Australian evidence concerning the information content of economic value-added.

Abstract:

Pooled time-series, cross-sectional data on 110 Australian companies over the period 1992-1998 is employed to examine whether the trademarked variant of residual income known as economic value-added (EVA[R]) is more highly associated with stock returns than other commonly-used accounting-based measures. These other measures of internal and external performance include earnings, net cash flow and residual income. Three alternative formulations for pooling data are also employed in the analysis, namely, the common-effects, fixed-effects and random-effects models, with the fixed-effects approach found to be the most empirically appropriate. Relative information content tests reveal returns to be more closely associated with EVA[R] than residual income, earnings and net cash flow, respectively. An analysis of the components' of EVA[R] confirms that the GAAP-related adjustments' most closely associated with EVA[R] are significant at the margin in explaining stock returns.

Keywords:

VALUE-RELEVANCE; RELATIVE AND INCREMENTAL INFORMATION CONTENT, ECONOMIC- VALUE ADDED; RESIDUAL INCOME.

1. Introduction

In recent years, Economic Value-Added (EVA[R])--a trademarked variant of residual income (net operating profits less a charge for the opportunity cost of invested capital)--has become increasingly popularised as a tool for financial decision-making. Its developer and principal advocate, US-based business consultants Stern Stewart, argue that 'earnings, earnings per share, and earnings growth are misleading measures of corporate performance [and that] the best practical periodic performance measure is economic value-added' (Stewart 1991, p. 66). Similarly, Stewart (1991, p. 66) contends that EVA[R] '... is the financial performance measure that comes closer than any other to capturing the true economic profit of an enterprise [and] is the performance measure most directly linked to the creation of shareholder wealth over time'. As a means of providing support for these claims, Stern Stewart has commissioned several in-house studies to link changes in EVA[R] with changes in shareholder wealth. For instance, Stewart (1994, p. 75) provides evidence that:

   EVA stands well out from the crowd as the single best measure of
   wealth creation on a contemporaneous basis [and] is almost 50%
   better than its closest accounting-based competitor [including EPS,
   ROE and ROI] in explaining changes in shareholder wealth.

Support for EVA[R] has also been forthcoming from other sources. Fortune has called it 'today's hottest financial idea', 'The Real Key to Creating Wealth' (Anonymous 1993) and 'A New Way to Find Bargains' (Topkis 1996). And Peter Drucker in the Harvard Business Review suggested that EVA's[R] growing popularity reflected the demands of the information age for a measure of 'total factor productivity' (Drucker 1991). McClenahen (1998) similarly observes that 'traditional corporate performance measures are being relegated to second-class status as metrics such as EVA[R] become management's primary tools'. Finally, there has been the widespread adoption of EVA[R] by security analysts since 'instead of using a dividend discount approach, these models measure value from the point of view of the firms' capacity for ongoing wealth creation rather than simply wealth distribution' (Herzberg 1998, p. 45) (emphasis added).

In response to these claims, an emerging literature has addressed the empirical issue as to whether EVA[R] is more highly associated with stock returns and firm values than other accounting-based figures. For example, Biddle, Bowen and Wallace (1997) used relative and incremental information tests to examine whether stock returns were more highly associated with EVA [R], residual income or cash flow from operations. They concluded that while 'for some firms EVA[R] may be an effective tool for internal decision making, performance measurement, and incentive compensation, it does not dominate earnings in its association with stock market returns' (p. 333).

Chen and Dodd (1997) likewise examined different dimensions of the EVA [R] system and concluded: '... not a single EVA[R] measure [annualised EVA [R] return, average EVA[R] per share, change in standardised EVA[R] and average return on capital] was able to account for more than 26% of the variation in stock return'. Lehn and Makhija (1997) Rogerson (1997) and Biddle, Bowen and Wallace (1997) reached similar conclusions. Clinton and Chen (1998) also compared share prices and returns to residual cash flow, economic value-added and other traditional measures, and recommended that companies using EVA[R] consider residual cash flow as an alternative.

However, Bao and Bao (1998, p. 262) in an analysis of price levels and firm valuations concluded that the 'results are not consistent for earnings and abnormal economic earnings, but are consistent for value-added, that is, value-added is significant in both levels and changes deflated by price analyses'. Similarly, Uyemura, Kantor and Petit (1996) demonstrated that EVA[R] has a high correlation with market value added (the difference between the firm's value and cumulative investor capital) and thereby stock price, while O'Byrne (1996) estimated that changes in EVA[R] explain more variation in long-term stock returns than changes in earnings. Finally, and from a stock selection perspective, Herzberg (1998, p. 52) concluded that the residual income valuation model (including EVA[R]) 'appears to have been very effective in uncovering firms whose stock is underpriced when considered in conjunction with expectations for strong earnings and growth'. Nevertheless, the bulk of empirical evidence indicates that the superiority of EVA[R] over earnings (as variously defined) has not been established.

However, when examining existing research in this area, two salient points emerge. First and foremost, and notwithstanding that EVA[R] figures are readily available and promoted in the UK, Australia, Canada, Brazil, Germany, Mexico, Turkey and France (see Stern Stewart 1999), no empirical studies of this type (as far as the authors are aware) have been conducted outside the United States. This is despite several international companies adopting EVA[R] for performance measure and/or incentive compensation packages. There is an obvious requirement to examine the usefulness of EVA[R] vis-a-vis traditional financial statement measures in an alternative institutional milieu.

Second, there has been an emphasis in previous empirical work in this area on either a cross-section of companies or limited pooled time-series, cross-sectional data. For example, Bao and Bao (1998) only employ a cross-section of 166 firms over the period 1992/93. Examination of extended time-series data would certainly permit greater empirical certainty on the usefulness of economic value-added. However, while data sets that combine time series and cross sections are increasingly common in financial analysis, modelling in these settings calls for some quite complex stochastic specifications. For example, past empirical studies have often employed pooled time-series, cross-sectional data without giving specific a priori justification for the choice of model formulation. More particularly, the simplest assumption of common-effects has usually been made. It is with these considerations in mind that the present study is undertaken.

The remainder of the paper is divided as follows. The second section briefly outlines the calculation of EVA[R] and discusses the empirical methodology employed. The results are dealt with in the third section. The paper ends with some brief concluding remarks.

2. Empirical Methodology

The calculation of EVA[R] consists of two separate but related steps. The primary adjustment is where a capital charge is subtracted from net operating profit after-tax. The capital charge is derived from multiplying the firm's overall financing cost, as reflected in the weighted average cost of capital by the amount of invested capital. Invested capital in turn is defined as total assets, net of non-interest bearing current liabilities. In this form, EVA[R] is essentially the same as residual income, though the latter measure is normally expressed as net income less a charge for the cost of equity capital (with the cost of debt already included in the calculation of net income). The second and more controversial step consists of a series of adjustments to GAAP-based numbers. These modifications to a company's conventional accounts may be meaningfully grouped as adjustments to research and development, deferred taxes, intangibles, depreciation, provisions for warranties and bad debts, restructuring changes, and macroeconomic conditions (see Stewart 1991, 1994; O'Hanlon & Peasnell 1998, 2000; Young 1999; Stern Stewart 1999; Worthington & West 2001 for a detailed discussion of these accounting adjustments).

The analysis contained in this paper consists of two closely related empirical questions. The first question relates to the purported dominance of EVA[R] over both residual income and the conventional accounting performance measures of earnings before extraordinary items and net cash flow from operations in explaining contemporaneous stock returns. The second empirical question concerns those components unique to EVA[R] that help explain these contemporaneous stock returns beyond that explained by residual income, earnings before extraordinary items and net cash flow from operations. Assuming that equity markets are (semi-strong form) efficient, stock returns may be used to compare the information content (or value-relevance) of these competing accounting-based performance measures (Bowen, Burgstahler & Daley 1987; Jennings 1990; Easton & Harris 1991; Ali & Pope 1995; Biddle, Seow & Siegel 1995). Both relative and incremental information content comparisons are made. In terms of specific studies, the approach selected in the current study is most consistent with that used by Biddle, Bowen and Wallace (1997) and Bao and Bao (1998).

2.1 Linkages Between EVA and EVA Components

The first methodological requirement is to describe the linkages that exist between the competing measures of firm performance; namely, earnings before extraordinary items (ERN), net cash flow from operations (NCF), residual income (RI) and economic value-added (EVA). Starting with ERN as the most basic indicator of firm value we have:

(1) ER[N.sub.t] = NC[F.sub.t] + AC[C.sub.t]

where ERN is the sum of net cash flow from operations (NCF) and accruals (ACC) with the t sub-script denoting the time-period. ACC is defined as total accruals relating to operating activities and is composed of depreciation, amortisation, changes in non-cash current assets, changes in current liabilities, and changes in the non-current portion of deterred taxes. Net operating profit after tax (PRF) is a closely related indicator of current and future firm performance and is calculated by adding after-tax interest expense (ATI) to ERN:

(2) PR[F.sub.t] = ER[N.sub.t] + AT[I.sub.t] = NC[F.sub.t] + AC[C.sub.t] + AT[I.sub.t]

As indicated, the most significant difference between ERN and PRF is that the later separates operating activities from financing activities by including the after-tax effect of debt financing (interest expense). As a measure of operating profit, no allowance is therefore made in (2) for the financing activities (both debt and equity) of the firm. One measure that does is residual income (RI) where operating performance is reduced by a net charge for the cost of all debt and equity capital employed:

(3) R[I.sub.t] = PR[F.sub.t] - (WAC[C.sub.t] x CA[P.sub.t-1]) = NC[F.sub.t] + AC[C.sub.t] + AT[I.sub.t] - C[C.sub.t]

where WACC is an estimate of the firm's weighted average cost of capital, and capital (CAP) is defined as assets (net of depreciation) invested in going-concern operating activities, or equivalently, contributed and retained debt and equity capital, at the beginning of the period. The product of the firm's WACC and the amount of contributed capital thereby forms a capital charge (CC) against which PRF is reduced to reflect the return required by the providers of debt and equity capital. A positive (negative) RI indicates profits in surplus (deficit) of that required by the suppliers of debt and equity capital and is associated with an increase (decrease) in shareholder wealth.

The primary point of departure for EVA from R/is the adjusting of both PRF and CAP for purported 'distortions' in the accounting model of performance. EVA- type adjustments are made to both accounting measures of operating profits (PRF), and accounting measures of capital (CAP). EVA thereby reflects adjustments to GAAP in terms of both operating and financing activities. Simplifying, EVA is thus determined by:

(4) EV[A.sub.t] = NC[F.sub.t] + AC[C.sub.t] + AT[I.sub.t] - C[C.sub.t] + AD[J.sub.t]

where the total EVA accounting adjustment (ADJ) is the net figure of adjustments to PRF (NCF + ACC + ATI) less the adjustment to capital in determining CC (WACC x CAP).

2.2 Specoqcation of 'Valuation' and 'Components' Models

The second methodological requirement is to specify the regression models used to: (i) calculate the relative and incremental content of the competing measures of firm performance; and (ii) calculate the relative and incremental content of the components of economic value-added (EVA[R]) itself. The first specification is referred to as the 'valuation model':

(5) ST[K.sub.it] = [b.sub.0] + [b.sub.1]NC[F.sub.it] + [b.sub.2]ER[N.sub.it] + [b.sub.3]R[I.sub.it] + [b.sub.4]EV[A.sub.it] + [e.sub.it]

The dependent variable in (5) is the compounded annual stock return (STK) for firm i in period t. A 12-month non-overlapping period ending three months following the firm's fiscal year end is chosen to allow time for information contained in the annual report to be impounded in market prices. The explanatory variables in the firm valuation model are net cash flows from operations (NCF), earnings before extraordinary items (ERN), residual income (RI) and economic value-added (EVA). The first three accounting measures are specified since they represent both components of EVA[R] and alternative measures of periodic performance. Selected descriptive statistics for these variables are given in table 1. Following the value-relevance literature on financial statement information, the hypothesis suggests positive coefficients for NCF, ERN, RI and EVA when specified as explanatory variables for stock returns, and the relative information content of these measures will be higher the more closely they approximate these returns.

This model is similar to that used in Biddle, Bowen and Wallace's (1997) EVA study save two respects. First, in the Biddle, Bowen and Wallace (1997) the independent variables are normalised by the lagged market value of equity to provide consistency with the lagged accounting values specified in an autoregressive model. In this study the independent variables are normalised by the number of outstanding shares with no requirement to normalise in the same manner due to the absence of lags. While both approaches are commonly used to reduce heteroskedasticity in firm-level data, White's (1980) heteroscedastic-consistent estimator is also employed, along with an equivalent correction for time-wise autocorrelation in the pooled time-series, cross-sectional least squares regression. Second, in Biddle, Bowen and Wallace (1997) the dependent variable is specified as market adjusted returns (each firm's 12-month compounded stock return less the 12-month compounded value-weighted market wide return) whereas in this analysis each firm's stock return remains unadjusted by the market return. This difference in specification maintains consistency with both the method of normalisation used for the independent variables and is largely nominal given that the regression sum of squares and the slope coefficients, which are the focus of this analysis, are unaffected.

The second specification examined is referred to as the 'components model':

(6) EV[A.sub.it] = [b.sub.0] + [b.sub.1] NC[F.sub.it] + [b.sub.2]AC[C.sub.it] +[b.sub.3]AT[I.sub.it] + [b.sub.4]C[C.sub.it] + [b.sub.5]AD[J.sub.it] + [e.sub.it]

This model is also estimated using a pooled time-series, cross-sectional least squares regression with corrections for heteroscedasticity and autocorrelation. The dependent variable is given as EVA. The independent variables are the five components of EVA: namely, net cash flows (NCF), operating accruals (ACC), after-tax interest (ATI), cost of capital (CC) and EVA accounting adjustments (ADJ). All variables are as previously defined and are equivalent to the left-hand side of equation (4). Descriptive statistics are provided in table 1. This set of explanatory variables are also normalised by the number of shares on issue. The first variable, NCF, is as previously defined. ACC is defined as earnings less net cash flow from operations (ERN--NCF). Accruals can either be positive or negative, but are usually negative (reflecting non-cash expenses such as depreciation and amortisation). The ex ante sign on the coefficient for accruals is thought to be positive when specified as an explanatory variable for EVA and stock returns. ATI is calculated as one minus the firm's tax rate (assumed to be 36%) multiplied by interest expense. A positive coefficient is hypothesised when EVA and stock returns are regressed against interest expense.

CC is defined as each firm's weighted-average cost of capital multiplied by the beginning of year capital (WACC x CAP). A negative coefficient is hypothesised. Finally ADJ reflects Stern Stewart's adjustments to earnings and capital, and is defined as economic value-added less residual income (EVA--RI). Given the fact that the direction of change for ADJ may vary across firms in the sample depending on both financing and operations (GAAP-related accounting adjustments can either be positive or negative), it is somewhat difficult to postulate the relationship between GAAP adjustments and EVA/stock returns. No a priori coefficient is hypothesised.

2.3 Data

The third methodological requirement is to specify the data sources for the financial statement numbers in (5) and (6) along with stock returns as the dependent variable. Three separate sources of data are used. Data for EVA[R] and its components is obtained directly from Stem Stewart's Australian EVA[R] Performance Rankings. These data contain EVA[R], the weighted average cost of capital (WACC), return on capital, net operating profit after-tax (PRF), capital (CAP) and average shareholder returns for Australia's 110 largest listed (non-financial) companies. The sample of firms consists of both adopters and non-adopters of the EVA[TM] Financial Management System over the period of 1992-1998. Financial statement data for ERN, ATI, RI, NCF, ACC and ADJ are collected from the Australian Stock Exchange's (ASX) Datadisk database and the Connect-4 Annual Report Collection database. Finally, share price data are obtained from the Australian Graduate School of Management's (AGSM) Share Price and Price Relative database (incorporating capitalisation adjustments and dividends). Descriptive statistics for these variables are detailed in table 1.

2.4 Choice of Pooling Technique

The fourth methodological requirement is to examine the different methods of pooling panel data. In the basic regression model, a simple assumption is that the parameters do not vary across sample observations. One advantage of pooled time series models is that it '... allows parameters to vary in some systematic and/or random way across partitions of the sample data, or even from observation to observation' (Judge, Hill, Griffiths, Lutkepohl & Lee 1988, p. 468). However, while pooling of data is used extensively in practice, most studies assume that financial relations, however defined, are homogeneous across firms (a common-effects model).

The two additional pooling models considered are the fixed-effects (or dummy variable) model and the random-effects (or error components) model. To start with, the fixed-effects model allows the differences in intercepts to be modeled using dummy variables, that is, fixed coefficients. Assuming we have i = 1, 2, ..., N cross-sectional observations, and t = 1, 2, ..., T time-series observations, the (i, t)th observation on the dummy variable model with which we are concerned can be written as:

(7) [y.sub.it] = [N.summation over (j = 1)] [[beta].sub.1j] [D.sub.jt] + [K.summation over(k = 2) [[beta].sub.k] [x.sub.kit] + [e.sub.it]

where [[beta].sub.1i] represents the intercept coefficient for the ith cross-sectional firm, [D.sub.jt], are dummy variables that take a value of unity for observations on firm j but will be 0 for observations on other firms, [[beta].sub.k] represent the slope coefficients that are common to all firms, [y.sub.it] is the dependent variable, [x.sub.kit] are the explanatory variables, and the [e.sub.it] are independent and identically distributed random variables with E[[e.sub.it]] = 0 and E[[e.sub.it.sup.2]] = [[sigma].sub.3.sup.2]. This specification is usually employed when specifying a different intercept coefficient for each cross-sectional unit can adequately capture differences in cross-sectional units. That is, cross-sectional identifiers explain changes from firm to firm (Judge et al. 1988).

An alternative to the fixed-effects model is a random-effects model that assumes that the coefficients are random variables drawn from some larger population:

(8) [y.sub.it] = [[beta].sub.1] + [K.summation over k = 2] [[beta].sub.k] [x.sub.kit] + [u.sub.i] + [e.sub.it]

where E[[u.sub.i]] = 0, E[[u.sub.i.sup.2]] = [[sigma].sub.u.sup.2], E[[u.sub.i][u.sub.j]] = 0 for i [not equal to] j, E[[u.sub.i][e.sub.it]] = 0 and all other variables are as previously defined. The structure of the model is such that, for a given firm, the correlation between any two disturbances in different time periods is the same, and unlike a first-order autoregressive model, does not decline as the disturbances become farther apart in time. Further, not only is the correlation constant over time, it is identical for all firms (Judge et al. 1988). The inference is that the results from this model may be generalised to the whole population from which the sample is taken.

In this manner, the distinction between the random and fixed-effects can be viewed as the distinction between conditional and unconditional inference (Judge et al. 1988, p. 491). With the fixed-effects model, inference is conditional on the firms in the sample, whereas the random-effects model is more appropriate when we are interested in (unconditional) inferences about a larger population. Bearing in mind the fact that the sample is nearly exhaustive of the Australian companies for which Stern Stewart calculate EVA, a reasonable assumption in most circumstances might be a fixed-effects formulation, however such assumptions should be tested empirically. This is especially important in studies of this type where the number of cross-sections (N) is relatively large and the number of time series (T) is relatively small. Under these conditions the results of the two models can differ significantly.

The procedures used to carry out tests between the models are as follows. Firstly, the model is estimated using common coefficients, and tested against the fixed and random-effects specifications using an F-test. The F ratio used for the test is:

(9) F(n-1, nT-n-K) = ([R.sup.2.sub.u])-([R.sup.2.sub.p])/(n-1)] / [(1-[R.sup.2.sub.u]/(nT-n-K)

where u indicates the unrestricted model (common-effects) and p indicates the pooled or restricted model with only a single overall constant term. Under the null hypothesis, the efficient estimator is pooled least squares (Greene 1993, p. 617). The second test is used to choose between a fixed or random-effects specification. This is accomplished using a Hausman test. Under this hypothesis, there are two sets of estimates; one of which is consistent under both the null and alternative hypothesis, and another that is consistent only under the null. The null hypothesis is that both the fixed and random specifications are consistent, whereas under the alternative the fixed effect model is, but the random-effects model is not. The test is based on a Wald criterion:

(10) W = [x.sup.2][K] = [[(b-[beta]].sup.2] / Var[b]-Var[[beta]]

which is asymptotically distributed as chi-squared with K degrees of freedom (Greene 1993, p. 613). From this, the preferred model is identified and used in the incremental and relative information content tests.

2.5 Relative and Incremental Information Content Tests

The 'valuation model' is estimated using a pooled time-series, cross-sectional least squares regression assuming cross-sectional heteroskedasticity and timewise autoregression (Greene 1993, p. 613). The first set of tests is joint hypothesis tests of equation (5), that NCF, ERN, EVA and RI have equal relative information content. To accomplish this, each of these variables is specified as the explanatory variable in separate univariate regressions with stock returns as the dependent variable (i.e. STK and NCF, STK and ERN, etc.) (Biddle, Bowen & Wallace 1997; Bao & Bao 1998). Comparisons of the R: of the regression results are made to determine which variable better explains variation in STK. Rejection of this hypothesis is viewed as evidence of a significant difference in the relative information content.

The second set of tests indicates whether one of these predictors of firm value provides value-relevance data beyond that provided by another measure. Rejection of this hypothesis is viewed as evidence of incremental information content. In these tests, each of the four explanatory variables in the valuation model is alternately paired with each other measure in a multivariate regression. As before, STK is specified as the dependent variable. For example, the incremental information content for EVA/ERN is obtained from a multivariate regression where both EVA and ERN are specified as explanatory variables. Taking the adjusted [R.sup.2] from this pairwise regression, and subtracting the individual [R.sup.2] for ERN obtained in the earlier univariate regression, yields the incremental information content of EVA over ERN.

Similar tests of relative and incremental information content are performed in the 'components model' (6), using the preferred pooling technique. The components in this instance are NCF, ACC, ATI, CC and ADJ. Additionally, in order to evaluate the sensitivity of these models to the specification of variables, two separate regressions are undertaken. These are identical in all respects except that in the first form all variables are expressed in levels (or undifferenced), while in the second the variables represent year-to-year changes in the variables (or differenced). Bao and Bao (1998) also evaluated the usefulness of value added measures using level and differenced variables. As an alternative, Biddle, Bowen and Wallace (1997, p. 309) specified the independent variables in levels along with a lagged value: 'it is in a more convenient form that allows the slope or 'response' coefficient to be observed directly (rather than being derived directly from separate coefficients on levels and changes)'.

3. Empirical Results

The first step in the analysis is to select the most appropriate pooling technique for both the 'valuation' (5) and 'components' (6) models. In the first instance, the explanatory variables are net cash flow (NCF), earnings before extraordinary items (ERN), residual income (RI) and economic value-added (EVA). In the second instance, the explanatory variables are net cash flows from operations (NCF), accruals (ACC), after-tax interest (ATI), the cost of capital (CC) and accounting adjustments (ADJ). The dependent variable in the first instance is the compounded annual stock return and in the second, economic value-added. An assumption of a linear relationship between these variables is made.

Table 2 presents the estimated coefficients, standard errors and t-statistics for both models, in both differenced and levels form, across the three alternative pooling techniques; namely, common, fixed and random-effects. In general, there is consistency in the signs on the estimated coefficients for both the valuation and components models across both the alternative panel data specifications and whether the regression employs levels or differenced variables. However, levels of significance do vary. For instance, the levels of significance are generally higher for the differenced regressions for the valuation model while the reverse holds for the levels regressions for the components model. Moreover, across the three alternative methods of pooling data [R.sup.2] is highest for the fixed-effects models and lowest for the models assuming common-effects. This reinforces the suggestion that at least some of the difference in information content found across past EVA studies may be attributable to differences in the chosen pooling technique.

As discussed, the significance of group effects (fixed or random) over the common-effects in the valuation model is tested using (9). The test for commoneffects in the undifferenced data [F = 8.438 ~ [F.sup..05.sub.109,656] rejects the null hypothesis that the efficient estimator is the unrestricted (common-effects model). Likewise, the test for common-effects in the differenced data [F= 5.170 ~ [F.sub..05.sub.109,656] also reflects the null hypothesis. Since both statistics are larger than the critical value we may reject the null hypothesis of no common effect (i.e. variation across cross-sections).

The next procedure (10) uses Hausman's test for fixed and random-effects. The underlying idea of the Hausman test is to compare two sets of estimates, one of which is consistent under both the null and alternative hypothesis, and another that is consistent only under the null hypothesis. Under the null hypothesis both the fixed and random specifications are consistent, whereas under the alternative the fixed effect model is, but the random-effects model is not. The Wald value calculated in the valuation model is 79.546 for the undifferenced data and 89.985 for the differenced data. Both of these are larger than the critical value of 9.48773 (chi-square at 5% level of significance), thus rejecting the null hypothesis. We may conclude that the fixed-effects specification is appropriate whether using levels (undifferenced) or differenced data. As the fixed-effects (or dummy variable) model is the preferred model it is used in the remainder of the analysis.

Table 3 presents the estimated coefficients, standard errors and t-statistics of the valuation model, for both differenced and levels, assuming a fixed-effects specification. The dependent variable is specified as compounded annual stock returns (with a lagged period of three months following fiscal year end) and the explanatory variables are variously specified as net cash flow, earnings before extraordinary items, residual income and economic value-added. Variance inflation factors (VIF) are also calculated using the [R.sup.2] for each independent variable when regressed on the remaining independent variables. As a rule of thumb, if the VIF of an independent variable exceeds 10, multicollinearity may be a problem. In the case of the regressors in the valuation model, the highest VIF is only 5.71 (ERN) while the highest VIF in the components model is 6.76 (ACC). These suggest that multicollinearity, while present, is not significant.

Table 3 indicates that all four accounting-based performance measures are positively associated with stock returns (except on four occasions net cash flow). These tables also show that only residual income (and on four occasions earnings and one occasion net cash flow) is significant in explaining stock returns over the period 1992-1998. Of the estimated 40 slope coefficients, only 14 are significant at the .10 level or lower and four are not in the predicted direction (NCF). This finding holds when a pairwise combination of performance measures is specified in the same regression. Table 3 shows that residual income is most significant by itself and when paired with net cash flow. Also earnings are significant by itself and when paired with net cash flow. The pairwise regression that most explains stock returns (STK) is EVA/RI (26.56%), EVA/NCF (25.99%), EVA/ERN (25.57%), RI/NCF (18.47%), ERN/RI(18.44%) and finally ERN/NCF (14.17%). As EVA is in the three pairwise regressions that best explain returns, there is already an indication that it is a highly significant explanatory factor.

The significance of these variables improves with the differenced variables as detailed in table 3, indicating that changes from year to year are important. The most significant explanatory pairwise combinations are in order of decreasing power are EVA/RI (41.91%), EVA/ERN (38.72%), EVA/NCF (38.17%), NCF/RI (23.72%), ERN/RI (22.97%) and ERN/NCF (17.80%). Again EVA has the most explanatory power, although it lacks significance. NCF is also periodically in the wrong direction, meaning that firms in the sample for the period under consideration may have experienced predominately negative cash flows, thus skewing the results and causing insignificance. Thus the market may not recognise EVA and net cash flows (NCF) as legitimate or reliable firm valuation measures. These results show that earnings are highly valued by the market, as is the long-established residual income concept. The summary results of these regressions in the form of relative and incremental information content tests are presented in tables 4 and 5.

Panel A of tables 4 and 5 indicates that there is a significant difference in relative information content between the accounting-based measures. The highest adjusted [R.sup.2] from the single coefficient regressions is shown on the left, with lower explanatory power in descending order to the right. The suggestion is that EVA better explains STK than RI, ERN and NCF alone, the explanatory power being slightly higher for differenced variables (table 5). EVA, however, does lack significance, whereas RI is highly significant. In terms of international comparisons, Biddle, Bowen and Wallace (1997) indicated that earnings (ERN) was more highly associated with stock returns than either RI or EVA, but that all three measures dominate net cash flow (NCF). Furthermore, the explanatory power of all four accounting-based measures is significantly higher than that found in a number of comparable studies. For example, Biddle, Bowen and Wallace (1997) estimated the relative information content of ERN, RI, EVA and NCF at 9.04, 6.24, 5.07 and 2.38% respectively.

The results in panel B of tables 4 and 5 provide incremental information content tests for the pairwise combinations of EVA, ERN, RI and NCF. For example, in table 4 EVA/ERN(11.15%) is equal to the information content of the pairwise comparison of EVA and ERN (25.57%) minus the information content of ERN (14.42%) from table 3. The pairwise combinations of EVA and ERN, NCF and RI indicate that explanatory power has increased by 11.15, 12.48 and 8.03% respectively over the EVA measure alone. A comparison with the incremental information tests contained in Bao and Bao (1998) for pooled data indicates that earnings have a zero impact on EVA alone, while residual income increases explanatory power by some 38%. Overall, the results indicate that EVA exhibits the largest relative information content among the measures, with RI (0.88%), NCF (0.31%) and ERN (-0.11%) providing only limited incremental information content beyond EVA. The most logical pairing of information variables in explaining stock returns is therefore composed of EVA and RI. These results persuasively support the claims made by Stern Stewart that EVA[R] outperforms other accounting-based performance measures in explaining stock returns.

The second phase of the study is to examine the components of EVA. These components are net cash flows from operations (NCF), accruals (ACC), after-tax interest (ATI), the cost of capital (CC) and accounting adjustments (ADJ). This part of the analysis addresses the empirical question of what component of EVA contributes most to variation in EVA, and hence explaining stock returns. Table 6 presents the results of the individual and pairwise regressions of the components of EVA, employing both levels and differenced data.

In table 6, the individual regressions show that two of the five variables (CC, ATI) are significant at the 0.01 level. The cost of capital (CC) is highly significant for all regressions, and interestingly, accruals (ACC) is never significant until included in the final regression. ADJ are only significant when paired with CC and ATI, and NCF is only significant when paired with CC. ATI losses significance when paired with CC, and CC loses significance itself in this pairwise regression, when compared to the others. The pairwise combinations show that CC/ADJ most explains EVA (78.08%), followed by ATI/ADJ (64.46%), CC/NCF (61.15%), CC/ATI (56.92%), ADJ/NCF (56.92%), CC/ACC (56.01%), ATI/NCF (55.72%), ACC/ADJ (51.34%), and ACC/NCF (56.92%). The final regression shows almost complete explanation (98.40%) with all variables highly significant and in the predicted direction.

Table 6 also shows that three out of the five variables are significant (ACC, ADJ and NCF) at the 0.01 level for the individual regressions. ATI is found never to be significant, even when included in the final regression with all of the variables included in the model. CC is significant when paired with ADJ and in the final regression. NCF loses significance when paired with ACC, compared to the other regressions. ACC loses significance when paired with ATI and ADJ, when paired with the other regressors. The pairwise regressions for the differenced variables show (in order of explanatory power) that ADJ/NCF (65.21%) better explains EVA than CC/ADJ (64.03%), ACC/ADJ (62.32%), ATI/ADJ (60.53%), ATI/NCF (50.44%), CC/NCF (50.13%), ACC/NCF (65.21%), ATI/ACC (49.74%), CC/ATI (48.86%), and CC/ACC (48.85%). A comparison of the two tables shows that differences in ACC, ADJ and NCF explain EVA better than these variables in levels analysis, and that the remaining two variables CC and A TI have more explanatory power in the levels analysis. In table 6, only in two of five of the single coefficient or pairwise regressions does the estimated sign for CC correspond with the ex ante sign, and ATI coefficients never demonstrate the postulated sign (positive). Also, even in the final regression specification in table 6 the ex post sign for ATI does not correspond with a priori reasoning.

Panel A of tables 7 and 8 give the results of relative information content tests of the components of EVA. In table 7, when specified as a single slope coefficient CC (56.16%) has greater explanatory power than ATI (54.36%), ADJ (50.26%), NCF (48.04%) and ACC (47.46%). Table 8 describes different results, namely that when specified as a single slope coefficient ADJ (60.57%) has greater explanatory power than NCF (49.67%), ATI (48.64%), ACC (48.42%), and CC (47.29%). The results from table 8 are consistent with the previous part of the analysis since ADJ is shared by EVA with ERN, RI and NCF, ATI and CC with RI, and ADJ by itself alone. Panel B of tables 7 and 8 present the incremental information content results. Starting with the base CC, ADJ adds 27.82% in explanatory power, NCF adds 13.11%, ACC 8.05% and ATI adds 2.56%. Overall, the component of EVA that explains most variation in stock returns is adjustments (ADJ), followed by net cash flow (NCF), accruals (ACC), after-tax interest (ATI), and capital charges (CC). In other words, the first part of the analysis is supported, because ADJ, which separates EVA from other accounting-based performance measures is the most significant factor.

The results indicate that EVA[R] does indeed explain more variation in stock returns than traditional accounting-based performance measures, namely earnings (ERN), residual income (RI), and net cash flows (NCF). All other things being equal, the relative information content of EVA is in the order of 25 to 38% (depending upon the specification), whereas it is 18 to 23% for residual income, 14 to 17% for earnings and 13 to 17% for net cash flow. The second part of the first phase of the study shows that for the differenced variables, Stern Stewart's accounting adjustments (ADJ) more adequately explains STK than cash flows (NCF), after-tax interest (ATI), accruals (ACC) or the cost of capital (CC). The importance of these adjustments is also suggested by the fact that EVA has a relative information content of some 8.03% over the closely related measure of residual income.

4. Concluding Remarks

A number of points emerge from the present study. The first part of the analysis uses pooled time-series, cross-sectional data of 110 listed Australian companies to evaluate the usefulness of EVA[R] and other accounting-based performance measures. The measures of relative and incremental information content indicate that over the period 1992 to 1998 some 27% of the variation in the level of stock returns could be explained by these measures, and 44% of the variation in returns defined as year-to-year changes. Notwithstanding the obvious importance of earnings figures in value-relevance studies, EVA[R] is significant at the margin in explaining variation in stock returns. This would support the potential usefulness of EVA-type measures for internal and external performance measurement.

In the second part of the analysis, the components of EVA[R] are specified as explanatory variables in regressions with EVA. When examining the components of EVA[R] (most of which are shared with closely-related performance measures) the capital charge and after-tax interest payments were found to be the most significant components explaining EVA[R] differences, and, accordingly, the level of stock returns. However, the accounting adjustments entailed in EVA[R] calculations were found to be more significant in explaining changes in EVA[R] and hence stock returns Net cash flow, after-tax interest, accruals and the capital charge followed this. Overall, the results are broadly comparable to other studies supporting the usefulness of economic value-added, including Uyemura, Kantor and Petit (1996), O'Byrne (1996) and Bao and Bao (1998), amongst others. However, the divergence between the results of this paper and that of Biddle, Bowen and Wallace (1997) and some other US studies requires explanation, of which two possibilities are thought likely.

One possibility is that GAAP differences between Australia and the US may account for at least some difference in incremental information content between these two institutional settings. Barth and Clinch (1996, p. 164), for example, concluded that '...in addition to domestic net income and shareholder equity, differences in accounting for goodwill, asset revaluations, deferred income taxes, and pensions (or, equivalently, the US GAAP for these items) provide incremental power in explaining share returns or prices for either, or both UK and Australian firms'. Though Barth and Clinch (1996) found that the direction and magnitude of differences in Australian and US GAAP varies by the type of accounting change, one might expect that the relatively less conservative nature of Australian GAAP would result in earnings numbers that are more reflective of 'economic income'. Since O'Hanlon and Peasnell (1998) point out that one of Stern-Stewart's main objectives in calculating EVA[R] is undoing (US) accounting conservatism to more closely reflect the economic substance of transactions, the significant marginal contribution of EVA[R] in explaining Australian stock returns is a surprise. Unfortunately, it is not possible to decompose the specific GAAP adjustments in the publicly available dataset in order to throw light on this issue. Nonetheless, it would be useful to more fully examine the institutional differences between the US and Australia in order to understand disparities in future empirical work.

The second possibility is that differences in research design are responsible for the differences in results between this analysis and that of, say, Biddle, Bowen and Wallace (1997). Without doubt, while there are differences in the specification of both the dependent and independent variables in these analyses, the singularly most important finding in this study is that differences in the explanatory power of accounting-based measures across firms could be captured by differences in the constant term (a fixed-effects formulation). The implication is that the results from similar studies that rely upon the simpler common-effects formulation of panel data could be questioned on a number of grounds. For instance, in Biddle, Bowen and Wallace (1997), a simple common-effects specification is employed. That is, no allowance is made for the cross-sectional specific variation in the valuation relationship likely to arise when a sample is drawn from different industries and at different stages in the firm's life cycle. One particular outcome is that regressions incorporating this sort of assumption would tend to understate the significance of accounting-based performance measures when specified as common explanatory factors for stock returns. By itself, this may account for the differences in information content between this study and earlier work by Biddle, Bowen and Wallace (1997), amongst others.

There are at least three ways in which this research may be extended. First, a limitation in this study is that a comparison could not be made of firms who use the EVA[R] Financial Management System (incorporating redesigned executive compensation plans) against firms that use traditional accounting earnings-based incentives. While the results in the present study are suggestive of the benefits of EVA[R] as a tool for internal performance measurement and compensation design, it is conceivable that the association between EVA[R] and returns is even stronger for EVA[R] adopters (Biddle, Bowen and Wallace 1997; Ferguson & Leistikow 1998; Garvey & Milbourn 2000).

Second, there is abundant empirical evidence to suggest that models relating accounting and stock returns have more explanatory power when the accounting returns are expressed by relative changes and the relation is a non-linear, convex-concave function (see Freeman & Tse 1992; Riahi-Belakoui 1996; Frankel & Lee 1998). A further avenue of research would therefore consist of alternative specifications of accounting and stock returns, along with the use of non-linear regression techniques.

Finally, there is scope for the investigation of the usefulness of EVA[R] as an internal and external performance measure in other settings. Stern Stewart also provide performance rankings for listed companies in the UK, Canada, Brazil, Germany, Mexico, Turkey, and France, amongst others (Stern Stewart 1999), and empirical evidence from these institutional milieus would provide additional evidence regarding the contextual and/or substantive usefulness of accounting-based value-added measures.

Table 1

Descriptive Statistics of Variables Employed in the Valuation and
Components Models

This table provides descriptive statistics for compounded annual stock
returns (STK), net cash flows from operations (NCF), earnings before
extraordinary items (ERN), residual income (RI) and economic
value-added (EVA), accruals (ACC), after-tax interest expense, cost of
capital (CC) and accounting adjustments (ADJ) as specified in the
valuation and components models over the period 1992 to 1998. All
variables except STK are scaled by the number of outstanding shares.

Variable       Mean      Std. Dev.     Skewness     Kurtosis

STK           0.1496      0.4015        1.8345       13.5283
NCF           0.4205      0.6693        5.2401       45.5583
ERN           0.3869      0.5061        3.8748       23.7392
RI           -0.1445      0.4773       -2.7882       17.6090
EVA          -0.0640      0.2997       -5.2973       55.5217
ACC          -0.0330      0.4710       -7.7296      113.6291
ATI          -0.0579      0.1111       -6.9512       68.6646
CC            0.4760      0.4808        3.2759       16.9221
ADJ           0.0873      0.3581        2.0636       16.7207

Table 2
Association with Market Returns and Economic Value-Added

This table presents the estimated coefficients, standard errors,
t-statistics and [R.sup.2] for the valuation (equation 5) and
components (equation 6) models. The upper panel provides the results
for the valuation model assuming common, fixed (equation 7) and
random (equation 8) effects in the pooled data for undifferenced
and differenced variables. The dependent variable in these models is
compounded annual stock returns (STK) and the independent variables
are net cash flows from operations (NCF), earnings before
extraordinary items (ERN), residual income (RI) and economic
value-added (EVA). The lower panel provides the results for the
components model assuming common, fixed (equation 7) and random
(equation 8) effects in the pooled data for undifferenced and
differenced variables. The dependent variable in these models is
economic value-added (EVA) and (ACC), after-tax interest expense (ATI),
cost of capital (CC) and accounting adjustments (ADJ).

                    Undifferenced Variables

Variable         Estimated     Standard   t-stat   [R.sup.2]
                 Coefficient   Error

                    Panel A: Valuation Model

Common-Effects
  CONS.               0.1176     0.0263     4.47        2.86
  NCF                 0.0275     0.0431     0.64
  ERN                 0.0714     0.0615     1.16
  RI                 -0.0090     0.0481    -0.19
  EVA                 0.1476     0.1011     1.46
Fixed-Effects
  NCF                -0.0228     0.1050    -0.22       27.59
  ERN                 0.1354     0.1337     1.01
  RI                  0.1977     0.1117     1.77
  EVA                 0.0610     0.1101     0.55
Random-Effects
  CONS.               0.1304     0.0305     4.28       18.96
  NCF                 0.0083     0.0423     0.20
  ERN                 0.0763     0.0639     1.19
  RI                  0.0402     0.0603     0.67
  EVA                 0.1237     0.0841     1.47

Panel B: Components Model

Common-Effects
  CONS.               0.0045     0.0031     1.45       97.59
  NCF                 0.9490     0.0249    38.11
  ACC                 0.9479     0.0252    37.61
  ATI                 0.9919     0.0300    33.06
  CC                 -0.9679     0.0187   -51.75
  ADJ                 0.9550     0.0264    36.17
Fixed-Effects
  NCF                 0.9031    -0.9614     0.93       98.42
  ACC                 0.9045     0.0299    30.25
  ATI                 0.9848     0.0361    27.27
  CC                 -0.9614     0.0160   -60.08
  ADJ                 0.9707     0.0181    53.62
Random-Effects
  CONS.               0.0055     0.0040     1.37       98.22
  NCF                 0.9431     0.0086   109.66
  ACC                 0.9427     0.0098    96.19
  ATI                 0.9917     0.0277    35.80
  CC                 -0.9657     0.0093   -103.8
  ADJ                 0.9588     0.0097    98.84

                    Differenced Variables

Variable         Estimated     Standard   t-stat   [R.sup.2]
                 Coefficient   Error

                    Panel A: Valuation Model

Common-Effects
  CONS.               0.1457     0.0169     8.62       11.01
  NCF                -0.0089     0.0364    -0.24
  ERN                 0.2320     0.1049     2.21
  RI                  0.3293     0.1454     2.26
  EVA                 0.0952     0.0864     1.10
Fixed-Effects
  NCF                -0.0066     0.0313    -0.21       44.06
  ERN                 0.2098     0.0991     2.12
  RI                  0.3097     0.1240     2.50
  EVA                 0.0806     0.0797     1.01
Random-Effects
  CONS.               0.1462     0.0293     4.99       41.41
  NCF                -0.0074     0.0321    -0.23
  ERN                 0.2168     0.0819     2.65
  RI                  0.3168     0.0593     5.34
  EVA                 0.0849     0.0562     1.51

                    Panel B: Components Model

Common-Effects
  CONS.              -0.0423     0.0101    -4.18       31.09
  NCF                 0.2386     0.1797     1.32
  ACC                 0.1929     0.1859     1.03
  ATI                -0.4203     0.6163    -0.68
  CC                 -0.2973     0.1170    -2.54
  ADJ                 0.5029     0.2499     2.01
Fixed-Effects
  NCF                 0.4166     0.0992     4.19       72.60
  ACC                 0.3690     0.0994     3.71
  ATI                -0.0256     0.3405    -0.07
  CC                 -0.3889     0.0806    -4.82
  ADJ                 0.4988     0.1358     3.67
Random-Effects
  CONS.              -0.0465     0.1736    -0.26       61.65
  NCF                 0.3645     0.0526     6.92
  ACC                 0.3171     0.0540     5.87
  ATI                -0.1536     0.1557    -0.98
  CC                 -0.3623     0.0478    -7.57
  ADJ                 0.4999     0.0322    15.52

Table 3
Association with Market Returns for the Valuation Model with
Fixed-Effects

This table provides the estimated coefficients, standard errors,
t-statistics, F-statistic and adjusted [R.sup.2] for the valuation
models in equation (5) assuming fixed-effects (equation 7) in the
pooled data. The variables are specified in undifferenced form in the
upper panel and differenced form in the lower panel. The dependent
variable is compounded annual stock returns (STK) and the independent
variables are net cash flows from operations (NCF), earnings before
extraordinary items (ERA), residual income (RI) and economic value-added
(EVA). The 11 models in each panel are comprised of four models
where each independent variable is specified univariately followed by
six models where the independent variables are specified in pairwise
combinations and finally jointly.

              NCF                                   ERN

Estimated     Standard             Estimated    Standard
Coefficient     Error    t-stat   Coefficient     Error    t-stat

Panel A: Undifferenced Variables

 0.4118         0.0367    11.22
                                       0.1867     0.0579     3.22
-0.0288         0.0879    -0.33
                                       0.0992     0.086      1.15
 0.0043         0.0403     0.11        0.1810     0.0648     2.79
                                       0.0504     0.0900     0.56
 0.0412         0.0424     0.97
-0.0228         0.1050    -0.22        0.1354     0.1337     1.01

Panel B: Differenced Variables

 0.0823         0.0629     1.31
                                       0.1673     0.0893     1.87
-0.0015         0.0311    -0.05
                                       0.2043     0.1288     1.59
 0.0790         0.0801     0.99        0.0960     0.1158     0.83
                                       0.0382     0.0897     0.43
 0.0789         0.0634     1.24
-0.0066         0.0313    -0.21        0.2098     0.0991     2.12

              NCF                                    RI

Estimated     Standard             Estimated    Standard
Coefficient     Error    t-stat   Coefficient     Error    t-stat

Panel A: Undifferenced Variables

 0.4118         0.0367    11.22
                                       0.2510     0.0878     2.86
-0.0288         0.0879    -0.33
                                       0.1971     0.1123     1.76
 0.0043         0.0403     0.11
                                       0.2350     0.0981     2.40
 0.0412         0.0424     0.97        0.2539     0.0884     2.87
-0.0228         0.1050    -0.22        0.1977     0.1117     1.77

Panel B: Differenced Variables

 0.0823         0.0629     1.31
                                       0.2731     0.0841     3.25
-0.0015         0.0311    -0.05
                                       0.3124     0.1204     2.59
 0.0790         0.0801     0.99
                                       0.2568     0.0838     3.06
 0.0789         0.0634     1.24        0.2554     0.0855     2.99
-0.0066         0.0313    -0.21        0.3097     0.1240     2.50

              NCF                                   EVA

Estimated     Standard             Estimated    Standard
Coefficient     Error    t-stat   Coefficient     Error    t-stat

Panel A: Undifferenced Variables

 0.4118         0.0367    11.22
                                       0.2326     0.1484     1.57
-0.0288         0.0879    -0.33        0.2205     0.1488     1.48
                                       0.2260     0.1459     1.55
                                       0.0835     0.1135     0.74
 0.0043         0.0403     0.11
 0.0412         0.0424     0.97
-0.0228         0.1050    -0.22        0.0610     0.1101     0.55

Panel B: Differenced Variables

 0.0823         0.0629     1.31
                                       0.1647     0.1022     1.61
-0.0015         0.0311    -0.05        0.1656     0.1023     1.62
                                       0.1504     0.1042     1.44
                                       0.0935     0.0767     1.22
 0.0790         0.0801     0.99

 0.0789         0.0634     1.24
-0.0066         0.0313    -0.21        0.0806     0.0797     1.01

              NCF

Estimated     Standard                         Adj.
Coefficient     Error    t-stat       F    [R.sup.2]

Panel A: Undifferenced Variables

 0.4118         0.0367    11.22     1.26       13.51
                                   10.41       14.42
                                    8.44       18.53
                                    2.60       25.68
-0.0288         0.0879    -0.33   138.01       25.99
                                  137.40       25.57
                                  142.78       26.56
 0.0043         0.0403     0.11    92.43       14.17
                                  111.19       18.44
 0.0412         0.0424     0.97   110.54       18.47
-0.0228         0.1050    -0.22    48.65       27.15

Panel B: Differenced Variables

 0.0823         0.0629     1.31     1.71       17.09
                                    3.51       17.31
                                   10.47       23.06
                                    2.60       37.69
-0.0015         0.0311    -0.05   175.79       38.17
                                  182.43       38.72
                                  206.66       41.91
 0.0790         0.0801     0.99    99.65       17.80
                                  117.11       22.97
 0.0789         0.0634     1.24   120.83       23.72
-0.0066         0.0313    -0.21    72.73       43.63

Table 4
Relative and Incremental Information Content for Valuation Model,
Undifferenced Variables

This table provides the hypothesis tests of equation (5) that EVA, RI,
ERN and NCF have equal relative (panel A) and incremental (panel B)
information content in undifferenced form. The adjusted [R.sup.2] for
the fixed-effects specification in table 3 where each independent
variable is specified univariately and provides the tests of relative
information content. Taking the adjusted
[R.sup.2] from each pairwise
regression in table 3, and subtracting the [R.sup.] obtained in the
earlier univariate regression obtain the tests for relative information
content.

Panel A: Relative Information Content

  EVA        >        RI         >       ERN       >        NCF

25.68%              18.53%              14.42%            13.51%

Panel B: Incremental Information Content

EVA/ERN   ERN/EVA   EVA/NCF   NCF/EVA   EVA/RI   RI/EVA   ERN/NCF

11.15%    -0.11%    12.48%     0.31%    8.03%    0.88%     0.66%

NCF/ERN   ERN/RI   RI/ERN   NCF/RI   RI/NCF

-0.25%    -0.09%   4.02%    -0.06%   4.96%

Table 5
Relative and Incremental Information Content for Valuation Model,
Differenced Variables

This table provides the hypothesis tests of equation (5) that EVA, RI,
ERN and NCF have equal relative (panel A) and incremental (panel B)
information content in differenced form. The adjusted [R.sup.2] for the
fixed-effects specification in table 3 where each independent variable
is specified univariately and provides the tests of relative
information content. Taking the adjusted [R.sup.2] from each pairwise
regression in table 3, and subtracting the [R.sup.2] obtained in the
earlier univariate regression obtain the tests for relative information
content.

Panel A: Relative Information Content

  EVA        >        RI         >       ERN       >        NCF

37.69%              23.06%              17.31%            17.09%

Panel B: Incremental Information Content

EVA/ERN   ERN/EVA   EVA/NCF   NCF/EVA   EVA/RI   RI/EVA   ERN/NCF

21.41%     1.03%    21.08%     0.48%    18.85%   4.22%     0.71%

NCF/ERN   ERN/RI   RI/ERN   NCF/RI   RI/NCF

 0.49%    -0.09%   5.66%    0.66%    6.63%

Table 6
Association with EVA for the Components Model, Fixed-Effects
Specification

This table provides the estimated coefficients, standard errors,
t-statistics, F-statistic and adjusted [R.sup.2] for the components
models in equation (6) assuming fixed-effects (equation 7) in the
pooled data. The variables are specified in undifferenced form in the
upper panel and in differenced form in the lower panel. The dependent
variable is economic value-added (EVA) and the independent variables
are net cash flow (NCF), accruals (ACC), after-tax interest expense
(ATI), cost of capital (CC) and accounting adjustments (ADJ). The 16
models in each panel are comprised of five models where each
independent variable is specified unit-stat

                  NCF                               ACC

Estimated     Standard             Estimated    Standard
coefficient     error    t-stat   coefficient     error    t-stat

Panel A: Undifferenced Variables

-0.0227         0.0428     0.53
                                       0.0532     0.0536     0.99
                                      -0.0019     0.0427     0.04
 0.1065         0.0396     2.69
                                      -0.0263     0.0514     0.51
 0.0908         0.0596     1.52
                                       0.0948     0.0668     1.42
 0.1019         0.1374     0.74        0.1565     0.1533     1.02
-0.0616         0.0561     1.10
 0.9231         0.0307    30.07        0.9045     0.0299    30.25

Panel B: Differenced Variables

 0.0817         0.0262     3.12
                                      -0.0696     0.0262     2.66
                                      -0.0685     0.0261     2.62
 0.0792         0.0254     3.12
                                      -0.0595     0.0253     2.35
 0.0674         0.0247     2.73
                                      -0.0513     0.0205     2.50
 0.1696         0.0689     2.46        0.0992     0.0651     1.52
 0.0621         0.0237     2.62
 0.4166         0.0992     4.20        0.369      0.0994     3.71

                  NCF                               ATI

Estimated     Standard             Estimated    Standard
coefficient     error    t-stat   coefficient     error    t-stat

Panel A: Undifferenced Variables

-0.0227         0.0428     0.53
                                       0.9548     0.3518     2.71
                                       0.4413     0.4504     0.98
 0.1065         0.0396     2.69
                                       0.9947     0.3754     2.65
                                       1.5673     0.3742     4.19
 0.0908         0.0596     1.52        1.1695     0.4337     2.70
 0.1019         0.1374     0.74
-0.0616         0.0561     1.10
 0.9231         0.0307    30.07        0.9848     0.0361    27.28

Panel B: Differenced Variables

 0.0817         0.0262     3.12
                                      -0.5757     0.4009     1.44
                                      -0.6302     0.4377     1.44
 0.0792         0.0254     3.12
                                      -0.4983     0.3972     1.25
                                      -0.0718     0.3142     0.23
 0.0674         0.0247     2.73       -0.4056     0.4086     0.99
 0.1696         0.0689     2.46
 0.0621         0.0237     2.62
 0.4166         0.0992     4.20       -0.0256     0.3405     0.08

                  NCF                                CC

Estimated     Standard             Estimated    Standard
coefficient     error    t-stat   coefficient     error    t-stat

Panel A: Undifferenced Variables

-0.0227         0.0428     0.53
                                      -0.3072     0.0795     3.86
                                      -0.2251     0.1082     2.08
                                      -0.3077     0.0839     3.67
                                      -0.7312     0.0948     7.71
 0.1065         0.0396     2.69       -0.4608     0.0729     6.32
 0.0908         0.0596     1.52
 0.1019         0.1374     0.74
-0.0616         0.0561     1.10
 0.9231         0.0307    30.07       -0.9614      0.016    60.09

Panel B: Differenced Variables

 0.0817         0.0262     3.12
                                       0.0546     0.0686     0.80
                                      -0.029      0.0809     0.36
                                       0.0497     0.0683     0.73
                                      -0.2258     0.0756     2.99
 0.0792         0.0254     3.12        0.0453     0.0826     0.55
 0.0674         0.0247     2.73
 0.1696         0.0689     2.46
 0.0621         0.0237     2.62
 0.4166         0.0992     4.20       -0.3889     0.0806     4.83

                  NCF                               ADJ

Estimated     Standard             Estimated    Standard
coefficient     error    t-stat   coefficient     error    t-stat

Panel A: Undifferenced Variables

-0.0227         0.0428     0.53
                                       0.1861     0.1401     1.33
                                       0.6996     0.1242     5.63
 0.1065         0.0396     2.69
                                       0.407      0.1543     2.64
 0.0908         0.0596     1.52
                                       0.2184     0.1441     1.52
 0.1019         0.1374     0.74
-0.0616         0.0561     1.10        0.2358     0.1869     1.26
 0.9231         0.0307    30.07        0.9707     0.0181    53.63

Panel B: Differenced Variables

 0.0817         0.0262     3.12
                                       0.3102     0.1172     2.65
                                       0.4134     0.1183     3.49
 0.0792         0.0254     3.12
                                       0.3051     0.1226     2.49
 0.0674         0.0247     2.73
                                       0.3021     0.1157     2.61
 0.1696         0.0689     2.46
 0.0621         0.0237     2.62        0.3528     0.1181     2.99
 0.4166         0.0992     4.20        0.4988     0.1358     3.67

                  NCF

Estimated     Standard                          Adj.
coefficient     error    t-stat        F    [R.sup.2]

Panel A: Undifferenced Variables

-0.0227         0.0428     0.53      0.28       48.04
                                     0.97       47.96
                                     7.37       54.36
                                    14.88       56.16
                                     2.30       50.26
                                   509.72       56.92
                                   500.21       56.01
                                  1396.67       78.08
 0.1065         0.0396     2.69    613.25       61.15
                                   458.28       54.29
                                   699.05       64.46
 0.0908         0.0596     1.52    481.76       55.72
                                   415.77       51.34
 0.1019         0.1374     0.74    368.74       48.52
-0.0616         0.0561     1.10    404.35       56.92
 0.9231         0.0307    30.07   5826.20       98.40

Panel B: Differenced Variables

 0.0817         0.0262     3.12      9.72       49.67
                                     7.07       48.42
                                     2.06       48.64
                                     0.63       47.29
                                     7.81       60.57
                                   314.36       48.86
                                   322.89       48.85
                                   600.51       64.03
 0.0792         0.0254     3.12    336.76       50.13
                                   327.56       49.74
                                   506.89       60.53
 0.0674         0.0247     2.73    333.76       50.44
                                   543.03       62.32
 0.1696         0.0689     2.46    336.72       49.90
 0.0621         0.0237     2.62    630.27       65.21
 0.4166         0.0992     4.20    210.66       72.28

Table 7

Relative and Incremental Information Content for the Components Model,
Undifferenced Variables

This table provides the hypothesis tests of equation (6) that NCF, ACC,
ATI, CC and ADJ have equal relative (panel A) and incremental (panel B)
information content in undifferenced form. The adjusted [R.sup.2] for
the fixed-effects specification in table 6 where each independent
variable is specified univariately provides the tests of relative
information content. Taking the adjusted [R.sup.2] from each pairwise
regression in 3, and subtracting the [R.sup.2] obtained in the earlier
univariate regression obtain the tests for relative information
content.

Panel A: Relative Information Content

CC        >         ATI       >         ADJ       >

56.16%              54.36%              50.26%

Panel B: Incremental Information Content

CC/ATI    ATI/CC    CC/ACC    ACC/CC    CC/ADJ    ADJ/CC

2.56%     0.76%     8.05%     -0.15%    27.82%    21.92%

ATI/ADJ   ADRATT    ATI/NCF   NCF/ATI   ACC/ADJ   ADJ/ACC

14.20%    10.10%    7.68%     1.36%     1.08%     3.38%

Panel A: Relative Information Content

NCF       >         ACC

48.04%              47.96%

Panel B: Incremental Information Content

CC/NCF    NCF/CC    ATI/ACC   ACC/ATI

13.11%    4.99%     6.33%     -0.07%

ACC/NCF   NCF/ACC   ADJ/NCF   NCF/ADJ

0.48%     0.56%     8.88%       6.66%

Table 8

Relative and Incremental Information Content for the Components Model,
Differenced Variables

This table provides the hypothesis tests of equation (6) that NCF, ACC,
ATI, CC and ADJ have equal relative (panel A) and incremental (panel B)
information content in differenced form. The adjusted [R.sup.2] for the
fixed-effects specification in table 6 where each independent variable
is specified univariately provides the tests of relative information
content. Taking the adjusted [R.sup.2] from each pairwise regression in
table 3, and subtracting the [R.sup.2] obtained in the earlier
univariate regression obtain the tests for relative information
content.

Panel A: Relative Information Content

  ADJ        >        NCF        >        ATI        >

60.57%              49.67%              48.64%

Panel B: Incremental Information Content

CC/ATI    ATI/CC    CC/ACC    ACC/CC    CC/ADJ    ADJ/CC

 0.22%     1.57%     0.43%     1.56%     3.46%    16.74%

ATI/ADJ   ADJ/ATI   ATI/NCF   NCF/ATT   ACC/ADJ   ADJ/ACC

-0.04%    11.89%     0.77%     1.80%     1.75%    13.90%

Panel A: Relative Information Content

  ACC        >        CC

48.42%              47.29%

Panel B: Incremental Information Content

CC/NCF    NCF/CC    ATI/ACC   ACC/ATI

 0.46%     2.84%     1.32%     1.10%

ACC/NCF   NCF/ACC   ADJ/NCF   NCF/ADJ

 0.23%     1.48%    15.54%     4.64%

(Date of receipt of final transcript: June, 2004. Accepted by John Lyon & Baljit Sidhu, Area Editors.)

References

Ali, A. & Pope, P.F. 1995, 'The incremental information content of earnings, funds flow and cash flow: The UK evidence', Journal of Business Finance and Accounting, vol. 22, no. 1, pp. 19-34.

Anonymous, 1993, 'The real key to creating wealth', Fortune, vol. 128, no. 6, pp. 38-43.

Bao, B-H. & Bao, D-H. 1998, 'Usefulness of value added and abnormal economic earnings: An empirical examination', Journal of Business Finance and Accounting, vol. 25, no. 1-2, pp. 251-65.

Barth, M.E. & Clinch, G. 1996, 'International accounting differences and their relation to share prices: Evidence from UK, Australian and Canadian Firms', Contemporary Accounting Research, vol. 13, no. 1, pp. 135-70.

Biddle, G.C., Bowen, R.M. & Wallace, J.S. 1997, 'Does EVA beat earnings? Evidence on associations with stock returns and firm values', Journal of Accounting and Economics, vol. 24, no. 3, pp. 301-36.

Biddle, G.C., Seow, G.S. & Siegel, A.F. 1995, 'Relative versus incremental information content', Contemporary Accounting Research, vol. 12, no. 1, pp. 1-23.

Bowen, R.M., Burgstahler, D. & Daley, L.A. 1987, 'The incremental information content of accrual versus cash flows', The Accounting Review, vol. 62, no. 4, pp. 723-47.

Chen, S. & Dodd, J.L. 1997, 'Economic value added (EVA): An empirical examination of a new corporate performance measure', Journal of Managerial Issues, vol. 9, no. 3, pp. 318-33.

Clinton, B.D. & Chen, S. 1998, 'Do new performance measures measure up?', Management Accounting, vol. 80, no. 4, pp. 38-44.

Drucker, P. 1991, 'The new productivity challenge', Harvard Business Review, vol. 69, no. 6, pp. 69-79.

Easton, P.D. & Harris, T.S. 1991, 'Earnings as an explanatory variable for returns', Journal of Accounting Research, vol. 29, no. 1, pp. 19-36.

Ferguson, R. & Leistikow, D. 1998, 'Search for the best financial performance measure: Basic are better', Financial Analysts Journal, vol. 54, no. 1, pp. 81-6.

Frankel, R. & Lee, C.M.C. 1998, 'Accounting valuation, market expectation and cross-sectional stock returns', Journal of Accounting and Economics, vol. 25, no. 3, pp. 283- 319.

Freeman, R.N. & Tse, S.Y. 1992, 'A nonlinear model of security price responses to unexpected earnings', Journal of Accounting Research, vol. 30, no. 2, pp. 185-209.

Garvey, G.T. & Milbourn, T.T. 2000, 'EVA versus earnings: Does it matter which is more highly correlated with stock returns?', Journal of Accounting Research, vol. 38, Supplement, pp. 209-45.

Greene, W.H. 1993, Econometric Analysis, Prentice-Hall, Englewood Cliffs, pp. 612-38.

Herzberg, M.M. 1998, 'Implementing EBO/EVA[R] analysis in stock selection', Journal of Investing, vol. 7, Spring, pp. 45-53.

Jennings, R. 1990, 'A note on interpreting incremental information content', The Accounting Review, vol. 65, no. 4, pp. 925-32.

Judge, G.G., Hill, R.C., Griffiths, W.E., Lutkepohl, H. & Lee, T.C. 1988, Introduction to the Theory and Practice of Econometrics, Wiley, New York, pp. 468-91.

Lehn, K. & Makhija, A.K. 1997, 'EVA, accounting profits, and CEO turnover: An empirical examination, 1985-1994', Journal of Applied Corporate Finance, vol. 10, no. 2, pp. 90-7.

McClenahen, J.S. 1998, 'Accounting for Change', Industry Week, vol. 247, no. 17, pp. 63-5.

O'Byrne, S.F. 1996, 'EVA and market value', Journal of Applied Corporate Finance, vol. 9, no. 1, pp. 116-25.

O'Hanlon, J. & Peasnell, K. 1998, 'Wall Street's contribution to management accounting: The Stern Stewart EVA[R] Financial Management System', Management Accounting Research, vol. 9, no. 2, pp. 421-44.

O'Hanlon, J. & Peasnell, K. 2000, 'Residual income and value creation: The missing link', paper presented at the British Accounting Association (BAA) Annual Conference, University of Exeter.

Riahi-Belakoui, A. 1996, 'Earnings-returns relation versus net-value-added-returns relation: The case for a nonlinear specification', Advances in Quantitative Analysis of Finance and Accounting, vol. 4, no. 1, pp. 175-85.

Rogerson, W.P. 1997, 'Intertemporal cost allocation and managerial investment incentives: A theory explaining the use of economic value added as a performance measure', Journal of Political Economy, vol. 105, no. 4, pp. 770-95.

Stern Stewart & Co. 1999, WWW <http:/www.sternstewart.com/>, Accessed February 1999.

Stewart, G.B. 1991, The Quest for Value, Harper, New York.

Stewart, G.B. 1994, 'EVA: Fact and fantasy', Journal of Applied Corporate Finance, vol. 7, no. 2, pp. 71-84.

Topkis, M. 1996, 'A new way to find bargains', Fortune, vol. 134, no. 11, pp. 265-66.

Uyemura, D.G., Kantor C.C. & Petit, J.M. 1996, 'EVA for banks: Value creation, risk management and profitability measurement', Journal of Applied Corporate Finance, vol. 9, no. 2, pp. 94-111.

White, H. 1980, 'A heteroskedasticity-consistent covariance matrix and a direct test for heteroskedasticity, Econometrica, vol. 48, pp. 817-38.

Worthington, A.C. & West, T. 2001, 'Economic value-added: A review of the theoretical and empirical literature', Asian Review of Accounting, vol. 9, no. 1, pp. 67-86.

Young, S.D. 1999, 'Some reflections on accounting adjustments and economic value added', Journal of Financial Statement Analysis, vol. 4, no. 2, pp. 7-13.

Andrew C. Worthington ([dagger]) Tracey West ([dagger])

([dagger]) School of Economics and Finance, Queensland University of Technology, GPO Box 2434, Brisbane, QLD 4001. Email: [e-mail]

The authors would like to thank conference participants at the British Accounting Association (BAA) Annual Conference, University of Exeter, seminar attendees at the University of Technology, Sydney and two anonymous referees for helpful comments on an earlier version of this paper. The financial support of a QUT Faculty of Business Research Initiative Grant is also gratefully acknowledged.