The Value Relevance of Earnings and the Prediction of Future Cash Flows

Transcription

1 The Value Relevance of Earnings and the Prediction of Future Cash Flows Oliver Kim University of Maryland Steve C. Lim Texas Christian University Taewoo Park University of Maryland December 2005 Very Preliminary Abstract In this paper we test the validity of the prediction test of future cash flows as a substitute for the test of value relevance of earnings. We theoretically show that the R 2 of the cash flows prediction regression is contaminated by (1) the relative presence of noise in future cash flows (2) the relative presence of noise in the covariance between future cash flows and current earnings. We test if either of the above two factors contribute to the result of Kim and Kross (2005) that the ability of earnings to predict one-year-ahead cash flows has increased over the recent decades, in contrast to the evidence of decreasing value relevance of earnings presented by others. We find empirical evidence that both factors contributed to their result. From the significant presence of noise in cash flows (97%) and also in the covariance(87%),we conclude that the cash flows prediction test is a poor substitute for the test of value relevance of earnings, and perhapsofotheraccountingnumbersaswell.

2 1 Introduction The purpose of this paper is to examine the implication of the prediction test of immediate future cash flows by current earnings on the value-relevance of earnings. We attempt to assess two seemingly conflicting views of cash flows manifested in the existing literature. On the one hand, we have long adopted accrual accounting over cash accounting, and archival evidence clearly shows that stock returns are more closely associated with earnings than cash flows. 1 On the other hand, the usefulness of an accounting number is often gauged by its ability to predict future cash flows. 2 Despite the appearance of a possible conflict between the two views, return association tests and cash flows prediction tests have been used as the two most prominent tests of the usefulness of an accounting number. A recent paper by Kim and Kross (2005), however, brings the conflict to the surface. They show that the ability of current earnings to predict one-year-ahead cash flows has significantly increased over recent decades. Combined with the evidence of deteriorating association between earnings and stock returns over the same recent decades provided by Francis and Schipper (1999) and others, the two sets of results beg for an explanation why value relevance and cash flows prediction diverge for earnings. Kim and Kross (2005) describe it well: If stock price is the present value of future cash flows, the deterioration in the association between accounting earnings and stock prices implies a growing inability of accounting numbers to forecast future cash flows, but that is not what we find. The theoretical foundation for value-relevance tests is firmly established in the 1 For example, Dechow (1994) shows that earnings explain 16.2% of the variation in annual returns while cash flows from operations explain only 3.2%. 2 This literature includes Finger (1994), Dechow, Kothari and Watts (1998), (and) Barth, Cram and Nelson (2001), and Kelly, Shores, and Tong (2003). 1

3 literature. For example, Kim and Verrecchia (1991) show in a short-window setting that if a significant fraction of investors use certain information in their investment decisions, the information will be impounded into the equilibrium price. Therefore, an association of certain information with stock price or its change can be considered as evidence that the information is used by investors. In a long-window setting, an association with return indicates that either the information was used by the market or at least the information reflects the beliefs of the market participants. In contrast, the theoretical foundation for cash flows prediction tests is rather weak. Investors as well as creditors are clearly concerned about a firm s future cash flows,andthisisreflected in the concept statement of the Financial Accounting Standard Board that a primary objective of financial reporting is to provide information to help investors, creditors, and others assess the amount and timing of prospective cash flows (FASB 1978, 37-39). The problem is, however, that prospective cash flows are elusive and difficult to pinpoint, because the term literally means all prospective flows of cash. A researcher who wants to find a number that represents prospective cash flows other than from the market (i.e., from the security price and its change) encounters many problems. First, he has to choose a subset of cash flows among the long cash flows series. Second, if cash flows of multiple years are chosen, he must determine what weights to assign to cash flows of different years when combining them. Third, observed cash flows of a particular period may contain period noise that cancel one another across multiple periods and is thus ignored or discounted by the users of accounting information. The above three problems are closely interrelated and must be dealt with simultaneously. For simplicity, most existing studies of cash flows prediction concentrate on a small number of immediate future years cash flows, and a majority including Kim and 2

4 Kross (2005) on one-year-ahead cash flows. This practice ignores the first problem and bypasses the second problem above, and has been accepted as a practical approach in the literature. 3 This study also investigates the prediction of one-year-ahead cash flows by current earnings and, as a result, concentrate on the third problem above. That is, one-year-ahead cash flows may be a very noisy proxy for all prospective cash flows because they contain significant value-irrelevant noise. The purpose of this paper is to resolve the conflict between result of decreasing value-relevance of earnings and the result of increasing ability of earnings to predict future cash flows focusing on noise in cash flows(andinearnings). Wefirst develop a theoretical model of stock returns, earnings, and cash flows in which earnings and cash flows each consists of two additive components, value-relevant component and valueirrelevant noise. As a result, all the variances and (contemporaneous and lagged) covariances among returns, earnings, and cash flows also consist of the value-driven portion and the noise-driven portion. We then define the value-relevance of earnings and cash flows as the percentages of the value-driven portions in the variances. Similarly, the value-relevance of predicting one-year-ahead cash flows by current earnings is defined as the percentage of the value-driven portion in their covariance. We then express the R 2 of the return-earnings regression and the R 2 of the cash flows prediction regression as the value relevance of earnings contaminated by noise. While the former is depressed by the presence of market noise, the latter is depressed by the presence of noise in cash flows and also boosted by the presence of noise in the covariance between one-year-ahead cash flows and current earnings. We test hypotheses that the two factors contribute to the result of improving cash prediction by earnings. 3 A notable related study is Dechow and Dicheve (2002) who consider past, current, and future cash flows, in their definition of earnings quality. 3

5 Our empirical results show that both of the above factors contributed to the observation of Kim and Kross (2005). Based on our results, we conclude that the cash flows prediction test is a poor substitute for the test of value relevance of an accounting variable. The paper is organized as follows. Section 2 present the model and section 3 analyzes the differences between the R 2 s of the return-earnings regression and the cash flows prediction regression and develops the two hypotheses. In section 4 we develop value relevance measures and present empirical results in section 5. Section 6concludes. 2 A Model of Return, Earnings, and Cash Flows In order to analyze the differences between the value-relevance test and the cash flows prediction test, we develop a model of return, earnings, and cash flows in which all three variables are noisy. We first use an equation that expresses stock returns consiting of two components. The first component reflects thechangeinthemarket s assessment of the value of the firm (i.e., in the market s expectation of all future payoffs bythefirm) and the second component is unrelated to it. We assume that the two components are additive and independent of each other. That is: R t = X t + δ t (1) for all t, wherer t is the year t stock return, X t is the value-related return, and δ t is value-irrelevant return which we call market noise. Given equation (1) and the fact that returns are serially independent, we assume that X t is serially independent and normally distributed with variance v. 4

6 We specify earnings and cash flows in relation to equation (1) as: E t = α 1 X t + α 2 X t 1 + ε t (2) and C t = β 1 X t + β 2 X t 1 + γ t, (3) where ε t and γ t are normally distributed independently of X t s and δ t s. In equations (2) and (3) earnings and cash flows are similarly characterized with different parameters and are each decomposed into three components. The first components, α 1 X t and β 1 X t, are priced in the same year and the second components, α 2 X t 1 and β 2 X t 1, are priced in earlier years. 4 The third components, ε t and γ t,arethosethat are not priced in any year. The three components are mutually independent. The second component is what is priced earlier than this year, and is due to the fact that earnings and cash flows are not completely timely (Collins et al. 1994). This untimeliness of an accounting measure is inevitable in order to guarantee a certain degree of objectivity to the measure. For earnings, revenues are recognized when earned and realizable and expenses are recognized when the matching revenues are recognized. For cash flows, thecriterionisthereceiptandpayment ofcash. Onthe contrary, the stock price responds to the expectations of future earnings and cash flows. For example, consider a company that developed a promising new product during a year. Stock price goes up reflecting the market s expectation of increased future earnings and cash flows due to the new product. However, earnings does not increase until the firm manufactures and sells the product. Earnings may even decrease in the current year due to the development costs of the new product that 4 Earnings and cash flows information is impounded into price more than one year early. However, we only consider year t 1 to make things simple here. Combining years t 2, t 3, andt 4 generate qualitatively similar theoretical and empirical results. 5

7 are expensed. Also, cash flows generally reflects the sales even later than earnings when cash is collected for the sales. While (the lack of) timeliness concerns about how swiftly an accounting number reflects information being incorporated in stock price, the other source of the imperfection of earnings and cash flowsthatisthefocusofthispaperisthecomponentsthat are never impounded into stock price, i.e., ε t and γ t. In this paper we consider these portions as the components of earnings and cash flowsthatarenotusedbyinvestors. In this sense we will call ε t and γ t earnings noise and cash flows noise, respectively. This period noise reflects the period fluctuations of earnings and cash flows that are not priced because the fluctuations of different periods cancel each other out. Under the current accounting system, this period noise reverses over time. In other words, if earnings (or cash flows) of many consecutive periods are added up, the noise significantly diminishes. While the reversal of the period noise automatically gives a degree of negative autocorrelation, it is also possible that the direction of intended (e.g., income management) or unintended (e.g., ones due to applying a certain accounting rules such as a decreasing-charge depreciation method) period noise may persist over multiple years, giving a degree of positive autocorrelation. It is also reasonable to assume that certain noise affects both earnings and cash flows, either in the same year or with a lag. The magnitudes of the noise in earnings and cash flows, ε t and γ t,, i.e., Var(ε t ) and Var(γ t ), and their auto- and cross-covariances, Cov(ε t,ε t 1 ), Cov(γ t,γ t 1 ), Cov(ε t,γ t ), Cov(ε t,γ t 1 ),andcov(γ t,ε t 1 ), seem largely an empirical issue and we do not make any assumption about their magnitudes at this point. 6

8 3 Value Relevance and Cash Flows Prediction 3.1 The Theoretical Differences In this section we analyze the differences between the value-relevance test and the cash flows prediction test using the model of section 2. The (theoretical value of the) R 2 of the contemporaneous return-earnings regression can be written as: R 2 (R t,e t ) = [Cov(R t,e t )] 2 Var(R t ) Var(E t ) = (α 1 v) 2 [v + Var(δ t )] [(α α 2 2)v + Var(ε t )] = v v + Var(δ t ) α 2 1v (α α 2 2)v + Var(ε t ) = VRR VRE cur. (4) In the third expression of equation (4) the R 2 is expressed as the product of two v terms. The first term, v+var(δ t, is the percentage of the value-related variance in ) thevarianceofreturnandisdirectlyaffected by the relative presence of market α 2 1 v noise. This second term, (α 2 1 +α2 2 )v+var(εt), is the percentage of the variance of the component of earnings directly related to current return, i.e., the variance of α 1 X t,in the total variance of earnings. We will call this ratio the value-relevance of earnings with respect to current return, denoted by VRE cur. The two terms above are not separately observable, and the R 2 can be considered as a noisy measure of the valuerelevance of earnings depressed by v v+var(δ t ).5 We now write the R 2 of the cash flows prediction regression and relate it to VRE cur 5 The regression coefficient, i.e., the earnings response coefficient, can be expressed as α 1 v (α 2 1 +α2 2 )v+var(ε t), which is different from the second term because α 1 is not squared. 7

9 as follows: R 2 (C t+1,e t ) = = = [Cov(C t+1,e t )] 2 Var(C t+1 ) Var(E t ) α1 β 2 v + Cov(γ t+1,ε t ) 2 β β2 2 v + Var(γt+1 ) [(α α 2 2)v + Var(ε t )] [α 1 β 2 v] 2 [(α α 2 2)v + Var(ε t )] β β 2 2 v + Var(γt+1 ) α1 β 2 v + Cov(γ t+1,ε t ) α 1 β 2 v α 2 = 1v (α α 2 2)v + Var(ε t ) β 2 2v β β2 2 v + Var(γt+1 ) 2 α1 β 2 v + Cov(γ t+1,ε t ) α 1 β 2 v 1 VRE cur VRC lag (%VPCE) 2. (5) Equation (5) expresses the R 2 of the cash flows prediction regression as the valuerelevance of earnings with respect to current return, VRE cur,multipliedbytwofactors. The first factor is the value-relevance of cash flows with respect to lagged return, i.e., VRC lag = β 2 2 v (β 2 1 +β2 2)v+Var(γ t+1 ) that depresses the measure. The second factor is the squared inverse of the percentage of value-driven portion in Cov(C t+1,e t ),denoted α by, i.e., %VPCE, that boosts the measure because %VPCE 1 β 2 v < 1. α 1 β 2 v+cov(γ t+1,ε t ) To summarize the implications of equations (4) and (5), while the R 2 of the contemporaneous return-earnings regression measures VRE cur with market noise, the R 2 of the cash flows prediction regression measures VRE cur with noise in one-yearahead cash flows and also with noise in the covariance between one-year-ahead cash flows and current earnings. 2 8

10 3.2 The Observed Trends and Hypotheses Development The analysis of last subsection enables us to track the sources of the discrepancy between the decreasing value relevance of earnings and the increasing ability of earnings to predict one-year-ahead cash flows. This can occur if either or both of the following have occurred. We divide the two factors into the following two hypotheses: Hypothesis 1: The value-relevance of cash flows with respect to lagged return, i.e., VRC lag, has increased over the recent decades. Hypothesis 2: The percentage of the value-driven portion in the covariance between current earnings and one-year-ahead cash flows, i.e., %VPCE, has decreased over the recent decades. The above two hypotheses, if they can be tested, would generate valuable insights on the usefulness of the cash flows prediction tests. However, variances and covariances are always observed as the sums of value-driven and noise-driven portions, and there is no easy way to separate them. The next section is devoted to developing measures of VRE cur, VRC lag,and%vpce in order to test hypotheses 1 and 2. 4 Measures of Value Relevance 4.1 Measuring the Percentage of the Value-Driven Variance and Covariance The value-relevance of a periodic performance measure such as earnings and cash flows is understood in this paper as the degree to which the measure reflects the firm s value. We first write their variances as: Var(E t )= α α 2 2 v + Var(εt ), Var(C t )= β β 2 2 v + Var(γt ). 9

11 The value-relevance of earnings, denoted by VRE, and the value-relevance of cash flows, denoted by VRC, are each defined as: VRE (α α 2 2) v (α α 2 2) v + Var(ε t ), VRC β β 2 2 β β2 2 v v + Var(γt ). In the above definition, the value-driven portion comes not only from the timely contemporaneous association, but also from the delayed lagged association between value and the performance measure. The definition is also similar to the concept of signal-to-noise ratio, which is defined as the value-driven variance divided by the noise-driven variance. 6 We also define the value-relevance of earnings and cash flows with respect to either the current return or lagged return. For example, the value-relevance of earnings with respect to current return, denoted by VRE cur,isdefined by: VRE cur α 2 1v (α α 2 2) v + Var(ε t ), and the value-relevance of cash flows with respect to one-year lagged return, denoted by VRC lag,isdefined by: VRC lag β 2 2v β β 2 2 v + Var(γt ). Though the above definitions of value-relevance are natural, there is usually a problem with measuring them because a firm s value or its change (X t )israrely observed separate from market noise (δ t ). The problem is illustrated below in the attempt to compute the value-relevance of earnings and cash flows. Given a sample, we first use the fact that the covariances between return and current or one-year-ahead or, wquiv- 6 There is a one-to-one relation between the two, which can be written as SNR = VR alently, as VR= SNR 1+SNR. 1 VR 10

12 earnings or cash flows can be measured and take the following simple forms: Cov( P t,e t ) = α 1 v, Cov( P t,e t+1 ) = α 2 v, Cov( P t,c t ) = β 1 v, Cov( P t,c t+1 ) = β 2 v. (6) Using equation (6), parameters α 2, β 1,andβ 2 can be converted to a multiple of α 1 : α 2 = Cov( P t,e t+1 ) Cov( P t,e t ) α 1, β 1 = Cov( P t,c t ) Cov( P t,e t ) α 1, β 2 = Cov( P t,c t+1 ) Cov( P t,e t ) α 1. (7) Equation (7) allows us to write the variances of earnings and cash flowsasfollows: " µ # 2 Cov( Pt,E t+1 ) Var(E t )= 1+ α 2 Cov( P t,e t ) 1v + Var(ε t ), (8) Var(C t )= " µcov( Pt 2,C t ) + Cov( P t,e t ) µ # 2 Cov( Pt,C t+1 ) α 2 Cov( P t,e t ) 1v + Var(γ t ). (9) We can also express the covariance between one-year-ahead cash flows and current earnings as follows using equation (7): Cov(C t+1,e t ) = α 1 β 2 v + Cov(γ t+1,ε t ) = Cov( P t,c t+1 ) Cov( P t,e t ) α2 1v + Cov(γ t+1,ε t ), (10) The percentage of value-driven portion in the above covariance, denoted by %VPCE, can be written as: %VPCE α 1 β 2 v α 1 β 2 v + Cov(γ t+1,ε t ) = Cov( P t,c t+1 ) Cov( P t,e t ) α 2 1v Cov( P t,c t+1 ) Cov( P t,e t α 2 ) 1v + Cov(γ t+1,ε t ). (11) We can similarly define %VPEE, %VPEC, and%vpcc for other covariances in the predictions of earnings by earnings, earnings by cash flows, and cash flows by cash flows, respectively. Equations (8), (9), and (10) indicate that for any given number of equations for variances and covariances of earnings and cash flows and their observed values, we 11

13 have one more unknowns, namely, α 2 1v, in addition to the variances and covariances of noise in earnings and cash flows. Our approach here is not to impose any strong assumptions about the relations between earnings and cash flows noise and increase the number of equations or reduce the number of unknowns. Instead, we attempt to investigate their upper bounds. 4.2 Procedure of Computing Value Relevance Measures We seek to obtain a sufficiently tight upper bound for α 2 1v thatwouldinturngenerate sufficiently tight upper bounds for VRE cur, VRC lag,and%vpce in equations (8), (9), and (10). Then, they can be used as reasonable proxies for VRE cur, VRC lag, and %VPCE themselves, respectively. An upper bound for α 2 1v is chosen for each year in the following way. v First, we assume that the value-relevance of return, VRR = v+var(δ t, stay the ) same for different years, since the extent of market noise is not observable. Under this assumption, the R 2 of the regression of current earnings on current and lagged returns is proportional to the value-relevance of earnings, VRE. Second, we seek the maximum R 2 of the above regression among all years and assign VRE =100%to the year and assign VRE of other years proportionally to the R 2 of the above regression. By doing this, we ensure that VRE of any given year does not exceed 100%. This value of VRE foranygivenyearcanbeunderstoodas the upper bound. Third, once VRE is computed for each year, α 2 1v can be solved from: ³ 2 1+ Cov( Pt,E t+1 ) Cov( P t,e t) α 2 1v VRE, (12) Var(E t ) whichisobtainedfromequation(8). 12

14 Fourth, compute VRE cur, VRC lag,and%vpce by using equation (8), (9), and (10). 5 Data and Empirical Results 5.1 Data Our sample includes all non-financial firms (excluding firms with SIC 6000s), of which accounting and return data are available from the CRSP and Compustat from 1970 through We follow the data screening procedure of Kim and Kross (2005). Sample firmsshouldhaveinformationonstock returnsofthetestperiod,andac- counting income, cash flow and asset amount up to next two years. After eliminating top and bottom 1% of distribution of earnings, cash flows and returns, we analyze total 98,149 observations over 33 years. Following Kim and Kross (2005), and Dechow, Kothari and Watts (1998), we define cash flows such that CFO = Operating Income before Depreciation Interest Expense (13) + Interest Revenue Taxes WC, where WC is the change in accounts receivable, inventory, other current assets from year t 1 to year t, minus change in accounts payable, taxes payable, other current liabilities and deferred taxes from year t 1 to year t. AndE(Earnings) is CFO + WC Depreciation. All variables are deflated by average assets. Stock returns of period t is the compounded monthly return from the 4th month of year t through the end of the third month after the fiscal year end. 13

15 5.2 Empirical Results Table 4 gives the values of α 2 1v, VRE cur, VRC lag,and%vpce for all years from 1970 to First, the results support both hypotheses 1 and 2. In other words, theincreaseinther 2 of the cash flows prediction by current earnings has increased due both to an increase in VRC lag (VRE cur ) and to a decrease in %VPCE. Once we have seen the evidence that the decreasing value relevance of earnings and improving prediction of one-year-ahead cash flows by earnings are caused by the two factors, VRE cur and %VPCE, a more interesting issue is the implications of these results on the validity of the cash flows prediction test as a test of the usefulness of an accounting number. Our answer is No based on Table 4. Note from equation (5) that: R 2 1 (C t+1,e t )=VRE cur VRC lag (%VPCE) 2. TheaveragevalueofVRC lag for 33 years is only 3.01% which means that on average 96.9% of the variance in one-year-ahead cash flows is due to value-irrelevant noise. Given the mean VRE cur of 15.35%, an immediately question that arises is how can one use earnings to predict future cash flows, a less value-relevant periodic performance measure, and expect to generate any information about the value-relevance of earnings? The mean value of %VPCEis 12.88%. This means that 87.12% of the covariance between one-year-ahead cash flows and current earnings is driven by noise, which further muddles the interpretation of the cash flows prediction test results. It is interesting to see that the value-relevance earnings is in general decreasing, while the value-relevance of cash flows is increasing. The value-relevance gap between the two measures is thus narrowing, but the gap still remains wide. 14

16 Another notable result is that the percentage of the value-driven portion is much greater (on average 23.31% and 27.05% for earnings predictions versus for cash flows predictions). This implies that earnings prediction results are more closely related to value than cash flows prediction results. 6 Conclusion In this paper we examine the sources of the discrepancy between decreasing value relevance of earnings and increasing ability of earnings to predict one-year-ahead cash flows. Our results cast doubts on the validity of using cash flows prediction tests as tests of the value-relevance or usefulness of an accounting number, method, or practice. Simply resorting to the FASB concept statement does not seem adequate because future cash flows in the statement means an appropriately discounted sum of all future cash flows without errors, and not cash flows of a particular period or periods which include only a fraction of future cash flows with significant noise. The noise in cash flows (and any periodic performance measure such as earnings) arises from fluctuations of cash flows that even out over multiple periods and thus are not priced. For example, a bird in hand does not count if a bird this year implies one less bird next year. Also, why do we regress to cash flows after having evolved from cash flows to earnings long time ago? Our analysis is limited in many ways. For example, we concentrate on one-yearahead cash flows, and do not provide guidance as to how to choose and combine multiple periods cash flows. One caveat is that our criticism of the cash flows prediction literature should in no way be construed as a statement that cash flow information is not useful to investors and creditors. Cash flows are in general a good source of information that 15

17 is complementary to information that can be extracted from earnings (e.g., DeFond and Hung 2003). Moreover, much of the noise in cash flows may be removed by observing the components of cash flows or other information such as footnotes and newspaper articles. We are just cautioning against the use of cash flows tests as tests of usefulness without qualifications. 16

18 References [1] Barth, M.E., W.H. Beaver, J.R.M. Hand, and W.R. Landsman. "Accruals, Cash Flows and Equity Values." Review of Accounting Studies 4 (1999): [2] Barth, M.E., D.P. Cram, and K.K. Nelson. Accruals and the Prediction of Future Cash Flows. The Accounting Review 76 (2001): [3] Collins, D., S.P. Kothari, J. Shanke, and R. Sloan. "Lack of timeliness and noise as explanations for the low contemporaneous return-earnings association." Journal of Accounting and Economics 19 (1994): [4] Dechow, P.M. and I.D. Dichev. "The Quality of Accruals and Earnings: The Role of Accrual Estimation Errors" The Accounting Review 77 (Supplement, 2002): [5] Dechow, P.M, S.P. Kothari, and R.L. Watts. "The Relation Between Earnings and Cash Flows." Journal of Accounting and Economics 25 (1998): [6] DeFond, Mark and Mingyi Hung. "An empirical analysis of analysts cash flow forecasts." Journal of Accounting and Economics 35 (2003): [7] Financial Accounting Standard Board. Statement of Financial Accounting Concepts No. 1: Objectives of Financial Reporting by Business Enterprises. Stamford, Conn.: FASB, [8] Financial Accounting Standard Board. Statement of Financial Accounting Concepts No. 5: Recognition and Measurementin Financial Statements of Business Enterprises. Stamford, Conn.: FASB, [9] Finger, C.A. "The ability of earnings to predict future earnings and cash flow." Journal of Accounting Research 32 (1994): [10] Francis, J. and K. Schipper. "Have Financial Statements Lost Their Relevance?" Journal of Accounting Research 38 (1999): [11] Kelly S., D. Shores, and Y. Tong. "Independence in Appearance, Earnings Conservatism, and Prediction of Future Cash Flows." Working Paper (2003). [12] Kim, M. and W. Kross. "The Ability of Earnings to Predict Future Operating Cash Flows Has Been Increasing-Not Decreasing." Journal of Accounting Research 43 (2005): [13] Kim, O. and R. Verrecchia. (1991) [14] Lev, B. and P. Zarowin. "The boundaries of financial reporting and how to extend them." Journal of Accounting Research 37 (1999): [15] Liu, J., D. Nissim, and J. Thomas. Equity valuation using multiples. Journal of Accounting Research 40 (2002):

19 [16] Ryan, S.G. and P.A. Zarowin. "Why Has the Contemporaneous Linear Returns- Earnings Relation Declined?" The Accounting Review 78 (2003): [17] Zach, Tzachi. "Predicting future cash flows: Current earnings or current cash flows?" Working Paper (Washington University)

20 Table 1 Descriptive Statistics Variables Mean Std. Dev. Median C E R The sample includes 98,149 firm observations during 1970 through C is the cash flow from operations such that C = operating income before depreciation - interest expense + interest revenue taxes - WC, where WC = changes in accounts receivable, inventory, other current assets from year t-1 to year t, minus changes in accounts payable, taxes payable, other current liabilities and deferred taxes from year t-1 to year t. E is earnings such that E = C + WC - Depreciation. All variables are at percentage, deflated by average assets. R is the compounded monthly return at percentage from the fourth month of the fiscal year through the end of the third month after the fiscal year end. 19

21 Table 2 Trends of the R 2 from return and cash flows prediction regressions (1) Earnings - Return Model : Et = α0 + α1rt + εt (2) Earnings - Return Model : Et = α0 + α1rt + α2rt 1+ εt (3) Cash Flow - Return Model : Ct = β0 + β1rt + β2rt 1+ γt (4) Cash Flow Forecasting Model : C t 1 δ 0 δ1e + t ϕt Model 1 Model 2 Model 3 Model Average R % point change per year t-value

22 The table reports percentage R 2 s from the return regression models and the cash forecasting model. Each yearly regression includes observations of which the fiscal year ends during the specific calendar year. C is the cash flow from operations such that C = operating income before depreciation - interest expense + interest revenue taxes - WC, where WC = changes in accounts receivable, inventory, other current assets from year t-1 to year t, minus changes in accounts payable, taxes payable, other current liabilities and deferred taxes from year t-1 to year t. E is earnings such that E = C + WC - Depreciation. All variables are at percentage, deflated by average assets. R is the compounded monthly return at percentage from the fourth month of the fiscal year through the end of the third month after the fiscal year end. The last two rows report the slope coefficients (% point 2 change per year) and t-values from the trend regression of the model R t = θ0 + θ1yeart + σ. In the t trend regression, the year of 1999 is excluded because the stock return is unusually negatively correlated with earnings or cash flows of the year. 21

23 Table 3 Variance, Covariance and R square over time Panel A : Earnings and Cash Flows Later/Earlier Ratio Var(E t ) Var(C t ) Cov(E t+1, E t ) Cov(C t+1, C t ) Cov(E t, C t ) Cov(E t+1, C t ) Cov(C t+1, E t ) Panel B : Stock Returns Later/Earlier Ratio Var(R t ) Cov(R t, E t ) Cov(R t, C t ) Cov(R t, E t+1 ) Cov(R t, C t+1 ) Panel C : Regression R squares Later/Earlier Ratio R 2 (E t, R t ) R 2 (C t+1, R t ) R 2 (E t, R t and R t-1 ) R 2 (C t+1, E t ) The table reports the averages of annual variances(var), covariances(cov) and regression R squares(r 2 ) over the earlier 16 years ( ) and the later 16 years ( , except for 1999). The year of 1999 is excluded because the stock return is unusually negatively correlated with earnings or cash flows of the year. C is the cash flow from operations such that C = operating income before 22

24 depreciation - interest expense + interest revenue taxes - WC, where WC = changes in accounts receivable, inventory, other current assets from year t-1 to year t, minus changes in accounts payable, taxes payable, other current liabilities and deferred taxes from year t-1 to year t. E is earnings such that E = C + WC - Depreciation. All variables are at percentage, deflated by average assets. R is the compounded monthly return at percentage from the fourth month of the fiscal year through the end of the third month after the fiscal year end. The last column reports the ratio of later period average over that of earlier period. 23

25 Table 4 Trends of Key Value Relevance Measures Percent Value-driven Year α 1 2 ν VRE cur VRC lag PEE PEC PCE PCC Average % changes per year t-value

26 The table reports trends of key value relevance measures at percentage αν 1 = ( R Rmax) Var( Et) (1 + Cov( Rt, Et+ 1) Cov( Rt, Et) ), 2 VREcur = αν 1 Var( Et ), VRC = ( Cov( R, C ) Cov( R, E ) ) αν Var( C ), lag t t+ 1 t t 1 t+ 1 2 t t+ 1 t t αν 1 t t+ 1 PEE = ( Cov( R, E ) Cov( R, E )) Cov( E, E ), 2 PEC = ( Cov( Rt, Ct) Cov( Rt, Et)) ( Cov( Rt, Et+ 1) Cov( Rt, Et)) αν 1 Cov( Et+ 1, Ct), 2 PCE = ( Cov( Rt, Et) Cov( Rt, Ct)) ( Cov( Rt, Ct+ 1) Cov( Rt, Ct)) αν 1 Cov( Ct+ 1, Et), 2 PCC = ( Cov( Rt, Ct+ 1) Cov( Rt, Ct)) αν 1 Cov( Ct, Ct+ 1), where R 2 is the R squares from the yearly regression of Et = α0 + α1rt + α2rt 1+ εt, and R 2 max is the maximum R 2 from the above yearly regression over 33 years. Var( ) and Cov( ) represent the variance or covariance. C is the cash flow from operations such that C = operating income before depreciation - interest expense + interest revenue taxes - WC, where WC = changes in accounts receivable, inventory, other current assets from year t-1 to year t, minus changes in accounts payable, taxes payable, other current liabilities and deferred taxes from year t-1 to year t. E is earnings such that E = C + WC - Depreciation. All variables are at percentage, deflated by average assets. R is the compounded monthly return at percentage from the fourth month of the fiscal year through the end of the third month after the fiscal year end. The last two rows report the slope coefficients (% changes per year) and t-values from the trend regression of the model Measurest = θ0 + θ1yeart + σ. In the t trend regression, the year of 1999 is excluded because the stock return is unusually negatively correlated with earnings or cash flows of the year. 25