Valuation of takeover targets and auditor quality

Transcription

1 Valuaton of takeover targets and audtor qualty Lasse Nem a - Hannu Ojala a - Tom Seppälä b Forthcomng, (2013) DBW De Betrebswrtscaft Busness Admnstraton revew (BARev), 4/13 a Department of Accountng, Aalto Unversty School of Busness b Department of Informaton and Servce Management, Aalto Unversty School of Busness Abstract Ths study nvestgates whether the market s percepton of audtor qualty makes a dfference to the market value of a frm usng a sample of takeover offers n the USA over the perod 1990 to The study fnds, as hypothessed, that the takeover process makes a smaller correcton to the market prce of the target when t has a Bg 4 audtor suggestng that less prvate nformaton becomes avalable. We fnd no emprcal evdence for the audtor swtch effect, whch s an alternatve explanaton for the lower CARs for Bg 4 targets. Keywords Audt qualty, valuaton, takeovers JEL classfcaton G34; M41; M42 Acknowledgements: We are grateful to Robert Knechel, Juha Knnunen, Pontus Troberg, Jll Colls, Kar Tovanen, Derek Oler, Matt Keloharju, Sam Torstla, Seppo Ikähemo, Jar Hukku and partcpants of the EAA 2007 Conference n Lsbon, EAA 2008 Conference n Rotterdam and the AAA 2009 Conference n New York for ther helpful suggestons and remarks. The fnancal assstance of the Helsnk School of Economcs Foundaton s gratefully acknowledged.

2 Valuaton of takeover targets and audtor qualty Abstract: Ths study nvestgates whether the market s percepton of audtor qualty makes a dfference to the market value of a frm usng a sample of takeover offers n the USA over the perod 1990 to The study fnds, as hypothessed, that the takeover process makes a smaller correcton to the market prce of the target when t has a Bg 4 audtor suggestng that less prvate nformaton becomes avalable. We fnd no emprcal evdence for the audtor swtch effect, whch s an alternatve explanaton for the lower CARs for Bg 4 targets. Key words: Audt qualty, valuaton, takeovers 1. INTRODUCTION Ths paper examnes whether dfferental levels of market-assessed audtor 1 qualty have an mpact on takeover bds. The am of our nvestgaton of takeover announcements s to provde evdence on the economc mpact of the market s percepton of audtor qualty. If the market perceves that a Bg 4 audt lends hgher credblty to the fnancal reportng than a non-bg 4 audt, the frms audted by the Bg 4 should carry a lower cost of equty captal. Gven that the nformaton rsk pror to takeover negotatons s lower for more transparent Bg 4 audted target frms, the takeover process makes a smaller correcton to the market prce of the target wth a Bg 4 audtor because less prvate nformaton becomes avalable. An alternatve explanaton for lower cumulatve abnormal returns (CARs) s that antcpated target s audtor 1 In ths study, wth the term audtor we refer to a frm provdng audtng servces, not an ndvdual workng as an audtor wth the frm.

3 2 swtches from non-bg 4 to Bg 4 audtors are perceved as good news by the market and vce versa 2. Our study s motvated by a gap n the lterature on the economc value of audtor reputaton. The economc value of audtor reputaton has been assessed n the stuaton when a frm decdes to offer ts shares to the publc n ntal publc offerngs (IPOs) (e.g. Ttman/Trueman, 1986; Smunc/Sten, 1987; Balvers et al., 1988; Beatty, 1989; Datar et al., 1991; Menon/Wllams, 1991; Frth/Smth; 1992; Clarkson/Smunc, 1994) or n seasoned equty offerngs (SEOs) (Slovn et al., 1990; Zhou/Elder, 2004; Km/Park, 2006; Rauterkus/Song, 2005). The credblty that audtor reputaton lends to the earnngs numbers reported n annual reports has also ganed research attenton (Teoh/Wong, 1993; Francs/Ke, 2006; Haw et al., 2008). Audtor swtches provde another nterestng context to examne the effects of the dfferental levels of reputaton captal of the audtors on frm value (e.g. Nchols/Smth, 1983; Echenseher et al., 1989; Johnson/Lys, 1990). Also studes on how audt reports ndcatng gong concern problems ( bad news ) affect share prces add to our understandng on the economc value of audtng at more general level 3. Lke SEOs (Slovn et al., 1990; Zhou/Elder, 2004; Km/Park, 2006), busness takeovers provde a good settng n whch to observe the systematc dfferences n frm values subsequent to IPO snce takeovers facltate a more powerful test of the economc value of audtor 2 Accordng to studes on audtor swtches n takeovers most takeovers result n a change n target's audtor, but some retan the ncumbent audtor (Andersson et al., 1993 and Frth, 1999) 3 In balance, event studes do not provde clear and consstent evdence on market reacton to gong concern reports (e.g. Ogneva/Subramanyam, 2007; Martnez/Martnez/Benau, 2004; Al-Thunebat/Khamees/Al-Fayoum, 2008), whereas some expermental studes (e.g. O Relly, 2010) have been able to document a negatve reacton. Ths dscrepancy may ether relate to methodologcal dfferences between event studes and experments or t may relate to effcency of captal markets. As Church et al. (2008:82) conclude n ther lterature revew artcle on audtor reportng: the market reacts pror to the ssuance of the audtor s report, whch suggests that the market s nterested n the underlyng economc event as opposed to the audtor s communcaton of the event. Consstent wth ths, Dodd et al. (1984) fnd a negatve market reacton pror to the release of a qualfed audt opnon, but lttle evdence of a stock prce effect when the opnon s dsclosed publcly.

4 3 reputaton than audtor swtches or earnngs numbers n annual reports. Audtor swtches and earnngs are fnancal quarter- or year-end nformaton and therefore dsclosed smultaneously wth other annual performance and corporate governance nformaton. However, the effect of audtor reputaton on frm valuaton n the takeover context has receved surprsngly lttle research attenton, whereas, for example, motves for takeovers have been a focus of pror research for many years (e.g. Jensen 1986; Roll 1986; Bradley et al., 1988; Dong et al., 2006; Hope/Wayne 2008). The man contrbuton of our study s two-fold. Frst, we are not aware of any prevous studes that have examned the lnk between audtor qualty and the valuaton of takeover targets. Therefore, through ts focus on the target audtor, our study contrbutes to the lterature on the economc value of audtor reputaton by provdng emprcal evdence that lower cumulatve abnormal returns (CARs) are assocated wth takeover announcements of Bg 4 audted target frms. Ths s an mportant fndng, as t shows that the effect of audtor reputaton on clent frm value s not lmted to the event when a prvately held frm goes publc. Instead, consstent wth studes on SEOs (Slovn et al., 1990; Zhou/Elder, 2004; Km/Park, 2006; Rauterkus/Song, 2005), the same reputaton effect s sustaned after the IPO event. The second contrbuton of the study s our emprcal examnaton, whch suggests that the expected audtor swtches do not have an effect on CARs. In the majorty of takeovers the target s audtor s replaced by the acqurer s audtor (Anderson et al., 1993; Frth, 1999). Therefore, market partcpants may antcpate audtor swtch n the target frm at the tme of the takeover announcement. Consequently, changes from non-bg 4 to Bg 4 (Bg 4 to non-bg 4) can be perceved as qualty upgrades ( downgrades ). Our analyss of these expected audtor swtches extends the lterature on the market reactons to audtor swtches. Consstent wth some

5 4 studes examnng market reactons to audtor qualty upgrades and downgrades (e.g. Nchols/Smth, 1983; Johnson/Lys, 1990), our fndngs suggest that the market does not react at the tme of the takeover announcement to the expected future upgrades or downgrades of audtor qualty. The remander of ths paper s organzed as follows. Secton 2 revews relevant pror lterature and develops the hypotheses. The data used n the emprcal tests, followed by the research desgn are explaned n Secton 3. Secton 4 descrbes the emprcal results, and Secton 5 concludes the study. 2. DEVELOPMENT OF HYPOTHESES The credblty that an audt lends to fnancal statements s the key element n the lnk between the qualty of audt servces and frm valuaton. As audt qualty s unobservable to outsders, are all supplers of audt qualty regarded as the same (cf. Akerlof, 1970) or s t possble for some audt frms to ncrease ther market-assessed credblty above that attached to the professon n general? When buyers cannot observe qualty pror to a purchase, the reputaton of the suppler provdes a mechansm that sgnals superor qualty (e.g. Shapro, 1983; Rley, 2001). Pror theoretcal work provdes two related explanatons for how audt frms can acqure an above average reputaton (.e. hgher level of market-assessed credblty). Both of these explanatons suggest that the reputaton s audt frm-specfc (.e. the same level of credblty s delvered to all clents of the frm). DeAngelo s (1981) well-known work proposes that the audtor s nvestments n clent relatonshps (start-up costs) enable the ncumbent audtor to earn

6 5 clent-specfc quas-rents, and that these quas-rents represent the collateral that s lost f promses are not kept (.e. audt falure). Thus, audtors wth a larger number of clents possess greater total collateral and consequently have more to lose n audt falure. DeAngelo argues that even f professonal competence s the same across the professon, varaton n market-assessed qualty levels wll arse from the probablty that the audtor wll report the errors found. In other words, varaton n market-assessed qualty levels arses from dfferental levels of ndependence n appearance, whch s mechancally related to the number of quas-rents (clents) of the audtor (.e. bgger s nevtably better). Another lne of argument s that n order to acheve a reputaton for above average servces, the suppler must frst nvest more than others n the qualty of servces. The market wll eventually learn about the hgher level of qualty and wll be wllng to pay for t (Klen/Leffler, 1981; Shapro, 1983). Because a reputaton s easly destroyed, and rebuldng t s costly, reputaton captal (brand name) serves as collateral that promses wll be kept. Consstent wth ths asserton, Ttman/Trueman (1986) defne audtor qualty as the accuracy of the nformaton provded, whch allows nvestors to make a more precse estmate of the clent frm s value. A frm wth more favorable nformaton wll be wllng to pay a hgher fee for a more accurate audtor. A frm wth less favorable nformaton has no ncentve to pay a hgher fee for a more accurate audtor, as the audtor s nformaton s lkely to be unfavorable. As a result, the choce of audtor per se serves as a mechansm sgnalng prvate nformaton about the frm s future prospects to the market. In lne wth Ttman/Trueman (1986), Bg 4 frms are assocated wth more accurate reports and more nformatve sgnals of fnancal dstress (Petron/Beasley; 1996; Lennox, 1999), lower ltgaton actvty (Palmrose, 1988), and ther clents fnancal statements are assocated wth

7 6 hgher complance wth GAAP dsclosure requrements (e.g. Krshnan/Schauer, 2000). The fnancal statements of Bg 4 clents are also assocated wth lower abnormal accruals, whch can be nterpreted as an ndcaton of hgher earnngs qualty (Becker et al., 1998; Francs et al., 1999) 4. Also ndcatve of hgher earnngs qualty of Bg 4 clents, Teoh/Wong (1993) found that the stock market reacton to earnngs surprses from Bg 4 clents was greater than that of other frms. Taken together, these studes are well n lne wth the vew that a hgher audtor reputaton reduces the uncertanty of future cash flows among nvestors and therefore reduces the cost of equty captal, and ths s reflected n a hgher value of the frm 5. However, Francs/Ke (2006) clearly show the sgnfcance of audtor ndependence n appearance. Utlzng change n the dsclosure requrements of fees pad to the ncumbent audtor, Francs/Ke (2006) found that the market valuaton of earnngs surprses was lower for frms wth hgh levels than for those wth low levels of non-audt fees, but there was no reducton n the precedng year when no dsclosures were requred. Therefore, ther fndngs suggest that dsclosng hgh levels of non-audt fees provded new nformaton to the market related to audtor ndependence and earnngs qualty. Ths fndng s mportant, as t shows that market perceved audt qualty s not entrely audt frm specfc. Ths s well n lne wth studes on ndustry specalzaton n audtng, whch suggest that audtor reputaton may vary to some extent across ndustres (e.g. Craswell et al., 1995). 4 Recently, usng matchng models Lawrence et al., 2011 fnd no statstcally sgnfcant dfference n ther three audt qualty proxes (dscretonary accruals, the ex ante cost-of-equty captal, and analyst forecast accuracy) between Bg 4 and non-bg 4 audtors. 5 Botosan (2006) provdes a lterature revew on the lnk between dsclosure and the cost of captal. She concludes that the sum total of the evdence accumulated across many studes usng alternatve measures, samples and research desgns lends consderable support to the hypothess that greater dsclosure reduces the cost of equty captal (Botosan 2006:39). A recent analytcal study, however, descrbes a stuaton when dsclosure qualty may ncrease the cost of captal (Gao 2010). Gao s model predcts that dsclosure qualty ncreases the cost of captal when t ncreases the overall rsk of the frm s cash flow, suggestng that dsclosure qualty does not always monotoncally reduce the cost of captal.

8 7 Also, there mght be specfc stuatons when non-bg 4 frms have compettve edge over Bg 4 frms. Lous (2005) who focuses on the lnk between acqurers abnormal returns and audtor qualty n connecton wth merger transactons, argues that when targets are small and n partcular are prvately held companes, non-bg 4 audtors provde better advsory servces to the managers of the acqurer frm because of ther local knowledge and because the Bg 4 tend to neglect small clents n favor of more lucratve busnesses wth larger clents. Accordng to Lous, due to the superor advsory servces that non-bg 4 audtors provde, acqurers audted by non-bg 4 frms outperform those audted by the Bg 4 at merger announcements. 6 However, even f there were some varaton n perceved audt qualty across audts of a gven suppler, t would not dspute the theores on audtor reputaton. As Francs (2004: 352) ponted out: the arguments smply mean that audts of Bg 4 frms as a group wll, on average, be of hgher qualty than other (smaller) accountng frms. Overall, consstent wth the vew that they are perceved as hgher qualty audtors, the Bg 4 audtors are found to charge hgher fees than other supplers of audt servces (Yardley et al., 1992; Walker/Johnson, 1996; Mozer, 1997; Taylor/Smon, 1999; Hay et al., 2006). One mportant reason for the wllngness to pay hgher Bg 4 fees seems to be the ablty of the Bg 4 to reduce the uncertanty of future cash flows of the clent frm, and therefore underprcng of ts shares (Balvers et al., 1988; Beatty, 1989; Hogan, 1997; Wllenborg, 1999). Followng these suggestons and emprcal fndngs, we post that pror to the takeover announcement, non-bg 4 audted takeover targets carry more nformaton rsk than Bg 4 6 However, ths argument does not apply to target frms and ther audtors, whch s the focus of ths paper. In other words, qualty of advsory servces provded by the target frm s audtor cannot be assumed to affect the acqurer s takeover decson (.e. bd) and market value of the target frm. Nevertheless, even f the argument of superor advce by non-bg 4 audtors does not apply to target frms, the study of Lous (2005) s related to ours, as t dentfes the factors that are expected to have an effect on takeover bds.

9 8 audted takeover targets. However, n the takeover, acqurers are lkely to have access to prvate nformaton on the target frm not avalable to other outsders, and hence acqurers potentally yeld more accurate valuatons than the equty market as a whole. The emprcal fndngs of Raman et al. (2008) are consstent wth a value dscount of more opaque lsted takeover targets. They fnd, among other thngs, that when the target s fnancal reportng qualty s poor, negotatons wth the target s management generate addtonal nformaton that leads to a hgher premum beng offered. In short, earnngs qualty and takeover premums are negatvely correlated, suggestng that the takeover process reveals valuable prvate nformaton to the bdder, especally when earnngs qualty s poor. Ths result s stronger for prvate bdders than publc bdders. The nverse relaton of earnngs qualty and takeover premums documented by Raman et al. (2008) s consstent wth our argument that Bg 4 audts lend hgher credblty to fnancal statements, and the surprses for more transparent Bg 4 clents are therefore smaller than for more opaque non-bg 4 clents, resultng n hgher abnormal returns for the non-bg 4 targets. In summary, gven that Bg 4 clent frms carry less nformaton rsk than non-bg 4 clents pror to the takeover process, acqurers access to prvate nformaton durng the takeover process should reduce the nformaton rsk to a lesser extent than n the case of non-bg 4 clents, ceters parbus. The bd (takeover announcement) sgnals ths new prvate nformaton about frm value to the equty market, reflectng an ncrease n the market prce of the target frm. For Bg 4 audted, more transparent targets, less prvate nformaton s avalable through the takeover process to be sgnaled n the bd, leadng to lower abnormal returns on average. Accordngly, we set our hypothess as follows:

10 9 H1: The cumulatve abnormal returns of Bg 4 audted takeover targets are lower than those of non-bg 4 audted takeover targets, ceters parbus. An alternatve explanaton for lower CARs for Bg 4 targets could be the target s future audtor swtch antcpated by market partcpants at the tme of takeover announcement. The antcpaton may be warranted as n the majorty of takeovers the target s audtor s replaced by the acqurer s audtor: Anderson et al. (1993) fnd a swtch n 73% and Frth (1999) n 80% of the takeovers. In general, studes on factors assocated wth audtor swtches fnd that a corporate takeover or a merger s one sgnfcant reason for audtor swtches (Beatte/Fearnley, 1995; 1998; Woo/Koh, 2001). Therefore, f the market s percepton of audtor qualty affects the cost of captal of the clent, audtor swtches may be perceved as qualty upgrades or downgrades. A stream of research has examned market reactons to audtor swtches (e.g. Nchols/Smth, 1983; Echenseher et al., 1989; Johnson/Lys, 1990). On balance, the results of these studes suggest that audtor swtches are generally vewed unfavorably by the market, but swtches to Bg 4 audtors tend to be vewed more favorably than other audtor swtches, supportng the vew that a change from non- Bg 4 to Bg 4 can be taken as a postve sgnal ( good news ) for future growth prospects. However, these studes do not provde clear and consstent evdence of market reactons to audtor swtches. For example, Johnson/Lys (1990) found that the market reacts pror to the swtches, ndcatng that changes n a clent s operatons and actvtes trgger the audtor swtch to a more cost-effcent audtor, and the swtch tself therefore has lttle nformaton content. However, n the context of takeover, the audtor swtch s not trggered by changes n the clent frm s operatons, but can be expected to happen smply because the audtor of the target frm s

11 10 lkely to be replaced by the acqurer s audtor. Therefore, the audtor swtch n a takeover may cause a stronger market reacton than n other stuatons. To examne the effect of nformaton rsk wthout the swtch effect, we examne H1 n a more powerful way by elmnatng the potental effect of antcpated swtch from the hypotheses: H2a: The cumulatve abnormal returns of Bg 4 audted takeover targets are lower than non-bg 4 audted takeover targets after the effects of the expected audtor swtch are controlled for, ceters parbus. The above hypotheszed valuaton dscount assocated wth the fact that a frm has a non-bg 4 audtor (H1) should no longer be warranted f the target frm swtches to a Bg 4 audtor. Therefore, f the target s taken over by a frm wth a Bg 4 audtor, t could be possble that the market antcpates ths qualty upgrade (as acqurer s audtor s known) and would result n larger abnormal returns to those non-bg 4 audted targets that are expected to swtch from non- Bg 4 to Bg 4. However, the swtch effect s two-way snce the audtor qualty of acqurers may also trgger an expected downward shft n the audtor qualty of the takeover target. As an alternatve explanaton for the lower CAR for Bg 4 targets, we examne the hypotheses H2b and H2c related to an expected audtor upgrade and downgrade swtch. Frst, we examne the effect of an expected audtor upgrade. H2b: The cumulatve abnormal returns are hgher for non-bg 4 audted takeover targets that are expected to upgrade to Bg 4 audtors than for those that are expected to mantan non-bg 4 audtors, ceters parbus.

12 11 Secondly, we examne the effect of an expected audtor downgrade: H2c: The cumulatve abnormal returns are hgher for Bg 4 audted takeover targets that are expected to mantan Bg 4 audtors than for those that are expected to downgrade to non-bg 4 audtors, ceters parbus. 3. DATA AND RESEARCH DESIGN Our analyss s based on a sample of frms selected from the Securtes Data Corporaton (SDC), Audt Analytcs and Thomson One Banker, whch contans fnancal data from the Worldscope, Compustat, I/B/E/S and Datastream databases. For the examnaton of H1 and H2, we retreve a total of takeover bds (268 non-bg 4 audted and Bg 4 audted) that meet the followng crtera. Frst, the takeover target frms are lsted n a US stock exchange, whch provde us wth nformaton on the changes n share value. Second, for homogenety, we only accept acqustons where the acqurer seeks to obtan over 50% of the shares. Thrd, we requre nformaton on the audtor of the target and acqurer. Fourth, we do not accept frms wth negatve shareholders equty, because they are lkely to be n fnancal dstress (Bowen et al., 1995). We nclude both successful and unsuccessful bds n our sample. In the examnaton of H1 and H2, for the computaton of abnormal returns, we couple the share prce data from the SDC wth the MSCI US total market ndex obtaned from Datastream. In the multvarate examnaton of the hypotheses, we have excluded observatons f the Belsley/Kuh/Welsch (1980) dagnostc ndcated that they were nfluental (absolute value of the studentzed resdual greater than 5 or Cook s D statstc greater than 1).

13 12 The examnaton of H1 and H2 requres us to dentfy the quantty of cumulatve abnormal returns. We compute the announcement-perod cumulatve abnormal returns (CARs) for the three-day wndow (one day before to one day after) around the announcement day,.e. day 0. For the market return, we use the Morgan & Stanley U.S. return ndex. Followng pror study by Fuller et al. (2002) and Dong et al. (2006), we employ the modfed market model where r s the target frm- return and r m s the market return: CAR r r m. (1) We estmate the followng multvarate regresson model for CAR, ncludng the audtor effect and control varables for factors known to affect CARs: CAR 0 ( audtor characterstcs) ( target ( acqurer frm characterstcs) frm characterstcs) ( deal characterstcs). (2) In the above, audtor characterstcs s a vector ncludng audtor characterstcs TBIG4, ABIG4, and nteracton term TBIG4 ABIG4 (explaned below)., 1 2, 3,, and are the regresson coeffcent vectors and s the error term of the model. In the followng, we descrbe the varables of our model n detal. Frst, we descrbe the varables related to audtor characterstcs, whch are the varables of man nterest. We then descrbe the control varables that capture the relevant target frm, acqurer frm, and deal characterstcs. These control

14 13 varables are based on Lous (2005), who dentfes the factors that are expected to have an effect on takeover bds. Audtor characterstcs Our man nterest les n the margnal effects of the dchotomous varables TBIG4 and ABIG4, and ther nteracton TBIG4 ABIG4. TBIG4 equals 1 f the target frm has a Bg 4 audtor, and 0 otherwse. ABIG4 equals 1 f the acqurer frm has a Bg 4 audtor, and 0 otherwse. Consequently, we wll examne ther regresson coeffcents, = 1, 2, 3. We nclude a dchotomous varable MIDTIER, whch dstngushes between second ter versus other target s audtors. It equals 1 f the target has been audted by BDO or Grant Thornton, and 0 otherwse. Frst, we focus on the regresson coeffcent 1 of TBIG4 n equaton (2): f the shareholders of Bg 4 audted target frms obtan lower cumulatve abnormal returns (CARs) than non-bg 4 audted frms, ceters parbus, then 1 < 0. H1 does not take a stance on acqurer audtor characterstcs. Hence, at ths pont we exclude the varables ABIG4 and TBIG4 ABIG4, whch measure the audtor qualty of the acqurer, and ther correspondng coeffcents 2 and 3 from the examnaton of H1. To further study the effects of audtor qualty on the CAR of takeover targets, and n partcular to nclude the effects of the expected audtor swtch n the examnaton, we have developed a 2x2 cross table analyss (Fgure 1) that shows how we can calculate the condtonal effects on CAR for each of the four combnatons of acqurer and target audtor qualty. Fgure 1 presents these combnatons by dsplayng the non-bg 4 and Bg 4 audted targets n the

15 14 horzontal dmenson of the cross-table, and the non-bg 4 and Bg 4 audted acqurers n the vertcal dmenson of the cross-table. We label the quadrants clockwse from the bottom-left as follows: Low qualty, Upgrade, Hgh qualty and Downgrade. H2a examnes the dagonal dfference between the Hgh qualty and Low qualty quadrants and posts that the CAR of the Low qualty quadrant s hgher than that of the Hgh qualty quadrant. In the latter par of quadrants, the target and the acqurer have the same audtor qualty classfcaton. Namely, n the quadrant Low qualty, the takeover targets and the acqurers have non-bg 4 audtors and n the quadrant Hgh qualty they both have Bg 4 audtors. Consequently, the acquston bds do not trgger expectatons of audtor swtches. Therefore, cost of captal revsons should not be expected. In the quadrant Low qualty, the target and the acqurer have non-bg 4 audtors, and therefore the hghest ntal nformaton rsk and the hghest CAR of the quadrants. The quadrant Hgh qualty dsplays a contrastng stuaton wth hgh qualty audtors where frms have the lowest ntal nformaton rsk and, as a result of ths, the lowest CAR of the quadrants. We wll measure the condtonal CAR of the Low qualty quadrant, n whch the target, acqurer, and deal characterstcs are controlled for, as the regresson coeffcent of the ntercept, 0, of regresson equaton (2); ths s because TBIG4 = ABIG4 = TBIG4 ABIG4 = 0 n ths case. The condtonal CAR of the hgh qualty quadrant wll be measured as the sum of regresson equaton (2), snce TBIG4 = ABIG4 = TBIG4 ABIG4 = 1 n ths case. Our statstcal, 0 test for H2a s thus related to the dfference n the condtonal effects and can be stated as 0, postng a lower CAR for the Hgh qualty quadrant compared to the Low qualty quadrant.

16 15 (Insert Fgure 1 here) In Fgure 1 we also present the operatonalsaton of H2b and H2c. H2b posts that the CAR of the quadrant Upgrade s hgher than that of the quadrant Low qualty. In the quadrants Upgrade and Low qualty, takeover targets have non-bg 4 audtors. In the Upgrade quadrant, acqurers have a Bg 4 audtor, whch rases expectatons of a decrease n the cost of captal. Ths, n turn, ncreases the CAR of the target, ceters parbus. In the quadrant Low qualty, no such expectatons of an audtor swtch exst. Therefore, the CAR of the quadrant Upgrade should be hgher than that of low qualty. The condtonal CAR of the Upgrade quadrant s measured as the sum 0 2 based on regresson equaton (2). Consequently, our statstcal test for H2b s formulated as 2 0. H2c examnes the opposte audtor swtch effect compared to H2b. In the quadrant Downgrade the targets have Bg 4 audtors and the acqurers have non-bg 4 audtors. The dfference n audtor qualty between the target and the acqurer trggers expectatons of a downgrade, ncreasng the cost of captal and decreasng the CAR of the quadrant Downgrade compared to the quadrant Hgh qualty. The condtonal CAR of the Downgrade quadrant s measured as the sum of regresson coeffcents 0 and 1 of regresson equaton (2). Our statstcal test for H2c s based on the dfference between the effects of these quadrants,.e., and the hypothess s 0, thus postng that downgradng should reduce the CAR. Regresson equaton (2) ncludes varous control varables for the target

17 16 frm, acqurer frm, and acquston characterstcs that may have an mpact on target returns around a takeover bd. These wll be explaned next. Target frm characterstcs We control for the followng target frm characterstcs. ACOV s analyst coverage, measured by the number of analysts forecastng the target s annual earnngs n the month mmedately pror to the earnngs announcement. CASH s the target s cash dvded by the total assets, a proxy for lqudty. E/P s the target s earnngs to prce rato. INHOUSE s a dchotomous varable, whch equals 1 f the target frm does not use an nvestment banker (as dentfed by the SDC), and 0 otherwse. LEVERAGE s the target s total debt dvded by the total shareholders equty. NIB s the number of nvestment bankers hred by the target (as dentfed by the SDC). P/B s the target s prce-to-book rato. TROE s the target s return-on-equty rato, whch we measure from the last fscal year that ended before the bd announcement. TSIZE s the natural logarthm of the target s total assets. VOLATILITY s the target s pre-bd stock volatlty measured as the standard devaton of the target s return over the perod from 60 to 259 days before the merger announcement. QUALIFIED s a dchotomous varable, whch equals 1 for target frms wth gong concern audt opnons, and 0 otherwse. We allow for separate ntercepts for frms n regulated SIC ndustres (REGIND) 49 (energy) or (fnancal nsttutons), agrculture (two-dgt SIC code 1-9), constructon (two-dgt SIC code 15-17), manufacturng (two-dgt SIC code 20-39), transport (two-dgt SIC code 40-48), wholesale (two-dgt SIC code 50-51), retal (two-dgt SIC code 52-59), and servce (twodgt SIC code 70-89).

18 17 Acqurer characterstcs We control for the followng acqurer frm characterstcs: the acqurer s sze (AASSETS), ndustry relatedness (INDR), and relatve sze (RELSIZE). AASSETS s the natural logarthm of the acqurer s total assets. INDR s a proxy for the ndustry relatedness of the target and the acqurer. It s equal to 1 f the target and acqurer are n the same two-dgt SIC code and 0 otherwse. RELSIZE s the rato of the target s total assets to the acqurer s total assets. Deal characterstcs We control for the followng acquston characterstcs: the atttude towards the acquston bd (FRIENDLY), the consoldaton method n the fnancal reports of the acqurer (POOL), potental rumors regardng the acquston bd (RUMORED), the payment method (STOCK), and fxed year effects. FRIENDLY equals 1 f the target s atttude to the proposed merger s characterzed as frendly by the SDC, and 0 otherwse. POOL s a dchotomous varable. It equals 1 f the acqurer wll use the poolng-of-nterest method n the consoldated fnancal statements as opposed to the purchase method of consoldaton, and 0 otherwse. RUMORED equals 1 f the SDC classfes the bd as rumored, and 0 otherwse. STOCK s the proporton of common stock used as payment for the target s shares. We allow for separate ntercepts for all calendar years (YEAR) except of the last calendar year of the examned perod, 2005, whch we leave n the ntercept. After characterzng the explanatory varables, we can now state the regresson model n full detal:

19 18 CAR TBIG4 k 1 k INDUSTRY k, ACOV CASH EP INHOUSE LEVERAGE NIB AASSETS PB TROE TSIZE VOLATILITY QUALIFIED ABIG4 TBIG4 ABIG4 MIDTIER POOL RUMORED STOCK t 1 INDR RELSIZE FRIENDLY YEAR t t, (3) where refers to the error term and to frm. Addressng potental self selecton bas usng two stage-modelng: audtor choce model There s a possblty that the sub-samples audted by Bg 4 and non-bg 4 audtors dffer n a large number of frm characterstcs, and that self-selecton bas may be present. To account for possble sample selecton bas n the audtor choce, we apply Heckman s (1979) two-step procedure to examne the mpact of two-stage modelng on our results as follows. We frst estmate the audtor choce model usng probt regresson Prob( TBIG4 1) ( Z ) where denotes the cumulatve normal densty functon, and _ (4) Z 1CASH 2E P 3LEVERAGE 4TSIZE 5VOLATILITY 6 0 / REGIND where =0, 1, 2, 3, 4, 5, 6 refer to the regresson coeffcents of the selecton model. The model predcts the probablty of the audtor choce based on the same explanatory varables that have earler been used n the two-step audtor choce model by Lous (2005).

20 19 These explanatory varables (CASH, E/P, LEVERAGE, TSIZE, VOLATILITY and REGIND) have been defned earler n the target frm characterstcs secton. Then, usng the ordnary least squares method, we re-estmate regresson model (3) for CAR. We nclude the nverse Mll s rato (LAMBDA) estmated from the probt model (4) n the frst step and by excludng those control varables that were already ncluded n the audtor choce model (.e. CASH, E/P, LEVERAGE, TSIZE, VOLATILITY and REGIND). 4. RESULTS Descrptve statstcs In Table 1 we report the descrptve statstcs of our data separately for the contnuous varables (Panel A) and dchotomous varables (Panel B). The mean, medan and standard devaton of the contnuous varables used n regresson equaton (3) are presented n Panel A. The results of the t-tests comparng the non-bg 4 and Bg 4 groups wth respect to each contnuous varable are shown there. We also report the non-parametrc Mann-Whtney- Wlcoxon U-test for the equalty of the medans of the two groups. The U-test s preferred n the case of skewed dstrbutons of the underlyng varables. The dchotomous varables and ther relatve frequences are presented n Panel B of Table 1. To compare the frequences of the non-bg 4 and Bg 4 groups we use Fsher s exact test, whch s more accurate than the conventonal 2 -test for 2x2 tables, especally f the proportons are close to 0 or 1.

21 20 For almost all of the consdered varables, the undmensonal test shows a statstcally sgnfcant dfference n the means of the non-bg 4 and Bg 4 groups. Of the contnuous varables, STOCK s the only non-sgnfcant varable n both the t-test and the U-test, ndcatng that the proporton of common stock used as payment does not dffer between the non-bg 4 and Bg 4 groups. (Insert Table 1 here) Interestngly, n relaton to hypothess H1, the mean CAR for the three days surroundng the takeover bd announcement s hgher for the non-bg 4 audted target frms (15.7%) compared to Bg 4 audted target frms (13.1%). The dfference s statstcally sgnfcant at the sgnfcance level p = Analyst coverage (ACOV) s broader (p < 0.001) and lqudty (CASH) s sgnfcantly hgher for Bg 4 audted targets (16.6%) compared to non-bg 4 audted targets (8.8%) (p < 0.001). Non-Bg 4 audted frms are rsker, as ther mean LEVERAGE s compared to for the Bg 4 audted targets (p < 0.001). We can also see that on average the non-bg 4 frms have fewer nvestment bankers (NIB) nvolved n the acquston process than the Bg 4 frms (p < 0.001). The mean of the prce-to-book rato n the subsample of non-bg 4 audted targets s lower (P/B = 2.190) than n the subsample of Bg 4 audted takeover targets (P/B = 3.363) (p < 0.001). The average proftablty (measured as return on equty, TROE) of Bg 4 audted targets s lower than that of non-bg 4 audted targets. Because the mean n the Bg 4 group s negatve and thus much lower than the medan, the proftablty dstrbuton of the Bg 4 group s extremely left-skewed,.e. t nvolves some frms wth very negatve proftabltes. The non-bg

22 21 4 group also has a left-skewed proftablty dstrbuton, but not as severe as the Bg 4 group. In addton, the standard devaton of the Bg 4 group s very hgh, Smlar arguments that apply to the target ROE also apply to the earnngs-to-prce rato (E/P). The Bg 4 audted targets are larger on average than the non-bg 4 audted targets (TSIZE), wth a sgnfcance level p < The VOLATILITY of the Bg 4 targets s hgher, beng 3.6%, compared to 3.0% for non-bg 4 targets (p < 0.001). The mean of the acqurer s assets (AASSETS) s sgnfcantly hgher at the 5% level when the target s a Bg 4 audted company than when t s not. The average relatve sze s sgnfcantly hgher at the 5% level when the target s Bg 4 audted compared to non-bg audted targets. Between non-bg 4 and Bg 4 audted targets, there s a statstcally sgnfcant dfference n how often the acqurer s Bg 4 audted (p < 0.001), the target and the acqurer are related by ndustry (p = 0.003), whether an INHOUSE advsor s used (p = 0.016). There are 76 md-ter audtors (28.36%) n the subgroup non-bg 4 audtors, and by defnton none n the subgroup of Bg 4 audtors. We can also see that there are dfferences n how often the bds have appeared n dfferent years and n dfferent ndustres. The manufacturng and servce sectors are the two largest groups n our sample, but both of them are less represented n the sample of non-bg 4 audted frms (11.6% and 9.0%, respectvely) compared to the sample of Bg 4 audted frms (24.9% and 21.0%, respectvely). There s no statstcally sgnfcant dfference n how often the merger s consdered FRIENDLY.

23 22 The audtor choce model We can see from Table 2 that the audtor choce model s sgnfcant at level p < It s ablty to classfy the observatons correctly s 82.1 %. Nagelkerke s (1991) coeffcent of determnaton Pseudo-R 2 s 27.2%. CASH, E/P, LEVERAGE, TSIZE and REGIND are all sgnfcant at 0.1 % level. Our model ft s almost dentcal to a smlar model developed by Lous (2005) for the acqurers. (Insert Table 2 here) The cumulatve abnormal returns of Bg 4 audted and non-bg 4 audted takeover targets (H1) Panel A of Table 1 provdes unvarate results that support H1 snce the mean of the CAR for non-bg 4 audted takeover targets s hgher than that of Bg 4 audted takeover targets (p = 0.019). Then, usng the ordnary least squares method, we estmate a regresson model (equaton 3) for CAR, ncludng the audtor effect and control varables as explanatory varables. In testng the sgnfcance of regresson coeffcents, we use Whte s heteroskedastcty corrected estmates for standard errors (Whte, 1980). The results of the regresson model and the expected sgns for those control varables that can be justfed based on pror studes are provded n Table 3. We present three models: the frst model tests for H1 and the second and thrd model (Heckman) test for H2. (Insert Table 3 here)

24 23 In model 1, consstent wth hypothess H1, the regresson coeffcent of the Bg 4 dummy ( 1 ) s negatve (coeffcent of ) and statstcally sgnfcant (p = 0.018), suggestng that non-bg 4 audted takeover targets obtan 2.77% hgher cumulatve abnormal returns compared to Bg 4 audted takeover targets. In Model 2 of Table 3 audtor characterstcs ABIG4, MIDTIER and nteracton TBIG4xABIG4 are added. The regresson coeffcent MIDTIER s negatve ( ) and sgnfcant (p = <0.001) suggestng that ths group of md-ter audt frms receves lower CAR than non-bg4 audted frms. The regresson coeffcent ACOV has the predcted negatve coeffcent ( ) and s sgnfcant (p = 0.005) suggestng that targets have lower CAR when there s less nformaton asymmetry. The coeffcent of E/P s negatve, as predcted ( ), and sgnfcant (p = 0.002) consstent wth CAR decreasng wth ncreasng E/P or decreasng growth expectatons. Smlarly to Dong et al. (2006: 747), we fnd that target valuaton (P/B) s negatvely assocated wth cumulatve abnormal returns (CAR). The coeffcent of P/B s , and sgnfcant (p = 0.004). The coeffcent of TROE s postve (0.0046), as predcted (Ismal and Davdson 2007), and sgnfcant (p = 0.037). The regresson coeffcent of AASSETS s postve (0.0113), and s statstcally sgnfcant (p < 0.001). The coeffcent of POOL s postve (0.0428), and sgnfcant (p < 0.001) consstent wth prce advantage of poolng frms (Vncent, 1997). The coeffcent of RUMORED s negatve, as predcted ( ), and sgnfcant (p = 0.019) suggestng that pror nformaton leaks dmnsh the CAR measured n an event wndow. Followng Bhagat et al. (2005) the regresson coeffcent of stock payment (STOCK) s negatve ( ), and s sgnfcant (p = 0.001). Other varables n model 2 are not sgnfcant.

25 24 The effects of an expected audtor swtch on cumulatve abnormal returns of Bg 4 audted and non-bg 4 audted takeover targets (H2) In Table 4 we present the statstcal results regardng hypotheses H2 related to the possble alternatve explanaton that expected audtor swtch effects CAR. These effects orgnate from Model 3 (Heckman) of Table 3. Our emprcal results are n accordance wth our hypothess H2a. We can see from the Panel A of Table 4 that the largest CAR (12.68%) of all quadrants appears n the quadrant Low qualty. The condtonal CAR of the quadrant Hgh qualty s 6.08%, whch s 6.60% ponts lower (two-taled p = 0.002) than that of the Low qualty quadrant. Hence, we fnd emprcal evdence suggestng that the takeover process makes a smaller correcton to the market prce of the target wth a Bg 4 audtor even n the absence of expected audtor swtches. In the examnaton of H2b, we fnd that there s no statstcally sgnfcant dfference (twotaled p = 0.299) between the condtonal CAR of the Upgrade quadrant (CAR = 10.36%) and the Low qualty quadrant (CAR = 12.68%). Hence, we do not fnd emprcal evdence supportng the audtor upgrade hypothess. In the examnaton of H2c, we fnd that the condtonal effect n CAR of Downgrade (CAR = 3.47%) s not statstcally sgnfcantly dfferent (two-taled p = 0.119) from the condtonal effect n CAR of Hgh qualty (CAR = 6.08%). Hence, we do not fnd emprcal evdence supportng the audtor downgrade hypothess. (Insert Table 4 here)

26 25 Robustness tests We perform several robustness tests, whch are descrbed below. Sze effect. Gven that the TBIG4 varable and TSIZE are hghly correlated (untabulated), a potental concern arses that the results reported could be drven by a clent frm sze effect (cf. Lawrence et al., 2007). To address ths concern, we restrct the sample by removng frms bgger than the largest non-bg 4 clent and frms smaller than the smallest Bg 4 clent. Ths does not change the man results of the study. S&P 500 ndex n measurng cumulatve abnormal returns. To compute abnormal returns, we use the S&P 500 return ndex as an alternatve to Morgan & Stanley U.S. return ndex. The results of these alternatve analyses are nferentally smlar to the orgnal analyss. Tme perod studed. The perceptons of audtor qualty mght change over tme. Enron scandal took place n 2001 leadng to the collapse of Arthur Andersen and to the Sarbanes- Oxley Act (SOX) that mpacted the structure of the audt market and the qualty of audt servces. To study ths effect, we splt our data and run two separate regressons for sub-perods and The regresson coeffcents and p-values of man nterest are very smlar when comparng the two tme perods. Dfferent tme wndows n measurng cumulatve abnormal returns. We have used a narrow event wndow to reduce nose relatng to other market or frm nformaton than what the acquston offer. As a senstvty test, we wden the event wndow from [-1, 1] to [-5, 5]. Ths does not change the man conclusons of the study. Non-audt servce fees. Dsclosng hgh levels of non-audt fees may mpar the percepton of audt qualty and audtor ndependence (Francs/Ke, 2006). The reportng of audt fees became

27 26 oblgatory for SEC regstrants n 2001 and therefore we were only able to study the subsample for the tme perod We measure the level of non-audt fees as the rato of non-audt fees to total audt fees. When controllng for ths effect our man results reman qualtatvely unchanged. The effect of rumored deals. Rumors of mmnent mergers and acqustons are notorous for leakng early nto share prces. We analyze and fnd that the evdence s robust to droppng deals that the SDC codes as rumored. Targeted over 10% ownershp. Because an effectve control may be secured wthout acqurng over 50% of the outstandng shares, we examne whether the results are senstve to mposng less demandng crtera. A resamplng wth targeted acqustons exceedng 10% of the shares show that the man results of our study are robust to ncludng targeted 10%-50% ownershp to the sample. Successful and unsuccessful deals. It s possble that the dstncton between successful and unsuccessful deals wll affect the results. We address ths possblty by deletng unsuccessful deals from the sample. Our results are robust to ths sample selecton choce. In all of the above alternatve tests, our results regardng the varables of nterest reman qualtatvely the same.

28 27 5. CONCLUSIONS The am of ths study s to nvestgate whether the market s percepton of audtor qualty makes a dfference to the market value of a frm usng a sample of takeover offers n the USA nvolvng lsted acqurers and targets over the perod 1990 to If the market perceves that a Bg 4 audt lends hgher credblty to the fnancal reportng than a non-bg 4 audt, the frms audted by the Bg 4 should carry, ceters parbus, a lower cost of equty captal. Consstent wth ths, the pror lterature supports the vew that a prvately-held frm gong publc can reduce the level of uncertanty of future prospects, and consequently the level of underprcng of ts shares, through audtor choce (e.g. Ttman/Trueman 1986; Beatty 1989). However, the mpact of audtor choce on frm value s not lmted to the event when a frm goes publc (Zhou/Elder, 2004; Km/Park, 2006; Teoh/Wong, 1993; Haw et al., 2008). Smlar to SEOs (Zhou/Elder, 2004; Km/Park, 2006), busness takeovers provde a good settng n whch to observe the systematc dfferences n frm values subsequent to IPO snce takeovers facltate a more powerful test of the economc value of audtor reputaton than audtor swtches or earnngs numbers (as used n prevous studes). Unlke audtor swtches and earnngs that are fnancal quarter- or year-end nformaton dsclosed smultaneously wth other annual performance and corporate governance nformaton, busness takeovers can be thought as random events wth less other nformaton dsclosed around takeover announcements. The man contrbuton of our study s two-fold. The frst contrbuton of ths study s that through ts focus on the target audtor t provdes emprcal evdence that lower CARs are assocated wth takeover announcements of Bg 4 audted target frms. Ths suggests that the takeover process makes a bgger correcton to the market prce of the target when t has a non-

29 28 Bg 4 audtor because more prvate nformaton becomes avalable. That addtonal nformaton s sgnaled to the stock market va the acqurer s bd. Average bds are lower for Bg 4 audted targets and produce lower CARs for takeover announcements of such frms. To llustrate the economc sgnfcance of our man result, the dfference n the value between Bg 4 and non-bg 4 audt for an average target frm n our sample (USD 1.2 bllon) s approxmately USD 33 mllon (2.77%). Our results suggest that the effect of audtor reputaton on clent frm value s not lmted to an IPO, but can also be observed after gong publc. An alternatve explanaton for lower CARs s that antcpated audtor swtches from non-bg 4 to Bg 4 audtors are perceved as good news by the market and vce versa. A stream of research has examned market reactons to audtor swtches (e.g. Nchols/Smth, 1983; Echenseher et al., 1989; Johnson/Lys, 1990). These studes do not provde clear and consstent evdence of market reactons to audtor swtches. On balance, however, the results of these studes suggest that audtor swtches are generally vewed unfavorably by the market, but swtches to Bg 4 audtors tend to be vewed more favorably than other audtor swtches, supportng the vew that a change from non-bg 4 to Bg 4 mght be taken as a postve sgnal ( good news ) for future growth prospects. However, we do not fnd statstcally sgnfcant effects of audtor swtches and therefore no support for ther economc sgnfcance n takeovers. Even, after controllng for the expected audtor swtches, we stll fnd that the takeover announcements of Bg 4 audted targets are assocated wth lower CARs. Moreover, when we develop a 2x2 cross table analyss and calculate the condtonal effects on CAR for each of the four combnatons of acqurer and target audtor qualty, we fnd no emprcal evdence of the effects of expected swtches on CARs. It may be that markets do not react to possble future audtor swtches at the tme of take over announcement as there s stll some uncertanty

30 29 nvolved. Even f n most cases the audtor of target frm s replaced by acqurer s audtor, sometmes the ncumbent audtor s allowed to contnue (Anderson et al., 1993; Frth, 1999). These results, by showng that a dfferental market reacton to takeover announcements between Bg 4 and non-bg 4 clents s attrbutable to ncumbent audtor, not expected future audtor adds to the studes on market reactons to audtor swtches, represent a second contrbuton of the study. Even f we control for the factors affectng CARs ncludng clent characterstcs dentfed n pror studes (e.g. Lous, 2005), employ Heckman modelng to address potental self-selecton bas, and conduct seres of robustness tests reported n the paper, t s stll possble that our fndngs are drven by clent characterstcs rather than audtor qualty. Employng matchng models such as propensty-scorng, Lawrence et al do not fnd sgnfcant dfferences n audt qualty between B4 and non-b4 audtors, but that observed dfferences are manly attrbutable to dfferences n clent sze. We do not use matchng models as the relatvely small number of non-bg 4 takeover targets would reduce our sample sze sgnfcantly. Instead, as an addtonal control for the clent sze effect, we restrct the sample by removng frms bgger than the largest non-bg 4 clent and frms smaller than the smallest Bg 4 clent. Ths does not change the man results of the study. Nevertheless, we acknowledge the possblty of nsuffcent controls for clent characterstcs as a lmtaton of our study. Our fndngs should be of nterest to those nvolved n the merger and acqustons market. Ths paper focuses on US takeover bds and there may be characterstcs n other markets that are not captured by ths study. However, n today s global markets, t seems lkely that such effects would be relatvely small. The fact that the evdence s drawn from the largest captal