range was employed. Therefore, only point predictions were used in the research. Also excluded from the predictions used in the research were those predictions that might be described as "Open-ended." These predictions are stated in terms of "at least" or "no more than," etc. The reason for excluding the Open-ended predictions is that it is difficult to mea— sure the prediction error. For example, if a firm predicts earnings to be "at least $1.00 per share," and actual earn- ings are $1.25 per share, technically the prediction error is zero. Because of these measurement problems, Open-ended predictions were excluded. Those firms that made a prediction and subsequently were acquired by another firm in the period to which the prediction applied were also excluded. These firms came under new management, therefore their Operating environment had changed and it is difficult to justify a comparison of predictions made under the old management with actual re- sults, partially earned under the new management. Addi- tionally, it is difficult to Obtain actual results if the acquired firm Operates as a segment of the acquiring firm. 48 StudpreriOd In order for the results of the research study to be as general as possible, the time period covered by the study included differing economic conditions. The period selected for this study was 1966 through 1970. Although this period is arbitrary, it does include some years of good economic climates, namely 1966, 1968 and 1969, a year of a slight downturn, 1967, and finally the reces— sion of 1970. As evidence that 1967 and 1970 were periods of economic downturn, the First National City Bank of New York survey of annual corporate profits1 found corporate profits rose in 1966, 1968 and 1969 by 10 percent, 9 per- cent and 3 percent, respectively. In 1967 and 1970, how— ever, the survey found that corporate profits fell by 1 percent and 8 percent, respectively. The claim that this period includes varying economic conditions is the researcher's Opinion. The reader must decide the degree of his agreement with this Opinion and how much the chosen period affects the outcome Of the study. 1First National City Bank of New York, Monthly Letter (New YOrk: First National City Bank Of New YOrk), April 1967, April 1968, April 1969, April 1970, and April and July 1971. 49 The Subpopulation Following the requirements set forth in the previous discussion, the data gathering process produced a suprpu- lation for the five-year study period Of 201 predictions Of future earnings. Table 6 presents the composition of the subpopulation by years. TABLE 6--Number of Predictions of Future Earnings Composing the Subp0pulation Number of Observations Year in Subp0pulation 1966 45 1967 33 1968 43 1969 44 1970 —£¥i Total 201 The 201 predictions of future earnings consisted of 152 firms for which one prediction was located in the five-year study period, 23 firms for which two predictions were located, and finally, one firm for which three predictions were found. It is interesting to note that fewer Observations were found in the years 1967 and 1970. These years coincide 50 with the years in the study period described as years of "downturns" in the economy. Although not specifically tested, this Observation seems to indicate a reluctance on the part Of firms to predict earnings when the economy is depressed. But, it cannot be determined whether this re- luctance is due to uncertainty about the economy and firms are therefore unable to make reliable predictions or whether it is due to a general reluctance by firms to predict lower earnings. To present a better picture of the firms composing the suprpulation, some additional data are summarized. To make the results of the research as general as possible, it is desirable to Obtain predictions of earnings from firms in different segments of the economy. The predic- tions composing the suprpulation are classified into some general industry groupings in Table 7. 51 TABLE 7--Firms Whose Predictions Compose the Subp0pulation Classified into General Industry Groupings £2221; teaming 2:22:23“ 1966 27 4 2 7 5 1967 19 4 7 l 2 1968 29 7 3 2 2 1969 28 8 - 3 5 1970 .22. .4 .9. _1 _2_ Totals 125 31 14 15 16 The size of the firms whose predictions compose the suprpulation is summarized in Tables 8 and 9. TABLE 8—-Size as Measured by Total Revenue for Firms Whose Predictions Compose the Subp0pulation Year Range Mean (Millions) (Millions) 1966 $5.7 -- $1,369.7 $257.5 1967 2.5 -- 975.3 178.6 1968 2.5 "' 4,558.5 467.2 1969 3.2 -- 4,953.3 364.6 1970 9.6 -- 1,248.5 274.4 52 TABLE 9--Size as Measured by Total Assets for Firms Whose Predictions Compose the Subp0pulation Year (MIIIISns) (MiIIIEns) 1966 $ 5.3 -- $2,269.2 $323.6 1967 4.2 -- 2,074.8 297.6 1968 2.7 -- 40,150.? 1,575.5 1969 2.9 —- 43,903.1 1,463.3 1970 16.7 -- 8,038.1 830.8 It should be noted that the mean total assets are much larger in 1968 and 1969 due to the inclusion of the predic- tions of the American Telephone and Telegraph CO. for those two years. It is the Opinion of the researcher that the compo- sition of the suprpulation does not make it inherently biased. Given the industry groupings and the range, in terms of size of the firms, the suprpulation does not appear to be biased on the basis of size or on the basis of concentration in particular industries. The reader must decide to what extent the composition of the subpopulation influences the results obtained and, therefore, any con- clusions reached from the study. 53 Limitations Introduced by the Data Composition It is Obvious that the method Of Obtaining predic- tions Of earnings, examining the Wall Street JOurnal, does not produce a random sample. The method employed produces, in fact, a subpopulation that exhibits the characteristics described above. Therefore, any statistical results Of the study cannot be generalized to the population of all firms. Any statistical generalizations may apply only to the sub— population from which the Observations were obtained. A possible bias of the study is due to the suprpu- lation being self-selecting. Making the predictions public was voluntary on the part of the firms. Therefore, it is possible that only those firms with an above-average ability to predict earnings made their predictions public, thus biasing the results. Second, it must be assumed that the predictions made public by the firms are the same predic- tions that would have been included in the published finan- cial statements. It is the Opinion of the researcher that this assumption is more valid than Daily's assumption that the predictions Obtained directly from the firms, that were not necessarily ever made public, are the same predictions 1 that would have been published. Daily, The Accountinngeview, p. 688. 54 The Measurement of Prediction Error The research reported on in the dissertation used the concept of reliability prOposed by Ijiri.1 If a system is reliable, then it generally works the way it is supposed to work. Reliability, howeven,is not an absolute yes-or-no concept, but can be thought of as the degree Of closeness of being correct. Ijiri proposed that the prOper measure— ment of reliability is the difference between the "alleged" value and the actual value. In this study the alleged value is a firm's prediction of future earnings. The measurement prOposed is actually a measurement of unreliability, or prediction error. The larger the dif— ference between the firm's prediction Of earnings and actual earnings the more unreliable the prediction. Additionally, the study will aggregate data and will present data for comparative purposes: therefore, it is desirable that the measurement Of prediction error produce a measure that is on the same basis as all other measures. The measurement that will be employed in this study is as follows: Relative Pre- = Actual Earnings - Predicted Earnings diction Error Predicted Earnings lYuji Ijiri, The Foundations Of Accounting Measure- ment (Englewood Cliffs, New Jersey: Prentice-Hall, Inc., 1967), pp. 137-142. 55 The measure produced is the prediction error expressed as a percentage of the firm's predicted earnings. This mea— surement scheme will produce measurements with the desired qualities mentioned above. The measurement of the earnings predictions errors could be expressed relative to either predicted earnings or actual earnings. In fact, the findings of Frank, discussed in Chapter I, and Green and Seagall, discussed in Chapter II, expressed prediction errors relative to actual earnings; while Daily's findings, discussed in Chapter II, expressed prediction errors relative to predicted earnings. The pre- dictions errors in the research reported in this disserta- tion will be expressed relative to predicted earnings be- cause in this form the prediction errors will be easier to use by investors. At the time the prediction is published the investor does not have knowledge of actual earnings. If the investor wishes to adjust the prediction for past tendencies to overpredict or underpredict, the adjustment is more convenient if past prediction errors are expressed relative to predicted earnings. A vast majority of the predictions located in the Wall Street Journal were given as per-share figures. Therefore, the study uses per-share figures in computing the relative prediction error. The necessary adjustments 56 for stock dividends and stock—splits were made. The actual earnings figures and the information needed to make the adjustments for per-share data were obtained from Moody's Investor Service, Inc. manuals.1 After the adjustments were made the Relative Prediction Errors were calculated. The re- sults are reported in the next section. Findings of the Descriptive Study Recall that the first research question suggested a descriptive study of the reliability of firms' predictions of future earnings. This section presents some descriptive statistics and figures for the suprpulation described in previous sections. Remember the statistic used is Relative Prediction Error. Figures 1-5, beginning on page 59, pre- sent the frequency distribution for the Relative Prediction Errors for the years 1966-1970. Figure 6, page 64, presents the frequency distribution for the entire five—year period covered by the study. The statistics for the frequency distributions in Tables l-6 may be found in Table 10. The statistics point out that the Relative Predic- tion Errors occur over a wide range and the mean error for 1Moody's Investors Service, Inc., Moody's Bank and Finance Manual, Moody's Industrial Manual, Moody's Public Utility Manual and Moody's Transportation Manual (New YOrk: Moody's Investors Service, Inc.), volumes for the years 1967-1971. 57 each year is negative. A negative error occurs when the firm overpredicts earnings. Examining the findings in Table 10 by years, it is obvious that 1966 was the best year with respect to reliability of predictions. As might be expected, 1967 and 1970 exhibit the poorest results and the predictions on the average are overstated. Recall that 1967 and 1970 were described as poor economic years. How- ever, no statement can be made about the absolute relia- bility of the predictions. Since there is a lack of knowl- edge of investors' materiality functions, it is not possible to declare the earnings predictions in the subp0pulation to either reliable or unreliable. The reader must assess the reliability of the earnings predictions according to his materiality function or some assumed materiality function. As can be seen in Figure 6, however, some of the in— dividual predictions appear to be sufficiently reliable to 1 be useful to financial statement readers. 0f the 201 pre— dictions, 35.3 percent are within 5 percent of actual earnings and 48.8 percent are within 10 percent of actual earnings. On the other hand, 39.8 percent are more than 15 percent from actual earnings. These results are comparable to Daily's findings that more than 33 percent of his sample had dif- ferences exceeding 15 percent and 53 percent of his sample 1With the possible exception of securities selling at a relatively high price-earnings ratio. 58 had differences exceeding 10 percent. The coefficient of variation is presented in Table 10 to describe the relative dispersion in the distribution in Figures l-6. The coefficient of variation is the ratio of the standard deviation of the distribution to the mean of the distribution. Caution should be exercised in inter- preting the results of the descriptive study using the co- efficient of variation. The coefficient of variation for the 1966 prediction errors is much larger than any of the other years, but much of its relative size is explained by the fact that the mean prediction error for 1966 is much smaller than the mean prediction error for the other years. The tendency to overpredict that is discussed above is further illuminated by the coefficient of skewness. The coefficient of skewness is computed as follows: 0 Coefficient of Skewness = M3 0'3 the third moment about the mean where: H3 0 standard deviation1 A symmetrical distribution has a coefficient of skewness of 0.0. A negative coefficient of skewness indicates that the 1Harald Cramér, Mathematical Methods of Statistics (Princeton, New Jersey, Princeton university Press, 1946), 59 FIGURE 1 O ('0 O N O H .0 O H I O N l O m I O b: I O In I O CD I E 3 .3 OOOH HON mHORHM GOHUUfifimHm U>fiUNHQM m0 fiOHHfiflfihumHQ hUQOSUQHWIIH NMDUHW 60 FIGURE 2 BOGH MOM WHOHHM GOHUUfiUmHm 0>HUmHOM HO COfiuSn—HHUWHQ ~AUG0370HWIIN NMDUHW 6]. FIGURE 3 om cm 0.0. o." o 07. out oml o3! oml owl om... P b pl - L H a a H. as. E ®©@H HOW mHOHHN flOfiUUfiUNHh 0>fiuwHOM MO COHUQQHHUWHQ phofiwfivmhnmllm EUHW 62 FIGURE 4 aoma How muouum coauowumum o>Humawm mo coauanfiuumwn hoamsvwhmllc MMDUHM 63 FIGURE 5 0H 0H 0 OH! owl cm! 0:! cm! owl om - n b P P . u . LP H H H H H H H. (M m m 1 m onma now muouum cofiuofipoum m>wumamm wo noausnwuumwa moamsgmHMIlm mmaon .64 FIGURE 6 okaHuocaH .eoHuom “mosnm>Hm pom muouum cowuofiwoum m>wumaom wo aofiuanwuumwo mocmsvaMIlo umDOHm 65 H.mI $.mom I o.He ©.MHI m.moa ou o.mmMI HON mummem>am w.NI a.am~ I m.mw H.NmI m.moa ou o.mmMI on omma m. I o.me I n.0H H.NHI w.HH ou N.Nm I «6 mega n. I «.mgw I m.m~ b.0HI o.mm ou ~.om I mq mead n. I c.moH I «.0N o.mHI m.mm Ou m.Hm I mm Roma o.HI Nm.¢HOHI Nm.NH N5.H I Nw.om ou Nc.mm I we mood mug swarm fin” ..s sews“. upsum m>aumauumwo ummew>Hm mo muHsmmm HNUHumfiumumIIoH mHm mo aofiuma>wo coax omsmm umbasz coaumowmammmao usoHUHmamoo usowoqwmooo cumvcmum mummem>Hm wsu pom mwswnsouo muumspsH Hmumsmo hm mvsum m>auewuomon mo mwaavaHmIIqH mgm¢e 73 o.HI m.NH~I N.~N N.0HI o.mm cu o.mw I naH q mummem>Hm q.HI m.mmNI m.H¢ m.qHI n.moa ou «.meI an N qua m. I m.N¢HI o.qa N.OHI w.HH Ou o.oe I we N mood o.HI m.mHNI m.mH m.m I m.mm Ou m.mm I cc m moma q. n.5maI m.wH q.a I m.mm Ou m.<< I on m mama H. I Nn.ooou No.qa Nu. Nn.om ou Nm.Nm I no N bead mmocsmxm mo soaumauw> mo soaumw>wn coax mmsmm maowum>ummno vm>osom new» ucmaofimmooo usmfioawmooo pumpcmum wsaaamamm mumwauso umbasz mo ambasz mumHHuso mo Hm>oamm uwum< xpsum o>wuefiuumon mo macaw hp mwsficawmllna m4m< U1 ll i the size of the firm in average total assets, and 89 Ei = the error term, or the amount of the vari- ation of the dependent variable from its mean not explained by the regression equation. Recall from Chapter III that the dependent variable, prediction errors, used in this research was a suprpulation and not a random sample of the population of all firms. Therefore, no inferences should be made from the multi- variate analysis about the universe of all firms. However, since no inferences should nor will be made to an underlying population, the strictness of the assumptions regarding the distributions of the dependent variable and the independent variables can be relaxed. When theldata are actually analyzed using multiple regression, it is desirable that the final regression equa- tion contain only those independent variables that are sig— nificant in explaining the variations in the dependent vari- able. For this reason a least squares deletion algorithm was employed in constructing the final regression equation. The least squares deletion algorithm begins with a regres- sion equation containing all independent variables. The algorithm then examines each independent variable for its contribution to the total variation in the dependent vari- able eXplained by the regression equation. The next step in the algorithm is to compare the independent variable with 90 the least significant contribution to the total variation explained to some preset desired level of significance. If the independent variable is not significant it is deleted from the regression equation. The algorithm is repeated, deleting one variable at a time, until all remaining vari- ables are significant according to the preset level of sig- nificance. In this study the level of significance was at .05 because this level of significance is commonly used in research of this type. For a more detailed discussion of the least squares deletion algorithm see Appendix A. Results of the Multivariate Analysis The purpose of the multivariate analysis was to attempt to find an association between the dependent vari- able, prediction errors, and the several independent vari- ables. This association consisted of associating the vari- ation of the dependent variable by variations of the inde- pendent variables. The ability of the regression equation to associate the variation of the independent variables with the variation of the dependent variable may be ex- pressed by R2, the coefficient of determination. The co- efficient of determination is computed as follows: n ~ 2 .2 (Y1 " Yi) R2 = 1:1 — 2 Z .. i=1 (Yi Yi) 91 where: Y. - Y. = the deviation of the predicted value of the dependent vari- able Obtained by using the regression equation from the mean of the observations of the dependent variable, Y. - Y. = the deviation of the Observed values of the dependent variable from the mean of the observations of the dependent variable.1 The coefficient of determination is the relative amount of total variation associated by the regression equation. The initial regression equation calculated before any of the independent variables were deleted due to insig— nificance has a value for R2 of .2409. This value of the coefficient of determination can be interpreted as the regression equation associated 24 percent of the variation in the dependent variable with the variations in the sig- nificant independent variables. The least squares deletion routine deleted three of the independent variables, in the order listed below, be- cause their contributions to the total variation associated by the regression equation were insignificant at the .05 level. X3 - Fluctuation in Past Earnings X5 - Size of the Firm X2 - Change in Industry Profits 1Ibid.. pp. 561-562. 92 After the deletion the regression equation is as follows: Yi = -l7.121 + 1.388Xli + .067X4i Recall that X1 is the change in aggregate corporate profits and X4 is the ratio of extraordinary gains and losses to Operating earnings in the year of the prediction. After the three independent variables were deleted the regression equation was able to explain 22.86 percent of the total variation of the dependent variable about its mean. The levels of significance-of the remaining variables are pre- sented in Table 17. TABLE l7--Level of Significance of Independent Variables in the Initial Model After Deletions Level of Variable Significance X1 - Change in Aggregate Corporate Profits .003 X4 - Relative Extraordinary Gains and Losses <.0005 Note that the variables deleted have the smallest correla- tion with prediction errors on a univariate basis as pre- sented in Table 16. Analysis of Residuals A residual is the difference between the predicted value of the dependent variable using the regression equa- tion and the observed value of the dependent variable. By 93 examining the residuals, especially the large residuals, a researcher can Often gain further information about the data. In the regression equation discussed above there were several residuals that were extremely large. These particular Observations were examined in an attempt to determine the reason for the large residuals. It was noted that in most of these cases, the reason for the large residuals appeared to be from a large change in earnings from the previous year. This observation sug— gested that the inclusion of a variable representing the relative change in earnings from the previous year might improve the ability of the regression equation to explain the variation of the dependent variable. Reformulation with an Additional Independent Variable To take advantage of the analysis that the observa- tions with large residuals were those firms with a sizable change in earnings, the multiple regression model was re- formulated with an additional independent variable. The new independent variable represented the change in Oper- ating earnings for each firm and was calculated as follows: Change in Operating Earningsi = OEOi - OE‘li . 100 CE -li where: OEOi = Operating earnings for firm 1 for the year covered by the prediction, 94 OE_li = Operating earnings for firm 1 for the year previous to the year covered by the prediction. The change in Operating earnings was used, rather than the change in net income, because independent variable X4, the ratio Of extraordinary gains and losses to Operating earn- ings for the year of the prediction, already provided for the effect of extraordinary gains and losses. To employ the change in net income would introduce collinearity be- tween X and the new independent variable, X Addition- 4 6' ally, the change in Operating earnings was divided by operating earnings for the year previous to the year covered by the prediction to get all the observations on a relative basis. Multiplying by 100 produced the percent— age change in Operating earnings. Analyzing the new independent variable, relative change in Operating earnings, on a univariate basis pro- duced a slightly higher correlation coefficient than any of the initial independent variables. Recall from the earlier discussion that the strongest association was be- tween the relative prediction errors and X4, the ratio of extraordinary gains and losses to Operating earnings, with a correlation coefficient of .43530. The correlation co- efficient for X was a slightly higher .44008. 6 To examine the affect of the new independent 95 variable on the multivariate analysis, the multiple regres- sion model was reformulated, producing a coefficient of determination of .3740 before any of the variables were deleted due to insignificance. In other words, the reformu— lated regression model can now explain 37.4 percent of the variation in the dependent variable, Relative Prediction Errors, whereas the initial model could only eXplain about 24 percent of the variation, before deletion. In the initial regression model, the deletion algorithm removed three variables: X3, X5 and X2, in that order. In the reformulated model the same variables were deleted but in a different order. The order of deletion was as follows: X2 - Change in Industry Profits X3 — Fluctuation in Past Earnings X5 - Size of the Firms After the deletion of insignificant variables, the remain- ing model was as follows: Y1 = -19.734 + 1.221 xii + .059 X41.- + .266 X6i The coefficient of determination for the regression model after deletion is .3666. The level of significance for each of the independent variables is presented in Table 18. 96 TABLE 18--Level of Significance of Independent Variables in the Reformulated Mbdel After Deletions Level of Variable Significance X1 - Change in Aggregate Profits .004 X4 - Extraordinary Gains and Losses <.0005 X6 - Change in Operating Earnings <.0005 All three of the remaining variables are much more signifi- cant than the minimum level of significance of .05. Analysis of Outliers When examining residuals, particular attention should be paid to Outliers, those far greater in absolute value than the other residuals. In the reformulated regres— sion model there were four such outliers, all from the 1970 predictions. Three of the residuals were more than three standard deviations from the mean of the residuals and the fourth was more than nine standard deviations from the mean. Two of the observations were for firms that predicted posi- tive earnings and in fact suffered a considerable loss. The third Observation was a utility that had a 600 percent increase in earnings due to a 19 percent rate increase. The firm, however, must have included the increase in its 97 prediction since the relative prediction error was only 18.7 percent. The fourth observation was a firm that eXpe— rienced an extraordinary gain approximately twice as large as Operating earnings. These four atypical Observations cloud the general ability of the regression model to explain the variance of relative prediction errors. As a final step in the analy- sis, these four outliers were removed from the Observations. The Correlation Coefficients after removing the outliers are presented in Table 19. TABLE l9--Correlation Coefficients Between Relative Prediction Errors and Several Variables After Removing Outliers Coefficient of Correlation Variable Being Analyzed With Outliers Without Outliers Change in Aggregate Corporate Profits .24772 .25003 Change in Industry Profits .14739 .15712 Fluctuation in Past Earnings .00099 -.01963 Relative Extraordinary Gains and Losses .43530 .65599 Size of the Firm .03974 .04734 Change in Operating Earnings .44008 .55366 98 It is interesting to note that after removing the four outliers the independent variable with the strongest association with Relative Prediction Errors was Relative Extraordinary Gains and Losses. This analysis is con- sistent with the American Institute of Certified Public Accountants' definition of extraordinary items. Extra- f ordinary items were defined as nonrecurring and unexpected. The fact that extraordinary items have the strongest asso— ciation with prediction errors points out that they must be unexpected or they would have been included in the firm's prediction of earnings, thereby reducing the predic- tion error and reducing the correlation between the predic- tion errors and the extraordinary items. After removing the four outliers, the reformulated multiple regression model was recalculated. The removal of the outliers increased the coefficient of determination from .3666 to .6419 before the deletion of insignificant independent variables. After removing the atypical obser- vations, considerably more than half the total variation of~ the dependent variable, prediction errors, was explained by the six independent variables of the reformulated model. The deletion algorithm produced a different regres- sion equation after removing the outliers. In fact, only two of the six posited independent variables were deleted 99 due to insignificance. The two variables in the order deleted were: X2 - Change in Industry Profits Xl - Change in Aggregate Corporate Profits. The independent variables remaining in the regression model and their level of significance are presented in Table 20. TABLE 20-—Leve1 of Significance of Independent Variables in the Reformulated MOdel with Outliers Removed and After Deletion Level of variable Significance X3 - Fluctuation in Past Operating Earnings .002 X4 - Relative Extraordinary Gains and Losses <.0005 X5 - Size of the Firm .005 X6 - Change in Operating Earnings <.0005 The deletion of the two insignificant independent variables reduced the coefficient of determination to .6334. The posited model continues to associate considerably more than half the total variation of the prediction errors. The fact that X1 and X2 were deleted means that those factors exogenous to the firm were insignificant in explaining the prediction errors for the suprpulation. The only variables remaining are the endogenous factors. 100 Summary Chapter IV reports upon an analysis of Relative Prediction Errors which sought to determine the association between the prediction errors and various variables. The variables chosen for analysis were both exogenous and endogenous to the firm. The variables were analyzed on both a univariate and a multivariate basis. In the initial formulation of the model, on a univariate basis the Relative Extraordinary Gains and Losses were found to have the strongest association with prediction errors with a corre— lation coefficient of .4353. A multiple regression model with Relative Prediction Errors as the dependent variable produced a coefficient of determination of .2409. An examination of the residuals of the regression analysis suggested that the inclusion of another variable would produce a better explanatory model. This new variable was the change in Operating earnings between the year previous to the year of prediction and the year covered by the earnings prediction. This new variable exhibited a cor- relation coefficient of .44008. The multiple regression model was reformulated with the new variable and a coeffi- cient of determination of .3740 was attained. Further examination of the regression residuals determined that four of the residuals were more than three 101 standard deviations from the mean of the residuals. Closer examination of the outliers suggested that they were atypical and possibly clouding the results of the analyses. After removing the four outliers, the variable with the strongest association with the prediction errors was Rela- tive Extraordinary Gains and Losses with correlation coef- ficient of .65599. The multiple regression analysis with the four outliers removed attained a coefficient of deter- mination of .6419. This means that the posited variables explained considerably more than half the variation of the prediction errors. Of the six independent variables, two variables were insignificant in their contribution to the variation explained. These two variables were the exogenous independent variables. The four significant independent variables were the endogenous variables. CHAPTER V SUMMARY, CONCLUSIONS AND RECOMMENDATIONS The first section of Chapter V contains a summary of the results obtained in the previous chapters. First the results of the descriptive study of earnings predictions are summarized, followed by the results of the univariate and multivariate analyses of the earnings predictions. The next section presents the conclusions drawn from the results of the empirical examination of the earnings predictions. The final section of Chapter V presents a discussion of the recommendations supported by the conclu- sions. Included in the recommendations are suggestions for further research efforts. Summary of Results In the literature of finance it is found that in- vestors should desire knowledge of future earnings. The theoretical price of a share of common stock is the present value of future dividends. Future earnings are generally used as a surrogate for future dividends, therefore, if 102 103 financial statements were to contain predictions of future earnings,the statements would be more useful to investors. The prOposals made by scholars that predictions of earnings be included in the financial statements is gaining acceptance in the business world. For example, the chair- man of the Securities and Exchange Commission called for I} the inclusion of earnings predictions in published financial statements. In addition, in England,firms are required to include predictions of future earnings in prospectuses for new stock issues and in connection with take-over bids. ‘” The formats suggested by various researchers for the presentation of the earnings predictions in the pub- lished financial statements not only include the prediction, but also attempt to aid the reader in assessing the reli- ability of previous earnings predictions of the firm. The aid in assessing reliability would be a comparison of the previous years' earnings predictions with the actual results of the Operations. These comparisons would indicate to the reader any past tendency of the firm to overestimate or underestimate future earnings. The reader could also ob- serve how the reliability of previous predictions as affected by such events as economic downturns, strikes, wars, etc. For the immediately previous year the comparison 104 between predicted earnings and actual earnings would be accompanied by management's explanation of any difference. This explanation would give the reader further information about the effect of economic and political events on the reliability of the earnings predictions. Additionally, the reader could gain an insight into the foresight of manage- ment and how well management reacted when the unexpected did occur. The prediction Of earnings for the current year could be presented in the form of a point estimate, or preferably in the form of a range estimate with its prob- ability distribution. As additional information, the assumptions underlying the earnings prediction, such as changes in Gross National Product, would be disclosed. Depending on the degree of agreement with the assumptions, the reader could decide how much weight to attach to the earnings predictions in his decision making process. The dominant criticism against the prOposals to include earnings predictions in the published financial statements is that the predictions would prove unreliable. If the earnings predictions are to be usable to the state- ment readers, then the predictions must be sufficiently raliable to be used in the decision making processes of the readers. The critics claim that, due to the uncertainty of 105 the business environment and the undevelOped prediction processes, firms cannot make predictions of sufficient reli— ability to be useful to statement readers. The criticism of unreliability served as motivation for the descriptive study reported on in this dissertation. To examine the reliability of published predictions of E future earnings, 201 predictions were obtained from the E Wall Street JOurnal. The predictions composing the suprpu- lation included the years 1966 through 1970. The predic- tions were further restricted to being point estimates be- cause of difficulties in error measurements, and restricted to those appearing within 120 days after the end of the firm's fiscal year to meet Securities and Exchange Commission annual report requirements. It was explicitly assumed that the predictions composing the subpopulation were the same predictions that would have appeared in the published finan- cial statements. Prediction errors were measured as the difference between actual earnings and predicted earnings expressed as a percentage of predicted earnings. A negative percentage indicates an overprediction, while a positive percentage indicates an underprediction. The overall results for the five—year period indicated an average prediction error of -l3.6 percent and a standard deviation of 41.6 percent. 106 For the individual years, 1970 had the largest average pre- diction error of -32.1 percent and a standard deviation of 83.5 percent, while 1966 had the lowest average prediction error of —l.7 percent and a standard deviation of 17.8 per- cent. For the entire five—year period, 35.3 percent of the firms had a prediction error between positive 5 percent and negative 5 percent and 48.8 percent of the prediction errors fell between positive 10 percent and negative 10 percent. The frequency distributions of the prediction errors indicate a tendency on the part of the firms to overpredict future earnings. Of the 201 predictions, 63.7 percent were overpredictions, 33.8 percent were underpredic- tions and 2.5 percent were exact predictions of future earn— ings. A chi-square one-sample test was performed comparing the frequency of occurrence of overpredictions to under- predictions and exact predictions. For all individual years, except 1966, the distributions were not consistent with a symmetrical distribution of overpredictions and underpredictions. The results of the entire five-year study were also not consistent with a symmetrical distri- bution. A one-sample runs test was also performed on the order in which overpredictions and underpredictions Occurred during the study period. For each year and for the entire five-year period the results were consistent with a random 107 order of occurrence of overpredictions and underpredictions. The prediction errors for the five-year study were also examined by general industry groupings. As expected, the utilities had the lowest average prediction error of —.6 percent. The industrials had the highest average pre— diction error of -l7.2 percent. The statistics computed for the five-year study period were influenced by several extreme observations. The results were examined to determine the effect of these outliers. An outlier is arbitrarily defined as an observa- tion more than two standard deviations from the mean. By removing four such outliers, the average prediction error for the five-year period was reduced from -13.6 percent to -10.2 percent. The standard deviation was reduced from 41.6 percent to 22.2 percent. Following the descriptive study of the prediction errors, several independent variables were analyzed for their association with the prediction errors. The inde- pendent variables included: Change in Aggregate Corporate Profits, Change in Industry Profits, Fluctuation in Past Operating Earnings, Relative Extraordinary Gains and Losses, and Size of the Firm. The first analysis of asso- ciation performed was the association between the prediction errors and each of the independent variables individually. 108 The statistical method employed was correlation analysis. Of the five independent variables, Relative Extraordinary Gains and Losses had the highest coefficient Of correlation of .43530, followed by the Change in Aggregate Corporate Profits with a coefficient of correlation of .24772. The remaining variables all had a coefficient of correlation r of less .15. i In addition to the univariate analysis described above, the combined association Of the independent vari- may. a. ables with the prediction errors was analyzed using a least squares deletion multiple regression algorithm. The results of the multiple regression analysis indicated that only two of the independent variables were significant at the .05 level in their contribution to the association between the prediction errors and the independent variables. The two significant independent variables were Change in Aggregate Corporate Profits and Relative Extraordinary Gains and Losses. The coefficient of determination of the model after deletion was .2286. An analysis of the regression residuals indicated that the addition of another independent variable, Change in Operating Earnings, would improve the multiple regres— sion model. The reformulated model contains three signifi- cant independent variables: Change in Aggregate Corporate 109 Profits, Relative Extraordinary Gains and Losses, and the Change in Operating Earnings. The coefficient of determina- tion for the reformulated model was .3666. An additional examination Of the large regression residuals indicated that four of the large residuals were atypical and their removal would improve the multiple re- gression model. The regression equations contains four significant independent variables after removing the four atypical observations. These variables were: Fluctuation in Past Operating Earnings, Relative Extraordinary Gains and Losses, Size of the Firm, and the Change in Operating Earnings. The coefficient of determination for the regres- sion equation after removal of the atypical Observations was .6334. Conclusions The research reported on in the dissertation was composed of two basic parts: the descriptive study and the analysis of association between prediction errors and the six independent variables. The conclusions are drawn from each of these parts. Conclusions from the Descriptive Study As discussed in Chapter III, the lack of knowledge of investors' materiality functions precluded reaching a 110 conclusion that the predictions in the subpopulation are suf— ficiently reliable to be useful or they are not sufficiently reliable to be useful. But there are some conclusions that can be drawn from the descriptive study. The first conclu- sion is that some of the predictions £923 to be reliable enough to be useful. With prediction errors measured as a F” percentage of predicted earnings, more than one—third of the : predictions, 35.3 percent, are within 5 percent of actual earnings, and almost one—half of the predictions, 48.8 per— cent, are within 10 percent of actual earnings. This conclu— Le sion is particularly true for the utilities which had an average prediction error of -.6 percent over the five-year study period. Of the predictions that fall within 5 percent of actual earnings, 28.2 percent were predictions of the utilities. This leaves 71.8 percent of the predictions that were within 5 percent of actual earnings as predictions of firms other than utilities. Therefore, firms other than utilities can also make reliable predictions, although on the average, utilities managements are superior predictors. The second conclusion from the descriptive study is that the firms whose predictions composed the subp0pulation exhibited a tendency to overpredict earnings. This conclu- sion is based on several facts. First, the average Relative Prediction Error in all cases was negative, indicating an lll overprediction. Secondly in all individual years and for the entire five-year period, the distributions of the Relative Prediction Errors had a negative coefficient of skewness. The is further supported by the fact that 63.7 percent of the Relative Prediction Errors were overpredictions. A chi- square one-sample test found this distribution to be sig- nificantly different from a symmetrical distribution at the .05 level of significance. This conclusion is not consis— tent with the argument that firms would intentionally under- predict earnings because of the conservative nature of businessmen. Conclusions from the Association Analysis The conclusions drawn from the association analysis, both univariate and multivariate, are based on the statisti- cal results Obtained after all refinements were made, i.e., the inclusion of an additional independent variable and the removal of atypical Observations. The first conclusion from the association analyses is that the endogenous independent variables have a signifi- cant association with the Relative Prediction Errors in the suprpulation. There is a possible explanation for the insignificance of the exogenous variables. This possible explanation is that both of the exogenous variables. 112 Change in Aggregate Corporate Profits and Change in Industry Profits, are indexes compiled from a large number of firms and are much less volatile in their movements than some of the endogenous variables. Although these exogenous vari— ables are positively correlated with the Relative Prediction Errors, their contribution to the multiple regression we model's assocation of the variation in Relative Prediction Errors is not significant. The second conclusion from the association analyses deals with method of prediction employed by firms that had L~ earnings predictions in the suprpulation. The independent variable with the second highest coefficient of correlation was the Change in Operating Earnings. This positive asso- ciation was consistent with firms predicting earnings with only a small change from the previous period. If there was a large increase in Operating earnings, then there is a large underprediction. If there is a large decrease in Operating earnings, then there was a large overprediction. These relationships brought about the relatively strong association found between the Relative Prediction Errors and the Change in Operating Earnings. The final conclusion from the association analyses concerns the relatively strong association between Relative Prediction Errors and Relative Extraordinary Gains and 113 Losses. The fact that this association was the strongest of the independent variables examined has two possible explanations. The first explanation is that the Extra- ordinary Gains and Losses were really unexpected and not considered in the prediction process. The second explana- tion is that the firms that had predictions in the subpopu- *- lation were actually predicting earnings before extra— ordinary items and were not wording their predictions accordingly. If the firms were actually predicting earn— ings before extraordinary items, then naturally Relative 1 Extraordinary Gains and Losses were significant in associ- ating differences between predicted earnings and actual earnings after extraordinary items. Recommendations Based partially on the findings of the research reported on in the dissertation and as a personal observa— tion of the author after doing the research, published annual financial statements should include predictions of earnings for the forthcoming fiscal year. The presentation of the earnings predictions should be in the format dis- cussed in Chapter II. This format would assist the state— ment reader in assessing the past ability of the firm to predict earnings, and knowledge of the assumptions 114 underlying the current earnings prediction and would further assist the reader in deciding whether or not to use the prediction in his decision model. The recommenda- tion for including the predictions in the published finan— cial statements is based on four considerations. 1. The descriptive study indicates that many firms p- can make predictions of future earnings that seem to be sufficiently reliable to be useful. If earnings predictions are required in the published financial statements, there would be motivation for firms to improve the reliability of the prediction process. Many predictions are currently published, as evi- denced by the number of "usable" predictions in the subp0pulation. The number of "unusable" pre- dictions greatly outnumber the usable predictions. These predictions, however, are published in an informal manner without the benefits of the sug- gested format of presentation. If inclusion in the published financial statements is required, the whole process would be formalized. If all firms were required to publish their earnings predictions, it would be a start at solving the problem of "inside information." 115 Firms that employ a budgeting system arrive at a prediction of earnings for the coming year. It seems possible that influential investors and security analysts gain access to these predictions. If firms are required to publish these predictions, then the pre- —u dictions will become public information. It is also recommended that further research be undertaken to learn more about predictions of future earn- ings. This research should not only be concerned with the reliability of predictions, but also the prediction pro- cesses should be studied to determine the methods employed by firms that do make reliable earnings predictions. Pre- dictions for periods other than a fiscal year should be examined. Any research finding would have implications for including predictions of earnings in the quarterly financial statements. Finally, further research is recommended for identifying those factors that are associated with a firm's ability to predict. This additional research should not only be concerned with different variables than those em- ployed in this study, but also in attempting to find better measurements for these variables. The final recommendation is for further research into the decision making processes of investors. The 116 results of these research efforts would provide knowledge about investors materiality functions. This knowledge would provide a benchmark for assessing the usefulness of earnings predictions. APPENDIX A MODEL CONSTRUCTION USING A LEAST SQUARES DELETION ALGORITHM The purpose for which multiple regression was used in the research reported on in the dissertation was to ex- amine the ability Of several posited variables to explain variations in the dependent variable, Relative Prediction Errors. While it is desirable for the sake of completeness " ‘Fl'n' n.JI' '_ to examine the ability of the entire set of independent variables to explain the variation, for the purpose of Intuit-3r. efficiency the final regression equation should contain only those independent variables that make a significant contribution to the total explained variation. The proce- dure chosen to accomplish these goals was a stepwise dele- tion algorithm, also known as a backward elimination pro-. cedure.1 This procedure was selected because it initially calculates the regression equation using all independent variables and then eliminates the independent variables that are insignificant in their contribution to total ex- plained variation. In detail the stepwise deletion algorithm consists basically of three steps. 1. A regression equation using all the posited independent variablesis calculated. 1N. R. Draper and H. Smith, Applied Regression Anal sis (New York: John Wiley and Sons, Inc., 1966), pp. -169. 117 118 2. For each independent variable in the equation a partial F-test is computed treating the variable as if it were the last variable to enter the equation. The partial F-test is a statistic computed comparing the contribution to the regression sum of squares of each vari- able, given that all other variables are already F 1 in the equation. 3. The lowest partial F-test value is then com- pared with the preset level of significance, .05 in this research: Finis (a) If the independent variable with the lowest partial F-test value is insignificant, it is deleted from the regression equation. The regression equation is then recomputed and step 2 is performed. (b) If the independent variable with the lowest partial F-test value is significant, then all remaining variables are also signifi- cant and the algorithm is completed.2 lIbid., pp. 71-72. 21bid., pp. 167-168. BIBLIOGRAPHY American Accounting Association. A Stgtement of Basic Accounting Theory. Evanston, Illinois: American Accounting Association, 1966. American Institute of Certified Public Accountants. Opinion of the Accounting Principles Board Opinion No. 9. New York: American Institute of”Certified PuEIic Accountants, 1967. Ball, Ray and Brown, Phillip. "An Empirical Evaluation of F- Accounting Income Numbers,” Journal of Accounting . Research, VI (Autumn, 1968), I59-178. Bates, Rex J. "Discussion of The Information Content of Annual Earnings Announcements," Empirical Research in Accounting: Selected Studies! 1968, a supplement to Volume VI 0 the‘fiOurnal of'Accounting Research, 93-95. Beaver, William H. "The Information Content of Annual Earnings Announcements," Empirical Research in Accounting: Selected StudiesLII968, a supplement to Volume VI of the Journal of Accounting Research, 67-92. Bernstein, Leopold A. "The Concept of Materiality," The Accounting Review, XLII (January, 1967), 86-95. Carey, John L., and Doherty, William 0. Ethical Standards of the Accounting Profession. New Yofk: American Institute of Certified Public Accountants, 1966. Clark, Charles T., and Schkade, Lawrence L. Statistical Methods for Business Decisions. Cincinnati, Ohio: South-western Publishing Company, 1969. Cooper, W. W., DOpuch, N., and Keller, T. F. "Budgeting Disclosure and Other Suggestions for Improving Accounting Reports," The Accounting Review, XLIII (October, 1968), 640-637. Cramer, Harald. Mathematical Methods of Statistics. Princeton, New Jersey: Princeton University Press, 1946. Daily, R. Austin. ”The Feasibility of Reporting Forecasted Information," The Accounting Review, XLVI (October, 1971), 686-692. 119 120 Draper, N. R., and Smith, H. Applied Regression Analysis. New York: John Wiley and Sons, Inc.,41966. Frank, Werner. "A Study of the Predictive Significance of Two Income Measures," Journal of Accounting Research, VII (Spring, 1969),l23-136. Fraser, Ian J. "Accountancy and the Merger Movement," The Accountant, Volume 165 (September 9, 1971), 353-355. Green, David, Jr., and Seagall, Joel. ”The Predictive ‘“ Power of First Quarter Earnings Reports," Journal of Business, XL (January, 1967), 44-55. Hendriksen, Eldon S. Accounting Theory. HOmewood, Illinois: Richard D. Irwin, Inc., 1970. . "Disclosure - Insights into Requirements in the United Kingdom,” The International Journal of Accounting, IV (Spring, 1969T,21-32. Ijiri, Yuji. The Foundationspf Accounging Measggement. Englewood Cliffs, New Jersey: Prentice-Hail, Inc., 1967. Institute of Chartered Accountants in England and Wales. "Accountants' Reports on Profit Forecasts," 815, June 26, 1969. Latane, Henry A., and Tuttle, Donald L. Securit Anal sis and Portfolio Management. New YorE: The Ronald Press Company, 1970. May, Robert G. The Influgnce ofyguarterly Earnin s Announcemepts onlinvestor DecisiOns as Re egted 13 Common Stock Price Changes. Unpublished Ph.D. dissertation, Michigan State University, 1970. Mennis, Edmund A. "Forecasting Corporate Profits," in HOw Business Economists Forecast. Edited by WIIIIam E. Butler and Robert A. Kavesh. Englewood Cliffs, New Jersey: Prentice Hall, Inc., 1966. Monthly Letter. New York: First National City Bank of New York (April, 1967-1971, and July 1971). Moody's Bank and Finance Manual, Moody's Industrial Manual, MOOdy‘s Public UtilitijanuaI, and MoodyTs Trans- portation Manual. New York: Moody‘s Investors Service, Inc., volumes for the years 1967-1971. 121 New York Stock Exchange Guide, Volume III -- Related Laws and Regulations. New York: Commerce Clearing House, Inc., 1962. Revsine, Lawrence. "Predictive Ability, Market Prices, and Operating Flows," The AccountingyReview, XLVI (July, 1971), 480-489. Robards, Terry. "S.E.C. Shifts Hinted on Earnings Fore- casts," The New York Times, November 19, 1971, 63—650 Schattke, Rudy. "Expected Income -- A Reporting Challenge," The Accounting Review, XXXVII (October, 1962), 670-676. Securities and Exchange Commission. Directory of Companies Filing Annual Reports with Securities and'Exchange Commission. washington D.C.: U. S.IGOvernment Printing Office, volumes for 1966, 1969 and 1970. Siegel, Sidney. Nonparametric Statistics for the Behavioral Sciences. New York: McGraw-Hill Book Company, Inc., 1956. Solomon, Ezra. The Theory of Financial Management. New York: Columbia University Press, 1963. wall Street Journal. January-April, 1966-1970. Wilkinson, T. R., and Doney, L. D. "Extending Audit and Reporting Boundaries," The Accounting_Review, XL (October, 1965), 753-7567 T— .I—I‘