THE INFORMATION CONTENT OF ACCOUNTING CHANGES DISSERTATION FOR THE DEGREE OF Ph.D. MICHIGAN STATE UNIVERSITY WALTER THOMAS HARRISON. JR. I975. Iii-1 I ' ' F‘t‘ 9'9 This is to certify that the thesis entitled THE INFORMATION CONTENT OF ACCOUNTING CHANGES presented by Walter T. Harrison has been accepted towards fulfillment of the requirements for Ph.D. degree in Accounting Major professor Date W6 0-7639 . I Lulu; III" l I I Ll III I! ”ll I I”; 13127: 3.311‘ ABSTRACT THE INFORMATION CONTENT OF ACCOUNTING CHANGES By Walter Thomas Harrison, Jr. This research is an attempt to answer some ques- tions that previous inquiries into the stock market effects of accounting changes (ACs) have left unanswered, and largely unaddressed. One is the general question of whether the market reacts systematically to A03. The failure of most earlier studies to report the results of a control group of nonchange firms casts some doubt on the ability of their research designs to answer this question. Also, no previous study has examined the market effect of nondiscretionary ACs. All previous AC studies have been based on the assumption that security return distributions are univariate in nature. This assumption precluded the detection of any dependency between the information effect of ACS and a particular risk class of firms. And finally, no previous study has examined both means and variances of security return distributions in assessing the informa- tional effects of ACs. In order to correct for the abovementioned omis- sions, this study compares rates of return on a large Walter Thomas Harrison, Jr. sample of firms that made ACs during the years 1968-1972 to a control group of nonchange firms that were individu- ally matched on fiscal year-end, relative risk, and less stringently, on industry membership. It tests for the effects of ACs that have four possible directional effects on net income: (1) positive; (2) negative; (5) zero or immaterial; or (4) directional effect not disclosed. It also tests for the effects of ACs that were made with three degrees of management discretion: (1) discretionary --made at the discretion of management; (2) nondiscre- tionary--made as the result of a pronouncement by an exog- enous body such as the Accounting Principles Board, or (5) both--the firm made atleast one discretionary AC and at least one nondiscretionary AC during the same year. Then tests are performed on each of the four directional effect categories in combination with each of the three degrees of relative discretion (e.g., discretionary ACs with positive effects on net income, etc.). Return dis- tributions are assumed to be multivariate in nature in order to give effect to the possibility of detecting any existing dependency between AC information and a particu- lar risk class. And finally, tests are conducted on both mean vectors and covariance matrices of returns. The study assumes that capital markets are efficient with respect to publicly available information and that the Walter Thomas Harrison, Jr. capital asset pricing model reflects the market's mecha- nism for establishing equilibrium values for firms. The test period extends from six months before the month when year-end AC disclosures are made to six months after, a thirteen-month period. The omnibus test for the information effects of ACs (of all types) in general produced results that are consistent with the null hypothesis of no information con- tent. The four tests for the effects of ACs with the four directional effects all produced results leading to the same inference. Also, the three tests for the effects of ACs with the three relative discretion characteristics, in general, implied no information content in ACs. However, all eight of the tests Just mentioned were shown to bring together ACs with widely different characteristics such that, in some cases, the effects of ACs with more narrowly defined (and competing) characteristics canceled each other. This was evident from the results of tests on ACs with a specific directional effect on net income and a specific relative discretion characteristic. A number of these tests produced results that led to rejection of the null hypothesis and hence to the inference that ACs are strongly associated with changes in firms' equilibrium values. On this basis, it appears that these ACs have information content. THE INFORMATION CONTENT OF ACCOUNTING CHANGES by Walter Thomas Harrison, Jr. ‘A DISSERTATION Submitted to Michigan State University in partial fulfillment of the requirements for the degree of DOCTOR OF PHILOSOPHY Department of Accounting and Financial Administration 1976 © Copyright by Walter Thomas Harrison, Jr. 1976 All rights reserved. ii D E D I C A T I O N This dissertation is dedicated to my beloved wife Nancy. Her mental, spiritual, and physical encouragement was an indispensable element that contributed mightily to the completion of this study. iii A C K N O W L E D G M E N T S I wish to eXpress my gratitude to my disserta- tion committee of Professor Daniel W. Collins, Professor James B. Greene, and Professor Melvin C. O'Connor (Chair- man). At virtually every stage of the development of this project they were available on an almost unlimited basis for suggestions as to tOpical areas for the research, ad- vice as to writing style, and encouragement toward its com- pletion. The research was partially financed by grants from the Haskins & Sells Foundation and the Department of Accounting and Financial Administration of the Graduate School of Business Administration at Michigan State Uni- versity. In addition, valuable data were supplied by the American Institute of Certified Public Accountants and by Merrill Lynch, Pierce, Fenner & Smith. To these organiza- tions I extend my hearty thanks. Walter Thomas Harrison, Jr. February 27, 1976 iv Chapter I. II. III. IV. C O N T E N T S INTRODUCTION 0 O O O O O O O O O C O O O O O O l THEORETICAL FOUNDATIONS AND EMPIRICAL SUPPORT FOR MODELS EMPLOYED . . . . . . . . . . . . . 13 Information Content . . . . . . . . . . . . . 13 Efficient Capital Markets . . . . . . . . . . 15 Capital Asset Pricing . . . . . . . . . . 17 The Distributional Properties of Security Returns . . . . . . . . . . . . . . . . 19 Integration of Theories and Evidence into the Present Design . . . . . . . . . . . . . . . 20 LITERATURE REVIEW . . . . . . . . . . . . . . 26 Archibald . . . . . . . . . . . . . . . . . . 26 Baskin . . . . . . . . . 28 Kaplan and Roll . . . . . 32 Ball . . . . . . . . . . . 55 . . . . . 37 revious . . . . . . . . . . 4O sunder O O O O O O O O O O O 0 Differences between This Study and Studies . . . . . . . . .UO 0 o 0 DATA AND DATA SELECTION . . . . . . . . . . . 51 External Validity . . . . . . . . . . . . . . 51 Sample Selection . . . . . . . . . . . . . . . 52 An Example . . . . . . . . . . . . . . . . . . 59 Problems in Internal Validity . . . . . . . . 62 Further Sample Description . . . . . . . . . . 65 HYPOTHESES AND DATA ANALYSIS PROCEDURES . . . 85 Hypotheses O I O O O O O O O O O O O O O O 0 O 83 Procedures for Grouping Firms and Computing Returns . . . . . . . . . . . . . . . . . . 85 Statistical Tests . . . . . . . . . . . . . . 88 Tests on Mean Vectors . . . . . . . . . . 88 Tests on Covariance Matrices . . . . . . . 91 Risk and Information Effects . . . . . . . . . 95 Chapter Page VI. EMPIRICAL RESULTS AND INTERPRETATION . . . . 98 Tests on Mean Vectors . . . . . . . . . . . . 99 Omnibus Test of ACs . . . . . . . . . . 102 Tests for Effects of ACs with Different Directional Effects on EPS . . . . . 104 Tests for Effects of ACs with Varying De- grees of Discretion . . . . . . . . 109 Tests for Effects of ACs with Different Directional Effects on EPS and Varying Degrees of Discretion . . . . . . . . 115 Test of Discretionary ACs That Increase EPS--ACF (+,D) . . . . . 116 Test of Discretionary ACs That De- crease EPS--ACF (-,D) . . . . . . 123 Test of Discretionary ACs That Do Not Affect EPS--ACF (0,D) . . . . 126 Test of Nondiscretionary ACs That Increase EPS--ACF (+,N) . . . . . 155 Test of Nondiscretionary ACs That Decrease EPS--ACF (-,N) . . . . . 159 Test of ACFs Whose Discretionary ACs and Nondiscretionary ACs Increase EPS--ACF (+,B) . . . . . 145 Other Mean Vector Tests Not Dis- cussed Previously . . . . . . . . 148 Tests on Covariance Matrices . . . . . . 150 Tests for Equality of Unconditional Co- variance Matrices during the B- -Estima- tion Period . . . . . . . . . . 152 Tests for Equality of Conditional Co- variance Matrices during the Thirteen- month Impact Period . . . . . . . . . 154 Risk and Information Effects . . . . . . . . 156 Appendix A . . . . . . . . . . . . . . . . . 167 Table VIII - Summary Statistics for Tests on Mean Return Vectors . . . . . 168 Table IX - Summary Statistics for Tests on Mean Return Vectors - Continued . . 169 Appendix B . . . . . . . . . . . . . . . . 170 Results of Texas Which Correct for Dif- ferences in the Relative Risks of Matched Groups of ACFs and NCFs . . . 171 Part 1 - Problems in Matching Point Estimates of BB . . . . . . . . . 172 vi Chapter Subpart A - Sample Revision Subpart B - Tests Which Cor- rect for Differences between the Point Estimates of Paired B . . . . . . . . . Part 2 - ProbleEs in Using Esti- mates of fig . . . . . . . . . . . VII. SUMMARY, CONCLUSIONS, CONTRIBUTIONS OF THIS STUDY, AND SUGGESTIONS FOR ADDITIONAL RESEARCH O O O O O O O O O O O O O O O O O O Summary . . . . . . . . . . . . . . . . . . . conClUSions O O O O O O O O O O O O O O O O O Contributions of This Study . . . . . . . . . Suggestions for Additional Research . . . . . FOOTNOTES O O O O O O O O O O O O O O O O O O O O O BIBLIOGRAPHY O O O O O O O O O O O O O O O O O O O O vii Page 172 175 177 181 181 185 189 195 197 217 Table II. III. IVO VI. VII. VIII. IX. XI. XII. XIII. XIV. XV. T A B L E S BASKIN'S DECISION RULES FOR THE ANALYSIS OF INFORMATION IN THE ANNUAL REPORT . . . SUMMARY OF ACCOUNTING CHANGE STUDIES . . CALENDAR YEARS FROM WHICH ACCOUNTING CHANGE FIRMS WERE SELECTED . . . . . . . INDUSTRIES OF SAMPLE FIRMS . . . . . . . TYPES OF ACCOUNTING CHANGES BY ACCOUNT AFFECTED, DISCRETION OF MANAGEMENT IN MAKING THE CHANGE, AND DIRECTIONAL EFFECT ON NET INCOME . . . . . . . . . . . . . . NUMBERS 0F PAIRS 0F FIRMS AND AVERAGE 68 FOR HYPOTHESIS GROUPS . . . . . . . . . . DISTRIBUTION OF 61 0F SAMPLE FIRMS . . . SUMMARY STATISTICS FOR TESTS ON MEAN RE TURN VECTORS . O O O O O O O O O O O O SUMMARY STATISTICS FOR TESTS ON MEAN RETURN VECTORS - CONTINUED . . . . . . . NUMBERS OF THE VARIOUS TYPES OF ACs MADE BY ACFS (+,D) . . . . . . . . . . . . . . TYPES OF ACs MADE BY HIGH-RISK ACFS (O,D) DECOMPOSITION 0F ACs MADE BY ACFs (+,B) . SUMMARY STATISTICS FOR TESTS ON COVARIANCE MATRICES O O O O O O O O O O O O O O O O MARKET REACTION TO THE RISK CLASS MORE AFFECTED BY ACs WITH SIGNIFICANT F-VALUES IN THE TESTS ON MEAN VECTORS . . . . . . RESULTS OF TESTS ON MEAN VECTORS FOR DIF- FERENT OVERALL GROUP RISKS . . . . . . . viii Page 50 59 67 7O 72 -76 81 168 169 121 131 146 153 158 164 F I G U R E S F’l'gure Page .1.- CUMULATIVE AVERAGE RETURN DIFFERENCE - ACF (°,') - NCF O o o o o o O o o o o o o O O o o 103 23.. CUMULATIVE AVERAGE RETURN DIFFERENCE - ACF (+,') ’ NCF o O o O o o o O o o o o o o c o o 106 .35 .. CUMULATIVE AVERAGE RETURN DIFFERENCE - ACF (-).) - NCF o o o o o o c o o o o o o o a o o 106 4L «- CUMULATIVE AVERAGE RETURN DIFFERENCE - ACF (-,.) - ACF (+,.) o o o o o o o o o o o o o o 106 ES - CUMULATIVE AVERAGE RETURN DIFFERENCE - ACF (0,.) - NCF o o o o O o o o o o c o o o o O o 106 ES . CUMULATIVE AVERAGE RETURN DIFFERENCE - ACF (?)°) ' NCF o o o o o o o o o o o O o o o o o 107 .7 . CUMULATIVE AVERAGE RETURN DIFFERENCE - ACF (.’D) - NCF o o o o o o o o O o o o o o o o o 111 8. CUMULATIVE AVERAGE RETURN DIFFERENCE - ACF (.’N) - NCF o o O o o o o o o o o o o o c o o 111 9. CUMULATIVE AVERAGE RETURN DIFFERENCE - ACF (“’B) ' NCF o o o o o o o o o o O o o O o o o 114 :10. CUMULATIVE AVERAGE RETURN DIFFERENCE - ACF (+)B) - NCF I O 0 0 o o o o 0.. o o o o o o o 117 .ll. CUMULATIVE AVERAGE RETURN DIFFERENCE - ACF ('yD) - NCF o o o O o o o O O o o O o o o O o 125 12. CUMULATIVE AVERAGE RETURN DIFFERENCE (O,D) ' NCF o o o o o o O o o o o o o o o c o 128 I b 0 q 14. CUMULATIVE AVERAGE RETURN DIFFERENCE (+,N) - NCF o o o o o o o o o o c o o o o o o 156 I 11> O '11 15. CUMULATIVE AVERAGE RETURN DIFFERENCE - ACF (”)N) ' NCF o o o o o o o o o o o O o o o o g 140 ix Figure Page 18. CUMULATIVE AVERAGE RETURN DIFFERENCE (+’B) - NCF O O O O O O O O O O O O O O O O O 144 I > 0 w 13. CUMULATIVE AVERAGE RETURN DIFFERENCE (?)D) ' NCF O O O O O O O O O O O O O O O O O 149 I b O w 16. CUMULATIVE AVERAGE RETURN DIFFERENCE (O’N) ' NCF O O O O O O O O O O O O O O O O O 149 I b O W 17. CUMULATIVE AVERAGE RETURN DIFFERENCE (7’N) - NCF 0 O 0 O O 0 O O O O O O O O O O O 149 I b C q AC ACF API APB ATE CAPM CRSP Eq. FASB FYE ITC NC NCF NYSE SIC A B B R E V I A T I O N S Accounting change Accounting change firm Abnormal performance index Accounting Principles Board Accounting Trends and Techniques Capital asset pricing model Center for Research in Security Prices Equation Financial Accounting Standards Board Fiscal year-end Investment tax credit Nonchange Nonchange firm New York Stock Exchange Standard Industrial Classification xi C H A P T E R 1 INTRODUCTION In recent years several researchers have con- ducted studies on various aspects of accounting changes (ACs). Some of the studies have focused on the smoothing 1 of income. One reported the degree to which firms have complied with specific AC disclosure requirements.2 Others have considered the behavioral characteristics of firms which made ACs.8 Several studies have focused on consistency exceptions arising from ACs.4 The intent of another group of AC studies has been to determine the effect, if any, of ACs on the rates of return5 on common stocks.8 The present study falls within the last cate- gory; the purpose is to assess whether ACs and disclosures of ACs have an effect on stock returns. Justification for additional AC-Stock return research lies in (l) the many questions left unanswered (and largely unaddressed) by previous AC research and (2) the substantial refine- ments in research design applied in the present study as compared to earlier ones. The first chapter is a dis- Cussion of these questions and refinements. The extent to which this study addresses ques- tions ignored by other researchers does not imply a weak- ness in their studies. Rather, analysis of their designs and results has led this researcher to ask additional questions which were beyond the scope of these earlier studies. Hopefully this research will provide the im- petus for additional research into the effect of ACS on stock returns. Lev states: Empirical evidence generally indicates that account- ing changes had no significant effect on stock prices, implying that investors are able to recognize economic reality de8pite the different reporting modes. However, a qualification is warranted here; in a few cases, accounting changes seemed to have some effect on stock prices.7 Whether such effects are systematic and of importance is yet to be deter- mined.8 After reviewing essentially the same AC studies referred to in footnote six, Gonedes and Dopuch state: Summing up, the above studies' results are consis- tent with the statement that the capital market does distinguish between changes that appear to be report- ing changes of no economic importance and those that appear to have substantive economic implications. This inference must, however, be qualified because not all aspects of the conditional distribution, F(eitIQit): have been examined. In addition, the studies suggest that some types of accounting changes (e.g., those examined by Sunder and Ball) are associ- ated with events affecting firms' relative risks. The nature of this association has not been thor- oughly investigated.9 Lev raises three issues: (1) whether the effects of ACs on stock prices are systematic; (2) whether the effects of ACS on stock prices are of economic importance; and (5) whether the market intelligently prices stocks of accounting-change firms (ACFS). Gonedes and Dopuch raise two additional issues: (1) the need to base in- ferences on an examination of more aspects of the dis- tribution of stock returns of ACFS than just the first moment, the mean; and (2) the nature of the association between ACs and firms' relative risks. Each of the five issues deserves consideration in order to eXplain the scope, the limitations, and the inferences to be drawn from this study. The design tests directly whether there is a contemporaneous association between ACs and unusual stock return movement. It was constructed to substantially reduce the effects of confounding variables and to test the market effects of ACs which had specific directional effects on primary earnings per share (EPS) and/or which were discretionary or nondiscretionary. Such refinements concerning the effects of ACs have been largely over- looked in previous studies. The economic importance of the effect of ACS on stock prices is complicated by the fact that there are a number <>f very different potential economic factors which may be associated (either causally or effectually) with an AC. For example, an AC may result from a firm's shift into a high technology field where accelerated de- preciation is more appropriate than previously-used straight line. The market's perception of that firm's inability to compete in the new field may cause the firm's stock price to drop contemporaneously with disclosure of the AC. In this hypothetical case the economic effects are very real, both internally—-from the possibility that the Shift may Significantly impair the firm's competitive position, and externally--from the effect this negative possibility may have on the firm's ability to obtain financing. In this case the AC is associated with a market effect and disclosure of the AC may be a significant part of the message to the marketplace that the firm has made a Shift in its production-investment_decision. 0r a1- ternatively, the AC may only be a by-product of the shift and merely be contemporaneously associated with the un- usual stock price movement. Another example of an economic effect of an AC is the change to LIFO inventory costing. In most cases this AC will actually cause a net cash flow increase be- cause of the reduction in taxable income. This internal economic effect could be associated withanzleast two opposite external effects, a reduction in stock price due to lower reported profits or an increase in stock price due to the higher real profits from reinvestment Of the cash saving.lo Here the AC actually causes the increase in cash flow which may affect stock prices. A third example of the economic importance of ACs is the firm which changes from the cost to the equity method of accounting for long-term investments in stock. Generally, if the investee earns more profit than it dis- tributes in dividends, the investor increases reported profits as a result of the change. This AC, which appears to have no internal effect on the firm (in terms of cash flows or other economic betterment) and which appears to convey no economic message about the firm (e.g., in terms of its production-investment decisions), may nevertheless 'have an economic effect insofar as the increase in re- ported profits makes it possible for the firm to obtain financing. In this example any economic effect associated with the AC is not induced by a Shift in production- investment decisions or altered cash flows, and yet there may be a very real effect on financing, which in turn may affect the firm's production-investment decisions and hence the return on the stock. The preceding discussion should make it apparent that the effect of some ACs on security prices is compli- cated in some cases by the presence of exogenous non- accounting factors. Accordingly, one must be cautious in interpreting the results of market-based research in this area. Many other examples could also serve to illustrate the difficulty in making general statements about the economic implications of ACS.11 Lev's third issue is related to the second be- cause an assessment of market reaction to an AC depends in part, at least, on the perceived economic substance of the AC. Therefore, the inherent difficulty in assessing the economic substance of an AC makes it equally difficult to make normative statements about how (or whether) the market should react in a given Situation. Consequently, the present study deals only indirectly with the second and third of Lev's issues. This will become apparent in the interpretation of the empirical findings of the study. The first issue raised by Gonedes and Dopuch, namely, the need to examine higher moments of the dis- tribution of returns in order to draw inferences about the underlying events or information transmissions related to ACs, is incorporated into the design of this study in that tests are performed on variances as well as on means. Except for Baskin's study previous AC research examined only distribution means. To ignore the higher moments is to say that they are not important to investors' decision processes, and this is simply not true. To illustrate, consider an investor with $100. He may choose between two stocks A and B which both have eXpected returns of $75. However, the return on A has a standard deviation of $25 and the return on B has a standard deviation of $100. The investor’s decision is not obvious, but it is obvious that the risk-averse investor should consider the variability of returns because stock B offers the greater probability of losing the entire $100. Skewness is not a consideration as long as return distributions no *1 I D (u <4 '5‘ B (b H. HI } .J 'cal, and monthly return distributions have ‘enerally been found to be symmetrical.l2 The point made by this simple illustration is supported inOthe theoretical work of Tobin and Richter, among others. In Tobin s work he gives an example of a risk-averse investor's utility map that is composed of quadratic utility functions that are concave to the ex- pected return axis (in a mean-variance Space).13 And Richter has shown that only the first two moments of the distribution Of returns are needed by the risk-averse investor with utility functions of degree three or lower and who views a symmetrical (or two-parameter, e.g., the normal) distribution of returns.14 Empirical support for the need to consider stock return variability is pro- vided by the fact that investors seek information about Betas (BS).15 Gonedes and Dopuch's second issue, the associ- ation between ACs and firm riskiness, has at least two facets. One is the possibility that ACs actually affect firm risk. This is one implication of the work of Ball and of Sunder. If ACs do affect firm risk, the other AC studies' results may be suspect because of their failure to control for the effect of this phenomenon in their designs. All the other studies assumed constant risk, which became the independent variable in their assumed return-generating functions. To the extent that the constant risk assumption was untrue, the dependent vari- ables (i.e., abnormal returns) in these studies were mis- specified. There appear to be two ways to control for a changing risk. One is some kind of moving average risk estimation procedure, which takes into consideration shifts in firm riskiness, and the other is the use of a control group whose risk is the same as the risk of the groups of ACFS. In this study the latter method is employed. Another facet of Gonedes and Dopuch's second issue is incorporated into the present design in that tests are performed on firms having different relative risks in order to determine whether there is a unique association between the effects Of ACs and specific risk classes. It is possible that the effect of ACs on stock prices is risk-dependent in the sense that there is an AC effect for firms in specific risk classes. For ex- ample, the market may react more demonstrably to high- risk firms' ACs which could be construed to be motivated by the desire to manipulate net income than to similar ACs made by low-risk firms due to the inherently greater riskiness associated with high-risk firms. 0r, following a different line of reasoning, if ACs are associated with changes in firms' risk levels, this change could affectiflmeearnings rate used to discount firms' future cash flows to a present value of the firm. If this is the case, then, other things being equal, the effect of an AC on the market value of low-risk firms will be greater than on high-risk firms. This possibility will be explored more fully in Chapter 5. Earlier AC research was not designed to detect any such risk dependence. AS a result, it is not known whether such an association exists. Thus, in summary, the present study was con- ducted in an attempt to overcome some of the limitations 10 in prior AC research and to implement some of the exten- sions suggested in the more recent literature. Specifi- cally, the association between the information content of ACs and particular risk classes of firms is examined. And tests are conducted on variances as well as on means. Furthermore, the use of a control group, in addition to the above refinements, provides a framework within which to assess whether ACs have a systematic effect on stock returns. .There are many different ways one can approach the determination of whether ACs have a systematic effect on stock returns. One way is to test whether all types of ACs aggregated together have an effect. This is es- sentially what Ball and Baskin18 tested. Another way is to test whether ACs involving Specific accounts have an effect. Archibald tested depreciation, Kaplan and Roll tested depreciation and the investment tax credit, and Sunder tested inventory. A third way is to test whether ACs with predictable effects on net income systematically affect profits. With the exception of the Ball and Baskin studies, previous AC studies provided tests of whether ACs with Specifically predictable effects on net income systematically affecuuistockLreturns. But none of them provided tests of the market effects of ACs with every 11 different directional effect on net income. The present study does provide tests of whether ACs having all types of directional effects (positive, negative, zero, and directional effect not disclosed) affect stock returns. A fourth way to approach the determination of whether ACs systematically affect stock returns is to test whether discretionary ACs and nondiscretionary ACs are viewed differently by the market. Archibald's, Kaplan and Roll's, and Sunder's studies all provide tests of discretionary changes involving Specific general ledger accounts. Ball's study aggregates the effects of dis- cretionary ACs and nondiscretionary ACs and thereby fails to test for the systematic effect which may be uniquely associated with each type of AC. The present study tests whether each of three levels of relative discretion in making ACs has a systematic effect. The three levels are discretionary, nondiscretionary, and both (for firms which in the same year made at least one discretionary AC and at least one nondiscretionary AC). In addition, the four levels of directional effect (positive, negative, zero, and directional effect not disclosed) are multiplied by the three levels of dis- cretion to include more refined tests of direction and discretion interactively, e.g., positive-discretionary, 12 negative-discretionary, . . ., directional effect not disclosed-—both discretionary and nondiscretionary. In this way tests can be performed on ACs having virtually all directional effects on net income and/or virtually all levels of discretion. There is one refinement in the present study, the use of a control group, that is found in Baskin's and in Kaplan and Roll's studies but not in the other AC studies. In the absence of a control group one wonders whether the observed behavior of the dependent variable is unique to the group being investigated or whether similar (nonunique) results would also have been obtained for a control group. Another refinement lies in the assumption of a multivariate normal distribution of stock returns. The multivariate nature of the return distribution provides for the inclusion of covariances in overall return vari- ability. This provision in turn allows a researcher to more convincingly infer statistically significant differ~ ences (shouId they result from the data) than is possible when a univariate distribution is assumed. All previous studies made the assumption of a univariate return dis- tribution. The literature review chapter will discuss in more detail the use of a control group and the advantages of the assumption of a multivariate return distribution. C H A P T E R 2 THEORETICAL FOUNDATIONS AND EMPIRICAL SUPPORT FOR MODELS EMPLOYED Chapter 2 is in five parts. The first four parts briefly summarize information content, efficient capital markets, capital asset pricing, and the distribu- tional prOperties of security returns. Part five inte- grates all four notions as they apply to the present re- search. Information Content The purpose of this study is to determine whether ACs and their disclosures have an effect on stock prices. Stated differently, the purpose is to determine whether ACs have information content. The latter (and equivalent) statement of the purpose is useful for providing a frame- work within which to test the hypotheses addressed here. An event y has information content if the con- ditional distribution r(st) is not identical to the uncon- ditional distribution f(s) for some S, where s is a random variable. Obviously for an experiment to have substance, 15 14 there must be some reason for believing that y and s are related in some way. If f(sly) = f(s), then y does not have information content because its presence is not associated with changes in s; in short, y makes no differ- ence to agents whose activities impact on s.1 This gen- eral notion of the concept of information is incorporated into the present study in the paragraphs that follow. Inferences can be made about the information con- tent of accounting events y by observing the behavior of some random variable 8 during a period of time when it is reasonable to believe the accounting event may affect the random variable. There is the need to identify three items: the event, the random variable, and the time period. The event y being considered here is an AC. The random (dependent) variable s to be examined is the stock return movement of ACFS. The association between the event and the random variable results from a prior belief that disclosures of accounting events are functionally re- lated to stock return movements because accounting dis- closures represent one element in the information set available to investors. Empirical research supports this belief.2 Selection of the relevant time period could present difficult problems if there were no evidence to indicate when the effects of accounting events are CI impounded in stock returns. However, evidence concerning the speed with which the securities market impounds infor- mation provides relevant insights in this regard. Efficient Capital Markets In order to assess the information content of accounting events (and disclosures of accounting events) one must identify some period of time within which any po- tential information effects could be expected to be mani- fest. Evidence supporting market efficiency suggests that security prices adjust instantaneously to new information once it becomes publicly available. Accordingly, one potentially relevant time reference for studying the ef- fects of ACs is the period immediately following the dis- closure of the change. Fama provides a summary of the theory and evidence of the efficiency of capital markets.3 Based on the findings of previous research (see footnote 2) it appears that the information effects of some economic events that are reflected in accounting num- bers (dividend announcements and preliminary reports of income numbers, for example) appear in stock prices.sev- eral months before public disclosure. The same could also be true of ACs. Some firms announce their intentions to 16 make ACs prior to release of the annual report. Other firms announce after the fact but before release of the annual report that they have made ACs.4 A third group of ACFS wait until the annual report to disclose their ACS, but the information content of some of these ACS (such as depreciation changes due to a shift in the firm's production-investment activities or changes induced by Accounting Principles Board directives of which market agents were aware) may have been impounded in stock prices prior to release of the annual report. Thus, in addition to the period immediately following disclosure of the AC, the period immediately preceding the disclosure of the AC is also examined. Sunder; points out that certain studies designed to measure the relationship between accounting information and market prices have stated that their evidence supports the efficiency hypothesis.6 At the same time these stud- ies relied upon capital market efficiency to measure the contemporaneous relationship between stock prices and ac- counting events. This type of circular reasoning has caused some confusion. Either the research design makes an assumption about the effects of accounting information and tests for market efficiency, or it assumes efficiency and tests for effects. It appears logically invalid to 17 test hypotheses about the efficiency of capital markets (which makes assumptions about the effects of events) and simultaneously to test hypotheses concerning the effects of events (which makes assumptions about the efficiency of capital markets). Accordingly, the present study as- sumes capital market efficiency and tests for the effects of accounting changes. Information theory provides the definition of information, in terms of differences between distribution functions. Market efficiency provides the temporal set- ting for the assessment of the effect of accounting events on stock prices. But one question still remains: How does one assess the effect of accounting events on stock prices? Capital Asset Pricing The normative capital asset pricing model (CAPM) of Sharpe7 provides an equilibrium expectation of the re- turn from holding an asset (such as a stock). It is pos- sible to compare various aspects of the unconditional expected return distribution to the same aspects of the realized return distribution conditioned on some informa- tion set.8 Assuming that the effects of other competing 18 factors have been controlled, differences between the two distributions represent the information content of the event under investigation. This method fits the formal definition of information presented earlier. It was em- ployed in all the AC studies referred to in footnote six of Chapter 1. The CAPM, plus the empirical evidence on the ef- ficiency of capital markets, provides a well-defined frame- work for assessing the information content of ACs. The form of the CAPM assumed here is the one proposed by Black9 and tested by Black, Jensen, and Scholeslo and by Fama and MacBeth.ll Jensen12 provides a summary of the various forms of the CAPM and tests thereof, as well as the model's assumptions, which are adopted here. The model is given by the expression: EQ' l Ei-‘xln-O i. ICCUQ(-. 39 .au»aano mung uo coauuom and: com: N®_muummémw Huston monomoHomuc ucm ouaaaa> o< can mo< 0» new» huaum .o sopamcu com mmémfiiflw—fim tam was» no» no» maoHan> -oaoa unsuns onweaxm ucomoam .omCOQnoa uoxaas ans omHA song -socpm on on: omHm o» weawcnno Hana “mate: omHm ca momcano sumac «snag "mooaan swan madeLOC onw ounaaa> omHm song can au< couauon Annmav use on: omHA o» mauwcmno msaum INH n«:: no» oz oz QLHA ca COwum«00mma magnum: aoocsm Oaaulu Hashes who: go .ucoauauuo ma poxaas amnw sumasm> «sauna newsman a“ ho "moofisa xooum poouum no: on mo< IHH vac: mo» oz 0: msofium> ncuuoauuo uoxaas have Hana .oumc u:o&ooc:0:¢a mwc«csmo oczoaa beamed xoozuoo .:0aum«ooan mafiuco cm so>o moowsn finance .00 vegans -onon can cofiuaaooaaoo Am 0» uaooom scam wcwmcmzo msawm .mxooz ma was: an 09 “09H on» Low mmoaaq unsaOCaonop cannon ho anyhow can some ucoeoocsoch mwcficsao mMunu oaauoesmm nob soak ocaosm mxoor OH sou ”woman onw sumasn> zwsoacu mo< go Haom ecu Hassocuo>on~ um: mammcmno 09H lam vac: oz no» no» uroHu as vacuum assume saunas: madam: on» Hashes mcoHunooxu mono» .mcoflunooxo accoumwmcoo o» uHoH oumwaa> -mumc00 .maOuuumm mo massages noses no: moon uoxumz 1 van: 0: new no» msofiam> outage assume spawns: cmemm .UOwst pmohuw acoscomnsm ceauawo :w madman Hashes use u< oaouon Haste: imaaou noun ceased Laohim :u mooasq Hesse: u oumaaa> uaoflooon mo< ou cofiu isomon on: menu» xomninouuzm .W nus: 0: oz on soak Am 08 tomes uoxans onuenxm eflmnwcoa< «u .A .u .wnuv. nnan. 0.8 DZu “—3. .d V o s a n 1.: u;do a.» J o u. n n w n m mum “am mm am. an n .q. r. I u o J m 9.0.“ a.» .d J 1.9. o o n a. D uvra tévo so 0 u o B J 8 v. a . D u T. .wrv r. a s a I. a u 9.0 a . D o 3. .Jr. 3 w .m. m : nHHanm moz d(_ _O> All 1: erms in Eq. 6 and Eq. 7 are as previously defined except N, which is the sample size, and Sd, which is the sample covariance matrix of the mean difference vector E. The elements of Sd are defined i’niEq. 8, where H denotes high- risk and L denotes low-risk. The information condi- ticfling argument, _Q, is omitted from Eqs. 6-8 to avoid Clutter. Var(d ) Cov(d .d ) Ems S _ H H’L Cov(dH,dL) Var(dL) The sample size N in Eq. 6 and Eq. 7 is the number of monthly observations (either six or thirteen) 90 used for estimating:EL§. A condition necessary for vir- tually all statistical applications is'that the sampling units be independent of each other. In this case, the sampling units are monthly return Observations, and the weak-form tests of market efficiency provide evidence that monthly price observations (and price changes) are very nearly serially independent. A further discussion of weak- form market efficiency can be found in Fama.a Eq. 6 reflects the fact that the multivariate statistic T2 forms a linear combination of the high- and low-risk elements in the difference vector E. The weight vector 1 can take on any set of values desired, but the r T2 statistic used the value of Kr which maximizes the value of T2. Thus the multivariate T3 is nothing more than the maximum squared univariate t(1r)-statistic. In utilizing the maximizing value of Er, the T2 implicitly considers all values of yr. The statistic t2(3r) is dimensionless and is unaffected by a change in the scale of Er’ Therefore, «Er can be normalized so that wl + wz = 1 by imposing the constraint,li'rsdjlr = 1, on the maximization done in Eq. 6. Then the normalized values of the elements Offlr can be -1.“ determined by multiplying the vector E = Sd (E “'50) by 91 R the scalar 1/ Z x r=l W1 and wz can then be viewed as legitimate portfolio r’ The resulting normalized values of weights. When the null hypothesis (Eq. 4) is true, Eq. 9 F =§E('N-—-R‘i_)' T2 has the F distribution with degrees of freedom R (= two, for the number of risk classes) and N - R (where N a six or thirteen). Departures of Edifi frompO increase the mean Of T2 so the decision rule is to accept the null hypothesis if the Observed F3value falls beneath the area of the F distribution corresponding to one's desired level of significance. Otherwise, reject the null hypothesis and infer that the ACS have information content. A thorough description of the T2 statistic can be found in Morrison4 or Anderson.S Tests on Covariance Matrices The test for equality of covariance matrices employs the M statistic prOposed by Bartlett: G Eq. 10 M = (ng)loseISl - gglngloselsgl, 92 where G is the number of groups being compared-~two in this case, 1 = ACF and 2 = NCF; n8 is the number of month- ly Observations used to estimate Sg (i.e., sample size Ng) - 1; S is the arithmetic average of the two sample co- variance matrices, 81 and $2 (in Eq. ll'below); and the parallel vertical lines denote a determinant. The elements of Sg are defined in Eq. 11. Var(RgH) Cov(RgH,RgL) ) Var(RgL) . Cov(RgH,R gL An important distinction between the sample covariance matrix Sd in the test on mean vectors (in Eqs. 6-8) and the sample covariance matrices S8 in the test on co- variance matrices (in Eq. 11) is that Sd is computed on the distribution Of return differences, while the S8 are computed on the raw return distributions of groups of ACFs and NCFs. Reference to Eq. 5 indicates that the multi- variate test for equality Of covariance matrices is per- formed without weighfln.g In the multivariate context, r0 simultaneous tests are performed on the differences Var(RlH) - Var(RZH), Var(RlL) - Var(RZL); and C°v(R1H’R1L) - Cov(RZH,R2L). Introduction of the weight vector,1i,into the covariance 95 matrix test reduces the null hypothesis of Eq. 5 to a univariate hypothesis because ggziigigr is a scalar repre- senting the variability of the return on only one asset. This asset is a linear combination of the returns on high- risk and low-risk firms, and the process of combining the two eliminates the duality necessary in the present study for a multivariate test. Reference to Eq. 10 indicates that departure of any of the three different elements of S (or 82) from its 1 corresponding average element in S causes M to have a non- zero value. The M statistic represents Bartlett's applica- tion of the generalized likelihood ratio criterion,.and in it the determinants of the sample covariance matrices as- sume the role of generalized variances. Morrison6 includes a discussion of the M statistic. For samples Of around twenty or more, M can be transformed into a chi-square random variable. For smaller samples, which apply to the principal tests conducted here, Box7 has prOposed a transformation of M into an F random variable with degrees of freedom fl = 1/2(G — 1)R(R + l) and f2 = (fl + 2)/(A§ - A2).8 The transformation is given by Eq. 14: 94 E 14 F sz q- = ’ where b = (f2)/(1 - Al + 2/f2). Box also presents evi- dence_that the F approximation is very good for the case of G = 2, R = 2 (the dimensions Of the present data), and n = 9 or more. He presents no evidence on samples Of less than ten. Therefore, the goddness of fit of the data to the F-distribution is not known for the tests conducted over six-month periodsin this study. Large (enough) Observed F-values lead to rejec- tion Of the null hypothesis Of no difference in variability, and hence to the inference Of information content in ACs. Values of F which fall below the selected fractile of the null hypothesis true distribution of Ffl’fz lead to ac- ceptance of the no information null hypothesis. The Chapter 2 section, Integration of Theories and Evidence, discussed the effect thatEl =-22 has on one's ability to infer that differences between zlLQl and ZZLQZ are due to E3. Because the condition-pligp2 says nothing about the equality of unsystematic variability of 21 and 22, additional steps must be taken in order to be able to attribute conditional variability differences to ACS. 95 It appears that the most straightforward way to approach the problem is to directly test whether 21 = 22 for the sixty-month period over which Bi were estimated. If 21 = 22 prior to the thirteen-month impact period, then Zngl % ZZLQZ during the impact period indicates that ACs possess information content. On the other hand, 21 # 22 indicates that the two groups, unsystematic vari- abilities were different prior to testing, and it will not be possible to attribute a nonzero zngl - ZZLQZ difference to Q1, even though'El.gIQ2. In this event, the test for equality Of conditional covariance matrices will not have much meaning. The appropriate test of Ho: 21 =22 also utilizes Bartlett's M-statistic, but the M statistic is transformed into a (large-sample) chi-square random variable with degrees Of freedom 1/2(G - 1)R(R + 1) because sixty monthly observations are used to estimate the M statistic in this case. The specific transformation is discussed in Morrison.9 Risk and Information Effects In order to determine whether ACS possess in- formation for particular risk groups, the analyses were 96 performed using different values of Er' The T2 test utilizes the]!r value which maximizes the value of T2 and thereby tests all values Ofufl against the null hy- r pothesis. Therefore any other 1 value will produce a r T2 value that is less than the value observed in the test on mean vectors. By varying 1r, insight can be gained into the existence of any risk dependency by observing whether dif- ferent T2 values result from different values of Er' Such tests on conditional mean vectors are carried out in the context Of the multivariate test because of the all-inclu- 3 test. The use of dif- sive nature of the maximizing T ferent‘gr values in the test for equality of conditional covariance matrices involves a series of univariate tests, because 1£2L§gr reduces to a scalar (i.e., the variability of one asset). In addition to the value of._w_r implicit in the T2 statistic, the following values of._w_r were used: .3; = [1 o]; g; = [O l]; 1; = [1/2 1/2]; and 1; = [x y], where x and y produce an overall group 68 of one, and x + y = 1. The vector 1; = [1 O] is analogous to a "high- risk" portfolio with 100 percent Of the assets invested in the risk-group of firms with the higher 8g.~ The 97 vector 1; a [O 1] is analogous to a "low-risk" portfolio. The vector 1; = [1/2 1/2] is analogous to a portfolio with equal investments in the high- and low-risk groups Of firms. The vector 1; == [x y] was also used because the ACFS in the sample had an average risk coefficient greater than one. If the average had been one, then 1; = [1/2 1/2] could be expected to produce an "average-risk" portfolio with (3;: 1. The vector 31', = [x y] is included to ac- complish this. In summary, a difference between returns of ACFs and returns Of NCFs on either parameter, Edig or ELQ, is sufficient for rejecting the no information hypothesis. Therefore it takes equality on both parameters to infer no information content in Ads. The design is multivariate in that tests are performed simultaneously on high- and low-risk groups of firms. In order to know whether in- formation effects, if any, are uniquely related to one of the risk classes, tests are performed using different values 0f.flr: which correspond to different overall group risk coefficients. The risk effect tests are multivariate on mean vectors and univariate on variances. C H A P T E R 6 EMPIRICAL RESULTS AND INTERPRETATION Chapter 6 is divided into three parts. The first part covers the tests on mean return vectors, the second part the tests on covariance matrices of returns, and the last part the association between risk and information ef- fects of ACs. All tests were performed on data for three time periods: t = -6 through -1; t = +1 through +6; and t = -6 through +6, where t = O is the post-year-end month when the AC is disclosed. The third, all-inclusive period tests whether ACs and AC disclosures are associated with unique market effects during a thirteen-month period around the month when the AC is disclosed. The two subperiods are of interest because effects may appear during either or bOth of them. Separate analyses on the two subperiods can indicate the relative timing of the information impact on the market. The first subperiod, prior to the AC dis— closure month, reveals whether the market anticipates in- formation associated with ACS prior to firms' disclosures of ACs on alternatively, whether there was a unique market effect that might have been associated with firms' de- cisions to make ACS during the latter part of a year. The 98 99 techniques used in this study will not provide evidence as to which is the case, if indeed unique market effects ap- pear during the first subperiod. The second subperiod, which follows the post-yearbend disclosure of ACs, reveals whether the market reacts to AC disclosures in the pre- liminary earnings report and/or the annual report. Throughout Chapter 6 the two subperiods will be referred to as "first" and "second." Also, the notational scheme with respect to different categories Of ACs and ACFs that was explained immediately prior to Table VI will also be retained in this chapter.1 Tests on Mean Vectors Tables VIII and IX (in appendix A) give the high-risk and low-risk average return differences,2 the standard errors of the estimates of average return dif- ference, the risk class weights implicit in the T2 sta- tistic, and the related F-values of the T2 tests. In Tables VIII and IX each of the eighteen hypothesis groups is identified with a specific line number, and each line contains the results of tests for the respective hypoth- esis group across all three periods. In addition to the tables, Figures 1-18 show plots of the cumulative average 100 N return difference (CARDt) over the thirteen-month period of the stud» as well as the number of pairs of firms used for estimation. Cumulative average return difference for month t is defined as: Eq. 15 CARDt = z gtlp , where EEIQJ (Eltlfll - EZtLQZ)’ is the 2 x 1 average return difference column vector equal to the difference between the two 2 x 1 average return column vectors of a pair, all conditional on a value Of'Ei for month t. The elements in the vectors are groups of high-risk firms and groups of low-risk firms. Tables VIII and IX also give the num- bers of pairs of firms pertaining to the test on each hy- pothesis group. In interpreting the figure related to a given hypothesis test, it is important to keep two things in mind. One is the fact that the average return differences plotted in the figures are cumulative, and except for the first monthly average return difference plotted in each figure, do not represent the monthly average return dif- ferences used in the statistical tests. The second is that the average return difference on high-risk firms is not formally compared in this study to the average return 101 difference on low-risk firms(although such a comparison may be apprOpriate in assessing the different market ef- fects associated with the ACs firms in different risk classes). Each Of the two risk—class plots in each figure represents a difference--between returns on ACFS and re- turns on NCFs. The related statistical tests make this comparison, between high-risk ACFS and high-risk NCFs and between low-risk ACFs and low-risk NCFs and not between the average return difference on high-risk firms and the average return difference on low-risk firms. Therefore, each risk-class plot should be visually compared to the horizontal zero-return line. Furthermore, it will become apparent that selection of the two six-month test periods examined here produces test results that differ from the results Of similar tests that could have been performed on average return differences Over different segments of the thirteen- month period. The rationale for selecting the three time periods examined in this study has been discussed pre- viously. In most cases where statistically significant differences occur, the risk-class weights in Tables VIII and IX provide a signal as to which risk class contributes more of the overall multivariate return difference to the 102 value Of the resulting F-value. Accordingly, discussions of the tests with significant differences are identified with a particular risk class. The related plot of the cumulative average return difference is also helpful in identifying information effects with a particular risk class. Where needed, tables showing a further breakdown of a group of firms' ACs on the basis Of accounts affected and percentage effect on EPS are also given. Omnibus Test Of ACs Results of the omnibus test of the information content of ACs are consistent with the no information hy- pothesis. See line one of Table VIII and Figure 1 below. This high-risk cumulative average return difference in the figure remains very close to the zero horizontal line. In the case of the low-risk firms, the market effect seems to be more pronounced during the second subperiod; how- ever, the variability in the second subperiod average re- turn difference is large enough tO render it statistically insignificant. (Note in particular the increase which oc- curs in month t = +5 to reverse the previously downward drift in the low-risk cumulative average return dif- ference.) Consequently, the vertical return distance for 105 lG-C 20f0 12.0 4-00 -4-00 -|6.0 -12.0 -8.00 I l 9-0 -2 l L i l l i l l l l L l -s.oo «.00 44:0 .0 2.00 4.00 LE Ficus 1. Coumvs AVERAGE m DIFFEREMI - NIH.) - [CF 560 pairs of firms 104 low-risk firms from t = +1 to +6 is statistically nominal. The Observed F-values in Table VIII for all three periods are substantially below the seventy-fifth fractile Of the appropriate F distributions. This test is essentially an updated replication of Ball's study, plus a separation of firms into risk classes and the use of a control group Of NCFs. The results Of tests performed on firms that made more homogeneous types of ACs suggest that the aggregation in the omnibus test cancelled significant effects related to several specific types Of ACS. As mentioned previously, the aggregation problem was a source of motivation for this study. Therefore, this result comes as no surprise. Tests for Effects Of ACs with Different Directional Effects on EPS As mentioned previously, there is evidence to indicate that the sign Of changes in net income is an im- portant piece of information to market agents.3 In fact, many of the stock valuation models of fundamental financial analysis capitalize net income or some derivative thereof to predict equilibrium values of firms.4 It follows, therefore, that since all ACs have Specific directional effects (+,-,0, or ?) on the net incomes of ACFS, ACS 105 with particular income effects may be associated with unique stock market behavior. However, results Of tests conducted on mean vectors to ascertain the information content of ACs with the aforementioned directional effects on EPS, when aggregated across discretionary and nondis- cretionary ACs,are consistent with the no information hy- pothesis. (See lines two through six Of Table VIII and Figures 2-6 below.) In some cases the average return dif- ferences depart from zero by wide margins, but in no case is the multivariate set Of differences large enough in re- lation to its variability to infer that ACs with different directional effects on income are associated with unique market behavior. The hypothesis test of differential market be- havior toward ACs that decreased net income and ACs that increase net income (see line four of table VIII and figure 4) was included along with the tests of ACF-NCF because of the possibility that the effects of ACF (+,-) and ACF (-,°) might be insignificant when compared to NCFs but significant when compared to other ACFS that im- plemented changes with the opposite signed effect on in- come. ?However, the result of this test, like that of the other tests of ACs with different directional effects on income, assigns a high probability to the null hypothesis (Eq. 4).5 16.0 20,0 0 IL!) 12 .00 '. Him-R151: V\ /) Lon-l)! 5x / \ A A 8 LOO-RISK 8 m .5 P .‘ t- O O D C) ‘5 ” s — O O a )- ~ _ T T D D g )— g _ O. 0 3-0- Col-J nJiilil liLili 711.111 111L411 .0100 «.00 -2.00 .0 2.00 4.00 5.00 «.00 «.00 4.00 .0 2.00 4.00 0.57 W W has: 2. umnvs Am Rents Disrespect - hinge) - in Flat: 3. Cmunvs m m 01mm - MI(-.-) - III 275 pairs of firms 43 pairs of firms a O 61.- 3-9- N N mu FERENT a O o '- g "— c! N '- 2‘ '- 2 __ s _ «5 o i / “lat-RISK - Him-Risk s _ , - 8 _ Lot-Russ T 0 °. ? b ‘3 N r- 7 °. to ._ 7 °. 4 N :_ 1 1 L l 1 1 1 J J 1 u l 1 l L 1 J l l I L l l “'o¢o «.00 -2.00 .0 2.00 0.00 030 «1.00 «.00 4.00 .0 2.00 0.00 0. m m f,“ ‘I- WNW m m 0|"ng - ”(3-) - 5“,.) FIG!!! 5. (MIN! m m DIM ' ”(0.“ - H 30 pairs of firms 142 pairs of firms FERENT 107 l2-0 18-0 20%;!) 8.00 T "lav-RISK )4 O O ‘1— o I 9 2+- I O o— l °. 9» l l 1 l in i L i i 1 l - 00 « 00 -z 00 .0 2 00 4 00 etc Hans 6. ununw m m Dim - ACF(?.') - MI 100 pairs of firms 108 All the tests Of ACs with specific directional effects on EPS were aggregated across all different cate- gories of relative discretion. As a result, these tests for the most part are not directly comparable to previous AC research. For example, the investment tax credit and the depreciation method ACs that Kaplan and Roll examined increased net income, but they were all discretionary ACs. The changes to LIFO and the changes to FIFO examined by Sunder decreased and increased reported net income, re- spectively, and they too were all discretionary. Archi- bald's changes to straight-line depreciation were all dis- cretionary. The only aspect of the directional effects tests that is comparable to earlier AC research is the set Of tests over the first subperiod. Ball conducted chi- square tests for association between the Sign of the in- come effect and the sign of the cumulative API up to month t = 0.6 He detected no statistically significant relationship. The present findings tend to corroborate Ball's results. Further refinement of the directional ef- fect analysis on the basis Of whether or not ACs were dis- cretionary suggests that the tests discussed above Obscure tile information effects associated with ACS more specif- 1.Ctally identified than just by the directional effect on IEZIDES. 109 Tests for Effects of ACs with Varying Degrees of Discretion In Chapter 4 it was hypothesized that the mar- ket may react differently to discretionary and to nondis- cretionary ACs. It was suggested that the possibility for manipulation of income is likely to be substantially greater for discretionary ACs than for nondiscretionary ACs. For example, the stock Of firms that make discre- tionary ACs which increase EPS in order to conceal other- wise unacceptable Operating results may be spurned by the market. On the other hand, the market may view firms that make nondiscretionary ACs with positive income effects differently inasmuch as they have been compelled to change their accounting techniques. Perhaps the imposition Of an AC could cause firms to alter their production and invest- ment activities. This shift could, in turn, be considered by market agents to be of substantive economic importance to the firm, and this may give rise to unique market be- havior. The results of the tests of the information con- ‘tent of ACs with varying degrees Of discretion generally .sxlpport the no information hypothesis. The test for the effect of discretionary ACS and the test for the effect CDJf’ nondiscretionary ACs both indicate that such ACs, when 110 aggregated across different directional effects on net in- come, make no difference to market agents. (See lines seven and eight of Table VIII, as well as Figures 7 and 8 below.) Thus, it appears that aggregations of discre- tionary and nondiscretionary ACS, like the aggregations of ACs with specific directional effects on EPS, bring tOgether ACs that are dissimilar enough to Obscure the ef- fects associated with a more narrowly defined groups of ACs, if indeed there is unique market behavior toward some types of ACS. One may wonder why the low-risk average return difference in Figure 7 is not statistically different from zero in view of the fact that the cumulative average re- turn difference reaches a value of around -.07 by month t = +6. Two explanations can be given, both of which were alluded to earlier. First, most of the nonzero be- havior occurs from t = -1 through +4. But no test was performed on the average return differences pertaining to this particular six-month period. The average dif- ferences over the last six-month period (from t = +1 to +5) were subjected to testing, but (the vertical distance beetween the cumulative average return differences at t = +1 and at +6 indicates that) the total return dif- f’€3:r-ence over this six-month period is only around negative PERI)" ~12.0 IJJL ZOIO IG-O 12-0 0 C3.— \\\\Efi:fiff\\\\V///—.—" \L -0 m -I2.0 -8-00 -4-00 I -lG-0 I L..- ~2P-0 l I L I I I l 4 l J— 1 I 3.00 «.00 -2.00 .0 2.00 4.00 0.50 . FINI‘H Finn: 7. Cuwmvs Aim Rams Dlmm - AIR-.0) - I0: 377 pairs of firms 20{C l6.0 O ‘3 - HIGH-RISK Q WIQ/ .0 / Lao-RISK \/ "200 -8.00 -4o00 I -1600 I p— -2P.C L l I l I L I L I L l I «V00 «.00 4.00 .0 2.00 0.00 6.50 HIHH Flex 8. Oman»: Am km DIFFERDCI - nan-,3) - no 135 pairs of firms 112 two percent. When averaged over six months, this return difference is very small indeed. SO (1) the fact that the figures plot cumulative average return differences and (2) the selection of the particular six-month periods for testing suggest that care must be exercised in interpreting the data in the figures in relation to the results of the corresponding statisti- cal tests. In this regard, the results of the T2 test imply that there is enough variability in the low-risk average return difference for the second six-month period that one could easily Observe by chance a difference as large as that Obtained in the present analysis. Also, the fact that the high-risk and low-risk plots differ markedly from each other is of no consequence to the tests performed in this study. This is because both average re- turn differences in each figure are formally compared to zero and not to each other. Somewhat paradoxically, the test for the infor- mation content Of ACs where a firm made both a discre- tionary AC and a nondiscretionary AC in the same year jrields significant differences between return of AC and 1VC3 firms. (See line nine of Table VIII.) The F-value of 6- 358 for the test over the first subperiod is significant a‘t: .less than .10, whereas the F-values for the other two 115 periods are not significant at .25 or less. As can be seen by the risk-class weights of 1.561 and -.561, the high-risk group appears to have contributed more to the significance of the statistic than the low-risk group. Figure 9 below shows that the path Of the high-risk cumu- lative average return difference (CARD) is well below zero and quite stable for the first six months, whereas the low-risk CARD stays very close to zero Over the entire test period. The purpose of including the both (B) category was to see whether it more closely resembles the discre- tionary group or the nondiscretionary group. If the re- sult of this test were similar to the result of either Of the other two tests, it could be interpreted in light of that similarity. But the test result on the both group resembles neither of the other two. The market apparently imputes some negative characteristics to the both group which do not apply in general tO discretionary ACFs or to nondiscretionary ACFs. Inspection of the sample sizes in the tests Of .ACFS that made both discretionary ACs and nondiscretionary JACLs, i.e., ACF (°,B), and of the ACFs that made both types (Dif' ACs with an overall positive effect on EPS, i.e., ACF ("f’,.B), may provide a clue as to the market's behavior PEIIBII 114 .00 12.0 I6.0 201.0 I I .00 4 [DP-RISK < -‘ .00 Him-RISK ‘5200 '8-00 I -l6.0 I -2P.0 I I I I I I I I I -G.OO o6.00 -2.00 .0 2.00 I 0.00 0 .Im I‘DIIII Has: 9. Wm: Minx: km: 01mm - ACF(-.B) - MT 48 pairs of firms 115 toward ACFs (’,B). (See line nine of Table VIII and line eighteen Of Table IX.) Thirty-two Of the forty-eight ACFS (°,B) are also included in the ACF (+,B) category, for which the test results are very similar but more pro- nounced. (Compare Figures 9 and 18.) Because Of this similarity, discussion of possible reasons for the re- sults of ACFs with both types of ACs will not be under- taken here. Rather, the related results of ACF (+,B) will be interpreted in the section on various combinations of directional effects on net income and relative discretion. Tests for Effects Of ACs with Different DirectionaLpEffects on EPS and Varying Degrees of Discretion The tests reviewed above tend, in general, to support one of the theses Of this study, namely that tests of the information effect Of aggregations of dissimilar types of ACs are likely to result in the inference that ACs are not associated with unique stock market behavior. The results Of tests reviewed in this section tend to support the related thesis, that such aggregation hides the unique information effects associated with several types of ACs that are homOgeneous with respect to direc- tional effect on EPS and management discretion. 116 lest of Discretionary ACS That Increase EPS-eACF (+,D). Neither discretionary ACs in general nor ACs that increase EPS in general are associated with unique market behavior. However, ACs with both character- istics do appear to be associated with differential mar- ket behavior, as can be seen from line ten of Table IX. The F-value of 5.546 for the full thirteen-month period is significant at less than .10, and the F-value Of 6.751 for the second subperiod is significant at close to .05. Thus, it appears that the market perceives some informa- tion content associated with discretionary ACs that in- crease net income. From Figure 10 below it is apparent that the low-risk group Of firms contributes more Of the overall difference in average returns. This is also evi- dent from the magnitudes of the average return differences on line ten Of Table IX. The low-risk differences are clearly larger, both in absolute terms and in relation to their standard errors, than the differences for high-risk firms. Furthermore, the average return differences (of ACF-NCF) are mostly negative. This suggests that the market imputes some negative characteristics to firms that voluntarily make ACs with positive net income effects. Perhaps this result is to be expected in light Of recent evidence provided by Bremser. He found that eighty firms 117 I6.0 ZULU I2-0 4.00 Him-RISK ~0 ~4-00 5' Law-RISK -IZ-0 -8.00 I -5600 I h— —2P~0 I I L .L L I I, I I I LL 1 ~0100 -4.00 -2 00 .0 2.00 0.00 0.50 IUNTH FIGIRE ID. Cmunv: AVERAGE Rmm DIFFEREMI - ACF(+,D) - NCF 183 pairs of firms 118 that made discretionary ACs with a positive effect on EPS exhibited a poorer pattern of EPS and return on stock- holders' equity than a randomly selected control group Of NCFs.7 Taken together, these results imply that firms which make ACs (+,D) are less desirable as investments than similar firms that do not make ACs. Thus, it is pos- sible that firms which make discretionary ACs that increase net income do So as a result of poor past performance (in terms of net income and/or stock price activity) or as a prelude to a poor future outlook. Of these two possibil- ities, the latter is implied here by the fact that the T2 test for the first subperiod, prior to the year-end AC disclosure, resulted in insignificant differences, while the test for the subperiod after disclosure yielded dif- ferences that are significant. (See the two subperiod F- values on line ten of Table IX.) However, the results of Archibald's test on firms that changed to straight-line depreciation suggest that the market's negative behavior toward depreciation switch-back firms began up to nine- teen months prior to the change announcement.a Also, Kaplan and Roll Observed mostly negative market reaction to their sample of ACFs during the thirty weeks prior to release of the year-end earnings summary. Even though the results of these two Studies are not strictly comparable 119 to the restuls of the present study, (because of differ- ent testing techniques),9 it appears that the negative relative market behavior toward depreciation method and ITC change firms began prior to the post-year-end dis- closure of the ACs. If firms that make ACs (+,D) do so to conceal otherwise poor performance, it is unlikely that the mar- ket is caught completely unaware because of quarterly earnings reports, the multitude Of investment letters, and other competing sources Of information. One possible ex- planation for the insignificant results in the first sub- period is that six months prior tO the year-end disclosure of ACs may not extend far enough back in time to reveal the negative price activity which might have influenced firms' decisions to make ACS. Or perhaps the market be- havior before year-end is simply not strong enough to be statistically significant. Neither of the possibilities with respect to a systematic market effect prior to year- end denies the result of the second subperiod testrcon- ducted here, namely that post-year-end disclosures of dis- cretionary ACs with positive income effects appear to con- vey information that the market uses to price stocks. The reader may seek an explanation of these re- sults in terms Of the accounts affected by the ACs. The 120 185 ACFs (+,D) made 244 ACS, which are summarized in Table X below. The discussion that follows examines similarities between this study and previous AC research in an attempt to explain why the market reacts more adversely tO the low-risk ACFs than to the high-risk firms in the sample, particularly during the second subperiod (after year-end). The types of ACs on which the two risk-classes differ most are changes to straight-line depreciation (examined by Archibald and by Kaplan and Roll), to flow-through of the ITC (examined by Kaplan and Roll), from LIFO inventory costing (examined by Sunder), and to capitalization of costs. The low-risk group of ACFs in Table X made more depreciation and ITC changes, while the high-risk ACFs made more changes from LIFO. The market reacted negatively after year-end to the ACs examined by Kaplan and Roll.10 Thus, it could be that the market without knowledge of the depreciation and tax credit ACs was misled into thinking the Operating performances Of these change firms were bet- ter than expected. But when the year-end report disclosed the reason for part of the change firms' net income, the market systematically bid down the share prices Of such firms. The market showed no distinct behavioral pattern toward the changes from LIFO examined by Sunder.11 No AC TABLE XL NUMBERS OF THE VARIOUS TYPES OF ACs MADE BY ACFs (+,D)a ——~ — High-Risk Low-Risk Type °f AC ACFs ACFs To Straight-line Depreciation 21 52 Pension-related Changes 27 24 To Capitalization of'Costs 15 10 To Inclusion of Subsidiary or to Equity from.Cost 11 ll From LIFO Inventory Costing ll 5 To Flow-through for Investment Tax Credit 4 10 Increase Depreciable Life of Asset 8 6 All Others (none over 6) 25 2; Total Number of ACs 123b 121 Average Increase in EPS per*F1rm .120 .107 l. thm: sum of A03 of a given type (such as depreciation method, etc.) made by the two risk classes does not necessarily equal the total number of discretionary ACs that increase EPS given in Table V. The difference is made up of the discretionary-positive effects ACs that were made by firms (separately classified as "both") that also made nondiscretionary A03 in the same year. bExcludes three extreme cases in which the ACs increased EPS by 169.6 percent, 198 percent and 472.5 percent. If only the last extreme case (+472.5 percent) is excluded, the average effect per high-risk firm is +15.8 percent. 122 studies have reported separately the results of tests con- ducted on ACs involving capitalization of costs that were formerly expensed as incurred. Therefore, one explanation of these results could be the preponderance of changes to straightqine and to flow-through made by the low-risk group, because otherwise the ACs made by the two risk- classes of ACFs (+,D) are quite similar. It is possible that some variant of the theory underlying the discounted cash flow models could explain why the low-risk firms' values were more strongly affected. That is, the market may have interpreted these particular ACs as indicative of shifts in firm relative riskiness, in which case, other things being equal, there would be a greater impact on the market values of low-risk firms. For an expanded discussion of the reasoning underlying this possibility, see the final section of Chapter 3. In summary, the results of the tests of the in- formation content of discretionary ACs that increase net income are difficult to compare with the results of pre- vious AC research because only the return diffence (of ACF-NCF) is dealt with in this study. Other researchers have either presented only the API of the change firms or the APIs of change and nonchange firms with no formal com- parison of the two. (See in particular footnote nine.) 123 Also, the fact that several different types of ACs are examined in this study makes comparisons difficult. Not- withstanding the different testing techniques used in this and other studies, there appears to be a negative market reaction (either in relative or in absolute terms) to firms that made discretionary ACs which increase net in- come. And it also appears that the year-end disclosures of ACs (+,D) convey information that is pertinent to valuing such firms, regardless of the competing sources of information. Test of_Discretionary ACs That Decrease EPS-- ACF (rJD). The results of the preceding test suggest that discretionary ACs which increase net income are viewed by the market as attempts to conceal otherwise poor operat- ing performance or as a positive signal regarding NCFs, or both. The fact that the market has been shown in other studies to respond favorably to increases in EPS from non-AC sources suggests that the market views such ACs as attempts to manipulate net income. If this is true, then it appears reasonable that the market may respond favor- ably to discretionary ACs that reduce reported profits. Such ACs could conceivably be viewed as a show of strength or as a means of smoothing income in unusually successful 124 periods. The negative response to depreciation changes to straight-line and investment tax credit changes to flow- through may be reversed for changes to an accelerated method of depreciation and for changes to deferral of the tax credit, for example. Or, the market may respond fav- orably to discretionary ACs that reduce taxable income because of a resulting cash savings in tax outlays. This is what Sunder observed for firms that changed to the LIFO method of costing merchandise inventory.12 The results of the tests on ACs (-,D) suggest, however, that the market does not assign a higher value to firms which voluntarily make ACs with a negative effect on net income. (See line eleven of Tablx IX.) The F-values for the three periods covered in this study led to rejec- tion of the null hypothesis of no difference only at sig- nificant levels of .25 or higher. Moreover, the signs of the average return differences (of ACFs-NCFs) for both risk classes over the full thirteen-month period are nega- tive. This is also evident in Figure ll below. The sample of ACs examined here does not pro- vide a rigorous test of the hypothesized positive market reaction to the discretionary ACs which were discussed above. The reason is that very few sample firms made in- verrtory, depreciation, or ITC changes that reduced net PERCEH 125 16 201.0 I 12 .00 LDPRISK 00 p "m 9 \/ O 9 ‘b I 0 C3 Him—RISK p- 0 I O. Np— I O «a». ‘ °. Q...— ‘T‘L11111111111 4100 «.00 -z.oo .o 2.00 LOO 0.50 Fun: 11. Cmume m m Dim - NIH-,0) - III 27 pairs of firms 126 income. For example, the twenty-seven ACFs (-,D) made only one change to LIFO, four changes to deferral of the ITC, and no changes to an accelerated depreciation method. However, the test does cover the following discretionary ACs that reduced reported profits: eight changes to ex- pense costs as incurred or to accrue expenses previously recognized when paid; four pension changes in method or amortization period; four changes to include subsidiaries or to the equity method from cost; three miscellaneous changes in revenue rec0gnition; and two changes to expense intangibles. The other five types of ACs were each made by only one firm in the sample. On balance, these results suggest that discretionary ACs with a negative effect on net income are not associated with unique stock market ef- fects. But due to the relatively small sample size under- lying this test (see figure ll),further evidence should be gathered before making any definitive statements regarding investors' perceptions of such changes. Test of Discretionary ACs That Do Not Affect EPS--ACF (O,D). The results of tests performed on discre- tionary ACs that increase net income and on discretionary ACs that reduce net income do not provide a basis for Predicting the market's reaction to discretionary ACs 127 which have no effect on net income. The results of the positive-effects test was explained in terms of the in- come manipulation hypothesis. The results of the negative- effects test did not reveal any unique market behavior, but the small sample size limits the strength of the state- ments that can reasonably be made about the results of the test. An attempt to apply the income manipulation hy- pothesis to an explanation of the no-effects test will not be fruitful because such ACs do not appear to have been motivated by the desire to manipulate income. Further- more, the heterOgeneity of the A03 in this category renders such broad generalizations useless. In explaining the re- sults of this test it is necessary to examine the A03 in- dividually in order to better understand why the market reacted uniquely to ACs (O,D). The results of this test suggest that, for the second subperiod and with the probability of a Type I er- ror less than .10, such ACs are associated with unique stock market behavior (see line twelve of table IX). It appears from the risk-class weights on line twelve of Table IX and from Figure 12 below that the high-risk group of firms contributes more to the overall multivar- iate average return difference than the low-risk group doses. Moreoveru the unique market reaction to the FERENT 128 l6.0 l2-0 j HIGH-RISK l i 'l2-0 -8000 ‘4-00 T '15-0 r -2p.0 I Loo-RISK L I l Flat! 12. Oman»: m m Dsrrm - ”(O,D) - 10 106 pairs of firms 129 high-risk firms during the second subperiod is noticeably positive, which implies that the market assigns a syste- matically higher value to high-risk firms that make dis- cretionary ACs with no income effect, than to a matched group of NCFs. This particular hypothesis group is one for which different test results would probably occur for subperiods other than the two examined in this study. For example, the strong negative movement in the high-risk- cumulative average return difference from t = -6 through -2 would probably be statistically significant by itself, were it not for the interrupting upward movement in month t = -1. Furthermore, Figure 12 reveals some interesting market behavior which is not formally analyzed in the re- lated statistical tests conducted in this study. It shows roughly offsetting market behavior toward the shares of each of the two risk classes of ACFs (O,D) during months t = -3 through -1. In the case of high-risk firms the market bid down the shares of ACFs in relation to the shares of NCFs during month t = -2 and then returned dur- ing the following month to the higher relative level of month t = -3. In the case of low-risk firms the opposite occurs. The market bid up ACF values during t = -2 and then returned to a lower value during t = -l. The market 150 behavior toward high-risk firms is more pronounced dur- ing the first subperiod than it is toward low-risk firms, but neither average return difference is statistically different from zero. Nevertheless, Figure l2 reveals some interesting offsetting market behavior'during the six months before the year-end AC disclosures. Table XI below focuses on the ACs made by the high-risk ACFs (O,D) in an attempt to explain why the market values such firms more highly than their control group. It presents the numbers of the various types of ACs made by the fifty-three high-risk firms, along with a subjective Judgment as to how the particular ACs in ques- tion related to each firm's wealth position.13 The messages of AC types T1, T7, and T10 are all neutral because there is nothing economically better about a firm's using consolidation, equity, or cost ac- counting for equity investments when the income effect is nil. The message of a type T2 AC can be viewed as either favorable or neutral. A favorable impression of a T2 AC could be due to the likelihood that a higher earnings rate on pension assets means the firm needs to fund less of its pension obligation, resulting in a net cash saving. But this interpretation is superficial because the cause of the higher fund earnings during 1968-1972 is almost 1351 TABLE XI TYPES or ACs MADE BY HIGH-RISK ACFs(_6.D) Type Type of Ac Number message Number of ACs T1 To Inclusion of Subsidiary from Cost or Equity 10 N T2 Increase Assumed Earnings Rate on Pension Assetsa-b 8 G or N T3 To Inclusion of Foreign Subsidiary or to Equity 5 G from Cost for Foreign Investee 7 2 N T4 To LIFO Inventory Costing 5 G T5 To LIFO Inventory Costing 5 B T6 To Straight-Line Deprecationb'c 4 B T7 To Equity from Inclusion of Subsidiary d 4 N T8 To Percent Completion from Completed Contract 2 B T9 To Percent Completion of Work Performed from Percent Completion of Shipments 1 N T10 To Equity from Cost 2 N T11 To GAAP for Insurance Subsidiary 1 N T12 To Amortization of Past Service Costd l G T13 Shorten Amortization geriod of Past Service Cost from 40 to 30 Years 1 G T14 Reclassify DoSses on Foreign Currency as Other Deduction 1 B T15 Increase Pension Funding Period8 1 G T16 Reclassify Deferred Tax Credit as Reserve 1 N T17 Change Pension Method 1 N T18 To Amortization of Goodwillc 1 G or B T19 To Reporting all Assets and Liabilities on Film Contract Rentals 1 N T20 To Capitalization of Research Development Costs 1 G or B 81 Total Number of ACs G--considered to be "good" or favorable information with respect to a firm's wealth position. B--considered to be "bad" or unfavorable information with respect to a firm's wealth position. N--considered to be neither "good" nor "bad" information about a firm, therefore "neutral." aOne firm made two pension related ACs. bOne firm made a change in the assumed earnings rate on pension assets and also a change to straight-line depreciation. cOne firm made a change to straight-line depreciation and also a change to begin amortizing goodwill. dOne firm made a change to percentage-of-completion for recognizing revenue on long-term contracts, a change to start amortizing past service cost. and a third change to shorten the amortization period for past service costs. 152 certainly attributable to a rise in interest rates which is probably related to a rise in the inflation rate. Em- ployees who are covered by pension plans are not unaware of the effect of inflation on their real earnings, and in inflationary times they can be expected to demand greater wages and/or pension benefits which should offset the cash saving from a higher pension fund earnings rate. The message of this AC could therefore be neutral because the overall effect on the firm is nil. Type T12 is believed to convey a favorable impression of the firm that has will- fully decided to record an expense. The same reasoning could also be applied to a type T18 AC. On the other hand, the amortization of goodwill could be viewed as the firm's admission that an asset has ceased to exist, an unfavor- able piece of information. Type T15 is viewed as good news because the firm is voluntarily hastening the rec0g- nition of an expense that is, by definition, related to prior periods. Type T15 conveys good news because of the cash saving from a lower periodic funding requirement. Type T5 reflects favorably on five of the seven firms be- cause the stated underlying reason for the change is im- proved politico-economic conditions in the domicile of the subsidiary or affiliate. The message for the other two type T5 ACFS is neutral for the same reason given for T1 155 and T10. Type T4 reflects favorably on a firm because of the cash savings that accompanies a change to LIFO when factor prices are increasing. Type T5 conveys the Oppo- site message about a firm whose inventory prices are in- creasing. This inference about the T4 changes to LIFO is supported by Sunder's work.l4 Type T6 reflects unfavor- ably on the firm if the intent is to manipulate earnings. This is implied by the results of the test on discretionary ACs that increase net income, where thirty-two of the 121 low-risk firms, to which the market reacted negatively, changed to straight-line depreciation (see table X). Type T8 suggests the firm is hastening the rec0gnition of rev- enue, a negative signal. Type T14 is considered to reflect negatively on the ACF because it implies an attempt to hide losses in the "other" section of the income statement. Type T20 is bad news if it is an attempt to arbitrarily defer expense rec0gnition. It is good news if it is mo- tivated by successful research activities which are ex- pected to benefit future periods. The remaining types of ACS are either reclassifications (T11, T16, T19) or are not susceptible to a normative evaluation (T9, T17) of the type accorded the other types. They are rated as neutral. 154 Based on this rather subjective analysis, it appears from the data in Table XI that the overall reflec- tion of all the ACs on the firms that made them is mildly favorable. On this basis it appears reasonable that an efficient market would have generally responded positively to this group of ACFs. This conclusion is, of course, conditional upon the messages which the ACS convey to the author. Another analyst may disagree with the normative signals inferred from the AC disclosures that the firms used to convey their AC decisions. Another explanation for these results could be sampling error. The Opening comments in this section state that there is little a priori reason for predicting any unique market behavior toward firms that willfully make ACs which do not affect income. Furthermore, the number of "good news" ACS in Table XI is not overwhelmingly greater than the number of "bad news" ACS, especially in relation to the large number which are considered to con- vey a neutral Signal about the firm's wealth position. Out of fifty-four hypothesis tests,15 around five can be expected to yield spurious results with a significance level of .10. This test may fall into that category. For this reason, the test needs to be replicated in order to determine whether the results are sample-specific or 155 whether such changes do, in fact, reflect favorably on the firms. Test of Nondiscretionary ACs That Increase EPS-- ACF (+,N). The dearth of evidence on the market's reaction to nondiscretionary ACs was a source of motivation for this study. The earlier test for the effect of nondiscre- tionary ACs in general, provided the expected result--that that level of aggregation may well have hidden any infor- mation effects that nondiscretionary ACs with particular directional effects on net income may have. The test for the effect of nondiscretionary ACs that increase EPS resulted in an F-value of 4.961 which, with a confidence level of greater than .95, suggests that the market assigns a systematically higher value to such firms than to a control group of NCFs. (For details see line fourteen of Table IX.) Figure 14 below reveals that most of market effect occurs during the second subperiod, after the preliminary and/or annual reports have been re- leased. The timing of the market behavior toward nondis- cretionary ACS seems at first somewhat surprising in light of the fact that ninety percent of the ACS in this sample were induced by APB Opinions. Consequently, market agents FEW 156 O onrstx D. r- Q M / Lat-RISK -l2.0 -0.00 -4.00 I "6I0 l —2P.0 l L l l 1 L l l l l l 1 4.00 «.00 -2.oo .o 2.00 4.00 5.1m MM Flare 1H. (mum: AVERAGE Raw DIFFERENCE - MIMI!) - [G 60 pairs of firms 157 should have known from exposure drafts and other publicity given the Accounting Principle Board's agenda that these firms would be making ACs. Perhaps they did not act on this knowledge long before year-end because they were un- able to predict what effect the ACS would have on net in- come. The insignificant differences in the first sub- period test, plus the Significant results in the posteyear- end period, lend support to this proposition. However, a closer inSpection of Figure 14 indicates that the unique behavior toward high-risk firms actually began prior to the disclosure of the full year's results. This suggests that analysts and other investors may have used quarterly reports or other media in establishing the higher values for the ACFS in this group. However, inasmuch as Figure 14 plots the cumulative average return difference, the up- ward trend that continues after year-end suggests that year-end AC disclosures are also associated with unique stock market behavior aS well. One could also hypothesize that the market re- action may have taken place well in advance of the test period examined here. There was generally a time lag of several months between the release of an exposure draft and the effective date of APB Opinions. And in many cases there was an additional time lag between the effective 158 date of the opinion and the FYE of affected companies. So it is reasonable to believe that the market may have be- haved abnormally toward affected firms prior to the first subperiod examined here. However, even if there was such an anticipation on the part of the market, that does not deny the post-yearsend return behavior observed in this test--that year-end disclosures of ACs contained informa- tion that was useful for establishing relative prices of firms that made the changes. The discussion above concerns only the timing of the market reaction; it does not address the reason for the market's favorable response to ACS (+,N). If such ACS are inherently related to increases in the real wealth of firms, then the result is easy to understand. But fifty of the sixty-two ACs made by the sixty firms in this test were changes to the equity method of accounting for in- vestments in common stock, and there is no apparent effect of this change on a firm's cash flows. Of these fifty ACs, twenty-five were made by the high-risk firms, to which the market reaction was stronger. The thirty high-risk firms also made five additional ACS: three increases in the service lives of assets, one AC to capitalize costs formerly expensed as incurred, and one miscellaneous depreciation change. The market's positive 159 reaction to these ACS could be due to investors' expecta- tions of higher dividends resulting from the higher re- ported profits that resulted from the ACs. Perhaps the market infers company strength from the fact that ACFs in this cateotry had formerly chosen a more conservative method of accounting and had to be forced to adopt a tech- nique which produces a larger income number. Either mar- ket perception could lead to higher relative stock prices for ACFS in this category. In addition to the possible price increase, firms may have actually raised their divi- dends due to their higher reported net income. Either or both of these factors could account for the higher rela- tive returns on this group of ACFS. Thus, in summary it appears that the market views the EPS improvement from nondiscretionary ACs in a very different light than it does increases from discretionary ACS. Test of Nondiscretionary ACs That Decrease EPS-- ACF (-.N). The test for the information effect of nondis- cretionary ACs which decrease EPS resulted in an F-value for the second subperiod that is significant at less than .10 (see line fifteen of table IX). The negative high- risk average return difference for the second subperiod contributes more to the value of the statistic, as can be Seen from Figure 15 below. 140 PLREEIT _ HIGH-RISK (:1- .)— a '3 16-0 00 0 00 -0 ‘8000 -‘.00 "200 I Luv-:1 O «3- T ‘3 3“ . 1 1 1 1‘1 1 1 1 1 1 1 4.00 «.00 -2.00 .0 2.00 0.00 0.50 me15 Onuuwehawmflnumbwnmmz-NIFJ)—KF 10 pairs of firms 141 It is interesting that the market reacted favorably to nondiscretionary ACS with positive income ef- fects (see line fourteen of table IX and figure 14) and unfavorably to these particular nondiscretionary ACs with negative income effects. One possible explanation for the results could be that, since four of the ten ACs in this group were changes to the equity method, the AC disclosures could be investors' initial notification that their firms held stock in other companies which were not earning a profit. This could lead investors to expect lower divi- dends. Four of the other six ACS were changes to accure income taxes on subsidiaries' undistributed incomes. An- other possible explanation of these results could be that the market views with disfavor the fact that these firms had willfully selected an accounting method that produces a higher income number than the more conservative method they were compelled to adOpt. Figure 15 indicates that, for both risk classes of firms, pronounced market movements were reversed in month t = -1. For high-risk firms the subsequent differ- ential (ACF-NCF) return movement was uniformly negative until month t = -5. Apparently investors' favorable ex- pectations up to month t = -l were not met, and they re- acted in a strongly negative manner to the net income 142 decreases that resulted from the ACS (or possible from some other sources). For low-risk firms the first subperiod movement was strongly negative, even considering the posi- tive return differences during months t =-5 and -2. Dur- ing month t = -1, however, the market's relative reaction toward low-risk ACFS was positive; thereafter there was very little difference between low-risk ACFs and NCFs as can be seen from the nearly horizontal low-risk CARD from t = 0 through +6. In summary, it appears that month t = -l was a turning point for both risk classes of firms that made nondiscretionary ACS with negative effects on net income. The fact that this month precedes the assumed disclosure month of t = O is consistent with the timing of the unique behavior toward firms whose nondiscretionary ACs increased net income (see figure 14). And this similarity is rea- sonable because market agents could have anticipated these ACS in advance of the year-end disclosures of them. But here the market reaction to the two risk classes of ACFs is very different, whereas the market reaction to the two Pi.sk classes of ACFS charted in Figure 14 is quite Similar. FLlrther research should be conducted to ascertain the CEluses for this observed difference in return performance. 145 Table IX reveals that the test of ACF (-,N) is based on observations of only ten matched pairs of firms, five in each risk class. Therefore, sampling error is likely to have affected the results of this test to a considerable degree. As a result, the test needs to be replicated on a larger sample before more definitive con- clusions are reached concerning the market behavior toward firms whose nondiscretionary ACS decrease EPS. Test of ACFS Whose Discretionary ACS and Nondis- cretionary ACS Incregse EPS--ACF (+,B). The test for the market behavior toward firms that made at least one dis- cretionary AC and at least one nondiscretionary AC in the same year (referred to as "Both") resulted in a significant F-value for the first subperiod covered in this study (see line nine of table VIII). At first this result appears to be an aberration because the tests of discretionary ACS in general and of nondiscretionary ACs in general produced insignificant F-values (see lines seven and eight of table VIII). As mentioned in the discussion of the results of teasts on the both category, however, the results of this teast appear to be dominated by the market behavior toward ifi.rms with both types of ACs, where the effect on net in- ‘3Crme is positive. Compare line nine of Table VIII with lWiIle eighteen of Table IX and Figure 9 with Figure 18 below. 1J14 l6.0 12-0 4.00 l Lav-Rm /\ ‘? D 9 ' 1 HlarRlsx O O 'h— D l v ‘1’ N»— I O 0_ | 9 e— 1111111111111 also «.00 -z.oo .0 2.00 0.00 3.50 "HRH Fun: 18. Datum: Mame Rtnm Dtrrame - Amati) - N2} 32 pairs of firms 145 At this point it becomes necessary to explain the results of the test of both types of ACS with a posi- tive overall effect on EPS. Line eighteen of Table IX in- dicates that the F-value of 11.555 for the first subperiod is significant with a confidence level of greater than .975. The risk-class weights, along with Figure 18, indi- cate that the negative average return difference (of ACF- NCF) on the high-risk firms contributes more to the overall multivariate difference than the low-risk difference does. Thus, the discussion which follows chiefly concerns the high-risk firms that made both types of ACs, where the ACS increased net income. To that end Table XII shows a breakdown of the ACs made by firms in this category. Table XII indicates that the sixteen high-risk firms made seventeen nondiscretionary ACS and twenty-one discretionary ACS. These numbers are comparable to the numbers of discretionary and nondiscretionary ACS made by the low-risk firms, to which the market reaction was slightly favorable (see figure 18). However, the high- risk firms made eighteen discretionary ACS with positive income effects compared to fourteen for the low-risk firms. Earlier it was shown that the market reacts negatively to ACs in this category (see line ten of table IX and figure 10). Moreover, the discretionary ACS made by the high-risk 146 .TABLE XII DECOMPOSITION 0F ACs MADE BY ACFs(+,B) High-Risk Low-Risk ACF(+,B) ACF(+,B) Number of Firms 15 15 Number of Nondiscretionary ACS 17 16 Effects on EPS Positive 11 11 Negative 5 1 'Zero 1? 1 Not Disclosed 2 5 Average Effect on EPS Per Firm +.152 +.O44 Adjusted Average Effect on EPS Per Firma 4‘955 +.O44 Number of Discretionary ACS 21 22 Effects on EPS Positive 18 14 Negative 1 5 Zero 1 5 Not Disclosed 1 2 .Average Effect on EPS Per Firm tgglg +.O4fi Average Effect of Both Types of ACs on EPS Per Firm +.211 +.O92 Adjusted Average Effect of Both Types of ACs on EPS Per Firma +.154 +.992 ‘ aOne high risk firm switched to the equity method of accounting for investments in stock (a nondiscretionary AC) and increased its EPS by 128.6 percent. Deletion of this extreme case reduces the average EPS effects to those referred to as "adjusted average effect . . . per firm." 147 firms increased EPS by an average of 7.9 percent, com- pared to 4.8 percent for the low-risk firms. Another relevant piece of information not given in Table XII is that the thirty-two ACFS in this test made only four non- discretionary ACs that decreased EPS. In every one of the four cases, the firms also made a discretionary AC which more than offset the negative EPS effect of the nondis- cretionary AC. Three of these cases occurred in the high- risk group. Thus, on balance, the results are consistent with the notion that market behavior associated with ACF (+,B) is influenced by the perceived manipulative intent of the discretionary ACS which increase net income. If this is true, it then becomes necessary to explain why the market reactions to high-risk (ACFS (+,B) and to 12!- ‘ggsk ACFS (+,D) were similar. The similarity probably re- sults from the fact that the ACs made by these two groups of firms have more potentially manipulative characteris- tics in common than do the two groups of high-risk ACFs or the two groups of low-risk ACFS. The above interpretation does not address the question 11s to why the market's negative behavior occurs during the six months prior to the year-end disclosure of ACS. Perhaps, as Bremser observed for firms that made discretionary ACS with positive income effects, the firms 148 that made both types of ACs were also less successful in terms of profitability than their control group. In this case, the negative stock price activity during the latter half of the fiscal years of high-risk firms may have in- duced them to increase their EPS by making discretionarm A08. The fact that most of their nondiscretionary ACs also increased net income could account for the tapering off of the market's negative behavior in the six months after the year-end disclosure month (see figure 18).16 Thus, it is possible that, for both risk classes, the po- tentially offsetting market effects of the discretionary and nondiscretionary ACS could explain the essentially horizontal movement in the cumulative average return dif- ferences during the second subperiodtusrevealed in Figure 18. Other Mean Vector TegtggNot Diggggsed Previougly. The results of three tests have not been discussed thus far. They are the tests of (l) discretionary ACs for which the income effect was not disclosed (see line thirteen of table IX and figure 15 below), (2) nondiscretionary ACS with no effect on net income (see line sixteen of table IX and figure 16 below), and (5) nondiscretionary ACs for 149 8.00 HIGOI'RISK m 4-00 Hist-RISK .0 I x. Loo-RISK HIGo-Rxsx -I2-0 -8.00 -4.00 -0600 .- -ZP-0 1 l J 1 l .L Lor-Rlsx - .00 -4.00 ~2.00 -0 F101! 13. 0mm: Meme km. W Dim - NZF(?.D) - MI 61 pairs of firms 20.0 FEED“ 0-00 12.0 16-0 4.00 Low-RISK \/A HIGH-RISK .0 - -12.0 -0.00 -4.00 ‘0600 p— ~2P-0 l -‘ .00 l L 2.00 l 1 4.00 l l 1, l -2.00 LI, 1, Lbs [NIH Flags 16. 0mm: Mm Etna: Dlrrmx - “(0.14) - 10' 30 pairs of firms .0 _1— O N 12.0 0.00 HlG-v-RISK 4 .0 Lav-R151: ~12.0 ~0.00 -4.00 '0600 .- -2P.0 I l Fume 17. (mum m m 0mm - ”(7.10 - IO 35 pairs of firms 150 which the effect on net income was not disclosed (see line seventeen of table IX and figure 17 below). The F- values for all of these tests suggest that the market reactions to these types of ACs are, in all likelihood, governed by a random process. The only one of the above tests which lends it- self to a ready a priori belief as to the outcome is the first one. It was reasoned earlier that the market may view a firm's nondisclosure of the income effect of its discretionary AC as a negative signal about that firm be- cause nondisclosure may be viewed as an attempt to hide unfavorable news about the firm. A competing belief is that firms choose not to disclose the income effect bee cause it is immaterial. The test results tend to support the latter belief. Except for case (1) above, there are no compelling reasons for predicting a market response to the above three categories of ACS, and accordingly the three tests are not discussed further. Tests on Covariance Matriceg The first part of this chapter covers the em- pirical results and interpretations of the mean vector tests called for by the research design utilized in this 151 study. This part of the chapter discusses the results of tests conducted on covariance matrices of returns, which define the variances of the mean vectors of returns on ACFS and NCFs. (See the Chapter 5 section entitled.1§§£§ on Covariance Matrices for a discussion of the nature of the covariance matrices being examined here.) Inequality between the two groups (in general ACFs and NCFs) on either parameter, the mean vector or the covariance matrix, is sufficient for rejecting the null hypothesis (Eq. 4 or Eq. 5) of no information content associated with ACS. The need to examine variances as well as means in order to assess the information effects of accounting events was discussed initially in Chapter 1, and also in the final section of Chapter 2. Chapter 5 covered the statistical procedures to be followed in performing the tests for equality of two covariance matrices. That sec- tion also discussed the need to establish that the uncon- ditional covariance matrices, 21 and'Zz (computed over the sixty-month B-estimation period), were equal in order to be able to attribute impact-period differences between the conditional covariance matrices, legl and Zzlgz, to the information variable E}. Thus, this section of Chapter 6 is divided into two subsections, the first covering tests on unconditional covariance matrices of returns and the 152 second covering tests on covariance matrices of returns conditional on the information effects of ACS. Tests for Equality of Unconditional Covariance Mgtrice§_during the B-Egtimation Period In Chapter 5 it was asserted that satisfaction of the condition 2 == 2 l 2 that E, has information content if legl # ZZIQZ. On the provides a basis for inferring other hand, the condition 21 # £2 fails to provide the basis for ascribing differences between lefil and Zzlfiz to E, because there were differences between 21 and 22 be- fore the information variable 5, was introduced. AS a re- sult, zllgl # zzlgz may be due to either g, or to other exogenous factors which cause the variability in Ellgl to differ from that in Ezlgz. In the event that 22 # 22, there is no unambiguous conclusion from the test of Hb: 2 IQ = 2 lg . o 1 l 2 2 The results of the test for equality of covar-. iance matrices during the B-estimation period are given in Table XIII in the column, t ==-66through -7, which indicates that four of the chi-square statistics17 are Significant at .01 or less. These tests are identified on lines three, four, eleven, and eighteen of Table XIII. 1155 TABLE XIII SUMMARY STATISTICS FOR TESTS ON COVARIANCE MATRICES t - 66 ‘ t - -6 t . +1 t u -6 gigger Hypothesis Group ggifggflarz through -1 through through +6 valuea F-valueb F-value F-value° Omnibus (1) ' Acr(-.-) - NCF 1.015 .046 .397 .068 Direction of Income Effect 1.024 .120 .006 .105 15.993 NA NA NA ,.) 12.052 NA NA NA 4.469 .078 .191 .226 6.489 .071 .223 .272 Relative Discretion 7 ACF o,D - NCF 3.434 .492 .762 .182 8 ACF -.N - NCF 1.425 1.112 .817 .424 9 ACF ..B - NCF 7.168 NA NA NA Direction of Income Effect and Relative Discretion 1.67 .388 .130 .099 NCF 20.423 NA NA NA NCF 6.311 NA NA NA NCF 6.456 NA NA NA NCF ..844 .847 .537 .932 NCF 3.931 1.026 .151 .067 NCF 7.214 NA NA NA NCF 4.353 .354 .170 .101 NCF 13.890 NA NA NA aSelected fractiles of Chi-square with 3 degrees of freedom are: are both: Fractile Value of x3 .900 6.251 .950 7.815 .975 9.837 .990 11.341 .999 16.268 to b Selected fractiles of F5; 13.000 and F5;104 680C Fractile Value of F go .900 2.86 ’ .950 3.98 .975 5.26 .990 7.21 NA--Not applicable because the antecedent conditions necessary for this test were not met. 154 When the significance level is raised to .10,five more of the chi-square values are significant. They are for the tests identified on lines six, nine, twelve, thirteen,and sixteen of Table XIII. These results indicate that be- cause 21 # 22 over the B-estimation period, the two un- conditional return distributions differ in unsystematic variability. A complete analysis of the reasons for the differences between 21 and 22 is not warranted because the test was conducted in order to determine whether the antecedent conditions of the tests of zllgl = ZZIQ2 were met. If the .10 level is retained, then the unconditional covariance matrices El and 22 are equal for only nine of the eighteen hypothesis groups. This means, of course, that only these nine tests of legz = 2 can be per- 1 |_9_2 formed in the manner called for by the research design developed earlier. Tests for Equality of Conditional ngariance '1£E££1£g§_gpring the Thirteen-Month Impact Period Because of the statistically significant dif- ferences found to exist between the unconditional covari- ance matrices,'£l and £2, fOr half of the eighteen hypoth- esis groups examined in this study, only a partial inquiry 155 was made into the equality of the conditional covariance matrices, Ellgl and ZZIQZ. For the nine hypothesis groups where the two Ei differ, it is not appropriate to infer that realized differences between EEIEE are associated with ACs. Table XIII indicates that none of the remaining twenty-seven F-values18 for the tests of legl = Zzlgz is significant at .25 or less. At face value the inference is that for all those hypothesis groups legl = Zzlgz with a confidence level of .75 or greater.19 But this inference may not be warranted. The tests over the B-estimation period and the tests over the impact period both use the M-Statistic, which is affected by sample size (see eq. 10). The far larger sample size (i.e., sixty monthly observations) in the test on unconditional co- variance matrices versus the six and thirteen monthly observations in the test on conditional covariance matrices gives the former test more statistical power than the lat- Zlfiz be powerful enough to detect true differences between the ter tests. Therefore, the tests of Ellgl = 2 may not two conditional covariance matrices. In summary, the results of tests for the equality of conditional covariance matrices are indeterminate. How- ever, the loss is not as great as it may appear at first. 156 In Table XIII the tests identified on lines nine, twelve, and eighteen yielded results in the mean vector tests which suggest that the return distributions of firms which made those types of ACs differ significantly from the re- turn distributions of their matched NCFs. Therefore, the results of tests on conditional covariance matrices for these three types of ACS, for the time periods when Sig- nificant differences were observed, could provide only limited additional insights into the information effects of ACs. For the other types of ACs, however, a design different from the one used in this study will be needed to provide more convincing evidence on the effects that ACs have on the variances of security returns.20 isk and Information Effects One of the principal general hypotheses of this study is that the information effects of ACS may be risk- dependent. This possibility served as the reason for dividing firms into two risk classes and conducting multi- variate tests instead of the univariate tests other re- searchers have used. The final section of Chapter 5 pro- vides the rationale for believing such a relationship may exist. In general, that discussion hypothesizes that simi- larities in relative risk are pervasive in the sense that 157 firms with homogeneous risk characteristics may be prone to make similar types of ACs that have similar stock mar- ket consequences. In the first section of this chapter ACs of several types (of directional effect on net income and relative management discretion) have been shown to have information content. The discussions of those ACS hint strongly that the information effects are stronger for one risk class than for the other. Moreover, the graphical results bear this out quite vividly. In some cases, even though the multivariate tests on mean vectors indicate no significance, there appears to be rather marked differences in the way the market reacted to Similar ACs made by high- risk and low-risk firms. See in particular Figures 2, 5, and 7. Table XIV below presents a summary of the hypoth- esis tests for which information effects were inferred, along with the market reaction to the risk class that was more affected. The market's negative reaction to low-risk firms that made discretionary ACs with positive income effects (Figure 10) was eXplained by the fact that they made a greater number of changes to the straight-line method of depreciation and to flow-through for the ITC than high- risk firms did (see table X). These results generally 158 TABLE XIV MARKET REACTION TO THE RISK CLASS MORE AFFECTED BY ACs WITH SIGNIFICANT F4VALUES IN THE TESTS ON MEAN VECTORS Nfiigzr Hypothesis Group MfiizkAgfzzied szzgizn (l) ACF(+,D) - NCF Low Negative (2) ACF(O,D) - NCF High Positive (5) ACF(+,N) - NCF High Positive (4) ACF(-,N) - NCF High Negative (5) ACF(+,B) - NCF High Negative (6) ACP(-,B) - NCF High Negative =======_ 1 J 159 seem to support the findings of studies conducted by Kap- lan and Roll and by Archibald, who also observed a nega- tive reaction to these particular types of discretionary ACS. The results of the test on firms that made both discretionary and nondiscretionary ACs in one year (Figure 9), as well as the test on the both-positive category of firms (Figure 18) was explained in terms of the large num- ber of disgretionary ACS with positive income effects made by the high-risk group of firms. Also, the high-risk firms in both of these tests made discretionary ACs with somewhat larger percentage increases in net income than did the low—risk firms (see table XII). Thus, in the three tests noted on lines one, five, and six of Table XIV, it appears that the risk-dependency can be explained in terms of the income-manipulation hypothesis. The market reacted favorably to nondiscretionary ACS with positive income effects (Figure 14), but demon- strably so only for the high-risk group. The explanation given for the positive market effect was that such firms had formerly adopted accounting techniques which produced a lower net income number than the method they were com- pelled to accept. Market agents may have (currently) viewed the firms' former election to "understate' profits 160 as a Show of strength, and this knowledge in turn could have led them to expect higher dividends or perhaps to view the firms as less risky. Other things being equal, either expectation would seem to reflect favorably on those firms as investments. Around eighty percent of the ACs made by both risk classes were changes to the equity method, prompted by APB Opinion No. 18. The only apparent difference between the two risk classes of firms is in the percentage effects their ACS had on net income. The av- erage effect on the net incomes of high-risk firms was +8.5 percent,21 as compared to +4.1 percent for the low- riSk firms. Thus, it appears that the more strongly posi- tive market effect observed for the high-risk firms is re- lated to this difference. Even though the AC, per se, does not enhance the wealth position of the firm, dis- closure of the AC could have been investors' initial noti- fication that their firms were earning, on the average, 8.5 percent more net income than they had thought. The risk dependency of the information effect of these ACS seems to stem from the fact that the high-risk firms were less willing than low-risk firms to use the latitude in generally accepted accounting principles to report a higher profit. This argument could be characterized as a negative version of the income manipulation hypOthesis. 161 It is interesting that in this test of nondis- cretionary ACs that increase net income, the high-risk firms appear to be less prone to use A08 to manipulate EPS (see line three of table XIV), while in the tests of ACs (both) and ACs (both-positive) (lines five and six), the high-risk class was shown to be the one which did use ACS to improve reported profits to the greater extent. These opposite test results point toward conflicting con- clusions with respect to the AC behavior of firms with high risk coefficients. Additional work will have to be done in this area before the nature of the relationship is more fully understood. The results of the other two tests where infor- mation content was inferred do not lend themselves to such plausible eXplanations of why the market reacted to only one of the two risk classes (see lines two and four of table XIV). For example, the discussion of the results of the discretionary-zero effects ACs (Figure 12) ignored the low-risk group of firms and only mentioned the high- risk firms, to which the stronger market reaction was posi- tive. The overall impression conveyed by the high-risk firms' AC disclosures was favorable with respect to those firms' wealth positions (see table XI). An analysis simi- lar to the one underlying Table XI was also performed on 162 the ACs made by the low-risk firms. Conditional on the appropriateness of the author's interpretation of the rela- tive messages of those ACs, the results suggest that the market should have reacted even more favorably to low-risk firms' ACs than to those of high-risk firms. This set of results remains a puzzle. The results in the test of nondiscretionary ACS with negative income effects also provide an ambiguous signal as to the reason for the risk-dependency of this AC information. In that test the market was shown to react negatively after year-end to the high-risk group of firms (Figure 15). Their ACs hardly differed from the low-risk firms' ACS either as to accounts affected or as to the percentage effect on EPS, but the ambiguity of these re- sults are not considered too surprising in view of the ex- tremely small number of firms involved in this test. In the tests in which there is no clear reason for the risk-dependency, as is indeed the case for all the other tests conducted in this study, the presence or ab- sence of an auditor's consistency exception could have af- fected the market's reaction to the firms that made ACs. In Chapter 5 Baskin's study was reviewed,and it was as- serted that his research design did not distinguish be- tween the AC or the consistency exception as the indepen- dent variable. As a result, it is not clear which event 165 Baskin was measuring. To some degree this study is sub- ject to the same criticism because a clear distinction between ACs and consistency exceptions was not made. It seems safe to assume, however, that this study measures the effects of ACs because all 560 firms did make ACs, while only 289 of the firms received consistency excep- tions. This means that a little under half of the ACFs did not receive consistency exceptions from their auditors. Additional work is under way to determine whether auditors' consistency exceptions are also associated with unique stock market effects. The foregoing discussion summarizes the tests for which information content in ACS was inferred and gives possible reasons for the observed risk dependency of the information. The second phase of the inquiry into risk and information effects is summarized in Table XV be- low. It gives, for the full thirteen-month period only, the F-values of mean vector tests using five different values of the weight vector, Er: in Eq. 6. The column labeled "Max t2" pertains to the weight vector which is implicit in the T2 test (see the chapter section entitled fiestg on mean Vectorg). The F-values in this column are the same as those in Tables VIII and IX for the full thirteen-month period. The column labeled "High-Risk" 164 TABLE XV RESUETS OF TESTS ON MEAN VECTORS FOR DIFFERENT OVERALL GROUP RISKS F2,lla for t = -6 to +6 Hypothesis Group Max t High- Lowe Equal Overall Risk Risk weight 68 Equal One bus ACF -,o) - NCF 1.064 .099 1.011 .067 .898 t o In ome f ect ACF(+, -) - NCF 1.614 .228 1.564 .666 1.612 ACF(-, -) - NCF .616 .266 .206 .669 .595 ACF(-, ~) - ACF(+,°) .215 .202 .045 .169 .104 ACF(O,-) - NCF .989 .504 .661 .057 .744 ACF(i, -) - NCF .175 .004 .144 .055 .155 R t v s r t on ACF(o ,D) - NCF 1.619 ..122 1.547 .267 1.465 ACF(o ,N) - NCF .611 .209 .420 .676 .527 ACF( ,B) - NCF .588 .565 .0005 .147 .056 t o o o t R t v t ACF + ,D) - NCF 5.546 .0005 2.697 5.007 2.904 ACF(- ,D) - NCF 1.511 .157 .556 .665 .657 ACF(O, D) - NCF .428 .527 .145 .176 .065 ACF(i, D) - NCF .079 .010 .072 .014 .072 ACF(+, N) - NCF 4.961 5.792 .622 4.960 .960 ACF(- ,N) - NCF .644 .509 .256 .591 .147 ACF(O, N) - NCF .461 .295 .457 .420 .460 ACF(7, N) - NCF .472 .002 .407 .152 .556 ACF(+,B) - NCF .694 .674 .179 .060 .065 8See Table VIII in appendix A for the fractiles of F2 11 when the null .7 hypothesis is true. 165 pertains to the weight vector, x; = [l O], which is analogous to the placement of one's entire investment in the high-risk group of securities. The column labeled "Low-Risk" pertains to the weight vector, 3; = [O l], which is analogous to placing one's entire investment in the low-risk securities in the sample. The column en- titled "Equal Weight" pertains to the weight vector, .3; = [1/2 1/2], which corresponds to equal investments in the high- and low-risk groups of firms. The final column of Table XV, entitled "Overall 38 Equal One" pertains to the weight vector, 3; = [x y], where x and y sum to unity and they also combine the high- and low-risk groups' ég (in the risk vector 3g) for each hypothesis group to produce an average'fig of one for the two risk classes in each hypothesis group. The purpose of examining the different F-values which result from mean vector tests using different values of Er is to investigate whether the widely different val- ues of Er give rise to widely different F-values. If they do, this would imply that AC information is risk-dependent. On the other hand, if they do not, then there is reason to believe that AC information is constant across the two risk classes being examined. Consistent with the discus— sion above, Table XV indicates that different values of 166 the weight vector are associated with different F-values. This is evident for virtually all eighteen hypothesis groups in Table XV. On the basis of these summary sta- tistics, it appears that there are some risk dependencies. However, as mentioned in the earlier part of this discus- sion, these results are only a first step and do not pro- vide a cohesive description of the nature of the relation- ship between risk and the information effects of ACs. That will have to await additional investigation which is beyond the scope of this study. Appendix A 167 168 a . a... . an. ..o. “so. a... 0.... .o~.. se~.. a... an... an... one. an... ac..- a.~. «:~. a... a... :3. 2h. 03.. So. R. . (we . .6... .6. can". - $38.. 2 86. .s 30.. ..8. .88.... 330.. 3.38.. .38. :96... 8. 8.. .on. 8.. on. 8.- $38.. :80. $38.. 09.3.- 2.28.. .88.. .8. 8.. $3... 828.. .28. .58.. .88.- 3&2... 3.8. 388.. 8.8.. .83.... 52.8.. .88.. 8.8. $88.. 0228.. «38... a $.09. 3 080.38 0.050: igdooa a... 2... i. 0.- £0 .35 3. . 80. Re. 80. 3... 83. an. 8. . 18...: 80.3. Ga. 3... «3. 2A . a: . 8... . an..- 5... g 838.. 38... .38.. .c~.8. £28.. 2.3.... $98.. :2... G... 8.. was... .2. a 9.. .88.- 82.8.. “008.- 3.38.. :88... .58.... 1.3.... 4.1.4.844... .2. 8.. 8.8. .3. 8.. .38. .28.. :88. 2.38.. .88.. .3. 8.. 038. ..an.o.. ao~.o.- .32.... 9.08.. 30.8.. 028. 3.8.. e.22.... cup-e: “a a, u. 01. 300.80 I 3 ‘0: '0 03108... at. 3430.. no no. .0008. as. .80... 230.10...- 3380 8608:... 0.3.0.. cull- ~0 .380 883:. on... .3: A... a... a... 3%. ..o. 8." 2n. 2... , «8. 8... n8. 3-. acts .8. 8.. 1. 9.3.8 03.3 sln I 3.8 .8. 833.80 Ell..- uunu 0.“ :3... cl..- 8... I 3.8 .8. I33...- Bull. «3183 3A.. as 08... n3.- 2.... 0... 80. 8.! «a. ‘6 lo. .2. :4. a... 1: 8 8.3.8. 3.88.... 288.. .28... :38. 2...... 3 8 . 3...! .. 8.. 22.8.. 8.... .38. an. 8 . :58. .88.... .98.... “.38.- 88°. Rn a . Elwma 2.38.. .88.. 386.- 868.. 8. B- - .238. 328.. .892. .38.- .88. . 9. 8- . .288. 32.... .28.. . 8.8. 82.8. R .2364 . ...-.B« .38.... 8...... 138.- 38.- a. 8 . .....ba :38. 8&8. . .38.. 88.... 8 . . . $8.. .88.. . .08.. $88.. I. 8 . .... is... i 0.80 . Ilia-.141} .o. .3 .s. 3. .3 .o. .3 .~. 2. 'l 3 169 .88.. So. 8... .8. «38. 838.. R... 2.... 2...- 2.8. 328.. .3 3:. a: 2.8.- 38:... 3... 2... n3. 8......- 23.8.. .8... a... 8.... .88. 338.. 8... 3.... S...- :8. .38.... 09. Rod .3... 8.8.- $58.. 2... ac... 2». ~38.- .88.. 3... 2... En. cues...- .88.. 8.8.. «.508; 908.0 3.38.. 2.8.- .8. . 9.. 9.5....- .28.... 3a.... 35...... 5.80.0 288.. Rec... 22...... 28.. 4.888. . es88. - rain. 05 Ni... 4.0.. fin...- Mla-qJ-I 55.81! a: .0) 0.9.0.5 Ill-.4638... .14 no... ..8. 8n. 3.... an... .8.- ..88 8...- 8a.. 93... 93. an“. .3... 2n. .3. .8. 03.. 8....- 8... .2...- .8." 08. ma... So. :53 nun—3% :25... no.3... .83.... $20. .128. . 2.8.... 3.8.. 2.8.- $38.. 88... $88.. 5.8. 238.. .88.- $88.. 8......- has... . 2.....- $.13 .28.... 28°. .32 9.. 3. 3. £88.. .95.... .82.... 89.....- 288.. 38... ..csoo.. ensue.- $38.. 58.... .2 2.... 5.8. ..38... . k8. - on..." 148.84.le .000... 3333 300300 823...... 530.. one»... so not. 158.0300 .2. ...~. e... in. 8 83...... 3.8.8.3.. 8.3.6.. .88.... nan... Sn. 3... ~88. .399.- .~.58.. .22.... R... «3.. «.3.- a~8.- 2.8. 888.. .82.... .8. 53.. a .. 2.8.. 8.8.. . . 838.. .38.... .8. 3...- Se: 33...- 9...... .2. 8.. 89.8.. 8... a... . no... a. . 8. 9.1.8. .88.... .92.... ..o. 3... 3..- 3.8. .38. .38.. .38.. .8. 8a.. 8a.. 28.... 92.8. :88. .89... 8... an... 3.. .38.. mo...- 28. .23.... as... 2....- c..88.-e98. . Hanna 08F I'd-0.54.. . .. 3.3.38 - 9.300- 038 so... so 308. .0.- 83033. as! = and“ :09 go 3.8 «no. ‘0. is a a. . 8...... an 8 - 3...... A 8 . .188 o. 8 - 2.-..3 3 8s - .358. 3 8 - 3.38 8. 8 . 3.0:: 0 8 - .050. .o.. .2. 3.. .... .2. .n.. .2. .... Appendix B 170 171 RESULTS OF TESTS WHICH CORRECT FOR DIFFERENCES IN THE RELATIVE RISKS OF MATCHED GROUPS OF ACFS AND NCFs Because of the importance of the fig matching re- quirement of the research design utilized in this study, this appendix considers two inherent problems in achieving a perfect matching of fig' The first problem,covered in Part 1, is the difficulty of matching the point estimates of‘gg. This problem is discussed in two subparts: A, which involves a sample revision which was undertaken to more nearly equalize matched 5g? and B, which involves sup- plemental tests that were conducted in order to determine whether the remaining 8g differences could be ekpected to cause spurious results in the principal tests of this study. The second problem,which is discussed in Part 2, results from the necessity of using statistical estimates of fig‘ Part 2 gives the results of additional supplemental tests which take into account the confidence intervals around the as which were used for matching purposes. In both parts of the appendix, the supplemental tests were performed on the hypothesis group which was most likely to have been affected by the problem discussed therein. 172 Part 1 - Problems in Matching Point Estimates ofggg Subpart A ~ Sample Revision The initial selection of a control group pro- duced some differences in excess of .05 between the esti- mated group relative risks LE8) of ACFs and NCFs. These differences were considered to be too large in view of the fact that the major requirement of the matching procedures called for by the research design is that the two as of a matched pair(of groups of firms)be equal. A case in point is the low-risk pair for which the AC was discretionary and increased EPS, i.e., ACF (+,D). Initially the NCF average fig was .973 for low-risk firms in the sample, or .049 greater than the low-risk ACFs' 6% of .924. The mean vector test for the period of t = -6 through -l using these firms produced a negative average return difference (of ACF - NCF) which was significant at less than .05. How- ever, when eleven of the NCFs with the risk coefficients that had the largest paired Bi differences were replaced by eleven NCFs whose g1 combined with the NCFs retained to reduce the average 58 difference to .016 (see table VI), the average return difference for the revised sample was not statistically different from zero. 175 In addition, several NCFs which were initially matched with ACFs in other hypothesis test groups were also replaced in order to reduce the differences between ACFs' and NCFs' matched ég' But in no case other than the one mentioned above did the replacement of firms in the sample significantly alter the results of the test. Subpart B_: Tests Which Correct for Differences between the Point Estimates of Pairedggg Even after several of the initial 560 NCFs in the sample were replaced in order to make the two risk vectors of a pair more nearly equal (as described above), some differences in fig naturally remain. The largest such difference of .038 occurs between the low-risk pair for the hypothesis test of nondiscretionary ACs with positive effects on EPS, i.e., ACF(+,N) (see table VI). Because the research design utilized in this study requires that $1 = a2 in order to be able to attribute differences in realized return distributions to the information content of ACs, it was feared that a difference as large as .058 could distort the results of the mean vector and covariance matrix tests on ACF (+,N). Table VI indicates that the two NCF 6g exceed the two ACF fig by .02 for high-risk 174 firms and by .038 for low-risk firms. This difference implies that the vector of expected returns on these par- ticular NCFs exceeds the vector of expected returns on ACFs (see eq. 1) In order to measure the effect that such a dif- ference could have on the tests conducted in this study, the mean vector test and the covariance matrix test of ACF(+,N) were performed a second time for all three time periods. In these cross-validation tests, corrections were made for the difference between the fig of ACFs and NCFs. In the test on mean vectors the correction in- volved incrementing the realized returns on the two risk- class groups of ACFs by the respective risk-class 68 dis- crepancies multiplied by the average realized risk premium (realized Rmt - th in Eq. 1) over the period covered in the test. The correction thus raised the average return for each risk-class group of ACFs to the theoretical equi- librium (in relation to the average return for the cor- responding risk-class group of NCFs) of Eq. I (because the fig of NCFs exceeds the fig of ACFs). The resulting incre- ment to the average monthly return on high-risk ACFs was .02 multiplied by .00908, or .00018, and the increment to the average return on low-risk ACFs was .038 multiplied by .00908, or .00035. The average monthly risk premium of 175 .00908 was the weighted average risk premium (the average monthly return on a portfolio consisting of the Standard and Poor's 500 stocks minus the average monthly return on 90-day U.S. Treasury bills) over the 1968-1972 period of this study. In the averaging process the weights were the relative proportions of firms in this particular test that came from each of the five years, l968-l972. The weighted average risk premium thus created was the actual risk pre- mium that investors in these particular firms faced during the test period. It was unusually high (.00908) because 48.5 percent of the firms came from 1971, when the average monthly risk premium was .00858, and 41.7 percent of the firms came from 1972, when it was .01285.1 The condition- al covariance matrix (Eq. 8) of the corrected conditional return difference vector (defined in the Hypotheses sec; tion of Chapter 5) was assumed to remain unchanged as a result of thegg correction. Although it may have, in fact, changed, there is no reason to believe the change would be substantial with such small changes in the mean difference vector. But because this is an empirical is- sue, there is no theoretical model to suggest what the change may be. The results of these cross-validation tests were essentially the same as the results on the major mean 176 vector tests conducted in this study, in which the E8 dif- ferences were allowed to remain. Specifically, in the test over the period t = -6 through -1, the inference from the result of the test was that returns on ACFs and NCFs were the same, with a confidence level of .75 or greater. In the tests over the periods t = +1 through +6 and t = -6 through +6, the inferences from the results were that returns on ACFs and NCFs were different, with confidence levels of .949 and .975, respectively. As can be seen from the results of these tests on line fourteen of Table IX, the differences in results are nominal. In the test for equality of covariance matrices, the correction for the difference between ACFs' and NCFs' fig uses the definition that the variance of the return on any portfolio p is equal to the systematic variability BZVBJ'(Rm), plus the unsystematic variability, Va1'(ep), from holding the portfolio. Thus, the effect of the E8 discrepancy on the results of the test may be corrected by incrementing the conditional covariance matrix (Eq. ll) of ACFs by 62Var(Rm) = (.02)2 multiplied by .034 for the vari- ance of the return on high-risk ACFs and by (.058)2 multi- plied by .034 for the variance of the return on low-risk ACFs.2 The covariance between the returns on high- and low-risk ACFs, as well as the covariance between the returns 177 on high- and low-risk NCFs, was assumed to remain constant. As with the tests on mean vectors, the results of the tests for equality of covariance matrices over the three periods were essentially the same as the results obtained in tests in which the fig difference was allowed to remain. Part 2 - Problems in Using Estimates of;g In an entirely separate set of cross-validation tests on mean vectors of ACFs that made nondiscretionary ACs with positive income effects, i.e., ACF(+,N), the two (ACF and NCF) 9g were assumed to differ by as much as would be allowed by a ninety percent confidence interval around the two Eg' That is, for the hypothesis test being considered, ACF +,N), where the fig of high-risk NCFs ex- ceeds the 68 of high-risk ACFs by .02 and the 58 of low- risk NCFs exceeds that of low-risk ACFs by .038, it is possible, because the fig are only estimates of the true 88, that the differences are greater than .02 and .038, respectively. Therefore, in order to determine whether this possibility is likely to have influenced the inferences drawn from the test of ACF(+,N), the high- and low-risk fig differences were allowed to vary up to .316 and .236, 178 respectively. These estimates were derived by hypothe- sizing that the high-risk fig ofuggfis may be as low as 1.412, which is equal to its point estimate of 1.56 minus the product of 1.645 (the number of standard deviations defining the ninety percent confidence boundary of a normally distributed random variable) multiplied by its estimated standard error of .09.3 The high-risk‘flgfis' Eg was allowed to reach its ninety percent confidence maximum of 1.728 (= 1.58 + (1.645 x .09)). The low-risk firms' fig difference was allowed to reach its estimated maximum in like manner, but using .06 as the estimated standard error of the two low-risk fig. Then the mean return vector of ACFs was incremented in the manner discussed above, and the T2 tests were conducted again. The results were es- sentially unchanged from the other (corrected and uncor- rected) test results. In the cross-validation test on covariance matrices in which the two E6 were allowed to reach their ninety percent confidence maximum difference, a different result emerges. The greater resulting differences between the variances of high-risk ACFs and NCFs and the variances of low-risk ACFs and NCFs were large enough to assign probabilities of greater than .975 to the inference that ACFs' covariance matrices of returns differ from NCFs' 179 covariance matrices of returns for tests over all three time periods. This suggests that the covariance matrix test is more sensitive to large differences in the matched E8 than the test On mean vectors is. This result means, of course, that the covariance matrix test is able to de- tect large differences. It also implies that the covari- ance matrix test conducted on samples of as few as six monthly observations may not be as lacking in statistical power as was implied in the Chapter 6 section entitled Tests on Covariance Matrices. But another implication is equally clear--that the estimation of fig with error can cause spurious results in the covariance matrix test. On balance then, it appears that the conclusion of in- determinacy which is reached in the Chapter 6 discussion is warranted. Furthermore, and unfortunately, there ap- pear to be no alternative means of estimating fig with sub- stantially less error than the method used here, given this research design. 0n the basis of the results of the cross-valida- tion tests described in Parts 1 and 2 above, it appears that the largest sample difference between the two fig of a matched pair does not distort the results of the mean vector test which is affected by that difference. By im- plication then, it is reasonable to assume that none of 180 the other hypothesis tests is adversely affected by the (smaller) differences between the fig of ACFsand NCFs. How- ever, it is possible that the same differences in fig can seriously affect the results of tests on covariance matrices. C H A P T E R 7 SUMMARY, CONCLUSIONS, CONTRIBUTIONS OF THE STUDY, AND SUGGESTIONS FOR ADDITIONAL RESEARCH Summary Previous research on the stock market behavior associated with accounting changes (ACs) has focused pri- marily on the question of whether the market was deceived by ACs. That is, researchers generally made assumptions about the information content of ACs and tested to see if the market behaved in the hypothesized manner. They also assumed that the market's behavior is constant across all securities, and only two of the studies have reported the results of tests conducted on a control group of firms that did not make ACs. Finally, previous researchers have all examined only the first moment, the mean, of security re- turn distributions in making their assessments of the ef- fects of ACs. This study utilized a recently devised research design in order to assess whether and how the stock market reacts to ACs. The market was assumed to be efficient with respect to publicly available data (such as AC 181 182 disclosures), and the capital sumed to reflect the market's equilibrium values for firms. changes during 1968-1972 were editions of Accounting Trends consists of 560 change firms. asset pricing model was as- mechanism for establishing Firms that made accounting taken from the apprOpriate The sample and Techniques. Also, 560 nonchange firms were selected as a control group from the population of firms that did not make ACs. at the firm level on the bases of fiscal year-end, tive risk, The two groups were matched rela- and less stringently on industry membership. The matching procedures were designed to eliminate syste- matic differences between the two groups in order to maxi- mize the probability that observed return differences re- flect the market's reaction to ACs. The firms were divided into two groups on the basis of their relative risks with the tOp half thus labeled as high-risk and the bottom half labeled as low-risk. Then multivariate statistical tests were conducted on mean vectors of returns and covariance ma- trices of returns in order to return distributions (of ACFs cally. determine whether the two and NCFs) differ statisti- The multivariate research design gives effect to the performance of simultaneous tests on the information effects of ACs for the two risk classes. The separation of firms into different risk classes provides the framework 183 for testing whether AC information is unique to a particu- lar risk class (as defined above). The test period consisted of the thirteen months centered at the second month after each firm's fi§cal year- end. This month is believed to include the date when mpg; investors become aware that firms have made ACs.l Where month t = 0 is the assumed post-year-end disclosure month, the entire period ran from t = -6 through t = +6. Tests were conducted on data pertaining to the full thirteen- month period, as well as to the two six-month periods, t = -6 through -1 and t = +l through +6. The separation into the first and second subperiods allows for testing the timing of any unusual market behavior. Conclusions Tests were conducted on various subgroups of the 560 matched pairs of firms. The largest such group was the entire sample, and the results of this omnibus test corroborated the findings of Ball, who concluded that ACs are not associated with a unique market effect. This test analyzed many different types of ACs, which could be ex- pected to have widely different market effects. Thus, the blanket inference of no effect could be erroneous in the 184 sense that it does not pertain to subgroups of ACs with more homogeneous characteristics. In order to determine whether the aggregation hypothesis has validity, tests were also conducted on ACs that had four different directional effects on net income: (1) positive; (2) negative; (3) zero or immaterial; and (4) directional effect not disclosed. The results of all of these tests were also consistent with the null hypothe- sis of no information content. Because each of these tests also aggregated across ACs with widely different character- istics, the results are not surprising. One such characteristic hypothesized to be of interest to stock market agents is whether the ACs are made at the discretion of management or whether the firm is compelled by some exogenous body such as the Accounting Principles Board (APB) or the Financial Accounting Stan- dards Board (FASB) to make the change. Accordingly, tests were also conducted on ACs that had three different dis- cretion characteristics: (1) discretionary; (2) nondis- cretionary; or (3) both, i.e., the firm made at least one discretionary AC and at least one nondiscretionary AC dur- ing the same year. All previous AC research in which tests were conducted on specific types of ACs concerned discre- tionary ACs only. This study provides the only evidence 185 of which the author is aware on the market effect of nondis- cretionary A08. The results of the tests on discretionary ACs, and the results of the tests on nondiscretionary ACs both led to the inference that ACs do not have an informational effect on security returns. The results of the test on the both category, however, led to rejection of the null hypothesis and hence to the inference of information con- tent. This test was shown later to have been significant- ly influenced by the presence of a large number of firms for which the overall effect on net income was positive. In general, however, the level of aggregation in these tests was still high enough to obscure the unique effect associated with some more narrowly defined ACs. In addition, tests were conducted on nine of the twelve subgroups of ACs defined by the various combina- tions of directional effect and relative discretion (e.g., discretionary ACs with positive net income effects, etc.). Three groups were dropped since the sample sizes were con- sidered too small to yield meaningful conclusions. Sever- al of these tests yielded statistically significant dif- ferences between the mean return vectors of AC and non- change firms. For example, the market reacted negatively to firms that made discretionary ACs that increased net 186 income, in relation to NCFs. These results are generally consistent with the findings of Kaplan and Roll, but they are more general because of the wider variety of accounts affected by the ACs examined in this study. The market reacted negatively to firms that made both a discretionary AC and a nondiscretionary AC with an overall positive effect on net income. A negative market reaction was also observed for firms that made non- discretionary changes that reduced net income (but the em- pirical results of this testtne likely to have been sig- nificantly affected by sampling error because of the small number of firms involved). “The two significantly positive market reactions were to nondiscretionary ACs that in- creased net income and to discretionary ACs that had no effect on net income. All other tests resulted in insig- nificant differences between the returns of AC and non- change firms. Thus, it appears that the types of ACs referred to in Table XIV are associated with unique stock market effects. Furthermore, the second six-month period (after the year-end disclosures of ACs) yielded the most pro- nounced market adjustments. The tests conducted in this Study can only firmly establish association, not cause- aJldreffect relationships. Nevertheless, one possible 187 implication of these findings is that the year—end dis- closures of ACs (chiefly in footnotes to financial state- ments) are used by market agents in establishing equilib- rium values of firms. This particular conclusion is not too startling in light of the fact that there are not many competing sources of AC information. However, to the extent that the information is an indicator of firm prof- itabilitm dividend policy, or other widely followed char- acteristics of firms, this finding places some new impor- tance on AC disclosures as a source of information. The two tests for which the significant differ- ences occurred during the six months prior to the year-end disclosure involved firms that made both a discretionary AC and a nondiscretionary AC. This result appears to have been influenced by the presence of a large number of firms that made discretionary ACs with relatively large increases in net income. An explanation of this result, which ap- pears to be consistent with evidence provided by Bremser,2 is that firms which make discretionary ACs with positive effects are significantly less successful than nonchange firms in terms of profitability and return on stockholders' equity. In short, such firms may be using AC8 to increase net income in order to conceal otherwise unsatisfactory Operating performance. This possibility raises a question 188 as to why the market did not react negatively during the first six-month subperiod to the discretionary ACs made by the discretionary-positive effects group of firms. If one were to accept the results of that particular test, subject to the probability of a type I error equal to around .20, then he could conclude that all discretionary ACs with positive income effects examined in this study are associated with a negative market reaction in the six- month period prior to the year-end disclosure month. How- ever, the mild inconsistency between the two test results still remains somewhat of a mystery. In each test that led to rejection of the no in- formation hypothesis, the return distributions of the two risk classes of firms exhibited substantially different behavioral patterns. This result suggests that the sep- aration of firms into risk classes was a worthwhile exer- cise in that it allowed for the risk dependencies of the AC information to be revealed. However, in several cases the nature of the risk dependency is far from clear. This study merely indicates that there appears to be a risk dependency associated with AC information. It offers rela- tively few clues as to the reasons for the risk dependen- cies observed. More evidence is needed before any firm conclusions are reached as to why the market reacts 189 uniquely to the AC information of a particular risk class of firms. The results of tests for the equality of covari- ance matrices of returns on change and nonchange firms were indeterminate because of a combination of factors re- lated to the research design, the failure to satisfy cer- tain antecedent conditions of the tests, and the statis- tical power of the tests conducted. As a result, the tests on covariance matrices yielded ambiguous conclusions. Contribntions of This Study The contributions of this research are two-fold. One aspect is technical, and the other concerns the content of the study and the related conclusions. The technical contribution stems from the application of a new research design (or technique) for testing the market effects asso- ciated with accounting events such as ACs. The major part of the technical contribution is merely a derivative of Gonedes' work because he formulated the basic design used here. However, several extensions of his work that were introduced in this study are believed to be of value. One is the set of procedures that were followed in order to obtain the control group of nonchange firms. The matching of firms on estimated relative risk and industry membership 190 largely eliminates those variables as competing explana- tions for the observed results. As a result, the return differences between the two groups are more likely to be related to ACs than in previous AC research. Inasmuch as the risk coefficients used for match- ing change and nonchange firms are only estimates that are subject to error, appendix B to Chapter 6 contains the re- sults of some cross-validation tests. These tests were based on the assumption that change firms' true relative risks differ from those of nonchange firms by the widest margin that a ninety percent confidence interval would allow. The results of the cross-validation tests on mean vectors are very little different from the results of tests in which the point estimates of risk are treated as though they were true values. Of course, this added step lends more support to the foundation of internal validity underlying the conclusions of the study. Results of the cross-validation tests on covariance matrices support the opposite conclusion--that possible differences between the true risk coefficients could lead to spurious results in the covariance matrix tests. However, because those re- sults do not, in general, lend themselves to meaningful interpretation, this finding was not considered much of a loss to the import of the study. 191 The contributions involving content chiefly center around the size of the sample--560 AC firms over the five years 1968-1972. The large sample, as well as the specific years included, allowed for tests to be per- formed on a wide variety of ACs, some of which had not been subjected to testing previously. The principal ex- ample is nondiscretionary ACs. This evidence complements the evidence that was already available on the market re- action to discretionary ACs. In addition, this study pro- vides evidence on a wider variety of discretionary ACs than any previous study has done. And finally, the results obtained here suggest that the information content of ACs is somewhat risk-dependent. Beaver suggests that the Financial Accounting Standards Board should consider evidence on the market consequences associated with various modes of reporting financial data. In this regard he states that, . . . although evidence cannot indicate what choice to make, it can provide information on the poten- tial consequences of the various choices. Without a knowledge of the consequences (e.g., as reflected in security prices), it is inconceivable that a policymaking body such as the FASB will be able to select optimal financial accounting standards.:3 Beaver is referring to choices such as straight-line versus accelerated depreciation, equity versus cost for invest- ments, etc. This study presents evidence on the consequences 192 of changes from one method to another. Following Beaver's line of reasoning, one concludes that bodies such as the FASB should be interested in the results obtained here. One reason is obvious: these results provide evidence on how investors react to various types of ACs and by impli- cation to the different reporting modes involved. The re- sults also provide an indication of the consequences of actions taken by the APB (and its predecessor organization, the Committee on Accounting Procedure) to require AC dis- closures.4 Finally, the fact that the market appeared to react systematically to nondiscretionary ACs also suggests that APB Opinions and other authoritative pronouncements make a difference to market agents. In summary, this study presents evidence which should be useful in understanding investors' decision models. How the FASB views these results depends, of course, on the subjective judgments they place on the re- sults. They may conclude that certain investor reactions were irrational and resulted in a serious misallocation of resources. 0n the other hand, they may accept the position taken here that the market is efficient and interpret the results in terms of the assumed rationality of investors. In either case, this study provides them with new and ex- panded evidence on the market consequences associated with ACs. 193 Another implication of the results of this study pertains to business managers. Financial theory states that they are motivated by the desire to maximize the values of the firms entrusted to them by shareholders. If they were pursuing this objective, why then did a large number of the firms make discretionary ACs that increased net income? Did they not know in advance that the market reaction to such an AC would be demonstrably negative? However, because it is easy after the fact to find an ex- planation for observed phenomena, and then to label the explanation a prediction, phggg particular managers are not to be condemned too harShly. However, future genera- tions of managers would do well to heed these results, and to predict future market reaction toward their A08 in light of this evidence. Suggestions for Additional Research This study is subject to one principal line of criticism--that it did not adequately distinguish between the AC or the resulting auditor's consistency exception as the independent variable being measured. However, because only around half of the sample AC firms received consis- tency exceptions, it seems safe to conclude that the AC was the event more closely associated with market effects. 194 Nevertheless, a detailed inquiry into the effect that consistency exceptions have on security returns is already under way. Also, it is possible that ACs with different magnitudes of effect on net income can affect returns dif- ferently. The question of materiality has received much attention in the professional and academic literatures, but has been subjected to very little empirical testing outside the laboratory environment. A project in this area is planned for the future inasmuch as the data needed for such an inquiry have already been gathered. Another refinement of the tests conducted in this study could be separate tests on ACs affecting specif- ic accounts. Depreciation-related and inventory-related ACs have already been examined, as well as changes in- volving the investment tax credit. The sample in the present study includes a large number of pension-related ACs, most of which increased net income. The fact that some firms increased the assumed earnings rate on their pension fund assets, while others did not, suggests that the former group of firms may have timed their A08 in order to increase net income during periods when their operating profits were not meeting expectations. It would be inter- esting to know the market's reaction to pension-related 195 ACs in particular. Similar reasons can be advanced for examining other types of ACs as well. A series of studies on the effects of ACs involving specific accounts would seem to be a natural way to complement the more general results obtained in this study. There is an extensive literature on the relative stability of risk coefficients, but relatively little is known about the effect ACs have on firms' relative risks. Ball and Sunder have examined this phenomenmm but their results by no means provide a complete description of the effect, if any, that ACs have on risk levels. This ques- tion may be related to the fact that each of the unique AC effects observed in this study pertained to only one of the two risk classes of firms. It is possible that the AC information actually affected the firms' risk coeffici- ents, and this effect may have been directly responsible for the differential behavior in security returns. This chain reaction of hypothetical effects is consistent with the major implication of the capital asset pricing model. But it would take additional evidence in order to determine whether this is in fact what occurred to give rise to the differences observed here. Finally, it would be interesting to look into the behavioral implications of ACs and their market effects. 196 As mentioned above, managers of firms are supposed to be motivated by the desire to maximize the values of their firms. Yet a large number of the change firms in this study made the (ex post) irrational decision to make an AC which would increase the firm's net income. Given this evidence, it would be enlightening to query managers who made those decisions in order to gain some insight into, for example, the source of their motivation for making the change. It would also be interesting to learn of their perceptions, both prior to and after the AC, of the mar- ket's reaction, as well as whether, in retrospect, they would make the same decisions again. FOO TNO TES 198 Footnotes - Chapter 1 1Two examples are: Ronald M. COpeland, "Income Smoothing," Empirical Reseagch in Accounting: Selected Studies, 1968, Supplement to Journal of Accounting Re- search 6 (1968): 101-121; and Barry E.Cushing, “An Empiri- cal Study of Changes in Accounting Policy," Journal of Accounting Research 7 (Autumn 1969): 196-203. 2Barry E. Cushing, "Accounting Changes: The Impact of APB Opinion 20," Journal of Accountancy 138 (November 1974): 54-62. 8Three examples are: T. Ross Archibald, "The Return to Straight-line Depreciation: An Analysis of a Change in Accounting Method," Empirical Research in Ac- counting: Selected Studies, 1967, Supplement to Journgl of Accounting Research 5: 164-180; Martin L. Gosman, "Charac- teristics of Firms Making Accounting Changes," The Ac- counting Review 48 (January 1973): 1-11; and Wayne G. Bremser, "The Earnings Characteristics of Firms Reporting Discretionary Accounting Changes," The Accounting Review 50 (July 1975): 563-573. ‘Two examples are: Fred Neumann, "The Auditing Standard of Consistency," Empirical Research in Account- ing: Selected Studies, 1968, Supplement to Journgl_of Accounting Research 6 (1968): 1-17; and Elba F. Baskin, "The Communicative Effectiveness of Consistency Excep- tions," The Accounting Review 47 (January 1972): 38-51. - + SReturn (R) = P1 P0 Dl where P = price, D = dividends, and the subscripts refer to arbitrarily-defined points in time. 6Examples are: Eugene E. Comiskey, "Market .ReSponse to Changes in Depreciation Accounting," The Ac- counting Review 46 (April 1971): 279-285; T. Ross Archi- bald, "Stock Market Reaction to the Depreciation Switch- Back," The Accounting Review 47 (January 1972): 22-50; Robert S. Kaplan and Richard Roll, "Investor Evaluation of Accounting Information: Some Empirical Evidence," Journal of Business 45 (April 1972): 225-57; Baskin, pp. 38-51; Raymond J. Ball, "Changes in Accounting Techniques and Stock Prices," Empirical Research in Accounting: Selected Studies, 1972, Supplement to Journal of Accounting Research 199 10 (1972): 1-38; and Shyam Sunder, "Stock Price and Risk Related to Accounting Changes in Inventory Valuation," The Accounting Review 50 (April 1975): 505-515. 7Over the long run stock returns and stock prices are synonymous because the only nonprice element in returns, i.e., dividends, is implicitly included in prices which adjust for dividend declarations and distributions. Therefore returns and prices will be used interchangeably. 8Baruch Lev, Financial Statement Anglysis: A New Approach (Englewood Cliffs, N.J.: Prentice-Hall, 1974), p. 241. 9Nicholas J. Gonedes and Nicholas Dopuch, "Capi- tal Market Equilibrium, Information Production, and Select- ing Accounting Techniques: Theoretical Framework and Review of Empirical Work," S ud' s on 'n nci 1 count'n 0b ec- tives: 1974, Supplement to Journ of ccountin ese rch 12 (1974): 48-129. In their quotation Gonedes and Dopuch use eit to refer to the return residual of the ith entity for the tth time period, which is conditioned on the in- formation variable 9 with the same subscripts. 10Sunder observed the latter result. See Sunder (1975), pp. 305-515. 11An expanded discussion of ACs with different economic implications is given in Gonedes and DOpuch, pp. 84-91. leSee, for example, Eugene Fama, "The Behavior of Stock Prices," Journal of Business 28 (January 1965): 34-105. Not enough is known about kurtosis to be certain of what it measures. laJames Tobin, "Liquidity Preference as Behavior towards Risk," Review of Egongmig Studies 25 (February 1958): 65-86. 1‘M. K. Richter, "Cardinal Utility, Portfolio Selection, and Taxation," Review of Economic Studies 27 (June 1960): 152-166. 15Value Line; Merrill Lynch, Pierce, Fenner & Smith and other firms periodically publish companies' Beta coefficients for investors' use. Intuitively a firm's 200 Beta is the measure of the sensitivity of that firm's stock return to the return on some market index. 0r stated differently, Beta measures the systematic portion of a firm's stock variability. A more rigorous definition will be given later when firms' Betas will be considered in some detail. 16Baskin's study ostensibly tested the effect of consistency exceptions. Yet, his design does not distin- guish whether the event of interest was the consistency exception or the AC; accordingly his study is considered an AC study. 201 Footnotes - Chapter 2 1This definition of information content is cov- ered in Joel S. Demski, Information Analysis (Reading, Mass.: Addison-Wesley, 1972), p. 14. 2See, for example: Raymond J. Ball and Phillip Brown, "An Empirical Evaluation of Accounting Income Num- bers," Journ l of Accountin ese rch 6 (Autumn 1968): 159- 177; Eugene F. Fama, L. Fisher, M. Jensen, and R. Roll, "The Adjustment of Stock Prices to New Information," lg- terngtiongl Economic Review 10 (February 1969): 1-21; and William H. Beaver, "The Information Content of Annual Earnings Announcements," Empirical Research in Accounting: Selected tudies 968, Supplement to Journal of Accounting Research 6 (1968): 67-100. aEugene F. Fama, "Efficient Capital Markets: A Review of Theory and Empirical Work," Journal oleingnce 25 (May 1970): 383-417. 4‘These two groups of firms are believed to be relatively small in relation of the pOpulation of ACFs be- cause examination of a full year's Wall Street_Journgl Index entries for each of twenty firms included in the present sample yielded disclosure prior to the annual re- port for only one firm. That one disclosure, in a quar- terly earnings summary, merely stated that the firm had made a type x change. In almost all cases that a firm discloses its AC prior to year-end, there is reason to believe that the market may react to the AC around year- end because it would be difficult to predict in advance exactly what effect the AC may have on net income. For example, the income effect of inventory changes depends on the value of the ending inventory balance. A change to the equity method of accounting for investments depends on the net income of the investee. Changes in deprecia- tion methods depend on the ages of depreciable assets (which are not normally disclosed) and current-year ac- quisitions. 5Shyam Sunder, "Relationships between Accounting Changes and Stock Prices: Problems of Measurement and Some Empirical Evidence," Empirical Research in Accounting: Selected Studies 1973, Supplement to Journal of Accounting Research 11 (1973): 1-45. 202 6See William H. Beaver and Roland Dukes, "Inter- period Tax Allocation, Earnings Expectations, and the Behavior of Security Prices," The Accounting,Review 47 (April 1972): 331, for one example. 7William F. Sharpe, "Capital Asset Prices: A Theory of Market Equilibrium under Conditions of Risk," Journal of Finance 19 (September 1964): 425-442. Sharpe's work was an extension of the portfolio theory work done by Harry Markowitz, Portfolio Selection: Efficient Diversifi- cation of Investments (New York: Wiley, 1959). 8One aspect of return distributions which has been examined is the individualistic portion, eit: of the conditional realized~return distribution where eitlei " E(eit) = Ritlei - E(Rit)° Here 31 represents the account- ing event, where i can take on two values, 1 for the pres- ence of the event and 2 for the absence of the event. Known as the Abnormal Performance Index (API) method, it draws inferences about the information effects of 31 de- pending on whether eitlei is equal to its equilibrium expectation of zero. A nonzero eit|9i means that 31 has information content. Conversely, a zero eitlei signals no information. Tilde (~) denotes a random variable. 8Fischer Black, "Capital Market Equilibrium with Restricted Borrowing," Journal of Business 45 (July 1972): 444-455. loFischer Black, Michael Jensen, and Myron Scholes, "The Capital Asset Pricing Model: Some Empirical Tests," in Studie§_in the Theory of Capital Markets, ed. Michael C. Jensen (New York: Praeger, 1972). 11Eugene F. Fama and James MacBeth, "Risk, Return, and Equilibrium: Empirical Tests," Journal of Political Economy 61 (May/June 1975): 607-656. 12Michael C. Jensen, "Capital Markets: Theory and Evidence," Bell Journal of Economics and Management Science 5 (Autumn 1972): 357-398. 13Fama (1965), pp. 54-105. 1‘Nicholas J. Gonedes, "Risk, Information, and the Effects of Special Accounting Items on Capital Market Equilibrium," Report No. 7429, Center for Mathematical Studies in Business and Economics, University of Chicago, June 1974, Table 10. 205 lsDemski, p. 14. 16Gonedes (1974a), pp. 2.1-2.15. 17Actually the design used here is the com ari- son ofAAPIs for two groups of firms with different 91 and equal 81. Letting 61 represent an AC and 92 represent no AC the group of ACFs' realized return is Rllel = E(R1) + e1 81. Where 81 = Bg, E(Rl) = E(R2): and the analysis reduces to a comparison of ellel and ezlez. This assumes, of course, that 81 = 82 = Bconstant: which may not remain true throughout the experiment. If 51 # 82 during the test period,then Rllel # R2l92 may be due to a differen- tial shift in the 81 of the two groups. For an expanded discussion of the effect of a changing 81, see the final section of Chapter 3. 18For a discussion of variance partitioning see Jack C. Francis and Stephen Archer, Portfolio Analyais (Englewood Cliffs, N.J.: Prentice-Hall, 1971), p. 179. 19See Kaplan and Roll, pp. 257-259. 2°See Sunder (1975), p. 515. 21Ball and Brown (1968), p. 167. 22Louis H. Rappaport, SEC Accounting Practice and Procedure, 3rd edition (New York: Ronald Press Com- Pany, 1972) , p. 14.5. 204 Footnotes - Chapter 3 1If, for example, Bi were increasing over the 204-month period of his study, Archibald's constant risk assumption could have caused his estimates of St to be understated in the prechange period and overstated in the postchange period. The reverse would be true if 5i were decreasing over the period of the study. If the rate of change in Bi were relatively constant, as Ball observed (see Ball, figure 4), the effect on the full four-year period would be small, but the effects on individual pre- and postchange periods may be substantial, depending on the rate of change. It is impossible to know what the ef- fect was without knowing the behavior of 8- during the test period. But if the Bi behavior in Archibald's design was similar to that observed by Ball, this could explain why the average et for the prechange months were signifi- cantly less than zero. 2The exact time period of this test is not clear. Because the later and more important test was con- ducted over five weeks, it appears reasonable to assume that this test also covered five weeks. 3To see why this is true, refer again to Ball, Figure 4, which indicates that the average estimated Bi of the 267 ACFs of Ball's study increased from approximately .91 in the 110th month before the AC disclosure to approx- imately 1.02 fifty months after the disclosure month. The rate of change during the one month after the disclosure was high, but in absolute terms the change is estimated to be .002, hardly enough to cause spurious results. 4'Ball makes this observation on page 31 of his study. 55cc Sunder (1975), pp. 505-515. 6For difficulties with normative assumptions about the market's response to particular types of ACs, see Chapter 1, page 3 and Gonedes and DOpuch, pp. 84-91. 7See Gonedes (1974a), Sections 2.2 and 2.5. aIbid., Section 5.5 and Table 7. 9Ibid., Section 2.2. 205 l°Gene v. Glass and Julian Stanley, Statistical Methods in Education and Psychology (Englewood Cliffs, N.J.: Prentice-Hall, 1970), pp, 491-494, 11See the Chapter 5 discussion following Eq. 9. 12Gonedes (1974a), p. 4.8. 13R. Darrel Bock and Ernest A. Haggard, "The Use of Multivariate Analysis of Variance in Behavioral Re- search," in H ndbook of 8 sur m n n Ass ssment Behavioral Sgiences, ed. Dean K. Whitla (Reading, Mass.: Addison-Wesley, 1968), p. 102. Book and Haggard are not writing about security price research in which returns on different risk classes of firms are correlated. Neverthe- less, their statement, which concerns different teachers and textbooks, is perfectly analogous to a statement con- cerning different risk classes and security returns, re- spectively. 1"See William H. Beaver, "The Behavior of Secur- ity Prices and Its Implications for Accounting Research (Methods)," Supplement to The Accounting Review 47 (1972): 407-436, for a discussion of the CAPM and the market model, as well as the assumptions of the two models. 206 Footnotes - Chapter 4 1The CRSP tape, deve10ped at the University of Chicago, includes various pieces of financial data for all NYSE-listed companies from December 1925 through dates that are constantly being updated. The CRSP data perti- nent to this study are the apprOpriate total monthly rates of return (see footnote five in chapter 1) on the common stock of ACFs and NCFs in this sample. 2A firm was deemed to have an insufficient num- ber of monthly returns available for estimating its 81 if its Bi was not included in Merrill Lynch, Pierce, Fenner and Smith, Inc.‘ Securit Risk Evalu tion, from which firms' Bi were taken. In general, Merrill, Lynch esti- mates each firm' 3 Bi over Sixty monthly return observa- tions. However, where fewer monthly returns are available for a given firm, the number of available monthly returns is used, down to a minimum of twelve. An expanded dis- cussion of the B- estimation procedures is given below. 3What constitutes a "suitably matched NCF" is the subject of the remainder of this chapter. ‘During the test period the APB issued opinions 13-24. Among this group of twelve, the Opinions which gave rise to ACs were those requiring a different account- ing method from the method then in use for assets, lia- bilities, or capital items already entered in the accounts from previously consummated transactions. The Opinion contributing the largest number (105) of nondiscretionary ACs to the present sample was Opinion No. 18, "The Equity Method of Accounting for Investments in Common Stock." The number of nondiscretionary ACS resulting from all other Opinions issued during the sample period was only eighteen. Of these, twelve result from Opinion No. 23, which mandated the accrual of income taxes on the undis- tributed earnings of subsidiaries. 5This group of nondiscretionary ACs was quite small, only ten. Principal accounts affected were inven- tories and fixed assets. 8The fact that twenty-two steelAfirms made the same AC, plus the fact that steel firms' Bi do not differ markedly from one another, provides evidence which supports 207 the need to examine whether there is an association be- tween informational effects of AC disclosures and par- ticular risk classes. 7Sunder (1975), p. 515. 8Benjamin J. King, "Market and Industry Factors in Stock Price Behavior," Journal of Business 39 (January 1966): 159-190. 9Stephen L. Meyers, "A Re-Examination of Market and Industry Factors in Stock Price Behavior," Journal of Finance 28 (June 1975): 695-705. l°The 81 of 1968 ACFS and NCFs were not taken from Security Riak Evaluation because the publication was not begun until 1969. Instead, 1968 firms' 8. were es- timated in the same manner that Merrill, Lynch estimates Bi! all over sixty months' data. llSecurity Riak Evaluation contains two ty es of estimates of firms' 5i: One type is the unadjusted i’ and the other is Merrill Lynch's attempt to adjust for the regression tendency of Si. The unadjusted Bi are used in this study. 12For the procedure that defines each firm as high- or low-risk, see the Chapter 5 section, Procedures for Grouping Firms and Computing Returns. A 13These estimates of the standard errors of groups' BB were not computed in the normal manner, i.e., from a time-segies offifi which would be taken from a regression of Rgt on Rmt over t time periods. This method is not feasible here because firms that made ACs in dif- ferent time periods t are aggregated into each group g. Consequently, there is more than one Rmt pertaining to each Rgt. Fortunately, however, Fama and MacBeth have deveIOped a way of approximating the standard error of a group (or in their case, a portfolig) 68 (8p). They aver- aged the standard errors of firms' Bi in their respective portfolios. Then they computed the time-series estimates of the standard errors of the 8 of their (legitimate) portfolios and found that the ratio of the simple average standard error of the in a portfolio to the standard error of the portfolio was between three and seven. This means that the 8p 0? portfolios (of around size 40 208 in their sample) can be estimated with between three and seven times thp precision of the average of the standard errors of the Bi in a portfolio. See Fama and MacBeth, pp. 615-621. 14See Table‘TIof this study and Table 5 of Gonedes (1974a) for a comparison of group 5g and portfolio 5p: respectively. For the standard errors of Gonedes' ép, see Gonedes (1974a),p . 5.6. 15Gonedes (1974a), p. 219. .fi£ and §€ in Gonedes can be thought of as a; and 82; respectively, in this study, omitting the time subscript. 16See Kaplan and Roll, pp. 257-259 and Sunder (1975), p. 515. In each study the API's nonzero behavior is contained within a period of around six to eight months of the release of the preliminary earnings report. 17For a discussion of the McNemar test, see W. J. Conover, Practical Nonparametric Statistics (New York: Wiley, 1971), pp. 141-145. 18The 525 ACFS and 451 NCFs that are listed in 521 each had five 8- estimation years, or totals of 2, 625 and 2,155, respectively (ignoring the fact that around five percent of the firms' Bi were estimated over fewer than five years' data). The unavailability of the list of 1966 5;: firms that made ACS accounts for the fact that these numbers of B-estimation years are not 2,625 and 2,155. Specifically, the number of B- estimation years analyzed for the 525 ACFs is the sum of five years (1967- 1971) multiplied by the 151 ACFs for 1972 that were listed in ATT plus four years multiplied by the 394 ACFS for 1968-1971 that were listed in ATTJ for a total of 2,231. . The number of B-estimation years analyzed for the 431 NCFs is the sum of five multiplied by ninety-six plus four multiplied by 335, for a total of 1,820. 19of the 600 firms listed in the 1965 edition of ATT, 302 changed to the flow-through method of accounting for the ITC. See Kaplan and Roll, p. 229. During 1971 and 1972 the numbers of the ATT 600 that changed to the equity method were 86 and 66, respectively. 20See Ball, Table 2 on p. 7. 209 21The contrast in the authority of American Institute pronouncements is clearly seen in American In- stitute of Certified Public Accountants, AccountinglRe- search Bulletin No. 51 (August 1959): Notes, and American Institute of Certified Public Accountants, Accounting Erin- cipleg Board Opinion No. 6 (October 1965): paragraph 1 and Notes. 220f Ball's 267 ACS, seventy-five decreased net income. 0f the 730 ACs analyzed here, only sixty decreased net income. See Ball, Table 7. 23Briefly stated, the regression effect is a statistical artifact which dictates that values of x es- timated for subjects selected from the tails of the dis- tribution of x are exaggerated. Black, Jensen, and Scholes observed in their empirical tests of the CAPM that the 51 of high-risk firms tend to be overestimated while the 81 of low-risk firms are typically underestimated. Thus, while the positive skewness in the present sample of ACFs' 81 is very real, it is likely to be somewhat less pronounced than Table VII makes it appear. 210 Footnotes - Chapter 5 1Four months before the FYE is equivalent to six months before two months after FYE, when the AC was assumed to have been disclosed. The latter is a more precise statement of the B-estimation period actually employed. 2He did observe that around seventy-six percent of his sample firms had December FYES. See p. 4.8 of Gonedes (1974a). s - Fama (1970), pp. 595 599. ‘Donald F. Morrison, Multivariate Statistical Mathods (New York: McGraw-Hill, 1967), pp. 117-124. 5T. W. Anderson, An Introduct'on to Multiv riate Statistical Analysis (New York: Wiley, 1958), pp. 101-108. 6Morrison, pp. 152-155. 7G. E. P. Box, "A General Distribution Theory for a Class of Likelihood Criteria," Biometrika 36 (1949): 317-346. 8The small sample correction factors are: 2 G = ZR + 3R'1 1.- 1 Eq. 12 Al 6 G-l R+l (gilng N) and G whereN=Zn . ) ’ s=l g G ( -1)( 2) ( l - Eq. 15 A2 = R6 G_§+ Eqngz For further details, see Box, p. 334 and p. 338. raw 9Morrison, pp. 152-153. 211 Footnotes - Chapter 6 1Let ACF (i,j) represent an accounting change firm, where i = directional effect of the accounting change on net income and i can take on one of four values: +, -, 0, or ?; j = relative discretion of management in making the change and j can take on one of three values: D for discretionary, N for nondiscretionary, or B for both a discretionary AC and a nondiscretionary AC in the same year. 2In Tables VIII and IX and throughout the re- mainder of the chapter "mean" average return differences are referred to simply as average return differences for purposes of clarity. The double averaging procedures that give rise to the "mean" average return differences result from (1) the cross-sectional averaging across firm returns to arrive at group returns for each month and (2) the averaging across the monthly group returns to obtain the intertemporal average group returns. Between the averag- ing steps (1) and (2) above, the difference between ACF group returns and NCF group returns is taken to obtain the return difference. 3Ball and Brown, p. 176; Beaver (1968), p. 84. 4See, for example, James H. Lorie and Mary T. Hamilton, The Stock Market: Theories and Evidence (Home- wood, Ill.: Irwin, 1973), Chapters 6-9. 5The reason for there being only thirty pairs of firms in this test, whereas there were forty-three pairs in the test of ACF (-,-) - NCF, is due to thg difficulty in matching ACF (+,t) and ACF (-,-) on FYE, Bi, and indus- try. Of course, the principal constraint is the small number affirms that made ACs with negative effects on net income. 6Ball, p. 28. 7Bremser, p. 572. 8Archibald (1972), p. 28. 9See Archibald (1972), p. 28 and Kaplan and R011, pp. 237-239. Archibald does not present the results on a control group of NCFs. Kaplan and Roll's Figure l-Panel A 212 gives the cumulative average API of ITC change firms, and Figure 1-Panel B gives the same data for the ITC control group, both for essentially the same period of time (in relation to FYE) covered in this study. In order for Kaplan and Roll's results to be comparable to the results of this study, the cumulative API of their NCFs must be subtracted from the cumulative API of their ACFs. When this is done, the relative market reaction toward(discre- tionary)ACs to flow-through for the ITC is mildly negative prior to the publication of the annual report and dis- tinctly negative thereafter. The cumulative average re- turn difference (on ACFs-NCFs) is around -.015 in the twenty-fourth week before the preliminary earnings report is announced, as compared to the average on both risk classes of close to .00 during month t = -6 in this study. Their cumulative average return difference is around -.015 in the announcement week, compared to -.01 during month t = 0 here. Twenty-four weeks later, their cumulative average return difference is around -.075, whereas the overall CARD in this study is -.06 six months later. (See Figure 10.) Kaplan and Roll's report of the results of firms that changed to straight-line depreciation was not accompanied by a report On a control group. However, the cumulative average API on depreciation change firms was negative through most of the sixty-week range around Kaplan and Roll's week 0. l°See Kaplan and Roll, pp. 257-259. llSunder (1975), p. 515. 12Ibid. 13The author is conscious of his own caveats with respect to ventures into the area of predicting better-or-worse wealth positions as related to ACs. How- ever, the market appears to perceive ACFS (O,D) as "better" than the matched control group, and there is some reason for this phenomenon. This exercise is an attempt to iden- tify the reason. l4Sunder (1975), p. 515. 15Each of eighteen different hypothesis groups is tested over three time periods for a total of fifty- four different hypothesis tests. 215 16Recall that the market reaction to nondiscre- tionary ACs with positive income effects was significantly positive in the six months after the year-end disclosure month. See line fourteen of Table IX and Figure 14. 17See the Chapter 5 discussion of Bartlett's M- statistic (Eq. 10), and particularly the chi-square trans- formation used to test the equality of covariance matrices with large samples. In all the statistical tests conduc- ted in this study, the samples are monthly observations of security returns. In this case, the sample size is large because the test spans the Sixty months of data over which the Bi of firms were estimated. 18See the Chapter 5 discussion of the transfor- mation of Bartlett's M-statistic (Eq. 10) into an F random variable (Eq. 14). 19Only eight of the twenty-seven F-values exceed one. So for almost all hypothesis tests the confidence level exceeds .75 by a wide margin. 20It appears that the only way to conduct the test so that meaningful conclusions can be reached is to increase the monthly observations of the impact period to a number approaching sixty. This would give the test of 2119:22192 statistical power comparable to that in the tes of 21 = 22. Then the results on the two sets of tests could be interpreted in terms of the information content of 61. Unfortunately, the focus on the thirteen months around the AC disclosure date makes this alteration impossible so long as returns on firms that made ACS in different years are aggregated in arriving at Bit'fli where t ==0 is the AC disclosure month. The alternative is a design similar to that used by Gonedes. He observed that his estimated portfolio risk coefficients were relatively constant across t. Because of this stability he treated the stocks in the portfolio for each year as though they were the same stocks in the portfolios for the other four years of his five-year impact period. This allowed him to aggregate monthly returns per year end-on-end for five years to provide sixty monthly observations. The reason this aggregation method was not used in the present study is the widely varying number of ACFs during the years 1968-1972 (see table III in chapter 4). This wide varia- tion would make the assets used for estimation in one year (for example, 1968, when there were forty-six ACFS) dif- ferent from the assets in another year (for example, 1971, 214 when there were 167 ACFS) in terms of their variabilities. This difference would render the end-on-end aggregation in- valid. 21This average does not include the effect of one equity-method AC, which increased the firm's EPS by 542.9 percent. Inclusion of this extreme case increases the average effect on EPS to +26.l percent for the high- riskpfirms. 215 Footnotes - Appendix B 1Roger C. Ibbotson and Rex Sinquefield, "Stocks, Bonds, Bills, and Inflation: Year-by-Year Historical Re- turns (1926-1974)." To appear in Journal of Business 49 (January 1976). 2The variance of the return on the market port- folio, Var(Rm), was computed on the monthly price changes of the Standard and Poor's 500 average for January 1968 through December 1972. 3See footnote thirteen of Chapter 4 and the text material pertaining thereto for the manner in which the estimated standard errors of 68 were derived. 216 Footnotes - Chapter 7 1A related examination of Wall Street Journal Index entries for a small group of the firms in the sample indicates that relatively few firms disclosed their ACs prior to year-end and that most of the year-end disclo- sures were made in the annual report. 2Bremser, p. 572. 3William H. Beaver, "What Should Be the FASB'S Objectives?" Journal of Accountanqy 136 (August 1973): 56. 4Were it not for the disclosure of ACs (by what- ever means and in whatever form the disclosures take), investors would not even know that the changes had oct curred. The very existence of AC disclosures probably arises from the auditing standard of consistency, and even that can be traced to the same institutional arrangements that created the Committee on Accounting Procedure, the APB, and the FASB. Thus, the inseparability of AC dis- closures and accounting rule-making bodies is undeniable. BIBLIOGRAPHY 218 BIBLIOGRAPHY American Institute of Certified Public Accountants. Ap- counting Principles Board Opinion No. 6. New York: American Institute of Certified Public Accountants, 1965. . Accounting Principles Board Opinion Nos. 13-24. New York: American Institute of Certified Public Accountants, 1969-1971. . Accounting Research Bulletin No. 43, Chapter 2, paragraphs 3-4. New York: American Institute of Certified Public Accountants, 1953. . Accounting Research Bulletin No. 51, Notes section. New York: American Institute of Certi- fied Public Accountants, 1953. . Accounting Trends and Techniques. Editions 23- 27 (1969-1973). New York: American Institute of Certified Public Accountants. Anderson, T. W. An Introduction to Multivariate Statisti- cal Analysis. New York: Wiley, 1958. Archibald, T. Ross. "The Return to Straight-Line Deprecia- tion: An Analysis of a Change in Accounting Method." Empirical Research in Accounping: Se- lected Studieaaal967. Supplement to Journal of Accounting Research 5: 164-180. . "Stock Market Reaction to the Depreciation Switch-Back." The Accounting Review 47 (January 1972): 22-50. Baesel, Jerome B. "On the Assessment of Risk: Some Further Considerations." Journal of Finance 29 (December 1974): 1491-1494. Ball, Raymond J. "Changes in Accounting Techniques and Stock Prices." Empirical Research in Accounting: Selected Studiaa, I972. supplement to Journal of Accounting Research 10: 1-38. , and Brown, P. "An Empirical Evaluation of Ac- counting Income Numbers." Journal of Accounting Rasearch 6 (Autumn 1968): 159-177. 219 . "Portfolio Theory and Accounting." Journal of AccountinglResearch 7 (Autumn 1969): 300-323. Baskin, Elba F. "The Communicative Effectiveness of Con- sistency Exce tions." The Accounting Review 47 (January 1972 : 38-51. Beaver, William H. "The Behavior of Security Prices and Its Implications for Accounting Research (Meth- ods)." Supplement to The Accounting Review 47 (1972): 407-456. . "The Information Content of Annual Earnings Announcements." Empirical Rasaarch in Accounting: Selected Studiea 1968. Supplement to Journal of Accounting Research 6 (1968): . "What Should be the FASB's Objectives?" Jour- nal of Agcountancy 136 (August 1973): 49-56. , and Dukes, R. "Interperiod Tax Allocation and 5-Depreciation Methods: Some Empirical Results." The Accounting Review 48 (July 1975): 549-559. . "Interperiod Tax Allocation, Earnings Expecta- tions, and the Behavior of Security Prices." The Accounting Review 47 (April 1972): 520-552. Black, F. "Capital Market Equilibrium with Restricted Bor- rowing." Journal of Business 45 (July 1972): 444-455. ; Jensen, M.; and Scholes, M. "The Capital Asset Pricing Model: Some Empirical Tests." In Studies in tha Thaory pf Capital Markets, edited by Michael C. Jensen. New York: Praeger, 1972. Blume, Marshall E. "Betas and Their Regression Tendencies." Joarnal of Finance 30 (June 1975): 785-795. . "On the Assessment of Risk." Journal of Fi- nance 26 (March 1971): 1-10. Box, G. E. P. "A General Distribution Theory for a Class of Likelihood Criteria." Biometrika 56 (December 1949): 517-546. 220 Bremser, Wayne G. "The Earnings Characteristics of Firms Reporting Discretionary Accounting Changes." The Accounting Review 50 (July 1975): 565-575. Comiskey, Eugene E. "Market Response to Changes in Depre- ciation Accounting." The Accounting Review 46 (April 1971): 279-285. Conover, w. J. Practical Nonparametric statistics. New York: Wiley, 1971. COpeland, R. "Income Smoothing." Empirical Research in Accounting: Selected Studiea, 1968. Supplement to Journal of Accounting Research 6 (1968): 101- 121. Cox, D. R. Planning of Experiments. New York: John Wiley & Sons, 1958. Cushing, Barry E. "An Empirical Study of Changes in Ac- counting Policy." Journal of Accounting Research 7 (Autumn 1969): 196-205. , and Deakin, E. "Firms Making Accounting Changes: A Comment." Tha Aggounting Baview 49 (January 1974): 104-111. Demski, Joel S. Information Analysis. Reading, Mass.: Addison-Wesley, 1972. Downes, David, and Dyckman, T. "A Critical Look at the Efficient Market Empirical Research Literature as it Relates to Accounting Information." The Accounting Review 48 (April 1975): 500-517. Fama, Eugene F. "Efficient Capital Markets: A Review of Theory and Empirical Work." Journal of Finance 25 (May 1970): 383-417. . "The Behavior of Stock Prices." Journal of Business 28 (January 1965): 54-105. ; Fisher, L.; Jensen, M.; and Roll, R. "The Adjustment of Stock Prices to New Information." Inparnapional Economic Raview 10 (February 1969): 1-21. 221' , and MacBeth, J. "Risk, Return and Equilibrium: Glass, G. Some Empirical Tests." Journal of Politicai Economy 81 (May/June 1975): 607-656. V., and Stanley, J. Statistical Methods in Educa- tion and Psychology. Englewood Cliffs, N.J.: Prentice-Hall, 1970. Gonedes, Nicholas J. "Efficient Capital Markets and Exter- nal Accounting." The ccountin eview 47(January 1972):'11-21. . "Risk, Information, and the Effects of Special Accounting Items on Capital Market Equilibrium." Report No. 7429, Center for Mathematical Studies in Business and Economics, University of Chica- go, 1974a. . "Risk, Information, and the Effects of Special Accounting Items on Capital Market Equilibrium." December 1974b. (Mimeographed.) . "Risk, Information, and the Effects of Special Accounting Items on Capital Market Equilibrium." May 1975. To appear in Journal of Accounting Research. , and DOpuch, N. "Capital Market Equilibrium, Information-Production, and Selecting Accounting Techniques: Theoretical Framework and Review of Empirical Work." Studias on Pinancial Acgounting Ob ective : 1974. Supplement to Journa; o; A - counting Research 12 (1974): 47-129. Gosman, Martin L. "Characteristics of Firms Making Ac- Hagerman, Ibbotson, counting Changes." The Accounting Review 48 (January 1975): 1-11. Robert L., and Richmond, R. "Random Walks, Mar- tingales and the OTC." Journal 0: Einance 28 (September 1973): 897-910. R. C., and Sinquefield, R. "Stocks, Bonds, Bills and Inflation: Year-by-Year Historical Returns (1926-1974)." To appear in Journal ochusinass 49 (January 1976). 222 Jensen, Michael C. "Capital Markets: Theory and Evidence." Bell Journal of Economics and Management Science 3 Autumn 1972 : 357-398. . "The Performance of Mutual Funds in the Period 1945-1964." Journal of Pinance 25 (May 1968): 589-416. Kaplan, R. S., and R011, R. "Investor Evaluation of Ac- counting Information: Some Empirical Evidence." Journal of Business 45 (April 1972): 225-257. King, Benjamin J. "Market and Industry Factors in Stock Price Behavior." Journal of Business 39 (January 1966): 159-190. Klemkosky, Robert C., and Martin, J. "The Adjustment of Beta Forecasts." Journal of Einance 30 (Septem- ber 1975): 1123-1128. . "The Effect of Market Risk on Portfolio Di- versification." Journal of Finance 30 (March 1975): 147-154. ‘ Lev, Baruch. Financial Statement Analysis: A New Approach. Englewood Cliffs, N.J.: Prentice-Hall, 1974. Lorie, James H., and Hamilton, Mary T. The Stock Markat: .Theorieg and Evidence. Homewood, Ill.: Richard D. Irwin, 1973. Magee, R.."Industry-Wide Commonalities in Earnings." aoar- nal of Accounting Research 12 (Autumn 1974): 270-287. Marcis, Richard G., and Smith, K. "Efficient Estimation of Multivariate Financial Relationships." Jparnal of Pinance 29 (December 1974): 1415-1424. Markowitz, H. Portfolio Selection: Efficient Diversifica- tion of Investments. New York: John Wiley & Sons, 1959. Merrill Lynch, Pierce, Fenner & Smith, Inc. Security Risk Evaluation. New York: Merrill Lynch, Pierce, Fenner'& Smith, Inc., December 1969-August 1972. 225 Modigliani, Franco, and Pogue, G. "An Introduction to Risk and Return." Financial Analysts gournal 30 (March/April 1974): 68-80. . "An Introduction to Risk and Return - II." Financial Analysts gournal 50 (May/June 1974): 69-86. Morrison, Donald F. Multivariate Statistical Methods. New York: McGraw Hill, 1967. Myers, Stephen L. "A Re-Examination of Market and Industry Factors in Stock Price Behavior." Journal 0: Finance 28 (June 1975): 695-705. . "The Stationarity Problem in the Use of the Market Model of Security Price Behavior." The Appopnting Review 48 (April 1975): 518-522. Neuman, Fred.."The Auditing Standard of Consistency." Empirical Research in Accounting: Selected Stu- dies, 1968. Supplement to Journal of Accounting Research 6 (1968): 1-17. O'Connor, Melvin C. "On the Usefulness of Financial Ratios to Investors in Common Stock." The Accounting Review 48 (April 1975): 559-552. Rappaport, Louis H. SEC Accounting Practice and Procedure. New York: Ronald Press Company, 1972. Richter, M. K. "Cardinal Utility, Portfolio Selection, and Taxation." Review of Economic Studies 27 (June 1960): 152-166. Sharpe, W. F. "Capital Asset Prices: A Theory of Market Equilibrium under Conditions of Risk." Journal p;_yiaaaaa 19 (September 1964): 425-442. , and OOOper, G. "Risk-Return Classes of New York Stock Exchange Common Stocks,l951-1967." Financial Analysts Journal 28 (March/April 1972): 46-81. Sunder, Shyam. "Relationships between Accounting Changes and Stock Prices: Problems of Measurement and Some Empirical Evidence." Empirical Researcp in Accounting: Selected Studies, 1973. Supplement to Journal of Accounting Research 11 (1973). 224 . "Stock Price and Risk Related to Accounting Changes in Inventory Valuation." The Accounting Review 50 (April 1975): 505-515. Tatsuoka, Maurice M. Multivariate Analysis: Tachnigues for Educapionai and Psychological Research. New York: Wiley, 1971. Tobin, James. "Liquidity Preference as Behavior towards Risk." Review of Economic Studies 25 (February 1958): 65-86. Van Horne, James 0. Financial Management and Policy, 2nd 1974 0 ed. Englewood Cliffs, N.J.: Prentice-Hall, Wilks, S. S. "Sample Criteria for Testing Equality of Means, Equality of Variances and Equality of Covariances in a Normal Multivariate POpula- tion." Annals of Mathematical Statistics 17 (1946): 257-281. 74m. 4.: Wm m, 2707 mm PARK 0 AUSTIN. TEXAS 70705 0 AC 512 472-3210 "11111111711711111)ITS