Emerging Markets Review

Transcription

1 Emergng Markets Revew 10 (2009) 1 22 Contents lsts avalable at ScenceDrect Emergng Markets Revew journal homepage: Frm dversfcaton and equlbrum rsk poolng: The Korean fnancal crss as a natural experment Robert Masson a, Heather Tookes b,, Taejong Um c a Cornell Unversty, Unted States b Yale School of Management, Unted States c Samsung Investments, Republc of Korea artcle nfo abstract Artcle hstory: Receved 25 August 2008 Receved n revsed form 4 November 2008 Accepted 6 November 2008 Avalable onlne 13 November 2008 JEL classfcaton: G31 G32 G33 Keywords: Asan Fnancal Crss Korean Chaebols Frm dversfcaton Bankruptcy and credt provson We use the Korean Fnancal Crss as a natural laboratory for examnng nteractons among frm dversfcaton, equlbrum captal structure and tal probablty events. When the crss ht n 1997, several major frms, ncludng a large number of hghly leveraged conglomerates (Chaebols), experenced bankruptces. We show how dversfed Chaebols obtan hgher equlbrum leverage than non-chaebols (a cosgner effect ). In the event of a low probablty macro-economc shock, the model predcts a systematc change n relatve bankruptcy rsks of Chaebol frms. To examne ths mplcaton, we ntroduce an emprcal methodology that decomposes equlbrum debt nto demand, supply and Chaebol-specfc factors, for use n a bankruptcy predcton model. We fnd that the prmary cause of Chaebol frm bankruptces was not dosyncratc leverage, but leverage systematcally related to greater equlbrum access to debt durng normal tmes Elsever B.V. All rghts reserved. 1. Introducton (A)n unprecedented number of hghly leveraged conglomerates (Chaebols) have moved nto bankruptcy. IMF Press Release, December 4, 1997 Correspondng author. Yale School of Management, P.O. Box , New Haven, Connectcut , Unted States. Tel.: ; fax: E-mal address: heather.tookes@yale.edu (H. Tookes) /$ see front matter 2008 Elsever B.V. All rghts reserved. do: /j.ememar

2 2 R. Masson et al. / Emergng Markets Revew 10 (2009) 1 22 When the Korean Fnancal Crss ht n late 1997, a number of major frms experenced bankruptces. At the rsk of over-smplfcaton, ths wave of corporate bankruptces can be traced to the peculartes of corporate fnancal structures n Korea, namely the domnance of Chaebol member frms n the economy and the hgh dependence on short-term debt. 1 In ths paper, we analyze the equlbrum determnaton of debt for a sample of Korean manufacturng frms for the years and the subsequent wave of bankruptces followng the 1997 crss. Our prmary goal s to characterze the role of frm dversfcaton n the determnaton of captal structure and crss-perod bankruptcy. An mportant aspect of the wave of Korean bankruptces was that Chaebol frms were dsproportonately affected. Ths led to a percepton that they were responsble for the severty of the crss. Chaebols are large, famly-controlled busness groups that consst of dverse, fnancally afflated frms. We present a very smple model based on Lewellen (1971) to show how the crss-perod Chaebol bankruptces may reflect ratonal market equlbrum outcomes negotated n normal tmes. The basc dea of Lewellen's consurance hypothess s that when combnng operatons wth mperfectly correlated cashflows, t s possble to reduce frm rsk and ncrease debt capacty. We extend the Lewellen (1971) ntuton to consder the mpact of a systemc crss on relatve bankruptcy rsks of dversfed versus undversfed frms. When a systemc shock (or crss) rases the bankruptcy probablty for every frm n an economy, the rsk of dversfed frms rses relatve to non-dversfed frms. The factors whch make dversfed frms better credt rsks n normal tmes can make them worse credt rsks durng crses. We are the frst (to our knowledge) to document a mechansm through whch systemc crses may dsproportonately mpact dversfed frms. In partcular, we use the Korean Fnancal Crss as a natural experment for examnng the mpact of dversfcaton on ex ante credt provson and ex post bankruptcy outcomes when negatve tal probablty events occur. The Korean case provdes a partcularly useful settng for analyzng the mpact of dversfcaton on captal structure for several reasons. Frst, there are a large number of dversfed conglomerates (Chaebols) n the economy. Second, both leverage and bankruptcy events are drectly observable at the ndvdual Chaebol member frm level. In fact, full fnancal and accountng statements are avalable at ths level. Ths s n contrast to U.S. dversfed frms whch are not requred to report detaled segment nformaton. In the U.S., by the Fnancal Accountng Standards Board's (FASB) Statements of Fnancal Accountng Standards (SFAS) 131 and SFAS 14, segment-level reportng conssts prmarly of sales and proft nformaton, makng t extremely dffcult to capture dvsonal captal structure. Fnally, the severty of the Korean crss provdes dentfcaton of a large, unantcpated negatve tal probablty event. We ntroduce a new bankruptcy predcton methodology and use the Korean fnancal crss as a laboratory for emprcal testng. Snce leverage ratos are mportant predctors of bankruptcy rsk, we begn wth a frst-stage regresson to estmate pre-crss equlbrum debt ratos. Ths allows us to dentfy the debt supply, debt demand and Chaebol-specfc factors n equlbrum debt rato determnaton. Next, we use the parameter estmates from the frst stage regresson to decompose pre-crss debt ratos accordng to Chaebol and non-chaebol supply and demand factors. For ntuton behnd the decomposton, we can thnk of a Chaebol frm's observed debt rato as the sum of ts predcted debt rato were t a non-chaebol frm, plus an ncremental debt rato due to Chaebol afflaton (plus an dosyncratc resdual). We then use the decomposed ratos to estmate the ncremental crss perod bankruptcy probablty due to each of these components. A novel aspect of our decomposton s that we use common debt rato unts. Ths makes all estmated coeffcents drectly comparable and allows for an assessment of relatve mportance of the sources of the observed debt levels. To the best of our knowledge, ths methodology s unque and we beleve that t provdes mportant nsghts whch would be dffcult to obtan from tradtonal bankruptcy predcton models. Our prmary fndngs are two-fold: Frst, Chaebol frms have greater equlbrum access to debt due to dversfcaton. We nterpret Chaebol membershp as one mportant measure of dversfcaton. We also explot cross-sectonal varaton n the dversfcaton across groups by ncludng the number of member frms n the debt determnaton regressons. Ths measure s also postvely and sgnfcantly related to debt 1 Hstorcally, due to government polcy and mperfect captal markets, Korean frms depended on short term bank debt from government-controlled banks. Korea lberalzed bankng n the 1990s and n the md-1990s lberalzed short-term foregn borrowng.

3 R. Masson et al. / Emergng Markets Revew 10 (2009) provson. Second, although pre-crss debt level s the most mportant determnant of bankruptcy, an mportant cause s leverage that s systematcally related to Chaebol frms' greater equlbrum access to debt durng normal tmes. In addton to mplcatons for frm dversfcaton and captal structure, the fndngs n ths paper also have relevance for post-crss perod regulaton. The Korea Far Trade Commsson has amended the Monopoly Regulaton and Far Trade Act to, among other thngs, elmnate cross-subsdary loan guarantees on new debt. Ths polcy s consstent wth the general vew n Johnson et al. (2000, p. 22), n whch the authors lst troubled frms n a group propped up usng loan guarantees by other lsted group members as an example of lootng by controllng shareholders durng the crses of the In ths paper, loan guarantees and crss perod dstress of Chaebol member frms are equlbrum outcomes that can exst even wthout such lootng. Moreover, durng perods n whch the probablty of systemc crss s low, an mplcaton of the analyss n ths paper s that prohbtng such loan guarantees may be neffcent. Ths paper organzed as follows. The next secton provdes a bref overvew of related lterature and a descrpton of the Korean fnancal crss. Secton 3 presents a smple theoretcal model of equlbrum debt determnaton and derves an emprcal specfcaton. Secton 4 descrbes the sample of Korean manufacturng frms. Results of estmaton are presented n Secton 5. Secton 6 concludes. 2. Background 2.1. Related lterature Ths paper adds to the lterature n three mportant ways. Frst, we study changes n the relatve bankruptcy rsk of dversfed (Chaebol member frms) versus stand-alone (non-chaebol) frms when negatve tal probablty events occur. Ths aspect of the value mplcatons of frm dversfcaton has not been prevously addressed n depth. A prmary observaton motvatng our analyss s Chaebol frms' relatvely hgh pre-crss leverage. 2 We argue that ths was attanable due to the protectve cosgner effect of afflated frms, as shown n Lewellen (1971) and, more recently, n Leland (2007). 3 The emprcal evdence we present s consstent wth that vew. In the context of mergers of U.S. frms, there s also some emprcal evdence n support of Lewellen's co-nsurance hypothess. For example, Km and McConnell (1977) report that leverage ncreases followng dversfyng mergers. However the general emprcal queston of whether dversfed frms borrow more s stll somewhat open. Sten (2003) reports that the emprcal evdence s weak on whether U.S. dversfed frms borrow more than ther stand-alone peers (e.g., Berger and Ofek (1995) report that U.S. dversfed frms borrow approxmately 1% more than focused frms). The econometrc approach that we take n ths paper allows us to shed further lght on ths ssue snce t provdes clearer dentfcaton of the porton of the debt that comes from frm dversfcaton. In a closely related paper, Dmtrov and Tce (2006) use the busness cycle to test whether dfferences n credt access cause dfferences between performance n U.S. dversfed versus stand-alone frms. They fnd that focused frms perform worse (.e., decreases n sales and nventory growth) than dversfed frms durng recessons. Ther use of fluctuatons n the macro-economy s smlar to the approach that we take n ths paper; however, our use of a large (tal probablty) shock mproves our understandng of the full economc mpact of dversfed frms' leverage. Note that the co-nsurance effect s one of many potentally value-relevant aspects of frm dversfcaton. 4 The more general lterature focuses on the dversfcaton dscount, the observaton that the stocks of dversfed frms trade at a dscount relatve to the sum of the mputed values of ther 2 Ferrs et al. (2003) report that Chaebols have greater debt capacty (ndustry-adjusted debt-to-asset ratos are 10% hgher for Chaebols). 3 Leland (2007) shows that fnancal synerges can actually be negatve f rsks facng operatons are dfferent. Ths s because credt spreads and nterest tax deductons may be hgher when fnancal contracts reman separate. A man mplcaton s that fnancal synerges should be explctly consdered n merger (and dvestture) analyses. An mportant assumpton underlyng both Lewellen (1971) and Leland (2007) s frms' greater debt capactes and leverage due to dversfcaton. 4 Potental costs of dversfcaton nclude opportuntes for cross-subsdzng unproftable dvsons, opportuntes for neffcent nvestment and dffcultes algnng ncentves of dvsonal managers. Benefts nclude productve synerges, economes of scale, effcent allocaton of nternal resources, ncreased debt capacty due to decreased cashflow varablty and the assocated tax savngs assocated wth debt.

4 4 R. Masson et al. / Emergng Markets Revew 10 (2009) 1 22 ndvdual segments. 5 For example, Lang and Stulz (1994) document a negatve relatonshp between Tobn's Q and dversfcaton. Smlarly, Berger and Ofek (1995) report a value loss for U.S. frms related to dversfcaton of 13 to 15% durng 1986 to1991. Importantly they also fnd that dversfed frms borrow more than sngle segment frms, whch partally offsets reductons n value, due to the debt tax sheld. The second area of the lterature to whch we contrbute s that on optmal captal structure. In Modglan and Mller (1958), captal structure s rrelevant; however, n the presence of frctons, debt levels may matter. These frctons nclude: 1) the balancng of the tax beneft of debt aganst bankruptcy and other leverage-related costs (e.g., Mller, 1977; 2) mtgatng conflcts of nterest between nsde managers (e.g., Jensen and Mecklng, 1976; Jensen, 1986) and stockholders and debt holders (e.g., Stulz, 1990; and 3) conveyng nsde nformaton to captal markets or reducng adverse selecton costs (e.g., Myers and Majluf, 1984). 6 Tax-based trade-off theores of captal structure have the closest relatonshp wth our work; however we account for all three of these n our emprcal estmaton of equlbrum debt levels. If Chaebol frms have hgher debt capactes (.e. lower bankruptcy probabltes, ceters parbus) due to the protectve effects of cosgners they wll be offered, and wll choose, hgher levels of borrowng due to the preferental tax treatment of debt. Fnally, the thrd contrbuton s the ntroducton of a new debt decomposton methodology applcable to bankruptcy predcton models. Begnnng wth Altman (1968), there has been a large body of emprcal research usng fnancal and accountng ratos to predct bankruptcy. By decomposng equlbrum debt ratos, we can estmate bankruptcy probabltes, condtonal upon demand and supply factors of Chaebol and non-chaebol frms (rather than the standard accountng ratos). Our methodology allows for economc nsghts that would be dffcult to obtan from the standard approach. In the next secton, we provde a bref descrpton of the Korean fnancal crss and mplcatons for Chaebol frms The Korean Fnancal Crss The Korean fnancal crss can be explaned largely by the chan of events that preceded t. Korean equty markets were not well-developed untl the md-1980s. Bank loans (manly short-term) have hstorcally been the prmary source of corporate nvestment fnancng. In 1996, followng sgnfcant lberalzaton of the fnancal sector (to secure OECD membershp), the Korean government took steps to lberalze short-term foregn fnancal borrowng. 7 The large gap n nterest rates (approxmately 12% domestcally, compared to 6.5% n the foregn market durng ) provded strong market ncentves for short-term foregn loans. Merchant banks took advantage of the lower rates avalable outsde of Korea. The Korean banks' relance on short term foregn debt, combned wth the short term nature of corporate leverage, left the economy vulnerable. When Korea subsequently experenced a severe macro fluctuaton, the result was devaluaton of the Won and a fnancal crss (begnnng n late 1997), n whch both frms and banks had trouble repayng ther loan oblgatons. Wth the devaluaton of the Won, Korean banks became unable to pay ther foregn denomnated loans and they called n ther short term Wondenomnated loans wthout extendng new credt. Ths left many frms wthout the lqudty necessary to operate at full capacty, even f they were otherwse solvent. The large number of bankruptces of Chaebol frms led to a percepton that Chaebols were a major part of the problem n the Korean crss. Ths paper examnes ths contenton and shows how the Chaebol bankruptces mght occur n an equlbrum framework. To understand the arguments, t s useful to dscuss what Chaebol frms are. A Chaebol s a group of fnancally afflated frms, typcally operatng n dfferent 5 The exstence of the dscount n U.S. dversfed frms has been the subject of sgnfcant recent debate. Mans and Reeb (2002) fnd that the observed dscount reflects a shft n value from equty holders to bondholders, due to decreased cashflow varablty. They hold that pror observatons of the dscount are a result of pror researchers' use of book, rather than market value of debt. Chevaler (2004) and Whted (2001) argue that measurement error explans some of the dscount. Vllalonga (2004) reports evdence of a dversfcaton premum. Campa and Keda (2002) account for possble endogenety n the relatonshp between dversfcaton and value. 6 For a comprehensve survey of the theoretcal captal structure lterature, see Harrs and Ravv (1991). Note that because many new share ssues n Korea are rghts offerngs (.e., offerngs of addtonal shares to exstng shareholders), the peckng order hypothess of Myers and Majluf (1984) mght be expected to be less applcable. 7 Long-term foregn loans remaned restrcted.

5 R. Masson et al. / Emergng Markets Revew 10 (2009) lnes of busness. Famly control s common, as s a group headquarters where certan nvestment decsons are made. For example, consder Samsung whch, pror to the crss, had more than forty separate frms bearng ts name and fnancal backng. One mght own stock n Samsung Electroncs (e.g., TVs and computers) and not own stock n the entrely separate Samsung frm, SDI, whch manufactures pcture tubes and montors (and mostly sells them to non-samsung frms). In normal tmes, f one of Samsung's forty frms were to experence fnancal dffculty, the remanng frms would act as cosgners on the loans of the dvson n fnancal trouble. The cosgner effect makes a Chaebol frm less rsky, ceters parbus (see Secton 3). Hence, n fnancal equlbrum, the leverage suppled to a Chaebol frm should be greater than ts non-chaebol equvalent. 8 Followng a low probablty economy-wde shock, cosgners would also have short term loans beng called and may have nsuffcent funds to pay the loans of the nsolvent sub-parts of a Chaebol. 9 In ths case, the leverage from normal tmes becomes a lablty n crses. 3. Equlbrum debt determnaton Our man objectves are to dentfy the factors that determne equlbrum debt ratos (ex ante), and then to model the relatonshp between these factors and the realzaton of bankruptcy. We focus on the queston of whether there was a systematc change n relatve bankruptcy rsks of conglomerates followng the Korean Fnancal Crss of late We start wth a heurstc model of a compettve credt market equlbrum durng a non-crss perod and use the ntuton n Lewellen (1971) to llustrate how dversfcaton can ncrease debt capacty. We then show how ths hgher equlbrum access to debt durng normal perods can ncrease the relatve bankruptcy rsks of dverse frms when negatve tal probablty events (.e., crses) occur. Ths framework gudes the decomposton methodology ntroduced for the emprcal mplementaton Hypothess development and emprcal methodology Heurstc model of debt determnaton Assume a sngle bank loan to an ndvdual frm n an amount equal to L. Let the probablty of bankruptcy equal β, an ncreasng functon of debt rato (D). 10 The payoff to lenders n the event of bankruptcy s equal to αl where α<1. If a frm does not go bankrupt, lenders receve (1+ρ)L, where ρ s the loan contract's equlbrum nterest rate spread, an ncreasng functon of debt rato (D) and bankruptcy probablty (β(d)). 11 The expected proft of a lender s: E½πŠ= βαl + ð1 βþ½ð1+ρþlš L: ð1þ Assumng a compettve equlbrum, the zero (expected) proft condton s: βα + ð1 βþð1+ρþ=1: ð2þ Solvng for ρ and wrtng ρ and β as functons of the debt to asset rato (D), the model mples a postve nterest rate spread: ρðdþ= 1 βðdþα 1 > 0: 1 βðdþ ð3þ 8 Ths adds value due to the debt tax sheld. 9 We are not the frst to use the Asan fnancal crss to examne the role of debt n ex post outcomes. Whle much dfferent n focus from our study, Allayanns et al. (2003) examne crss-perod changes n market values as a functon of foregn versus local currency debt. Baek et al. (2004) examne the mpact of foregn ownershp on crss-perod value changes. They fnd that frms wth more foregn nvestors had lower decreases n share value whle Chaebols experenced larger value decreases. 10 L denotes total debt amount; D denotes total debt relatve to total assets of the frm. 11 Each ndvdual loan has ts own supply and demand, so the endogenously determned ρ pad by frm I wll generally not equal ρ pad by frm j.

6 6 R. Masson et al. / Emergng Markets Revew 10 (2009) 1 22 Table 1 Model and methodology: notaton Parameter/functon L β α ρ D Π Z X Defnton Loan amount to an ndvdual frm Probablty of bankruptcy Fracton of loan recovered n bankruptcy Loan contract's equlbrum nterest rate Debt rato Lender's proft Parameters affectng the supply of debt Parameters affectng the demand of debt An upward slopng supply curve (assumed) mples 12 : ρvðd Þ= β VðD Þð1 αþ > 0: ð4þ 2 ð1 βðdþþ Dfferentate the equlbrum ρ n Eq. (3) wth respect to β to descrbe the relatonshp between the probablty of bankruptcy and the nterest rate spread: Aρ Aβ = ð1 α ð1 βþ Þ > 0: ð5þ 2 As one would expect, Eq. (5) mples that lenders charge a hgher nterest rate as the probablty of bankruptcy ncreases (along wth the hgher debt rato). If we consder a zero proft condton and mpose hghly stylzed condtons n whch α=0 and ρ s very small, then β s approxmately equal to ρ. Ths mples that a hgher exogenous probablty of bankruptcy translates nto a hgher nterest rate spread. Defne the supply and demand for debt as: ρ s = sz; ð DÞ and ρ d = dx; ð DÞ: ð6þ (where D s debt rato and Z and X are parameters affectng the supply and demand for debt, respectvely). The model's notaton s summarzed n Table 1. Ths basc framework allows us to ntroduce a new debt decomposton methodology, as well as ntuton regardng the mpact of an unantcpated shock on the bankruptcy probabltes of dversfed frms Emprcal mplementaton (debt decomposton methodology) Our objectve s to establsh the predctons of ths model and to derve a testable emprcal specfcaton. Assume that for the th elements of Z and X, represented at by z and x, respectvely, ρ s / z <0 and ρ d / x >0. The frst nequalty mples that greater z shfts the supply curve to the rght, ndcatng lower ρ (lower rsk of bankruptcy) and a greater debt rato for any gven demand curve, whle the second nequalty mples that greater x shfts demand to the rght, leadng to hgher ρ (hgher rsk of bankruptcy) and a greater debt rato, D, for any gven supply curve. For a tractable functonal form, let: ρ s = a 0 + a 1 Z + a 2 D and ρ d = b 0 + b 1 X + b 2 D: ð7þ Based on our assumptons on supply and demand: a 1 <0; a 2 >0: b 1 >0; b 2 <0. Equatng ρ s wth ρ d,we can wrte the equlbrum debt rato and nterest rate spread as: D = γ 0 + γ 1 X + γ 2 Z; where γ 1 = b 1 a 2 b 2 > 0 and γ 2 = a 1 a 2 b 2 < 0: ð8þ 12 Also, as D ncreases, loans become rsker on the margn. Contractng frctons (e.g., nformaton problems) can generate upward slopng supply.

7 R. Masson et al. / Emergng Markets Revew 10 (2009) ρ = η 0 + η 1 X + η 2 Z; where η 1 = b 1a 2 a 2 b 2 > 0 and η 2 = b 2a 1 a 2 b 2 < 0: ð9þ To mplement ths n a regresson format, suppose there s data on D for each frm and we consder the regresson: D = γ 0 + γ 1 X + γ 2 Z + e : ð10þ To ntroduce our debt decomposton methodology, we partton the observed D to examne the separate effects of demand and supply factors on the probablty of bankruptcy. Let: h D = ˆD X; Z + ˆD X ; Z h ˆD X; Z + ˆD ðx ; Z Þ ˆD X ; Z + e : ð11þ D s the frm's observed debt rato whch s equal to the parttoned predcted debt ratos plus the resdual; X s a vector of demand factors and Z s a vector of supply factors; X and Z are mean levels of demand and supply factors, respectvely. The frst term n Eq. (11) s the equlbrum debt rato mpled for a frm wth mean values for X and Z. The second term reflects the equlbrum devaton n the debt rato of frm that s due ts dosyncratc demand factors not equalng sample mean demand factors, X. Note that ths term stll assumes sample mean supply factors, Z. The thrd term then adds the change n equlbrum debt rato due to Frm 's dosyncratc supply factors. Fnally there s an error term, whch reflects Frm 's dosyncratc devatons from the predcted equlbrum debt rato for the frm (.e., the dosyncratc devaton from Frm 's equlbrum debt-to-asset rato gven the observed demand and supply factors, X and Z ). If we knew the true probablty of bankruptcy, β, we could regress ths on the elements of D (predcted and resdual values of the debt rato) from the debt rato regresson (Eq. (11)): h β = C 0 ˆD X; Z + C 1 ˆD X; Z h ˆD X; Z + C 2 ˆDðX; ZÞ ˆD X ; Z + C 3 e: ð12þ Importantly, the constant s scaled to D (X, Z ) to facltate nterpretaton n common debt rato unts. Eq. (12) s novel (to our knowledge) and s the bass of our emprcal model, although the partton s more complex n our actual mplementaton. As one mght expect, Eqs. (7) (9) mply a postve relatonshp between X (loan demand) and observed D and a postve relatonshp between Z and both ρ and β (approxmately equal under our assumptons), at least durng normal tmes. The model also mples a postve relatonshp between Z (loan supply) and observed D and a negatve relatonshp between Z and both ρ and β, durng normal tmes Applcaton to the Korean Fnancal Crss [Lewellen (1971) ntuton] Gven the basc equlbrum models presented above, how mght we extend the analyss to provde a better representaton of the Korean case? More specfcally, how can ths framework explan the hgh levels of debt of Chaebol frms as well as the ncdence of post-crss bankruptcy for Chaebols? Consder the followng smple example of Chaebol bankruptcy: Suppose that the world s composed of many frms that are dentcal n all respects other than Chaebol afflaton. Assume that all frms carry dentcal debt levels and that the probablty of llqudty for each frm s drawn from an..d. process and s denoted φ. All frms are ether Chaebol members or ndependent frms. We defne a Chaebol to be smply a loan cosgnng agreement. 13 For smplcty, we consder a Chaebol consstng of two member frms. In the orgnal model, the rsk that a frm became llqud at a gven level of debt was dentcal to the bankruptcy probablty β(d). Clearly, the bankruptcy probablty of a Chaebol frm s no longer equal to the probablty of ndvdual frm llqudty snce there are some states of the world n whch the full loan and nterest are repad, even though the ndvdual borrower s llqud. From the perspectve of a lender, the probablty of llqudty for each ndvdual frm s φ. However, a Chaebol frm secures fnancng based on the fact that t has an afflated frm whch also has a probablty of falure equal to φ and that these falure probabltes are ndependent. 14 Suppose φ=0.01. Then the relatve bankruptcy rsk of a non-chaebol to a Chaebol frm s φ/φ 2 =100. Now suppose that there s an economy- 13 In prncple, non-chaebol frms could also enter nto cosgnng arrangements; however, t s lkely that the tes between Chaebol member frms are stronger than they are between unafflated frms, makng contractng easer for Chaebol members. 14 We assume ndependence for purposes of llustraton. The ntuton holds as long the repayment abltes (.e., cashflows) of the borrower and the cosgner are not perfectly correlated. Note also that Chaebols are typcally groups of unrelated busnesses.

8 8 R. Masson et al. / Emergng Markets Revew 10 (2009) 1 22 wde shock that shfts the probablty of bankruptcy by.09 for all frms (note the crss bankruptcy rsk n our sample s more than 15%). 15 Now φ=0.10 and the relatve bankruptcy rsk of the non-chaebol to Chaebol frm s φ/φ 2 =10. Thus, not only have bankruptcy rsks for ndvdual frms ncreased, but also the relatve rsk for the Chaebols relatve to the non-chaebol has ncreased ten-fold. The protectve effect of Chaebol membershp, whch would have been a consderaton n the determnaton of pre-crss debt levels, may become a lablty when there s a low probablty fnancal crss. 16 In ths example, we have assumed dentcal debt levels for Chaebol and non-chaebol frms n order to llustrate the effect of Chaebol membershp on bankruptcy rsk. In equlbrum (all else equal), the reducton of bankruptcy rsk due to Chaebol membershp wll allow Chaebol frms to secure hgher amounts of debt. Wth β close to ρ, banks would be wllng to lend larger amounts to Chaebol frms. If they loaned funds to Chaebol frms untl the bankruptcy rsks, β, were smlar to those for non-chaebol frms, the relatve precrss rsk at φ=0.01 (probablty of ndvdual frm non-chaebol llqudty when debt levels were fxed at a constant D) wll rse. If debt ncreases value (.e., va a debt tax sheld) then, n equlbrum, Chaebol frms would amass greater debt due to ths ncreased supply. 17 As the Chaebol frm's rsk approaches that of an ndependent frm, ts rsk n crss tmes clmbs. One should then expect to see far more Chaebol falures n the event of a crss when the captal markets are compettve, effcent and have ratonal expectatons (assumng that the perceved probablty of a crss s low). Ths smple model and dscusson suggest that a lower exogenous rsk of Chaebol frms (φ 2 versus φ n the example above) shfts the loan supply curve to the rght, so that ther nterest rates are lower and equlbrum debt ratos are hgher. Ths smple model therefore predcts a negatve relatonshp between extra debt due to Chaebol membershp and the probablty of bankruptcy. However, n the event of an unlkely economy-wde shock n whch β ncreases for all frms, the rsk poolng mechansm that leads to a lower bankruptcy probablty and greater leverage of Chaebol frms n normal tmes s precsely what causes them to be rsker n crses Data Our analyss s based on data from a sample of 385 Korean manufacturng frms lsted on the Korea Stock Exchange (KSE) n and Accountng data are from the Korea Investment Servce and Chaebol membershp data are from the Korea Far Trade Commsson. The Korea Investment Servce data ncludes both the fnancal statements and the ncome statements of 625 manufacturng frms lsted on the KSE n We requre a complete record of the varables for each year ncluded n our analyss, leadng to a balanced panel of 385 frms (82 Chaebols and 303 non-chaebols). Bankruptcy data for lsted manufacturng frms are from the stock prce secton n the Dong-A Ilbo. 19 The bankruptces, nfluenced by the Korean fnancal crss, began to occur n late For our defnton of a bankruptcy, we are nterested n dentfyng a farly broad measure of a frm's fnancal falure (e.g., frms that are techncally nsolvent as well as those that are forced to lqudate). We therefore classfy as bankrupt, all frms on the Korea Stock Exchange that fell nto techncal nsolvency and were downgraded to frm under prce survellance by the KSE durng the perod October 1, 1997 through July 31, Table 2 provdes summary statstcs of our sample of Korean manufacturng frms. The most pronounced characterstc of captal structure of Korean manufacturng frms les n the hgh level of corporate borrowng (captured by the debt to total assets varable, DASSET). The sample mean DASSET s 65.9%, of 15 From the dscusson of Table 2 below, 17.4% of frms n the sample are classfed as bankrupt durng the crss perod. 16 Wth more frms n a Chaebol, ths effect becomes even stronger. 17 If banks recognze the remote possblty of crss then the full rsk for a Chaebol frm, ncludng the crss possbltes, wll be somewhat hgher than ts normal tmes rsk, whch n turn should be lower than that for a non Chaebol. 18 An alternatve nvolves credt-ratonng wth a usury constrant. In practce, Korean frms often accessed the curb market. Gven ths fact and gven that the demand and supply factors behave as we suggest n our heurstc theory, we stay wth ths model. 19 One of the largest daly natonal newspapers. Publshed n Seoul, South Korea. 20 Insolvent frms fall nto the category frm under prce survellance under KSE regulatons. Frms n ths category are permtted to reman lsted on the KSE for a lmted tme perod. Ths bankruptcy defnton would nclude frms that are under prce survellance and then eventually delsted. It would not nclude frms n some fnancal dstress but stll able to meet ts debt oblgatons. Presumably, n ex ante debt provson, credtors care most about whether ther clams wll be mpared (.e., techncal nsolvency), and less about the fnances of the frm after oblgatons are met.

9 R. Masson et al. / Emergng Markets Revew 10 (2009) Table 2 Descrptve statstcs All frms Chaebol frms Non-Chaebol frms Mean Std. Dev. Mean Std. Dev. Mean Std. Dev. DASSET LDASSET ROA TANSET DEPSET LGSALE WC CFDEBT ADSALE GROWTH EXPORT BUSRISK CFIRMS CSALES CPROFIT N Ths table presents summary statstcs for the sample of 385 Korean Manufacturng Frms, Varables are defned as follows: DASSET s the rato of total debt to total book value of assets (multpled by 100). LDASSET s the rato of long term debt (maturng n more than 1 year) to assets ( 100). ROA s return on assets. Rato of earnngs before nterest less tax to total assets ( 100). TANSET s the rato of tangble assets to total assets ( 100). DEPSET s the rato of deprecaton to total assets ( 100). LGSALE s log of sales. WC the rato of workng captal (current assets mnus current debt) to total assets. CFDEBT s the rato of cashflow to total debt ( 100). ADSALE s the rato of advertsng expendture to total sales ( 100). GROWTH s annual growth rate of total assets ( 100). EXPORT s the rato of exports to sales ( 100). CFIRMS The number of member frms n the Chaebol group (natural log). CSALES Log of Chaebol group sales, excludng member frm sales. BUSRISK s tme trend controlled busness rsk (hgh BUSRISK means low busness rsk). CPROFIT Chaebol group proft, defned as total Chaebol proft dvded by Chaebol group total. whch most s short term debt. The mean long term debt to asset rato s 25.7%. Also mportant s the fact that the frms were hghly dependent upon foregn debt (both drectly and ndrectly though natonal banks' debt postons). Total foregn debt n Korea amounted to bllon U.S. dollars n 1996, accountng for 22.6% of gross natonal product, and short-term foregn debt maturng wthn 1 year made up 58.8% of total foregn debt as of November 1997, when the Korean government offcally requested rescue loans from the IMF. We nvestgate the queston of whether the Korean fnancal crss of late 1997 can be attrbuted to ths hgh dependence on debt and whether any further nsght can be ganed by a systematc decomposton of debt. From Table 2, the data also suggest that Chaebol frms had hgher levels of borrowng than non-chaebol frms. The average debt-to-total assets ratos of.74 and.64, respectvely are consstent wth the observaton n Ferrs et al. (2003) that Chaebol frms pay less n taxes than non-chaebol frms due to the tax advantage of leverage. Ths dfference would be expected, gven the dscusson presented n Secton 3. We also note that Chaebol frms have a hgher mean value of collateral value of assets (TANSET) but a lower mean value for the advertsng propensty (ADSALE) compared to non-chaebol frms. The apparently large dfference between Chaebol and non-chaebol frms n TANSET and ADSALE mght be an ndustry effect snce Chaebol frms are more lkely to be n the heavy ndustres and chemcals, whch have hgher tangble assets and less dfferentated products. We use ndustry fxed effects n our emprcal analyss n order to control for ths type of varaton. Fnally, note that we observe lower busness rsks (.e., BUSRISK) n Chaebol member frms The BUSRISK varable s constructed from OLS regresson: ROA T =at+e T. ROA s Return on Assets. BUSRISK s the coeffcent on ndependent varable T (tme) mnus standard error. A Hgher value of the BUSRISK varable means lower busness rsk.

10 10 R. Masson et al. / Emergng Markets Revew 10 (2009) 1 22 Table 3 Bankruptces Bankrupt frms Non-bankrupt frms Chaebol frms 16 (19.5%) 66 (80.5%) Non-Chaebol frms 51 (16.8%) 303 (83.2%) All frms 82 (17.4%) 303 (82.6%) Ths table reports crss perod (October 1, 1997 to July 31, 1998) bankruptcy events for the sample of 385 Korean manufacturng frms lsted on the Korean Stock Exchange. Table 3 dvdes our sample of 385 Korean manufacturng frms accordng to fnancal falure and Chaebol membershp. 22 The table shows that 17.4% of the frms n our sample faled durng the sample perod. The rate of falure for Chaebol frms s 19.5%, not statstcally dfferent from the 16.8% rate for non-chaebol frms. However, Chaebol and non-chaebol frms dffered substantally n ther economc sgnfcance. The average sze of non-chaebol frms n our sample s U.S. $91 mllon n annual sales. Chaebol frms are 5 tmes larger, wth average annual sales at U.S. $451 mllon Emprcal specfcaton and results of estmaton 5.1. Prelmnary model/estmaton Our man goal s to analyze the relatonshp between bankruptcy and the contrbutons of demand, supply and Chaebol related factors to equlbrum debt ratos. As a prelmnary step, we examne whether debt levels durng the 1991 to 1994 pre-crss perod are, n fact, good predctors of bankruptcy. In order to ensure that ex ante levels of debt were obtaned durng a tme n whch crss probablty was low (.e., the crss was not yet antcpated, consstent wth the framework presented n Secton 1), we end the analyss of pre-crss equlbrum debt determnaton n 1994, 3 years pror to the crss. For our sample of frms, average pre-crss debt rato (debt/book-value-of-assets) was 82.1% for Chaebol frms that went bankrupt, compared to 72.7% for Chaebols that dd not. Smlarly, for non-chaebol frms, the debt ratos were 76.9% for frms whch became bankrupt and 61.5% for those whch dd not. Our model wll depend entrely on debt ratos, the factors whch lead to them and ther nfluence on bankruptcy. The justfcaton for focusng on the debt to asset rato can be seen n Table 4a, n whch we use a prelmnary logt model to evaluate the mportance of a set of explanatory varables n the determnaton of bankruptcy. The response varable n our model equals 1 f a frm became llqud or bankrupt between October 1, 1997 and July 31, 1998 and zero f t dd not. Ths model employs four ndependent varables, chosen based on pror bankruptcy studes. 24 The explanatory varables are: sze of frm (log of sales, LGSALE); debt to total assets rato (DASSET); lqud assets (the rato of workng captal to total assets WC); and debt coverage (the rato of cashflow to total debt CFDEBT). These varables capture the mportance of the frm n the economy, long-term solvency, short-term ablty to pay down debt, and the fnancal performance of the frm relatve to ts debt level, respectvely. We estmate the model separately for each year, (3 to 6 years pror to bankruptcy). The results of estmaton are presented n Table 4a. The debt rato (DASSET) s the only sgnfcant predctor of bankruptcy for all of the years of estmaton (the estmated coeffcent s postve, as one would expect). Gven the heavy relance on short term debt pror to the crss, debt maturty mght play a role n bankruptcy outcomes. To examne ths possblty, we rerun the model presented n Table 4a, but decompose debt ratos nto short- and long-term components. The results are presented n Table 4b. As 22 The Korean Far Trade Commsson lsts the 30 largest groups of two or more frms n total assets for each year, whch we use to defne frms as Chaebols. There s lttle turnover n these top 30, but we nclude any frm whch made the top 30 lst. 23 Chaebols are defned by sales of two or more afflated frms. In prncple, a large Chaebol could be made up of several very small frms, but on average the Chaebol frms n our sample are large. There s a substantal sample overlap between the frm szes, 10.7% of our Chaebol frm annual sales observatons are smaller than the mean for the non-chaebols. 24 We also replcated the basc logt usng the varables n Altman (1968). Altman's fve varables are ratos: workng captal to total assets; retaned earnngs to total assets; earnngs before nterest and taxes to total assets; sales to total assets; market value of equty to book value of debt. We found that only the retaned earnngs to total assets rato was protectve aganst bankruptcy.

11 R. Masson et al. / Emergng Markets Revew 10 (2009) Table 4a Logt regresson results: crss perod bankruptces Years pror to crss Varable name Intercept LGSALE DASSET WC CFDEBT Dependent varable: bankrupt=(0,1) 6 (1991) Coeffcent a a Std. Error Ch-square (1992) Coeffcent b a Std. Error Ch-square (1993) Coeffcent a Std. Error Ch-square (1994) Coeffcent a b b Std. Error Ch-square Ths table presents results of a logt model where the dependent varable s an ndcator equal to 1 f the frm experences a bankruptcy event durng the perod October 1, 1997 to July 31, The sample conssts of 385 Korean manufacturng frms. The explanatory varables are defned as follows: LGSALE: the log of sales. DASSET: rato of total debt to total book value of assets. WC: rato of workng captal (current assets-current debt) to total assets. CFDEBT: total debt coverage, defned as the rato of cashflow to total debt. a Sgnfcant at the 1% level; b sgnfcant at the 5% level. shown n the table, both components have smlar mpacts on ex post bankruptcy. In fact, long and short term debt are substtutes (they are negatvely and sgnfcantly correlated, wth a correlaton coeffcent of.17). The data suggest that t s total debt, regardless of type, that matters. Ths would be expected f, for example, frms were refnancng some of ther long term debt wth new short term debt. Gven the results n Tables 4a and 4b, we focus on total debt ratos n all subsequent analyss. Havng establshed that pre-crss debt levels are an mportant ndcator of bankruptcy, we proceed to the man focus our analyss, n whch we generate estmates of predcted debt and then decompose these Table 4b Logt regresson results: crss perod bankruptces wth long/short debt decomposton Years pror to crss Varable name Intercept LGSALE LDASSET SDASSET WC CFDEBT Dependent varable: bankrupt=(0,1) 6 (1991) Coeffcent a a a Std. Error Ch-square (1992) Coeffcent b a a Std. Error Ch-square (1993) Coeffcent a a Std. Error Ch-square (1994) Coeffcent a b b Std. Error Ch-square Ths table presents results of a logt model where the dependent varable s an ndcator equal to 1 f the frm experences a bankruptcy event durng the perod October 1, 1997 to July 31, The sample conssts of 385 Korean manufacturng frms. The explanatory varables are defned as follows: LGSALE: the log of sales. LDASSET: rato of long term debt to total book value of assets. SDASSET: rato of short-term debt to total book value of assets. WC: rato of workng captal (current assets-current debt) to total assets. CFDEBT: total debt coverage, defned as the rato of cashflow to total debt. a Sgnfcant at the 1% level; b sgnfcant at the 5% level.

12 12 R. Masson et al. / Emergng Markets Revew 10 (2009) 1 22 debt ratos based on Chaebol and non-chaebol demand and supply factors. We then estmate a logt model of bankruptcy usng the predcted/resdual value approach descrbed n Secton 3. Ths enables dentfcaton of any systematc effects of Chaebol membershp, debt demand and supply on the probablty of bankruptcy Equlbrum debt ratos In ths step, we generate predcted values for pre-crss equlbrum debt levels usng two-stage least squares. The dependent varable n our model s pre-crss debt-to-assets and the ndependent varables are the supply and demand varables for the years 1991 to 1994, defned n Table 5 and dscussed n more detal below. Varables n Table 5 that affect the demand sde of the debt rato are proftablty (ROA), growth rate of assets (GROWTH), deprecaton (DEPSET), and the collateral value of assets (TANSET). ROA s a potentally mportant demand varable; however the predcted sgn s ambguous accordng to agency explanatons for captal structure and the peckng order hypothess. Jensen (1986) presents the free cashflow hypothess that debt can mtgate the ablty for managers to consume perqustes when cashflows are hgh. Under ths theory, we expect a postve relatonshp between ROA and debt demanded. If, on the other hand, nternal funds are cheaper than external funds (.e., the peckng order hypothess n Myers and Majluf (1984)) ths would mply a negatve relatonshp between ROA and debt demanded. We use DEPSET (deprecaton/total assets) to examne the tax tradeoff theory. To the extent that alternatve tax shelds exst (.e., deprecaton), we expect to observe less debt demanded when deprecaton s greater. The GROWTH varable allows us to estmate the role of under-nvestment ( debt overhang ) problems. Frms wth hgh growth opportuntes wll demand less debt n order to avod value loss due to under-nvestment when debt levels are too hgh (.e., when postve net present value projects from the perspectve of the frm are unattractve to equty holders because the benefts of these projects are enjoyed by debtholders). Because the measured growth rate of total assets mght be correlated wth the resdual n the debt determnaton equaton, we use an nstrument to measure growth. Fnally, TANSET (tangble assets/total assets) allows us to examne the role of rsk-shftng. When frms have hgher tangble assets, costs due to potental rskshftng when frms are near dstress are reduced. Therefore these frms wll, all else equal, demand hgher levels of debt (see e.g., Rajan and Zngales (1995) for nternatonal evdence of the postve relatonshp between tangble assets and debt levels). Table 5 Varable defntons for decomposton analyss (varables are for frm n year t) Dependent varable DASSET Explanatory varables Demand factors ROA DEPSET GROWTH TANSET Common supply factors LGSALE ADSALE EXPORT BUSRISK Chaebol-specfc supply factors CHAEBOL CHLGSALE CFIRMS CPROFIT CSALES Industry Debt level, defned as total debt dvded by book value of assets. Return on assets: proft (earnngs before nterest, less taxes) dvded by book value of assets. Deprecaton dvded by book value of assets. Annual growth rate of assets. Tangble assets dvded by book value of assets. Frm sze varable. (Log) Total sales Total advertsng expendture dvded by total sales. Total exports dvded by total sales. Busness rsk. Varable constructed from OLS regresson: ROA T = at+e T. BUSRISK s the coeffcent on ndependent varable T (tme) mnus standard error. A hgher value of the BUSRISK varable means lower busness rsk. Dummy varable equal to 1 f frm s a Chaebol frm, 0 otherwse. Chaebol frm sze, defned as the log of total sales Chaebol dummy (CHAEBOL). Number of member frms n the Chaebol group (natural log). Chaebol group proft, defned as total Chaebol proft dvded by Chaebol group total assets (n year t). Log of Chaebol group sales, excludng member frm. 21 ndustry dummy varables.

13 R. Masson et al. / Emergng Markets Revew 10 (2009) Supply varables are assumed to be drven by asymmetrc nformaton, the ablty of the frm to repay (busness rsk) and the consurance effect. We expect to observe greater debt ratos when asymmetrc nformaton s low. 25 That s, when frm sze (LGSALE) and advertsng propensty (ADSALE) are large. We also expect greater debt when busness rsk s low (.e., BUSRISK value s hgh; and frm sze, LGSALE s hgh) and when the consurance effect s strong. 26 To the extent that advertsng propensty reflects the degree of product dfferentaton, the earnngs volatlty of hgh advertsng frms would also be lower due to hgher barrers to entry and moblty, whch wthstand the shocks comng from new competton. A lower probablty of bankruptcy stemmng from the reduced earnngs volatlty for hgh advertsng frms would ncrease the supply of debt. We expect supply of debt to be negatvely related to export proporton (EXPORT). To the extent that the export proporton reflects the rsk of exchange rate fluctuatons, the debt capacty of hgh export frms would tend to be lower and ths would have the effect of decreasng the supply of debt. Busness rsk s a measure of earnngs volatlty over tme and affects the supply of debt. Because of the potental endogenety of both growth (GROWTH) and busness rsk (BUSRISK), these varables are nstrumented n a frst stage regresson wth ndustry dummes as the nstruments. 27 The frst stage regressons are hghly sgnfcant (at the.01% level), wth F-values of 9.89 for the GROWTH regresson and for the BUSRISK regresson. We are able to capture 17 and 45% of the varaton n these varables, respectvely. As we descrbed n Secton 3, Chaebol membershp reduces the bankruptcy rsk (and ncreases debt supply) of a Chaebol member frm through rsk poolng among Chaebol members. Ths s because, whle ndvdual member frms tend to operate n focused ndustres, Chaebol groups are dverse. Therefore we nclude a Chaebol membershp dummy varable (CHAEBOL) as a supply factor capturng dversfcaton. In order to shed further lght on the underlyng mechansm drvng ths ncreased debt of dversfed frms, we also nclude a Chaebol group dversfcaton measure (CFIRMS, one plus the log number of member frms). Ths varable s ncluded to capture cross-sectonal varaton n the effects of dversfcaton. When the number of cosgners (CFIRMS) ncreases, we would expect debt supply to ncrease as well. 28 Proftablty of the entre Chaebol group mght also matter. One would expect that lenders are more wllng to provde captal to member frms of Chaebols that are more proftable (captured by CPROFIT, Chaebol proftablty multpled by the Chaebol dummy) snce the perceved rsk of falure of a member of a hgh proftablty Chaebol group s low. Sze (CSALES, the log sze of the Chaebol group, excludng the member frm) may also ndcate hgher rsk protecton. Because sze s expected to have a protectve effect then, all else equal, we would expect a postve relatonshp between Chaebol group sze and member frm debt provson. Before movng to the results, the nterpretaton of the mpact of Chaebol membershp deserves dscusson. Chaebols are dversfed groups of (manly sngle-ndustry) member frms; however, one potental concern s that the CHAEBOL dummy s capturng somethng unrelated to group dversfcaton. For example, ths varable mght be related to poltcal power of controllng famles or the Korean government's export-orented polcy. Note that the frm sze, Chaebol group sze, and export controls should capture these potental factors. In addton, the CFIRMS varable, whch captures cross-sectonal varaton n the number of cosgners across Chaebol frms, allows dentfcaton of a dversfcaton effect beyond Chaebol group membershp. The frst step n the analyss requres a decomposton of equlbrum debt ratos based on the supply and demand varables presented n Table 5. Results of estmaton of the 2SLS model of equlbrum debt on these varables are presented n Table Of the demand sde varables, the results suggest that 25 When we dscuss supply and demand, we generally thnk of downward slopng demand and upward slopng supply curves. That s, negatve coeffcents on demand varables and postve coeffcents on supply varables; however, n ths framework, the busness rsk varable s actually reverse, so the expected sgns are flpped (value of ths varable means less busness rsk.). 26 Large frms are perceved to have lower exogenous rsk compared to small frms n that they tend to be more dversfed and may also have easer access to rescue loans when they face lqudty constrants or fal. 27 Industry dummes are based on Korean Stock Exchange ndustry codes (smlar to 3-dgt SIC codes). The dummes, as well as all other exogenous varables n the system are used n the frst stage regressons. 28 Cho and Cowng (2002) report that the number of afflate frms also mportant n explanng Chaebol performance. 29 As noted above, growth (GROWTH) and busness rsk (BUSRISK) are endogenous and are nstrumented n a frst stage regresson wth ndustry dummes as the nstruments. These nstrumented varables are used n the second stage regresson of equlbrum debt ratos on supply and demand factors.