The Initial Going-concern of Delisting Firms: An Application of Proportional Hazard Model

Transcription

1 The Intal Gong-concern of Delstng Frms: An Applcaton of Proportonal Hazard Model Ch-Chen Wang Department of Fnancal Management, Natonal Defense Unversty Yueh-Ju Ln Department of Accountng, Kanan Unversty Yunsheng Hsu (Correspondng author) Department of Accountng, Natonal Chung-Hsng Unversty Receved: May 27, 2013 Accepted: October 12, 2013 Publshed: December 1, 2013 do: /ajfa.v URL: Abstract Ths paper examnes the survval perod and the factors of busness falure of frms who have been ssued wth an ntal gong concern opnon (IGCO) by audtors. Emprcal results show that fnancal varables are not sgnfcant predctors for future delstng crss, but the corporate governance varables are especally for frms under deteroratng fnancal condton. Important factors causng the hgher rate of delstng rsk nclude shorter lstng years, lower rate of retaned earnngs to total assets, lower rate of market value of equty to total debts, and hgher rate of pledged shares of drectors and supervsors wthn 7.5 quarters after the IGCO ssued, the number of delstng frms reaches ts peak, consstent wth the exstence of self-fulfllng prophecy. The hazard delstng functon frst rses to a peak at the 38 th quarter and then declnes rapdly, showng that after the dsclosure of IGCO, frst nne years s the delstng crss perod for Tawan publc frms. Keywords: Gong concern, Busness falure, Proportonal hazard model, Survval analyss, Self-fulfllng prophecy 139

2 1. Introducton Most fnancal statements users apply audt opnons as an operatonal health checkup pont to evaluate f the frm could operate normally or would fal n the future (Casterlla et al., 2002). Therefore, when frms are ssued wth an ntal gong-concern modfed opnon by audtors, related partes may have varous concerns about the IGCO frms. Fnancal statements users need to evaluate the length of future survval perod n the lstng market for the IGCO frms. Shareholders need to make sellng decsons at an approprate tmng to avod more losses. Credtors need to evaluate whether to demand for a debt repayment. Partcularly when a hgher credt rsk s assocated wth borrowers, banks wll consder the acceptance of future fnancng requests or smply demand for debt repayments. Government agences also need to determne approprate nterventons to mantan fnancal orders. Therefore, further studes about the survval crss for the IGCO frms are essentally useful to the related partes of delstng frms. When frms are ssued wth IGCO, the frms have a hgher probablty to occur fnancal crss. In fnancal accountng lterature, there are two ssues related to IGCO. Frst, after the dsclosure of an adverse IGCO, whether the frms would go bankruptcy or whether the audt opnon could be an early warnng predctng busness bankruptcy has become an mportant research topc. Prevous studes focus on the predcton powers of busness bankruptcy by comparng audt opnons wth statstcal forecastng models. Most emprcal evdences show that audt opnons do not exert hgher predcton powers than those statstcal models (Altman and Mcgough, 1974; Koh, 1991). These results cause doubts on audtors professonal judgment, and some studes even suggest audtors to use statstcal forecastng models to lower possble msjudgment n ther evaluaton (Altman, 1982; Koh and Kllough, 1990; Levtan and Knoblett, 1985). Prevous research smply uses a bnary choce of falure or survval n the bankruptcy forecastng models n whch only an accurate bankruptcy rate wthn the exact research perod could be learned. However, because busness falure s a lengthy and complcated process and even after ssued wth IGCO, some frms may stll survve whle others may bankrupt when tme passes. However, snce frms have dfferent survval perods and some may occur bankruptcy out of a research perod, analyss assumng a same tme perod as a comparson bass would even causes the observed data to be censored. The problem wll underestmate the probablty of hazard rate of bankruptcy and cause errors n the analyss by wrongfully assumng the busnesses are stll n normal operaton but actually ts future falure s unpredctable. Therefore, the bnary choce model does not consder the survval perod of busness falure n ts forecastng models and thus lowers ts effectveness n future preventons and treatments. Understandng the tme process about how frms go bankrupt s more useful than smply knowng ts probablty durng the research perod by the choce model. The survval analyss can further measure the survval tme n addton to the probablty of falure. The second ssue dscussed n the fnancal accountng lterature s to nvestgate the exstence of self-fulfllng prophecy by examnng f the bankruptcy probablty ncreases durng the frst or second year after an IGCO ssued. These studes, although consderng the 140

3 tme process of busness falure, merely lmt ther dscussons to the prmary effects of IGCO at an early stage. Actually after the dsclosure of IGCO, fnancal mpacts to corporate operatons do not occur mmedately but generally prolong to the future. Focusng on only one or two years after an adverse IGCO ssued and when bankruptcy probablty may not ncrease yet and mpacts of future perods are gnored, result bas may occur n emprcal nferences (Louwers et al., 1999). Therefore, after the dsclosure of IGCO frms, for any pont n tme durng the survval perod, bankruptcy probablty should be evaluated carefully. The survval analyss focuses on the tmng ssue by measurng a possble future bankruptcy rsk for the IGCO frms when fnancal crss occurs. In addton, a related ssue for fnancal statements users s about when to make ther decsons. To overcome ssues n prevous studes and solve for actual problems for fnancal statements users, the entre tme process, from the ssuance of IGCO to ther delstng, should be consdered for these frms n fnancal crss. Evaluaton for the survval perod before delstng and ts delstng probablty durng the perod s necessary for dfferent users of fnancal statements to determne ther further strateges. Ths man characterstc n ths paper s to connect audt opnons wth the dynamc process of busness falure, amng at dscussng the tme process for a survval perod untl delstng for frms ssued wth an IGCO. We dscuss three related ssues of an IGCO, whch are the hazard delstng perod for Tawan IGCO frms, the exstence of self-fulfllng prophecy, and factors nfluencng the delstng. Followng prevous fnancal crss studes, we separate delstng factors based on corporate structures, the general, fnancal, operatonal, and shareholdng structure. We construct a survval perod model for fnancal crss frms, coverng from the tme IGCO ssued tll delstng. Busness falure occurs when many ntervenng factors cause ts operaton to cease or even wthdraw from ndustry. Most related studes employ the choce model n the matchng process of a busness falure forecastng model to measure the most lkely probablty for survval or falure at a certan pont n tme. Ther models use a statc analyss whch can merely evaluate the hghest hazard rate for busness falure or delstng durng a certan perod. Unfortunately, the complcated tme process nvolved durng the survval perod s not nvestgated, for the process of ether delstng from fnancal crss or busness falure from normal operaton. Thus, possble tmng for preventon and treatment has lost for frms n fnancal crss. For nstance, when a fnancal warnng of IGCO ssued, possble delstng tme and ts probablty are stll unknown. Then, f apply the survval analyss, delstng factors and the survval perods and probablty can be estmated. Ths paper employs the proportonal hazard models to estmate the delstng factors for Tawan publc companes. Our emprcal results show that factors causng a hgher delstng hazard rate are shorter lstng tme, lower rate of retaned earnngs to total assets, lower rate of market value of equty to total debts, and hgher rate of pledge shares of drectors and supervsors. We dscover that after the sample been ssued wth an IGCO, ts delstng rsk rses contnuously and reaches ts peak at the 38 th quarter and then declnes rapdly. Thus we conclude that Tawan publc frms have a nne-year delstng hazard perod after the dsclosure of an IGCO. 141

4 In addton to the above ntroducton, the second secton s lterature revew and the thrd secton explans the desgn for econometrc models. The fourth secton descrbes the sample and data statstcs, wth the ffth secton concludng our fndngs. 2. Lterature Revew Ths secton ams at mprovng prevous fnancal crss studes whch gnore tmng factor and by employng the survval analyss model, tmng factor s ncorporated wth two research topcs nto our study. In addton, we also dscuss the applcaton of survval analyss n the emprcal studes focusng on the ndustry and fnancal structure. The frst research topc s the comparson of predcton accuracy for corporate bankruptcy between audt opnons and statstcal forecastng models. When corporate contnuous operaton s doubtful, audtors ssue ther gong-concern opnons to cauton publc nvestors but not to predct bankruptcy. However, users of fnancal statements stll consder the opnons as an early warnng for busness falure and regard a modfed opnon as a fnancal checkup pont (Casterlla et al., 2002). A study of Altman and Mcgough (1974) frst connects audt opnon of gong-concern qualty wth bankruptcy predcton. They collect a sample of 34 frms one year before bankruptcy durng the years of 1970 to 1973 and employ a Z-score statstcal forecastng model (Altman, 1968) to compare accuracy n bankruptcy predcton based on audt gong-concern opnons. Ther study fnds that the statstcal forecastng model has an accuracy rate of 82%, almost doublng the audt opnons. Thus they suggest audtors to apply statstcal models to mprove accuracy n ther opnons. Other related studes also have smlar fndngs that statstcal forecastng models are superor than those of audt opnons n bankruptcy predcton (Altman, 1982; Koh, 1991; Koh and Kllough, 1990; Levtan and Knoblett, 1985). Besdes, a report by the Cohen commsson (AICPA, 1978) also concerns the ssue and Internatonal Federaton of Accountants (1989) even requres more audtors attenton when ssung gong-concern opnons. Although these suggestons cast a reasonable doubt on audt opnons, they also provde an objectve auxlary tool for audtors to apply. Studes n bankruptcy predcton models are developed from the unvarate to multvarate models, most of whch have revewed and based ther methods from Altman (1968), Deakn (1972), and Ohlson (1980). Zavgren (1983) and Jones (1980) provde audtors wth bankruptcy models as a professonal reference. Ther models are a bnary choce model n whch only accuracy but not bankruptcy probablty s provded. Unfortunately, users of fnancal statements also concern the occurrng tme for bankruptcy n addton to ts probablty. Durng the survval perod after an IGCO ssued, users must evaluate assocate rsk and probablty of bankruptcy at varous tmes n searchng for earler strateges. However, the bnary choce model does not dscuss the survval perod and the tme process for busness falure or delstng, and thus has lowered opportuntes for early preventons and treatments. Therefore, understandng the changng process durng survval perod s meanngful n actual busness practces when frms are n fnancal and delstng crss. The second research topc s an nvestgaton about the exstence of self-fulfllng prophecy. After an IGCO s ssued, f bankruptcy probablty ncreases wthn one or two years, the 142

5 effect of self-fulfllng prophecy may exst. Louwers (1999) ndcated that the IGCO opnon accelerates busness falure and thus has mpacts on current and potental nvestors, credtors, supplers, and customers. Prevous studes test the self-fulfllng prophecy by the percentage approach, calculatng the rato of bankrupt IGCO frms to the total IGCO frms. For example, n a sample of 78 IGCO Australan frms durng the years of 1980 to 1992, Psaros and Zhang (1994) dscover that after the frst or second year of IGCO ssued, 24 frms (37.2%) declares bankruptcy. Wthn the same research perod, Barnes and Hoo (1987) fnd that only 3 (5.9%) out of a total sample of 51 IGCO frms occur bankruptcy wthn two years after IGCO ssued. Besdes, Altman (1982) states that about 25% IGCO frms are bankrupt after IGCO ssued. In concluson, above studes do not support the exstence of self-fulfllng prophecy. On the other hand, n a sample of 157 Amercan IGCO frms durng the years of 1983 to 1990, Nogler (1995) fnds a 33% sample declares bankruptcy after the frst year of IGCO ssued and supports the effect of self-fulfllng prophecy. In addton, n a sample of 210 IGCO Amercan frms durng the years of 1984 to 1991, Louwers (1999) states that 38 (18%) frms are bankrupt after the frst year of IGCO ssued, 17 (8%) frms after the second year. If applyng the DTSA (dscrete tme survval analyss) modelng, t s 27% after the frst year and 18% after the second year, whch are smlar to a 24% after the frst year n the study of Ctron and Taffer (1992) for a Brtsh IGCO sample. Both studes support the effect of self-fulfllng prophecy. Therefore, there s no concluson about the exstence of self-fulfllng prophecy for frms been ssued wth IGCO. The nconclusve fndng for self-fulfllng prophecy s due to the lack of an objectve standard n the measurements of percentage approach. To present the effect of self-fulfllng prophecy, there s a need to nvestgate objectve measurements for the effect of self-fulfllng prophecy. Zhang and Suzanne (1997) suggest applyng the length of tme for survval to measure the effect. However, the self-fulfllng prophecy seems to mply that IGCO must be ssued before bankruptcy to occur. Because t s a complcate process for a normal corporaton to declare bankruptcy, many factors could nvolve. Merely applyng audt opnons to determne whether the frm wll bankrupt after an early stage of IGCO ssued seems ncomprehensve. Besdes, an IGCO s expected to exert ts effect further to the future and other lagged effects may nvolve as well. Therefore, changes n the tme process for audt opnons to exert mpacts should be nvestgated further. Fnally, prevous emprcal studes fnd that the log-logstc model n survval analyss s the most approprate method to match frms survval perod. Through the lkelhood rato test, an approprate matchng model can be obtaned. However, f a quarter of the sample data s censored and f the selected parameter model assumes a fnal falng rate of one, the rate could easly be overestmated. In addton, the use of Webull dstrbuton assumpton could also be used to buld an early warnng model for fnancal nsttutons. Bandopadhyyaya and Jagga (2001), n a sample of 107 bankrupt frms durng the years of 1979 to 1990, dscuss the length of tme and factors causng fnal bankruptcy after reorganzaton. They employ a splt populaton duraton model to perform tests and realstcally consder the fnal bankruptcy rate not necessary equal to one. However, they do not explan the selecton process of log-logstc modfed model n ther splt populaton duraton model. In concluson, 143

6 although the survval analyss s emprcally appled to busness studes, t s rarely appled to the study of audt opnons. Snce fnancal dstress occurs at dfferent ponts n tme whch s a contnuous process, t s more reasonable to use the survval perod model to overcome the tmng ssue for our study. 3. Econometrc Models Logc or Probt method s generally used n the fnancal crss lterature whch dscusses the relaton between the probablty of fnancal dstress and ts factors. These methods mply that fnancal crss occurs at the same tme whch smply separate the sample nto occurrng crss or not. In order to dstngush dfferent occurrng tme, ths paper uses a perodcal model to further separate the sample based on dfferent delstng tme. Our models consder the survval perod so that the relaton between corporate operatng factors and delstng can be nvestgated correctly. The perodcal model, also called the transton model, focuses on the length of tme a status condton remans or the possblty of changng status at a certan pont n tme. The possblty s called the hazard rate, and the hgher the rate s the shorter tme the status remans. Ths paper dscusses the relaton between the hazard rate and corporate operatng condtons (delstng). A hgher hazard rate represents a worse operatng condton, n whch when the possblty of changng status from un-delstng to delstng s hgher, the frm s more lkely to occur delstng wthn a shorter tme perod. The perodcal model s appled n prevous bankruptcy studes for fnancal nsttutons n whch two models are generally used, the proportonal hazard model and the accelerated falure tme model (Lane, Looney and Wansley, 1986; Wheelock and Wlson, 1994). The two models have dfferent assumpton about ts baselne hazard. The baselne hazard s set respectvely as the followng: the proportonal hazard model: h ( t x) = h ( t)exp( x ), 0 β the accelerated falure model: h ( t x) = h ( t, xα)exp( x ), 0 β n whch h ( t x) s the hazard rate, h ( ) s the baselne hazard, x s the varous predctors, 0 α and β s the coeffcent for ts baselne and exponental functon. From the above settngs, the baselne hazard s assumed not to be affected by the predctors n the proportonal hazard model, but otherwse n the accelerated falure model. Both models can apply parameter estmaton methods, but Cox (1972) n hs partal-lkelhood approach for the estmaton of proportonal hazard model, assumes no baselne hazard for any pattern of dstrbuton. Both models have dfferent thresholds, characterstcs, and applcatons. The followng explans the use of proportonal hazard model n ths paper. For the random varablet, representng days after IGCO, follows the probablty dstrbuton of f(t x) wth an cumulatve densty functon of F(t x). Wthn t days after the IGCO, the delstng probablty could be wrtten as: 144

7 t prob( T t x) = F( t x) = f ( u x) du (1) The delstng probablty after t days n the survval functon of S ( t x) could be wrtten as: 0 S( t x) = 1 F ( t x). (2) Thus, the hazard rate h( t x) for delstng occurrng at t days could be wrtten as the followng: prob( t T t + δ T t, x) h ( t x) = lm = f ( t T t, x) = δ δ 0 x f ( t x). (3) S( t ) From equaton (1), (2) and (3), we can obtan the followng three equatons: d log S( t x) h( t x) = dt (4) f ( t x) = S( t x) h( t x) (5) t S( t x) = exp( h( u x) du). (6) To estmate the relaton between varables and the survval perod, any functon set for f ( t x), F ( t x), h ( t x), or S ( t x) must be known. Ths paper follows Cox s (1972) partal log-lkelhood method to estmate the proportonal hazards model. The hazard rate s set as the followng: 0 h ) xβ ( t x) = h0 ( t e (7) of whch x descrbes the structures for lstng frms and β s the coeffcent. Dfferent characterstcs of x may cause delstng effects dfferently, so the sgn for each correspondng coeffcent could be postve or negatve. The baselne hazard h ( ) s the unobserved varable effects on the hazard rate. From equaton (7), we can obtan the parameter functon for h ( t x) but not h ( ) 0 t. Therefore, the proportonal hazard model s called the sem-parametrc estmaton method. The estmaton for β s based on the partal-lkelhood approach. Ths approach frst arranges the sample n the order of the delstng and assumes n number of lstng frms n the sample. Thus, each t n a lstng frm 0 t 145

8 satsfes t t. When at t 1, the condtonal delstng probablty for the frst frm s 1 2 t3 as the followng: At L 1 h( t1 x1β ) e = = n n h( t x t= 1 1 1β ) t= t j, the condtonal delstng probablty for the j th frm s as the followng: x1β 1 e xβ. (8) L j h( t j x j β ) e = = n n h( t = xβ ) 1 = x jβ 1 e xβ. (9) From equaton (8) and (9), we can wrte the partal lkelhood functon, PL as n the followng: PL = n L = n = 1 = 1 [ n = 1 exp( xβ ) ] y exp( x β ) j j δ. (10) In equaton (10), f t j t, then let y j = 1 and f t j t, then let y j = 0. If the data s censored, then δ = 0, and f not censored, then δ = 1. Therefore, a set of coeffcents β can be obtaned n whch the partal lkelhood functon reaches ts maxmum value. The β set s our estmated targets. From the above, the proportonal hazard model merely assumes the relaton between the hazard rate and the baselne hazard. In equaton (7), the hazard rate and the followed dstrbuton n the survval perod are not set. The man dfference between the two models (the proportonal hazards and the accelerated falure tme) s n the assumpton that the survval perod follows a certan dstrbuton. The model s set as the followng: T = exp( xβ + σε ), (11) of whch T follows a certan dstrbuton of f(t),σ s the scale parameter, and ε s the standard error term. We can wrte the lkelhood functon L as the followng: L = n n δ 1 δ [ f ( t )] [1 F ( t )] = t= 1 = 1 [ f ( t )] δ [ S ( t )] 1 δ. (12) When L s maxmzed, β can be estmated. Note that the obtaned sgns n both models must be opposte, because when the survval perod s shorter, the hazard rate wll be hgher. In the accelerated falure model, the shape for hazard rate s dfferent because t follows varous dstrbuton of T. In equaton (11), non-lnear relaton between T and varables can be found. Dstrbuton n the standard error term ε s not the same as T s, but a one-on-one 146

9 swtchng relaton. Table one s dstrbutons for the survval perod, the standard error term, and characterstcs n ts correspondng hazard rates. For these dstrbutons, hazard rates have four dfferent shapes, ncreasng, decreasng, no changng, or ncreasng frst and then declnng later. Therefore, when set the dstrbuton for survval perod, a lfe table must be used to descrbe the hazard rate frst. Then based on the shape and characterstcs n the hazard rate, a reasonable dstrbuton for the survval perod can be set, or there could be a specfcaton error presented n the model. Table 1. Dstrbutons for T,ε, and hazard rate characterstcs Dstrbuton T Exponental Gamma Log-logstc Log-normal Webull Dstrbuton ε Extreme value (1 parameter) Log-gamma logstc Normal Extreme value (2 parameter) Characterstcs n h (t) h(t) s a fxed constant h(t) ncreasng or decreasng h(t) frst ncreasng/decreasng, then the opposte h(t) frst ncreasng, then decreasng h(t) frst decreasng, then ncreasng Source:Allson (1995) Ths paper studes a total sample of 153 delstng frms after been ssued wth an ntal gong-concern opnon (IGCO) by audtors. Research perod covers from January 1, 1980 to June 30, 2006, a total of 16 years (64 quarters). Table 2 s the lfe table for the sample, arranged by the order of occurrng perod of delstng (n quarter tme) based on a non-parametrc analyss. As shown n Table 2, the IGCO sample have a hgh number of delstng frms durng the quarters of 0 to 7.5 (31 frms), 7.5 to 15.0 (10 frms), 15.0 to 22.5 (29 frms), and 22.5 to 30.0 (13 frms) respectvely. On the other hand, the rest of survval perods have only delstng frms of sx, one, or zero. Partcularly, after IGCO ssued untl the 7.5 th quarter, most IGCO frms occur delstng, showng that the effect of self-fulfllng prophecy exsts n Tawan captal market for our research perod. Durng the same perod (0 to 7.5 th quarter), the delstng probablty contnuously rses from to the hghest rate, wth a correspondng survval perod from 30.0 to 37.5 quarters. Then, the hazard rate declnes rapdly, meanng the average survval perod for the IGCO frms to occur delstng s about 38 quarters (9 years). Wthn ths perod, the delstng probablty s the hghest, and then the hazard rate declnes quckly untl the quarter of 45 to 52.5, the rate rses to agan. Graph 1 s a bar chart for the hazard rate and survval perod based on table 2. The dstrbuton pattern for hazard rate s dfferent from those of correspondng hazard rates n 147

10 table 1. Therefore, we decde not to apply the accelerated falure tme model and employ the proportonal hazard model nstead by applyng the Cox s partal log-lkelhood approach. Table 2. Lfe table 1 Survval perod (quarters) Sample sze Censored frms Possble Actual delstng delstng frms frms Survval rate Hazard rate (0.000) (0.006) (0.035) (0.005) (0.040) (0.014) (0.049) (0.024) (0.044) (0.038) (0.033) (0.000) (0.033) (0.038) (0.031) (0.000) (0.031) (0.000) (0.031) (0.000) Source: calculaton from the sample Remark 1: ths table s based on Cutler and Ederer method. 2:standard devaton s n brackets for the survval rate. 3:standard devaton s n brackets for the hazard rate Mean RATE TIME Graph 1. Hazard rate 148

11 4. Data and sample statstcs Based on audt opnons on fnancal statements, excludng the unqualfed opnon, ths paper nvestgates audtors ntal gong-concern opnon (IGCO) on lstng companes. We employ mportant predctors for delstng from related lterature to construct a survval analyss model. Corporate structures constructed n the predctors are fnancal nformaton, stock prces, the pledged shares rato of drectors and supervsors, and the nsder shareholdng rato. Sample data s derved from the quarterly modules of cumulatve fnancal nformaton, full delvery share, or delstng company n the Tawan Economc Journal (TEJ) databank. The research perod ncludes 64 quarters (16 years) from January 1, 1980 to June 30, 2006 and a total sample of 153 IGCO frms. Because a long complcated process s nvolved for a delstng crss to occur, our study consders the survval perod from the ssuance of IGCO to the tme of delstng (the falure event) or to research perod stop date (rght censored data). We take two ponts n tme when the IGCO s ssued and when the delstng occurs to calculate the actual survval perod n quarters. Durng the research perod for IGCO frms, a total fnal sample of 153 frms s obtaned for modelng. To qualfy as a delstng frm n ths paper, the frm must be delsted by Tawan Securtes and Futures Bureau but not by voluntary delstng. Durng the research perod (1980/1/1~2006/6/30), f a survval perod could be measured, a completed data for an IGCO delstng sample frm s obtaned. However, f the delstng tme s after the research stop date, the IGCO sample frm s consdered un-delstng (rght censored data) and has an ncomplete survval data. For ths ncomplete survval data (rght censored), ts survval perod s calculated by deductng the research stop date from the IGCO date. These ncomplete survval data can only provde partal nformaton due to the lmtaton on the research perod stop date. Table 3 lsts the selected sample numbers, n whch panel A s the sample selecton process for fnancal dstress frms. The frst sample selecton standard s to nclude only those IGCO frms who have complete fnancal statements and related nformaton. Thus, four bankng and securtes IGCO frms wth sgnfcant dfferent accountng systems and 25 other IGCO frms wth ncomplete nformaton are excluded. Fnally, 153 IGCO frms are derved as our research sample at the research stop date of June 30, Panel B n table 3 lsts annual sample numbers of IGCO frms, delstng and un-delstng. Untl the research stop date (2006/6/30), 68 IGCO frms are delsted and 85 IGCO frms are un-delsted. Based on the sample data, snce 1998, numbers of IGCO frms ncrease, and delstng occurs mostly after The man reason may due to the Asan fnancal crss n 1997 whch overturns economc condton, leadng operatng dffcultes and depressng stock market contnuously. Durng the perod, many adverse cases such as napproprate corporate nvestments or msappropratng corporate captal by large shareholders to mantan stock prces have caused corporate bankruptcy. 149

12 Table 3. Sample Selecton Panel A: Sample selecton process # of IGCO frms durng Less: Bankng and securtes companes 4 Less: Incomplete data nformaton 25 Total number of fnal IGCO sample frms 153 Panel B: Annual sample selected Year # of IGCO frms # of delstng frms # of un-delstng Total

13 Table 4 s the sample statstcs for the 153 IGCO sample frms by ndustry and lstng market. It s generally beleved publc frms lstng n the OTC market have smaller captal and unhealther fnancal structure than those lstng n the man stock market. Ths paper test the effect of type of lstng market on delstng and fnd t s nsgnfcant at a 0.05 confdence level wth a correlaton ph coeffcent of wth p-value of Thus, emprcally there s no sgnfcant correlaton between delstng and lstng market type n Tawan for our IGCO sample frms. There s no dfference between lstng market type for delstng IGCO frms. Table 4. Sample statstcs for the IGCO frms Delsted IGCO frms Un-delsted IGCO frms Man OTC Man OTC Total market market market market Total # % # % # % # % # % # % Cement Food Plastc Textle Machnery Applance Bo-chem Glass Paper Steel Rubber Auto Electroncs Constructng Tradng Toursm Communcaton Software Others Manageral total Prevous studes suggest audtors to apply corporate bankruptcy models as a tool to evaluate audt opnons. Due to ts objectve measurement and convenence n obtanng data, most 151

14 statstcal models employ fnancal ratos as man predctors. Cybnsk and Wnnsor (2005) also advse audtors to select approprate nformaton to reflect clent operatng condtons such as the fnancal ratos. Followng prevous studes (Altman, 1982, 1986; Chen and Lee, 1993; Ohlson, 1980, 1993), we classfed the factors for fnancal dstress nto four corporate structures, the general, fnancal, operatonal, and ownershp structure. The followng explans these four corporate structures n detal. Corporate general structure ncludng frm sze (SIZE) and frm age (AGE). Chen and Lee (1980) and Ohlson (1993) stated that the larger the frm sze s, the less lkely for fnancal crss to occur. Chen and Lee (1993) showed that the hgher the frm age, the less lkely for delstng to occur. Corporate fnancal structure could be evaluated by corporate short- and long-term solvency, and the worse the ablty, the more lkely fnancal crss could occur. Ths paper employs four fnancal ratos to measure corporate solvency. Current rato (CACL) s Current assets dvded by current labltes s current rato. Ths rato measure corporate lqudty, and the hgher t s, the better short-term solvency s. Ohlson (1980) emprcally proves that current rato measures lqudty whch s a sgnfcant ndcator for corporate bankruptcy. Workng captal to total assets rato (WCTA) s the dfference between total short-term assets and labltes. Altman (1968) states that reducng of the frm s net current assets and lqudty wll dampen ts short-term solvency and ncrease the lkelhood of delstng. Cash flows from operaton to total labltes (FUTL) s cash flows from operaton dvded by total labltes. Beaver (1966) beleves that ths rato reflects the amount of total labltes ts cash flows could bear. Total labltes to total assets (TLTA) s total labltes dvded by total assets. Beaver (1966) and Ohlson (1980) reveal that ths debt rato has a sgnfcant predcton power for busness fnancal crss. Fnancal crss generally occurs due to a declnng proftablty from neffectve use of corporate captal. Ths paper studes the effectveness n applyng corporate captal and beleves that the survval perod for a fnancal crss s related to ts operatonal structure. Corporate operatonal structure explans the effectveness n the use of captal. Returns on total assets (NITA) measure corporate proftablty meanng the return that each dollar asset creates. Studes reveal that ths rate s sgnfcantly assocated wth corporate fnancal crss and the hgher the rate, the better the proftablty (Altman, 1968; Beaver, 1996; Ohlson, 1980). Altman (1968) uses retaned earnngs to total assets rato (RETA) to measure the degree of earnngs accumulated when tme passes. Younger or proftless frms accumulate lower earnngs whch can lower ths rato and rase the delstng probablty. Altman (1968) employs market value of equty to total labltes (MVETL) to evaluate how the value of corporate assets declnes when total debts exceed assets. If ths rato s low, a serous asset value declnng problem and a solvency crss may exst, whch may thus ncrease the delstng probablty. Corporate governance studes nvestgate nternal control mechansm. It s generally beleved frms can effectvely supervse and restrant managers behavors and lower ther prvleged consumpton at corporate costs, resultng ncreases n frm value (Fama and Jensen, 1983; 152

15 Jensen and Mecklng, 1976). Ths lne of lterature focuses on the mpacts of board characterstcs and ownershp structure on frm values. We use the pledged share rato of drectors and supervsors and nsder shareholdng rato to proxy corporate ownershp structure n our paper. Pledged share rato of drectors and supervsors (RATIO) s measured by the total pledged shares of drectors and supervsors to ther total shareholdngs. When drectors and supervsors pledge ther shares, t seems that they wthdraw ther nvestments but reman ownershp control. If the rato s hgh, dstorton n ownershp structure become serous whch may rase ts delstng probablty. Insder shareholdng rato (OWN) s measured by the total shareholdngs of nsders (managers and drectors) to the total outstandng shares. Benesh (1997) ndcate that n larger companes, managers have lower shareholdng ratos and thus less ncentve to maxmze frm values. Laporta et al. (1999) dscover that when ownershp s rather concentrated, frm value wll be relatvely hgher. Ths paper uses shareholdngs of managers, drectors, and supervsors to represent the nsder shareholdng rato. Table 5. Predct sgns for predctors of delstng probablty Predct Explanatory varables* Sgn Defnton Frm sze(size) - Log (total assets) Frm age n years (AGE) - Years after establshment Current rato (CACL) Current assets/current debts Workng captal to total assets (WCTA) - (current assets current debts)/total assets Total labltes to assets (TLTA) + Total debts/total assets Market value of equty to total labltes (MVETL) - (market values of common and preferred shares)/total debts Cash flows to total labltes - Operatng cash flows/total debts (FUTL) Returns on assets (NITA) - Income before extraordnary tems/average total assets Retaned earnngs to total assets - Retaned earnngs/total assets (RETA) Pledged shares % of drectors and supervsors (RATIO) + Pledged shares of drectors and supervsors /ther total shares Insder shareholdng rato (OWN) - Shares of managers, drectors, and supervsors/total outstandng shares *above predctors use quarterly data n the nformaton of fnancal, stock prces, pledged share ratos of drectors and supervsors, and nsder share holdng rato. 153

16 Based on the descrptve statstcs n table 6, we fnd that the average of the IGCO observatons s 15 quarters. Because certan frms have censored data, the average survval perod s thus wthout actual meanngs. The average age s 13 quarters, but the standard devaton s qute large at , meanng that our sample frms are not concentrated at a certan year range, whch s measured from the establshment to the begnnng of our research perod. The current rato has an average value of 68%, revealng a depressed short-term solvency for the IGCO sample. Table 6. Descrptve statstcs and Mean dfferences Varable Mean Standard devaton Mean dfferences * TIME AGE SIZE CACL WCTA b TLTA b FUTL % RETA % MVETL a RATIO % a OWN % a s sgnfcant at a 0.01 level, and b s sgnfcant a 0.05 level. *s the dfference between the delstng frms and those remanng operaton at the end of Emprcal results Informaton s costly and some are even lmted or not obtanable when nvestors need useful nformaton to make nvestment decsons. Ths paper apples both fnancal and no fnancal nformaton to nvestgate delstng factors for IGCO frms. Our study examnes four models of corporate structure combnatons under nsuffcent or suffcent nformaton assumpton. For nsuffcent nformaton assumpton, three structure models, the general fnancal, general operaton, and general ownershp are tested respectvely. The suffcent nformaton model has all corporate fnancal and non-fnancal structures nto the model, ncludng the general, fnancal, operatonal, and ownershp structure. We dscuss the relaton between dfferent corporate structures and ts delstng and search for sgnfcant mpact predctors for corporate delstng. Table 7 ncludes our emprcal results. Accordng to model 1 n table 7, nsgnfcant fnancal predctors for delstng are current rato (CACL), workng captal to total assets rato 154

17 (WCTA), debt rato (TLTA), and market value of equty to total debt rato (MVETL). The result shows that ratos n fnancal structure are neffcent predctors for delstng. Model 2 ncludes three ratos of returns on assets (NITA), retaned earnngs to total assets (RETA), and market value of equty to total debts (MVETL) n corporate operatonal structure to ts mpacts on delstng. Among these ratos, the retaned earnngs to total assets (RETA) rato has a sgnfcant and negatve mpact on delstng. When n fnancal dstress, corporate performance and proft declne whch reduce accumulated earnngs as a result. Ths shows that napproprate applcaton of corporate resources may cause ts assets to functon ncompletely. In addton, the market value of equty to total debts rato (MVETL) s also sgnfcant and negatve. After the ssuance of IGCO, corporate stock prces fall, market values shrnk, and debt payng ablty even lowers, resulted n a hgher delstng probablty. Model 3 has two ratos n the corporate ownershp structure to predct delstng, ncludng the pledged share rate of drectors and supervsors and nsder shareholdng rato. Among these two ratos, the nsder shareholdng rato has a sgnfcant and postve effect on delstng. Large ownershp shareholders pledge ther shares to obtan captal and make napproprate nvestment decsons, causng corporate fnancal crss as a result. Ths stuaton s normally observed n Tawan publc companes and eventually they are dessted from the market. Model 4 ncorporates all four corporate structures nto the model when nformaton s suffcent to exam ts mpacts on delstng. The emprcal results show two sgnfcant and negatve mpacts on delstng, the retaned earnngs to total assets rato (RETA) and the market value of equty to total debts rato (MVETL). The pledged share rato of drectors and supervsors has a sgnfcant and negatve mpact, meanng that applyng operatonal and non-fnancal ownershp structure together nto the model s useful n predctng corporate delstng. 155

18 Table 7. Emprcal results based on the proportonal hazard model Asan Journal of Fnance & Accountng Mode 1 Model 2 Model 3 Model 4 CACL (0.276) (0.289) WTCA 0.216(0.241) (0.270) TLTA 0.209(0.351) 0.156(0.451) FUTL 1.895(3.812) 4.535(3.948) NITA (48.36) (56.557) RETA (0.404) * (0.416) * MVETL (0.295) *** (0.416) *** RATIO 0.978(0.364) *** 0.857(0.407) ** OWN (3.120) (3.337) AGE (0.014) *** (0.014) *** (0.014) *** (0.014) *** SIZE (0.099) (0.098) (0.097) (0.104) Log-lkelhood Restrct Log-L slope=0 Standard devaton n the brackets, *, **, and ***represent sgnfcant at the level of 0.05, 0.025, and Model 1 ncludes corporate general and fnancal structure. Model 2 ncludes corporate general and operatonal structure. Model 3 ncludes corporate general and ownershp structure. Model 4 ncludes corporate general, fnancal, operatonal, & ownershp structure. 6. Concluson After the ssuance of an ntal gong-concern opnon (IGCO), users of fnancal statements concern about whether the frm wll be delsted from the market, or t wll depart from fnancal crss and contnue ts operaton. Prevous studes suggest employng the bnary choce model as a tool for audtors to evaluate the ssue of gong-concern for the frm. The choce model s used to predct future busness falure possblty, and early relevant mpacts of IGCO. On the other hand, mportant ssues that captal markets concern about s the occurrng tme for delstng, the maxmum delstng probablty, or the length of survval perod after the opnon by IGCO frms. Furthermore, because the bnary choce model can only smply to predct the maxmum delstng probablty, the tmng process for delstng cannot be fully understood, whch may lower the preventon and treatment functon n crss forecastng models. We employ the proportonal hazard model on the sample of Tawan IGCO delstng frms to dscuss the delstng dynamc process after been ssued wth an IGCO. Frst, we obtan a lfe table from the research perod data by a non-parametrc analyss and learn that after

19 quarters from IGCO ssued, the sample has a hghest delstng probablty durng our research perod. After ssued wth an IGCO, the delstng hazard perod s about nne years, wth a hgh delstng probablty of 10%. Ths paper apples four corporate structures to measure mpact factors on the delstng functon and fnd that except for the ratos of equty to total debts and retaned earnngs to total assets, explanatory varables n corporate fnancal structure are all nsgnfcant factors. However, the pledged share rato of drectors and supervsors n ownershp structure, representng corporate governance has a sgnfcant and postve mpact on delstng. Generally, objectve fnancal nformaton can be used to measure operatonal performance. Audtors ssue the gong-concern modfed opnon based on ther professonal judgment about clents fnancal statements and relevant nformaton. When such a modfed IGCO s ssued, unhealthy fnancal structure of the frm has already exsted for a long tme. Then t would be msled by these fnancals to evaluate ts future operatonal performance. By assstng wth non-fnancal nformaton, decsons about a delstng crss for IGCO frms can be evaluated. Among our results, corporate governance structure s proved to be an mportant ndex. However, our results reflect that by merely applyng fnancal nformaton to evaluate factors for delstng n Tawan IGCO sample s nsuffcent. More mportantly, the result shows that corporate governance mechansm s related to the ssue of gong-concern operaton. References Altman, E. I., & T. P. McGough. (1974). Evaluaton of a company as a gong concern. Journal of Accountancy, Altman, E. I. (1982). Accountng mplcatons of falure predcton models. Journal of Accountng, Audtng and Fnance, Bandopadhyaya, A., & S. Jagga. (2001). An analyss of second tme around bankruptces usng a splt-populaton duraton model. Journal of Emprcal Fnance, Beaver, W. (1996). Fnancal ratos as predctors of falure. Journal of Accountng Research, Benesh, M. D. (1997). Detectng GAAP volaton: mplcatons for assessng earnngs management among frms wth extreme fnancal performance. Journal of Accountng and Publc Polcy, Barnes, P., & Hoo, D. (1987). The stranger case of the qualfed success. Accountancy, Casterella, J. R., B. L. Lews, & P. L. Walker. (2001). The relaton of audt opnon and audtor change wth bankruptcy emergence. AAA audt secton mdyear meetng. Ctron, D. B., & R. J. Taffler. (1992). The audt report under gong concern uncertantes: An emprcal analyss. Accountng and Busness Research, 22(4),

20 Chen, Ko, & J. Lee. (1993). Fnancal ratos and corporate endurance: A case of the ol and gas ndustry. Contemporary Accountng Research, 9(2), Deakn, E. B. (1972). A dscrmnant analyss of predctors of busness falure. Journal of Accountng Research, Jensen, M., & W. Mecklng. (1976). Theory of the frm: Manageral behavor agency costs and ownershp structure. Journal of Fnancal Economcs, Koh, H. C., & L. N. Kllough. (1990). The use of multple dscrmnant analyss n the assessment of the gong-concern status of an audt clent. Journal of Busness Fnance and Accountng, Koh, H. C. (1991). Model predctons and audtor assessments of gong-concern status. Accountng and Busness Research, 2L(4), Levtan, A., & Knoblett, J. (1985). Indcator of exceptons to the gong concern assumpton. Audtng: a Journal of Practce & Theory, 5, Louwers, T, Messna, F., & Rchard, M. (1999). The audtors gong concern dsclosure as a self-fulfllng prophecy: A dscrete tme survval analyss. Decson Scences, 30(3), Laporta, R., F. Lopez-de-Slanes, & A. Shlefer. (1999). Corporate ownershp around the world. Journal of Fnance, 54, Lancaster, T. (1990). The Econometrc Analyss of Transton Data. Brown Unversty. PMCd:PMC53443 Zhang, M. W., & H. Suzanne. (1997). Gong, gong gone? Is a GCQ a self fulfllng Prophecy. Australan Accountant, Nogler, G. E. (1995). The resoluton of audtor gong - concern opnons. Audtng: A Journal of Practces & Theory, 14(2), Ohlson, J. (1980). Fnancal ratos and the probablstc predcton of bankruptcy. Journal of Accountng Research, 18(1), Psaros, J. and M. Zhang. (1994). The gong concern audt opnon: Australan evdence. Perspectve on Contemporary Audtng, Cybnsk, P., & W. Carolyn. (2005). The effcacy of audtor s gong concern opnons compared wth a temporal and an temporal bankruptcy rsk model: analyzng U.S trade and servce ndustry falures Pacfc Accountng Revew, Schmdt, P., & A. D. Wtte. (1989). Predctng crmnal recdvsm usng splt populaton survval tme model. Journal of Econometrcs, 40,