European Accounting Review, 17 (3):

Transcription

1 Provided by te autor(s) and University College Dublin Library in accordance wit publiser policies. Please cite te publised version wen available. Title A Comparison of Error Rates for EVA, Residual Income, GAAP-earnings, and Oter Metrics Using a Long-window Valuation Approac Autor(s) Forker, Jon; Powell, Ronan Publication date 2008 Publication information European Accounting Review, 17 (3): Publiser Taylor and Francis Item record/more information Publiser's statement Publiser's version (DOI) ttp://dl.andle.net/10197/7490 Tis is an electronic version of an article publised in European Accounting Review. European Accounting Review is available online at: ttp://dx.doi.org/ / Downloaded T23:30:50Z Te UCD community as made tis article openly available. Please sare ow tis access benefits you. Your story matters! (@ucd_oa) Some rigts reserved. For more information, please see te item record link above.

2 A Comparison of Error Rates for EVA, Residual Income, GAAP-earnings, and Oter Metrics Using a Long-Window Valuation Approac JOHN FORKER Queen s University Belfast, Nortern Ireland RONAN POWELL University of New Sout Wales, Sydney, Australia Te autors are grateful to Stern Stewart & Co., for providing te data for te US and UK on wic tis study is based, to Micael Kollo for researc assistance and for financial support from te Centre for Business Performance, ICAEW, in funding tis researc. Te paper as benefited from comments received from two anonymous reviewers, seminar participants at Queen s University Belfast, and te annual conferences of te Britis Accounting Association and European Accounting Association in 2003 and 2004, respectively. Address for correspondence: Ronan Powell, Scool of Banking and Finance, te University of New Sout Wales, Sydney, NSW 2052, Australia. Tel: ; Fax: ; r.powell@unsw.edu.au. 1

3 A Comparison of Error Rates for EVA, Residual Income, GAAP-earnings, and Oter Metrics Using a Long-Window Valuation Approac ABSTRACT Predictability and variability are two measures commonly used in te empirical literature to gauge te quality of earnings and ence, decision usefulness to investors. We adopt bot measures to investigate empirically te relative quality of Stern Stewart s measure of economic value added (EVA) compared to GAAP earnings, residual income, cas flows and oter mandated metrics in te US and UK. We proxy for accounting quality by applying a long-window metodology to obtain indsigt valuation errors based on te difference between ex ante market value and discounted ex post metrics. Decision usefulness, in terms of ease of forecasting, is proxied by differences in valuation errors between te bencmark and alternative accounting metods. Contrary to te Biddle, Bowen and Wallace (1997) finding tat mandated earnings were superior to EVA and residual income, we find tat EVA and oter residual income metrics consistently give rise to lower average valuation errors and tus ave iger predictability across a variety of windows and terminal dates. Furter, on te basis of our second measure of accounting quality, te variability of valuation errors, EVA performs best in te US and tird in te UK. Te results strongly indicate tat differences between residual income constructs, including EVA, are generally small but tat earnings quality will be improved by recognition of a cost of equity capital in measuring reported income. Keywords: valuation errors, residual income, EVA, MVA, GAAP, IASB 2

4 A Comparison of Error Rates for EVA, Residual Income, GAAP-earnings, and Oter Metrics Using a Long-Window Valuation Approac 1. Introduction Te measurement and presentation of financial performance is central to te process by wic investors set and revise expected cas flows and serves as te basis for setting sare prices and te efficient allocation of resources in market-based economies. Te quality of mandated financial accounting information for tis purpose is, owever, increasingly under scrutiny. Alternative proprietary financial performance metrics, one of wic is te measure of economic value added (EVA), 1 devised by Stern Stewart ave been proposed. Te quest to improve te quality of financial statements is also ig on te agenda of accounting regulators. Te International Accounting Standards Board (IASB) and te Financial Accounting Standards Board (FASB) are engaged in a joint project to improve te content and presentation of financial statements for te purpose of assisting users of financial statements to predict cas flows. Te aim of te current project on Financial Statement Presentation (IASB, 2007a) is to establis a ig-quality standard of information in te financial statements, including te classification and display of items and te aggregation of line items into subtotals and totals. Te objective of tis study is to investigate te relative decision usefulness of differences between GAAP-based accounting and alternative metods, suc as Stern Stewart s measure of EVA to te prediction of cas flows. We focus in particular on te value relevance of te recognition of a cost of equity capital in measuring reported income wic as long been advocated by Antony (1975; 1983) and wic is noticeably absent from te issues under consideration by te IASB in its quest to improve financial reporting quality. Prior researc demonstrates te teoretical equivalence of equity valuation models based on eiter discounted dividends, cas flows or GAAP-based earnings (Penman and Sougiannis, 1997). However, te Edwards, Bell and Olson residual income valuation model 3

5 as been empirically identified as providing te best explanation for sare prices (Penman and Sougiannis, 1998; Lee, Myers and Swaminatan, 1999; and Francis, Olsson and Oswald, 2000). A likely explanation is tat te difficulty of forecasting dividends or cas flows is greater tan taking book value of equity as a starting point and ten estimating te residual of future earnings less a carge for te cost of equity (Lee, 1996). However, in a related study of te association between market returns and alternative performance metrics, Biddle et al., (1997) find tat mandated earnings ave a iger association wit equity returns compared to, in descending order, residual income, EVA and cas flow. In particular, Biddle et al., (1997: 332) conclude Furter, wile te carge for capital and Stern Stewart s adjustments for accounting distortions sow some marginal evidence of being incrementally important, tis difference does not appear to be economically significant. Te finding tat te capital carge as minimal value relevance is puzzling given its central role in te residual income valuation model. To investigate tis issue furter, we adopt a long-window design to test for te significance of te cost of capital carge in measuring equity valuation errors for a sample of UK and US firms. Many of te oter Stern Stewart adjustments to GAAP earnings relate to accounting treatment and timing differences and we expect, based on te prior literature, tat te effect of tese to be less evident in our long-window researc design. For example, O Hanlon and Pope (1999) find little evidence tat dirty-surplus flows (e.g., goodwill, prioryear adjustments) are value relevant in explaining valuation errors using long-window tests similar to Easton, Harris and Olson (1992). Furter, in a cross-country comparison, Isidro, O Hanlon and Young (2006) report similar findings for te UK, but also report some weak predictable evidence between dirty-surplus flows and valuation errors for te US. To contribute to tis literature, we investigate te relative value relevance of different line items in te income statement, before and after financing carges, exceptional items and, in te UK, for all recognised gains reported in te Statement of Recognised Gains and Losses (FRS 3, Accounting Standards Board, 1992). 4

6 We follow Scipper and Vincent (2003) by identifying predictability and variability as earnings quality constructs and in-line wit previous empirical studies measure tese as te mean and dispersion of valuation errors, respectively. To implement our researc design we employ a metodology, adapted from Siller (1981) and Penman and Sougiannis (1998), to obtain valuation errors, defined as te difference between te indsigt value for eac performance metric and te ex ante market value. We report te mean and variability of valuation errors for alternative windows and performance metrics to gauge te relative quality and, ence, te decision usefulness of eac metric. We also contribute to prior empirical work (Penman and Sougiannis, 1998; Lee et al., 1999; and Francis et al., 2000) by adopting te entity perspective, in wic Stern Stewart s EVA and market value added (MVA) are grounded. Valuation equations identify te impact of different accounting metods and Stern Stewart data on EVA, MVA and te cost of capital are used to construct a bencmark against wic to compare and rank indsigt valuation errors for different accounting metods. Tis allow us to assess te contribution to earnings quality of a carge for te cost of equity capital and to assess te decision usefulness, in terms of ease of forecasting, of mandated requirements for te display of specific earnings line items relative to Stern Stewart specific accounting practices. Te study utilises te Stern Stewart data sets for te UK and US covering 11 and 16 years, respectively. Tese data permit te calculation of valuation errors for windows of up to 10 years. Selective substitution of alternative measures for capital and earnings provides insigts into a range of measurement and presentational issues. Permanent differences between Stern Stewart s EVA and mandated earnings exist were GAAP applies dirty surplus accounting compared to te application of clean surplus accounting in EVA. For example, Stern Stewart and Penman (2003) advocate expense recognition of non-cas costs incurred by sareolders arising from te exercise of employee sare options, wic GAAP ignored prior to IFRS 2 (IASB, 2004). Like Antony (1975; 1983) Stern Stewart also strongly advocate recognition of an expense for sareolders cost of equity capital, wic is also ignored under GAAP. Timing or transitory differences also exist between GAAP and EVA in te form of 5

7 capitalisation and amortisation of value creating expenditure, suc as researc and development costs, wic are typically written off immediately as an expense under GAAP. Oter timing differences arise wen Stern Stewart reverse managerial discretion in accounting for provisions and reserves in accounting for EVA. Te key results reported ere are different to tose found by Biddle et al. (1997). First, EVA and oter residual income measures outperform mandated earnings by generating smaller valuation errors. Tis result provides strong support for Edwards and Bell (1961), Antony (1975; 1983) and Stewart (1991) wo advocate te recognition of a cost of equity capital in measuring financial performance. Te result also confirms previous US findings (Penman and Sougiannis, 1998, Lee et al., 1999; and Francis et al., 2000) wo also use longwindow metodologies to compare a residual income metric based on mandated earnings wit dividends and cas flow metrics. Second, for te set of residual income measures investigated, te differences in rankings are generally small. Focussing on te mean and variability of te distributions of errors, te best performing metric in te UK is residual income calculated using mandated earnings and Stern Stewart s measure of capital. In te US, te best performing metric is also residual income, calculated using mandated earnings and accounting book value of assets. However, and most importantly, EVA as te smallest variability across all metrics for te US wic is consistent wit te latter being easier to forecast compared to mandated practices. Te currently mandated reporting of total recognised gains and losses in te UK is te best performing conventional metric. Te relative rankings of metrics are fairly consistent across different windows, indicating a ig degree of robustness in te results. Te paper proceeds as follows. Section 2 provides an overview of te teoretical background and related researc and Section 3 describes te researc design and te ypoteses to be tested. Te data and results are reported in Section 4 and Section 5 concludes. 6

8 2. Teoretical background and related researc Earnings measurement and te valuation of equity are teoretically linked wen two conditions are satisfied. Te first requires application of te clean surplus relation wen measuring income to take into account all factors impacting on sareolder wealt. Te second is recognition of an expense for te cost of equity capital to report residual income. Tis gives te result tat te current value of equity is equal to te book value of equity plus te discounted present value of te future stream of residual equity income. Tis relationsip, first identified by Preinreic (1938), provided te teoretical core for Edwards and Bell s classic treatise, Te Teory and Measurement of Business Income (1961). Edwards and Bell s measure of business income is based on valuing assets at replacement cost, and is equivalent to te concept of compreensive income now advocated by te IASB as te basis for reporting financial performance. An integral part of te Edwards and Bell contribution to te teory for measuring business income was te deduction of a cost of equity capital based on te start of period value of assets. Te resulting measure of residual income was termed excess current income. Prior to tis, te concept of residual income attained prominence in management accounting for te purpose of exercising control in diversified companies based on its application in General Motors and General Electric (Solomons, 1965). Stern Stewart s measure of EVA meets te two requirements for a measure of residual income by adering to te practice of clean surplus accounting and by te recognition of a cost of equity capital. For tese reasons it is teoretically superior to mandated earnings. However, its role as a tool of management accounting in controlling and rewarding managerial performance is at least as prominent as its claimed contribution to investors seeking a relevant basis for equity valuation (Bromwic and Walker, 1998; O Hanlon and Peasnell, 1998; Stark and Tomas, 1998). A main finding in te literature on residual income is tat a single-period residual income figure is not a reliable indicator of te periodic cange in sareolder wealt (see, e.g., Bromwic and Walker, 1998; and O Hanlon and Peasnell, 1998); ence a long window researc design is more appropriate tan a one period window. 7

9 Biddle et al. (1997) focus on claims tat EVA is more igly associated wit sareolder returns tan conventional accrual-based earnings. Residual income-type measures migt be expected to ave a iger association wit firm value or security returns tan mandated earnings as residual income features in te valuation equation, wile mandated earnings does not. However, Biddle et al. (1997) note tat investors only observe past and current data as te basis for predicting residual income and, suggest it may be te case tat oter metrics, suc as mandated earnings, provide a better basis for predicting residual income tan do residual income metrics, including EVA. Teir study tus addresses te empirical issue of identifying te metric tat provides more information about future residual income. Biddle et al, (1997) regress contemporaneous sareolder returns on cas flow from operations, mandated earnings, residual income and EVA for te period Teir findings based on measures of relative information content were contrary to Stern Stewart s claims for EVA, tat earnings ave a iger association wit security returns tan EVA. In order of relative information content, te ranking was first, mandated earnings, ten residual income followed by EVA and cas flow from operations. Furter, an investigation of te Stern Stewart capital carge and te accounting adjustments added little in explaining contemporaneous returns. Te Biddle et al. (1997) researc design is, owever, subject to te limitation tat sareolders return (an equity metric) is regressed on contemporaneous measures of performance tat are measured at te entity (operating) level of te firm. Restricting te analysis to a single period contemporaneous association wit firm values and returns does not address te problem tat one period measures of residual income are not necessarily associated wit te sareolder canges in wealt reflected in security returns. Also, an association between one period returns and a carge for te cost of capital is potentially mitigated by te carge aving little variation across a sample of large firms. Subsequent to te Biddle et al. (1997) study are tree studies based on US data (Penman and Sougiannis, 1998; Lee et al., 1999; and Francis et al., 2000) tat use long window metodologies to compare te relative accuracy of earnings, dividends and cas 8

10 flows in explaining sare prices. Penman and Sougiannis (1998) use a indsigt approac similar to tat applied in tis paper and Francis et al. (2000) discount forecasted variables to explain te cross sectional variation in prices. Lee et al. (1999) set out to explain te timeseries relation between intrinsic value and sare prices. Eac study finds tat residual income metrics provide te best explanation of market prices. Tis study extends tis researc by examining bot US and UK data and by extending te metric set to include Stern Stewart s EVA and oter conventional metrics, including (1) tose tat include/exclude extraordinary/exceptional items; (2) metrics tat are based on equity accounting profits; (3) metrics based on operating cas flows; and (4) metrics wic report separately recognised gains/losses and operating profit from continuing activity. A comparison of different metrics addresses issues under consideration in te IASB/FASB project on Financial Statement Presentation Researc Design and Testable Hypoteses We investigate te accounting quality of different performance metrics using an entity-based residual income valuation model. EVA is an example of entity-based residual income tat applies clean surplus accounting (Lee, 1996; O Hanlon and Peasnell, 1998). A long window metodology in te manner of Siller (1981) is employed. Based on fundamentals as reflected in ex post performance metrics, a range of indsigt intrinsic values are calculated by discounting different measures of performance, togeter wit a orizon term for te value of discounted future residual earnings. Valuation errors are te differences between ex post indsigt values and ex ante actual values. Tese provide te basis for investigating te earnings quality of different metrics. We begin by expressing te value of te firm ( V 0 ) in terms of future cas flows ( CF t in period t) up to a orizon date plus a terminal value for te expected value of cas flows ( V ) from te orizon to infinity, all discounted at k te weigted average cost of capital: 3 9

11 V 0 CFt V = + (1) t = 0 (1 + k ) (1 + k ) t t Application of clean surplus accounting provides a link to te residual income valuation model for different clean surplus accounting metods i were cas flow is defined as net distributions to sareolders and debt-olders and earnings before interest is defined as: i i i EBIt = TAt + CFt TAt 1 (2) were i EBI t is earnings before interest carges but after tax and i TA t is book value of total assets at te end of te period. Residual income (RI) is ten defined in te usual way as: RI = EBI k TA 1 (3) i t i t t i t It follows from (3) tat we can rewrite (1) as: V RI V TA (4) i i t 0 TA0 = + t t= 0 (1 + kt) (1 + k) Next we expand te set of accounting metods to include dirty surplus alternatives j. Ten, EBI = TA + CF TA + DIRTY (5) j j j j t t t t 1 t and j DIRTY t is dirty-surplus flows tat ave bypassed earnings. Including dirty flows allows us to define income wic is consistent wit te clean surplus requirements of te residual income model. Relaxing te clean surplus requirement in equation (5) and rewriting equation (3) gives us: 10

12 j j j j DIRTY _ RIt = EBIt DIRTYt ( kt* TAt 1) (6) Were j DIRTY _ RI represents residual income calculated using dirty flows (e.g., earnings t before extraordinary items or foreign currency translation differences). Now we ave: V DIRTY _ RI V TA DIRTY j j j j t t 0 TA0 = + + t t t= 1 (1 + kt) (1 + k) t= 1 (1 + kt) (7) If we include te final term in (7), we get te same result as in (4). Next, valuation differences are measured relative to te opening and closing MVA for a bencmark accounting metod b and for tis purpose we coose Stern Stewart s clean surplus measure of MVA. To identify te source of valuation differences between te baseline metod and clean surplus metods i we rewrite (4) as: i b b RIt V TA i V0 TA0 = + + TADIFF (8a) t = 1 (1 + k ) (1 + k ) t t and for dirty surplus metods j we rewrite (7) as: DIRTY _ RI V TA DIRTY V TA = TADIFF b 0 0 j b j t t j t (8b) t t= 1 (1 + kt) (1 + k) t= 1 (1 + kt) b TA TA were TADIFF = ( TA0 TA0) (1 + k ) a a a b and a = i or j for clean and dirty surplus models, respectively. Note tat it follows from (4), (8a), and (8b) tat excess value, or V TA is te same for te bencmark (b), oter clean surplus (i) and dirty surplus (j) 0 0 accounting metods. 11

13 If we coose Stern Stewart s MVA as te bencmark model b, ten V TA = MVA. Using Stern Stewart s MVA as a bencmark model allows te source of b b value differences in RI and TA for clean (i), or dirty surplus (j) accounting metods to be identified and measured against te Stern Stewart bencmark b. We operationalise te insigts obtained from equations (1) to (8b) by calculating, wit indsigt, te ex post excess values for different metrics using actual realisations for income flows and interest rates and te actual orizon value (MVA) for te bencmark model. Under uncertainty, actual (ex post) income flows and discount rates will differ from expected (ex ante) values up to and including te orizon. We denote te actual (indsigt) values by ^. Tus, te indsigt value for Stern Stewart s MVA, EVA, terminal value and discount rates over any given orizon is: MVA ˆ EVA ˆ MVA ˆ (9) b t 0 = + ˆ t ˆ t= 1 (1 + kt) (1 + k) and te indsigt valuation error, given by te difference between actual (indsigt) and expected values for flows and discount rates for Stern Stewart s MVA is: ˆ b Valuation error b = MVA b 0 MVA0 (10) Equations (8a) and (8b) suggest tat valuation estimates could differ because of: (1) differences in te frequency or magnitude of dirty surplus flows; (2) differences in te measurement of assets; or (3) a combination of bot. To igligt te impact of incorrect expectations for flows and interest rates we rewrite te valuation error for Stern Stewart s MVA as: 12

14 MVA ˆ ( EVA ˆ ) ( EVA ) ( MVA ˆ ) ( MVA ) b b b b b b t t 0 MVA0 = + ˆ t t ˆ t= 1 (1 + k ) t 1 (1 kt) (1 ) (1 k) t = + + k + (11) and for oter clean surplus metods i as: ( ˆ ) ( ) ( ˆ ˆ RI EVA MVA ) ( MVA ) MVA MVA TADIFF i b b b i b t t i 0 0 = + + ˆ t t ˆ t= 1 (1 + k ) t 1 (1 kt) (1 ) (1 k) t = + + k + (12) Finally, te difference in valuation errors between te bencmark metod and dirty surplus metods j is: MVA ˆ t= 1 ( DIRTY _ RIˆ ) ( EVA ) ( MVA ˆ ) ( MVA ) = + + j b b j b t t 0 MVA0 ˆ t t ˆ t= 1 (1 + k ) t 1 (1 kt ) (1 ) (1 k) t = + + k + DIRTYt (1 ˆ t + k ) t j + TADIFF j (13) As is te case for te ex ante equations (4), (8a) and (8b), te valuation errors in equations (11), (12) and (13) are equal. However, our primary focus in tis paper is on ow well te forecasting needs of investors are served by different accounting measures of performance. Te use of MVA as a bencmark allows us to compare different errors tus: ( MVA ˆ MVA ) ( MVA ˆ MVA ) (14) b b a b and te difference in valuation errors between MVA and a callenger metric a is given by: ˆ b 0 ˆ a MVA - MVA 0 (15) were a is eiter a clean or dirty surplus metod. 13

15 We follow equation (9) and measure ex post MVA ˆ a 0 as: MVA ˆ Iˆ MVA ˆ (16) a b a t 0 = + ˆ t ˆ t= 1 (1 + kt) (1 + k) were I ˆa t is actual clean or dirty surplus income for te callenger metods. To estimate te contribution of income flows to valuation errors, we assume tat te difference between expected and actual terminal asset values for te bencmark metod is te same for te callenger metods. 4 Negative values for equation (15) indicate te bencmark metod is of iger quality and more decision useful, consistent wit a greater ease of forecasting compared to te alternative model. Rater tan focussing primarily on metric-specific errors, we focus on error rankings across metrics. We interpret tese rankings as indicating decision usefulness measured by te relative extent to wic metric-specific realisations reflect te accounting data on wic investors confirm and revise teir expectations for te purpose of setting security prices. Some accounting metrics may be easier to forecast tan oters and tus will be more appealing to investors, and we expect tis to be reflected in lower valuation errors. Our ranking of accounting quality and te decision usefulness of different performance metrics is based on te mean and variability of te respective valuation errors and pair-wise comparisons between errors using windows of 3, 5 and 10 years. Te longer te window, te greater is te influence of te caracteristics of te respective performance metrics. To control for bias relating to te coice of any particular start or terminal date, errors are calculated for all available terminal dates in te period for wic data was available. For example, in te UK (US) for te period ( ) tere are 9 (13) different errors for te 3-year windows tat begin wit ( ) and end wit For te 5-year windows in te UK (US) 7 (11) different errors are calculated and for te 10-year window 2 (6) different errors. Te errors for eac metric across te 14

16 number of sub-periods are ten averaged and tese are reported for te 3, 5 and 10-year windows as te basis for ranking te respective metric. Te extent to wic rankings are similar across different windows and terminal dates is a feature of te researc design tat provides an informal metod for assessing te reliability of te findings. Furter, by examining te absolute differences between errors for different performance measures, we directly address te ypoteses outlined in Section 4. Insert Table 1 about ere A full list of te different measures investigated in tis paper is provided in Table 1. In accordance wit te differences in ex post valuation errors between te Stern Stewart bencmark metod and oter clean and dirty surplus metods identified in equations (8a) and (8b) tese are grouped into clean and dirty surplus measures of residual income-based metods tat use Stern Stewart s estimates of capital to estimate te capital carge (Panel A) and tose were te capital carge is estimated on GAAP-based book values reported in financial statements (Panel B) and, to reflect conventional practice, metods were no capital carge for equity is recognised (Panel C). Arguments in favour of recognising a carge for te cost of equity capital in financial accounting performance metrics ave been made by Edwards and Bell (1961), Antony (1975; 1983), Edwards (1977; 1980) and Stewart (1991). For entity-based metrics, a direct test of weter investors factor a cost of equity capital into teir security pricing decisions is provided in tis paper by comparing, ceteris paribus, te valuation errors using EVA (Table 1, Panel A) and oter residual income metrics (Table 1, Panel B) to tose based on oter earnings and cas flow entity-based and mandated equity-based metrics (Table 1, Panel C). Te former include a carge for te cost of equity capital wile tose in Panel C do not. Consistent wit efficient pricing, te prediction is tat compared to tose for EVA and oter residual income metrics, te errors (means and standard deviations) for te conventional before interest alternatives will be larger. A furter prediction is tat as te lengt of te indsigt window increases, so too will te difference between te EVA and non-residual metod valuation errors. Te differences in valuation errors for EVA and NOPAT, for 15

17 example, will be solely attributable to te omission of te capital carge, wile tose between EVA and dirty surplus metods will combine te net effect of differences attributable to dirty accounting flows and book values for assets (DIRTY and TADIFF) plus te omission of te capital carge. Te ypoteses are presented in te null form. Hypotesis 1 Ceteris Paribus, tere will be no difference in valuation errors for residual income-based performance metrics tat include a carge for te cost of equity capital compared to te conventional metrics tat do not include a capital carge for equity capital. Te size of te error will be unrelated to te window used to calculate indsigt excess value. Holding constant Stern Stewart s clean surplus-based measurement of income, furter insigt into te significance of te different measures of capital (average or ending operating capital) in calculating te capital carge is obtained by comparing te respective errors for Stern Stewart-based measures of residual income EVA and RI(SS) (Table 1, Panel A) tat differ only by using Stern Stewart s measures of average or end of period operating capital. Hypotesis 2 Tere is no difference in valuation errors based on different Stern Stewart measures of capital. Considerable empasis is placed by Stern Stewart on te adjustments tey make to items in te conventional profit statements and balance seets. Compared to mandated data tese adjustments give rise to permanent and transitory differences in te recognition and measurement of assets and liabilities wit consequent impact on te performance metric (DIRTY) and asset values (TADIFF). Te effect of tese differences will be reflected in te respective valuation errors for EVA compared to oter metrics. Examples of permanent differences are te non-recognition in GAAP of non-cas costs incurred by sareolders for 16

18 te costs of employee sare options and equity sare capital. If investors treat tese items as costs ten GAAP-based performance metrics will be over-stated relative to investors expectations and positive valuation errors will arise. Tese errors will be positively related to te lengt of te window and will be greater tan te valuation errors for EVA. Te significance of transitory differences will be more evident for sorter windows. To test te accounting quality of tese adjustments compared to mandated accounting measures, valuation errors are calculated first using GAAP-based accounting performance data, but retaining te Stern Stewart measure of capital. Tese are te metrics for entity earnings (before interest) before exceptional and extraordinary items [RI(1), Table 1, Panel A], and for entity earnings after exceptional items [RI(2), Table 1, Panel A]. We ten investigate te significance of Stern Stewart s adjustments by substituting GAAP-based accounting measures of total assets employed (TAE) [RI(1-TAE), Table 1, Panel B and RI(2- TAE)], Table 1, Panel B) for estimating te capital carge. A major focus of te Stern Stewart accounting adjustments is te correction of wat are regarded as errors in te GAAP-based measurement of capital employed. Significant amongst tese adjustments is te capitalisation of researc expenditure, marketing and oter value creating expenditures, te amortisation of goodwill and te deduction of marketable securities and construction in progress. If tese differences are significant tey will result in lower valuation errors for te Stern Stewart metric. Hypotesis 3 Ceteris paribus, for te set of residual income metrics, tere is no difference in te valuation errors for EVA and tose for conventional entity earnings before or after exceptional items and irrespective of te definition of capital employed. A major focus of te claims of Stern Stewart for EVA and of te callenge posed by te findings of Biddle et al., (1997) is te relevance to investors of conventional equity earnings compared to EVA. Comparing valuation errors for EVA and te equity-based flow metric of 17

19 earnings for ordinary provides a basis for assessing te relevance to investors of tese metrics for te purpose of setting and revising investors expectations of MVA. A feature of te Biddle et al. (1997) researc design is te regression of an equity-based measure of sareolder returns on entity flows. In tis study, te use of te entity-based MVA metric as te ex ante market value bencmark for measuring te respective error metrics provides a test tat is consistent for entity metrics. Mandated equity-based metrics are subject to te same permanent and transitory differences as GAAP-based entity metrics, but te former are reported after te impact of financing decisions. Inspection of te differences between te valuation errors for EVA and mandated equity earnings provide a basis for assessing relative decision usefulness in terms of ease of forecasting for valuation purposes. Hypotesis 4 Ceteris paribus, tere is no difference in te valuation errors between EVA and conventional equity-based accounting profit. A significant callenger to earnings-based performance measures is cas flow reporting. To compare te relevance of earnings and cas flow, two errors are calculated using a smooted measure of cas flow from operations after tax, RI(OCF-SS) and RI(OCF-TAE), (Table 1, Panel B) based on Stern Stewart and conventional measures of capital, respectively. Hypotesis 5 Ceteris paribus, tere is no difference between Stern Stewart or conventional earnings-based metrics and a cas flow from operations metric. Finally, to cast ligt on issues relating to conventional accounting measurement (Table 1, Panel C), insigts into te debate on te provision of information on separate components of income are provided by calculating valuation errors for smooted [PBIT(1), before extraordinary items] and unsmooted [PBIT(2), after extraordinary items] accounting profit 18

20 before interest and tax, operating profit for continuing activity [OPCO(1)], continuing and acquired activities [OPCO(2)] and finally for two equity-based metrics, clean surplus income (CSUR) and total recognised gains and losses (SRGL). Hypotesis 6 Tere is no difference in te accounting quality of EVA, NOPAT and accounting based entity and equity metrics tat feature (a) all recognised gains (b) operating profit from continuing activity (c) sareolder earnings before and after exceptional items. 4. Data and Results Te starting point in our sample construction is te Stern Stewart 2002 UK and US datasets, wic were provided by Stern Stewart & Co. Te USA dataset contains 1,000 firms for a period of 16 years (1986 to 2001). Te UK dataset contains 500 firms for a period of 12 years (1990 to 2001). Firms are allocated to bot datasets if tey rank in te top N firms (N=1,000 for te US and 500 for te UK) according to MVA, measured at Not all firms, owever, are ranked, since Stern Stewart & Co. limit teir rankings to only te largest listed firms in eac country. Eac dataset contains up to 11 variables, including: (1) MVA; (2) EVA; (3) NOPAT; (4) WACC (k); and (4) ending operating capital (EOC). To supplement te datasets wit conventional accounting metrics, we extract oter variables for bot te US and UK Stern Stewart list of firms from Compustat (US) and Datastream (UK). Definitions of all metrics are given in Table 1. Naturally, not all Stern Stewart firms ave data for all years. Table 2 below sows te distribution of firm-years across te different performance metrics. Descriptive statistics are also reported for te different performance measures. Tey suggest tat in terms of EVA and alternative definitions of RI, te median firm in te UK over te period 1990 to 2001 as generated profits in excess of a carge for capital in te region of 2 to 3 million. Te equivalent results for te US, owever, tend to be sensitive to te definition of RI, but also suggest value added. As expected, te results indicate tat US firms are 19

21 substantially larger tan teir UK counterparts in terms of market value (MV) and total assets employed (GAAP-based TAE). 5 Differences in Stern Stewart s measure of ending operating capital (EOC) compared to GAAP-based total assets employed (TAE) as reported by Datastream also tend to be muc more pronounced in te US wit a mean value of $4.1 billion for te former compared to $7.8 billion for te latter. Median values, owever, are muc closer at $1.3 billion for EOC compared to $1.4 billion for TAE. Also noteworty, is te median WACC over te sample period, wic is about 1% iger in te US at 10%. Insert Table 2 about ere Te results from Table 2 suggest tat we need a treatment for outliers in calculating error rates. Several options are available, from simply deleting observations identified as outliers to winzorising. We cose te latter approac since we take te view tat wile outliers could distort results, reducing teir influence troug winzorising is preferred. We identify outliers as tose observations tat lie ±3 standard deviations from te cross-sectional mean. Values identified as outliers are simply reverted to ±3 standard deviations from te mean. Tis procedure is applied to all error metrics employed in te paper. We first report te cross-sectional mean valuation errors (i.e., predictability) and te differences between mean paired valuation errors for 3, 5 and 10-year windows (Table 3 and Table 4). Second, we report te variability of te individual valuation errors for all performance metrics (Tables 5). Tird, Appendix B provides an overview of te actual cross sectional errors and differences in errors across all windows and terminal dates. Te smaller are te mean and variability statistics for valuation errors te iger is accounting quality and decision usefulness in forecasting future values. By calculating means across all firms in te sample we rely on a portfolio effect to control for te extent to wic expectations differ from realisations at te level of te individual company. Te valuation errors reported in te tables are a combination of systematic market errors arising from market optimism (negative errors) or pessimism (positive errors) and valuation errors tat are metric specific. For any given window te starting (ex ante) and terminal MVA for eac firm are eld constant across metrics wic controls for systematic market errors. Tus, for a given window, te cross- 20

22 sectional variation in valuation errors, as identified in equation (15), is attributable to valuation differences between metrics arising from te use of dirty surplus flows, different asset values and te recognition of te cost of capital. Insert Table 3 about ere Inspection of te rankings reported in Table 3 indicates notable differences from te results reported by Biddle et al., (1997). First, for Hypotesis 1, olding te window constant to control for systemic market valuation errors, tere is no US or UK evidence to support te null of no difference between errors for residual income-based metrics compared to tose for conventional metrics. All residual-based metrics reported in te top alf of Panels A (US) and B (UK) ave lower valuation errors. Te consistency of tis finding across 3, 5 and 10-year windows confirms its robustness. A measure of te size of te error attributable to permanent differences between Stern Stewart and GAAP-based metrics, for example, te omission of te cost of equity capital, te magnitude or frequency of dirty accounting flows or a combination of bot for eac of te tree windows is reported in Table 4. Te most precise estimate of te effect on estimating intrinsic value is given by te difference between te valuation errors for EVA and Stern Stewart s measure of operating profit [NOPAT (SS)], were te difference is, oterwise, only attributable to te capital base for measuring te carge for te cost of capital. Te valuation error for EVA is always lower tan for te first five performance measures, including mandated equity earnings, wic omit a carge for te cost of equity capital. Tese differences, across all windows, are significantly different from zero using a standard t-test (similar results are obtained using a Wilcoxon signed ranked pairs test). We tus reject Hypotesis 1 tat ignoring a capital carge in GAAP-based metrics as no impact on estimates of intrinsic value. Insert Table 4 about ere To test ypotesis 2 we use tree different measures of capital employed in calculating te capital carge to investigate te significance for intrinsic valuation of te permanent and transitory differences in accounting recognition and measurement between Stern Stewart and GAAP. Tese are average capital (EVA), ending operating capital [RI (SS), 21

23 RI(1) and RI(2)], (te latter two use an accounting equivalent for NOPAT entity income, before interest and after tax) and GAAP-based total assets employed (TAE). Te negative difference between EVA and RI(SS) across all windows reported in Table 4 indicate tat average Stern Stewart capital gives lower valuation errors to end of period Stern Stewart operating capital in te US, and in te UK for te 10-year window. Te overall ranking across different windows is different. In Table 3 in te US, EVA based on average capital is ranked 5 t and based on end of period operating capital (RI SS) it is ranked 7 t. In te UK, EVA ranks 7 t for average capital and 4 t for operating capital. However, tese differences are not statistically significant and we cannot reject Hypotesis 2 tat error metrics are not sensitive te use of Stern Stewart's average or ending operating capital. Comparing valuation errors for te set of residual income metrics (ypotesis 3) identifies metric specific accounting quality. In general, tese differences are small and generally insignificant. In te US, (Table 3, Panel A), for a ranking of te summed ranks across 3, 5 and 10-year windows, te best performing metric is GAAP-based RI(1-TAE), residual income before exceptional and extraordinary items wit GAAP-based capital employed, and RI(2-TAE) residual income after exceptional and extraordinary items wit GAAP-based capital employed. Compared to te UK, it is te before exceptional and extraordinary alternative tat is consistently te best performing metric. In te UK, te best performing metrics over all windows are RI(1), residual income before exceptional and extraordinary items wit Stern Stewart end of period capital employed and RI(2), residual income after exceptional and extraordinary items wit Stern Stewart end of period capital employed. It is also te case tat RI(1), measured before exceptional and extraordinary items, dominates RI(2) in te overall ranking, wic is measured after tese items. However, as indicated in Table 4, differences between te set of residual income-based metrics are not significant in te UK, indicating tat differences in dirty flows (exceptional and extraordinary items) and asset values do not significantly impair valuation estimates. In te US, RI(1) is only significantly different for longer windows. We cannot reject Hypotesis 3 for te UK of no difference between EVA and oter residual income-based metrics. For te US, owever, 22

24 tere is evidence tat RI(TAE), based on GAAP capital employed, is superior to EVA for longer windows. In contrast to te Biddle et al., (1997) finding tat GAAP-based earnings are more value relevant tan residual-based metrics, including EVA, we reject Hypotesis 4 of no difference between EVA and GAAP-based earnings. It is clear tat te omission of a cost of capital from conventional earnings is a major source of error in estimating intrinsic value. In Tables 3, 4 and 5, for bot te UK and US, we report significant differences between EVA and EBEI (GAAP earnings before exceptional and extraordinary items) and EVA and EFO (GAAP earned for ordinary before a carge for cost of capital). In Table 3, EVA is consistently ranked iger tan EBEI (UK) and EFO (US). In Table 4, differences in errors between EVA and EBEI (UK) and EFO (US) are significant at te 1% level. Te negative sign indicates tat in eac case, EVA gives lower valuation errors. Finally, and importantly, in Table 5, EVA as lower variability in errors tan EBEI (UK) and EFO (US). Insert Table 5 about ere In common wit GAAP earnings, operating cas flow performs poorly compared to any residual-based measure and it is clearly inferior to EVA in te UK and US in its residualbased form. Tus, te null of no difference between EVA and operating cas flow in Hypotesis 5 is rejected. Tis finding is consistent wit te view tat cas flows, like dividends, are relatively more difficult to predict, ence te iger valuation errors. Te difference between GAAP-based earnings before and after exceptional items for NOPAT and PBIT reported in Table 3 are not statistically significant (Table 4). However, Stern Stewart NOPAT(SS) and GAAP-based NOPAT(1) and NOPAT(2) is superior to operating profit from continuing operations (OPCO) for 3 and 5-year windows. Finally, total recognised gains and losses reported in te UK (before a carge for te cost of equity capital) under FRS 3 is superior to NOPAT for a 5-year window. Tus we reject te null for Hypotesis 6 of no difference between EVA/NOPAT and GAAP metrics, altoug te results are mixed regarding te superiority of Stern Stewart compared to GAAP metrics. 23

25 A clearer picture emerges wen we consider te relative size of te measures of variation of te valuation errors, wic arguably are a better test of te performance of different metrics in reflecting te basis on wic inventors set expectations. Standard deviations of te absolute average valuation errors across performance metrics and te tree windows are reported in Table 5. In te US (Panel A), EVA consistently as te lowest variation for eac of te tree windows and is ranked best. Next best are tose residual-based metrics tat include Stern Stewart measures of capital employed. Tese findings are consistent wit relatively greater ease of forecasting using Stern Stewart accounting practices. It is noticeable tat te metrics comprising mandated accounting earnings and capital values, RI(1-TAE), perform poorly, as do te accounting metrics tat are measured before recognition of a cost of equity capital. Te UK results are reported in Table 5, Panel B. Here metrics tat combine accounting-based residual income wit Stern Stewart capital employed ave te lowest variation across te tree windows. EVA is ranked 3 rd. Te UK results provide evidence tat Stern Stewart capital measures contribute to forecasting accuracy. It is also noticeable tat te accounting-based NOPAT(1) metric performs well for sort windows. Te accounting-based residual metrics are in te middle of te rankings and te poorest performing metrics are te GAAP-based metrics tat ignore a carge for te cost of equity capital. 5. Conclusions In tis paper we contribute to te existing empirical literature on te quality of earnings measurement and presentation by using Stern Stewart data on EVA, MVA and te cost of capital in te context of a long window metodology. Te quality of earnings and decision usefulness in terms of forecasting accuracy is assessed by te mean and variability of valuation errors for different measures of earnings. Compared to te residual income valuation model, conventional accounting practice is limited by te non-recognition of te cost of equity capital. 24

26 A feature of Stern Stewart s measure of EVA is tat it satisfies two conditions for te residual income valuation model. First, by recognising a carge for te cost of equity capital and second, by te application of clean surplus accounting. In an earlier study, Biddle et al., (1997) investigate weter EVA is more igly associated wit sare returns and firm value tan mandated earnings. Contrary to te claims of Stern Stewart, mandated earnings ad te iger association and, furter, te inclusion of a capital carge ad little incremental explanatory power. However, te Biddle et al. (1997) study was subject to te limitations of considering only te contemporaneous one-period relation between returns and earnings. Te present study applies a metodology adapted from Sciller (1981) and Penman and Sougiannis (1998) tat measures valuation errors for long windows based on te difference between te ex post or indsigt values for different metrics and te ex ante equivalent observed value. Tese differences are described as valuation errors and provide te basis for assessing te ease of forecasting different accounting practices. Te results of te study are different to tose reported by Biddle et al. (1997), but consistent wit US findings in Penman and Sougiannis (1998), Lee et al. (1999) and Francis et al. (1999) regarding te superiority of te residual income valuation model. First, consistent wit teory te set of residual-based metrics, including EVA, are superior to GAAP-based metrics tat do not include a carge for te non-cas cost of equity capital. Second, residual-based metrics, including EVA, are superior for forecasting purposes to operating cas flow-based metrics. Tird, differences between EVA and residual-based measures are small. Tere is no significant difference between using Stern Stewart s measure of average or ending operating capital for te capital carge. An all, GAAP-based residualbased metrics gives te smallest valuation errors in te US, but ave relatively ig variation. Overall, EVA emerges from tese tests in a favourable ligt. Recognition of a carge for te cost of equity capital clearly improves te accuracy of estimates of intrinsic value. Altoug its valuation errors are not significantly lower tan oter residual income-based metrics, EVA does ave te lowest variation of valuation errors in te US indicating relative forecasting accuracy. 25

27 Our findings ave two main policy implications for improving te quality of earnings measurement reported in financial statements. First, te evidence reported in tis paper warrants active consideration by accounting regulators of mandating te recognition of a cost of equity capital wen reporting equity income in financial statements. Consideration sould also be given to weter tis can be accomplised on a pro forma basis, as part of a managerial analysis of performance or, as an integrated component of te financial accounting double entry system (Antony, 1976). Secondly, te finding of a lower variability, and ence greater reliability for forecasting purposes, of te valuation error for EVA in te US warrants consideration of te elements of te Stern Stewart agenda for te reform of GAAP wic ave not yet been adopted. 26