Is the Whole Greater than
the Sum of Its Parts?
Synergies in Accounting Measures
Gary C. Biddle,^{a} Robert M. Bowen,^{b }and James S. Wallace^{c}
^{a}School
of Business and Management, Hong Kong University of Science & Technology,
Clear Water Bay, Kowloon, Hong Kong
^{b}Business
School, University of Washington, Seattle, WA 981953200
^{c }Peter F. Drucker and Masatoshi Ito Graduate School of Management, Claremont Graduate University, Claremont, CA 91711
Current draft: February 2005
Please do not quote. Comments and suggestions are welcome.
We appreciate comments of Jane Kennedy Jollineau, Clive Lennox, Terry Shevlin, and seminar participants at the American Accounting Association Annual Meetings, Colorado State University, and Hong Kong University of Science and Technology Summer Symposium on Accounting Research. We also gratefully acknowledge financial support received for this research. Gary Biddle received support from the Research Grants Council of Hong Kong. Robert Bowen received support from the Accounting Development Fund and the Herbert O. Whitten Professorship at the University of Washington. James Wallace received support from the University of California, Irvine Committee on Research Award. We also gratefully acknowledge the expert research assistance of Ms. Fenny Cheng. EVA is a registered trade name of Stern Stewart & Co., New York.
Is the Whole Greater than
the Sum of Its Parts?
Synergies in Accounting Measures
1. Introduction
The
concept of synergy conveys a whole that is greater, or less, than the sum of
its parts arising from interactions among its parts, rather than from their
individual contributions.[1] In this study we examine synergies in value
relevance among components of summary accounting performance measures. That such synergies may exist is intuitive,
given that the components of accounting measures like earnings are designed
specifically to convey valuation relevant inferences when considered jointly.[2] Further, executives have incentives to manage
accounting components to convey future prospects, by say modifying one
component to offset a perceived overage or shortfall in another. In these ways, accounting components may work
together, like pieces in a puzzle, to reveal more than the sum of their
parts. However, the existence and nature
of such synergies are open questions.
We propose two definitions of synergy. One definition captures joint interactions among components of a summary accounting measure when considered collectively to reveal whether they work together, or against, each other in creating value relevance. Specifically, positive (negative) joint synergy arises when components collectively explain more (less) of the variation in a dependent variable (such as stock returns) than when they are considered separately and then summed. The second definition of synergy reveals how a given component’s interactions contribute to, or subtract from, value relevance when it is considered individually. Specifically, positive (negative) individual synergy arises when the interactions of a given component with other components enhance (diminish) variation explained. Corresponding statistical tests are developed and three settings are examined for possible joint synergy. Our synergy definitions and tests apply equally well in any setting that models a dependent variable as a linear additive combination of unrestricted predictors.[3] We are unaware of any prior tests for joint and individual synergy.
In this study we illustrate positive, negative and insignificant joint synergy. Our first illustration reveals “bottom line” income statement components (i.e., revenues and expenses) to exhibit insignificant joint synergy in value relevance, implying a whole indistinguishable from the sum of its parts.[4] Our second illustration finds noncash earnings accrual components (i.e., those reconciling earnings to operating cash flows) to exhibit positive joint synergy, implying a whole greater than the sum of its parts. Our third illustration examines the components of economic value added (EVA^{®}). Notwithstanding arguments by proponents that EVA corrects “distortions” in GAAPbased earnings, EVA components are found to exhibit negative joint synergy, implying a whole less than the sum of its parts. Thus, EVA components interact collectively to diminish value relevance. We similarly examine individual synergies for income, cash flows, and EVA^{®} components, thus revealing their separate interactions with other components.
Finally, we show that our definitions of synergy reconcile and extend prior concepts of incremental and relative value relevance, providing a more complete view. Whereas relative value relevance assesses “who wins,” and incremental value relevance assesses “who adds,” synergy reveals “who works together or against each other,” which in turn explains “who wins” and “who adds.”[5] Much like the three sides of a triangle, any two explain the third, and each is of interest in itself. For example, tax expense exhibits the largest relative value relevance among income statement line items, followed by sales revenues and special items. However, sales revenue exhibits the largest incremental value relevance, followed by costofgoodssold expense and selling, general and administrative expense. These findings are reconciled by positive individual synergies for sales revenue, costofgoodssold, and selling, general, and administrative expense, and negative individual synergy for tax expense. In other words, sales revenue, costofgoodssold, and selling, general and administrative expense interact with other components to enhance value relevance, whereas tax expense works against them, with synergies explaining the observed patterns of relative and incremental value relevance.[6]
Our research provides several contributions. First, the concepts of joint and individual synergy offer a more comprehensive characterization of how accounting components relate and combine that differs from, and extends, existing descriptions of incremental and relative value relevance. Second, our test results for earnings and EVA components illustrate the three possible outcomes of joint synergy and their implications. Third, our findings provide new evidence on how individual earnings and EVA components add to, and subtract from, value relevance. Ultimately, synergy is of interest as a characteristic of accounting measures as it reveals whether they work together or against each other. Synergy also constitutes a source of value relevance that can be influenced by measure design, institutional features, and the exercise of management discretion. For example, if an analyst or policy maker wants to find or create incremental value relevance, there are two ways to do so: increase separate value relevance holding synergy constant, or increase positive synergy holding separate value relevance constant. Without knowing how synergy contributes to incremental value relevance, and without having a way to measure it, one is left in the dark here. We provide the framework to measure and test for synergy.
The remaining sections are organized as follows. Section 2 reviews related prior research. Section 3 provides definitions and statistical tests for joint and individual synergy. Section 4 describes the components of earnings and EVA examined and our sample selection procedures. Section 5 describes our research design. Section 6 presents our empirical findings. Section 7 provides a summary and discussion.
In this study we illustrate synergy in the context of value relevance, where value relevance (or information content) is defined as the ability of an accounting measure or its components to explain contemporaneous equity returns. Given the key role played by accounting measures in informing firm valuations and share prices, it is natural that accounting researchers, managers, practitioners, and standards setters have long been interested in these associations, inspiring extensive prior research. One main stream of this research addresses questions of incremental value relevance (information content), asking if a given accounting measure adds to value relevance beyond another measure(s). A second main stream addresses questions of relative information content, asking if one measure provides greater value relevance than another.[7] Biddle, Bowen, and Wallace (1997), for example, find GAAP earnings to exceed EVA in value relevance, with both exceeding operating cash flows when considered as summary measures.
The present study extends this literature by introducing the concept of synergy, which describes how the components of accounting measures interact to enhance or diminish value relevance. These interactions explain patterns of incremental and relative information content, and reconcile the concepts. Our empirical tests below reveal that popular accounting performance measures exhibit a range of joint and individual synergies: positive, negative and insignificant. Before proceeding to these results, we develop two formal definitions and corresponding statistical tests for synergy, one for assessing joint synergy for a set of regression predictors, and the other for assessing the synergy of individual predictors.
In this study we consider synergies in value relevance among components of summary accounting measures modeled as unrestricted linear additive regression predictors.[8] Linear additive specifications arise naturally in accounting because many summary accounting measures (e.g., earnings) are constructed by adding their components, with additional motivation provided by their use in prior studies of value relevance (see Section 2 above). By unrestricted we mean that no restrictions beyond usual regression assumptions are placed on the coefficient estimates. This unrestricted case is more general and more consistent with real world financial analysis as it assumes investors can observe all components and assign their own weights, rather than observing only a summary total that implicitly restricts all component coefficients to be equal.[9] In linear additive functional form, synergies arise not from multiplicative interactions of components, but rather from correlations among them.[10]
3.1 Definitions of joint and individual synergy
Joint synergy captures the intuition of a whole greater or less than the sum of its parts. Specifically, we define positive (negative) joint synergy as the case where regression predictors considered altogether explain more (less) of the variation in a dependent variable than they do in sum, when considered individually. For instance, earnings revenue and expense components would be said to exhibit positive (negative) synergy if, when considered collectively, they provide larger (smaller) value relevance than do the components considered separately in sum. That given sets of accounting components interact collectively to enhance or diminish value relevance is of intrinsic interest, as it conveys whether or not they “work together” to comprise an informationally complementary set.
It also is of natural interest to know how a given predictor interacts with other predictors when it is considered individually. For example, we might ask whether an accounting expense, either by its construction or implementation, interacts with other revenues and expenses to enhance or detract from value relevance. Following this intuition, positive (negative) individual synergy is defined as the case where the interactions of a given regression predictor with other predictors increase (decrease) variation explained.
Formalizing these definitions, denote as the variation in a dependent variable explained by the set of unrestricted predictors collectively when all are included in a single multivariate regression. Denoting as the sum of the variations explained by the set of predictors when each is considered separately, the concept of joint synergy examined in this study is expressed as  .
Further, denote as the variation explained by the combined set of predictors after omitting predictor i. It follows that the contribution of predictor i to variation explained, including its interactions with other predictors, is given by  . Denoting as the ‘separate value relevance’ or variation explained by predictor i when it is considered individually, individual synergy is expressed as   . In other words, the effects of the interactions of predictor i with the other predictors, its individual synergy, is given by its contribution to variation explained in their presence, minus its contribution to variation explained in their absence.
3.2
Synergy for the twopredictor case
Applying these definitions, first consider the twopredictor case regressing dependent variable Y on predictors and , where in our context, and comprise the set of components comprising summary accounting measure Z. For example, let Y be stock returns, Z earnings, and and its two additive components, operating cash flows and accruals, respectively. In linear additive form we obtain:
, , (1)
, , (2)
, , (3)
where are intercepts, are slope coefficients, are random error terms; is the variation in dependent variable Y explained by and considered together; and , and are variations in dependent variable Y explained by and when considered separately.
Applying the definition of joint synergy from above to this twopredictor case, we obtain  =  ( + ). Applying the definition of individual synergy, = and = . Thus, individual synergy for is   =   =  ( + ) and individual synergy for is   =   =  ( + ). Thus, for the twopredictor case, there is no difference between joint and individual synergy. See Appendix 1 for further insights into the nature of synergy for the case of two predictors.
3.3
Synergy, relative and incremental value relevance for the twopredictor
case
Relative value relevance (information content) is the difference in variations explained by predictors considered separately. Incremental value relevance (information content) is the incremental variation explained by predictor i beyond other predictors (Biddle, Seow, and Siegel (1995)). In this study we choose to measure incremental value relevance differently than in many previous studies; specifically, as the decrement to variation explained when a given predictor is removed from the combined set of predictors, thus º  .[11]^{,[12]}
Applying these definitions to the twopredictor case, and relying on relations (2a), (3a), and (4a) in Appendix 1, we obtain:
= incremental variation in Y explained by beyond (4)
º  =  = = = + Synergy,
= incremental variation in Y explained by beyond (5)
º  =  = = = + Synergy,
 = relative value relevance of and (6)
=  = difference in separate value relevance between and .
We now generalize to the case of three predictors, as illustrated in Figure 1. The area of each rectangle represents the total variation in Y (not drawn to scale). The total areas of shapes within each rectangle represent variation in Y explained with the combined area of the shapes, , and other terms are as defined above. The top panel of Figure 1 illustrates a case of negative joint and individual synergies. Joint synergy is  = 0.54 – 0.65 = 0.11. Individual synergy for is  = 0.24  0.30 = 0.06; for it is  = 0.12 – 0.20 = 0.08 and for it is  = 0.08  0.15 = 0.07.[13] The bottom panel of Figure 1 illustrates a case of positive synergies.[14] Joint synergy is  = 0.75 – 0.65 = 0.10. Individual synergy for is  = 0.06; for it is  = 0.08 and for it is  = 0.07. Thus, the predictor interactions captured by joint and individual synergies reconcile relative and incremental value relevance. Also notice that with more than two predictors, joint and individual synergies differ, and individual synergies “double count” the effects of predictor interactions represented by the overlapping areas.
_______________________________
Place Figure 1 About Here
_______________________________
3.5 Statistical tests for joint synergy
We begin by stating a null hypothesis for zero joint synergy: the sum of the variations explained by the predictors considered separately equals their variation explained when considered collectively:
H_{0}: () = = _{.}
This is equivalent to a test of the independence of the predictors. Under independence, the sum of the variation explained by the predictors considered separately and the sum of their incremental variations explained both equal to R_{} (Wonnacott 1979):
= R_{ }and
= R_{}.
Consider the following general linear model (in the matrix form), D = MB + e, where D is an n x 1 standardized dependent variable vector (e.g., returns), M is an n x k predictor matrix (n is the sample size, k is the number of predictors), B is a kvector of regression coefficients, and e is an nvector of unobserved normal disturbances. To access the variation explained by subsets of predictor variables, we define M_{i} as an n x 3 component matrix (with 1’s in the first column and a predictor and its lag vector as the other two columns, respectively) and B_{i} as the subset of coefficients of B, where i = 1 to m, represents m sets of predictors. Define N_{C}_{i} as the remaining columns of M and B_{C}_{i} as the remaining coefficients of B. Here m represents the total number of sets of predictors (e.g., CFO and its lag comprise one set of predictors) and k vectors contain m predictor sets, k = 1 + 2m).
The regression D on M, with all the predictors, gives , and the regression D on M_{1}, with only one subset of predictors (e.g., CFO and its lag variable), gives . The relation can be expressed as:
=  ,_{ }or
SS_{reg1}/SST = SS_{regC}/SST  SS_{IC}_{1}/SST,
where _{ }is the incremental variation in D beyond M_{1}. Its numerator also can be interpreted as the bias in the residual sum of squares (S^{2}) due to the omitted parameters B_{C}_{1}, and is described as the squared semipartial correlation (type II SS/SST) in the SAS manual. SS_{regC} is the sum of squared regression of the combined Mmodel, SS_{reg1 }is that of the M_{1}model, SS_{IC}_{1} is the type II sum of squares, and SST is the sum of squares total for all M or M_{i} models. Under independence from above, = , and thus,
SS_{reg1}/SST + SS_{reg2}/SST +…+ SS_{regm}/SST = SS_{regC}/SST, or
SS_{reg1} + SS_{reg2} +…+ SS_{regm} = SS_{regC,},
where SS_{regC} = R_{} * SST = B`M`MB = Q, (when D is standardized), SS_{reg1} = SS_{regC} – SS_{I}_{ C1}, and SS_{I}_{ C1} = B_{C}_{1}`N_{C}_{1}`[I_{n}  M_{1}(M_{1}`M_{1})^{1}M_{1}`]N_{C}_{1}B_{C}_{1} = Q_{1. } SS_{reg2},...,SS_{regm }and Q_{2 ,.}..,Q_{m }are obtained similarly (using the same SST). All Q’s contain the quadratic forms of regression coefficients (B’s) and all M’s and N’s are known values.
Thus, the null hypothesis for joint synergy can be written as:
H_{0}: [SS_{reg1} + SS_{reg2} + … + SS_{regm}]  SS_{regC} = 0, or
H_{0}: [(SS_{regC}  SS_{Ic1}) + … + (SS_{regC}  SS_{IC}_{m})]  SS_{regC} = 0, or
H_{0}: (Q_{ }– Q_{1}) +… + (Q – Q_{m}) – Q = 0, or_{ }
H_{0}: [Q_{1 }+ Q_{2} +…+ Q_{m}]  (m1)[Q] = 0.
This is a nonlinear hypothesis in quadratic forms of regression coefficients B. Following Biddle, Seow, and Siegel (1995), statistical significance is assessed via a Wald chisquared test and is used in conjunction with White’s (1980) correction for heteroskedastic errors.
3.6 Statistical tests for individual synergy
Following similar reasoning, individual synergy is tested by examining the difference between the incremental () and separate () value relevance of a given predictor i. For example, if i = 1, the null hypothesis for individual synergy is:
H_{0 }:  = 0, or
H_{0 }: (  )  = 0, which after rearranging is
H_{0 }: + = R_{},
where we follow a general linear model (in the matrix form), D = MB + e and the regression D on M including all predictors gives R_{}, the regression D on M_{1} predictors gives R_{}(e.g. CFO and its lag variable)_{,} and the regression D on M_{C}_{1} gives R_{, }where M_{C}_{1}, M_{C}_{2 }, …, M_{C}_{m} represents the M matrix that contains all but one omitted predictor set, such as CFO and its lag.
Viewing as analogous to in the statement of the null hypothesis for joint synergy above, it follows that individual synergy can be tested as:
H_{0}: R_{} + R_{} = R_{}, or
H_{0}: [Q_{1 }+ Q_{C1}] – [(21)Q] = 0.
In other words, the test for individual synergy is like a test for joint synergy between two predictor sets, where the “second predictor set” contains all sets of predictors except the first.
We illustrate definitions and statistical tests for synergy for: 1) earnings expressed as revenue and expense components; 2) noncash accruals that reconcile earnings with operating cash flows; and 3) EVA components that reconcile EVA with earnings.[15] Figure 2 lists the earnings and EVA components we examine. Nine revenue and expense components sum to earnings, where earnings is defined as GAAP aftertax net income before extraordinary items and discontinued operations (EARN). Ten noncash accrual components reconcile earnings with Cash Flow from Operations (CFO). Ten EVA adjustments reconcile EVA with earnings.[16] All components and summary measures are obtained from the Standard & Poor’s 1500 Compustat database, where the EVA adjustments are created for Standard & Poor’s by Stern Stewart & Co.[17] Additional details regarding component definitions are provided in the Appendix 2.
_______________________________
Place Figure 2 About Here
_______________________________
Our sample comprises all available observations from the Standard & Poor’s 1500 database for years 19881999.[18] The year 1988 was chosen because it is the first year in which data on cash flow from operations are available for all firms; the year 1999 was chosen to avoid any effects of the “dot com” market correction of 2000. Stern Stewart stopped supplying EVA data to Compustat after 2001. The initial sample of 1,500 firms (18,000 firmyear observations) was reduced by 169 firms (5,486 firmyear observations) due to missing Compustat or CRSP data, and by 36 firms (1,332 firmyear observations) due to missing lagged observations required by equation (7) below. Deleting observations more than 6 standard deviations from the medians reduced the sample by a further 68 firms (1,515 firmyear observations). The resulting final sample has 1,227 firms and 9,667 firmyear observations with data items winsorized to their medians plus or minus eight standard deviations. Descriptive statistics are presented in Table 1.
_______________________________
Place Table 1 About Here
_______________________________
5. Research Design
Following Biddle, Bowen, and Wallace (1997), value relevance is estimated using regressions of marketadjusted equity returns on the levels and lagged levels of predictors in the following generic form, here illustrated for one predictor:
(7)
where:
MAR_{t} = Marketadjusted equity return for the 12month period ending 3 months after fiscal year t,
MVE_{t1} = Market value of equity at the end of year t1,
X_{t} = Summary measure or component predictor for year t,
e_{t} = Error term for year t.
Each predictor is deflated by the market value of equity to control for scale differences and heteroskedasticity.[19]
In results presented below we apply expression (7) in three ways to the components of summary accounting measures. First, we estimate (7) for the combined set of components comprising a summary measure to obtain . Second, we estimate (7) for each component considered separately to assess its value relevance in the absence of other components, . Summing the to obtain provides for our test of joint synergy measured as  . Third, we estimate a version of (7) that successively omits each component from the combined set to obtain . By comparing the value relevance of the combined set with the value relevance of the set that omits one component, we obtain a measure of the incremental value relevance of each component that includes its synergies with other components, º  , and a test of our measure of individual synergy,  =   .
6. Test Results
We present below estimates of the value relevance, joint synergy, incremental value relevance, and individual synergy of earnings and EVA components, where earnings components are expressed two ways – as individual income statement line items and as noncash accruals beyond cash flows from operations in the operating section of the statement of cash flows. As a benchmark for other results, Panel A of Table 2 presents the variation in market adjusted returns explained by the summary performance measures of earnings (EARN), cash from operations (CFO), and EVA, where each enters equation (7) as a definition, rather than as the set of its components. Consistent with findings in Biddle, Bowen, and Wallace (1997), earnings has the highest R^{2} (7.88%), followed by EVA (3.72%) and CFO (1.78%). As observed below, allowing their components to enter expression (7) in unrestricted, rather than in restricted definitional form, increases the variations explained to 10.70% for income statement line items (Table 2, Panel B), 11.17% for earnings noncash accruals beyond CFO (Table 5), and 12.22% for noncash accruals plus EVA components (Table 8).
_______________________________
Place Table 2 About Here
_______________________________
6.1 Results for income statement line items
6.1.1 Test of joint synergy for income statement
line items
Panel B of Table 2 presents results on the value relevance of income statement line items (revenues and expenses) when considered separately, in the notation of Section 3. Eight of the nine income statement line items exhibit variation explained that differs significantly from zero at conventional levels (Ftest of the null R^{2} = 0 is rejected at the 0.05 level or less); the one exception is depreciation expense (DP). , the sum of the s from the separate univariate regressions, equals 10.66%. for the unrestricted multivariate regression that includes all nine components equals 10.70%. Thus, income statement line items considered collectively provide value relevance approximately equal in magnitude to that of the sum of their parts when considered separately. Confirming this, the formal test for no joint synergy (  = 0) is not rejected at conventional levels (p = 0.90). Therefore, revenue and expense components exhibit insignificant joint synergy in value relevance.
6.1.2 Incremental value relevance of income
statement line items
Table 3 presents estimates of incremental value relevance for income statement line items where incremental value relevance is measured as the reduction in variation explained when one item is omitted from the combined set, º  . For example, when sales revenue (SALES) is omitted from the multivariate regression of all income statement line items, R^{2} falls from 10.70% (= ) to 7.20% (= ), for a loss of 3.50% (= ). This drop in R^{2}, or incremental value relevance, is highly significant (partial F = 188.80) and the null of no change in R^{2} is rejected at the 0.000 level. Overall, the results in Table 3 reveal that seven of the nine income statement line items exhibit significant incremental value relevance, with income tax expense (TXT) and minority interest (MII) being the exceptions. These results are depicted graphically in Figure 3.
_______________________________
Place Table 3 About Here
_______________________________
_______________________________
Place Figure 3 About Here
_______________________________
6.1.3 Individual synergies of income statement line
items
Recall from Section 3 that synergy is absent (present) in R^{2} measures of separate (incremental) value relevance; thus, the individual synergy of a component is described by the difference between a component’s incremental and separate value relevance,  . Table 4 combines the results in Tables 2 and 3 to obtain measures of individual synergy for income statement line items and test their statistical significance. Ranks for each also are presented. For example, whereas income tax expense (TXT) ranks first in separate value relevance, it ranks last in incremental value relevance. These contrasting findings are explained by its interactions with the other revenue and expense components as reflected by its significant negative individual synergy. In other words, the interactions of income tax expense with other components detract significantly from variation explained. [20] The largest positive individual synergy is exhibited by costofgoodssold expense (COGS) and sales revenue (SALES). Overall, five income statement line items exhibit positive individual synergy that is significant at conventional levels (p < .05) (COGS, SALES, XINT, XSGA, and DP), and three exhibit significant negative individual synergy (TXT, SPI and MII). These results are depicted graphically in Figure 4.
_______________________________
Place Table 4 and Figure 4 About Here
_______________________________
6.2 Results for earnings comprised of cash from
operations and nonaccrual components
6.2.1 Test of joint synergy for earnings comprised
of cash from operations and noncash accrual components
We next contrast the above results for revenue and expense components with those for noncash accrual components that reconcile earnings with cash flows from operations. Corresponding with Table 2 (Panel B), Table 5 presents results on the variation explained by earnings noncash accrual components and CFO when considered separately, i.e.,. CFO and nine of the ten noncash accruals exhibit variation explained that differs significantly from zero at conventional levels (Ftest p < 0.05); the only exception is depreciation expense (DP). The sum of the variations explained by the separate univariate regressions, , equals 9.82% whereas the variation explained by the unrestricted multivariate regression that includes all eleven components, , equals 11.17%. Thus, earnings noncash accrual components and CFO considered collectively provide value relevance larger in magnitude than the sum of their parts when considered separately. Confirming this, the formal test for joint synergy (  = 0) is rejected at conventional levels (p = 0.027). Therefore, earnings noncash accrual components exhibit significant positive joint synergy with interactions that collectively enhance value relevance.
_______________________________
Place Table 5 About Here
_______________________________
6.2.2
Incremental value relevance for earnings comprised of cash from
operations and noncash accrual components
Table 6 presents changes in variation explained (i.e., R^{2} lost) when CFO and noncash accrual components are omitted individually from their combined set, º  . For example, when CFO is omitted from the multivariate regression of all components, R^{2} falls from 11.17% to 5.63%, for a loss of 5.54%. This drop in R^{2} is highly significant (partial F = 300.54) and the null of no change in R^{2} is rejected at the 0.000 level. Overall, the results in Table 6 reveal that all eleven components exhibit significant incremental value relevance. These results are presented graphically in Figure 5.
_______________________________
Place Table 6 About Here
_______________________________
_______________________________
Place Figure 5 About Here
_______________________________
6.2.3 Individual synergies for earnings comprised
of cash from operations and noncash accrual components
Table 7 combines the results in Tables 5 and 6 to summarize measures of individual synergy for CFO and noncash accruals, and their statistical significance. For example, the change in accounts receivable (RECCH) ranks first in separate value relevance and second in incremental value relevance. These results are reconciled and explained by its significant positive individual synergy. In other words, the interactions of RECCH with the other components adds significantly to variation explained. The largest positive individual synergy is exhibited by CFO and miscellaneous noncash accruals (MISC); the largest negative individual synergy is exhibited by the change in accrued income taxes (TXACH) and change in accounts payable and accrued liabilities (APALCH). Overall, CFO plus five earnings noncash accrual components exhibit positive individual synergy that is significant at conventional levels (p < .05) (CFO, MISC, INVCH, RECCH, TXD, and SPPIV), and two exhibit significant negative individual synergy. These results are presented graphically in Figure 6.
_______________________________
Place Table 7 About Here
_______________________________
_______________________________
Place Figure 6 About Here
_______________________________
6.3 Results for EVA components
6.3.1 Test of joint synergy for EVA components
In our third illustration, we contrast the above results for two versions of earnings components with those of EVA components. Following Tables 2 and 5, Table 8 presents results on the variation explained by EVA adjustments (which begin with the prefix SS for Stern Stewart) beyond income statement line items when considered separately, . Fourteen of the eighteen components exhibit variation explained that differs significantly from zero at conventional levels (Ftest p < 0.05); seven of eight income statement line items and seven of ten EVA adjustments.[21] The sum of the variations explained by the separate univariate regressions, , equals 14.00% whereas the variation explained by the unrestricted multivariate regression that includes all eighteen components, , equals 12.22%. Thus, EVA components considered collectively provide value relevance smaller in magnitude to that of the sum of their parts when considered separately. Confirming this, the formal test for joint synergy (  = 0) is significant at p = .044. Therefore, EVA components exhibit significant negative joint synergy, with interactions that collectively diminish value relevance.[22]
_______________________________
Place Table 8 About Here
_______________________________
6.3.2 Incremental value relevance of EVA components
Table 9 presents changes in variation explained when EVA components are omitted individually from the combined set, º  . For example, when SALES is omitted from the multivariate regression of all components, R^{2} falls from 12.22% to 9.57%, for a loss of 2.65%. This drop in R^{2} is highly significant (partial F = 145.31) and the null of no change in R^{2} is rejected at the 0.000 level. Overall, the results in Table 9 reveal that ten components exhibit significant incremental value relevance; seven of eight income statement line items (with income tax expense being the exception); and three of ten EVA adjustments (Stern Stewart’s deferred tax, research and development, capital charge adjustments) with a fourth adjustment for pension costs being marginally significant at p = 0.057. These results are presented graphically in Figure 7.
_______________________________
Place Table 9 About Here
_______________________________
_______________________________
Place Figure 7 About Here
_______________________________
6.3.3 Individual synergies of EVA components
Table 10 combines the results in Tables 8 and 9 to obtain a measure of individual synergy for each EVA component and test its statistical significance. Income tax expense (TXT) ranks first in separate value relevance, as it did in Table 4, but now thirteenth in incremental value relevance. This result is explained by its significant negative individual synergy. In other words, the interactions between income tax expense and the other EVA components detracts significantly from variation explained. The largest positive individual synergy is exhibited by costofgoods sold expense (COGS) and sales revenue (SALES); the largest negative individual synergy is exhibited by income tax expense (TXT) and Stern Stewart’s estimated capital charge (SSCHG). Overall, three components exhibit positive individual synergy that is statistically significant at conventional levels, SALES, COGS, and DP: seven exhibit significant negative individual synergy. Tellingly, none of the ten Stern Stewart adjustments exhibits significant positive individual synergy although deferred taxes (SSDT) is close at p = 0.072. Six exhibit insignificant individual synergy and four exhibit significant negative individual synergy, namely, Stern Stewart’s adjustments for cost of capital (SSCHG), aftertax unusual items (SSUAT), capitalized research and development expense (SSCRD), and aftertax interest expense (SSIAT). These results are presented graphically in Figure 8.
_______________________________
Place Table 10 About Here
_______________________________
_______________________________
Place Figure 8 About Here
_______________________________
These findings suggest that EVA adjustments created by Stern Stewart & Co. exhibit no positive synergies, either jointly or individually. On the contrary, their individual interactions either contribute insignificantly or detract from value relevance, as do their collective interactions. In contrast to noncash earnings accruals that seem to work together as a coherent set to enhance value relevance, EVA adjustments behave more like an ad hoc collection without constructive interactions that contribute to value relevance. We offer a possible explanation suggested below. The interactions revealed by these synergy findings also explain why prior research found EVA to provide lower value relevance than accrual earnings.
6.4 Discussion of
test results
The findings above illustrate the concepts and tests for joint and individual synergy and three possible outcomes of joint synergy. They provide new results on the contributions of individual components, and they illustrate how joint and individual synergy, incremental and relative value relevance combine to provide a comprehensive view of how components of earnings and EVA components interact to reflect value relevance.
Although it may initially seem counterintuitive that income statement line items should exhibit insignificant joint synergy, whereas noncash accruals exhibit positive joint synergy, one intuition is that revenues and expenses act more like substitutes in explaining share returns, while noncash accruals are more complementary, reflecting the effects of accrual mechanics and management discretion in creating valuationrelevant earnings accruals from operating cash flows. That EVA components exhibit negative joint synergy is more intuitive given that EVA is designed primarily as a management incentive tool, rather than as a performance indicator for market participants (advertising claims notwithstanding). Beyond assertions that EVA adjustments correct “distortions” in GAAPbased earnings to create a measure more useful for valuation (see, for example, Stewart 1991, 1994; Stern, Stewart, and Chew 1995; Young and O’Byrne 2001, Feltham, et al. 2004), Stern Stewart & Co. further describe EVA as serving a broader role that includes not only shareholder communication, but also strategic planning, operating decisions, and management compensation (Stern, Stewart, and Chew 1995). This gives rise to differing motives for EVA adjustments, as described by Ehrbar (1998, p. 167):
“Some adjustments are necessary to avoid mixing operating and financing decisions. Some provide a longterm perspective. Some avoid mixing stocks and flows. Some convert GAAP accruals to a cashflow basis, while others convert GAAP cashflow items to additions to capital. Still others … resolve organizational interface problems that distort decisions.”
And, as characterized by Young and O’Byrne (2001, p. 205):
“There is no accepted cannon of these adjustments, because they are directed at a variety of accounting, performance measurement, and incentive issues. Not only is there disagreement over the importance of each issue, but also in some cases EVA proponents disagree on the correct way to address it.”
Thus, it is not so surprising that EVA adjustments designed to address adverse management incentives may detract from value relevance collectively if they were not designed specifically for this purpose.
New insights also are provided by findings regarding the individual synergy and incremental value relevance of earnings and EVA components; for example, tax expense provides the largest separate value relevance, but smallest incremental value relevance, a result explained by its significant negative individual synergy.[23] That sales revenue and costofgoodssold expense exhibit significant positive individual synergy is intuitive, in that they are major determinants of profitability in interaction with each other and with other earnings accruals. Put another way, individual synergy reveals the amount of incremental value relevance, positive or negative, that arises from the interactions of a component with others, rather than from its separate value relevance.
Among noncash earnings accruals beyond CFO, the change in accounts receivable exhibits both the largest incremental information content and the largest separate value relevance, a result explained by its significant positive individual synergy. To illustrate how individual synergy may arise, notice that the change in receivables interacts with CFO to reveal sales revenues, an arguably valuerelevant fundamental. The change in inventories interacts similarly with CFO to reveal costofgoodssold expense. That CFO ranks first in individual synergy follows.
The test results above also provide the first comprehensive view of value relevance for individual EVA adjustments. Broadly speaking, EVA adjustments exhibit generally smaller separate value relevance, incremental value relevance, and individual synergy than observed for earnings components. Three of ten EVA adjustments fail to exhibit significant separate value relevance (SSLIFO, SSGWA, SSCM) and seven of ten fail to exhibit significant incremental value relevance (SSLIFO, SSBD, SSGWA, SSCM, SSPEN, SSUAT, SSIAT). Completing the picture, none of the EVA adjustments exhibits significant positive individual synergy, and four exhibit significant negative individual synergy. In explanation, and extending the reasoning above, notice that EVA derives from residual income defined as net income minus a charge for equity capital. Many EVA adjustments designed by Stern Stewart cancel earnings accruals, since otherwise, executives could manipulate them to boost their EVAbased bonuses without enhancing firm value.[24] That an EVA adjustment that removes an earnings accrual should interact with other components to diminish value relevance is consistent with prior evidence that earnings, which includes noncash accruals, dominates operating cash flows, which excludes them, in value relevance (e.g., Biddle, Seow, and Siegel, 1995).
As illustrated, synergies reveal whether components interact collectively and individually with others to enhance or diminish value relevance. With regard to the desirability of synergy as a characteristic of accounting data, notice that joint (individual) synergy accounts for combined (incremental) value relevance beyond separate value relevance. Thus, combined (incremental) value relevance can be thought of as arising from two sources: innovations that create separate value relevance, and innovations that create positive joint (individual) synergy.[25] These innovations can arise from accounting conventions and mechanics, the exercise of management discretion, management incentives, and institutional features.[26]
7. Summary
Synergy
conveys a whole that is greater or less than the sum of its parts due to
interactions among its parts. This study
examines the concept of synergy as applied to the value relevance of summary
accounting performance measures. We
provide two definitions of synergy, corresponding statistical tests, and three
illustrative applications. Positive
(negative) joint synergy arises when
components considered collectively explain more (less) of the variation in a
dependent variable than when considered separately and then summed. Positive (negative) individual synergy arises when a given component interacts with
remaining components to enhance (diminish) variation explained.
It may
be counterintuitive to some that both positive
and negative synergies may
exist. We therefore show both
analytically and empirically such existence by applying these synergy
definitions and tests to the components of aftertax earnings expressed as
income statement line items, noncash accruals beyond operating cash flows, and
the components of EVA. Aftertax
income statement line items exhibit insignificant
joint synergy for the firmyear sample examined, implying a whole
indistinguishable from the sum of its parts in value relevance; noncash
accrual components exhibit positive joint
synergy, implying a whole greater than the sum of its parts. Economic value added (EVA^{®})
components exhibit negative joint synergy,
implying a whole less than the sum of its parts. Put another way, noncash accrual components
comprise a more complementary set than EVA components, which behave more like
ad hoc adjustments whose collective interactions detract from value relevance.
We also relate our definitions of synergy to the concepts of incremental and relative value relevance, showing that incremental value relevance (as we define it) includes individual synergy, whereas relative value relevance does not. Put another way, relative value relevance assesses “who wins,” incremental value relevance assesses “who adds,” and synergy reveals how components “work together or against each other.” Synergy is of interest in itself, as it reveals the influence of component interactions on value relevance distinct from separate value relevance. As such, it constitutes a source of incremental value relevance that can be managed and otherwise influenced by accounting procedures and institutional features.
Further,
we provide new evidence on the synergies of individual earnings and EVA
components and their separate and incremental value relevance. These findings reconcile and extend prior
findings regarding the incremental and separate value relevance of earnings
components and the relative value relevance of earnings and EVA summary
measures.
Finally, we might ask what our findings mean to users of financial statements. In other words, what are the takeaways? One answer is that synergy completes the picture of separate and incremental value relevance. In this sense it is like the third angle of a triangle that both completes the triangle, and reveals its existence. This new third angle also is of interest in itself, independent of the other two, as it isolates the effects of component interactions. Thus, the concept of synergy completes a value relevance triangle whose angles can now all be described, measured, and tested independently. To illustrate, notice that the incremental value relevance of income tax expense of 0.03 (Table 4) reveals only that it contributes little beyond the components of pretax earnings. But its separate value relevance of 6.00 exceeds that of any other income component. These results are explained and reconciled by its significant individual negative synergy of –5.97. To know that income tax expense “works against” other pretax earnings components in value relevance is of interest in itself. Likewise, it is of independent interest that noncash accruals beyond CFO and pretax earnings components exhibit positive synergy and that EVA components exhibit negative synergy, revealing how their components interact.
A final answer emerges if one asserts that incremental value relevance is what matters most, and arguably it does. Then one can ask “where does incremental value relevance come from and how it is created?” Our findings show is that it arises from two sources – from separate value relevance and from synergy. Thus, if an analyst or policy maker wants to find or create incremental value relevance, there are two separate ways to obtain it: increase separate value relevance holding synergy constant, or increase synergy positively holding separate value relevance constant. Without knowing that synergy contributes to incremental value relevance, and without having a way to measure it, one is left in the dark here.
Obvious avenues for further research include applying the synergy definitions and tests developed in this study to other summary accounting measures to determine if they also exhibit synergies. Examples include comprehensive income, proforma income, core earnings, and cash flow components. Another accounting application is to examine synergies exhibited among components defined by different country GAAPs. Because our synergy definitions and tests apply equally well in any setting that models a dependent variable as a linear additive combination of unrestricted predictors, they also extend to behavioral, management, and marketing contexts, and to fields beyond management where unrestricted linear additive measures apply. For example, one could assess synergies among the determinants of consumption expenditures, consumer choices, crop yields, medical effectiveness, and housing prices.
Appendix 1
Nature of synergy in the twopredictor case
Additional insights are obtained into the nature of synergy for the twopredictor case from the following relations (Gujarati (1995, p. 214)):
, (1a)
, (2a)
, (3a)
where is the correlation between and , and = is the square of the partial correlation coefficient, reflecting the proportion of variation in Y not explained by alone that is explained by the further inclusion of .
From expressions (1a), (2a), and (3a), it follows that (joint and individual) synergy for the twopredictor case can be expressed as:
 = . (4a)
Thus, synergy depends on correlations among Y, and , plus the variation explained by both components () multiplied by their squared correlation (). Because > 0 by construction, it follows that synergy can be negative for the twopredictor case only when is positive and greater than [27]
Appendix 2
Compustat
data item 125. DP represents the
noncash charge allocating the cost of capitalized expenditures to periods
subsequent to the cash outlay. This item
is, on average, the largest of the GAAP accruals.
Compustat
item 124. XIDO represents items that are
both unusual and infrequent (extraordinary items), and also the effects of
discontinued operations. These two items
appear after income before extraordinary items and discontinued operation on
the income statement. These items appear
further down on the income statement because they are considered less
persistent in nature.
Compustat
item 126. TXD represents the income
statement account deferred tax expense.
This account represents tax expense based on GAAP (as opposed to
taxable) income. This item includes
investment tax credits.
Compustat
item 106. ESUB represents the unremitted
portion of an unconsolidated subsidiary’s earnings reported under the equity
method on the income statement. ESUB is
calculated as the firm’s ownership share in the subsidiary times the
subsidiaries earnings, reduced by the firm’s share of the subsidiaries dividend
distributions.
Compustat
item 213. SPPIV represents gains and
losses from the disposal of assets.
Compustat
item 302. RECCH represents changes in
the balance sheet account, accounts receivable.
Compustat
item 303. INVCH represents changes in
the balance sheet inventory account.
Compustat
item 304. APALCH represents changes in
the balance sheet accounts payable and accrued liabilities accounts.
Compustat
item 305. TXACH represents changes in
the balance sheet accrued income taxes (income taxes payable) account.
Compustat
items 217 and 307. MISC represents
changes in miscellaneous current balance sheet accounts and items not
specifically included in another category within the operating activities
section of the statement of cash flows.
Examples of items included are changes in current deferred taxes,
amortization of negative intangibles, minority interest reported in operations,
special items, amortization of goodwill on unconsolidated subsidiaries,
provision for losses on accounts receivable, and unrealized gains/losses on
sale of property, plant, and equipment.
SSDT
is an adjusted difference in Compustat data item 35, deferred taxes and
investment tax credit, where the adjustment is for the longterm portion of the
deferred tax liability. The rational
behind this adjustment is that the deferred taxes will never be paid as long as
the firm replenishes the assets that give rise to the tax deferral. Since this replenishment is implied in the
going concern concept, it is felt that the deferral is really the equivalent of
permanent equity. This adjustment serves
to bring tax expense closer to a cash basis method of accounting for
taxes. Specifically, the adjustment adds
the deferred tax reserve back to equity capital and the current increase in the
reserve to NOPAT.
SSLIFO
is the change in Compustat item 240, LIFO reserve. The rational for this adjustment is that in
times of rising prices LIFO tends to understate ending inventories. To make the adjustment, the LIFO reserve is
added to capital and the increase in the LIFO reserve is added to NOPAT. This serves to make ending inventory a better
approximation of current replacement cost and bring into earnings the
unrealized gain attributable to holding appreciating inventory. A second purported advantage is that it eases
the task of comparing firms that use different inventory cost flow assumptions.
SSBD
is the change in Compustat item 67, receivables – estimated doubtful. The rational for this adjustment is that this
reserve obscures the actual timing of cash receipts and disbursements, and
provides an opportunity for earnings management. The adjustment consists of adding the reserve
back to capital and adding to NOPAT the increase in the reserve.
Compustat
item 65, amortization of intangibles.
SSGWA includes amortization of intangibles other than just goodwill, but
is used because Compustat does not have a data item that represents only
goodwill amortization. The adjustment
consists of adding cumulative goodwill amortization back to capital and adding
goodwill amortization expense to NOPAT.
The rational for this adjustment is that the GAAP requirement to
amortize goodwill against earnings may deter management from consummating
sensible acquisitions. The adjustment
serves to make goodwill a nonissue so as not to influence a manager’s
acquisition decisionmaking.
SSCM
is the aftertax capitalization of all marketing and advertising expenses
through the balance sheet date, net of accumulated amortization. The rationale behind this adjustment is that
marketing and advertising expenditures have a useful life beyond the current
accounting period and should therefore be capitalized and amortized over their
useful life. To make this adjustment,
the amount of capitalized marketing and advertising is added to capital and
increases in capitalized marketing are added to NOPAT by reducing operating
expenses.
SSCRD
is the after tax capitalization of all research and development expenses
through the balance sheet date, net of accumulated amortization. The rationale behind this adjustment is that
research and development expenditures have a useful life beyond the current
accounting period and should therefore be capitalized and amortized over their
useful life. This adjustment is similar
to that for capitalized marketing expenses with the amount of capitalized
research and development added to capital and increases in capitalized
marketing added to NOPAT.
Compustat
item 43, pension and retirement expense, less item 331, pension plans – service
costs. SSPEN represents the pension and
retirement expense included as an expense in the income statement less the
present value of expected future pension payments attributed to employee
service performed during the current year.
Compustat
item 49. MII represents the portion of
the consolidated subsidiary income applicable to common stock not owned by the
parent company. MII does not appear in
Tables 8, 9, and 10 because it is subtracted from accrual earnings to create
EVA, thus canceling the earnings MII component.
SSUI
is the amount of any losses less gains, aftertax, arising from nonrecurring
property disposals, restructurings, or other unusual events. The rationale
behind this adjustment is similar to that given for using the full cost
approach instead of successful efforts.
The adjustment consists of excluding from NOPAT gains and losses when
they are unusual and nonrecurring.
Aftertax cumulative losses less gains are added back to the balance
sheet, as the losses represent investment strategy, the gains a recovery of
capital.
SSIAT
is the aftertax impact of interest expense less interest income. Interest expense encompasses interest
measured in Compustat item 15, along with capitalized interest item 147 and
also an amount for the impact that capitalizing operating leases would have on
taxable income. Taxable income consists
of Compustat item 62. Aftertax interest
is removed from NOPAT in order to separate financing items from operating
items. Interest is then included in the
capital charge that is subtracted from NOPAT in the computation of EVA.
SSCHG
represents the charge on all capital, both debt and equity capital. It is computed as the firm’s weighted average
cost of capital multiplied by invested capital.
Invested capital represents funds invested in a business to support
operations. Included in capital is
shareholder’s equity adjusted for Stern Stewart equity adjustments, debt, lease
obligations, and other Stern Stewart capital adjustments.
References
Barth, M., W. Beaver, and W.
Landsman. 2001. The relevance of the value relevance literature for financial
accounting standard setting: Another view. Journal of Accounting and Economics 31 (13): 77104.
Biddle, G., R. Bowen, and J. Wallace. 1997. Does EVA beat earnings? Evidence on associations with stock returns and firm values. Journal of Accounting and Economics 24: 301336.
Biddle, G., G. Seow, and A. Siegel. 1995. Relative versus incremental information content. Contemporary Accounting Research 12: 123.
Bowen, R., D. Burgstahler, and L. Daley. 1987. The incremental information content of accrual versus cash flows. The Accounting Review 62: 723747.
Burgstahler, D. and I. Dichev. 1997. Earnings, adaptation, and equity value. The Accounting Review 72: 187215.
Collins, D., E. Maydew, and I. Weiss. 1997. Changes in the valuerelevance of earnings and book values over the past forty years. Journal of Accounting and Economics 24: 3967.
Collins, D., M. Pincus, and H. Xie. 1999. Equity valuation and negative earnings: The role of book value of equity. The Accounting Review 74 (1): 2961.
Ehrbar, A. 1998. EVA: The Real Key To Creating Wealth, Wiley, New York.
Feltham, G., G. Isaac, C. Mbagwu, and G. Vaidyanathan. 2004. Perhaps EVA does beat earnings – Revisiting previous evidence. Journal of Applied Corporate Finance 16: 8388.
Francis, J., K. Schipper, and L. Vincent. 2002. Expanded disclosures and the increased usefulness of earnings announcements. The Accounting Review 77 (3): 515546.
Gujarati, D. 1995. Basic Econometrics. New York: McGrawHill.
Hayn, C. 1995. The information content of losses. Journal of Accounting and Economics 20: 125153.
Holthausen, R., and R. Watts. 2001. The relevance of the valuerelevance literature for financial accounting standard setting. Journal of Accounting and Economics 31 (13): 376.
Kothari, S. 2001. Capital markets research in accounting. Journal of Accounting and Economics 31 (13): 105232.
Lee. C. 2001. Market efficiency and accounting research: A discussion of ‘capital market research in accounting’ by S.P. Kothari. Journal of Accounting and Economics 31 (13): 233254.
Lev, B. and T. Sougiannis. 1996. The capitalization, amortization, and valuerelevance of R&D. Journal of Accounting and Economics 22 (1): 107138.
Lo, Kin. 2004. The effects of scale differences on inferences in accounting research: Coefficient estimates, tests of incremental association, and relative value relevance. Working paper. University of British Columbia and MIT.
O’Byrne, S. 1996. EVA® and market value. Journal of Applied Corporate Finance 9 (1): 116125.
Sougiannis, T. 1994. The accountingbased valuation of corporate R&D. The Accounting Review 69 (1): 4468.
Stern, J., B. Stewart III, and D Chew Jr. 1995. The EVA Financial Management System. Journal of Applied Corporate Finance 8: 3246.
Stewart III, B. 1991. The Quest for Value. New York: Harper Business.
Stewart III, B. 1994. EVA: fact or fantasy? Journal of Applied Corporate Finance 7: 4670.
White, H., 1980. A heteroskedasticityconsistent covariance matrix estimator and a direct test for heteroskedasticity. Econometrica 48: 817838.
Wonnacott, R. and Wonnacott, T. 1979. Econometrics. New York: Wiley.
Young, S. and S. O’Byrne. 2001. EVA and ValueBased Management: A Practical Guide To Implementation. McGrawHill, New York.
Table 1
Descriptive statistics for earnings, EVA^{®},
and their components
Variable^{1} 
Obs. 
Mean 
Median 
Std. Dev. 
Obs. 
Mean 
Median 
Std. Dev. 

All observations^{2,3} 
All nonzero value observations^{2,4} 

EARN 
9,667 
9,667 

CFO 
9,667 
398.64 
79.32 
1,276.27 
9,667 
398.63 
79.32 
1,276.21 
EVA 
9,667 
36.43 
2.51 
758.73 
9,667 
36.43 
2.52 
758.73 
Panel B. Earnings components expressed as revenue and expense line items: 

SALES 
9,667 
3,486 
844.38 
9,608.8 
9,662 
3,487 
844.90 
9,610.94 
COGS 
9,667 
2,363 
533.63 
7,149.7 
9,662 
2,365 
533.83 
7,151.32 
SGA 
9,667 
536.41 
89.86 
1,669.3 
7,681 
675.11 
148.99 
1,847.52 
DP 
9,667 
172.68 
33.47 
575.56 
9,585 
174.16 
33.99 
577.79 
XINT 
9,667 
97.66 
12.15 
629.81 
8,853 
106.64 
15.50 
657.41 
NOPI 
9,667 
31.81 
3.37 
158.57 
9,060 
33.94 
3.98 
163.52 
SPI 
9,667 
26.82 
0.00 
253.45 
4,067 
63.75 
5.00 
387.76 
TXT 
9,667 
114.03 
24.80 
339.48 
9,480 
116.27 
25.70 
342.43 
MII 
9,667 
3.32 
0.00 
20.75 
1,637 
19.60 
2.88 
47.17 
Panel C. Earnings components as noncash accruals beyond cash flows from operations 

DP 
9,667 
182.65 
35.37 
602.42 
9,514 
185.59 
36.65 
606.80 
XIDO 
9,667 
2.05 
0.00 
95.87 
1,096 
18.10 
0.13 
284.35 
TXD 
9,667 
3.85 
0.00 
74.91 
7,341 
5.07 
0.39 
85.92 
ESUB 
9,667 
0.26 
0.00 
41.81 
1,745 
1.46 
0.66 
98.43 
SPPIV 
9,667 
11.32 
0.00 
119.27 
3,976 
27.53 
0.54 
184.78 
RECCH 
9,667 
34.78 
5.25 
189.33 
8,956 
37.54 
6.74 
196.44 
INVCH 
9,667 
17.41 
0.44 
129.98 
7,592 
22.17 
3.28 
146.31 
APALCH 
9,667 
18.09 
0.13 
116.21 
7,231 
24.19 
3.89 
133.83 
TXACH 
9,667 
2.07 
0.00 
61.91 
4,407 
4.54 
0.43 
91.64 
MISC 
9,667 
51.32 
1.94 
543.07 
9,658 
51.37 
1.95 
543.32 
SSDT 
9,667 
3.90 
0.01 
107.16 
9,226 
4.08 
0.20 
109.68 
SSLIFO 
9,667 
0.23 
0.00 
73.02 
2,872 
0.78 
0.03 
133.98 
SSBD 
9,667 
2.17 
0.00 
26.15 
6,360 
3.30 
0.21 
32.18 
SSGWA 
9,667 
6.20 
0.00 
35.98 
2,457 
24.40 
4.40 
68.22 
SSCM 
9,667 
1.80 
0.00 
23.76 
3,236 
5.39 
0.40 
40.84 
SSCRD 
9,667 
6.27 
0.00 
52.61 
4,617 
15.21 
2.23 
75.34 
SSPEN 
9,667 
0.95 
0.46 
79.84 
7,934 
1.16 
1.05 
88.13 
SSUAT 
9,667 
20.03 
0.00 
187.14 
4,386 
44.14 
3.00 
275.93 
SSIAT 
9,667 
57.91 
7.62 
376.50 
9,612 
58.24 
7.79 
377.55 
SSCHG 
9,667 
267.3 
46.08 
991.13 
8,699 
297.1 
59.37 
1,040.58 
^{ }
^{ }
^{1} EARN 
GAAP earnings before
extraordinary items and discontinued operations. Compustat data item 18. 
CFO 
Cash flow from
operations. Compustat data item 308. 
EVA 
Economic Value Added. NOPAT less a charge for all capital based
on Stern Stewart standard adjustments.
A computed Compustat concept. 


SALES 
Sales revenue. Compustat data item 12. 
COGS 
Cost of goods sold. Compustat
data item 41. 
SGA 
Selling, general, and
administrative expenses.
Compustat data item 189. 
DP 
Depreciation and
amortization. Compustat data item 14. 
XINT 
Interest expense. Compustat
data item 15. 
NOPI 
Nonoperating income
(expense). Compustat data item 61. 
SPI 
Special items representing
unusual or infrequent items reported before income taxes. Compustat
data item 17. 
TXT 
Income tax expense. Compustat
data item 16. 
MII 
Minority interest. Compustat data item 49. 


DP 
Depreciation and
amortization. Compustat data item 125. 
XIDO 
Extraordinary items and
discontinued operations. Compustat
data item 124. 
TXD 
Deferred taxes. Compustat data item 126. 
ESUB 
Equity in subsidiary
earnings. Compustat data item 106. 
SPPIV 
Loss (gain) from sale of
property, plant, and equipment and sale of investments. Compustat data item 106. 
RECCH 
Change in accounts
receivable. Compustat data item 302. 
INVCH 
Change in inventory. Compustat data item 303. 
APALCH 
Change in accounts payable
and accrued liabilities. Compustat
data item 304. 
TXACH 
Change in accrued income
taxes. Compustat data item 305. 
MISC 
Other noncash operating
accruals. Compustat data items 207 and
307. 


SSDT 
Increase in deferred income
taxes. A concept created within
Compustat for a standard Stern Stewart adjustment. 
SSLIFO 
Increase in LIFO
reserve. A concept created within
Compustat for a standard Stern Stewart adjustment. 
SSBD 
Increase in bad debt
reserve. A concept created within
Compustat for a standard Stern Stewart adjustment. 
SSGWA 
Goodwill amortization. Compustat data item 65. A standard Stern Stewart adjustment. 
SSCM 
Capitalized marketing
expense. A concept created within
Compustat for a standard Stern Stewart adjustment. 
SSCRD 
Capitalized research and
development expense. A concept created
within Compustat for a standard Stern Stewart adjustment. 
SSPEN 
Pension and retirement
expense (Compustat item 43) less pension plansservice cost (Compustat item
331). A concept created within Compustat for a standard Stern Stewart
adjustment. 
SSUAT 
Aftertax unusual
items. A concept created within
Compustat for a standard Stern Stewart adjustment. 
SSIAT 
Aftertax
interestexpense. A concept created
within Compustat for a standard Stern Stewart adjustment. 
SSCHG 
Capital charge. This represents a charge at the firm’s
weighted average cost of capital on all invested capital, both debt and
equity. This concept is created within
Compustat for a standard Stern Stewart adjustment. 
^{2 }All amounts (except the number of observations) are
expressed in millions of US dollars.
^{3} These statistics are based on all firmyear
observations, regardless of whether the firmyear observation includes a
reported amount for the respective variable.
Thus, the reported statistics may be influenced by the assumption that
missing values are set to zero.
^{4} These statistics are based on only firmyear
observations in which a nonzero value is reported for the variable.
^{5} MII is also an EVA adjustment (with descriptive
statistics in Panel A above).
Table 2
Value relevance of summary performance measures (Panel A) and of
individual earnings components compared with their combined set (Panel B)^{1}
(7)
Panel A: Value relevance of summary measures
Variable 
R^{2} 
EARN 
7.88 
CFO 
1.78 
EVA 
3.72 
Component 
^{} (% of ) 
Ftest ^{2} (pvalue) 
SALES 

COGS 
0.55 5.1% 
26.66 (0.000) 
XSGA 
0.95 8.9% 
46.31 (0.000) 
DP 
0.00 0.0% 
0.07 (0.930) 
XINT 
0.13 1.2% 
8.22 (0.000) 
NOPI 
0.29 2.7% 
13.99 (0.000) 
SPI 
0.97 9.1% 
47.23 (0.000) 
TXT 
6.00 56.1% 
308.67 (0.000) 
MII 
0.22 2.1% 
10.57 (0.000) 
SUM OF ABOVE COMPONENTS 
10.66 99.6% 
TEST FOR JOINT SYNERGY (0.90) 
COMBINED SET 
10.70 
^{ }
^{1 }See Table 1 for variable definitions.
^{2} Ftest for the null
hypothesis: R^{2} = 0. Test for synergy described in text.
Table 3
Incremental value relevance of earnings revenue and expense components
compared with their combined set^{1,2}
Component i omitted 

(% R^{2} Lost) 
Partial F test^{3} (pvalue) 
SALES 
188.80 (0.000) 

COGS 
7.45 
30.4% 
175.32 (0.000) 
XSGA 
8.88 
98.33 (0.000) 

DP 
10.34 
0.36 3.4% 
19.13 (0.000) 
XINT 
9.64 
1.06 9.9% 
56.92 (0.000) 
NOPI 
10.47 
0.23 2.1% 
12.10 (0.000) 
SPI 
10.15 
0.55 5.1% 
29.72 (0.000) 
TXT 
10.67 
0.03 0.3% 
0.97 (0.377) 
MII 
10.65 
0.05 0.5% 
2.37 (0.094) 
^{1 }See Table 1 for variable definitions.
^{2} Incremental variation
explained, , is
the difference in R^{2} between the regression (7) for the combined set
of components (R^{2} = 10.70%) and the same regression omitting a
single component. The “% R^{2}
Lost” is the ratio of the reduction in R^{2} to the R^{2} for
the combined set of components (10.70%).
^{3} Partial Ftest for the null
hypothesis: = 0.
Table 4
Individual component synergies of earnings revenue and expense
components^{1,2}
Component 
(1) Separate Value Relevance 
Rank 
(2) Incremental Value Relevance 
Rank 
(3) Individual Component Synergy (col 2 –col 1) (pvalue)^{3} 
Rank 
SALES 
2 
3.50 
1 
1.94 (0.002) 
2 

COGS 
0.55 
5 
3.25 
2 
2.70 (0.000) 
1 
XSGA 
0.95 
4 
1.82 
3 
0.87 (0.003) 
4 
DP 
0.00 
9 
0.36 
6 
0.36 (0.004) 
5 
XINT 
0.13 
8 
1.06 
4 
0.93 (0.000) 
3 
NOPI 
0.29 
6 
0.23 
7 
0.06 (0.150) 
6 
SPI 
0.97 
3 
0.55 
5 
0.42 (0.002) 
8 
TXT 
6.00 
1 
0.03 
9 
5.97 (0.000) 
9 
MII 
0.22 
7 
0.05 
8 
0.17 (0.034) 
7 
^{1 }See Table 1 for variable definitions.
^{2} Individual (incremental)
value relevance in column 1 (2) excludes (includes) synergy. Thus, the individual component synergy of
each component with the remaining
components is the difference in R^{2},
i.e., column 2 – column 1. A positive
(negative) difference indicates positive (negative) synergy.
^{3
}Test for
synergy is described in text.
Table 5
Value relevance of CFO and individual earnings noncash accrual
components
compared with their combined set^{1}
(7)
Component 
^{} (% of ) 
Ftest ^{2} (pvalue) 

CFO 


DP 
0.02 0.2% 
1.14 (0.319) 

XIDO 
0.08 0.7% 
3.73 (0.025) 

TXD 
0.17 1.5% 
8.12 (0.000) 

ESUB 
0.08 0.7% 
3.73 (0.025) 

SPPIV 
0.10 0.9% 
4.61 (0.010) 

RECCH 
2.51 22.47% 
124.24 (0.000) 

INVCH 
0.68 6.08% 
33.00 (0.000) 

APALCH 
2.47 22.1% 
122.66 (0.000) 

TXACH 
1.84 16.5% 
90.83 (0.000) 

MISC 
0.09 0.9% 
4.73 (0.009) 

SUM OF ABOVE COMPONENTS 
9.82 87.9% 
TEST FOR JOINT SYNERGY 

COMBINED SET 
11.17 
^{1 }See Table 1 for variable definitions.
^{2} Ftest for the null
hypothesis: R^{2} = 0. Test for synergy described in text.
Table 6
Incremental value relevance of CFO and earnings noncash accrual
components compared with their combined set^{1,2}
Component i omitted 

(% R^{2} Lost) 
Partial F test^{3} (pvalue) 
CFO 
5.54 49.6% 
300.54 (0.000) 

DP 
10.79 
3.4% 
20.41 (0.000) 
XIDO 
11.07 
5.33 (0.005) 

TXD 
10.33 
0.84 7.5% 
45.43 (0.000) 
ESUB 
11.08 
0.09 0.8% 
4.89 (0.008) 
SPPIV 
10.91 
0.26 2.3% 
14.16 (0.000) 
RECCH 
6.49 
4.68 41.9% 
254.0 (0.000) 
INVCH 
8.11 
3.06 27.4% 
165.97 (0.000) 
APALCH 
10.11 
1.16 10.4% 
57.59 (0.000) 
TXACH 
10.87 
0.30 2.7% 
16.16 (0.000) 
MISC 
8.65 
2.52 22.6% 
136.59 (0.000) 
^{1 }See Table 1 for variable definitions.
^{2} Incremental variation
explained, , is
the difference in R^{2} between the regression (7) for the combined set
of components (R^{2} = 11.17%) and the same regression omitting a
single component. The “% R^{2}
Lost” is the ratio of the reduction in R^{2} to the R^{2} for
the combined set of components (11.17%).
^{3} Partial Ftest for the null
hypothesis: = 0.
Table 7
Individual component synergies of CFO and noncash earnings accrual
components^{1,2}
Component 
(1) Separate Value Relevance 
Rank 
(2) Incremental Value Relevance 
Rank 
(3) Individual Component Synergy (col 2 –col 1) (pvalue)^{3} 
Rank 
CFO 
4 
5.54 
1 
3.76 (0.000) 
1 

DP 
0.02 
10 
0.38 
7 
0.36 (0.061) 
6 
XIDO 
0.08 
8 
0.10 
10 
0.02 (0.840) 
8 
TXD 
0.17 
6 
0.84 
6 
0.67 (0.006) 
5 
ESUB 
0.08 
8 
0.09 
11 
0.01 (0.200) 
9 
SPPIV 
0.10 
7 
0.26 
9 
0.16 (0.001) 
7 
RECCH 
2.51 
1 
4.68 
2 
2.17 (0.000) 
4 
INVCH 
0.68 
5 
3.06 
3 
2.38 (0.000) 
3 
APALCH 
2.47 
2 
1.16 
5 
1.31 (0.000) 
10 
TXACH 
1.84 
3 
0.30 
8 
1.54 (0.000) 
11 
MISC 
0.09 
11 
2.52 
4 
2.43 (0.000) 
2 
^{1 }See Table 1 for variable definitions.
^{2} Individual (incremental)
value relevance in column 1 (2) excludes (includes) synergy. Thus, the individual component synergy of
each component with the remaining
components is the difference in R^{2},
i.e., column 2 – column 1. A positive
(negative) difference indicates positive (negative) synergy.
^{3
}Test for
synergy is described in text.
Table 8
Value relevance of individual EVA^{®} components compared with
their combined set^{1}
(7)
Component 
Adj. R^{2} (% of ) 
Ftest ^{2} (pvalue) 
SALES 

COGS 
0.55 4.5% 
26.66 (0.000) 
XSGA 
0.95 8.0% 
46.31 (0.000) 
DP 
0.00 0.0% 
0.07 (0.930) 
XINT 
0.13 1.1% 
8.22 (0.008) 
NOPI 
0.29 2.3% 
13.99 (0.000) 
SPI 
0.97 8.2% 
47.23 (0.000) 
TXT 
6.00 49.1% 
308.67 (0.000) 
SSDT 
0.22 1.8% 
10.60 (0.000) 
SSLIFO 
0.01 0.1% 
0.15 (0.857) 
SSBD 
0.10 0.73% 
4.64 (0.010) 
SSGWA 
0.01 0.1% 
0.66 (0.519) 
SSCM 
0.02 0.2% 
1.06 (0.347) 
SSCRD 
0.98 8.0% 
47.17 (0.000) 
SSPEN 
0.14 1.1% 
5.81 (0.003) 
SSUAT 
0.60 4.9% 
29.34 (0.000) 
SSIAT 
0.32 
15.54 (0.000) 
SSCHG 
1.15 9.4% 
56.16 (0.000) 
SUM
OF ABOVE COMPONENTS 
14.00 115.3% 
TEST FOR JOINT SYNERGY
4.03
(0.044) 
COMBINED
SET 
^{1 }See Table 1 for variable definitions.
^{2} Ftest for the null
hypothesis: R^{2} = 0. Test for synergy described in text.
Table 9
Incremental value relevance of EVA^{®} components
compared with their combined set^{1,2}
Component i omitted 

(% R^{2} Lost) 
Partial Ftest^{3} (pvalue) 
SALES 
2.65 21.7% 
145.31 (0.000) 

COGS 
9.71 
20.5% 
137.55 (0.000) 
XSGA 
10.71 
82.52 (0.000) 

DP 
11.77 
0.45
3.7% 
24.41 (0.000) 
XINT 
12.02 
0.20
1.6% 
10.54 (0.000) 
NOPI 
12.11 
0.11
0.9% 
6.00 (0.003) 
SPI 
11.85 
0.37
3.0% 
19.95 (0.000) 
TXT 
12.19 
0.03
0.2% 
1.71 (0.180) 
SSDT 
11.67 
0.55 4.5% 
(0.000) 
SSLIFO 
12.21 
0.01 0.1% 
0.43
(0.652) 
SSBD 
12.21 
0.01 0.1% 
0.62
(0.540) 
SSGWA 
12.19 
0.03 0.2% 
1.34
(0.261) 
SSCM 
12.19 
0.03 0.2% 
1.02
(0.362) 
SSCRD 
11.64 
0.58 4.7% 
31.40
(0.000) 
SSPEN 
12.16 
0.06 0.5% 
2.86
(0.057) 
SSUAT 
12.18 
0.04 0.3% 
2.19
(0.113) 
SSIAT 
12.21 
0.01 0.1% 
0.38
(0.682) 
SSCHG 
12.04 
0.18 1.5% 
9.79
(0.000) 
^{1 }See Table 1 for variable definitions.
^{2} Incremental variation
explained, , is
the difference in R^{2} between the regression (7) for the combined set
of components (R^{2} = 12.22%) and the same regression omitting a
single component. The “% R^{2}
Lost” is the ratio of the reduction in R^{2} to the R^{2} for
the combined set of components (12.22%).
^{3} Partial Ftest for the null
hypothesis: = 0.
Table 10
Individual component synergies of EVA^{®} components^{1,2}
Component 
(1) Separate Value Relevance 
Rank 
(2) Incremental Value Relevance 
Rank 
(3) Individual Component Synergy (col 2 –col 1) (pvalue)^{3} 
Rank 
SALES 
2 
2.65 
1 
1.09 (0.035) 
2 

COGS 
0.55 
8 
2.51 
2 
1.96 (0.000) 
1 
XSGA 
0.95 
6 
1.51 
3 
0.56 (0.229) 
3 
DP 
0.00 
18 
0.45 
6 
0.45 (0.001) 
4 
XINT 
0.13 
13 
0.20 
8 
0.07 (0.592) 
6 
NOPI 
0.28 
10 
0.11 
10 
0.17 (0.008) 
12 
SPI 
0.97 
5 
0.37 
7 
0.60 (0.002) 
16 
TXT 
6.00 
1 
0.03 
13 
5.97 (0.000) 
18 
SSDT 
0.22 
11 
0.55 
5 
0.33 (0.072) 
5 
SSLIFO 
0.01 
16 
0.01 
16 
0.00 (0.825) 
9 
SSBD 
0.10 
14 
0.01 
16 
0.09 (0.114) 
11 
SSGWA 
0.01 
16 
0.03 
13 
0.02 (0.586) 
7 
SSCM 
0.02 
15 
0.03 
13 
0.01 (0.897) 
8 
SSCRD 
0.98 
4 
0.58 
4 
0.40 (0.004) 
14 
SSPEN 
0.14 
12 
0.06 
11 
0.08 (0.153) 
10 
SSUAT 
0.60 
7 
0.04 
12 
0.56 (0.004) 
15 
SSIAT 
0.32 
9 
0.01 
16 
0.31 (0.019) 
13 
SSCHG 
1.15 
3 
0.18 
9 
0.97 (0.000) 
17 
^{1 }See Table 1 for variable definitions.
^{2} Individual (incremental)
value relevance in column 1 (2) excludes (includes) synergy. Thus, the individual component synergy of
each component with the remaining
components is the difference in R^{2},
i.e., column 2 – column 1. A positive
(negative) difference indicates positive (negative) synergy.
^{3
}Test for
synergy is described in text.
Figure 1
Negative and positive joint synergy
Negative Joint Synergy
Variation in Y
Variation in Y
Figure 2
=
Earnings
EVA adjustments to earnings expressed as
revenues and expenses:
+ Increase in deferred income taxes (SSDT)
+ Increase in LIFO reserve (SSLIFO)
+ Increase in bad debt reserve (SSBD)
+ Goodwill amortization (SSGWA)
+ Capitalized marketing expenses (SSCM)
+ Capitalized research and development expenses (SSCRD)
+ Pension expense (SSPEN)
+ Aftertax unusual income (SSUAT)
+ Aftertax net interest expense (SSIAT)
= Net Operating Profits After Tax (NOPAT)
 Capital charge measured by Stern Stewart & Co. (SSCHG)
= EVA^{®}
Noncash accruals that reconcile cash flow
from operations with earnings:
Cash Flow From Operations (CFO)
+ Depreciation and amortization (DP)
+ Extraordinary items and discontinued operations (XIDO)
+ Deferred taxes (TXD)
+ Equity in net loss (earnings) of subsidiary (ESUB)
+ Loss (Gain) from sale of property, plant, and equipment and sale of investments (SPPIV)
+ Change in accounts receivable (RECCH)
+ Change in inventory (INVCH)
+ Change in accounts payable and accrued liabilities (APALCH)
+ Change in accrued income taxes (TXACH)
+ Other operating accruals (MISC)
=
Earnings
^{ }
^{1} See Table 1 for
definitions of components.
^{1} See Table 1 for
definitions of components.
^{1 }See Table 1 for
definitions of components.
Figure 6
^{1 }See Table 1 for
definitions of components.
Figure 7
^{1 }Stern Stewart
EVA adjustments beyond income statement line items. See Table 1 for definitions of components.
^{1 }Stern Stewart
EVA adjustments beyond income statement line items. See Table 1 for definitions of components.
[1] It may at first seem counterintuitive that synergy can be both positive and negative. Occurrences of negative synergy appear quite frequently in a wide variety of contexts, including the life sciences, philosophy and business. Negative synergy is sometimes referred to as dysergy.
[2] For example, jointly considering sales revenues and cost of goods sold may be more informative than considering each component individually. Similarly, the information content of cash from operations could be embellished by jointly considering changes in accounts receivable, inventories and payables.
[3] Here we focus on valuerelevance. Synergies similarly could be examined among accounting measures used in making audit or loan decisions. Applicability also extends to other fields, for example, synergies among determinants of crop yields, consumption expenditures, and healthcare outcomes.
[4] This finding of insignificant joint synergy for components of “bottom line” (aftertax) income contrasts with a finding of positive joint synergy for components of pretax income. These different outcomes are revealed later to be attributable to significant negative individual synergy for the income tax expense component.
[5] Specifically, incremental value relevance is shown to include individual synergies among components, whereas the separate value relevance of a component does not. We measure incremental value relevance as the change in variation explained from omitting a predictor from the collective set (see Section 3.3 below).
[6] Put another way, income tax expense exhibits large relative value relevance because it is a close proxy for net income. Income tax expense exhibits small incremental value relevance because it largely substitutes for the other income statement components that comprise pretax income, thereby interacting (with negative individual synergy) to diminish their contributions to value relevance.
[7] For definitions of incremental and relative value relevance (information content) and a discussion of related statistical tests, see Biddle, Seow, and Siegel (1995). For comprehensive reviews of prior value relevance research, see Holthausen and Watts (2001) and Kothari (2001), and related discussions by Barth, Beaver and Landsman (2001) and Lee (2001).
[8] It is of interest to note that these synergies arise in the linear additive form common to accounting metrics, in the absence of multiplicative interactions, and can arise among all predictors, not just specific multiplicative pairs.
[9] Francis, Schipper, and Vincent (2002) provide recent evidence that investors utilize information conveyed by the set of separate earnings components beyond that conveyed in their restricted sum.
[10] It may at first appear that we are describing the familiar incremental value relevance (information content) setting used extensively in prior research, and to a degree this is true. Prior studies of incremental value relevance have often applied models with unrestricted coefficients, at least on incremental variables. However, restrictions are sometimes present, as when assessing incremental information content beyond earnings, as the earnings summary measure itself restricts all of its components to have equal coefficients. Also, one should not infer that incremental value relevance formulations are unrestricted, and relative value relevance formulations are restricted. Rather, it is perfectly reasonable to consider either incremental or relative value relevance in an unrestricted (component set) or restricted (definitional) setting, as detailed below.
[11] In SAS, this measure of incremental value relevance is termed “semipartial correlation.” Collins, Maydew, and Weiss (1997) assess its statistical significance using a method applicable to variation explained, but as we show below, a different test is required.
[12] Our definition of negative synergy closely parallels the Webster dictionary definition where synergy is defined as the difference between the combined effect and the sum of individual effects resulting from the interaction of a group of humans, agents or forces. We define individual synergy as the difference between a predictor’s individual contribution to variation explained and its contribution to the set as revealed when the predictor is removed from the set. Thus, unlike some definitions of synergy that only allow nonnegative values, our definition allows a variable to have negative synergy.
[13] More generally, with more than two predictors individual synergies can differ in sign from each other, as illustrated below.
[14] It may at first be surprising that components of summary accounting measures can exhibit positive synergy since prior studies have documented negative synergies (Bowen, Burgstahler, and Daley (1987, p. 727) and Biddle, Seow, and Siegel (1995, p. 4)).
[15] We do not present results for EVA components with earnings expressed as noncash accruals because Stern Stewart’s EVA components are described and implemented as adjustments to accrual components of earnings, rather than to noncash accruals.
[16] There are actually 11 EVA adjustments, but one, MII, adds back minority interest income to earnings, thus canceling the MII earnings component. Descriptive statistics are as reported in Panel A of Table 1.
[17] In creating this EVA measure, Stern Stewart follows the socalled financing approach building from earnings, rather than the operating approach building from revenues. This is the same version of EVA prepared by Stern Stewart for published rankings such as those appearing in Fortune magazine, and it is the most comprehensive definition that can be gathered for a large sample of firms. It may not, however, be the definition of EVA provided by Stern Stewart to clients that adopt EVA for management incentive purposes. Stern Stewart has identified over 160 potential adjustments to GAAP earnings. Rarely, however, does any one company make more than 15 adjustments to create its ‘tailored’ EVA (Ehrbar 1998, pp. 164166.).
[18] The S&P 1500 database combines large firms from the S&P 500, midsize firms from the S&P Midcap 400, and smaller firms from the S&P Smallcap 600 databases.
[19] For example, see Hayn (1995), Burgstahler and Dichev (1997), Collins, Pincus, and Xie (1999), Biddle, Bowen, and Wallace (1997) for more complete descriptions of similar empirical models. Lo (2004) lends support to scaling.
[20] In the perhaps more familiar context of forward stepwise regression, TXT enters first among revenue and expense components, but later drops out as other components enter – a result attributable to its negative individual synergy. Consistent with these findings, it is of interest to note that the set of components comprising “pretax” earnings (omitting TXT, not reported separately) exhibit = 4.62% and = 10.68%, and thus positive joint synergy, with a high level of statistical significance (Test for Joint Synergy = 49.65, p = 0.000).
[21] The finding that research and development costs are value relevant is consistent with the findings of prior research (e.g., Sougiannis 1994; Lev and Sougiannis 1996).
[22] That earnings accruals themselves exhibit insignificant joint synergy (Table 2 above) implies that the negative joint synergy arises primarily from interactions of EVA adjustments with other predictors.
[23] This finding further explains why aftertax earnings components exhibit insignificant joint synergy whereas pretax earnings components exhibit positive joint synergy.
[24] Said another way, these adjustments attempt to correct incentive problems with EVA rather than correct “distortions” in investor communications by earnings as their marketing might suggest.
[25] These sources are illustrated by earnings accrual components SALES and COGS, which exhibit similar incremental value relevance in Table 4. Separate value relevance is a larger source of incremental value relevance for SALES than for COGS, whereas the incremental value relevance of COGS derives primarily from individual synergy.
[26] We also conducted the above tests by year. Keeping in mind that sample sizes are much smaller and that they are not entirely independent due to the use of lagged observations, the results by year are broadly similar to those for the pooled sample. Noncash earnings accruals beyond CFO exhibit significant positive joint synergy (p < 0.10) in three years (1989, 1994, and 1998), with no years exhibiting significant negative joint synergy. EVA components exhibit significant negative joint synergy in six years (198991, 1993, 1996, and 1999), with no years exhibiting significant positive synergy. Earnings components exhibit significant negative synergy in four years, 198993, with no years exhibiting statistically significant positive synergy. Pretax earnings components exhibit significant positive joint synergy in five years (1989, 199495, 199798), with no years exhibiting significant negative joint synergy. These latter two findings are consistent with the significant negative individual synergy noted above for TXT.
[27] Further insights into the origins of synergy are
obtained by considering special conditions of expression (4a). For instance, consider the condition where
both predictors and are positively correlated
with Y, i.e., , > 0. Applying expressions (2), (3) and (4a), it
follows that negative synergy can obtain only when > 0, in other
words, when and are themselves
positively correlated. Further notice
that because is the product of
squared terms of less than one, it will be small, and in most cases smaller in
absolute value than the other term in (4a), ; when so, the sign of synergy for this condition depends solely on the sign of .