Top Banner
Classificatory income smoothing: The impact of a change in regime of reporting financial performance Vasiliki E. Athanasakou, Norman C. Strong, Martin Walker * Manchester Business School, The University of Manchester, MBS Crawford House, Manchester M13 9PL, UK Abstract Financial Reporting Standard No. 3 (FRS3) regulated the reporting of financial perfor- mance by UK firms from 1993 until the adoption of International Financial Reporting Standards in 2005. FRS3 outlawed extraordinary items, but allowed a clearer distinction between recurring and transitory income by giving firms discretion over the classifications of unusual (i.e. exceptional) items and the option to disclose alternative EPS. Through these provisions FRS3 increased the scope for classificatory choices as a means to highlight persistent profitability. We examine the impact of FRS3 on classificatory smoothing by UK firms and document a significant rise in this practice post-FRS3. We find that this increase is due mainly to deviations of net income from expected earnings inducing a sig- nificantly higher level of classificatory smoothing post-FRS3. Additional analysis shows that earnings are substantially more persistent at the pre-exceptional level post-FRS3. Overall, our results suggest greater use of classificatory choices to highlight sustainable profitability after the change in performance reporting regime. Ó 2007 Elsevier Inc. All rights reserved. 0278-4254/$ - see front matter Ó 2007 Elsevier Inc. All rights reserved. doi:10.1016/j.jaccpubpol.2007.05.002 * Corresponding author. Tel.: +44 (0) 161 2754008; fax: +44 061 2754023. E-mail address: [email protected] (M. Walker). Journal of Accounting and Public Policy 26 (2007) 387–435 www.elsevier.com/locate/jaccpubpol
49

Classificatory income smoothing: The impact of a change in ... Classificatory income smoothing The impact of a change in...alongside the basic figure. In all, FRS3 increased the

Jan 30, 2020

Download

Documents

dariahiddleston
Welcome message from author
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Page 1: Classificatory income smoothing: The impact of a change in ... Classificatory income smoothing The impact of a change in...alongside the basic figure. In all, FRS3 increased the

Journal of Accounting and Public Policy 26 (2007) 387–435

www.elsevier.com/locate/jaccpubpol

Classificatory income smoothing:The impact of a change in regime of

reporting financial performance

Vasiliki E. Athanasakou, Norman C. Strong,Martin Walker *

Manchester Business School, The University of Manchester, MBS Crawford House,

Manchester M13 9PL, UK

Abstract

Financial Reporting Standard No. 3 (FRS3) regulated the reporting of financial perfor-mance by UK firms from 1993 until the adoption of International Financial ReportingStandards in 2005. FRS3 outlawed extraordinary items, but allowed a clearer distinctionbetween recurring and transitory income by giving firms discretion over the classificationsof unusual (i.e. exceptional) items and the option to disclose alternative EPS. Throughthese provisions FRS3 increased the scope for classificatory choices as a means to highlightpersistent profitability. We examine the impact of FRS3 on classificatory smoothing byUK firms and document a significant rise in this practice post-FRS3. We find that thisincrease is due mainly to deviations of net income from expected earnings inducing a sig-nificantly higher level of classificatory smoothing post-FRS3. Additional analysis showsthat earnings are substantially more persistent at the pre-exceptional level post-FRS3.Overall, our results suggest greater use of classificatory choices to highlight sustainableprofitability after the change in performance reporting regime.� 2007 Elsevier Inc. All rights reserved.

0278-4254/$ - see front matter � 2007 Elsevier Inc. All rights reserved.doi:10.1016/j.jaccpubpol.2007.05.002

* Corresponding author. Tel.: +44 (0) 161 2754008; fax: +44 061 2754023.E-mail address: [email protected] (M. Walker).

Page 2: Classificatory income smoothing: The impact of a change in ... Classificatory income smoothing The impact of a change in...alongside the basic figure. In all, FRS3 increased the

388 V.E. Athanasakou et al. / Journal of Accounting and Public Policy 26 (2007) 387–435

JEL classification: M41

Keywords: Classificatory income smoothing; Extraordinary and exceptional items; Sustainableprofitability; Earnings persistence

1. Introduction

In their current convergence project on financial statement presentation, theaim of the International Accounting Standards Board (IASB) and the USFinancial Accounting Standards Board (FASB) is to establish a common stan-dard for presentation of information in financial statements that improves theability of investors to appraise current performance and forecast future profit-ability. The two boards have decided that financial statements should include asingle statement of earnings and comprehensive income. They have also agreedon a set of presentation principles, but have yet to determine a commonapproach for disaggregating financial information and the totals and subtotalsto report in the income statement (IASB, 2006). At the same time, US regula-tors are concerned about increasing investor reliance on adjusted earnings (e.g.pro forma or street earnings) disclosed in earnings announcements, as thesedisclosures give managers considerable discretion over which items to includeor exclude from the alternative earnings definition. In the light of these currentdevelopments, we conduct a timely policy experiment on the potential effect ofdisaggregating income components and highlighting alternative measures ofperformance within the income statement. We explore the impact of FinancialReporting Standard No. 3 (FRS3): Reporting Financial Performance on thepractice of income smoothing via classifications of transitory items as a wayof highlighting recurrent profitability by UK firms.

The UK Accounting Standards Board issued FRS3 in 1992, changing rad-ically how UK firms report financial performance. Before FRS3, SSAP3required firms to report basic EPS on ordinary income after exceptional itemsand taxation, while SSAP6 offered a flexible definition of extraordinary items.1

This encouraged inconsistent classifications of items below ordinary incomeand classificatory smoothing through extraordinary items. It raised concernsamong UK authorities that companies were increasingly classifying transitoryitems as extraordinary mainly when they were expenses or losses, and as excep-tional (above ordinary income) when they were revenue or profits. FRS3 effec-tively reclassified all extraordinary items as exceptional and mandatedcalculation of basic EPS on net income, thereby disabling classifications of

1 FRS3 amended Statement of Standard Accounting Practice No. 3 (SSAP3), issued by the ASB’spredecessor, the Accounting Standards Committee (ASC), and superseded SSAP6.

Page 3: Classificatory income smoothing: The impact of a change in ... Classificatory income smoothing The impact of a change in...alongside the basic figure. In all, FRS3 increased the

V.E. Athanasakou et al. / Journal of Accounting and Public Policy 26 (2007) 387–435 389

extraordinary items to smooth or increase EPS. At the same time, based on therationale that a single number (basic EPS) cannot summarize the performanceof a complex organization, FRS3 encouraged an ‘information set’ approach tofacilitate the analysis and interpretation of various aspects of performance. Inaddition to requiring firms to highlight a range of important income compo-nents in a ‘layered’ format, FRS3 contained new provisions for items of anunusual size or nature (exceptional items) to allow for more informative dis-tinctions between recurring and non-recurring income and between operatingand non-operating income. FRS3 further allowed firms to disclose alternativemeasures of EPS as long as they reconciled any alternative EPS to the basic fig-ure, disclosed it consistently over time, and gave it no greater prominence in theannual report than basic EPS. The option to disclose alternative EPS consti-tuted a major change in the reporting practices of UK firms and a distinctiveGAAP provision, as US GAAP does not allow disclosures of pro forma earn-ings alongside basic EPS, and the IASB and the FASB have decided not toaddress pro forma measures in their convergence project.

FRS3’s provisions enabled UK firms to smooth profit and loss sub-totals viaclassifications of exceptional items, having several implications for this prac-tice. First, instead of a uniform list of exceptional items, FRS3 provided a widedefinition of exceptional items encouraging management to classify these itemsaccording to the nature of the firm’s activities. Wider flexibility in classificatorychoices affected primarily operating exceptional items, which firms needed todistinguish from non-operating exceptionals. Second, firms could report alter-native EPS on earnings before all exceptional items, not just those formerlyclassified as extraordinary. Third, FRS3 enhanced substantially the disclosureof exceptional items and of any alternative EPS metrics firms chose to presentalongside the basic figure. In all, FRS3 increased the flexibility and disclosureof classificatory choices.

In examining the impact of FRS3 on classificatory smoothing, we focus onthe information theory of accounting choice (Barnea et al., 1976; Holthausenand Leftwich, 1983), and assess whether FRS3’s provisions increased the scopefor classificatory choices to highlight sustainable profitability. The informationtheory links directly to the regulatory framework for reporting financial perfor-mance. This is because the theory addresses the issue of information asymme-try and limited communication between managers and the market (Fields et al.,2001), while the regulatory framework impacts upon the channels of commu-nication. The information theory links particularly to FRS3, as FRS3 enrichedthe information set of the income statement to enable invertors to derive amore meaningful indicator of sustainable performance. In addition, the infor-mation theory assumes that investors can identify smoothing techniques, andFRS3 coupled greater flexibility in classificatory choices with rigorous trans-parency restrictions. So while we allow for the possibility that greater flexibilityin classificatory choices under FRS3 could have encouraged opportunistic clas-

Page 4: Classificatory income smoothing: The impact of a change in ... Classificatory income smoothing The impact of a change in...alongside the basic figure. In all, FRS3 increased the

390 V.E. Athanasakou et al. / Journal of Accounting and Public Policy 26 (2007) 387–435

sifications of exceptional items to inflate core earnings, we consider this a sec-ond-order effect. This is because the transparency requirements of FRS3restricted the scope for opportunistic classifications by increasing the probabil-ity that investors identify misclassifications of exceptional items.

For the pre-FRS3 period we examine classificatory smoothing via extraor-dinary items. Post-FRS3, in line with the gradual increase in the number offirms reporting alternative EPS on earnings before exceptional items (Choiet al., 2005), we examine classificatory smoothing via exceptional items. Werefer to extraordinary and exceptional items as classification items (CIs). Wemeasure classificatory smoothing as the extent to which removing CIs reducesthe variability of reported earnings. Consistent with our predictions, our resultsreveal a significant increase in the practice of classificatory smoothing post-FRS3. The rise is due to absolute unexpected earnings inducing a higher levelof classificatory smoothing through exceptional items post-FRS3. These resultspersist after controlling for other smoothing incentives and factors affecting thevariation of our measure of classificatory smoothing. They are also robust toalternative measures, test specifications, and explanations. We complementthese findings with earnings persistence tests, which show that removing classi-fication items increases earnings persistence to a greater extent post-FRS3 rel-ative to the pre-FRS3 period.

Our study makes several contributions to the earnings management litera-ture. First, we provide valuable information for accounting standard settersby shedding light on the impact of regulatory intervention on firms’ incomesmoothing practices. Consistent with our hypotheses, we find that FRS3brought about an increase in the practice of classificatory smoothing. In addi-tion to providing important insights to the IASB and the FASB for their cur-rent project on financial statement presentation, our study is the first toexamine the impact of a change in financial performance reporting regimeon classificatory income smoothing, directly responding to the call of Healyand Wahlen (1999) for research on the impact of accounting regulation onfirms’ financial reporting choices.

Second, we extend prior research on the impact of FRS3 on financial com-munication. Acker et al. (2002) and Lin (2002) find that by increasing the infor-mation content of the income statement, FRS3 improved the accuracy ofanalyst forecasts. Lin and Walker (2000) find that post-FRS3 firms conveyedmore value relevant earnings measures. We add to these findings with evidencethat FRS3 enhanced the role of classificatory choices in highlighting permanentprofitability.

Third, our results offer insights into the ongoing debate on the role ofadjusted earnings (Bradshaw and Sloan, 2002; Bhattacharya et al., 2003;Brown and Sivakumar, 2003; Lougee and Marquardt, 2004; Choi et al.,2005). Consistent with evidence on the superior value relevance of earningsbefore non-recurring items (Elliott and Hanna, 1996; Bradshaw and Sloan,

Page 5: Classificatory income smoothing: The impact of a change in ... Classificatory income smoothing The impact of a change in...alongside the basic figure. In all, FRS3 increased the

V.E. Athanasakou et al. / Journal of Accounting and Public Policy 26 (2007) 387–435 391

2002; Eames and Sepe, 2005), we find that extraordinary and exceptional itemsdo not affect core income 1-year ahead and that removing these items results inmore persistent earnings measures than net income, especially post-FRS3.

Fourth, we add to prior research on classificatory smoothing, which in con-trast to research on abnormal working capital accruals, focuses mainly onsmall datasets (Beattie et al., 1994; Godfrey and Jones, 1999). We extend Beat-tie et al. (1994), who examine classificatory choices within an incentives-basedframework, by providing evidence based on a large sample. We also develop analternative measure of classificatory smoothing that relaxes the assumption ofa single smoothing object across firms. This is necessary, as disclosing alterna-tive EPS was optional under FRS3 and firms could choose any level of profit asan alternative to net income.

2. Prior literature, institutional overview, and development of hypotheses

2.1. Literature review

The income smoothing literature begins with the income smoothing hypoth-esis of Gordon (1964). Within his framework income smoothing arises asrational behavior based on the assumptions that: (a) managers maximize theirutility; (b) managerial utility depends on firm value and shareholder satisfac-tion; and (c) shareholder satisfaction and stock price increase with earningsgrowth and stability. This literature implicitly assumes the market is inefficient,i.e. it functionally fixates on bottom line earnings, irrespective of the account-ing choices that determine income. However, it is not necessary for the marketto be inefficient for income smoothing behavior to exist. It is sufficient thatmanagers believe the market is inefficient. In either case, managers smoothincome to reduce the riskiness of the firm arising from a perceived cause-and-effect relation between earnings fluctuations and market risk (Moses,1987, p. 366). Lev and Kunitzky (1974) provide evidence that the extent ofincome smoothing correlates with both total and systematic risk.

In contrast to this original income smoothing hypothesis, positive account-ing theory (PAT), assumes the market is efficient and can detect smoothingbehavior (Watts and Zimmerman, 1978). PAT rationalizes earnings manage-ment through positive contracting costs. Performance-related contracts inducemanagers to make accounting choices to maximize their own wealth or to min-imize agency costs with contracting parties. Alternatively, managers makeaccounting choices to reduce political costs associated with public visibilityor regulation. The earliest smoothing study adopting a positive accountingframework is Ronen and Sadan (1981), who examine income smoothingbehavior arising from managers’ desire to maximize their utility and to reducethe costs of potential litigation. Using a similar framework, Moses (1987)

Page 6: Classificatory income smoothing: The impact of a change in ... Classificatory income smoothing The impact of a change in...alongside the basic figure. In all, FRS3 increased the

392 V.E. Athanasakou et al. / Journal of Accounting and Public Policy 26 (2007) 387–435

examines income smoothing through accounting policy changes and docu-ments a significant association between smoothing and proxies for politicalcosts and agency costs of equity.

As an alternative to PAT, the information theory of accounting choice, ascoined by Holthausen and Leftwich (1983), identifies smoothing as a meansfor managers to convey their expectations for future profitability within theframework of conventional accounting practices, which do not allow disclo-sures of earnings forecasts. Barnea et al. (1976) were the first to examinesmoothing of ordinary income through extraordinary items to improve the pre-diction of future earnings. Under the information theory of accounting choice,the market is efficient and for smoothing to be effective investors must observemanagers’ smoothing techniques.

How do firms smooth income? Ronen and Sadan (1981) show that manag-ers can smooth income either inter-temporally by timing transactions or byallocating the effect of transactions over time to change the income of particu-lar periods, or via classifications of non-recurring items. Classificatory smooth-ing is feasible when the earnings figure managers seek to smooth—thesmoothing object—is a level of profit other than net income. Barnea et al.(1976) and Ronen and Sadan (1981) argue that ordinary income is a desirablesmoothing object, as transitory income components leave it unaffected and itdominates net income in predicting future profitability. Michelson et al.(1995) find that firms that smooth ordinary income have significantly lowerbetas and higher equity market values. Using an incentives-based framework,Beattie et al. (1994) examine classificatory smoothing via extraordinary itemsby UK firms. Consistent with Moses (1987), they find a significant associationbetween classificatory smoothing, agency costs, and accounting risk. Godfreyand Jones (1999) verify the link between political costs and classificatoryincome smoothing. Using the 1989 redefinition of ordinary activities for Aus-tralian firms to construct a measure of classificatory smoothing throughextraordinary items, they find that managers of companies with high laborrelated political costs (high employee union membership) smooth net operatingprofit through classifications of recurring gains and losses.

Recent research investigates adjusted earnings disclosures as a means ofincreasing the predictive power of reported earnings for future performance.Such disclosures enable a form of classificatory smoothing that is less restric-tive than classificatory smoothing through extraordinary items, as managersmay exclude any income components they consider transitory from theadjusted earnings figure. Evidence from Bradshaw and Sloan (2002) and Bhat-tacharya et al. (2003) suggests that pro forma earnings disclosures by US firmsare more informative and more persistent than GAAP earnings. However, thead hoc nature of pro forma disclosures and the fact that they are outside theaudited profit and loss statement have raised concerns about opportunisticclassificatory choices. Doyle et al. (2003) and Landsman et al. (2007) substan-

Page 7: Classificatory income smoothing: The impact of a change in ... Classificatory income smoothing The impact of a change in...alongside the basic figure. In all, FRS3 increased the

Pre-FRS3 Post-FRS3

Turnover Turnover

Cost of Sales Cost of sales

Operating exceptional items

Gross Profit Gross Profit

Net operating expenses Net operating expenses

Operating exceptional items

Operating profit Operating profit

Exceptional items Non-operating exceptional items:

Profits or losses on the sale or termination of an operation

Costs of a fundamental reorganization or restructuring

Profits or losses on the disposal of fixed assets

Taxation Taxation

Ordinary profit Ordinary profit

Extraordinary items (after tax) Extraordinary items (after tax)

Net profit Net profit

Basic EPS

BasicEPS

AlternativeEPS

Fig. 1. Presentation requirements for extraordinary and exceptional items and EPS calculationwithin the profit and loss statement pre- and post-FRS3 for UK companies.

V.E. Athanasakou et al. / Journal of Accounting and Public Policy 26 (2007) 387–435 393

tiate these concerns with evidence that US firms remove value relevant itemsfrom pro forma earnings.

The main difference in adjusted earnings reporting between the UK and theUS over our sample period is that FRS3 regulates UK practice. We considerthis difference in examining classificatory smoothing by UK firms post-FRS3. Similar to Beattie et al. (1994) and Godfrey and Jones (1999), we usean incentives-based framework to examine classificatory smoothing. Unlikethese two studies, which are constrained by the use of small datasets due toresearch design limitations,2 we use a large panel of UK firms, as we seek toexamine the impact of FRS3 on the overall practice of classificatorysmoothing.

2.2. FRS3 and classificatory choices

Fig. 1 illustrates the main changes in presentation requirements for extraor-dinary and exceptional items, and EPS calculation following the introductionof FRS3. SSAP6 required firms to disclose extraordinary items separately inthe income statement after profit and loss on ordinary activities. As EPS wascalculated on ordinary profit after exceptional items and taxation, many firmswere tempted to shift exceptional items to extraordinary when the former were

2 Beattie et al. (1994) use hand collected data from annual reports for 163 companies included inthe 1989–90 survey of UK published accounts that disclosed extraordinary and exceptional items.Godfrey and Jones (1999) use a sample of 58 Australian listed firms that restated their 1989extraordinary item figure in their 1990 accounts, and had available data for 1980–90.

Page 8: Classificatory income smoothing: The impact of a change in ... Classificatory income smoothing The impact of a change in...alongside the basic figure. In all, FRS3 increased the

394 V.E. Athanasakou et al. / Journal of Accounting and Public Policy 26 (2007) 387–435

losses. To restrict potential misclassifications of extraordinary items, a revisionof SSAP6 redefined ordinary activities as activities that the company under-takes usually, frequently or regularly and provided a list of specific items thatfirms could classify as extraordinary.3 Similar to the revised SSAP6, FRS3defined extraordinary items as material items arising from events or transac-tions falling outside the ordinary activities of the firms. However, FRS3 rede-fined ordinary activities4 to include the effects of any event irrespective of itsfrequency or unusual nature. Furthermore, FRS3 referred to extraordinaryitems as items possessing a high degree of abnormality and considered themso rare that it provided no examples. In effect, FRS3 outlawed extraordinaryitems by requiring firms to disclose as exceptional any items previously classi-fied as extraordinary.

While reclassifying extraordinary items as exceptional, FRS3 increased flex-ibility in classificatory choices over exceptional items by enhancing their disclo-sure and widening their definition. With regard to the disclosure of exceptionalitems, instead of being aggregated in one heading under SSAP6, FRS3 requiredfirms to distinguish between operating and non-operating exceptionals. FRS3required disclosure of non-operating exceptionals under separate headingsafter operating profit. Non-operating exceptional items included three catego-ries: (a) profits or losses on the sale or termination of an operation; (b) costs ofa fundamental reorganization or restructuring; and (c) profits or losses on thedisposal of fixed assets. FRS3 required firms to include any exceptional itemfalling outside these three categories under the statutory format heading towhich it related, and to disclose it either in a note or, if necessary to give a trueand fair view, on the face of the profit and loss statement. In practice, mostfirms disclosed exceptional items on the face of the profit and loss statement(Tonkin and Skerratt, 1995). In either case, FRS3 required firms to providean adequate description for each exceptional item to allow users to understandits nature and to indicate whether it related to continuing or discontinuedoperations.

The effect of FRS3 on exceptional items extended beyond these disclosurerequirements. With the option to disclose alternative EPS on other levels ofprofit, firms could remove exceptional items from alternative EPS to highlight

3 This list included rationalization, reorganisation and redundancy costs arising from discon-tinuance of a business segment, profits and losses on disposal of assets arising from anextraordinary event, provision for a permanent diminution in value of fixed assets, profits or lossesarising from the disposal of subsidiaries, and the effect of a change in the basis for taxation.

4 Ordinary activities are ‘any activities, which are undertaken by a reporting entity as part of itsbusiness and such related activities in which the reporting entity engages in furtherance of,incidental to, or arising from these activities. They include the effects on the reporting entity of anyevent in the various environments in which it operates, including the political, regulatory, economicand geographical environments irrespective of the frequency or unusual nature of the events’(FRS3§2).

Page 9: Classificatory income smoothing: The impact of a change in ... Classificatory income smoothing The impact of a change in...alongside the basic figure. In all, FRS3 increased the

V.E. Athanasakou et al. / Journal of Accounting and Public Policy 26 (2007) 387–435 395

a more meaningful measure of profitability. FRS3 widened the scope for clas-sificatory choices further through its broader definition of exceptional items.FRS3 defined exceptional items as any ‘material items which derive fromevents or transactions that fall within the ordinary activities of the reportingentity and . . .need to be disclosed by virtue of their size or incidence if financialstatements are to give a true and fair view’ (FRS3§5). Apart from the threenon-operating exceptional items, this definition allowed UK firms to classifyas exceptional not only highly unusual items falling within the ordinary activ-ities of the firm but also items arising regularly, but being unusually large in thecurrent year.5 Also, contrary to SSAP6’s uniform list of exceptional items,6

FRS3 did not offer examples of operating exceptionals, allowing managers toclassify items according to the nature of the firm’s operations. The broaderscope for classifying exceptional items is evident in the wide variety of itemsfirms disclosed as operating exceptional in practice. An abbreviated listincludes: reorganization costs (e.g. redundancy and rationalization costs),restructuring costs, provisions for permanent diminution in the value of prop-erty and land, provisions for business rationalization, provisions for environ-mental liabilities, current asset write downs, provisions for losses incontracts, legal costs, aborted merger costs, customer settlement costs, andgoodwill written-off. The Appendix illustrates sample disclosures of operatingexceptional items.

The enhanced disclosure of exceptional items and their wider definitionmotivated firms to adopt various approaches to presenting these items onthe face of the income statement to highlight a measure of underlying operatingperformance excluding the effect of significant one-off charges or credits. Theseapproaches included: (a) a multicolumn format, usually reporting four col-umns: before exceptional items, exceptional items, discontinued operations,and total (Appendix, Table A); (b) a single column format with boxes orsub-totals highlighting operating profit before and after exceptionals (Appen-dix, Table B); (c) a separate column segregating the effect of exceptionalsand a single column presenting remaining results from continuing and discon-tinued operations (Appendix, Table C); and (d) disclosure of exceptionals onthe face of the income statement, perhaps with additional commentary in theoperating and financial review (Appendix, Table D). An advantage of colum-nar approaches is that they allowed firms to report the total for each statutory

5 In the latter case the exceptional effect on the results was not the whole amount of the item, butonly the excess over a ‘normal’ amount. For example, where a firm revised the estimated useful lifeof an asset, the firm could classify the excess amount of the revised depreciation charge over the oldcharge, if material, as exceptional.

6 This list included redundancy and rationalization costs relating to continuing businesssegments, profits or losses on disposal of fixed assets arising from an exceptional event, abnormalbad debt charges, abnormal write-off of contract stocks, and surpluses from insurance claims.

Page 10: Classificatory income smoothing: The impact of a change in ... Classificatory income smoothing The impact of a change in...alongside the basic figure. In all, FRS3 increased the

396 V.E. Athanasakou et al. / Journal of Accounting and Public Policy 26 (2007) 387–435

heading on the face of the income statement as well as giving prominence to theresults before exceptional items. These approaches were usually combined withdisclosures of alternative EPS on pre-exceptional earnings.

2.3. Development of hypotheses

Classificatory choices offer managers an income smoothing device that hasthree advantages over inter-temporal smoothing devices (e.g. accruals, realbusiness transactions). First, they do not affect net income and so are less likelyto be challenged by auditors or trigger GAAP violations. Auditors and theboard of directors tend to react to income-increasing discretionary accountingchoices affecting bottom line earnings. Second, as they do not flow through theaccounting system, classificatory choices do not affect future income, reducingtheir cost as an income smoothing device. Third, classificatory choices do nothave tax implications.

UK firms have been able to engage in classificatory smoothing throughextraordinary items since the introduction of SSAP6 in 1974. Evidence onUK data predating the introduction of FRS3 documents the widespread useof extraordinary items as an income-smoothing device (Smith, 1992; Beattieet al., 1994). Pope and Walker (1999) report evidence of UK firms usingextraordinary items to classify bad news earnings components, mainly in theform of write-offs of large transitory losses. Consistent with this evidence, Peas-nell et al. (2000) report that pre-FRS3 the majority of extraordinary itemsreported by UK firms had the effect of increasing income, with their meanvalue approximating £2.7 million.

While eliminating extraordinary items, FRS3 increased the flexibility in clas-sificatory choices relating to exceptional items by enhancing their disclosureand widening their definition. Evidence that firms exploited this flexibilitycomes from Choi et al. (2005), who examine data for the 500 largest UK listednon-financial firms in 1994, 1996, and 2001 and document a gradual rise in thenumber of firms reporting alternative EPS on pre-exceptional earnings levels.They find that the main reason for disclosures of alternative EPS is the lowinformation content of net income for sustainable performance. Furthermore,they observe a gradual increase in the magnitude of negative exceptional itemspost-FRS3, due mainly to the growth of negative operating exceptionals. Thisevidence is consistent with an increase in total classification items post-FRS3resulting from managers’ desire to report a better indicator of sustainableprofitability.

FRS3’s disclosure requirements also increased the transparency of classifica-tory smoothing by enabling investors to assess the nature of all exceptionalitems and of any adjustments made to arrive at the alternative earnings figure.Higher disclosure helps investors and analysts identify core profitability andmake more accurate forecasts of future performance (Penman, 2003, Chapters

Page 11: Classificatory income smoothing: The impact of a change in ... Classificatory income smoothing The impact of a change in...alongside the basic figure. In all, FRS3 increased the

V.E. Athanasakou et al. / Journal of Accounting and Public Policy 26 (2007) 387–435 397

12 and 17). Acker et al. (2002) examine the effect of FRS3 on analyst forecasterrors based on pre-exceptional levels of earnings.7 They find that after the firstyear of FRS3, pre-exceptional earnings were associated with lower forecasterrors. Lin (2002) examines whether analyst forecasts reflect the post-FRS3information in unexpected earnings components. He finds that analysts revisedtheir current and future earnings forecasts to impound information containedin exceptional items and to improve the accuracy of their predictions. A signif-icant implication of higher transparency is that it constrains opportunistic clas-sificatory choices, as it increases the likelihood that investors detect attempts toremove income components opportunistically. Opportunistic firms may preferless transparent income smoothing devices. However, for firms wishing to con-vey a more sustainable measure of performance, classificatory smoothingthrough FRS3 provisions offers an efficient communication channel.

The higher flexibility and transparency of classificatory choices post-FRS3provided scope for an increase in classificatory smoothing to highlight a bettermeasure of persistent profitability. Accordingly, we predict that FRS3increased the practice of classificatory smoothing by UK firms to highlightrecurrent profitability and to signal future performance. We form two hypoth-eses, the first relating to the overall effect of FRS3 on the practice of classifica-tory smoothing, the second on the incentive underlying this effect,

Hypothesis 1: The practice of classificatory smoothing by UK firms increasedpost-FRS3.

Hypothesis 2: UK firms used classificatory smoothing to a greater extent post-FRS3 to increase earnings persistence.

3. Research design

3.1. Measuring classificatory smoothing

We measure classificatory smoothing as the extent to which removingextraordinary and exceptional items (CIs) reduces the variability of reportedearnings. We start by constructing an index of income smoothing followingEckel (1981) as

SI ¼ rEARNINGSrREV

7 Post-FRS3 analysts forecasted earnings before exceptional items, in contrast to the pre-FRS3regime in which analysts excluded only extraordinary items.

Page 12: Classificatory income smoothing: The impact of a change in ... Classificatory income smoothing The impact of a change in...alongside the basic figure. In all, FRS3 increased the

398 V.E. Athanasakou et al. / Journal of Accounting and Public Policy 26 (2007) 387–435

where rEARNINGS is the standard deviation of earnings and rREV is thestandard deviation of sales revenue.8 Scaling by sales volatility controls forthree effects. First is natural income smoothing that results from an accountingprocess that yields a genuinely smooth income stream. Second is any part ofinter-temporal smoothing incorporated in sales. Sales include changes inreceivables, which offer managers a device to smooth income. As we seek to fo-cus on classificatory smoothing, it is important to control for any inter-tempo-ral smoothing.9 Third is genuine shocks in operating performance reflected inrevenue that might materially affect earnings variability. These controls arecrucial to mitigate the effects on earnings volatility of natural and inter-tempo-ral smoothing and of genuine shocks in operating activities, in order to focuson classificatory income smoothing. Low values of SI indicate that, ceteris par-ibus, managers smooth reported earnings. We calculate the ratio at the firm le-vel and separately for the pre- and post-FRS3 periods.

To measure the level of classificatory income smoothing, we follow theframework of Ronen and Sadan (1981) on the interrelation between smoothingobjects and smoothing dimensions. In this framework, firms can smooth netincome only inter-temporally, but can smooth levels of earnings other thannet income by a combination of inter-temporal and classificatory smoothing(Fig. 2). Based on this framework, we calculate the smoothing index for netincome (EARN) and for alternative levels of earnings that are potentialsmoothing objects. We calculate a classificatory smoothing index (CSI) asthe difference between the smoothing index of net income and the smoothingindex of the smoothing object. Positive values of CSI capture smoothing ofreported earnings through classifications of extraordinary and exceptionalitems. We calculate CSI separately for the pre- and post-FRS3 periods.

Pre-FRS3, consistent with prior research (Beattie et al., 1994), we assumethe smoothing object is earnings before extraordinary items (EARNbXI), com-municated to investors through basic EPS. We calculate smoothing indices forEARN (SI1) and EARNbXI (SI2). Table 1, Panel A describes the calculations.CSI is the difference between SI1 and SI2, and captures the magnitude ofsmoothing resulting from classifications of extraordinary items.

Post-FRS3, identifying the smoothing object is not as straightforward. Theoption to disclose alternative EPS on levels of earnings other than net income

8 Following Eckel (1981), Michelson et al. (1995, p. 1182) compare the coefficient of variation ofearnings changes with the coefficient of variation of changes in sales and identify firms as incomesmoothers if the former is smaller than the latter. The coefficient of variation is a dimensionlessmeasure of variation, as it scales the standard deviation of a variable by its mean value. We do notfollow this approach because we seek a measure of the overall level of smoothing, rather than amethod to detect the relative incidence of smoothing. To mitigate concern over dimension effects,we add the magnitude of classification items to the main empirical specification (see Section 3.2).

9 As an additional control for inter-temporal smoothing we add accruals to the main specification(see Section 3.2).

Page 13: Classificatory income smoothing: The impact of a change in ... Classificatory income smoothing The impact of a change in...alongside the basic figure. In all, FRS3 increased the

Adapted from Figure 3.2 of Ronen and Sadan (1981, p.44) on the relation between smoothing objects and smoothing dimensions. When the smoothing object is operating or ordinary income, managers can smooth either inter-temporally from above or frombelow through classificatory items (e.g. classifications of depreciation, discretionary fixed charges and other charges, and non-recurring and non-operating items). Firms smooth net income only inter-temporally through all items above it.

Sales

Cost of sales

Gross Margin

Operating discretionary expenses

Operating Income

Depreciation

Discretionary fixed charges

Other charges

Non-recurring and non-operating items

Income statement items Smoothing dimensions

Ordinary income

Net income

Inter-temporal

less

less

less

less

less

less Classificatory

Fig. 2. Smoothing objects and related dimensions.

V.E. Athanasakou et al. / Journal of Accounting and Public Policy 26 (2007) 387–435 399

creates the possibility of multiple smoothing objects. We examine two smooth-ing objects: earnings before non-operating exceptional items (EARNbNOEI)and earnings before all (both non-operating and operating) exceptional items(EARNbEI). EARNbNOEI is similar to headline earnings that the UK profes-sional institute for investment analysts, the Institute of Investment Manage-ment and Research (IIMR), introduced in 1993 as an earnings metric thatfocused on the firm’s trading performance excluding most exceptional itemsformerly classified as extraordinary (mainly non-operating exceptionalitems).10 We also allow for EARNbEI as a smoothing object, because Choiet al. (2005) document a substantial increase in the frequency of firms disclos-ing alternative EPS on earnings before all exceptional items from 1994 to 1996.This increase coincides with a gradual rise in operating exceptional charges. Toaccommodate this evidence, we assume that within our sample there is a mix-ture of firms smoothing either EARNbNOEI or EARNbEI, and identify theleast volatile figure as the smoothing object.11 As this might be an invalidassumption for firms not disclosing an alternative EPS, we perform a robust-ness test (see Section 5.2.3). We calculate smoothing indices for EARNbNOEI

10 For a detail description of headline earnings introduced by the IIMR, see Lin and Walker(2000).11 We repeat our analysis based on the more rigid assumption that all firms smoothed earnings

before non-operating exceptional items post-FRS3. Though slightly weaker, because we restrictclassificatory choices to non-operating exceptional items (see Section 5.2.2), the core results remainunaltered.

Page 14: Classificatory income smoothing: The impact of a change in ... Classificatory income smoothing The impact of a change in...alongside the basic figure. In all, FRS3 increased the

Table 1Definition of variables

Period Level of earnings Smoothing index (SI) Classificatory smoothing index (CSI)

Panel A: Classificatory smoothing index (CSI)

Pre-FRS3 EARN SI1 ¼ rEARNrREV

EARNbXI SI2 ¼ rEARNbXIrREV CSI = SI1 � SI2

Post-FRS3 EARN SI1 ¼ rEARNrREV

EARNbNOEI SI3 ¼ rEARNbNOEIrREV

CSI ¼ SI1 � SI3

orCSI ¼ SI1 � SI4

9=; In this with the smoothing object

EARNbEI SI4 ¼ rEARNbEIrREV

Panel B: Definition of remaining variables

Variable Definition

EARN Earnings after extraordinary items. Pre-FRS3 EARN is DS625 + DS193, where DS625 is earnings before extraordinary itemsand DS193 is total extraordinary items. Post-FRS3 EARN is DS1087, where DS1087 is earnings after extraordinary items

EARNbXI Earnings before extraordinary items (DS625)EARNbXI & EI Earnings before extraordinary and exceptional items (DS625 � DS194), where DS194 is total exceptional itemsEARNbNOEI Earnings before non-operating exceptional items [DS1087 � (DS1083 � DS1094 � DS1097)], where DS1083 is total non-

operating exceptional items and includes profits or losses on the sale or termination of operations, costs of fundamentalreorganisation or restructuring, and profits or losses on the sale of fixed assets; DS1094 is tax on non-operating exceptionalitems; and DS1097 is the minority interest on non-operating exceptional items

EARNbEI Earnings before all exceptional items (DS1087 � DS194)Smoothing

ObjectPre-FRS3 the smoothing object is EARNbXI. Post-FRS3 the smoothing object is the least volatile of EARNbNOEI andEARNbEI

DIVERGENCE The absolute value of the difference between pre-managed earnings (EARN) and expected earnings scaled by total sales (DS104).Expected earnings is the lagged value of the smoothing object

400V

.E.

Ath

an

asa

ko

uet

al.

/J

ou

rna

lo

fA

ccou

ntin

ga

nd

Pu

blic

Po

licy2

6(

20

07

)3

87

–4

35

Page 15: Classificatory income smoothing: The impact of a change in ... Classificatory income smoothing The impact of a change in...alongside the basic figure. In all, FRS3 increased the

GAINLOSS Equals 1 if EARN < 0 and the smoothing object > 0, 0 otherwisePERMLOSS Equals 1 if the smoothing object < 0, 0 otherwiseRISK(b) The beta coefficient derived from firm specific regressions of stock return on the FTSE All Share Index return over a 60 month

window ending at the financial year end in questionLeverage (LEV) Total book value of debt (DS1301) over total assets (DS392)SIZE Decile portfolios formed each year by sorting observations into 10 groups based on the log of the lagged market value of equity

(DSHMV) (0 is the lowest, 9 the highest decile)Profitability

index (PI)Decile portfolios formed each year by sorting observations into 10 groups based on EARN (0 is the lowest, 9 the highest decile)

Change inprofitabilityindex (DPI)

Decile portfolios formed each year by sorting observations into 10 groups based on annual change in EARN (0 is the lowest, 9the highest decile)

Level impact ofclassificationitems (LI)

The income-increasing (decreasing) effect of negative (positive) extraordinary items (�XI) pre-FRS3, and of non-operatingexceptional items (�NOEI) or all exceptional items (�EI) post-FRS3, in line with the smoothing object. XI is DS193, NOEI isDS1083 � DS1094 � DS1097 and EI is DS194. LI is scaled by lagged total assets and is set to zero when CSI is zero or negative.Datastream reports exceptional and extraordinary items as negative when they are costs or losses, and positive when they arerevenues or profits

Working capitalaccruals(WCA)

Change in debtors (DS370), plus change in inventory (DS364), minus change in creditors (DS385) and change in provisions(DS380). We calculate WCA indirectly from the balance sheet since direct measurement through the cash flow statement is onlypossible in the UK for accounting periods ending after FRS1 ‘Cash Flow Statements’ in 1992. WCA is scaled by lagged totalassets.

Long-termaccruals(LTA)

EARN minus operating cash flows and WCA. Operating cash flows is operating profit before depreciation and amortizationminus working capital accruals (DS993 + DS696 �WCA) pre-FRS1, and cash flows from operations from the cash flowstatement (DS1015) post-FRS1. LTA is scaled by lagged total assets.

DS = Datastream code.

V.E

.A

tha

na

sak

ou

eta

l./

Jo

urn

al

of

Acco

un

ting

an

dP

ub

licP

olicy

26

(2

00

7)

38

7–

43

5401

Page 16: Classificatory income smoothing: The impact of a change in ... Classificatory income smoothing The impact of a change in...alongside the basic figure. In all, FRS3 increased the

402 V.E. Athanasakou et al. / Journal of Accounting and Public Policy 26 (2007) 387–435

(SI3) and EARNbEI (SI4). CSI is the difference between SI1 and SI3 or SI4

according to the smoothing object, and captures the magnitude of smoothingresulting from classifications of either non-operating exceptional items whenthe smoothing object is EARNbNOEI, or all exceptional items, when thesmoothing object is EARNbEI.

3.2. Research design for classificatory smoothing

To test our hypotheses we use descriptive statistics and two multivariatespecifications. In the first specification, Eq. (1) below, we test the effect ofFRS3 on the level of classificatory smoothing (Hypothesis 1) via the coefficienton FRS3, while in the second, Eq. (2) below, we test the incentive underlyingthis effect (Hypothesis 2) via the interaction term FRS3 · DIVERGENCE,

CSIi;p ¼ a0 þ a1FRS3i;p þ a2DIVERGENCEi;p þ a3GAINLOSSi;p

þ a4PERMLOSSi;p þ a5RISKðbÞi;p þ a6LEV i;p þ a7SIZEi;p

þ a8PIi;p þ a9DPIi;p þ a10LIi;p þ a11WCAi;p þ a12LTAi;p

þ ui;p; i ¼ 1; . . . ;N ; p ¼ 1; 2 ð1ÞCSIi;p ¼ a0 þ a1FRS3i;p þ a2aDIVERGENCEi;p þ a2bFRS3�DIVERGENCEi;p

þ a3GAINLOSSi;p þ a4PERMLOSSi;p þ a5RISKðbÞi;p þ a6LEV i;p

þ a7SIZEi;p þ a8PIi;p þ a9DPIi;p þ a10LIi;p þ a11WCAi;p þ a12LTAi;p

þ ui;p; i ¼ 1; . . . ;N ; p ¼ 1; 2 ð2Þ

CSI is our measure of classificatory smoothing, which we calculate at the firmlevel for pre- and post-FRS3 periods. In Eqs. (1) and (2) p = 1 refers to the pre-FRS3 period and p = 2 to the post-FRS3 period. These equations are thereforepanel regressions involving a cross-section of N firms, each with two observa-tions (pre- and post-FRS3). FRS3 is a GAAP regime indicator, taking the va-lue 0 when p = 1 and 1 when p = 2. The coefficient on FRS3, a1, estimates thechange in the level of classificatory smoothing attributable to FRS3. Consistentwith our first hypothesis, we expect a1 to be positive. We calculate all remainingexplanatory variables, discussed below, annually and use pre- and post-FRS3averages.

Our predictions focus on the information perspective of income smoothing.By smoothing income, managers produce a stable and predictable earningsstream to facilitate forecasts of future profitability. In this signaling frame-work, Moses (1987) argues that managers seek to report earnings that are clo-ser to expectations and that incentives to smooth income increase with thedivergence between actual and expected earnings. Accordingly, we expect apositive association between CSI and the extent to which pre-managed earn-ings diverge from expectations. We introduce DIVERGENCE as the absolute

Page 17: Classificatory income smoothing: The impact of a change in ... Classificatory income smoothing The impact of a change in...alongside the basic figure. In all, FRS3 increased the

V.E. Athanasakou et al. / Journal of Accounting and Public Policy 26 (2007) 387–435 403

value of unexpected earnings to capture this effect.12 We measure DIVER-

GENCE as the absolute value of the difference between pre-managed earningsand expected earnings scaled by total sales.13 We use EARN as pre-managedearnings, since classificatory choices do not affect net income. For expectedearnings, we assume a random walk model and choose the earnings level in linewith the smoothing object. To the extent that high deviations from expectedearnings induced classificatory smoothing pre-FRS3, we expect a2a in Eq. (2)to be positive. If UK firms used classifications of non-recurring items to agreater extent post-FRS3 to report earnings closer to expectations and high-light persistent profitability, we expect the positive association betweenDIVERGENCE and CSI to be more pronounced post-FRS3. The interactionterm, FRS3 · DIVERGENCE, captures this structural shift. Consistent withour second hypothesis, we expect a2b to be positive.

The practice of classificatory smoothing relies on a firm’s incentives to high-light adjusted earnings. We add two proxies to capture this. Lougee and Mar-quardt (2004) find that firms reporting GAAP losses are more likely to disclosepro forma earnings. This incentive is particularly keen when adjusted earningsare positive, as managers wish to highlight an indicator of the firm’s earningsgenerating ability that is unaffected by transitory losses. Choi et al. (2005) doc-ument a positive association between the likelihood of alternative EPS disclo-sure and the frequency of firms with net income losses and positive adjustedearnings.14 We add an indicator (GAINLOSS) for cases where net income isnegative while adjusted earnings are positive as a proxy for the incentive tohighlight pre-exceptional earnings. GAINLOSS also accommodates opportu-nistic classificatory choices to discard bad news earnings components and high-light a favorable measure of profitability. We expect a positive associationbetween GAINLOSS and CSI. Choi et al. (2005) further argue that firms areunlikely to disclose a negative adjusted earnings figure, as this is unlikely tobe informative about future prospects. Consistent with this claim, they findthat the probability of alternative EPS disclosure is lower when adjusted

12 DIVERGENCE also serves as a control, as deviations of pre-managed earnings fromexpectations affect the remaining smoothing incentives. Moses (1987) provides evidence thatlarger deviations from expectations are necessary to induce certain smoothing incentives (e.g.political costs).13 Table 1, Panel B gives the precise definitions of all variables.14 This evidence is consistent with opportunistic disclosures of alternative EPS to discard ‘bad

news’ and highlight a more favourable earnings number. However, Choi et al. (2006b) find that theitems (both gains and losses) managers exclude from alternative earnings are value irrelevant,consistent with alternative EPS measuring sustainable performance. The only evidence they find ofopportunism relates to gains that managers include in alternative earnings, which appear to bevalue irrelevant. The authors conclude that although opportunism may drive a subset of alternativeEPS disclosures, it is of second order importance compared with managers’ need to highlight ameasure of core profitability.

Page 18: Classificatory income smoothing: The impact of a change in ... Classificatory income smoothing The impact of a change in...alongside the basic figure. In all, FRS3 increased the

404 V.E. Athanasakou et al. / Journal of Accounting and Public Policy 26 (2007) 387–435

earnings are negative. Accordingly, as a second proxy we add an indicator ofnegative pre-exceptional earnings, PERMLOSS, and expect a negative associ-ation between PERMLOSS and CSI.

In addition to dampening deviations from expected earnings to signal futureperformance, managers smooth income to reduce risk resulting from the actualor perceived relation between earnings volatility and risk (Moses, 1987; Levand Kunitzky, 1974; Beattie et al., 1994; Michelson et al., 1995). Accordingly,we include market risk (RISK(b)). We expect a positive association betweenCSI and RISK(b), as riskier firms have greater incentives to smooth earnings.Consistent with PAT, we add gearing (LEV) and size (SIZE) as proxies foragency costs of debt and political costs. More highly levered firms are morelikely to smooth ordinary (or operating) income to avoid fluctuations thatmight breach debt covenants, especially in relation to the interest cover ratio.Even in the absence of interest cover constraints, managers may smoothincome to create an impression of financial stability. Therefore, we expect apositive association between CSI and LEV. Larger firms have greater incen-tives to dampen earnings fluctuations and minimize the costs of external inter-vention, thus we expect a positive association between CSI and SIZE.Following Ashari et al. (1994), we control for firm profitability (profitabilityindex, PI). Fluctuations in income have a more severe impact on low profitabil-ity firms giving them a stronger incentive to smooth income. We also controlfor the change in profitability (DPI) in response to White’s (1970) evidence thatfirms with declining profitability tend to smooth income.

Removing a classification item alters both the level and the variability ofreported earnings. Moses (1987) provides evidence that managers are con-cerned about the joint effect of accounting choices on earnings levels and var-iability. Since managers are concerned about both effects and some of thefactors in our model could motivate managers to adjust the level of earnings,we add a control for the impact that removing CIs has on the level of reportedincome (LI). We measure LI as the income-increasing (decreasing) effect ofnegative (positive) extraordinary items pre-FRS3, and non-operating excep-tional items or further all exceptional items post FRS3. Adding LI is particu-larly important in view of the evidence of Beattie et al. (1994) that incentives tosmooth depend on the effect of CIs on the level of earnings (i.e. this effectshould reach a specific level to trigger smoothing behavior). Including LI alsocontrols for the ability to engage in classificatory smoothing, because CIs usu-ally depend on the occurrence of specific events (e.g. restructurings, divesti-tures, events triggering fair value adjustments and changes in accountingestimates etc.). Furthermore, a positive correlation between the magnitudeand variance of CIs might cause a mechanical association between DIVER-

GENCE (or any of the remaining explanatory variables) and CSI. To mitigateconcern over mechanically induced relations, we add LI as a control for thelevel of CIs.

Page 19: Classificatory income smoothing: The impact of a change in ... Classificatory income smoothing The impact of a change in...alongside the basic figure. In all, FRS3 increased the

V.E. Athanasakou et al. / Journal of Accounting and Public Policy 26 (2007) 387–435 405

Finally, as we focus on classificatory smoothing and some CIs are booked asworking capital accruals (e.g. inventory write downs) or non-operating accru-als (e.g. provisions for restructuring costs), we control for total accruals. Wesplit total accruals into working capital accruals (WCA) and long-term accruals(LTA) to account for their different properties. Changes in working capitaloccur frequently and involve a high degree of managerial judgment in theirestimation, offering a flexible device to managers to smooth income streams.Long term-accruals represent large, visible, one-off costs or losses (e.g. write-downs, provisions for restructuring costs, impairments, losses on disposal ofassets), which are more likely to result from conservative accounting choices.Therefore, adding both accrual components controls for any variation inCSI resulting from inter-temporal smoothing through WCA or accountingconservatism through LTA. To the extent that inter-temporal smoothingthrough WCA is an alternative income smoothing mechanism to classificatorysmoothing, we expect a negative association between CSI and WCA. Consis-tent with the inverse relation between conservative reporting and incomesmoothing, we expect a positive association between LTA and CSI (e.g. largewrite-offs or impairments decreasing the extent of classificatory smoothing).

Because CSI takes zero and positive values, Eqs. (1) and (2) belong to the classof censored regression models. Even though CSI is a continuous variable overstrictly positive values, it takes the value zero with positive probability, in thesense that for some managers the optimal choice is zero classificatory smoothing.This type of response variable is a ‘corner solution outcome’ (Wooldridge, 2002,Chapter 16). Accordingly, we estimate the coefficients of Eqs. (1) and (2) using acorner solution Tobit model. To alleviate bias in the standard errors of the Tobitestimates due to heteroskedasticity or cross-sectional dependence, we use boot-strap standard errors to determine the statistical significance of the coefficients. 15

3.3. Earnings persistence tests

To explore whether UK firms use classificatory smoothing to a greater extentpost-FRS3 to increase earnings persistence, we complement our multivariateanalysis with earnings persistence tests. First, we assess the extent to which

15 Kayhan and Titman (2007) use a similar approach to control for heteroskedastic, autocorre-lated, and cross-correlated errors. To obtain bootstrap standard errors, we draw 500 bootstrapsamples from the 50 industry clusters in the original dataset. We draw samples of firm clusters toaccount for the panel structure of our data. In the random drawing, some industry clusters appearonce, others more than once, and some not at all. We apply the regression model to the bootstrapsamples to obtain coefficients. We calculate the bootstrap standard errors asPðh�i � �h�Þ=ð500� 1Þ

� �1=2; where �h� is the average bootstrap statistic and h�i is the statistic from

the ith bootstrap sample. This approach resembles block bootstrapping, which is appropriate whenthe errors are not independent or homoskedastic (MacKinnon, 2006). We obtain qualitativelysimilar results, when we draw 30 industry clusters in each replication.

Page 20: Classificatory income smoothing: The impact of a change in ... Classificatory income smoothing The impact of a change in...alongside the basic figure. In all, FRS3 increased the

406 V.E. Athanasakou et al. / Journal of Accounting and Public Policy 26 (2007) 387–435

removing classification items increases the persistence of reported earnings inthe pre- and post-FRS3 periods. Second, we test whether this effect increasesas a result of the introduction of FRS3. Initially, we estimate earnings persis-tence regressions for net income (EARN) and for different levels of earningsbefore classification items (EARNbCIs) for the pre- and post-FRS3 periods,

EARN i;tþ1 ¼ b0 þ b1EARN i;t þ vi;t ð3ÞEARNbCIsi;tþ1 ¼ c0 þ c1EARNbCIsi;t þ vi;t ð4Þ

Pre-FRS3 EARNbCIs is either earnings before extraordinary items (EARN-

bXI) or earnings before extraordinary and exceptional items (EARNbXI &EI). Post-FRS3 EARNbCIs is either earnings before non-operating exceptionalitems (EARNbNOEI) or earnings before all exceptional items (EARNbEI). Ifremoving classification items increases the persistence of earnings, we expectc1 to be larger than b1, and Eq. (4) to have higher explanatory power (adjustedR-square) than Eq. (3). To capture any structural break in earnings persistenceand the role of classification items post-FRS3, we add classification items (CIs)to Eq. (4) and an interaction term between FRS3 and the earnings componentsas follows:

EARNbCIsi;tþ1 ¼ d0 þ d1EARNbCIsi;t þ d2CIsi;t þ d3FRS3i;t þ d4FRS3

� EARNbCIsi;t þ d5FRS3� CIsi;t þ ei;t ð5Þ

We estimate Eq. (5) using three combinations of EARNbCIs: (a) EARNbXI

pre-FRS3 and EARNbNOEI post-FRS3; (b) EARNbXI pre-FRS3 and EARN-

bEI post-FRS3; and (c) EARNbXI & EI pre-FRS3 and EARNbEI post-FRS3.CIs is extraordinary items (cases a and b) or further exceptional items (case c)pre-FRS3, and non-operating exceptional items (case a) or all exceptionalitems (cases b and c) post-FRS3. If EARNbCIs proxies for persistent profitabil-ity, while classification items are truly non-recurring, we expect d2 to be zero.Conversely, if classification items are value relevant items that managers re-move opportunistically to inflate EARNbCIs, then we expect d2 to be positive(current costs or losses predicting declines in future profitability). If the persis-tence of pre-exceptional earnings is higher post-FRS3, we expect d4 to be po-sitive. FRS3 · CIs captures structural shifts on the predictive ability of CIspost-FRS3. Similar to the Tobit equations, we use bootstrap standard errorsfor the OLS estimates of Eqs. (3)–(5) to control for cross-sectional dependenceand heteroskedastic and autocorrelated residuals.

4. Sample

We use data for all UK (dead and live) non-financial listed firms from Data-stream for the period 1986 until the enforcement of FRS10 in December 1998.

Page 21: Classificatory income smoothing: The impact of a change in ... Classificatory income smoothing The impact of a change in...alongside the basic figure. In all, FRS3 increased the

V.E. Athanasakou et al. / Journal of Accounting and Public Policy 26 (2007) 387–435 407

We exclude financial firms because they have a substantially different financialreporting framework. Choosing FRS10 as the sample period cut-off is impor-tant due to the potential effect of FRS10 on classificatory smoothing. FRS10required UK firms to charge amortization of goodwill and intangible assetsto the income statement, providing an additional motive to disclose alternativeEPS on pre-exceptional earnings. As our objective is to examine the effect ofFRS3, the sample period ends with the introduction of FRS10 (12/23/1998).

We employ a balanced sample to assess whether there is a structural break inthe practice of classificatory smoothing following the introduction of FRS3.With a balanced sample design, each firm has at least one observation in boththe pre- and post-FRS3 periods. Each sample firm therefore acts as its own con-trol, minimizing the effects of temporal differences in sample composition (e.g.industry representation) on the results (Peasnell et al., 2000).16 The balancedsample includes 10,646 observations for 993 firms. Earnings persistence testsrequire data 1-year ahead, reducing the observations to 9653 observations.Due to this elimination, our sample period ends in 1997. To calculate firm andperiod specific classificatory income smoothing indices, we keep firms with atleast two observations in both the pre- and post-FRS3 periods. The final sampleis 914 firms and 9222 firm-year observations (for Eqs. (3)–(5)). As we calculate thesmoothing indices at the firm level for the pre- and post-FRS3 periods, we regressEqs. (1) and (2) on a panel of 1828 observations, that is a cross-section of 914firms each with two observations (pre- and post-FRS3).

5. Results

5.1. The effect of FRS3 on classificatory income smoothing

5.1.1. Descriptive statistics

Table 2, Panel A reports descriptive statistics for extraordinary items (XI) andexceptional items (EI) for the pre-FRS3 period and for non-operating excep-tional items (NOEI) and operating exceptional items (OEI) for the post-FRS3period. Pre-FRS3 mean XI (�0.004) and EI (�0.002) are both negative and sig-nificant. Post-FRS3 mean NOEI (�0.007) and OEI (�0.005) are also negativeand significant but are larger in magnitude. Since FRS3 treated most formerXI as NOEI, we calculate their mean difference as well as the mean differencebetween EI pre-FRS3 and OEI post-FRS3. Both differences are significant, indi-cating an increase in the magnitude of negative exceptional items post-FRS3.

To provide information on variation over time, Table 2, Panel B reportsannual statistics for XI and EI pre-FRS3 and for NOEI and OEI post-FRS3. Analysis of mean XI indicates increasingly negative values starting in

16 Due to the balanced sample requirement, there is no need to control for industry effects in ourspecification; the sample has the same industry composition pre- and post-FRS3.

Page 22: Classificatory income smoothing: The impact of a change in ... Classificatory income smoothing The impact of a change in...alongside the basic figure. In all, FRS3 increased the

Table 2Descriptive analysis of classification items (CIs)

Panel A: Descriptive statistics for XI and EI pre-FRS3, NOEI and OEI post-FRS3

Period N Variable Mean Std. dev. Median

Pre-FRS3 5415 XI (a) �0.004*** 0.036 0.000EI (b) �0.002*** 0.030 0.000Total �0.007*** 0.049 0.000

Post-FRS3 3807 NOEI (c) �0.007*** 0.044 0.000OEI (d) �0.005*** 0.031 0.000Total �0.012*** 0.060 0.000

Diff (c) � (a) (p-value)A �0.003 (<0.001)Diff (d) � (b) (p-value)A �0.003 (<0.001)

Panel B: Descriptive statistics by year for XI and EI pre-FRS3, NOEI and OEI post-FRS3

Year N Variable Mean Std. dev. Median

1986 598 XI �0.003*** 0.033 0.000EI 0.001 0.018 0.000

1987 654 XI �0.002 0.035 0.000EI 0.001 0.023 0.000

1988 710 XI 0.001 0.034 0.000EI 0.002** 0.019 0.000

1989 781 XI 0.002 0.039 0.000EI 0.002*** 0.016 0.000

1990 849 XI �0.005*** 0.036 0.000EI �0.002* 0.032 0.000

1991 898 XI �0.008*** 0.032 0.000EI �0.007*** 0.037 0.000

1992 and 1993(pre)a 923 XI �0.011*** 0.037 0.000

EI �0.009*** 0.041 0.0001992 and 1993

(post)b 885 XI – – –NOEI �0.009*** 0.048 0.000OEI �0.004*** 0.030 0.000

1994 874 NOEI �0.006*** 0.042 0.000OEI �0.004*** 0.024 0.000

1995 843 NOEI �0.006*** 0.041 0.000OEI �0.006*** 0.031 0.000

1996 786 NOEI �0.006*** 0.044 0.000OEI �0.005*** 0.032 0.000

1997 421 NOEI �0.004* 0.044 0.000OEI �0.008*** 0.039 0.000

Panel C: Frequencies of positive, negative, and zero XI and EI pre-FRS3, NOEI and OEI post-FRS3

Variable Freq (%)c Freq (%)d p-ValueB Variable Freq (%)c Freq (%)d p-ValueB

Pre-FRS3

XI > 0 15.16 33.20 EI > 0 46.99 57.92XI < 0 30.50 66.80 <0.001 EI < 0 34.14 42.08 <0.001XI = 0 54.34 EI = 0 18.87N 5.415

408 V.E. Athanasakou et al. / Journal of Accounting and Public Policy 26 (2007) 387–435

Page 23: Classificatory income smoothing: The impact of a change in ... Classificatory income smoothing The impact of a change in...alongside the basic figure. In all, FRS3 increased the

Table 2 (continued)

Variable Freq (%)c Freq (%)d p-ValueB Variable Freq (%)c Freq (%)d p-ValueB

Post-FRS3

NOEI > 0 23.61 50.50 OEI > 0 41.45 50.67NOEI < 0 24.09 49.50 0.690 OEI < 0 40.35 49.33 0.463NOEI = 0 52.30 OEI = 0 18.20N 3,807

*/**/*** indicate significance at 0.1/0.05/0.01 levels (two-tailed).A p-Value corresponds to a Wilcoxon non-parametric test (two-sided) for the difference in means

between the pre- and post-FRS3 period.B p-Value corresponds to a binomial test (two-sided) for the difference in frequencies of positive

and negative non-recurring items conditional on non-zero disclosures. The sample consists of9222 firm-year observations over the period 1986–97 for 914 UK non-financial firms with atleast two observations in both the pre and post-FRS3 periods.

a Includes pre-FRS3 observations of 1992 and 1993.b Includes post-FRS3 observations of 1992 and 1993.c In relation to the total number of observations.d In relation to the total number of non-zero observations. XI is total extraordinary items

(DS193). EI is total exceptional items (DS194). NOEI is non-operating exceptional items OEI isoperating exceptional items All variables are scaled by lagged total assets.

V.E. Athanasakou et al. / Journal of Accounting and Public Policy 26 (2007) 387–435 409

1990 and reaching �0.011 in the 92/93 pre-FRS3 observations. These resultsare consistent with prior evidence on the income-increasing role of XI reportedby UK firms (Smith, 1992; Pope and Walker, 1999; Peasnell et al., 2000). Asmean earnings before extraordinary items in the 92/93 pre-FRS3 observations(not tabulated) is 0.031, the income-increasing effect of XI for this period issubstantial. EI displays a similar pattern to XI, despite the pre-FRS3 positionof exceptional items above ordinary income. Negative values of mean EI startin 1990 and reach �0.009 in the 92/93 pre-FRS3 observations.17 Post-FRS3results show that NOEI and OEI are negative throughout, reaching �0.013immediately after the implementation of FRS3 and �0.012 in 1997. The ratioof OEI to NOEI rises from 0.44 (�0.004/�0.009) directly following the imple-mentation of FRS3 to 2 (�0.008/�0.004) in 1997. This suggests an increase inthe relative magnitude of negative OEI, as Choi et al. (2005) also document.

Table 2, Panel C reports the frequency of positive, negative, and zero XI andEI for the pre-FRS3 period and NOEI and OEI for the post-FRS3 period. Forboth XI and NOEI, the majority of observations are zero in both periods (54%and 52%), indicating that a substantial proportion of firms do not report theseitems. For firms disclosing these items, the majority (67%) of XI are negative.A binomial (two-sided) test for the difference in frequencies of positive andnegative extraordinary items conditional on non-zero disclosures is highly

17 The economic recession prevailing in the UK in the early 1990s affects trends for this period.The average growth rate of real GDP for 1990–93 was approximately 0.59% compared with 4.44%for 1986–89.

Page 24: Classificatory income smoothing: The impact of a change in ... Classificatory income smoothing The impact of a change in...alongside the basic figure. In all, FRS3 increased the

410 V.E. Athanasakou et al. / Journal of Accounting and Public Policy 26 (2007) 387–435

significant, confirming the predominance of extraordinary losses. In contrast,NOEI exhibits a roughly equal frequency of positive and negative values. A bino-mial test for the overall post-FRS3 period fails to reject equality of the two fre-quencies (p = 0.690).

For EI and OEI the frequency of zero observations is 19% and 18%, indicatingthat only a small proportion of the population do not disclose such items. Wherefirms report EI pre-FRS3, the majority are positive (58%). A binomial test is sig-nificant, confirming that EI is positively skewed. This evidence appears to justifyregulatory concerns at the time over the prevalence of negative extraordinary andpositive exceptional items pre-FRS3. Finally, similar to the pattern of NOEI,OEI appears equally distributed between positive and negative values post-FRS3. A binomial test for the overall post-FRS3 period does not reject equalityof the two frequencies (p = 0.463). FRS3 therefore appears to have successfullyremoved the skewness in the distribution of classification items.

Moving to the core empirical analysis for the first hypothesis, Panel A of Table3 reports the mean and median values of the classificatory smoothing index (CSI)partitioned into the pre- and post-FRS3 periods. Pre-FRS3, CSI measuresincome smoothing from classifications of XI. Post-FRS3, CSI measures incomesmoothing from classifications of either NOEI or both NOEI and OEI. For thepre-FRS3 period, the results show a significant mean and median CSI of 0.063and 0.004. For the majority of firms (527/914 = 58%) CSI is positive. Theseresults confirm the smoothing property of XI for ordinary income documentedby Beattie et al. (1994). Post-FRS3, mean and median CSI increase to 0.157and 0.023. The difference across periods is highly significant for both the meanand the median. The fraction of uncensored observations (CSI > 0) increasesto 78% (715/914), indicating a substantial rise in the frequency of firms engagingin classificatory smoothing via exceptional items post-FRS3. Panel A also reportsthe mean and median values of the smoothing indices. Mean and median SI1

(smoothing index for net income) increase from 0.301 and 0.161 pre-FRS3 to0.364 and 0.166 post-FRS3. The increase in mean SI1 is significant.18 Additional

18 The significant increase in the smoothing index of net income is subject to differentinterpretations. First, scaling net income volatility by sales volatility may not control adequatelyfor genuine business shocks in operating performance and the frequency of such shocks may haveincreased post-FRS3, increasing net income volatility. Alternatively, as we do not control for thesmoothing effect of all working capital accruals (especially those not reflected in revenue), apotential reduction in the use of working capital accruals as a smoothing device may have inflatednet income volatility post-FRS3. Higher net income variability could also reflect an overall increasein conservative accounting practices post-FRS3. The latter two scenarios appear consistent with theincreased oversight and governance following FRS3 and the Cadbury report in 1993. To ensurethese scenarios do not confound our results, we include accruals (both working capital and long-term accruals) and additional measures of operating performance (e.g. profitability index, change inprofitability index) in our multivariate specification (Eqs. (1) and (2)). This point also holds for theresults in Table 5.

Page 25: Classificatory income smoothing: The impact of a change in ... Classificatory income smoothing The impact of a change in...alongside the basic figure. In all, FRS3 increased the

Table 3Descriptive statistics for key variables

Variable Statistics Pre-FRS3 (N = 914) Post-FRS3 (N = 914) p-Value

Panel A: Descriptive statistics for the classificatory smoothing index (CSI) and smoothing indices

pre- and post-FRS3

CSI Mean 0.063*** 0.157*** <0.001Median 0.004*** 0.023*** <0.001

C: 387 UC: 527 C: 199 UC: 715

Smoothing indicesSI1 ¼ rEARN

rREV Mean 0.301 0.364 0.032Median 0.161 0.166 0.512

SI2 ¼ rEARNbXIrREV Mean 0.237

Median 0.121SI3 ¼ rEARNbNOEI

rREV Mean 0.259Median 0.131

SI4 ¼ rEARNbEIrREV Mean 0.207

Median 0.118

Panel B: Descriptive statistics for variables affecting the variation of the classificatory smoothing

index

DIVERGENCE Mean 0.070*** 0.057*** 0.012Median 0.026 0.024 0.064

GAINLOSS Mean 0.032*** 0.053*** <0.001Median 0.000 0.000 0.001

PERMLOSS Mean 0.138*** 0.131*** 0.015Median 0.000 0.000 0.001

RISK(b) Mean 0.824*** 0.753*** <0.001Median 0.849 0.770 0.003

Leverage (LEV) Mean 0.179*** 0.185*** 0.999Median 0.163 0.163 0.870

SIZE Mean 4.352*** 4.463*** 0.356Median 4.143 4.333 0.322

Profitability index (PI) Mean 4.540*** 4.463*** 0.415Median 4.613 4.333 0.295

Change in profitabilityindex (DPI)

Mean 4.531*** 4.507*** 0.794Median 4.613 4.667 0.597

Level impact ofclassification items (LI)

Mean 0.004*** 0.012*** <0.001Median 0.000 0.000 <0.001

Working capital accruals(WCA)

Mean 0.027*** 0.015*** <0.001Median 0.016 0.007 <0.001

Long-term accruals(LTA)

Mean �0.100*** �0.092*** <0.001Median �0.094 �0.083 <0.001

*/**/*** indicate significance at 1% (two-tailed) using a t-test for the means and distribution freeconfidence intervals for the medians. p-Value corresponds to a Wilcoxon non-parametric test (two-sided) for the difference in means and medians between the pre- and post-FRS3 period.The sample consists of 9222 firm-year observations over the period 1986–97 for 914 UK non-financial firms with at least two observations in both the pre and post-FRS3 periods. Table 1defines the variables. CSI measures income smoothing from classifications of extraordinary itemspre-FRS3 and of either non-operating or all exceptional items post-FRS3, in line with thesmoothing object. The smoothing object is ordinary income pre-FRS3 and the least volatile of

V.E. Athanasakou et al. / Journal of Accounting and Public Policy 26 (2007) 387–435 411

Page 26: Classificatory income smoothing: The impact of a change in ... Classificatory income smoothing The impact of a change in...alongside the basic figure. In all, FRS3 increased the

Table 3 (continued)

earnings before non-operating exceptional items or earnings before all exceptional items post-FRS3. We calculate CSI at the firm level for the pre- and post-FRS3 periods. C is the number ofcensored observations (CSI 6 0). UC is the number of uncensored observations (CSI > 0). Wecalculate all remaining variables annually and use pre- and post-FRS3 averages.

412 V.E. Athanasakou et al. / Journal of Accounting and Public Policy 26 (2007) 387–435

analysis (not tabulated) comparing mean SI2 (0.237) pre-FRS3 with mean SI3

(0.259) and mean SI4 (0.207) post-FRS3, provides no evidence of a significantchange in the indices post-FRS3.

Panel B of Table 3 reports descriptive statistics for the control variables ofEq. (1) partitioned into the pre- and post-FRS3 periods. Post-FRS3 there isevidence of a significant decrease in DIVERGENCE, in the frequency of firmswith negative pre-exceptional earnings (PERMLOSS), in RISK(b) and inWCA, and of a significant increase in the frequency of firms with pre-excep-tional earnings and net income losses (GAINLOSS) and in LTA. The impactof CIs on the level of reported earnings (LI) is income-increasing in both peri-ods and significantly higher post-FRS3. Consistent with their fundamentallydifferent nature, median WCA is positive, while median LTA is negative inboth periods. Additional analysis shows that LTA is negative in 97% of theobservations, consistent with long-term accruals representing primarily largeaccrued expenses or non-cash losses.

5.1.2. Multivariate analysis

Table 4 reports Tobit regression results for the effect of FRS3 on CSI. Asthis is a corner solution application, we also report the partial effects (economicsignificance) of the explanatory variables. Model 1 gives the results of estimat-ing Eq. (1). FRS3 is positive and significant (0.098, t = 6.35), consistent with anoverall increase in the classificatory smoothing index post-FRS3. Model 2reports the results of estimating Eq. (2). DIVERGENCE is positive and signif-icant (0.279, t = 2.87), suggesting a positive association between deviations ofnet income from lagged ordinary income and classificatory smoothing viaextraordinary items pre-FRS3. FRS3 · DIVERGENCE is positive and signifi-cant (0.671, z = 2.53), increasing the coefficient on DIVERGENCE to 0.950post-FRS3. This increase suggests that deviations of net income from laggedearnings before exceptional items induce a higher level of classificatorysmoothing post-FRS3.

With regard to the remaining factors affecting the variation of CSI, GAIN-

LOSS, PERMLOSS, RISK(b), LEV, SIZE, DPI, WCA and LTA in Model 2are all significant and consistent with predicted signs. LI is positive (4.715,z = 6.72), indicating that increasing adjusted earnings through income-increas-ing CIs (e.g. restructuring costs, losses from disposals of assets, exceptionaloperating costs) also results in smoother earnings. However, as CSI is left cen-sored, our main specification does not take into account cases where classifica-

Page 27: Classificatory income smoothing: The impact of a change in ... Classificatory income smoothing The impact of a change in...alongside the basic figure. In all, FRS3 increased the

Table 4

Regressions of the classificatory smoothing index (CSI) on an FRS3 indicator, divergence from expected earnings, and a set of control variables

Variables Predicted sign Entire sample Firms with at least five

observations in each period

Model 1 Model 2 Model 3

Coefficient

(z-statistic)

Partial

effects

Coefficient

(z-statistic)

Partial

effects

Coefficient

(z-statistic)

Partial

effects

Intercept �0.079 (�1.44) �0.066 (�1.27) �0.065 (�1.10)

FRS3 0.098*** (6.35) 0.054 0.057*** (4.26) 0.032 0.019 (1.22) 0.011

DIVERGENCE + 0.580*** (3.73) 0.320 0.279*** (2.87) 0.155 0.307 (1.39) 0.179

FRS3 · DIVERGENCE + 0.671** (2.53) 0.373 1.429*** (3.65) 0.833

GAINLOSS + 0.293*** (2.68) 0.162 0.313*** (2.88) 0.174 0.468*** (3.67) 0.273

PERMLOSS � �0.090 (�1.53) �0.050 �0.092* (�1.67) �0.051 �0.013 (�0.15) �0.007

RISK(b) + 0.042** (2.01) 0.023 0.044** (1.98) 0.024 �0.005 (�0.20) �0.003

LEV + 0.322*** (3.33) 0.178 0.308*** (3.11) 0.171 0.174* (1.71) 0.101

SIZE + 0.011*** (3.74) 0.006 0.011*** (3.75) 0.006 0.013*** (3.07) 0.007

PI � 0.003 (0.51) 0.001 �0.001 (�0.08) �0.001 �0.002 (�0.33) �0.001

DPI � �0.020*** (�2.65) �0.011 �0.016** (�2.39) �0.009 �0.001 (�0.04) �0.001

LI + 5.142*** (8.01) 2.841 4.715*** (6.72) 2.623 4.571*** (4.48) 2.664

WCA � �0.423*** (�2.89) �0.234 �0.465*** (�3.75) �0.259 �0.648*** (�3.54) �0.379

LTA + 0.684*** (4.63) 0.378 0.589*** (4.55) 0.327 0.581*** (3.04) 0.338

N 1828 (C: 586/UC: 1242) 1828 (C: 586/UC: 1242) 1218 (C: 348/UC: 870)

Chi-square 786.06 833.94 623.93

p-Value <0.001 <0.001 <0.001

*/**/*** indicate significance at 0.1/0.05/0.01 levels (two-tailed). z-statistics in parentheses are calculated on bootstrap standard errors to control forheteroskedasticity and cross-sectional dependence (see footnote 15). The coefficients are estimated using a corner solution Tobit model. The dependentvariable is left censored at zero. C is the number of censored observations. UC is the number of uncensored observations.The original sample consists of 9222 firm-year observations over the period 1986–97 for 914 UK non-financial firms with at least two observations in both

the pre and post-FRS3 periods. The sample of firms with at least five observations in each period consists of 609 UK non-financial firms that have at leastfive observations in each of the pre- and post-FRS3 periods. Table 1 defines the variables. CSI measures income smoothing from classifications ofextraordinary items pre-FRS3 and of either non-operating or all exceptional items post-FRS3, in line with the smoothing object. The smoothing object isordinary income pre-FRS3 and the least volatile of earnings before non-operating exceptional items or earnings before all exceptional items post-FRS3. Wecalculate CSI at the firm level for the pre- and post-FRS3 periods. We calculate all remaining variables annually and use pre- and post-FRS3 averages.These are panel regressions involving a cross-section of 914 (Models 1–2) or 609 firms (Model 3), each with two or five observations (pre- and post-FRS3).

V.E

.A

tha

na

sak

ou

eta

l./

Jo

urn

al

of

Acco

un

ting

an

dP

ub

licP

olicy

26

(2

00

7)

38

7–

43

5413

Page 28: Classificatory income smoothing: The impact of a change in ... Classificatory income smoothing The impact of a change in...alongside the basic figure. In all, FRS3 increased the

414 V.E. Athanasakou et al. / Journal of Accounting and Public Policy 26 (2007) 387–435

tory choices magnify the volatility of reported earnings. As a result we cannotcapture the effect of income-increasing CIs that increase the variability of earn-ings, biasing our results against finding a negative association between LI andCSI. To ensure robustness of this result, we repeat the regressions using anunrestricted LI (not set to zero where CSI is zero or negative). We also runOLS regressions of the uncensored CSI on the unrestricted LI. In both casesthe result of a positive association between CSI and LI remains.

In summary, the results in Table 4 provide evidence of an increase in thelevel of classificatory smoothing post-FRS3, consistent with our first hypothe-sis. Furthermore consistent with our second hypothesis, the results indicatethat UK firms use classificatory smoothing to a greater extent post-FRS3 toreport earnings closer to expectations and to highlight recurrent profitability.Table 4 also shows that the level and smoothing effects of removing CIs onreported earnings converge with income-increasing CIs resulting in smootherearnings, especially post-FRS3.

5.2. Additional analysis

5.2.1. Refining the estimation of CSI

For the initial calculation of CSI we require firms to have at least two obser-vations in both the pre- and post-FRS3 periods. This is the minimum requirementto calculate standard deviations. With this restriction we estimate CSI for eachfirm based on a mean (median) frequency of 5.57 (6) observations per period.For more efficient calculation of standard deviations, we repeat the analysis ona subset of 609 firms—approximately 67% of the original sample—with at leastfive observations in each period. The mean (median) frequency of observationsper period for each firm rises to 6.12 (7) on this subset. When we repeat Eq. (1)on this subset (results not tabulated), FRS3 remains positive and significant(0.081, z = 6.82). When we repeat Eq. (2), Model 3 in Table 4, FRS3 · DIVER-

GENCE is also positive and significant (1.429, z = 3.65), and higher in magnitudecompared to Model 2. These results lend further credence to our hypotheses.

5.2.2. Alternative scenarios for the post-FRS3 period

For the post-FRS3 period, we calculate CSI assuming two smoothing objectsacross firms: EARNbNOEI and EARNbEI. In addition to accommodating thegreater latitude for classifications of operating exceptional items post-FRS3, thisassumption is more realistic as it accords with evidence of a gradual rise in thefrequency of firms disclosing alternative EPS on earnings before all exceptionalitems (Choi et al., 2005). However, choosing the least volatile figure as thesmoothing object biases results in favor of an increase in CSI post-FRS3, con-sistent with our first hypothesis. To mitigate concern over this bias, we repeatour analysis based on the more restrictive assumption that all firms smoothedearnings before non-operating exceptional items post-FRS3 (CSI = SI1 � SI3).

Page 29: Classificatory income smoothing: The impact of a change in ... Classificatory income smoothing The impact of a change in...alongside the basic figure. In all, FRS3 increased the

V.E. Athanasakou et al. / Journal of Accounting and Public Policy 26 (2007) 387–435 415

The results of this analysis (not tabulated) show that, though lower in magni-tude, FRS3 remains positive and significant (0.029, z = 3.04). We also obtainconsistent results for FRS3 · DIVERGENCE (0.365, z = 2.09). The lower coef-ficients are due to restricting classificatory choices to non-operating exceptionalitems post-FRS3 and ignoring any smoothing activity through classifications ofoperating exceptionals. We also repeat the core tests on the assumption that allfirms smoothed EARNbEI post-FRS3 (CSI = SI1 � SI4). The core results arequalitatively similar under this assumption (FRS3: 0.081, z = 5.28; FRS3 ·DIVERGENCE: 0.633, z = 2.44).

5.2.3. Repeating the analysis on firms disclosing alternative EPS

In calculating CSI for the post-FRS3 period we assume the least volatileearnings number is the firm’s smoothing object. This might be a misleading cri-terion if the firm does not disclose this number to investors as an alternativeEPS.19 On the other hand, the lack of an alternative EPS does not necessarilymean that a firm’s smoothing object is basic EPS. Ronen and Sadan (1981)argue that the smoothing object is the variable that managers perceive to havethe greatest impact on investors’ actions. The original intention of the ASB inissuing FRS3 was to provide users with a range of important components ofperformance to enable them to convert net income to a more sustainable per-formance indicator and to form better informed expectations about futureresults and cash flows. To the extent managers were confident that analystsand investors make the necessary adjustments to arrive at core earnings, theymay have smoothed a profit and loss subtotal without reporting alternativeEPS on this number. This is plausible given the various approaches that firmsadopted to highlight the effect of exceptional items (see the Appendix). Priorevidence suggests that investors focused on pre-exceptional earnings. In twopost-FRS3 accounting periods, 1993 and 1996, Lin and Walker (2000) find thatheadline earnings were more value relevant for stock prices than basic EPS.The rise in the frequency and types of operating exceptional items graduallydiverted attention to earnings before all exceptional items. Analyst trackingservices (e.g. I/B/E/S, JCF) indicate that analysts predict earnings before allexceptional and non-recurring items. In two post-FRS3 accounting periods,1994 and 1996, Choi et al. (2005) find that the EPS figure I/B/E/S reports asactual is more value relevant than basic EPS in terms of earnings predictabilityand price-earnings and return-earnings associations.

19 This is a plausible scenario especially for the period directly following implementation of FRS3.As disclosing alternative EPS was not mandatory under FRS3, and as it was inevitably not anestablished policy in the initial period, some firms may have been reluctant to make such disclosuresimmediately. Possible reasons are confusion over the new definitions, uncertainty over thedisclosure choices of other firms, and the time needed to choose an alternative level to discloseconsistently over time.

Page 30: Classificatory income smoothing: The impact of a change in ... Classificatory income smoothing The impact of a change in...alongside the basic figure. In all, FRS3 increased the

Table 5Descriptive statistics for the classificatory smoothing index (CSI) and smoothing indices pre- andpost-FRS3, for firms disclosing alternative EPS in 1996

Variable Statistics Pre-FRS3 (N = 197) Post-FRS3 (N = 197) p-Valuea

CSI Mean 0.071*** 0.191*** <0.001Median 0.016*** 0.068*** <0.001

C: 58 UC: 139 C: 33 UC: 164

Smoothing indices

SI1 ¼ rEARNrREV Mean 0.228 0.339 0.001

Median 0.145 0.188 0.021SI2 ¼ rEARNbXI

rREV Mean 0.159Median 0.094

SI3 ¼ rEARNbNOEIrREV Mean 0.218

Median 0.134SI4 ¼ rEARNbEI

rREV Mean 0.164Median 0.115

The sample consists of 197 non-financial firms included in the 500 largest UK firms (by marketcapitalization) that reported alternative EPS in 1996. Of the total number of disclosure firmsincluded in the largest 500 UK firms (254 firms), we excluded 57 firms that did not satisfy oursample criteria. Retained firms have at least two observations in both the pre- and post-FRS3periods. Table 1 defines the variables. CSI measures income smoothing from classifications ofextraordinary items pre-FRS3 and of either non-operating or all exceptional items post-FRS3, inline with the smoothing object. The smoothing object is ordinary income pre-FRS3 and the leastvolatile of earnings before non-operating exceptional items or earnings before all exceptional itemspost-FRS3. We calculate CSI at the firm level for the pre- and post-FRS3 periods. C is the numberof censored observations (CSI 6 0). UC is the number of uncensored observations (CSI > 0).

*/**/*** indicate significance 1% (two-tailed) using a t-test for the means and distribution freeconfidence intervals for the medians.

a p-Value corresponds to a Wilcoxon non-parametric test (two-sided) for the difference in meansand medians between the pre- and post-FRS3 period.

416 V.E. Athanasakou et al. / Journal of Accounting and Public Policy 26 (2007) 387–435

To investigate this issue, we repeat the core tests on a subset of firms that dis-closed alternative EPS post-FRS3. We use data on alternative EPS disclosures in1996 for the 500 largest UK listed non-financial firms by market capitalization.20

From the initial number of 254 disclosers of alternative EPS, 21 197 satisfy the cri-teria for the original sample. Table 5 reports the mean and median CSI parti-tioned into the pre- and post-FRS3 periods. Mean and median CSI increasefrom 0.071 and 0.016 pre-FRS3 to 0.191 and 0.068 post-FRS3. The increase issignificant both for the mean and the median. The fraction of firms with positive

20 We thank Dr. Young-Soo Choi (Lancaster University) for providing his data on disclosures ofalternative EPS.21 This is approximately 50% of the top 500 UK firms. We repeat the analysis for firms disclosing

alternative EPS in the first post-FRS3 financial statements published in 1993/1994 (approximately38% of the top 500 UK firms). The core results remain.

Page 31: Classificatory income smoothing: The impact of a change in ... Classificatory income smoothing The impact of a change in...alongside the basic figure. In all, FRS3 increased the

Table 6Regressions of the classificatory smoothing index (CSI) on an FRS3 indicator, divergence from expected earnings and a set of control variables forfirms disclosing alternative EPS in 1996

Variables Predicted sign Firms disclosing alternative EPS Entire sample

Model 1 Model 2 Model 3

Coefficient(z-statistic)

Partialeffects

Coefficient(z-statistic)

Partialeffects

Coefficient(z-statistic)

Partialeffects

Intercept �0.199 (�1.22) �0.124 (�0.87) �0.045 (�0.92)FRS3 0.092*** (3.52) 0.058 �0.050* (�1.88) �0.033 0.036*** (2.72) 0.020DIS �0.036** (�1.96) �0.020DIVERGENCE + 1.075 (0.95) 0.683 0.577 (0.96) 0.379 0.217** (1.98) 0.122DIS · DIVERGENCE + 0.513 (0.94) 0.287FRS3 · DIVERGENCE + 4.212*** (6.13) 2.768 0.692*** (2.77) 0.386DIS · FRS3 · DIVERGENCE + 2.496*** (3.89) 1.396GAINLOSS + 0.441** (2.13) 0.280 0.300 (1.42) 0.197 0.271*** (2.68) 0.152PERMLOSS � 0.087 (0.24) 0.055 �0.097 (�0.37) �0.064 �0.095* (�1.83) �0.053RISK(b) + 0.030 (0.58) 0.019 0.064 (1.50) 0.042 0.043* (1.89) 0.024LEV + 0.315 (1.49) 0.200 0.132 (0.88) 0.087 0.296*** (3.11) 0.165SIZE + 0.028** (1.98) 0.018 0.025** (2.33) 0.016 0.009*** (2.68) 0.005PI � 0.003 (0.29) �0.002 �0.006 (�0.78) �0.004 �0.002 (�0.38) �0.001DPI � �0.023 (�1.54) �0.015 �0.019 (�1.39) �0.013 �0.015** (�2.12) �0.008LI + 3.190* (1.86) 2.028 3.079* (1.71) 2.024 4.832*** (7.95) 2.705WCA � �0.280 (�0.95) �0.178 �0.206 (�0.78) �0.135 �0.450*** (�3.40) �0.252LTA + 0.495 (1.02) 0.314 0.316 (0.90) 0.208 0.573*** (4.42) 0.321

(continued on next page)

V.E

.A

tha

na

sak

ou

eta

l./

Jo

urn

al

of

Acco

un

ting

an

dP

ub

licP

olicy

26

(2

00

7)

38

7–

43

5417

Page 32: Classificatory income smoothing: The impact of a change in ... Classificatory income smoothing The impact of a change in...alongside the basic figure. In all, FRS3 increased the

Variables Predicted sign Firms disclosing alternative EPS Entire sample

Model 1 Model 2 Model 3

Coefficient(z-statistic)

Partialeffects

Coefficient(z-statistic)

Partialeffects

Coefficient(z-statistic)

Partialeffects

N 394 (C: 91/UC: 303) 394 (C: 91/UC: 303) 1828 (C: 586 /UC: 1242)Chi-square 138.14 201.97 860.98p-Value <0.001 <0.001 <0.001

*/**/*** indicate significance at 0.1/0.05/0.01 levels (two-tailed). z-statistics in parentheses are calculated on bootstrap standard errors to control forheteroskedasticity and cross-sectional dependence (see footnote 15). The coefficients are estimated using a corner solution Tobit model. The dependentvariable is left censored at zero. C is the number of censored observations. UC is the number of uncensored observations.The sample of firms disclosing alternative EPS consists of 197 non-financial firms included in the 500 largest UK firms (by market capitalization) thatreported alternative EPS in 1996. Of the total number of disclosure firms included in the largest 500 UK firms (254 firms), we excluded 57 firms that didnot satisfy our sample criteria. Retained firms have at least two observations in both the pre- and post-FRS3 periods. The entire sample consists of 9222firm-year observations over the period 1986–97 for 914 UK non-financial firms with at least two observations in both the pre and post-FRS3 periods.Table 1 defines the variables. CSI measures income smoothing from classifications of extraordinary items pre-FRS3 and of either non-operating or allexceptional items post-FRS3, in line with the smoothing object. The smoothing object is ordinary income pre-FRS3 and the least volatile of earningsbefore non-operating exceptional items or earnings before all exceptional items post-FRS3. We calculate CSI at the firm level for the pre- and post-FRS3 periods. We calculate all remaining variables annually and use pre- and post-FRS3 averages. These are panel regressions involving a cross-section of 197 (Models 1–2) or 914 firms (Model 3), each with two observations (pre- and post-FRS3).

Table 6 (continued)

418V

.E.

Ath

an

asa

ko

uet

al.

/J

ou

rna

lo

fA

ccou

ntin

ga

nd

Pu

blic

Po

licy2

6(

20

07

)3

87

–4

35

Page 33: Classificatory income smoothing: The impact of a change in ... Classificatory income smoothing The impact of a change in...alongside the basic figure. In all, FRS3 increased the

V.E. Athanasakou et al. / Journal of Accounting and Public Policy 26 (2007) 387–435 419

CSI rises by 13% ((164 � 139)/197) post-FRS3, consistent with an increase in thefrequency of firms engaging in classificatory smoothing post-FRS3.

Model 1 of Table 6 reports the regression results of Eq. (1) for firms dis-closing alternative EPS in 1996. FRS3 remains positive and significant (0.092,z = 3.52). Model 2 reports regression results for Eq. (2). DIVERGENCE isnot significant, suggesting that deviations from expected earnings do notinduce classificatory smoothing by disclosing firms pre-FRS3. Post-FRS3there is a considerable change in the practice, as FRS3 · DIVERGENCE ispositive, significant (4.212, z = 6.13), and substantially higher than in Table4 (0.671). To test whether absolute unexpected earnings induce a greaterincrease in classificatory smoothing post-FRS3 for firms disclosing alternativeEPS in 1996 compared with the remaining firms in the sample, we run testson the full sample adding an interaction term between an indicator of disclo-sure of alternative EPS (DIS) and FRS3 · DIVERGENCE. Model 3 in Table 6reports the regression results. While FRS3 · DIVERGENCE remains signifi-cant (0.692, z = 2.77), DIS · FRS3 · DIVERGENCE is also significant(2.496, z = 3.89), indicating that absolute unexpected earnings induce anincremental increase in classificatory smoothing post-FRS3 for disclosingfirms compared with the remaining firms in the sample.

In summary, the results in Table 6 provide evidence of a greater increase inclassificatory income smoothing post-FRS3 to reduce deviations of net incomefrom expectations for firms disclosing an alternative EPS compared with theremaining firms in the sample. Even though this result underlines the impor-tance of FRS3’s option to disclose alternative EPS, evidence on the increaseremaining within the remaining firms suggests that our core findings are notsolely attributable to this option. This might be because firms that did not dis-close alternative EPS still smoothed pre-exceptional earnings in the belief thatinvestors would focus on these levels post-FRS3.

5.2.4. Other structural shiftsTo test whether structural breaks for any of the remaining drivers of varia-

tion in classificatory smoothing confound our core findings, we extend Eq. (2)to allow for interactions between these drivers and FRS3. Model 1 in Table 7reports the regression results for the entire sample. FRS3 · DIVERGENCE

remains positive and significant (0.664, z = 2.25). There is also evidence of astructural shift in LI. The coefficient on LI increases from 2.850 pre-FRS3 to5.327 (2.850 + 2.477) post-FRS3. The shift is more pronounced when werepeat the analysis for firms with at least five observations in both the pre-and post-FRS3 period in Model 2, and suggests that firms use CIs to a greaterextent post-FRS3 to increase and smooth reported earnings simultaneously.We obtain no evidence of a structural change for the remaining smoothingincentives post-FRS3. Therefore, deviations of net income from expected earn-ings induce a higher level of classificatory smoothing post-FRS3 even after

Page 34: Classificatory income smoothing: The impact of a change in ... Classificatory income smoothing The impact of a change in...alongside the basic figure. In all, FRS3 increased the

Table 7Regressions of the classificatory smoothing index (CSI) on an FRS3 indicator, divergence from expected earnings, a set of control variables andinteraction terms

Variables Predicted sign Entire sample Firms with at least fiveobservations in each period

Firms disclosing alternativeEPS

Mode1 Model 2 Model 3

Coefficient(z-statistic)

Partialeffects

Coefficient(z-statistic)

Partialeffects

Coefficient(z-statistic)

Partialeffects

Intercept �0.077 (�1.29) �0.062 (�1.11) 0.074 (0.70)FRS3 0.061 (0.81) 0.034 0.014 (0.15) 0.008 �0.492* (�1.73) �0.315DIVERGENCE + 0.272*** (3.00) 0.152 0.267 (1.51) 0.158 0.377 (0.64) 0.253FRS3 · DIVERGENCE + 0.664** (2.25) 0.371 1.357*** (2.99) 0.804 3.968*** (4.96) 2.661GAINLOSS + 0.386*** (2.62) 0.216 0.585*** (4.01) 0.347 0.437 (1.63) 0.293PERMLOSS � �0.107* (�1.87) �0.060 0.055 (0.75) 0.033 �0.286* (�1.86) �0.192RISK(b) + 0.059 (1.45) 0.033 0.013 (0.40) 0.008 0.097 (1.62) 0.065LEV + 0.392*** (3.02) 0.219 0.174* (1.92) 0.103 0.046 (0.21) 0.031SIZE + 0.006* (1.87) 0.004 0.009*** (3.09) 0.005 0.003 (0.35) 0.002PI � �0.002 (�0.29) �0.001 �0.000 (�0.04) �0.001 �0.013 (�1.46) �0.009DPI � �0.015* (�1.84) �0.008 �0.003 (�0.30) �0.002 �0.023 (�1.38) �0.015LI + 2.850*** (2.80) 1.591 0.094 (0.11) 0.056 �2.698* (�1.70) �1.809WCA � �0.345* (�1.94) �0.193 �0.459** (�2.28) �0.272 �0.556** (�2.05) �0.373LTA + 0.474*** (2.72) 0.265 0.521** (2.38) 0.309 0.043 (0.14) 0.026FRS3 · GAINLOSS �0.108 (�0.58) �0.060 �0.242 (�1.30) �0.144 �0.246 (�0.65) �0.165FRS3 · PERMLOSS 0.021 (0.19) 0.012 �0.162 (�1.27) �0.096 0.684 (1.07) 0.458

420V

.E.

Ath

an

asa

ko

uet

al.

/J

ou

rna

lo

fA

ccou

ntin

ga

nd

Pu

blic

Po

licy2

6(

20

07

)3

87

–4

35

Page 35: Classificatory income smoothing: The impact of a change in ... Classificatory income smoothing The impact of a change in...alongside the basic figure. In all, FRS3 increased the

FRS3 · RISK �0.026 (�0.54) �0.015 �0.015 (�0.33) �0.009 �0.039 (�0.48) �0.026FRS3 · LEV �0.146 (�1.09) �0.082 �0.065 (�0.38) �0.036 0.227 (0.70) 0.152FRS3 · SIZE 0.007 (1.03) 0.004 0.003 (0.36) 0.002 0.033 (1.41) 0.022FRS3 · PI 0.002 (0.23) 0.001 �0.006 (�0.51) �0.003 0.025 (1.25) 0.017FRS3 · DPI 0.000 (0.01) 0.000 0.004 (0.30) 0.003 0.014 (0.48) 0.009FRS3 · LI 2.477* (1.90) 1.382 6.147*** (4.30) 3.646 9.317*** (3.19) 6.249FRS3 · WCA �0.308 (�1.16) �0.172 �0.438 (�1.50) �0.260 0.570 (0.76) 0.382FRS3 · LTA 0.078 (0.25) 0.044 �0.301 (�0.73) �0.178 0.549 (0.88) 0.368

N 1828 (C: 586/UC: 1242) 1218 (C: 348/UC: 870) 394 (C: 91/UC: 303)Chi-square 417.38 583.77 246.45p-Value <0.001 <0.001 <0.001

*/**/*** indicate significance at 0.1/0.05/0.01 levels (two-tailed). z-statistics in parentheses are calculated on bootstrap standard errors to control forheteroskedasticity and cross-sectional dependence (see footnote 15). The dependent variable is left censored at zero. C is the number of censoredobservations. UC is the number of uncensored observations.The entire sample consists of 9222 firm-year observations over the period 1986–97 for 914 UK non-financial firms with at least two observations in

both the pre- and post-FRS3 periods. The sample of firms with at least five observations in each period consists of 609 UK non-financial firms that haveat least five observations in each of the pre- and post-FRS3 periods. The sample of firms disclosing alternative EPS consists of 197 non-financial firmsincluded in the 500 largest UK firms (by market capitalization) that reported alternative EPS in 1996. Of the total number of disclosure firms includedin the largest 500 UK firms (254 firms), we excluded 57 firms that did not satisfy our sample criteria. Retained firms have at least two observations inboth the pre- and post-FRS3 periods. Table 1 defines the variables. CSI measures income smoothing from classifications of extraordinary items pre-FRS3 and of either non-operating or all exceptional items post-FRS3, in line with the smoothing object. The smoothing object is ordinary income pre-FRS3 and the least volatile of earnings before non-operating exceptional items or earnings before all exceptional items post-FRS3. We calculate CSI atthe firm level for the pre- and post-FRS3 periods. We calculate all remaining variables annually and use pre- and post-FRS3 averages. These are panelregressions involving a cross-section of 914 (Model 1), 609 (Model 2) or 197 firms (Model 3), each with two or five observations (pre- and post-FRS3).

V.E

.A

tha

na

sak

ou

eta

l./

Jo

urn

al

of

Acco

un

ting

an

dP

ub

licP

olicy

26

(2

00

7)

38

7–

43

5421

Page 36: Classificatory income smoothing: The impact of a change in ... Classificatory income smoothing The impact of a change in...alongside the basic figure. In all, FRS3 increased the

422 V.E. Athanasakou et al. / Journal of Accounting and Public Policy 26 (2007) 387–435

controlling for the structural change in the relation between CSI and the mag-nitude of CIs.

In Model 3 we repeat the analysis for firms disclosing alternative EPS.The coefficient on LI switches from negative (�2.698) pre-FRS3, to positive(�2.698 + 9.317 = 6.619) post-FRS3, implying that for disclosure firms,while pre-FRS3 the level and smoothing effects of removing CIs on reportedearnings work in opposite directions, post-FRS3 the two effects convergewith income-increasing CIs resulting in smoother earnings. Consistent withresults in models 1 and 2, we obtain no evidence of a structural shift forthe remaining smoothing incentives, including GAINLOSS, which capturesopportunistic classificatory choices to highlight a more favorable perfor-mance indicator.

5.2.5. Alternative explanations for the increase in classificatory smoothingThe main problem in measuring classificatory smoothing is that the magni-

tude and incidence of CIs is primarily due to business fundamentals. Failure tocontrol for these factors can result in measurement error in the proxy for clas-sificatory smoothing. For example, an increase in business restructurings in thepost-FRS3 period could lead to an increase in both the magnitude of negativenon-operating exceptional items and in the extent to which removing theseitems reduces the variability of alternative measures of performance. Thus,an apparent increase in the magnitude of classificatory income smoothingcould reflect an increase in the frequency of real events. Elliott and Hanna(1996) document a dramatic increase in the frequency of negative special itemsfor US firms in the period 1980–94, despite the same reporting standard gov-erning the disclosure of special items throughout the period. The authors pointout that the rising trend coincides with an increasing frequency of corporaterestructurings during the 1980s.

To limit the risk of an increase in the frequency of real events driving ourresults, we scaled earnings volatility by revenue volatility in calculating thesmoothing indices. Revenues reflect the effect of events such as mergers,acquisitions, and other structural business changes, which give rise tonon-operating exceptional items. The fact that the documented increase innegative exceptional items post-FRS3 results mainly from gradual growthin the relative magnitude of operating exceptional items further mitigatesthis risk. Events giving rise to operating exceptional items are even morelikely to be reflected in revenue. As for restructuring charges, which are amajor component of non-operating exceptional items, contrary to US evi-dence, evidence in the UK documents a decreased frequency of corporaterestructurings in the years following the introduction of FRS3 up until1998 (Mak, 2002).

As an additional robustness check on our control for business fundamen-tals, we recalculate CSI based on smoothing indices that scale the standard

Page 37: Classificatory income smoothing: The impact of a change in ... Classificatory income smoothing The impact of a change in...alongside the basic figure. In all, FRS3 increased the

V.E. Athanasakou et al. / Journal of Accounting and Public Policy 26 (2007) 387–435 423

deviation of earnings by the standard deviation of operating cash flows. Con-trolling for cash flows captures the effect of events giving rise to exceptionalitems reflected in costs instead of revenue. We repeat the core tests with thealternative CSI measure and obtain consistent results (not tabulated). FRS3remains positive and significant (0.526, z = 6.64). This also holds forFRS3 · DIVERGENCE (2.468, z = 2.01). The results are stronger since oper-ating cash flows are smoother than revenue due to the offsetting effect of thevariability of costs.

To further limit the risk of an increase in the frequency of real events con-founding our results, our multivariate specification includes additional controlsfor real business performance (e.g. profitability and changes in profitabilitybased on net income, working capital accruals, and long-term accruals). Accru-als act as a key control variable as they are highly correlated with unusual oper-ating and non-operating performance.

An alternative explanation for the documented increase in the classifica-tory smoothing index post-FRS3 might be an increase in the variance ofCIs resulting from a higher upward trend in the level of CIs post-FRS3, rel-ative to the pre-FRS3 period. If smoothing flexibility and thus actual classi-ficatory smoothing activity rises post-FRS3, we expect a rise in the magnitudeof CIs (see Table 2), but not necessarily a rise in the upward trend. To testthis alternative scenario, we regress CIs on a time trend (T), an FRS3 indica-tor, and an interaction term FRS3 · T. Results (not-tabulated) show that T isnegative and significant pre-FRS3 (�0.004, z = �8.20), indicating a gradualrise in negative extraordinary and exceptional items in that period (as inTable 2). However, while FRS3 is negative and significant, consistent withthe reported increase in the magnitude of negative CIs post-FRS3, FRS3 · T

is positive and significant and of similar magnitude to T (0.004, z = 4.55),eliminating the time trend in exceptional items post-FRS3. This evidence isinconsistent with the alternative explanation, lending further credence toour hypothesis of an increase in the practice of classificatory smoothingpost-FRS3.

5.3. The effect of FRS3 on earnings persistence

To explore whether UK firms use classificatory smoothing to a greaterextent post-FRS3 to increase earnings persistence, we complement our multi-variate analysis with earnings persistence tests. Initially we examine the extentto which removing CIs increases the persistence of reported earnings in the pre-and post-FRS3 periods and then we test whether there is an increase in the per-sistence of pre-exceptional earnings post-FRS3. Panel A of Table 8 reports theresults of earnings persistence regressions for EARN (Eq. (3)) and different lev-els of earnings before classification items (EARNbCIs) for the pre- and post-FRS3 periods (Eq. (4)). The panel reports tests of differences in the persistence

Page 38: Classificatory income smoothing: The impact of a change in ... Classificatory income smoothing The impact of a change in...alongside the basic figure. In all, FRS3 increased the

Table 8Earnings persistence tests

Panel A: Earnings persistence regressions for different levels of earnings

EARNi,t+1 = b0 + b1EARNi,t + vi,t

EARNbCIsi,t+1 = c0 + c1EARNbCIsi,t + vi,t

Pre-FRS3 Post-FRS3

b0 b1 Adj. R2 b0 b1 Adj. R2

EARN(a) 0.023 0.521 0.2640 0.026 0.492 0.2391

EARNbCIs c0 c1 Adj. R2 c0 c1 Adj. R2

EARNbXI(b) 0.021 0.586 0.3475EARNbNOEI(c) 0.022 0.640 0.3977EARNbXI & EI(d) 0.020 0.636 0.4236EARNbEI(e) 0.019 0.728 0.5188

N 5415 3807Difference

Within periods Mean (p-value)a Vuong z-test/(p-value)b Mean (p-value)a Vuong z-test/(p-value)b

Pre-FRS3 j Post-FRS3(b) � (a) j (c) � (a) 0.065 (0.101) 12.48 (<0.001) 0.148 (0.005) 10.20 (<0.001)(d) � (b) j (e) � (c) 0.050 (0.178) 12.11 (<0.001) 0.088 (0.078) 12.85 (<0.001)(d) � (a) j (e) � (a) 0.115 (0.003) 20.09 (<0.001) 0.236 (<0.001) 20.62 (<0.001)

424V

.E.

Ath

an

asa

ko

uet

al.

/J

ou

rna

lo

fA

ccou

ntin

ga

nd

Pu

blic

Po

licy2

6(

20

07

)3

87

–4

35

Page 39: Classificatory income smoothing: The impact of a change in ... Classificatory income smoothing The impact of a change in...alongside the basic figure. In all, FRS3 increased the

Panel B: Earnings persistence regressions for EARN and different levels of earnings (EARNbCIs) with classification items (CIs) and

interaction terms capturing the change in persistence post-FRS3

EARNbCIsi,t+1 = d0 + d1EARNbCIsi,t + d2 CIsi,t + d3FRS3i,t + d4FRS3 · EARNbCIsi,t + d5FRS3 · CIsi,t + vi,t

EARNINGS d0 (z-stat) d1 (z-stat) d2 (z-stat) d3 (z-stat) d4 (z-stat) d5 (z-stat) Adj. R2

EARN 0.023** (8.24) 0.521*** (17.37) 0.003 (0.96) �0.030 (�0.79) 0.25501. EARNbCI (b) � (c) 0.021*** (8.53) 0.587*** (21.95) �0.035 (�0.66) 0.001 (0.16) 0.056 (1.24) 0.008 (0.11) 0.36592. EARNbCI (b) � (e) 0.023*** (9.92) 0.579*** (22.15) �0.032 (�0.56) �0.004 (�1.19) 0.152*** (4.50) 0.018 (0.31) 0.40423. EARNbCI (d) � (e) 0.019*** (8.69) 0.638*** (25.84) �0.018 (�0.51) 0.000 (�0.12) 0.093*** (2.71) 0.004 (0.07) 0.4581N 9222

The sample consists of 9222 firm-year observations over the period 1986–97 for 914 UK non-financial firms with at least two observations in both thepre- and post-FRS3 periods. Table 1 defines the variables. We use three combinations for earnings before classification items (EARNbCIs): (1)EARNbXI pre-FRS3 and EARNbNOEI post-FRS3; (2) EARNbXI pre-FRS3 and EARNbEI post-FRS3; and (3) EARNbXI & EI pre-FRS3 andEARNbEI post-FRS3. Accordingly, CIs is extraordinary items (1 and 2) or further exceptional items (3) pre-FRS3, and non-operating exceptionalitems (1) or all exceptional items (2 and 3) post-FRS3. All variables are scaled by lagged total assets.

*/**/*** indicate significance at 0.1/0.05/0.01 levels (two-tailed). z-statistics in parentheses are calculated on bootstrap standard errors to control forcross-sectional dependence and heteroskedastic and autocorrelated residuals (see footnote 15).

a p-Value (two-sided) for difference in the persistence coefficients within each period based on a t-value calculated by dividing the difference betweenthe two coefficients by the square root of the sum of the variances of the two coefficients; variances based on bootstrap standard errors.

b p-Value for the difference in adjusted R-squares among the equations, based on a Vuong z-test.

V.E

.A

tha

na

sak

ou

eta

l./

Jo

urn

al

of

Acco

un

ting

an

dP

ub

licP

olicy

26

(2

00

7)

38

7–

43

5425

Page 40: Classificatory income smoothing: The impact of a change in ... Classificatory income smoothing The impact of a change in...alongside the basic figure. In all, FRS3 increased the

426 V.E. Athanasakou et al. / Journal of Accounting and Public Policy 26 (2007) 387–435

coefficients within each period.22 It also reports the Vuong test for the differ-ence in the explanatory power (adjusted R-squares) of the equations withinperiods. Pre-FRS3 the persistence coefficient increases from 0.52 for EARN

to 0.59 for EARNbXI and to 0.64 for EARNbXI & EI. Even though the incre-mental increases in persistence are insignificant, the overall growth in persis-tence from EARN to EARNbXI & EI pre-FRS3 (0.12) is significant. TheVuong test provides stronger results, as incremental increases in the explana-tory power of the earnings persistence regressions are highly significant(z = 12.48 and 12.11). The overall increase in explanatory power pre-FRS3 is15.96%.

Post-FRS3, the persistence coefficient increases from 0.49 for EARN to 0.64for EARNbNOEI and to 0.73 for EARNbEI. The incremental increases of 0.15and 0.09 are significant, indicating that EARNbNOEI is more persistent thannet income and EARNbEI is in turn more persistent than EARNbNOEI. Theoverall increase in persistence from EARN to EARNbEI post-FRS3 is 0.24,which is double the increase in the pre-FRS3 period. The Vuong tests give sim-ilar results as the incremental rises in adjusted R-squares are highly significant(z = 10.20 and 12.85). The overall increase in the explanatory power of theearnings persistence equations post-FRS3 (27.97%) is almost double theincrease in the pre-FRS3 period.

Shedding further light on the effect of FRS3 on earnings persistence,Panel B reports the results of Eq. (5) for net income (EARN) and threemeasures of earnings before classification items (EARNbCIs). The interac-tion terms, FRS3 · EARNbCIs and FRS3 · CIs, capture changes in persis-tence of the profit levels and in the predictive power of CIs post-FRS3.There is no evidence of a significant change in the persistence of EARN

post-FRS3. So even though net income is communicated to investorsthrough basic EPS, it appears to be as difficult to predict post-FRS3 as itis pre-FRS3. The coefficient on FRS3 · EARNbCIs for the first measureof EARNbCIs is not significant, indicating that EARNbNOEI post-FRS3is not significantly more persistent than EARNbXI pre-FRS3. However,FRS3 · EARNbCIs is positive and significant for the two remaining mea-sures of EARNbCIs (0.152, z = 4.50 and 0.093, z = 2.71), indicating thatEARNbEI post-FRS3 is more persistent than both EARNbXI and EARN-

bXI & EI pre-FRS3. The coefficient on CIs is insignificant in all three spec-ifications, indicating that extraordinary and exceptional items pre-FRS3 do

22 For the tests of differences in persistence coefficients, we follow Brown and Sivakumar (2003)and use the more conservative approximation technique based on unequal variances. In particular,we estimate the t-value by dividing the difference between the two coefficients by the square root ofthe sum of the variances of the two coefficient estimates. We calculate the variances based onbootstrap standard errors (see footnote 15).

Page 41: Classificatory income smoothing: The impact of a change in ... Classificatory income smoothing The impact of a change in...alongside the basic figure. In all, FRS3 increased the

V.E. Athanasakou et al. / Journal of Accounting and Public Policy 26 (2007) 387–435 427

not possess any predictive power. This holds also for exceptional items post-FRS3, as FRS3 · CIs is consistently insignificant.

In sum, the results in Table 8 reveal that while net income persistenceremains unaltered post-FRS3, indicating no fundamental change in earningspredictability, earnings before all exceptional items post-FRS3 are more persis-tent than ordinary income and pre-exceptional earnings pre-FRS3. The resultsfurther show that, consistent with their transitory nature, classification itemsare not associated with future recurrent profitability and that removing theseitems increases the persistence of earnings in both periods, but to a substan-tially greater extent post-FRS3. Moreover, excluding all and not just non-oper-ating exceptional items results in more persistent earnings post-FRS3. Overall,this evidence is in line with FRS3’s provisions enhancing the role of classifica-tory choices, especially of operating exceptional items, in identifying recurringprofitability.

6. Conclusion

We examine the effect of FRS3 on the practice of classificatory incomesmoothing. Our results based on a large balanced sample, provide evidenceof an overall increase in the practice of classificatory smoothing post-FRS3.This result is robust to controls for real business activity (e.g. sales, accruals,profitability), inter-temporal smoothing through working capital accrualsand conservative accounting practices through long term-accruals, mitigatingthe possibility of an increase in an event or practice other than intentionalsmoothing of income through classificatory choices. The increased practiceof classificatory smoothing coincides with growth in the magnitude of CIsand greater use of income-increasing CIs (e.g. transitory losses, exceptionalcosts) to smooth pre-exceptional earnings levels. We also find that unlikeextraordinary items pre-FRS3, exceptional items post-FRS3 are equally dis-tributed between positive and negative values, indicating that FRS3 succeededin removing the asymmetry in the distribution of CIs. Results from a multivar-iate analysis further show that deviations of net income from target earnings(lagged adjusted earnings) induce a higher level of classificatory smoothingpost-FRS3. Consistent with this evidence, earnings persistence tests show thatremoving classifications items increases the persistence of reported earnings toa greater extent post-FRS3, compared with the pre-FRS3 period. Our resultssuggest that FRS3 increased the use of classificatory choices to highlight anindicator of sustainable earnings.

Our results shed light on managerial smoothing practices following theintroduction of FRS3 and have important implications for both investorsand accounting standard setters. For investors the main implication is thegreater transparency of classificatory smoothing post-FRS3. Even when

Page 42: Classificatory income smoothing: The impact of a change in ... Classificatory income smoothing The impact of a change in...alongside the basic figure. In all, FRS3 increased the

428 V.E. Athanasakou et al. / Journal of Accounting and Public Policy 26 (2007) 387–435

firms reported basic EPS as the sole performance indicator, greater disclo-sure of exceptionals enabled investors to extract their preferred measureof persistent profitability. Higher persistence of pre-exceptional earningspost-FRS3 further facilitated investors’ assessments of firms’ futureprospects.

With regard to regulators, FRS3 facilitated communication between manag-ers and the market by enabling classificatory smoothing as a means to conveyinformation about sustainable performance necessary for security valuation.At the same time, FRS3 made the practice of classificatory smoothing espe-cially challenging for opportunistic managers through rigorous transparencyrequirements that helped users to ascertain whether firms reasonably removedincome components. Overall, coupled with evidence of a decrease in accrualsmanagement post-FRS3 (Peasnell et al., 2000; Choi et al., 2006a), our evidencesuggests that FRS3 increased the more transparent and less costly practice ofincome smoothing.

FRS3 is a reference point for developing a global model for reportingfinancial performance. In line with FRS3’s approach, the IASB, the FASB,and the ASB are considering deemphasizing reliance on a single performanceindicator (IASB, 2006). Our results indicate that FRS3’s requirements for amore informative distinction between permanent and transitory income com-ponents along with the option to disclose alternative performance measuresencouraged the practice of classificatory smoothing. To the extent that man-agers remove transitory items from permanent income, they assist users inevaluating their firms’ future prospects. Our findings appear consistent withthis notion. To the extent, however, that managers remove value relevantitems from the alternative measure of profitability to increase the perceivedvalue of the firm’s ongoing earnings potential (Doyle et al., 2003), usersmay be misled and firm valuations become tentative. The increased level oftransparency under FRS3 may not have eliminated opportunistic classifica-tory choices completely. A possible avenue for future research is to investi-gate the role of classificatory choices within a setting where managerialincentives for engaging in income-increasing earnings management are partic-ularly strong.

Acknowledgements

We gratefully acknowledge the comments of two anonymous referees,Professor Kenneth Peasnell, Dr. Steven Young, Dr. Wendy Beekes (Lancas-ter University) and the participants of the 10th Annual Conference of theFinancial Reporting and Business Communication Panel (Cardiff BusinessSchool).

Page 43: Classificatory income smoothing: The impact of a change in ... Classificatory income smoothing The impact of a change in...alongside the basic figure. In all, FRS3 increased the

Appendix. Approaches to presenting exceptional items in the income statement

Table AMulticolumn analysis reporting four columns: before exceptional items, exceptional items, discontinued operations and total

Imperial Chemical Industries plc, 31 December 1993Extract from the group profit and loss account

Before exceptionalitems, £m

Exceptionalitems, £m

Discontinuedoperations, £m

Total, £m

Turnover 8430 – 2202 10,632Operating costs (8218) – (1937) (10,155)Other operating income 123 – 33 156Trading profit (loss) 335 – 298 633Share of profits less losses on associated

undertakings45 – 2 47

Losses less profits on sale or closure ofoperations

– (94) (59) (153)

Profit on ordinary activities before

interest

380 (94) 241 527

Net interest payable (90) – (63) (153)Profit on ordinary activities before

taxation

290 (94) 178 374

Tax on profit (loss) on ordinaryactivities

(105) (18) (71) (194)

Profit (loss) on ordinary activities after

taxation

185 (112) 107 180

Attributable to minorities (38) (4) – (42)Net profit (loss) for the financial year 147 (116) 107 138Dividends (562)Loss retained for the year (424)Earnings (loss) per 1£ ordinary share 20.4p (16.1)p 14.9p 19.2p

Source: Tonkin and Skerratt (1995, p. 74).

V.E

.A

tha

na

sak

ou

eta

l./

Jo

urn

al

of

Acco

un

ting

an

dP

ub

licP

olicy

26

(2

00

7)

38

7–

43

5429

Page 44: Classificatory income smoothing: The impact of a change in ... Classificatory income smoothing The impact of a change in...alongside the basic figure. In all, FRS3 increased the

Table BSingle column analysis with boxes highlighting operating profit before and after exceptional items

Anglo United plc, 31 March 1993Extract from the consolidated profit and loss account

1993 £000 (Restated) 1992 £000

1. Turnover from continuing operations 544,742 555,959Cost of sales (including exceptional items)

from continuing operations(463,224) (458,291)

Gross profit 81,518 97,6682. Net operating expenses (including

exceptional items)(80,009) (63,943)

Operating profit before exceptional items 20,313 34,3453. Exceptional items (18,804) (620)Operating profit from continuing

operations1509 33,725

Notes to the accounts (extract)3. Exceptional items

Cost ofsales £000

Distributioncosts £000

Administrativeexpenses £000

1993 Total£000

1992 Total£000

Legal, professional and banking fees inrespect of debt restructuring

– – 8894 8894 –

Costs in connection with aborteddisposals and acquisitions

– – 2360 2360 –

Future discounted rents receivable – – (1370) (1370) –Reorganization and restructuring costs of

ongoing business1450 720 4853 7023 –

Write-off of loan to Employees’ ShareOwnership Plan Trust – – 1367 1367 –

430V

.E.

Ath

an

asa

ko

uet

al.

/J

ou

rna

lo

fA

ccou

ntin

ga

nd

Pu

blic

Po

licy2

6(

20

07

)3

87

–4

35

Page 45: Classificatory income smoothing: The impact of a change in ... Classificatory income smoothing The impact of a change in...alongside the basic figure. In all, FRS3 increased the

Table CA separate column segregating the effect of operating exceptionals and a single column presenting the remaining results from continuing anddiscontinued operations

Booker plc, 31 December 1994Extract from the consolidated profit and loss account

Before exceptional items, £m Exceptional items, £m Total, £m

TurnoverContinuing operations 3688.7 – 3688.7Acquisitions 10.5 – 10.5

3699.2 – 3699.2Discontinued operations 23.1 – 23.1Total turnover 3722.3 – 3722.3Operating costs (3620.1) (20.8) (3640.9)

Operating profitContinuing operations 98.1 (18.3) 79.8Acquisitions 1.6 (2.5) (0.9)

99.7 (20.8) 78.9Discontinued operations 2.5 – 2.5

Total operating profit 102.2 (20.8) 81.4

Professional costs in respect ofenvironmental matters

– – 530 530 620

1450 720 16,634 18,804 620

The comparative cost for 1992 relates wholly to administrative expenses.In order to assist in understanding the Group’s results for the year, and in view of the unusual materiality of exceptional items to the current year’sresults, the directors believe that it is appropriate to show separately the operating profit of the Group before exceptional items on the face of the profitand loss account as additional information

Source: PricewaterhouseCoopers (2000, pp. 7047–7048).

(continued on next page)

V.E

.A

tha

na

sak

ou

eta

l./

Jo

urn

al

of

Acco

un

ting

an

dP

ub

licP

olicy

26

(2

00

7)

38

7–

43

5431

Page 46: Classificatory income smoothing: The impact of a change in ... Classificatory income smoothing The impact of a change in...alongside the basic figure. In all, FRS3 increased the

Table C (continued)

Before exceptional items, £m Exceptional items, £m Total, £m

Notes to the accounts (extract)2. Operating costs

Continuingoperations, £m

Acquisitions,£m

Discontinuedoperations, £m

Pre-exceptionalitems, £m

Exceptionalitems, £m

Total 1994,£m

Cost of sales 3459.0 7.4 18.9 3485.3 14.8 3500.1Distribution costs 63.3 0.8 1.9 66.0 0.5 66.5Administrative expenses 66.4 0.7 1.7 68.8 5.5 74.3

Total operating costs 3588.7 8.9 22.5 3620.1 20.8 3640.9Income from investments – – – – – –Total operating costs less other income 3588.7 8.9 22.5 3620.1 20.8 3640.9

Gross profit as defined by theCompanies Act 1985

229.7 3.1 4.2 237.0 (14.8) 222.2

Source: PricewaterhouseCoopers (2000, p. 7049).

Table DDisclosure of operating exceptionals on the face of the income statement with commentary

Cable and wireless plc, 31 March 1995Extract from the consolidated profit and loss account

Note 1995, £m 1994, £m

Turnover 5132.8 4699.2Operating costs before exceptional items 3.4 (3938.6) (3608.0)Exceptional items: charged against operating costs 5 (60.5) –Total operating costs 5.9 (3999.1) (3608.0)Operating profit 4 1133.7 1091.2

Notes to the accounts (extract)5. Operating costs (extract)

432V

.E.

Ath

an

asa

ko

uet

al.

/J

ou

rna

lo

fA

ccou

ntin

ga

nd

Pu

blic

Po

licy2

6(

20

07

)3

87

–4

35

Page 47: Classificatory income smoothing: The impact of a change in ... Classificatory income smoothing The impact of a change in...alongside the basic figure. In all, FRS3 increased the

1995 1995 1994After exceptionalitems, £m

Exceptional itemstherein, £m

£m

Outpayments to other telecommunications administrations and carriers 1564.6 – 1459.9Employee costs 799.2 30.3 686.6Pension costs (Note 8)–principal schemes 51.1 – 48.6–other schemes 8.9 – 11.2Rental of transmission facilities 71.5 – 58.9Hire of plant and machinery 19.9 – 20.7Other operating lease rentals 23.0 – 21.3Other operating costs 942.1 30.2 833.6Depreciation of owned tangible fixed assets 510.9 – 461.1Depreciation of tangible fixed assets held under finance leases 6.3 – 4.5Auditors’ remuneration –for audit services 1.6 – 1.6

3999.1 60.5 3608.0

Having regard to the special nature of the Group’s business an analysis of operating costs in the manner prescribed by the Companies Act 1985 is notmeaningful. In the circumstances therefore the Directors have, as required by paragraph 3(3) of schedule 4 to the Companies Act 1985, adapted theprescribed format to the requirements of the Group’s business.

9. Exceptional items (extract)1995, £m 1994, £m

Mercury reorganization costs shown asCharge to operating profit 60.5 –Profit less losses on sale and termination of operations 17.7 –Losses on disposal of fixed assets 43.7 –

121.9 –The exceptional pre-tax charge of £121.9m relates to the Mercury reorganization announced in December 1994 to streamline its operations andstrengthen its competitive position in the UK telecommunications market. Applicable to this exceptional charge is a tax credit of £13.2m relatingto the operating cost element of the charge, and a minority interest share of £21.7 m

Source: PricewaterhouseCoopers (2000, p. 7050).

V.E

.A

tha

na

sak

ou

eta

l./

Jo

urn

al

of

Acco

un

ting

an

dP

ub

licP

olicy

26

(2

00

7)

38

7–

43

5433

Page 48: Classificatory income smoothing: The impact of a change in ... Classificatory income smoothing The impact of a change in...alongside the basic figure. In all, FRS3 increased the

434 V.E. Athanasakou et al. / Journal of Accounting and Public Policy 26 (2007) 387–435

References

Acker, D., Horton, J., Tonks, I., 2002. Accounting standards and analysts’ forecasts: The impact ofFRS3 on analysts’ ability to forecast EPS. Journal of Accounting and Public Policy 21, 193–217.

Ashari, N., Koh, H.C., Tan, S.L., Wong, W.H., 1994. Factors affecting income smoothing amonglisted companies in Singapore. Accounting and Business Research 24, 291–301.

Barnea, A., Ronen, J., Sadan, S., 1976. Classificatory smoothing of income with extraordinaryitems. The Accounting Review 51, 110–122.

Beattie, V., Brown, S., Ewers, D., John, B., Manson, S., Thomas, D., Turner, M., 1994.Extraordinary items and income smoothing: A positive accounting approach. Journal ofBusiness Finance and Accounting 21, 791–811.

Bhattacharya, N., Black, E.L., Christensen, T.E., Larson, C.R., 2003. Assessing the relativeinformativeness and permanence of pro forma earnings and GAAP operating earnings. Journalof Accounting and Economics 36, 285–319.

Bradshaw, M.T., Sloan, R.G., 2002. GAAP versus the street: An empirical assessment of twoalternative definitions of earnings. Journal of Accounting Research 40, 41–66.

Brown, L.D., Sivakumar, K., 2003. Comparing the value relevance of two operating incomemeasures. Review of Accounting Studies 8, 561–572.

Choi, Y.S., Lin, S., Walker, M., Young, S., 2005. Voluntary disclosure of alternative earnings pershare numbers. Working paper, Lancaster University.

Choi, Y.S., Walker, M., Young, S., 2006a. Earnings reporting and analysts’ earnings forecasts: Theperceptions of UK analysts and financial managers. Working paper, Lancaster University.

Choi, Y., Walker, M., Young, S., 2006b. Disagreement over the persistence of earningscomponents: Evidence on the properties of management-specific adjustments to GAAPearnings. Working paper, Lancaster University.

Doyle, J.T., Lundholm, R.J., Soliman, M.T., 2003. The predictive value of expenses excluded frompro forma earnings. Review of Accounting Studies 8, 145–174.

Eames, M.J., Sepe, J., 2005. The valuation of special items. Journal of Applied Business Research21, 61–70.

Eckel, N., 1981. The income smoothing hypothesis revisited. Abacus 17, 28–40.Elliott, J.A., Hanna, J.D., 1996. Repeated accounting write-offs and the information content of

earnings. Journal of Accounting Research 34 (Supplement), 135–155.Fields, T.D., Lys, T.Z., Vincent, L., 2001. Empirical research on accounting choice. Journal of

Accounting and Economics 31, 255–307.Godfrey, J.M., Jones, K.L., 1999. Political cost influences on income smoothing via extraordinary

item classification. Accounting and Finance 39, 229–254.Gordon, M.J., 1964. Postulates, principles and research in accounting. The Accounting Review 39,

251–263.Healy, P.M., Wahlen, J.M., 1999. A review of the earnings management literature and its

implications for standard setting. Accounting Horizons 13, 365–383.Holthausen, R.W., Leftwich, R.W., 1983. The economic consequences of accounting choice:

Implications of costly contracting and monitoring. Journal of Accounting and Economics 5,77–117.

International Accounting Standards Board (IASB), 2006. Projects update: Financial StatementPresentation (Formerly known as Financial Performance Reporting by Business Enterprises)(Revised 26th May 2006). http://www.fasb.org/project/financial_statement_presentation.shtml.

Kayhan, A., Titman, S., 2007. Firms’ histories and their capital structures. Journal of FinancialEconomics 82, 1–32.

Landsman, W.R., Miller, B.L., Yeh, S., 2007. Implications of components of income excluded frompro forma earnings for future profitability and equity valuation. Journal of Business Financeand Accounting 34, 650–675.

Page 49: Classificatory income smoothing: The impact of a change in ... Classificatory income smoothing The impact of a change in...alongside the basic figure. In all, FRS3 increased the

V.E. Athanasakou et al. / Journal of Accounting and Public Policy 26 (2007) 387–435 435

Lev, B., Kunitzky, S., 1974. On the association between smoothing measures and the risk ofcommon stocks. The Accounting Review 49, 259–270.

Lin, S., 2002. The association between analysts’ forecast revisions and earnings components: Theevidence of FRS 3. British Accounting Review 34, 1–26.

Lin, S., Walker, M., 2000. FRS3 earnings, headline earnings, and accounting-based valuationmodels. Accounting and Business Research 30, 299–306.

Lougee, B., Marquardt, C., 2004. Earnings quality and strategic disclosure: An empiricalexamination of ‘pro forma’ earnings. The Accounting Review 79, 769–795.

MacKinnon, J.G., 2006. Bootstrap methods in econometrics. The Economic Record 82 (SpecialIssue), S2–S18.

Mak, C.Y., 2002. Effects of corporate refocusing on UK listed industrial companies’ valuations,reported earnings and financial analysts’ forecasts. Ph.D. thesis, University of Manchester.

Michelson, S.E., Jordan-Wagner, J., Wootton, C.W., 1995. A market based analysis of incomesmoothing. Journal of Business Finance and Accounting 22, 1179–1193.

Moses, O.D., 1987. Income smoothing and incentives: Empirical tests using accounting changes.The Accounting Review 62, 358–377.

Peasnell, K.V., Pope, P.F., Young, S., 2000. Accrual management to meet earnings targets: UKevidence pre- and post-Cadbury. British Accounting Review 32, 415–445.

Penman, S.N., 2003. Financial Statement Analysis and Security Valuation, second ed. McGraw-Hill/Irwin, New York.

Pope, P.F., Walker, M., 1999. International differences in the timeliness, conservatism, andclassification of earnings. Journal of Accounting Research 37 (Supplement), 53–87.

PricewaterhouseCoopers, 2000. Manual of Accounting: The Definitive Guide to UK AccountingLaw and Practice. Gee Publishing Limited, London.

Ronen, J., Sadan, S., 1981. Smoothing Income Numbers, Objectives, Means and Implications.Addison-Wesley Publishing Company, Reading.

Smith, T., 1992. Accounting for Growth. Century Business, London.Tonkin, D.J., Skerratt, L.C.L., 1995. Financial reporting 1994–1995: A survey of UK reporting

practice.Watts, R.L., Zimmerman, J.L., 1978. Towards a positive theory of the determination of accounting

standards. The Accounting Review 53, 112–134.White, G.E., 1970. Discretionary accounting decisions and income normalization. Journal of

Accounting Research 8, 260–273.Wooldridge, J.M., 2002. Econometric Analysis of Cross Section and Panel Data. The MIT Press,

Cambridge, MA.