Scholarship Project Paper 02/2012

Transcription

1 Scholarshp Proec Paper 02/2012 HE DEERMINANS OF CREDI SPREAD CHANGES OF INVESMEN GRADE CORPORAE BONDS IN HAILAND BEWEEN JUNE 2006 AND FEBRUARY 2012: AN APPLICAION OF HE REGIME SWICHING MODEL reerapo Kongorann Deparmen of Busness Admnsraon, Marn De ours School of Managemen and Economcs, Assumpon Unversy November 2012 Absrac Mos emprcal sudes on cred spread change precludng dosyncrac and sysemac rsk facors are faled o explan he cred spread puzzle. In pracce, cred spread change vares n response o he cred cycle whch s dfferen from he busness cycle. hs sudy proposes an emprcal model wh wo-sae swchng regme model of low and hgh varance regmes whch are exraced by Markov swchng model o explan he varaon of cred spread change n haland. he resuls sugges ha he model can explan he varaon of cred spread change more effcen han he sngle-regme model. he sensves of neres rae, macroeconomc, and lqudy facors are conssen wh assocaed heores and he cred spread change are more sensve o hese facors n hgh regme across cred rang and me-o-maury groups. he low rang group s no sensve o he lqudy rsk. JEL Classfcaon: G12 Keywords: cred spread, corporae bond, Markov swchng model E-Mal Address: reerapo@gmal.com Dsclamer: he vews expressed n hs workng paper are hose of he auhor(s) and do no necessarly represen he Capal Marke Research Insue or he Sock Exchange of haland. Capal Marke Research Insue Scholarshp Papers are research n progress by he auhor(s) and are publshed o elc commens and smulae dscusson.

2 Capal Marke Research Insue, he Sock Exchange of haland 02/2012 Conen Page Chaper 1 Inroducon 1 Research Obecve 3 Expeced Oucomes 3 Defnon 4 Chaper 2 Revew of Leraure 5 Chaper 3 Mehodology 9 he Sngle-Regme Mulple Regresson Model 9 he Mulple-Regme Mulple Regresson Model 10 Chaper 4 Fndngs 15 he Emprcal Resuls 18 Chaper 5 Dscusson 22 Chaper 6 Concluson and Recommendaons 23 References able 1. Esmaed Parameers from Markov Regme Swchng Model Descrpve Sascs Deermnans of Cred Spread Changes n haland by Rangs and me-o-maury Groups 19

3 Capal Marke Research Insue, he Sock Exchange of haland 02/2012 Chaper1 Inroducon he corporae bond marke n haland s growng dramacally due o he need for dversfcaon n porfolo rsk. However, he secondary marke s relavely neffcen n comparson o he equy marke. he ransacon daa and frequency are a lower levels due o he emergng sage of he marke. Mos sudes sugges ha he cred spread change s affeced by he cred rsk ncludng neres rae rsk, macroeconomc rsk, and lqudy rsk, assumng he defaul rsk s conrolled by he cred rang of he corporae bond (Colln- Dufresne, Goldsen & Marn, 2001; Huang & Huang, 2003). hs leads o he problem so called cred spread puzzle. I canno be solved by eher heorecal models or ordnary emprcal models of dosyncrac or sysemac rsk facors. Sudes on he regme-swchng model n economc leraure have been wdely appled n explanng he unusual macroeconomc evens for he dependen varable. For cred spread leraure, he mehod s mplemened o explan he cred spread puzzle. he cred cycle effec showed ha he sensvy of cred spread change o s deermnan was parally dfferen under low and hgh regmes (Daves, 2008; Maalaou, Donne & Franços, 2008). he cred cycle s relaed o he recovery rae of he frm as n he heorecal model of Meron (1973). he emprcal sudes ha focused on he deermnans of corporae bond yeld spread n haland are lmed o he cred spread level (Mongkonkacha & Paarahammas, 2010). Alhough, he explanaory power of he model was up o 80% n low cred rang corporae bonds wh specfc monh dummy varables, he sudy was a cross-seconal sudy a an ndvdual-level usng radonal bond characersc as explanaory varables, whch canno explan he dynamcs of he cred spread process n dfferen cred cycles. 1

4 Capal Marke Research Insue, he Sock Exchange of haland 02/2012 hs sudy focuses on explanng he facors affecng he aggregae cred spread change of corporae bonds n haland beween June 2006 and February 2012 durng dfferen cred cycles whle conrollng he me-o-maury and cred rang. I hghlghs he use of a regme-swchng model n explanng cred spread n dfferen cred cycles. he research quesons are: Wha are he deermnans of cred spread change n haland n overall and n wo-regmes of cred cycle.e. low and hgh regme? he resuls fnd ha he equy marke reurn, he change of neres rae, and he change of marke lqudy are key deermnans of cred spread change across rang and meo-maury group. However, he low rang group does no have a relaonshp wh lqudy facors. he cred spread change s more sensve o he facors n he hgh regme and more robus n he low regme. I suggess ha he ndvdual nvesors should ener he bond marke when he marke equy marke reurn, he change of neres, and he lqudy are low. he consequence s he hgh cred spread change, n parcular n he hgh regme. When he cred cycle s n he low regme, he nvesors should sell her corporae bonds, n parcular n he low regme, when he equy marke reurn, he change of neres rae, and lqudy are hgh. On he supply sde he corporae bond ssuers should make decsons abou capal srucure opmzaon o maxmze shareholder prof hrough callable bonds before he maury of bonds. For he noncallable opon, s mporan for he ssuer o know he opmal cred spread as ndcaed by he marke rae. he cred spread change dynamc suppors he ssuer n seng he IPO prce of he corporae bonds. For fnancal nsuons, a good undersandng n explanng cred spread change can affec he measuremen of he probably of defaul due o he change n ndependen varables. For he regulaor sde, hough s no clear whch facor makes cred cycle swch, he regulaor should be aware of he moneary polcy o appropraely regulae he neres rae. Moreover, he ncrease of lqudy n he marke can reduce he cos of fundng o he nvesor. he auhory should provde more nformaon o smulae hgher radng n he secondary marke. 2

5 Capal Marke Research Insue, he Sock Exchange of haland 02/2012 Research Obecve: he frs obecve s o fnd he sascally sgnfcan facors affecng he change of cred spread. he deermnans of he cred spread are relaed o he rsk affeced on he value of corporae bonds, whch nclude neres rae rsk, macroeconomc rsk, and lqudy rsk. I s neresng o affrm he cred spread puzzle for he corporae bond marke n haland. he second obecve s o sudy he effec of he laen varables of he cred cycle on he swch of he sensves of he explanaory facors of he cred spread. he deermnans of he cred spread n low and hgh regmes of cred cycle are esed. he focus s on he swch of he sensves durng low and hgh regmes of cred spread. he cred spread puzzle can be affrmed afer usng swchng-regme models. Expeced oucomes: he resuls from he mulple-regme model can deermne he drvers of cred spread of corporae bond n haland n he low and he hgh regmes. he sgn on boh regmes can be eher smlar or dfferen, due o he regme swches. he res of he paper s organzed as follows. In Secon II he revew of leraure on cred rsk model and relaed rsks on corporae bond are examned. In Secon III, he daa and mehodology are dscussed. In Secon IV, he resuls are provded and he dscusson of he resuls s presened n Secon V. Lasly he concluson can be drawn n Secon VI. 3

6 Capal Marke Research Insue, he Sock Exchange of haland 02/2012 Defnon: Corporae Bond Cred Rsk Cred Spread Cred Spread Puzzle Markov Sae Swchng Model A deb nsrumen ha oblgaes he ssuer o pay a specfed percenage of he bond s par value on desgnaed daes and o repay he bond s par or prncpal value a maury" (Fabozz, Mann & Wlson, 2005, pp. 305). he dsrbuon of fnancal losses owng o unexpeced changes n he cred qualy of he counerpary n a fnancal agreemen" (Backshall e al., 2005, pp. 779). "I ncludes cred defaul rsk and cred spread rsk" (Fabozz e al., 2005, pp. 327). he dfference beween a corporae bond s yeld and he yeld on a comparable-maury benchmark reasury secury. Cred spreads are drven by boh macroeconomc forces and ssuespecfc facors" (Fabozz e al., 2005, pp. 329). he puzzle ha such facors [mpled by he heory] explan lle [explanaory power] of hese spreads [cred spread]" (su, 2005, pp. 1) A ype of specfcaon whch allows for he ranson of saes as an nrnsc propery of he economerc model. Such ype of sascal represenaons are well known and ulzed n dfferen problems n he feld of economcs and fnance" (Perln, 2010, pp. 1). 4

7 Capal Marke Research Insue, he Sock Exchange of haland 02/2012 Chaper2 Revew of Leraure In hs secon he lnk beween he heores and cred spread, ncludng he man varables s demonsraed. he heores are classfed accordng o he rsks affecng he cred spread,.e. defaul rsk, neres rae rsk, macroeconomc rsk, lqudy rsk, and me-o-maury rsk. A) Defaul Rsk - he srucural model, developed from opon prcng heory of Black and Scholes (1973) s used o value he cred spread. he model explans relaonshp beween cred spread and he defaul probably, such ha he ncrease n frm defaul probably ncreases he cred spread. When he frm value ncreases, he defaul probably s low. he frm value s proxed by frm equy reurn (Kwan, 1996; Campbell & aksler, 2003; Anramov, Josova & Phlpov, 2007; Chen & Kou, 2009; Mongkonkacha & Paarahammas, 2010) or nversely measured hrough he leverage rao (Colln-Dufresne e al., 2001). B) Ineres Rae Rsk - Whle n he srucural model, he rskfree rae were assumed o be consan, he reduced-form model allowed he facors o be a sochasc process. he model can f he emprcal daa beer han he srucural model, bu sll canno explan he deermnans of cred spread varaon (Jarrow & urnbull, 1995; Lesseg & Sock, 1998; Elon, Gruber, Agrawal & Mann, 2004; Longsaff, Mhal & Nes, 2005). he negave relaonshp beween he change of neres rae level and cred spread s expeced, due o he fac ha he ncrease n rsk free rae ncreases he drf of he rsk-neural process of frm value, and herefore he probably of defaul decreases and he cred spread narrows (Longsaff & Schwarz, 1995). Moreover, f he frm values are correlaed wh he neres rae, he frm value volaly s relaed wh boh equy volaly and neres rae volaly. Increasng neres rae volaly ncreases he frm volaly and herefore he cred spread ncreases (Huang & Kong, 2003). Pure expecaon heory explans ha he relaonshp beween forward rae and fuure 5

8 Capal Marke Research Insue, he Sock Exchange of haland 02/2012 rae, ha hey are equal (Cox, Ingersoll & Ross, 1981). herefore he slope of he erm srucure can mply he fuure neres rae. he seeper he slope of he erm-srucure, he hgher he fuure neres rae. he fuure probably of defaul decreases, when he expecaon of fuure neres rae ncreases; herefore he cred spread narrows (Colln-Dufresne e al., 2001). C) Macroeconomc Rsk - Snce he srucural model canno explan cred spread change effcenly, he sysemac rsk facors are wdely seleced as he choce of deermnans of cred spread change (Colln-Dufresne e al., 2001). he fnancal nsably hypohess from Mnsky (1992) explaned he cred crss of he deb marke caused by he accumulaon of he deb from hree mbalance number of hree groups of borrowers,.e. hedge borrowers, speculave borrowers, and Ponz borrowers. he cred cycle s dfferen from he busness cycle snce he assumpon of he heory s based on dfferen economy srucure. he effcen markes hypohess shows ha here s a relaonshp beween he equy and bond markes hrough non-synchronzed percepons of prvae nformaon. herefore he equy marke wh low ransacon cos reacs o he nformaon before he bond marke, and herefore he relaonshp beween he lagged reurn of equy marke can explan he bond marke reurn (Kwan, 1996). D) Lqudy Rsk - Analogous o he sock marke, he aggregae lqudy facors can explan he sock reurns, such ha he socks wh hgher lqudy gve hgher expeced reurn o he nvesors (dos Sanos Pava & Savoac, 2009). Lqudy heory explans why he nvesors requre hgher premum on nvesng on he slow-cash-convered secures, such as corporae bonds, han he fas-cash-convered secures, such as reasury bonds (Nakashma & Saob, 2009). he cred spread puzzle of he Meron model may cause hs by no ncludng he lqudy facor n he model, especally n very shor-erm corporae bonds (Covz & Downng, 2007). he hgher he lqudy, he lower he cred spread (Lo, Mamaysky & Wang, 2004). 6

9 Capal Marke Research Insue, he Sock Exchange of haland 02/2012 E) me-o-maury Rsk - Snce he volaly of he yeld of corporae bonds s relaed o he me-o-maury, he cred spread s affeced by he me-o-maury. hs relaonshp s called erm-srucure, whch can be ncreasng, decreasng or hump shaped. he man heory relaed o he ncreasng shape s lqudaon heory, whch explans ha he nvesor demands hgher premums on longer maury bonds n order o hold a longer one. herefore, he yeld o maury ncreases wh he me-o-maury (Fabozz, 2005). In explanng cred spread change here are hree maor concepual models, ) sngle-regme model, ) smulaon model, and ) mulple-regme model. For he sngle-regme model, he deermnans of cred spread are based on defaul facors, neres rae facors, macroeconomc facors, and lqudy facors. he man par of he cred spread varable remans unexplanable (Colln-Dufresne e al., 2001; Campbell & aksler, 2003; Huang & Kong, 2003; Nakashma & Saob, 2009; Chebb & Hellara, 2010). he bond prcng models from Meron (1974), Geske (1977), Longsaff and Schwarz (1995), Leland and of (1996), and Colln-Dufresne and Goldsen (2001) were compared as n he work of Eom, Helwege and Huang (2004) n predcng bond prce. hey are based on relaed heores, however, he predced error of all he models were sensve o he sysemac rsk. he las ype of he framework s he use of mulple-regme model o explan he cred spread n low and hgh regmes. he measuremen of cred cycle was based on he Markov swchng model (MS model) from Hamlon (1990). he framework of Daves (2008) and Maalaou e al. (2008) used me-dummy varable based on he regme exracon from cred spread change. Alernavely, he busness cycle can be observed by macroeconomc ndcaors as follows. Daves (2008) used he SEAR model (Self Exracng hreshold) based on annual CPI nflaon, whle Maalaou e al. (2008) used he recesson announcemen from NBER and se he me-dummy varable accordngly. Mongkonkacha and Paarahammas (2010) se he me-dummy varable as one when he monh s assocaed wh a bear perod of he equy marke. he cred cycle can mprove explanaory of he model beer han he 7

10 Capal Marke Research Insue, he Sock Exchange of haland 02/2012 busness cycle (Maalaou e al., 2008). he model can explan cred spread beer han he sngle-regme model based on he mnmze AIC (Akake Informaon Crera). In he nex secon he measuremen of dependen and ndependen varables and he mehodology are explaned. 8

11 Capal Marke Research Insue, he Sock Exchange of haland 02/2012 Chaper 3 Mehodology follows. he regresson model of sngle- and mulple-regme models can be demonsraed as he Sngle-Regme Mulple Regresson Model he model s smply an ordnary mulple regresson analyss wh a general form expressed as porfolo, varables CS where 0 N = X (1) =1 ( ), CS s he cred spread change of cred rangs and me-o-maury 0 s a consan, are esmaed sensves of he observed ndependen y X,, ncludng he rsk free rae ( r 2 10 y 2 y ), he slope of erm srucure ( ) and 10 y slope r he neres rae volaly ( ), lag of he equy marke reurn ( se 2 ), lag of he hsorcal se equy marke reurn ( ), mssng prce raos of each cred rang group ( ms ) and 2 urnover rao ( urn ), and s unexplaned par. he models are conrolled by four rang groups (e.g. AAA, AA, A, and BBB) and hree me-o-maury groups (e.g. shor, medum, and long me-o-maury) for he defaul and me-o-maury rsk (Colln-Dufresne e al., 2001; Maalaou e al., 2008). he sensves should be compared whn he rang and me-o-maury groups, wheher hey are changed monooncally wh rang and me-o-maury changes. 9

12 Capal Marke Research Insue, he Sock Exchange of haland 02/2012 he Mulple-Regme Mulple Regresson Model he mulple-regme mulple regressons focus on he regme change of he cred cycle by addng addonal neracon erms (Daves, 2008; Maalaou e al., 2008). he general erm of he mulple regme model s as follows CS N N ( ) ( ) 0 X, 0 ner X, ner =1 =1 N, L ( ) = 0 X, when ner = 0 =1 N, H ( ) = ( 0 0 ) ( ) X, when ner =1 = (2) CS (3) CS = 1 (4) where 0 and he margnal effec erms. he dummy varables are parameers of he low regme, and are parameers of 0 ner are equal o zero when he cred cycle regme of porfolo s n he low regme, oherwse one (Daves, 2008; Maalaou e al., 2008). and can be nerpreed as sensves of explanaory varables n low and hgh regmes respecvely. he proxes of dependen and ndependen varables are dscussed as follows. A) Cred Spreads - he ndvdual corporae bond daa are obaned from he marko-marke of he fxed-ncomes prepared by he habma va he Bond daabase, ncludng sac cred spread 1, me-o-maury, and cred rang. o ensure ha he porfolos have no addonal rsk, he corporae bonds wh an embedded opon and a floang coupon ype are flered ou. Moreover, hey are dscarded whenever a bond has less han one year me-o- 1 he ndvdual sac spread, CS, for bond a monh, s calculaes from he prce of he corporae bond and he spo rae of he governmen yeld curve as follows, P c = [1 ( zc c CS )] [1 ( zc2 CS )] [1 ( zcn 10 c Par where P s a prce of he corporae bond a me, c s a coupon paymen, Par s a par value of he corporae bond, zc k spo rae for he coupon paymen me k, where k s he coupon paymen perod from 1 o n. CS was he sac spread s assumed o be consan across all he coupon paymen perod, k. he calculaon of he sac spread can be found from SOLVER and GOALSEEK from Mcrosof Excel (Swamogsahem, 2010). CS )] n

13 Capal Marke Research Insue, he Sock Exchange of haland 02/2012 maury (Colln-Dufresne e al., 2001; Maalaou e al., 2008). he corporae bonds whou a cred rang eher RIS or FICH are also flered ou. he non-nvesmen bonds,.e. wh a cred rang lower han BBB, are flered ou. he aggregae sac spread curves, e.g. AAA, AA, A, BBB 2, hrough he lognormal funcon as n equaon 5. CS where ( ) = a a 0 ln(1 ) 1 CS, are consruced for each cred rang,, (5) CS s he sac spread curve of porfolo a monh. I s a funcon of me-o-maury, and a 0 and a 1 are he parameers from he OLS of CS ( ) and s relaed me-o-maury (Swamogsahem, 2010). he lognormal funcon gves an ncreasng funcon whch s reasonable for he cred spread of corporae bonds ha s ncreasng wh he me-o-maury. he CS (, ) s a seres of cred spread a me-o-maury can be calculaed from he parameer of obaned from he OLS. B) Ineres Rae Facors he neres rae rsk s capured by he neres rae level, he slope of he yeld curve, and he neres rae volaly. he neres rae level ( obaned from he yeld of wo-year reasury yeld habma zero coupon yeld curve. he proxy of he slope of he yeld curve ( slope reasury bonds. he hsorcal neres volaly ( y r 2 ) s 10 y 2 y) s he spread beween 10-year and 2-year maury 10 y r ) s calculaed from he one-year hsorcal neres rae volaly usng he daly spo rae of en-year-maury reasury bond yeld from he habma zero coupon yeld curve. C) Macroeconomc Facors Macroeconomc rsk ncludes equy marke reurn and hsorcal volaly of equy marke reurn. he equy marke ndex s obaned from he Bloomberg daabase va Sock Exchange of haland (SE). he reurn of he sock marke ( 2 o form a porfolo of corporae bond n cred rangs, he cred rangs from RIS or FICH are consoldaed by he one wh he lower rae. Snce RIS and FICH use he same ermnology of cred symbols, a rang converson s no requred. Secondly, he rangs wh and are grouped o he mddle rang whch are AA, A, and BBB (Maalaou e al., 2008). 11

14 Capal Marke Research Insue, he Sock Exchange of haland 02/2012 se ) s he percenage change of he sock ndex. he hsorcal volaly of equy marke reurn ( se Bloomberg daabase. ) s calculaed from he sandard devaon of daly reurn back for 180 days from he D) Lqudy Facors Lqudy facors are measured by he mssng prce, he rao of non-radng days o he number of days n he monh of each bond. and he urnover rao ( urn ) s a rao beween he oal radng volume n he monh o he number of ousandng bonds (Maalaou e al., 2008). he hgher he rao, he more frequen rades are n he secondary marke. he radng volume daa s from he prcng daa from he habma webse. he ousandng volume s calculaed by usng value of he ousandng a he end of he monh dvded by par value of 1,000 HB. E) Ineracon erms he neracon erms of each porfolo ner are obaned from he Markov Swchng Model usng he me-seres daa of cred spread of each porfolo as an npu of he analyss. he cred spread of each porfolo s assumed o have he followng process ln CS = (6) S where = 1,2 S S, s he mean of ln a sae S, and follows a normal S dsrbuon wh zero mean and varance 2 S CS. he means and varances n regme 1 and 2 are dfferen. he ranson of he sae s assumed sochasc or s unceran wheher he sae wll change or no. Snce he swchng process s assumed o be known, a ranson marx ha conrols he probably of makng a swch from one o anoher sae s represened by he probably of a swch from sae n o sae m beween me and 1 ( p mn ). he esmaon of he parameers n he Markov Swchng model s based on he maxmum lkelhood usng Hamlon s fler 3. he calculaon of he smoohed probables can S 3 he model follows he Bayesan pror for wo regmes seng and he generalzaon of he Maxmum Lkelhood Esmaon obecve funcon s 2 2 ( ) = log p ( y1,, y ; ) ( 1 )/(2 ) 1 12

15 Capal Marke Research Insue, he Sock Exchange of haland 02/2012 be obaned from an erave process. he smoohed probably s hen used o wegh y. he OLS calculaon of he weghed y s performed o generae he new esmae parameer. he process s repeaed agan unl he value of he lkelhood funcon s maxmum and he fxed parameer of s found (Maalaou e al., 2008) ( 2 )/(2 2 ) log1 log 2 / 2 1 / 2 2 where,, are specfc Bayesan pror. he parameers ˆ ncludng n are calculaed from he erave algorhms (Hamlon, 1990) as where y =1 ˆ n = 2 ˆ n =1 p( S p( S = n y,, y 1 = n y,, y 1 ; ˆ) ; ˆ) 1 = 1/2 p( S = n y,, y ; ˆ ) =1 [ n 1 =1 2 1/2 ( y ˆ ) p( S = n y,, y ; ˆ ) 1/2 ˆ 2 n ] pˆ 11 = [ p( S = 1, S 1 = 1 y1,, y ; ˆ)]/ =2 [ p ( S 1 1 =2 = 1 y,, y ; ˆ) ˆ p( S = 1 y1,, ; ˆ)] y pˆ 22 = [ p( S = 2, S 1 = 2 y1,, y ; ˆ)]/ =2 [ p ( S 1 1 =2 = 2 y,, y ; ˆ) ˆ p( S = 1 y1,, ; ˆ)] y 1 pˆ 22 ˆ = (1 pˆ ) (1 pˆ ) ˆ,, ˆp 11, ˆp 22 he calculaon process s o esmae he parameers by employng Expecaon-Maxmzaon (EM) algorhm, he erave process o fnd he maxmum lkelhood of he model wh a laen varable (Dempser, Lard & Rubn, 1977). ˆ n

16 Capal Marke Research Insue, he Sock Exchange of haland 02/2012 For hs sudy, he calculaon of smoohed probably and parameer s performed by MALAB package MS_Regress wren by Perln (2010). Whenever he smoohed probably s lower han 0.5, he cred spread s n he low regme and vce versa. he smooh probably s convered o dummy varable, ner, usng hs crera (Maalaou e al.,. 2008). 14

17 Capal Marke Research Insue, he Sock Exchange of haland 02/2012 Chaper 4 Fndngs he daa are groups o four cred rang groups and he lognormal funcon fs he cred spread and me-o-maury. Afer runnng he algorhm n MALAB under he assumpons of wo regmes wh dfferen means and varances and fxed condonal probably of 11 p and p 22, he model canno dsngush beween wo dfferen regmes. One of he reasons ha wo dfferen regmes canno be found s ha he daa range s relave small (only 69 observaons). Durng he sample perod, here s only one cred cycle. However, wh an alernave assumpons of one common mean wh wo dfferen varances and fxed condonal probably of p 11 and p 22, mos of he cred spreads can be deermned n wo dfferen regmes,.e. low and hgh varance regmes. he summary of esmaed parameers from Markov regme swchng model s n able 4.1 Across all porfolos, mean of cred spread ncrease wh longer me-o-maury and lower cred rang. he volales of cred spread n sae 1 s lower han sae 2. he condonal probables p 11 s hgher han p 22 whn he range of 0.67 o hey are more perssen and relucan o swch beween dfferen regmes. o consruc neracon erm ( ner ), whenever he connuous probably s lower han 0.5, he cred spread s n he low regme, oherwse s n he hgh regme. he Augmened Duckey Fuller es s performed o all dependen and ndependen varables o avod he non-saonary problem. Mos of he varable are faled o reec he null hypohess ha here s a un roo. herefore hey are non-saonary daa, exceped for he volaly of long-erm neres rae and he equy marke reurn, whch can reec he null hypohess a 1% sgnfcance level. o solve he non-saonary problem, he frs dfference s appled o he non-saonary seres and reesed wh ADF. he resul shows ha he null 15

18 Capal Marke Research Insue, he Sock Exchange of haland 02/2012 able 1: Esmaed Parameers from Markov Regme Swchng Model hypoheses are reeced for all he frs dfference of he non-saonary daa wh sgnfcance level a 1%. he resuls of he es s no shown here. able 2 presens he descrpve sascs of he dependen and ndependen varables ncludng her mean, medan, maxmum and mnmum values, sandard devaons, skewness, Kuross, Jacquc-Bera and s p-value. Snce he bond marke has much less lqudy han he sock marke, he nformaon flowng o he bond marke s much slower han he sock marke. he approprae lags of ndependen varables should be n consderaon. he unvarae lnear regressons are performed beween each ndependen varable wh dfferen lags and welve cred spread changes. he crera o selec he approprae lag s usng he mnmum Akake Creron. he resuls are no shown here. Mos of he ndependen varables do no need lag erms, excep for he reurn of he equy marke and s hsorcal volaly, whose lag erms of wo monhs can mnmze he AIC. he las prelmnary resul s o check wheher he ndependen varables havng correlaon beween each oher. If he correlaon beween par of ndependen varable are hgher han 0.80 or lower han -0.80, ncludng of boh varables n he analyss would lead o 16

19 Capal Marke Research Insue, he Sock Exchange of haland 02/2012 mulcollneary problem (Guara, 2003).. he correlaon marx, no showng here, shows no par of ndependen varables wh hgh correlaon. able 2: Descrpve Sascs 17

20 Capal Marke Research Insue, he Sock Exchange of haland 02/2012 he Emprcal Resuls A. Sngle Regme Model - he resuls of he OLS analyss of a sngle regme model are presened n panel A n able 4.3. he consan erms are no sascally sgnfcan for all porfolos. he change of neres rae s sascally sgnfcan across all porfolos. he sensvy o he change of shor-erm neres rae ncreases across cred rang groups, bu decreases across me-o-maury. Only he AAA rang group, he change of slope of he erm srucure s slghly posvely relaed wh he cred spread change. he coeffcen value ncreases wh he me-o-maury. he lag of reurn of he equy marke s negavely correlaed wh cred spread change boh economcally and sascally. Moreover, he sensves ncrease conssenly across cred rang and me-o-maury groups. However, he change of he hsorcal volaly of reurn of equy marke s no sascally sgnfcan. he urnover rao s sascally sgnfcan n AA and A rang groups. he sensves ncrease wh he me-o-maury. he mssng prce rao s only sascally sgnfcan n AAA rang group. he sgn of boh lqudy facors are as expeced. Surprsngly, he BBB rang group s he only group ha s no relaed wh lqudy facor. B. Mulple Regme Model Panel B n able 4.3 shows he resuls from he regresson of mulple-regme model. For he low regme, mos of he explanaory varables are no sascally sgnfcan, excep for he change of neres rae level n AA and A rang groups, he reurn of equy marke and s hsorcal volaly n shor-erm AA rang group, he urnover rao n AAA rang group, and he mssng prce rao n AA and A rang groups. Mos of he sgns are as expeced from he heores, excep for he reurn of equy marke and urnover rao. 18

21 able 3: Deermnans of Cred Spread Changes n haland by Rangs and me-o-maury Groups Capal Marke Research Insue, he Sock Exchange of haland 02/

22 able 3: Connued Capal Marke Research Insue, he Sock Exchange of haland 02/

23 Capal Marke Research Insue, he Sock Exchange of haland 02/2012 For he hgh regme, he margnal pars of explanaory varable are dscussed. he change of shor-erm neres rae of mos of he porfolo s srongly sascally sgnfcan. he relaonshp beween neres volaly s only sgnfcan n medum-erm AAA, shor- and longerm AA. However, he sgns are mxed. Mos of he porfolos have srong negave relaonshps beween he reurn of equy marke and he cred spread change. However, he hsorcal volaly of he equy marke s no sgnfcan. he lqudy facors only play mporan role for medum-erm AA and A rang group and long-erm A rang group wh expeced sgn. he adused R-squared for mulple-regme models are beween 0.44 and 0.83, whle he adused R-square for sngle-regme models are beween 0.29 o he mulpleregme models can explan he cred spread change beer han he sngle-regme models. For all porfolos, he Akake crerons n mulple-regme models are lower han sngle-regme models. 21

24 Capal Marke Research Insue, he Sock Exchange of haland 02/2012 Chaper 5 Dscusson For he sngle regme and he hgh regme, he change of shor-erm neres for mos of he porfolos are conssen wh he emprcal fndngs of Longsaff and Schwarz (1995), Duffee (1998), Colln-Dufresne e al. (2001), and Maalaou e al. (2008), ha an ncrease n he shor-erm neres rae decreases he rsk-neural probably of defaul. However n he low cred rang he neres facor s less sensve. For he slope of erm-srucure, n he hgh regme, he ncrease of he slope wdens he cred spread. hs shows ha he fuure ncrease of neres rae durng he hgh cred cycle s no good news for he nvesors; herefore he cred spread wdens. he reurn of equy marke s sensve across all porfolos. Beween equy and bond marke here s a negave relaonshp and he equy marke reacs faser o he nformaon faser han he bond marke, as he lag of equy reurn can explan change of cred spread. In he hgh regme, he ncrease of he equy reurn, ncrease he frm value away from he defaul hreshold as n Meron (1973). he lqudy facors have relaonshp as expeced for all porfolos excep for he low cred rang. hese resuls are dfferen from oher sudes ha he low cred rang s more sensve o he lqudy change (Colln-Dufresne e al., 2001; Maalaou e al., 2008). One of he reason s ha he low cred rang s raded very few n secondary marke; herefore he mark-o-marke value s manly deermned he value of he bonds from he neres rae and macroeconomc facors. 22

25 Capal Marke Research Insue, he Sock Exchange of haland 02/2012 Chaper 6 Concluson and Recommendaons hs sudy nvesgaes he use of mulple-regme models n explanng he deermnans of cred spread change n haland. Several resuls are obaned. Frs, he mulple-regme model can explan he cred spread change n haland beer han he sngleregme model. Second, he resuls from mulple-regme model show ha n he low regme, he cred spread change s less sensve han n he hgh regme. hrd, he key deermnans across all porfolos are neres rae change and reurn of equy marke. he lqudy facor can only explan for hgh-cred rang. he fndngs sugges ha durng he low regme, he cred spread reman sable and s a good me for ssuers o ssue he new bond wh compeve cos. For he nvesor, he cos of he bond s less durng he hgh regme. As he neres rae and reurn of equy marke decreases, he value of frm s less, he cred spread ncreases sgnfcanly. Invesng he corporae bond a hgh rang s recommended. In general, he hgher lqudy of he secures marke can reduce he cred spread. he sudy suggess ha he need for he furher work on oher echnque on oher regme swchng model, such as SEAR, PSAR/LSAR for he more sophscaed model o mprove he explanaory power of he model. Furhermore, he lqudy proxes have many varees of choces. he hgher frequences of he daa and more aspecs of he daa, e.g., prce and volume of porfolo or bond marke or co movemen marke, f avalable, should be ncluded no he model (Houwelng, Mennk & Vorsb, 2005; Mahana, Nashkkarc, Subrahmanyamc, Chackod & Mallk, 2008; Acharya, Amhud & Bharah, 2010). Snce he reurn of equy marke has he srong relaonshp wh cred spread, s neresng o apply he even sudy o precsely exrac he mng of cred spread changes. I should be aware of 23

26 Capal Marke Research Insue, he Sock Exchange of haland 02/2012 asymmercal effec durng hgh and low varance of equy marke reurn as dscussed n Colln- Dufresne e al. (2001), or mplemen he even sudy ogeher wh neracve erm o refne he resul durng he ranson perod. 24

27 Capal Marke Research Insue, he Sock Exchange of haland 02/2012 References Acharya, V. V., Amhud, Y., & Bharah, S.. (2010). Lqudy rsk of corporae bond reurns. AFA 2010 Chcago Meengs Paper. Anramov, D., Josova, G., & Phlpov, A. (2007). Undersandng changes n corporae cred spreads. Fnancal Analyss Journal, 63(2), Campbell, J. & aksler, G. B. (2003). Equy volaly and corporae bond yelds. Journal of Fnance, 58(6), Chebb,. & Hellara, S. (2010). Corporae yeld spreads and soveregn defaul rsk. Inernaonal Revew of Appled Fnancal Issues & Economcs, 2(1), Chen, N. & Kou, S. (2009). Cred spreads, opmal capal srucure, and mpled volaly wh endogenous defaul and ump rsk. Mahemacal Fnance, 19(3), Colln-Dufresne, P., Goldsen, R. S., & Marn, J. P. (2001). he deermnans of cred spread changes. Journal of Fnance, 56(6), Covz, D. & Downng, C. (2007). Lqudy or cred rsk? he deermnans of very shor-erm corporae yeld spreads. Journal of Fnance, 62(5), Cox, J., Ingersoll, J., & Ross, S. (1981). A re-examnaon of radonal hypoheses abou he erm-srucure of neres raes. Journal of Fnance, 48(2), Daves, A. (2008). Cred spread deermnans: An 85 year perspecve. Journal of Fnancal Markes, 11, Dempser, A., Lard, N., & Rubn, D. (1977). Maxmum lkelhood from ncomplee daa va he EM algorhm. Journal of he Royal Sascal Socey B, 39,

28 Capal Marke Research Insue, he Sock Exchange of haland 02/2012 dos Sanos Pava, E. V. & Savoac, J. R. F. (2009). Prcng corporae bonds n Brazl: 2000 o Journal of Busness Research, 62, Duffee, G. R. (1998). he relaon beween reasury yelds and corporae bond yeld spreads. Journal of Fnance, 53(6), Elon, E. J., Gruber, M. J., Agrawal, D., & Mann, C. (2004). Facors affecng he valuaon of corporae bonds. Journal of Bankng & Fnance, 28(11), Fabozz, F. J. (2005). Cred analyss for corporae bonds. In F. J. Fabozz (Ed.), he handbook of fxed ncome secures (7h ed.). chaper 32, (pp ). Mc Graw Hll. Geske, R. (1977). he valuaon of corporae lables as compound opon. Journal of Fnancal & Quanave Analyss, 12, Guara, D. N. (2003). Basc Economercs (fourh ed.). Sngapore: McGraw Hll. Hamlon, J. D. (1990). Analyss of me seres subec o regme changes. Journal of Economercs, 45, Houwelng, P., Mennk, A., & Vorsb,. (2005). Comparng possble proxes of corporae bond lqudy. Journal of Bankng & Fnance, 29, Huang, J.-Z. & Huang, M. (2003). How much of he corporae-reasury yeld spread s due o cred rsk? PhD hess, Sanford Unversy. Huang, J.-Z. & Kong, W. (2003). Explanng cred spread changes: New evdence from oponadused bond ndexes. Journal of Dervaves, 11(1), Jarrow, R. & urnbull, S. (1995). Prcng dervaves on fnancal secures subec o cred rsk. Journal of Fnance, 50(1), Kwan, S. H. (1996). Frm-specfc nformaon and he correlaon beween ndvdual socks and bonds. Journal of Fnancal Economcs, 40, Lesseg, V. & Sock, D. (1998). he effec of neres raes on he value of corporae asses and he rsk prema of corporae deb. Revew of Quanave Fnance and Accounng, 11,

29 Capal Marke Research Insue, he Sock Exchange of haland 02/2012 Lo, A., Mamaysky, H., & Wang, J. (2004). Asse prces and radng volume under fxed ransacon coss. Journal of Polcal Economy, 112, Longsaff, F., Mhal, S., & Nes, E. (2005). Corporae yeld spreads: Defaul rsk or lqudy? New evdence from he cred defaul swap marke. Journal of Fnance, 60(5), Longsaff, F. & Schwarz, E. (1995). A smple approach o valung rsky fxed and floang rae deb. Journal of Fnance, 50(3), Maalaou, O., Donne, C., & Franços, P. (2008). Cred Spread Changes under swchng regmes. PhD hess, HEC Monréal. Mahana, S., Nashkkarc, A., Subrahmanyamc, M., Chackod, G., & Mallk, G. (2008). Laen lqudy: A new measure of lqudy, wh an applcaon o corporae bonds. Journal of Fnancal Economcs, 88, Meron, R. (1973). A raonal heory of opon prcng. Bell Journal of Economcs and Managemen Scence, 4(1), Meron, R. (1974). On he prcng of corporae deb: he rsk srucure of neres raes. Journal of Fnance, 29, Mnsky, H. (1992). he fnancal nsably hypohess. Workng Paper No. 74, he Jerome Levy Economcs Insuue of Bard College. Mongkonkacha, S. & Paarahammas, S. (2010). Lnkage beween sock volaly and corporae bond yeld spread n haland. Chna-USA Busness Revew, 9(1), Nakashma, K. & Saob, M. (2009). Cred spreads on corporae bonds and he macroeconomy n Japan. Journal of he Japanese and Inernaonal Economes, 23, Perln, M. (2010). MS_Regress - he MALAB package for Markov regme swchng models. Avalable a SSRN:hp://ssrn.com/absrac= Swamogsahem,. (2010). Developmen of habma cred and lqudy spread. Workng Paper. 27