Voluntary Financial Restatements: The Perspectives of Management Incentives and Pressures

Transcription

1 Chan-Jane Ln & Hu-Chun Huang Voluntary Fnancal Restatements: The Perspectves of Management Incentves and Pressures (Receved Feb 22, 2010; Frst Revson Jul 13, 2010; Second Revson Aug 8, 2014; Thrd Revson Jan 16, 2015; Accepted Oct 13, 2015) Abstract The recent fnancal msrepresentaton and accountng restatements have rased concerns regardng the relablty of fnancal statements. Fnancal restatement s the ex-post confrmaton of pror poor fnancal statements, whch may provde useful nformaton for evaluatng the credblty of present fnancal statements. Ths study nvestgates the ncentves and pressures of frms that voluntarly restate fnancal statements. A sample of 162 fnancal restatements (117 frms) announced durng s examned by usng a logstc regresson model. We fnd that frms are more lkely to voluntarly restate ther fnancal statements when the severty of msrepresentaton s less materal (.e., the msrepresentaton s not rregularty-related, does not volate the revenue recognton prncple, and s an earnngs understatement or low msrepresented earnngs), mplyng a low restatement propretary cost and threshold level of dsclosure. Next, frm CEOs wth less than 1-year tenure are nclned to voluntarly restate pror msrepresented fnancal statements because of the ncentve to respond to expectaton management or mandate. Fnally, frms wth hgh rato of outsde drectors and supervsors, a proxy for board ndependence, are more lkely to restate fnancal statements voluntarly, suggestng that board ndependence plays a role n the effectve mplementaton of governance montorng functon after msrepresented fnancal statements are released. Keywords: Msrepresentaton, Voluntary Restatement, Threshold Level of Dsclosure, New CEO, Board Independence 1. Introducton * Snce the early 2000s, scandals nvolvng Enron, WorldCom, and other corporatons have damaged the credblty of nvestors and credtors regardng fnancal statements. Moreover, the aforementoned companes earnngs decrease enormously after restatements, leadng to huge economc losses for nvestors because of ther relance on the msrepresented fnancal statements. Market partcpants may not have nformaton on frm msrepresentaton f frms volate related provsons and do not get caught (Karpoff et al. 2008). Fnancal restatement ndcates that the pror fnancal statements of a frm were msrepresented and sgnals management ntenton of earnngs manpulaton or frm turnng-down stuaton. Therefore, fnancal restatement s the prmary means through whch market partcpants can learn of msrepresentaton by frms. The announcement of restatements may cause a reducton n frm value, ncrease n frm equty or debt cost, and cut-down adjustment n management salares or bonuses. Addtonally, more and more U.S. companes attempt to hde restatements from nvestors by falng to fle the proper Form 8-Ks accordng to the report of Glass Lews & Co. (2007). 1 Even though restatements may cause adverse consequences whch frms may try to evade, frms may ntate restatements under no oblgaton from regulators or audtors. Thus, ths study nvestgates voluntary restatements by frms. Pfarrer et al. (2008) focus on how external socal and regulatory factors nduce frms voluntary dsclosure of past fnancal msconduct and Marcukatyte, Szewczyk, and Varma (2009) study the effect of corporate governance on voluntary restatements. However, lttle research examnes the nfluence of frms voluntary restatement decsons by accountng for nternal management factors. We consder restatement from the perspectve of Chan-Jane Ln s a Professor, Department of Accountng, Natonal Tawan Unversty. Hu-Chun Huang s an Assstant Professor, Department of Accountng Informaton, Natonal Tachung Unversty of Scence and Technology.(Correspondng Author) Emal: hch@nutc.edu.tw 1 Effectve Aug. 23, 2004, the U.S. Securtes and Exchange Commsson (SEC) mplemented new 8-K dsclosure rules that created a separate tem wthn the 8-K for announcng restatements and alertng nvestors to not rely on the fnancal statements affected by the restatements. Management Revew Vol. 35 (Jan 2016),

2 management and argue that the ncentves or pressures that the management encounters may be crucal factors. Accordng to the lterature, these ncentves or pressures may arse from the cost-beneft consderaton of restatement, decson property of management, or montorng of corporate governance. Frst, based on Palmrose, Rchardson, and Scholz s (2004) emprcal fndngs, restatement announcements may lead to frms market value declne, and the degree of ths declne s negatvely assocated wth the severty of msrepresentaton. The declne n market value nduced by nformaton dsclosure s the representaton of propretary cost proposed by Verreccha s (1983) dscretonary dsclosure theory. Generally, the probablty a frm dsclosng nformaton decreases wth ncreasng potental propretary costs attached to a dsclosure. Moreover, Schpper (1989) assumes that managers have prvate nformaton as to whether fnancal statements are msrepresented and the severty of msrepresentaton from the nformatonal perspectve on earnngs management. In summary, restatement, resultng n the declne n market value, may nduce propretary cost and dsclosure propretary cost ncreases wth the severty of msrepresentaton; hgh propretary cost may lead to severe market value declne. When msrepresentaton exsts n ssued fnancal statements and the severty of msrepresentaton s more materal, frms may not restate pror msrepresented fnancal statements wth the expectaton that frms market value wll declne further. Accordngly, we examne whether frms wth hgher msrepresentaton severty (.e., a hgher propretary cost of dsclosure) are less lkely to reveal past msrepresentaton (.e., to restate voluntarly). Second, pror research ndcates that whether the chef executve offcer (CEO) s at the begnnng of hs or her tenure affects management decsons. The crtcal msson of CEOs at the ntal stage s managng expectatons, respondng to mandate, adjustng to organzatonal change, and reformng performance (Hambrck and Fukutom 1991; Vancl 1987). New CEOs have an ncentve to use accountng decsons strategcally to transmt mproved performance nformaton or expose the msconduct of former CEOs. Consequently, we examne whether new CEOs are more lkely to reveal past msrepresentaton after learnng of t, relatve to those at the mddle or end of ther tenure. Fnally, accordng to a case report of restatement by the U.S. General Accountng Offce (GAO, 2002), t s mostly after specal commttee or audt commttee nvestgatons that the management of a frm s supervsed and requred to restate. Hence, the effect of ex-post detecton and montorng of corporate governance on msrepresented fnancal statements and the decson process regardng restatement may be crucal factors n a frm s decson to voluntarly restate. Thus we examne whether restatement frms wth more ndependent boards are more lkely to exert hgher decson montorng pressure on management, leadng to voluntary restatement. Subsequently, we examne whether the severty of msrepresentaton, CEO at ntal tenure, or board ndependence affects voluntary restatement f frms experence fnancal statement msrepresentaton. The emprcal results show that restatement frms are more lkely to restate voluntarly when the severty of msrepresentaton s less materal, tenure of frm CEOs s less than 1 year, and the board of drectors s more ndependent. In summary, despte restatements possbly causng adverse consequences, whch frms typcally attempt to avod, frms may stll choose to reveal msrepresentaton by restatng voluntarly after a cost beneft analyss of nformaton dsclosure, a new CEO s tendency to uncover former manager s msrepresentaton, or the ex-post supervsng and montorng pressure of an ndependent board on management regardng the dsclosure of msrepresented fnancal statements. Ths study examnes the determnants of restatement frm voluntary restatement decson and contrbutes to the lterature n the followng aspects. Frst, ths study provdes further knowledge of msrepresentaton and fnancal restatements by emphaszng the dfference between voluntary restatements and nvoluntary restatements and consderng how ncentves or pressures mposed on management nduce frms voluntary dsclosure of past msrepresentaton. Second, we consder fnancal restatements as a crtcal accountng decson and determne the relatonshp between restatement and management decson property, suggestng that fnancal restatements may be a strategc accountng decson that CEOs at an early tenure can use. Thrd, relatve to pror restatement research on the effect of corporate governance concernng Tawanese companes (Ln and Chang 2009; Shue, Ln, and Ln 2008), ths study provdes nsght nto the ex-post montorng functon of board ndependence after msrepresented fnancal statements are released. Fnally, from the perspectve of regulaton complance, frms wth new CEOs or a hghly ndependent board are ncreasngly nclned to comply wth regulatons and restate voluntarly, suggestng that the captal market can be ndrectly montored because of voluntary complance and the montorng resource can be effcently employed for the regulator. 2. Lterature and Research Hypotheses Based on ROC Statement of Audtng Standards (SAS) No. 43, fnancal statement msrepresentaton may be caused by error or fraud. 2 2 The term "error" refers to an unntentonal msstatement n fnancal statements, contanng the omsson of an amount or a dsclosure. The term "fraud" refers to an ntentonal act by one or more ndvduals among 114 Management Revew, January 2016

3 Fnancal restatement s the ex-post confrmaton of pror msrepresented fnancal statements, whch may be caused by management s aggressve accountng polcy or msuse of generally accepted accountng prncples (GAAPs) and rregulartes or errors, and the ncapablty of those charged wth governance or audtng to detect and correct error or fraud (Abbott, Parker, and Peters 2004; Elfsen and Messer 2000; Knney and McDanel 1989). Restatements can be ntrnscally classfed nto two categores accordng to normal frm actvtes or accountng polcy change, ncludng transactons regardng mergers and acqustons or dscontnued operatons, and GAAP volaton, prmarly nvestgated n prevous research (e.g. Arthaud-Day et al. 2006). Therefore, we restrct our analyss to GAAP volaton restatements. After the announcement of restatement, shareholders suffer severe economc losses (Desa, Hogan, and Wlkns 2006; Hennes, Leone, and Mller 2008; Palmrose, Rchardson, and Scholz 2004); the cost of equty captal ncreases (Hrbar and Jenkns 2004); bank loans have hgher spreads, shorter maturtes, and more covenant restrctons (Graham, L, and Qu 2008); CEOs and chef fnancal offcers are more lkely to ext ther frms and ther subsequent employment prospects are poorer (Arthaud-Day et al. 2006; Collns et al. 2009; Desa, Hogan, and Wlkns 2006); and CEOs face a decrease n opton-based compensaton (Cheng and Farber 2008). These adverse consequences nduced by restatements are the cost that may be ncurred when the management consders restatng voluntarly. Despte ths restatement cost, restatements may have potental benefts. For example, restatement helps report current fnancal nformaton more correctly because t reports on the ex-post accuracy of a frm s pror estmates (Lundholm 1999). Moreover, Palmrose, Rchardson, and Scholz (2004) suggest that detecton and revelaton by the company provde some ndcaton of relatvely stronger nternal controls and oversght by management, boards and audt commttees (p. 64). A frm s credblty may be damaged less when t restates voluntarly than when t s forced to restate by audtors or the SEC (Akhgbe, Kudla, and Madura 2005). Fnancal restatement s the predomnant means for market partcpants to dentfy frm msrepresentaton. Frms may voluntarly restate ther pror poor fnancal statements f management decdes that publcly dsclosng msrepresentaton s a cost beneft or desrable, whch consttutes voluntary restatement. Otherwse, audtors or regulators force frms to restate, whch consttutes nvoluntary restatement. 3 management, those charged wth governance, employees, or thrd partes, nvolvng the use of cheat to obtan an unjust or llegal advantage. 3 Pror research may classfy restatements ntated by companes or audtors as voluntary restatements (Agrawal and Chadha 2005); or may classfy those ntated by regulators or audtors as nvoluntary restatements (Akhgbe, Kudla, and Madura 2005; We compare voluntary and nvoluntary restatement. 2.1 Dsclosure Threshold Hypothess Based on dscretonary dsclosure theory proposed by Verreccha (1983), dsclosure may nduce a propretary cost, whch may reduce frms market value. Accordng to Verreccha (1983, p. 182), the propretary cost should be a functon of the nformaton observed by the manager. A manager exercses dscreton n dsclosng nformaton by settng a threshold level of dsclosure, above whch the observed nformaton s dsclosed, and below whch the nformaton s wthheld. The exstence of a threshold level of dsclosure suggests that as the (constant-level) propretary cost ncreases, so does the threshold level (Verreccha 1983, p. 182); the probablty of the specfc nformaton beng dsclosed may be lower f the threshold level s hgh (Verreccha 1990). Generally, the probablty of a frm dsclosng nformaton s a decreasng functon of the potental propretary costs attached to a dsclosure and s postvely assocated wth the favorableness of the news n the dsclosure. Some emprcal fndngs reveal that management has an ncentve to wthhold specfc nformaton (e.g., envronmental labltes or defned beneft penson plans), and a low propretary cost may reduce the threshold level of dsclosure and create a greater ncentve to dsclose voluntarly (L, Rchardson, and Thornton 1997; Scott 1994). Restatement announcement could reduce frm credblty, ncrease nvestor uncertanty about the future performance of the frm; and thus nvestor expectatons of future company prospects and cash flow may be reduced, and the market value of the frm s also reduced (Desa, Hogan, and Wlkns 2006; Hennes, Leone, and Mller 2008; Palmrose, Rchardson, and Scholz 2004). Accordngly, restatement may nduce propretary cost. The hgher the severty of msrepresentaton s, the more negatve the stock return s (Palmrose, Rchardson, and Scholz 2004). Ths suggests that hgh severty of msrepresentaton may ncur a hgh dsclosure propretary cost. Schpper (1989) proposes that the nformatonal perspectve on earnngs management assumes that managers have prvate nformaton that they can use when optng for accountng dscreton. Consequently, whether the fnancal statement s msrepresented and the severty of the msrepresentaton consttutes the prvate nformaton of managers. In summary, we clam that management may select the threshold level of dsclosure accordng to the severty of msrepresentaton to determne whether to restate pror msrepresented fnancal statements voluntarly. Restatement frms may experence addtonal negatve or adverse consequences f the msrepresentaton severty s more materal, mplyng a hgh Arthaud-Day et al. 2006; Marcukatyte, Szewczyk, and Varma 2009). Perspectves of Management Incentves and Pressures 115

4 propretary cost (dsclosure-related cost). Therefore, management may decde not to restate voluntarly f they expect a more severe restatement to cause an ncreased value reducton. Our frst research hypothess s constructed as follows: H1: When msrepresentaton exsts, frms wth hgher msrepresentaton severty are less lkely to restate voluntarly. 2.2 New CEO Hypothess Upper echelon theory proposed by Hambrck and Mason (1984) posts that executve demography presents compellng explanatons for varatons n ther strategc choces and consequent changes. In partcular, any CEO at the begnnng of hs or her tenure, must manage the expectatons of offcers and drectors. In other words, CEOs arduously attempt to meet the expectatons that the board or a predecessor have mplctly or explctly expressed (Vancl 1987). Based on leader lfe cycle theory (Hambrck and Fukutom 1991), newly selected CEOs are lkely to show a strategc change or seek to respond to mandate. New CEOs, beng adaptvely progressve, may provde a fresh vewpont and attempt to mplement novel deas and methods (Henderson, Mller, and Hambrck 2006). New CEOs may have a crtcal mpact on corporate strategc accountng decsons. It s suggested that new CEOs manage accruals and record wrte-offs n a way that reduces earnngs n the year of the executve change and ncreases earnngs the followng year, on the bass of blamng the prevous executves for poor performance, allowng the new management to take credt for mprovng performance (Pourcau 1993). New CEOs also have smlar ncentves to manpulate the mpressons created by the graphs n fnancal reports, presentng unfavorable (favorable) mpressons of the key fnancal varables graphed n the year of (after) CEO change (Godfrey, Mather, and Ramsay 2003). Furthermore, new CEOs focus more on nternal controls and are lkely to be more concerned about ensurng that there are no materal weaknesses; for nstance, Krshnan, Rama, and Zhang s (2008) emprcal fndngs ndcate that the appontment of new CEOs s postvely assocated wth the costs of complyng wth Secton 404 of the Sarbanes Oxley Act (SOX). New CEOs often make organzatonal or accountng changes, nvolvng more audtng, and may ask audtors to perform ncremental audtng because they want assurance on the fnancal poston of the company, leadng to hgher audtng fees (Frth 2002). Consequently, new CEOs may attempt to blame the prevous executves for msrepresentaton to manage expectatons or respond to mandate and to delver ther emphass on fnancal reportng qualty, whch can be accomplshed through fnancal restatements, leadng to the followng predcton: H2: When msrepresentaton exsts, frms wth new CEOs are more lkely to restate voluntarly. 2.3 Board Independence Hypothess Corporate governance plays a vtal role n hgh-qualty fnancal reportng (Cohen, Krshnamoorthy, and Wrght 2004), and board ndependence s crtcal to an effectve governance mechansm. For example, the board of drectors beng prmarly composed of outsde board members or havng outsde drectors wth substantal equty nterests s an essental factor (Beasley 1996; Jensen 1993). Addtonally, SOX (passed n 2002) renders more strct board ndependence regulaton; The R.O.C. Fnancal Supervsory Commsson (FSC) mplemented an ndependent drector system n 2002, whch has been appled to all lsted companes snce 2007 to enhance the effectveness of boards of drectors. Accordngly, board ndependence receves ncreasng emphass both n academc research and n practce. DeFond and Jambalvo (1991) assert that the probablty that errors are commtted s reduced by favorable control envronment that ncreases the probablty of error detecton, whch emphaszes the ex-ante montorng mechansm before fnancal statements are released. Ther emprcal fndngs ndcate that overstatements are less lkely among frms that have audt commttees. Furthermore, the ncdence of restatement exhbts a sgnfcant and negatve assocaton wth the ndependence of an audt commttee (Abbott, Parker, and Peters 2004; Agrawal and Chadha 2005) and a sgnfcant and postve assocaton wth the sze of a board (Shue, Ln, and Ln 2008). Consequently, from the perspectve of an ex-ante montorng mechansm, greater control or a more effectve board can promote hgh qualty fnancal reportng, as a result of decreasng restatement. Fama and Jensen (1983) post the exstence of an effcent drector labor market and argue that outsde drectors who perform ther board functons effectvely are lkely to be rewarded wth addtonal benefts and gan further prestge and reputaton from ths market. Thus, outsde drectors may consder that ex-post montorng after fnancal statements are released s as crucal as ex-ante montorng because of the benefts from the drector labor market. Moreover, frms wth more effectve board and audt commttee structures are more lkely to make or update an earnngs forecast (Karamanou and Vafeas 2005) and the probablty of voluntary restatements s postvely related to the ndependence of both the board of drectors and the audt commttee (Marcukatyte, Szewczyk, and Varma 2009), suggestng that effectve corporate governance may reduce the management ncentve to conceal nformaton and thus facltate voluntary dscretonary dsclosure. In summary, we clam that management may propose restatement because of the requrement 116 Management Revew, January 2016

5 Table 1 Restatement Sample Selecton Observatons Fnancal restatements announced from 1999 to Less: Restatements n the Fnance and Insurance ndustry 21 Restatement proposed by state-owned enterprse 1 Restatements due to changes n reportng entty 26 Restatements caused by changes n long-term constructon accountng methods 3 Restatements caused by vagueness of GAAPs or relevant regulatons 26 Restatements wthout restated fnancal data 23 Fnal sample 162 Table 2 Dstrbuton of Sample by Year and Industry Panel A: Dstrbuton of Sample by Year Year All TWSE/GTSM % of Sample over Voluntary Involuntary All Sample Companes All Lsted Restatement Restatement Number % Number % Companes , , , , Total , % Panel B: Dstrbuton of Sample by Industry Industry VR a IR a All Sample Industry Number % Number b % Cement Food Plastc Textle Electrc Machnery Electrcal and Cable Botechnology & Medcal Care Glass and Ceramc Paper and Pulp Iron and Steel Rubber Automoble Electronc Buldng Materal and Constructon Shppng and Transportaton Tradng and Consumers' Goods Other Total , a VR = Voluntary Restatement, IR= Involuntary Restatement b Total number of all TWSE or GTSM frms n the ndustry at the end of of ndependent drectors, mplyng effectve corporate governance. Our thrd research hypothess s fomulated as follows: H3: When msrepresentaton exsts, frms wth more ndependent boards are more lkely to restate voluntarly. 3. Research Desgn 3.1 Sample Selecton and Descrpton Palmrose, Rchardson, and Scholz (2004) suggest that attrbuton to outsde partes sgnals that Perspectves of Management Incentves and Pressures 117

6 Reason for Restatement Table 3 Summary of Reason for Restatement All Sample Voluntary Restatement Number % Number % n Group Involuntary Restatement Number % n Group Dfference Test a Perspectve of ex-ante Earnngs Management: Accrual-based EM Real EM Perspectve of ex-post Restatement: Msrepresented valuaton of long-term nvestment Audtor s nonstandard opnon Irregular related-party transacton *,**, *** denotes sgnfcance at 10%, 5%, and 1%, respectvely. a χ 2 test. χ 2 *** *** ** company montorng functons faled not only to prevent, but also to dentfy and correct a materal msstatement (p. 64). Therefore, nvoluntary restatement, mplyng that management s aware of msrepresentaton and chooses not to restate, can be used as a proxy of nformaton concealment. We use companes restatng prevous msrepresented fnancal statements as our research sample to test our research hypotheses wth respect to voluntary fnancal restatements. Accordng to the attrbuton, the sample s dvded nto voluntary restatement (ntated by the company) and nvoluntary restatement (ntated by regulators or audtors) as test and contrast samples, respectvely. 4,5 Fnancal restatements announced by TWSE or GTSM frms from January 1, 1999 to December 31, 2005 are retreved from the Tawan Economc Journal (TEJ) fnancal restatement database. To dentfy and verfy the reasons and ntators for each restatement, we further examne restatement announcements from the Market Observaton Post System to gather detaled and specfc data. Table 1 lsts our restatement sample selecton. To be ncluded n our sample, the fnancal restatements are restrcted to those caused by GAAP volaton. Intally, 262 (118) restatements (frms) are obtaned. 6 We exclude 21 restatements n the fnance and nsurance ndustry 4 Involuntary restatements n ths study nclude those ntated by the audtors, FSC, the Tawan Stock Exchange (TWSE), GreTa Securtes Market (GTSM), Accountng Research and Development Foundaton, or other governmental agences. 5 Whether the fnancal statement s msrepresented cannot be ascertaned and the severty of msrepresentaton s not measurable f nonrestatement companes are chosen as a contrast sample. The research desgn used n the current study was adopted from Marcukatyte, Szewczyk, and Varma (2009). An addtonal analyss for matchng nonrestatement companes as a control sample s reported n Secton 4. 6 Over research perod, restatng frms may announce multple restatements because of dfferent restatement reasons. Restatement announcement caused by dentcal reason for a frm s dentfed as one restatement. (TWSE ndustry code M2800), one restatement proposed by a state-owned enterprse, 26 restatements because of changes n reportng entty, 7 three restatements caused by changes n long-term constructon accountng methods, and 26 restatements resultng from vagueness of GAAPs or relevant regulatons. 8 Another 23 observatons are lost because of mssng restated earnngs data. Ths procedure yelds a sample of 162 (117) restatements (frms): 94 restatements are voluntarly ntated, and 68 restatements are forced. In Table 2, Panel A presents the chronologcal dstrbuton of our restatement sample observatons and all TWSE and GTSM companes, whch covers a seven-year perod from January 1999 to December All sample observatons account for 2.33 percent of accumulated TWSE and GTSM companes over research perod. Compared wth the steady-growth trend of all TWSE and GTSM frms over the study perod, t shows bg fluctuaton for our sample observatons. A decrease of restatement n 2001 may be caused by the deregulaton of prescrbed restatement requrement. 9 Because many accountng 7 Ths category ncludes cases nvolvng wth mergers and acqustons or dscontnued operatons. 8 Frms recognzed specfc transactons n earler ssued msrepresented fnancal statements based on relevant provsons whch were uncertan or not clearly explaned. Subsequently, these companes are nevertheless forced to restate because of regulators new explanaton for those recognzed transactons based on vague provsons. The purpose of ths procedure s to confrm that all restatement frms n the research sample are aware of the msrepresentaton. 9 Pursuant to Artcle 6 of Securtes and Exchange Act Enforcement Rules, amended n 2001, the fnancal report, beng publcly dsclosed and fled under Artcle 36 of the Securtes and Exchange Act was not made n conformance wth the relevant laws and regulatons and such non-conformance s to be corrected, shall be restated and then publcly dsclosed when the followng descrpton apples to the fnancal report: the total adjusted sum of the after tax proft and loss s equvalent to or more than NTD10,000,000, and ( or before 2001amendment) reaches one percent (1%) of the 118 Management Revew, January 2016

7 scandals occurred n the Unted States over the perod from the end of 2001 to 2002, both the regulators and the companes n Tawan pad close attenton to msrepresentaton, whch may be the reason of an ncrease of restatement n The number of voluntary and nvoluntary restatement s 94 and 68, respectvely, and accounts for and percent of the total number of sample observatons, respectvely. Panel B shows that the sample ncludes a wde range of ndustres and restatement announcements n the electronc ndustry (40, percent) represent a pluralty. Compared wth the aggregate ndustry dstrbuton, frms n the food and automoble ndustry are more lkely and those n the botechnology & medcal care and electronc ndustry are less lkely to msrepresent fnancal statements. Table 3 provdes a breakdown of the number and percentage for the reasons for restatement from the perspectves of ex-ante earnngs management and ex-post restatement, and restatement types (voluntary vs. nvoluntary restatement). From the perspectve of 10 ex-ante earnngs management, accrual-based earnngs management s the major reason for msrepresentaton (96, 59.26%); ts dstrbuton across restatement types s not statstcally sgnfcant (χ 2 = 0.305). The frequency of real earnngs management s low (33, 20.37%); however, ts dstrbuton across restatement types s statstcally sgnfcant (χ 2 = ). From the perspectve of ex-post restatement, the frequency of restatement attrbuted to msrepresented valuaton of long-term nvestment s the hghest (64, 39.51%); however, ts dstrbuton across restatement types s not statstcally sgnfcant (χ 2 = 1.810). The dstrbutons of audtor s nonstandard opnon and rregular related-party transacton across restatement types are statstcally sgnfcant (χ 2 = and 5.910, respectvely). 3.2 Emprcal Model and Varable Measurement We employ logstc regresson analyss to examne the effect of the severty of msrepresentaton, new CEOs, and board ndependence on the probablty of a frm to voluntarly restate ts earnngs. The emprcal model orgnally audted operatng revenue, or s fve percent (5%) or more of the pad-n captal of the company. 10 Earnngs management can be classfed nto two categores: real earnngs management and accrual-based earnngs management. Real earnngs management focuses on real actvtes manpulaton by changng the tmng or structurng of an operaton, nvestment, and /or fnancng transacton; for example, Typhone Food Corporaton ntentonally arranged a land sale transacton and was afterward forced to restate by FSC because the transacton was not completed. Accrual-based earnngs management relates to GAAP accountng choces to cover true economc performance. s constructed as follows. 11 The dependent varable (VOLUNTARY) was an ndcator varable that equals 1 f the restatement s voluntarly ntated, and 0 otherwse. VOLUNTARY 0 1IRREGULAR 2REVREC 3OVERSTATE 4ABSMISEARN 5NEWCEO 6OUTBS 7DUALITY 8BOARDS 9DEVIATION 10BIG4 11SIZE 12LEV 13ROA 2004 kyeark k 1999 On the bass of pror restatement studes, we construct four test varables to measure the severty of msrepresentaton to test H1. Frst, restatements nvolvng fraud are assocated wth more negatve returns (Palmrose, Rchardson, and Scholz 2004) and are treated as the most severe type by the SEC and nvestors (Hennes, Leone, and Mller 2008). Based on ROC SAS No. 43, the operatonal defnton of msrepresentaton related to rregulartes s whether the msrepresentaton s ntentonal; however, the management motvaton or ntenton cannot be judged merely by readng restatement documents. Consequently, by referrng to Hennes, Leone, and Mller (2008) and the msrepresentaton types descrbed n ROC SAS No. 43, we classfy restatements as rregulartes f the TEJ restatement descrpton mentons stuatons related to fraud or rregularty n reference to the msrepresentaton, such as msappropraton of funds, transactons nvolvng rregular or ntentonal arrangement, management dranng of funds, and Ant-Corrupton Investgaton Center nvestgatons. Then IRREGULAR s set 1 f the restatement s related to rregulartes, and 0 otherwse. Second, restatements nvolvng revenue recognton, decreasng reported ncome, or decreased reported earnngs lead to decreasng corporate market value and severe economc consequences after restatement announcements (Akhgbe, Kudla, and Madura 2005; Desa, Hogan, and Wlkns 2006; Palmrose, Rchardson, and Scholz 2004; Palmrose and Scholz 2004). Then the second measurement of msrepresentaton of severty s REVREC, equalng 1 f the restatement s related to revenue recognton, and 0 otherwse; the thrd measurement s OVERSTATE, equalng 1 f the aggregate amount of restated earnngs s less than 0, and 0 otherwse; the fourth measurement s ABSMISEARN, measured by the percentage of absolute value of the aggregate amount of restated earnngs over total assets the year pror to restatement announcement. In the lterature on the US market, management 11 Four test varables of msrepresentaton severty are measured based on all msrepresented fnancal statements for each restatement. All other varables are measured usng data from the year pror to the year of restatement announcements. Outler s treated by replacement of second outler. Perspectves of Management Incentves and Pressures 119

8 Table 4 Comparson of Characterstcs for Companes Makng Voluntary and Involuntary Restatements Varables a Voluntary Involuntary Restatement Restatement Dfference Test b Mean Medan Mean Medan t-test W test IRREGULAR *** *** REVREC *** *** OVERSTATE *** *** ABSMISEARN ** *** NEWCEO OUTBS *** 2.59 *** DUALITY BOARDS DEVIATION BIG SIZE LEV ROA *,**, *** denotes sgnfcance at 10%, 5%, and 1%, respectvely. a Varable defntons are as follows: VOLUNTARY equals 1 f the restatement s voluntarly ntated, and 0 otherwse. IRREGULAR equals 1 f the restatement s related to rregulartes, and 0 otherwse. REVREC equals 1f the restatement s related to revenue recognton, and 0 otherwse. OVERSTATE equals 1 f the aggregate amount of restated earnngs s less than 0, and 0 otherwse. ABSMISEARN s the percentage of absolute value of the aggregate amount of restated earnngs over total assets the year pror to restatement announcement. NEWCEO equals 1 f CEO tenure s less than 1 year at the restatement announcement, and 0 otherwse. OUTBS s the percentage of outsde drectors and supervsors over the total board and supervsor sze. DUALITY equals 1 f the charman s also n the poston of CEO, and 0 otherwse. BOARDS s total number of drectors and supervsors of the frm. DEVIATION s the percentage of controllng shareholder votng rghts over cash flow rghts. BIG4 equals 1 f frms are audted by the Bg Four audtors, and 0 otherwse. SIZE s natural log of total assets at end of year. LEV s the percentage of total labltes over total assets at end of year. ROA s the percentage of net ncome before nterest and tax over year-end total assets. YEAR k equals 1 f restatement announcement year s k, and 0 otherwse; k denotes from 1999 to 2004, respectvely. b We use t-test for test of dfferences n means and Wlcoxon Rank-Sum test for test of medan dfferences. s represented by the CEO; however, the top management n Tawan s often represented by the general manager, who s requred by laws and regulatons to sgn both orgnal fnancal statements and restated fnancal statements. Therefore, the operatonal defnton of a CEO s the same as that of a general manager. Some studes treat CEO tenure less than 1 year as representng a new CEO (Frth 2002), whereas others consder CEO tenure less than 2 years as representng a new CEO (Krshnan, Rama, and Zhang 2008). To test H2, NEWCEO s set as the ndcator varable f CEO tenure s less than 1 year at the restatement announcement and assumng that pror msrepresented fnancal statement s not proposed by the same CEO. Because Tawan dd not mplement the ndependent drector system unversally after 2007, we employ the percentage of outsde drectors and supervsors over the total board and supervsor sze (OUTBS) to proxy board ndependence to test H3. Accordng to the concept of ultmate control reported by La Porta, Lopez-de-Slanes, and Shlefer (1999) and Claessens, Djankov, and Lang (2000), drectors and supervsors who are not ultmate controllng shareholders are dentfed as outsde drectors and supervsors and those related to ultmate controllng shareholders through blood, marrage, or frendshp tes are excluded from outsde drectors and supervsors to objectvely evaluate the ndependence of outsde drectors. In addton, we control for other factors that pror research has suggested are lkely related to voluntary restatement decsons. If a CEO s nfluental on a board, the board s effectveness n montorng managers can be reduced and the board ndependence may be damaged (Beasley 1996; Marcukatyte, Szewczyk, and Varma 2009). Frms are postulated to be less lkely to restate voluntarly when the CEO has more nfluence on the board, mplyng a less effectve governance mechansm. Conventonal wsdom suggests that larger boards lkely reduce board effcency and thus damage montorng effectveness (Beasley 1996; Jensen 1993). The probablty of voluntary restatements s postulated to be negatvely assocated wth board sze. The larger dvergence between control rghts and ownershp rghts of the ultmate controllng shareholders represents the greater motves for ther expropratons (Claessens, Djankov, and Lang 2000; La Porta, Lopez-de-Slanes, and Shlefer 1999), mplyng a less effectve corporate governance mechansm. Therefore, the assocaton between the probablty of voluntary restatement and the dvergence between ultmate controllng shareholders control rghts and ownershp rghts s expected to be negatve. In summary, to consder the potental effect of other corporate government factors, we ncluded DUALITY (f the charman s also n the poston of CEO), board sze (BOARDS, total number of drectors and supervsors of the frm), and the 120 Management Revew, January 2016

9 Table5 Correlaton Matrx a Varable b VOL IRR REV OVER AMISE NEW OUT DUAL BOAR DEV BIG4 SIZE LEV ROA VOLUNTARY *** *** *** ** *** IRREGULAR *** *** ** REVREC *** *** * OVERSTATE *** ** * *** * ABSMISEARN *** *** * *** *** *** NEWCEO * OUTBS *** * *** ** ** DUALITY *** *** BOARDS ** *** *** *** ** DEVIATION ** ** * * BIG * *** ** * SIZE ** ** *** *** *** *** * * LEV ** ** *** ** *** ROA *** *** *** * *** *,**, *** denotes sgnfcance at 10%, 5%, and 1%, respectvely. a The upper (lower) trangle of matrx lsts Pearson (Spearman rank sum) correlaton. b VOL = VOLUNTARY, IRR = IRREGULAR, REV = REVREC, OVER = OVERSTATE, AMISE = ABSMISEARN, NEW = NEWCEO, OUT = OUTBS, DUAL = DUALITY, BOAR = BOARDS, DEV = DEVIATION. See Table 4 for varable defntons. dvergence between control rghts and cash flow rghts of the ultmate controllng shareholders (DEVIATION, measured accordng to the percentage of controllng shareholder votng rghts over cash flow rghts) n the model. The voluntary dsclosure lterature suggests that hgh-qualty audtors are nclned to requre companes to dsclose more nformaton (Frth 1979). Accordngly, we nclude audt qualty as a control varable because hgh-qualty audtors may requre a company to restate wth hgher probablty when they perceve that fnancal statements are msrepresented. On the bass of pror emprcal research (e.g., Beatty 1989), BIG4 (f frms are audted by the Bg Four audtors) s used to proxy audt qualty that may affect voluntary dsclosure. Frms wth bgger sze, hgher debt rato, and superor operatng performance are consdered to have a greater ncentve to dsclose voluntarly to reduce the agency cost and mtgate nformaton asymmetry (Dye 1985; Frth 1979; Jensen and Mecklng 1976; Watts and Zmmerman 1978); therefore, these frms are more lkely to restate voluntarly when they dscover msstatements. Thus, SIZE (natural log of total assets at end of year), LEV (debt rato, the percentage of total labltes over total assets at end of year), and operatng performance (ROA, the percentage of net ncome before nterest and tax over year-end total assets) are ncluded to be the proxes of the ncentve of voluntary dsclosure. The year ndcator (YEAR) s ncluded to control for envronmental and chronologcal dfferences. 4. Emprcal Results and Dscusson 4.1 Summary Statstcs and Correlaton Analyss Table 4 presents a comparson of the varous characterstcs of frms ssung voluntary restatements wth those forced to restate. A t test and a Wlcoxon Rank-Sum test are used for testng mean and medan dfferences, respectvely. The mean proportons of IRREGULAR and REVREC were 3% and 2% and 32% and 19% for voluntary and nvoluntary Perspectves of Management Incentves and Pressures 121

10 Table 6 Logstc Regresson Results for All Voluntary vs. Involuntary Restatements VOLUNTARY 0 1IRREGULAR 2REVREC 3OVERSTATE 4 ABSMISEARN NEWCEO OUTBS DUALITY BOARDS DEVIATION 5 BIG4 10 SIZE LEV ROA k k 1999 YEAR Varable Predct. Model 1 Model 2 Model 3 Model 4 Full Model Sgn Coef. χ 2 Coef. χ 2 Coef. χ 2 Coef. χ 2 Coef. χ 2 CONSTANT? IRREGULAR *** *** REVREC *** *** OVERSTATE *** ** ABSMISEARN ** NEWCEO ** ** * *** OUTBS *** *** ** *** *** DUALITY BOARDS DEVIATION BIG4? SIZE LEV ROA YEAR * ** ** YEAR *** *** *** *** *** YEAR ** ** ** * YEAR ** ** ** ** ** YEAR ** ** ** *** ** YEAR Model Statstcs Pseudo R Wald χ *** ** *** ** *** *,**, *** denotes sgnfcance at 10%, 5%, and 1%, respectvely. See Table 4 for varable defntons. k 9 restatements, respectvely. Intentonal msrepresentaton or msrepresentaton related to rregulartes or revenue recognton s sgnfcantly less prevalent n voluntary restatements than that n nvoluntary restatements (t = and z = for IRREGULAR; t = and z = for REVREC). The mean proporton of OVERSTATE s 54% and 84% for voluntary and nvoluntary restatements, respectvely, and s sgnfcantly dfferent (t = and z = -3.93). The mean value of ABSMISEARN s 4.49% and 9.96% for voluntary and nvoluntary restatements, respectvely, and s sgnfcantly dfferent (t = and z = -3.47). Overall, the msrepresentaton severty of voluntary restatements s lower than that of nvoluntary restatements. The mean proportons of NEWCEO for voluntary and nvoluntary restatements are 26% and 19%, respectvely. We fnd no support for H2, but obtan strong support for H3 (the effectve montorng by ndependent drectors for voluntary restatements). The proporton of outsde drectors on boards (OUTBS) for the frms that ssued voluntary restatements s hgher than that for the frms forced to restate earnngs (sgnfcant at 1%). Table 5 presents a correlaton matrx for testng varables. All of the testng varables, except for NEWCEO, are sgnfcantly correlated wth VOLUNTARY, as predcted n the hypotheses. Four test varables for the msrepresentaton severty are hghly correlated because they are constructed as a proxy for the same concept. Overall, the correlaton between ndependent varables s low, although several sgnfcant relatonshps stll occur Regresson Model Analyss and Dscusson Table 6 shows the emprcal results of the hypothess test. The frst four models n Table 6 present an analyss of the effect of ndvdual msrepresentaton severty on voluntary restatng decsons and the fnal model s a full model. The maxmum lkelhood method s employed to estmate logstc regresson. The Wald χ 2 n the full model s sgnfcant at 0.01 and the pseudo R 2 s All 12 Varance nflaton factor (VIF) dagnostc statstcs do not ndcate multcollnearty as a problem (all VIFs are less than 2.0). 122 Management Revew, January 2016

11 four estmated coeffcents of the proxes for the severty of msrepresentaton n the frst four models are negatve and sgnfcant at a level of at least 5% (p-value of IRREGULAR < p-value of REVREC = p-value of OVERSTATE < p-value of ABSMISEARN = ). In the full model, ABSMISEARN s not sgnfcant (p-value = ) and the other three testng varables are sgnfcantly negatvely correlated wth VOLUNTARY, as predcted n H1. Overall, the results reveal that a low msrepresentaton severty, mplyng a low restatement propretary cost and threshold of dsclosure, ndcates a hgh lkelhood for restatement frms to restate voluntarly. Ths result supports H1. NEWCEO exerts a sgnfcantly postve effect on VOLUNTARY (the coeffcent s and p-value = 0.011), suggestng that the restatement frms tend to restate voluntarly when CEOs are at the early phase of ther tenure to manage the expectaton or respond to mandate. The estmated coeffcent of the proxy for board ndependence (OUTBS) s sgnfcantly postve (the coeffcent s and p-value = ), whch supports H3. Ths result suggests that the restatement frms are nclned to restate voluntarly when the boards are hghly ndependent, whch s consstent wth the emprcal fndngs of Marcukatyte, Szewczyk, and Varma (2009). However, the emprcal results regardng Tawanese companes do not show that the ncdence of restatement or fraud exhbts a sgnfcant assocaton wth board ndependence (Ln and Chang 2009; Shue, Ln, and Ln 2008). The dfference of board ndependence between orgnal year of fnancal statements and the testng year (the pror year of restatement announcement) s further compared and tested to provde explanatons for ths dfference. Untabulated results show no sgnfcant dfference (t value of pared t test = and p-value = 0.506) for OUTBS n the full sample between the orgnal year (mean = and medan = 30.00) and testng year (mean = and medan = 30.00). In the orgnal year, OUTBS of voluntary restatements (mean = and medan = 30.00) s sgnfcantly hgher than that of nvoluntary restatements (mean = and medan = 29.29; t = 1.73 and z = 1.51, only sgnfcant at 10%). Nevertheless, the sgnfcance of ths comparson s weaker than that for the testng year (Table 4, sgnfcant at 1%). Because the research sample comprses restatement frms, the result of the aforementoned comparson may mply that the ex-ante montorng of ndependent boards for both voluntary and nvoluntary restatements s neffectve and thus both ssue msrepresented fnancal statements. However, n the year pror to restatement announcements, a more ndependent board of voluntary restatement frms may montor management and msconduct more effectvely when the msrepresentaton s dscovered because ndependent drectors ntend to gan more prestge and reputaton from the market and then exert greater decson pressure on management, resultng n ex-post voluntary restatement. The mplcaton s that ex-post montorng and detecton for ndependent boards s more effectve than ex-ante montorng. Regardng control varables, only YEAR s sgnfcant n the full model and the others (DUALITY, BOARDS, DEVIATION, BIG4, SIZE, LEV, and ROA) are not. All frms n the research sample are restatement companes; therefore, they may have the same characterstcs. 4.3 Addtonal Tests and Analyss We perform several addtonal tests to verfy our emprcal results. These tests nvolved usng matched nonrestatement frms, all frms as a control sample, a dfferent dentfcaton type for voluntary restatements, CEO tenure less than 2 years as NEWCEO, and the consderaton of the multrestatement effect on board ndependence. After the emprcal models are reperformed, we observe no sgnfcant dfference or contradcton n the results. The addtonal tests are descrbed as follows. (I) Senstvty to Selecton of Control Samples Although restatements caused by vagueness of GAAPs or relevant regulatons (see footnote 8 for more detaled explanaton) are excluded from the research sample to confrm that all restatement frms n the research sample are aware of the msrepresentaton, nvoluntary restatements may not be an approprate proxy for the decson not to restate. Pfarrer et al. (2008) suggest that t s common among publc corporatons to manpulate corporate earnngs through ncome smoothng, earnngs management, or explctly fraudulent behavor, and they assume that all frms n the sample are at rsk of restatng ther earnngs. Hence, nonrestatement frms may be an approprate proxy for choosng not to restate when they have msrepresentaton. Frst, we construct a control set of nonrestatement frms matched based on ndustry, sze, and close dscretonary accruals (DAs) wth reference to Shue, Ln, and Ln (2008). 13 Second, we nclude all nonrestatement frms throughout the research perod as a control sample, n accordance wth Pfarrer et al. (2008). In emprcal models for control sample testng, test varables for msrepresentaton severty are replaced by DAs and the other testng and control varables reman the same. 14 We reestmate two models for Samples A (the matched nonstatement control sample) and B (the nonrestatement contrast 13 We match each voluntary restatement sample frm wth a control frm n the same ndustry wth total assets wthn 40% of the sample frm n the year precedng the restatement announcement. We select two control frms wth close DA for each sample frm from among those meetng the aforementoned crtera. 14 DA s estmated usng Ball and Shvakumar s (2006) model. Perspectves of Management Incentves and Pressures 123