Retail Mortgage Backed Securities, Commercial Asset Backed Securities and Corporate Bonds: a Credit Spread Comparison +

Transcription

1 Retal Mortgage Backed Securtes, Commercal Asset Backed Securtes and Corporate Bonds: a Credt Spread Comparson + LORIANA PELIZZON * Unversty of Padua ENRICO RETTORE Unversty of Padua EMANUELA SOTTANA Fnanzara Internazonale Abstract In ths paper we nvestgate whether and how Retal Mortgage Backed Securtes (RMBS), Commercal Asset Backed Securtes (CABS) n the European market dffer from other more tradtonal nvestment products, n partcular n terms of the trade-off between rsk and return. Therefore n ths paper we compared credt spread on RMBS and CABS to credt spread on corporate bonds wthn the same ratng class and fnd some explanaton of the dfferences between them. We fnd that the market requres the same remuneraton to nvest n a bond of a sngle frm and n a bond on a portfolo wth the same ratng, f ths ratng s hgh. On the contrary, the market requres a dfferent remuneraton f the ratng s low. Moreover, the spread s not statstcally sgnfcant between Retal Mortgage Backed Securtes and Commercal Asset backed securtes when the ratng s hgh, but s hgher for Commercal Asset Backed Securtes f the ratng s low. Potental explanatons are provded. December We gratefully acknowledge conversatons wth Stephen Schaefer, Rajv Guha and Davde Menn. * Dpartmento d Scenze Economche, Unversty of Padua, Va del Santo 33, Padova, Italy, Phone , Fax , pelzzon@decon.unpd.t

2 INTRODUCTION Securtzaton s one of the most nnovatve fnancal technques; t conssts of a transformaton of llqud assets nto negotable securtes. The structure of securtzaton may dffer sgnfcantly from one ssue to another, but the bass scheme s always the same: the orgnator sells part of the assets he has n hs balance sheet to a Specal Purpose Vehcle (SPV), whch s a company created ad hoc n order to securtze the assets. Then the SPV ssues bonds, the Asset-Backed Securtes (ABS), n order to fnance the asset purchase. Securtzaton has been spreadng n Europe very fast snce the last years: n 1996 ABS ssues accounted for 37.2 bllons, whle n 2001 they reached bllons, almost doubled compared to 2000 level 1. A reason for ths exponental growth can be found n the fact that many European countres ntroduced new securtzaton laws durng the 1990s, gvng a steady legal framework to ths knd of transactons. In partcular Italan ABS market has grown quckly and became the second largest n Europe, snce the ntroducton of law 130/99 namely the Securtzaton Law. The advantages that make securtzaton attractve to orgnators are several: captal relef, lqudty, assets-labltes management, fundng, for example. Moreover securtzaton can be a useful tool for transferrng credt rsk because the orgnator may reduce the geographcal and ndustry concentraton of hs assets as well as concentraton of ndvdual exposure. Consderng the spread of securtzaton and thus the growth of ABS market n Europe n the last few years, t seems nterestng to nvestgate whether and how Retal Mortgage Backed Securtes (RMBS), Commercal Asset Backed Securtes (CABS) dffer from other more tradtonal nvestment products, n partcular n terms of the trade-off between rsk and return. Therefore n ths paper we compared credt spread on RMBS and CABS to credt spread on corporate bonds wthn the same ratng class and we try fnd some explanaton of the dfferences between them. 1 Data are from European Securtzaton Forum (2002). 2

3 The paper s organsed as follows: secton 1 deals wth securtzaton as a credt rsk management tool; secton 2 overvews economc lterature about credt spread both on corporate bonds and on Asset-Backed Securtes. Secton 3 presents the emprcal analyss developed n ths paper: frst, data set s brefly descrbed; secondly, a descrptve analyss s shown n order to underlne the excess spread of RMBS and CABS compared to corporate bonds; then some possble explanatons for the excess spread are dscussed; later on results of the regresson analyss s presented. Concludng remarks follows n secton Credt rsk management through securtzaton As sad above, securtzaton can be used to manage credt rsk. In fact sellng assets through a true sale the orgnator can transfer credt rsk, whch s, thus, shared by other agents n the securtzaton process, especally nvestors and credt enhancers. In ths way orgnator s balance sheet s more dversfed, snce geographcal, ndustry and ndvdual exposure concentraton s reduced. Moreover, usng synthetc securtzaton, the orgnator can buy protecton aganst credt rsk and, at the same tme, mantan the assets on hs balance sheet. So securtzaton assures from rsk fragmentaton; t s worth to underlyng that t s not convenent for the orgnator to behave n an opportunstc way. In fact, f he wanted to clean-up hs balance sheet by sellng only the worst assets and so damagng nvestors, he wouldn t have other chances to rse funds n the future snce hs reputaton gets bad. So orgnator has an ncentve to behave correctly to keep hs reputaton good. Standard and Poor s (2002) recently publshed a study on Italan bank-orgnators, n whch they assert that lttle rsk s transferred outsde the orgnatng group. In fact t s common for the orgnator to retan the subordnated tranche to credt enhance the ssue. Ths practce causes the entre rsk to be concentrated n the subscrbed junor notes, that are classfed among the orgnator s securtes. Ths mples that credt rsk s not transferred out through securtzaton; so, f one would use securtzaton as a credt rsk management tool, he must pay attenton to project a correct trade-off between credt rsk transferrng and credt enhancement provdng 2. 2 Credt enhancement s useful because t enables orgnator to ssue hgher rated bonds and thus to fund at a lower cost (n terms of yeld offered to nvestors). 3

4 2. Credt spread The yeld of a bond s nfluenced by several factors, among whch a very mportant one s credt ratng. Ratng reflects the credt rsk of the bond, so credt spread defned as the dfference between the yeld on a (rsky) bond and the yeld on a rsk-free bond (such as, for example, a T-bond) s n large part determned by credt rsk, but other factors may be ncluded as well. Lterature about credt spread on corporate bonds s plentful; He, Hu and Lang (2000) studed the shape of credt spread curves and fnd that hgh-rated bonds have postve sloped curves, low-rated bonds show negatve sloped curves, whle spread curves for mddle-rated bonds are hump-shaped wth peak ponts. Jakson and Perraudn (1999) found evdence for securtes ssued by U.S. banks to offer hgher spreads compared to bonds ssued by other ndustral and utltes U.S. companes. Elton, Agrawal and Mann (2001) found that credt spread s due to default and taxaton premum and also to a premum that covers a systematc rsk. Ths knd of rsk s due to the fact that spreads on bonds are n some part nfluenced also by the same factors that affect stock prces. Colln-Dufresne, Goldsten and Martn (2001) used a lnear regresson to fnd out whch factors determne changes of credt spread. Ther most nterestng result s that factor thought to be mportant varable to explan credt spread changes have lttle explanatory power, whle there s a common factor whch s able to explan the remanng spread; however they could not determne such factor. Annaert and De Ceuster (1999) analysed credt spread on European corporate bonds and found evdence for the presence of a lqudty premum n the yeld of these securtes. Huang and Huang (2002) estmated how much of credt spread s due to credt rsk and found that lttle part of t s explaned by credt rsk for hgh-qualty bonds, whle credt rsk s more mportant n explanng credt spread on junk bonds. There are two partcularly nterestng artcles about credt spread determnants of bonds ssued by a securtzaton transacton: Mars and Segal (2002) and Rothberg, Nothfat Gabrel (1989). The former used a lnear regresson to fnd the determnants of credt spread on CMBS 3 ; results dentfy several varables that are related to spreads. In partcular nterest rate volatlty and a recesson ndex capture the default probablty on underlyng mortgages and so nfluence spread on CMBSs; tranche sze has a negatve slope, provdng some support for the noton that larger tranches are assocated wth lower lqudty prema; at the opposte, total 3 Commercal Mortgage Backed Securtes. 4

5 deal sze are postve and sgnfcant, ndcatng that larger deals are assocated wth hgher spread. The authors explan the last result wth the fact that a larger transacton requre larger spreads to attract a suffcent number of nvestors to place the ssue. Moreover an mportant factor explanng spread on CMBS s a tme dependent varable whch has a negatve and sgnfcant coeffcent that ndcates that CMBS credt spread decrease from year to year. Ths provdes evdence to support the noton that a learnng process s takng place and nvolves ssuers, ratng agences and nvestors. In fact lack of confdence wth analyss and evaluaton of ths knd of securtes caused nvestors to overtmate the rsk and then explans the hgher yeld spreads of CMBSs n the early 1990s. Rothberg, Nothaft and Gabrel (1989) assert that mortgage pass-through and Treasury securtes may reflect dfferences n taxaton and compensaton for default, call and marketablty rsk on mortgages. Usng a lnear regresson they found that nterest rate volatlty and the term structure of rates have grown n mportance n recent years as exercse of the prepayment opton has ncreased. They also found evdence that lqudty and credt concerns affect the prcng of pass-through securtes. 3. Emprcal Analyss 3.1. Data descrpton Data used n ths work for Asset-Backed securtes are publshed by BNP Parbas, whle data for corporate floatng bonds are provded by Bloomberg. Both types of securtes are ssued n Europe and are ndexed to 3 or 6 months Eurbor and to 3 months Lbor; ratng categores are AA, A, BBB, BB, followng S&P s notaton. Many ABSs are trple A rated, but t s dffcult to fnd European corporate-floatng bonds rated AAA, so n the regresson analyss ABSs rated AAA are excluded. The tme perod covered s for corporate bonds, whle for ABS data prevous to 2001 (second quarter) were not avalable. The full ABS sample ncludes 448 securtes (101 are AAA - rated), whle corporate bonds ncluded n the analyss are For each bond we needed spread, sze of the ssue, maturty, ssue date, ratng, dex-rate, type of ABS. Followng Mars and Segal (2002), data for each securty are ssue date (n partcular 4 In the followng regresson (secton 3.4, 3.5 and 3.6) some bonds are elmnated and some more are ncluded, so t wll be ponted out each tme n the approprate secton the proper sample sze. 5

6 spread s the launch spread). The category ABS ncludes ABS n a strct sense (leases, credt cards, auto loans, etc.) 5, CMBS, RMBS, CDO, CLO and CBO Descrptve analyss The frst am of ths paper s to test whether the theoretcal relaton between spread and ratng holds for our data set. In fact theory predcts that lower rated bonds should offer hgher spreads snce they are rsker than hgher rated ones, whch should, thus, have lower yelds. Evdence supports ths argument n lterature 6 and also n our sample the relaton holds: let us look at fgure 1; the graphs depct ratng on the horzontal axs and spread (n bass ponts) n the vertcal axs. Fgure 1: Relaton between average spreads and credt ratng and comparson between average spreads for corporate bonds and ABSs. COMPARISON BETWEEN AVERAGE SPREADS AVERAGE SPREADS (b.p.) ABS_3ME CORP_3ME AAA AA A BBB BB RATING COMPARISON BETWEEN AVERAGE SPREADS 200 AVERAGE SPREAD (b.p.) ABS_3ML CORP_3ML AAA AA A BBB RATING COMPARISON BETWEEN AVERAGE SPREADS AVERAGE SPREADS (b.p.) ABS_6ME CORP_6ME AAA AA A BBB RATING 5 When a dstncton s needed, ABS n a strct sense are labelled classcal ABS. 6 For example see Sarg and Warga (1989); Mars and Segal (2002). 6

7 The fgure shows that the average spread for each ratng category wthn the same ndex group grows as credt qualty worsens; ths holds for corporate bonds (blue bars) as well as for Asset-Backed Securtes (yellow bars). The same relaton holds also for the aggregated sample as well that s wthout dstngushng between ndex-rate groups. Fgure 1 s useful also to compare spreads on ABSs and on Corporate bonds holdng for the ratng class and ndex-rate. As one can easly notce, ABSs spreads (represented by the yellow bars) are always much hgher than spreads on corporate bonds (depcted as the blue bars), holdng for credt ratng. We can say that there s an excess spread on ABSs compared to corporate bonds, whch must be due to factors other than those that affect corporate bonds yeld. Fgure 2 depcts the excess spread.e. the dfference between the spread on ABSs and the spread on corporate bonds. In the horzontal axs credt ratng s reported, whle the vertcal axs shows credt spread n bass ponts. As we can see, the curve has a postve slope ndcatng that the dfference n spreads grows at a hgher pace when ratng gets worse. So low-rated ABSs has a hgher excess spread than hgher-rated ones. An explanaton may be the followng: snce a goal of securtzaton s to reach a hgher ratng than that of the orgnator (so that the orgnator can fund at a lower cost), senor tranches have the best underlyng assets, whle subordnated tranches are less attractve and so must offer a hgher spread to be subscrbed by nvestors. Fgure 2: ABSs excess spread. EXCESS SPREAD (ABS-CORP) AA A BBB BB We can now try to dscuss some possble explanatons for the excess spread on ABSs. 7

8 3.3. Explanatons for the excess spread Ratng agences evaluate the credt rsk of an ssue and so classfy bonds nto dfferent rsk classes. Assumng that ratng s relable, two dfferent securtes wthn the same ratng category should be exposed to credt rsk n the same way. When one assgns ratng to a corporate bond, he consders the varables that capture the capacty of the ssuer to fulfl hs commtments. However to evaluate an ABS ssue t s necessary to deeply examne the structure of the transacton, because the credt qualty of the bond does not (entrely) depends on the credt qualty of the ssuer but on the qualty of the underlyng assets. In partcular economc and fnancal performances of the assets are analysed (geographcal and ndustry dversfcaton, cash-flows, tranche allocaton, etc.); dfferent stress scenaros are studed n order to calculate default probablty and recovery rates; fnally the legal structure and credt enhancement of the transacton are analysed. Even f we assume that ratng exactly reflects the credt rsk of a bond, to conclude that dfferent types of securtes wth the same ratng have the same prce, we should assert that credt rsk s the unque rsk to whch bonds are exposed. For ths s not true, we must say that dfferent types of securtes wth the same ratng are not necessarly prced the same. So we can say that there s a part of spread that s the same for the two categores of bonds holdng for credt ratng (and ths part of the spread s due to credt rsk), whle the remanng spread the excess spread shown n fgure 2 s not caused by credt rsk and must be explaned ntroducng other factors. These factors can be summarsed by () the dffculty of evaluatng a securtzaton transacton; the reasons for ths dffculty s the complexty of the transacton, () the presence of some nformatonal problems and () the lack of experence of the market. In fact, securtzaton s dffcult to evaluate snce t s a complcated transacton that nvolves many partes and a large number of contracts (and thus several small sub-transactons ). Ths mples that experence and knowledge s needed n order to understand securtzaton and to asses the rght prce to the securtes ssued. Moreover we can say that there s a nformatonal problem: whle nformaton about ssuer of corporate bonds s easly accessble and qute prompt to be nterpreted, nformaton about the credt qualty of all the assets underlyng a securtzaton transacton s not easy to collect and not smple to evaluate. In fact ABSs are collateralsed by several assets (often thousands), so the subscrber should look for nformaton about all these assets and then study how the cash flow are allocated throughout the subordnated tranches; on the opposte a bondholder only 8

9 has to gan nformaton about the ssuer of the securtes, whch s a smpler and less costly process. The dffculty to evaluate ABSs ssues s also due to the relatve lack of experence of the market: t takes tme for experts to get confdent wth ths knd of fnancal product and then to evaluate t n a somewhat automatc way. In fact, although the legal framework provdes good guarantes to nvestors n order to avod the presence of structural rsks, there s not a adequate case hstores to be sure that law provsons wll be fulflled. In other words, f one subscrbes a corporate bond and the frm suffers default, law provsons and the jurdcal practce assures that bondholders are repad before stockholders (and thus there s no uncertanty about what nvestors have to expect f such an event happens). But what does t happen f a large proporton of the borrowers underlyng a securtzaton transacton go bankrupt? Legal and structural guarantes are clear n a theoretcal level but n practce cases lke ths dd not happen (at least not n a suffcent number) and so nobody can say whether the legal provsons wll be enough to assure nvestors from the happenng of such an event. So we can dentfy somethng lke a learnng by dong process, whch needs tme to be completed and whch should make spreads get lower as tme passes. A smlar argument can be found n Mars and Segal (2002) about Amercan CMBS: they argue that nvestors (and ratng agences) were unfamlar wth analysng commercal mortgages and rsks they posed. Lack of famlarty mght have caused nvestors to overestmate the rsks, and could explan why yeld spreads were hgh n the early 1990s.As nvestors ganed confdence wth the product they developed a greater ablty to asses the rsks and so requred lower rsk prema Emprcal analyss of factors that nfluence the excess spread We can now try to determne factors that affect the excess spread usng the data set descrbed n secton 3.1. Securtes employed n the regresson analyss are 444, among whch 347 are ABSs 7 and 97 are corporate bonds. Lnear regresson s estmated wth the Ordnary Least Squares method usng the spread as dependent varable, whle usng a number of j regressors as ndependent varables, as descrbed below. In partcular, followng Mars and Segal (2002), we ntroduce varables that consder the sze of the ssue, the ratng, the ssue date are classcal ABSs; 141 are CDOs; 127 are RMBSs. 9

10 The frst varable s CORP_ABS, whch dscrmnates corporate bonds from ABSs; t s equal to 0 f the securty s ssued by a frm and t has value 1 f the note s ssued by a securtzaton transacton. We expect ths varable to have a postve coeffcent, ndcatng that Asset-Backed Securtes have hgher spreads compared to corporate bonds wthn the same ratng class and holdng for the other factors. T1, T2 and T3 are dummy varables as well and are defned as follows: 1, f the bond s ndexed to 3 months T 1 = 0, otherwse 1, f the bond s ndexed to 3 months T 2 = 0, otherwse 1, f the bond s ndexed to 6 months T 3 = 0, otherwse Eurbor Lbor Eurbor These dummes eventually capture the dfferences n spread due to the dfferent ndex rate of the bond. Ratng s also represented by dummy varables, R1, R2, R3, R4: 1, f the bond s rated AA R 1= 0, otherwse R 2 R 3 R 4 1, f the bond s rated A = 0, otherwse 1, f the bond s rated BBB = 0, otherwse 1, f the bond s rated BB = 0, otherwse In order to avod the sngularty of the estmaton matrx, we must exclude from the regresson one varable of each dummy-group; for the ratng group we chose to elmnate R4. R4 represents the lowest-ratng securtes n the sample ( BB ) and the coeffcents of the remanng dummes must be nterpreted as the average dfference between spread on, respectvely, the bonds rated AA, A and BBB and the spread on the dummy excluded; hence we expect the coeffcents of R1, R2 and R3 to be negatve and decreasng n ther absolute value. We argue that the excess spread of Asset-Backed Securtes depends on the ratng of each bond, so we nclude n the regresson other four varables n order to capture the dfferences n 10

11 excess spread due to the ratng class of the securty: CORPABS_R1, CORPABS_R2, CORPABS_R3, CORPABS_R4. They are constructed, respectvely, as the varable CORP_ABS tmes R1, R2, R3, R4; excludng CORPABS_R4 from the regresson, we expect CORPABS_R1-2-3, to be statstcally sgnfcant and to have negatve coeffcent, decreasng n ther absolute value, as for R1, R2, R3. Another explanatory varable s SIZE,.e. the sze of the ssue expressed as thousands of Euro. The greater the sze, the greater the lqudty n the market, so the lower the spread should be. So we expect the coeffcent of sze to be negatve, as found n Mars and Segal (2002). Another explanatory varable s MATURITY whch s defed as follows: for corporate bonds t represents the fracton of year whch elapses from the ssue date to the maturty date; for ABS t represents the Weghted Average Lfe (WAL). WAL s the average number each dollar of unpad prncpal due on the mortgage remans outstandng; t s computed as the weghted average tme to the recept of all future cash flows weghts the dollar amounts of the prncpal pay-downs. WAL s generally used by market operators whle dealng wth ABSs and other structured-fnance products, hence also n ths work WAL s used. The longer the term to maturty, the greater the uncertanty about the future, so the greater the rskness s. Thus we expect a postve estmated coeffcent for maturty. NWMKT s a group of dummes whch dscrmnate securtes consderng the ssung quarter of the year. NWMKT1 s equal to 1 f the securty was ssued n Aprl, May or June 2001; 0 otherwse. NWMKT2 s equal to 1 f the ssue date of the bond s July, August or September 2001; 0 otherwse. NWMKT3 equals 1 f the bond was ssued n the forth quarter of 2001; 0 otherwse. NWMKT4 and NWMKT5 are equal to 1 f the bond was ssued, respectvely, n the frst or second quarter of 2002; they are zero otherwse. In ths way we can capture also non-lnear relatonshps between tme and spread. The varable s zero for all corporate bonds. NWMKT1 s not ncluded n the regresson n order to avod the sngularty of the estmaton matrx, hence the coeffcents of NWMKT must be nterpreted as average spreads of bonds ssued, respectvely, n the thrd, forth quarter of 2001 and frst or second quarter of 2002 n excess compared to the excluded category s spread. The later a securtes was ssued, the lower the spread on t should be, because nvestors should have acqured a hgher confdence wth ABSs evaluaton and so they should ask for lower spreads. Therefore we expect a negatve coeffcent for Nwmkt1-5 (whch means that, for example, 11

12 a bond ssued n the second quarter of 2002 should have lower spreads compared to securtes ssued on the second quarter of 2001 that s the excluded category). The estmated equaton s: (1) SPREAD = β + β CORP_ ABS + β T1 + β T2 + β R1 + β R2 + β R3 0 + β CORPABS_ R1 + β CORPABS_ R2 7 + β + β SIZE + β NWMKT4 11 SCADENZA + β + β NWMKT5 + ε β CORPABS_ R3 NWMKT2 + β NWMKT3 6 where ndcates the -th securty. T3, R4, CORPABS_R4 and NWMKT1 are excluded from equaton 1 to avod the estmaton matrx to become sngular Results Results are shown n table 1. The frst column reports the explanatory varables. The second column shows the estmated coeffcent for each varable. Standard error s reported n column three and column four exhbts the t-statstcs (calculated as the rato of the estmated coeffcent and the related standard error). The last column dsplays the p-value; the last two rows report R 2 and ts adjusted verson. We can frst note that the regresson explan almost 68% of the excess spread whle remanng 32% s not captured by the varables. From table 1 we can frst notce that R1, R2 and R3 are hghly sgnfcant and ther coeffcents has the expected features: they are negatve and decreasng n absolute value when ratng worsens. So the results ndcate that bonds rated AA have spreads 254 b.p. hgher on average than bonds rated BB ; smlar arguments hold for R2 and R3. The coeffcent of CORPABS_R1 represents the average excess spread between AA-rated Asset-Backed Securtes and AA-rated corporate bonds compared to the category excluded (bonds rated BB ), holdng for all the other condtons; smlarly, the coeffcents of CORPABS_R2 and CORPABS_R3 represent the average spread, respectvely, between A- rated ABSs and A-rated corporate bonds and between BBB-rated Asset-Backed Securtes and BBB-rated corporate bonds, holdng for the other condtons. All these coeffcents are statstcally sgnfcant and have the expected sgns. Ths does not necessarly mply that for all ratng categores ABSs have hgher spreads compared to corporate bonds; n order to 12

13 verfy whether the excess spread holds for all ratng groups, we employed three coeffcent tests. The null hypothess are: TEST 1 H 0 : β 1 + β 7 = 0; TEST 2 H 0 : β 1 + β 8 = 0; TEST 3 H 0 : β 1 + β 9 = 0. If we accept the null hypothess for the frst test, we can conclude that AA-rated ABSs have not hgher average spreads than AA-rated corporate bonds. On the opposte rejectng H 0 for test 1 mples to confrm the excess spread between ABSs and corporate bonds for AA-ratng category. Smlar arguments hold for test 2 and 3, that s for A and BBB groups, respectvely. Therefore f we accept H 0 for all the tests we can conclude that the overall average excess spread s entrely determned by the average excess spread on BB-rated securtes. We reject H 0, at a 5% sgnfcance level, f the p-value s lower than The results of the tests show that p-value s for the frst test; for the second test; for the thrd test. Therefore we can conclude that the excess spread does exsts only for securtes rated BBB (and for BB-rated bonds), but evdence suggest that for the groups AA and A Asset-Backed Securtes have not hgher spreads than corporate bonds. The sze and the maturty of the bonds do not seem to nfluence the spread, snce the coeffcents of the two varable are not sgnfcant. Moreover T1 s not statstcally sgnfcant, whle T2 s weakly sgnfcant; ts coeffcent s negatve, ndcatng that 3-months-Lbor bonds have lower spreads compared to 6-month-Eurbbor (represented by T3,.e. the excluded dummy). The results do not provde evdence for a tme dependence of the spreads: only the coeffcents for NWMKT3 and NWMKT5 are sgnfcant 8 and the value of the coeffcents of entre group does not show a clear postve or negatve trend. Ths result contrast wth the one obtaned by Mars and Segal (2002) wth US data. A potental explanaton s that the European market s new and so only after fve or ten years we could expect to observe ths effect. It mght be nterestng to evdence how mportant each regressor s n determnng the excess spread. To do so we can estmate agan equaton (1), excludng one varable and repeat ths procedure as many tmes as the number of the varables. In such a way we can see how much the R 2 reduces when a varable s not ncluded n the regresson; the lower the R 2, the more 13

14 mportant the excluded varable s. Fgure 3 shows the dfference between R 2 n the total regresson and R 2 n the j-1 regresson. It s easy to notce that ratng, corp_abs and corp_abs*r1-2-3 are the most mportant varables. Table 1: Estmated coeffcents for equaton (1). The regresson has SPREAD (n bass ponts) as dependent varable, whle the followng varables are ntroduced as explanatory varables. CORP_ABS dscrmnates corporate bonds from ABSs; t s equal to 0 f the securty s ssued by a frm and t has value 1 f the note s ssued by a securtzaton transacton. T1 s equal to 1 f the bond s dexed to 3-months Eurbor; 0 otherwse. T2 s equal to 1 f the securty s dexed to 3-months Lbor; 0 otherwse. R1 dentfes AA-rated bonds; R2 represents A-rated securtes and R3 ndecates BBB-rated bonds. CORPABS_R1 s equal to CORP_ABS*R1; CORPABS_R2 s equal to CORP_ABS*R2; CORPABS_R3 s equal to CORP_ABS*R3. NWMKT s a group of dummes whch dscrmnate securtes consderng the ssung quarter of the year. NWMKT2 dentfes bonds ssued n the thrd quarter of 2001; NWMKT3 represents bonds ssued n the forth quarter of 2001; NWMKT4 dentfes securtes ssued n the frst quarter of 2002; NWMKT5 ndcates bonds ssued n the second quarter of SIZE s the sze of the ssue expressed as thousands of Euro. MATURITY s defed as follows: for corporate bonds t represents the fracton of year whch elapses from the ssue date to the maturty date; for ABS t represents the Weghted Average Lfe (WAL). Varable Coeffcent Std. Error Prob. C * CORP_ABS * T T * R * R * R * CORPABS_R * CORPABS_R * CORPABS_R * SIZE -2.03E E SCADENZA NWMKT NWMKT * NWMKT NWMKT * R-squared Adjusted R- squared NWMKT5 s weakly sgnfcant. 14

15 Fgure 3: Margnal contrbute of each explanatory varable. Nwmkt Maturty Sze Corp_abs_ R1-2-3 Ratng Index-rate corp_abs 0 0,02 0,04 0,06 0,08 0, The subsamples: classcal ABSs, CDOs and RMBSs. It mght be worth to analyse what happens f we regress only selected types of Asset-Backed Securtes and corporate bonds n order to understand whether the type of ABSs nfluence credt spread; ths s exactly what we do n ths secton. Before presentng the regresson and the results, t s worth to say that not all the types of Asset-Backed Securtes are charactersed by the same level of newness. In fact RMBS are the frst type of securtes ssued by a securtzaton transacton n the Unted States n the md 1970s, so nvestors should have acqured more experence and confdence wth ths knd of bonds and less wth more recently created products of the structured fnance. A smlar argument holds for classcal ABSs, such as the notes ssued by securtzaton of leasng, auto loans, recevables, etc. So we now try to regress separately classcal ABSs, CDOs and RMBSs wth corporate bonds; fnally we compare classcal ABSs and RMBSs Classcal ABSs The frst sub-sample analysed s classcal ABSs. The observatons nclude 176 bonds, among whch 97 are corporate and the remanng 79 are Asset-Backed Securtes. The estmated equaton s equaton (1), as the regresson mplemented n secton 3.4. Thus consderatons about the explanatory varables are the same as n secton 3.4. Moreover smlar consderaton as n secton 3.5 can be made for almost all the varables. In fact the 15

16 greatest dfferences between table 1 and table 2 deal wth the varables NWMKT : none of these are statstcally sgnfcant. Three tests smlar to the ones exposed n secton 3.5 are mplemented for the sub-sample classcal ABSs. The results are: p-value equal to for test one; p-value equal to for test 2; p-value equal to 0 for the thrd test. Therefore we can reject the null hypothess for two of the ratng categores A and BBB and conclude that evdence provdes support for the exstence of a postve excess spread between ABSs and corporate bonds for these two groups. On the other hand the frst test makes us argue that AA group has not an excess spread between the two types of bonds. Table 2: Estmated coeffcents for equaton (1) for the subsample classcal ABSs. The regresson has SPREAD (n bass ponts) as dependent varable, whle the followng varables are ntroduced as explanatory varables; only classcal ABSs and corporate bonds are ncluded n the analyss. CORP_ABS dscrmnates corporate bonds from ABSs; t s equal to 0 f the securty s ssued by a frm and t has value 1 f the note s ssued by a securtzaton transacton. T1 s equal to 1 f the bond s dexed to 3-months Eurbor; 0 otherwse. T2 s equal to 1 f the securty s dexed to 3-months Lbor; 0 otherwse. R1 dentfes AArated bonds; R2 represents A-rated securtes and R3 ndcates BBB-rated bonds. CORPABS_R1 s equal to CORP_ABS*R1; CORPABS_R2 s equal to CORP_ABS*R2; CORPABS_R3 s equal to CORP_ABS*R3. NWMKT s a group of dummes whch dscrmnate securtes consderng the ssung quarter of the year. NWMKT2 dentfes bonds ssued n the thrd quarter of 2001; NWMKT3 represents bonds ssued n the forth quarter of 2001; NWMKT4 dentfes securtes ssued n the frst quarter of 2002; NWMKT5 ndcates bonds ssued n the second quarter of SIZE s the sze of the ssue expressed as thousands of Euro. MATURITY s defed as follows: for corporate bonds t represents the fracton of year whch elapses from the ssue date to the maturty date; for ABS t represents the Weghted Average Lfe (WAL). Varable Coeffcent Std. Error Prob. C * CORP_ABS * T T * R * R * R * CORPABS_R * CORPABS_R * CORPABS_R * SIZE -2.03E E MATURITY NWMKT NWMKT NWMKT NWMKT R-squared Adjusted R-squared

17 3.6.2 RMBSs In ths secton 127 RMBSs and 97 corporate bonds are analysed; the regresson equaton s the same used n secton 3.4 and The results, reported n table 3, are smlar as well; n partcular none of the varables NWMKT s statstcally sgnfcant. Sze, Maturty, T1 and T2 are not sgnfcant whle CORP_ABS, R1-2-3 and CORPRMBS_R1-2-3 are hghly sgnfcant. Interpretaton of the coeffcents of these varables s smlar to that exposed n secton 3.5. Table 3: Estmated coeffcents for equaton (1) for the subsample RMBSs. The regresson has SPREAD (n bass ponts) as dependent varable, whle the followng varables are ntroduced as explanatory varables; only RMBSs and corporate bonds are ncluded n the analyss. CORP_ABS dscrmnates corporate bonds from ABSs; t s equal to 0 f the securty s ssued by a frm and t has value 1 f the note s ssued by a securtzaton transacton. T1 s equal to 1 f the bond s dexed to 3-months Eurbor; 0 otherwse. T2 s equal to 1 f the securty s dexed to 3-months Lbor; 0 otherwse. R1 dentfes AArated bonds; R2 represents A-rated securtes and R3 ndcates BBB-rated bonds. CORPABS_R1 s equal to CORP_ABS*R1; CORPABS_R2 s equal to CORP_ABS*R2; CORPABS_R3 s equal to CORP_ABS*R3. NWMKT s a group of dummes whch dscrmnate securtes consderng the ssung quarter of the year. NWMKT2 dentfes bonds ssued n the thrd quarter of 2001; NWMKT3 represents bonds ssued n the forth quarter of 2001; NWMKT4 dentfes securtes ssued n the frst quarter of 2002; NWMKT5 ndcates bonds ssued n the second quarter of SIZE s the sze of the ssue expressed as thousands of Euro. MATURITY s defed as follows: for corporate bonds t represents the fracton of year whch elapses from the ssue date to the maturty date; for ABS t represents the Weghted Average Lfe (WAL). Varable Coeffcent Std. Error Prob. C CORP_RMBS T T R R R CORPRMBS_R CORPRMBS_R CORPRMBS_R SIZE 6.22E E SCADENZA NWMKT NWMKT NWMKT NWMKT R-squared Adjusted R-squared

18 The coeffcent tests were carred out n the way already explaned n the prevous sectons. P- values for the frst and second tests are, respectvely, and , whch make us accept the null hypothess. The thrd test s charactersed by a p-value equal to , mplyng that the null hypothess s rejected. So we can agan conclude that the exstence of the excess spread s verfed only for lower-rated bonds CDOs The last sub-sample that has been analysed n ths paper s CDOs. The sample ncludes 141 CDOs and 97 corporate bonds. The estmaton output s summarsed n table 4. As we can easly notce, the results are the same as the prevous analyss except for the regressors NWMKT. In fact three of the four varables n the group are sgnfcant, provdng evdence for a relaton between spread and tme. However, as underlned n secton 3.5, the coeffcents of these varables does not show a clear trend ether postve nor negatve. The coeffcent tests agan support the dea that the dfference n spread between securtes ssued n a securtzaton transacton and corporate bonds does exsts only for low-rated securtes 9. Table 4: Estmated coeffcents for equaton (1) for the subsample CDOs. The regresson has SPREAD (n bass ponts) as dependent varable, whle the followng varables are ntroduced as explanatory varables; only CDOs and corporate bonds are ncluded n the analyss. CORP_ABS dscrmnates corporate bonds from ABSs; t s equal to 0 f the securty s ssued by a frm and t has value 1 f the note s ssued by a securtzaton transacton. T1 s equal to 1 f the bond s dexed to 3-months Eurbor; 0 otherwse. T2 s equal to 1 f the securty s dexed to 3-months Lbor; 0 otherwse. R1 dentfes AA-rated bonds; R2 represents A-rated securtes and R3 ndcates BBB-rated bonds. CORPABS_R1 s equal to CORP_ABS*R1; CORPABS_R2 s equal to CORP_ABS*R2; CORPABS_R3 s equal to CORP_ABS*R3. NWMKT s a group of dummes whch dscrmnate securtes consderng the ssung quarter of the year. NWMKT2 dentfes bonds ssued n the thrd quarter of 2001; NWMKT3 represents bonds ssued n the forth quarter of 2001; NWMKT4 dentfes securtes ssued n the frst quarter of 2002; NWMKT5 ndcates bonds ssued n the second quarter of SIZE s the sze of the ssue expressed as thousands of Euro. MATURITY s defed as follows: for corporate bonds t represents the fracton of year whch elapses from the ssue date to the maturty date; for ABS t represents the Weghted A verage Lfe (WAL). Varable Coeffcent Std. Error Prob. C CORP_CDO T T R In fact the frst and second tests have a p-value equal to , whle the p-value of the thrd one s nearly zero (0.0009). 18

19 R R CORPCDO_R CORPCDO_R CORPCDO_R SIZE -1.66E E SCADENZA NWMKT NWMKT NWMKT NWMKT R-squared Adjusted R-squared RMBS and CABS: a comparson In the prevous sectons corporate bonds were compared to securtes ssued by securtzaton transactons; we would lke now to nvestgate whether and how an excess spread exsts between RMBSs and classcal ABSs and/or between RMBSs and CDOs. In fact, as sad above, RMBSs were the frst type of Asset-Backed Securtes to be ssued, so we mght ague that RMBSs have lower spreads than classcal ABSs and n partcular lower spreads than CDOs ths would confrm that the newness of the securty does matter for the spread requred by the market. Hence, the frst analyss comprses 127 RMBSs and 79 classcal ABSs. The explanatory varables are the same as n equaton (1), however RMBS_ABS s defned n a somewhat dfferent way: t s equal to zero f the bond s a RMBS, whle t s equal to 1 f the securty s a classcal ABS. In such a way the varable dscrmnates RMBSs from classcal ABSs and ts coeffcent gves a measure of the average excess spread f t exsts between the two types of securtes. Estmaton output s reported by table 5. Results suggest that there s an excess spread between RMBSs and classcal ABSs, snce the coeffcent of RMBS_ABS s postve and statstcally sgnfcant. Moreover RMBSABS_R1-2-3 are sgnfcant as well, therefore, as n the prevous sectons, three coeffcent tests are run. The results make us accept the null hypothess n all the case, meanng that for the ratng groups AA, A, and BBB the excess spread s not statstcally sgnfcant. We can, thus, conclude that the excess spread only exsts for lower rated bonds labelled BB whch are the excluded category and are ncorporated n the coeffcent of RMBS_ABS. NWMKT are not statstcally 19

20 sgnfcant, mplyng that spread does not seem to be affected by the ssue date of the bond. Surprsngly SIZE s weakly sgnfcant and has a negatve estmated coeffcent; ths means that the market requres slghtly lower spreads for bonds ssued n a large sze. Table 5: Estmated coeffcents for equaton (1) for RMBSs and classcal ABSs. The regresson has SPREAD (n bass ponts) as dependent varable, whle the followng varables are ntroduced as explanatory varables; only RMBSs and classcal ABSs are ncluded n the analyss. CORP_ABS dscrmnates RMBSs from classcal ABSs; t s equal to 0 f the securty s ssued by a frm and t has value 1 f the note s ssued by a securtzaton transacton. T1 s equal to 1 f the bond s dexed to 3- months Eurbor; 0 otherwse. T2 s equal to 1 f the securty s dexed to 3-months Lbor; 0 otherwse. R1 dentfes AA-rated bonds; R2 represents A-rated securtes and R3 ndcates BBB-rated bonds. CORPABS_R1 s equal to CORP_ABS*R1; CORPABS_R2 s equal to CORP_ABS*R2; CORPABS_R3 s equal to CORP_ABS*R3. NWMKT s a group of dummes whch dscrmnate securtes consderng the ssung quarter of the year. NWMKT2 dentfes bonds ssued n the thrd quarter of 2001; NWMKT3 represents bonds ssued n the forth quarter of 2001; NWMKT4 dentfes securtes ssued n the frst quarter of 2002; NWMKT5 ndcates bonds ssued n the second quarter of SIZE s the sze of the ssue expressed as thousands of Euro. MATURITY s defed as follows: for corporate bonds t represents the fracton of year whch elapses from the ssue date to the maturty date; for ABS t represents the Weghted Average Lfe (WAL). Varable Coeffcent Std. Error Prob. C * RMBS_ABS * T * T R * R * R * RMBSABS_R * RMBSABS_R * RMBSABS_R * SIZE -2.59E -05* 1.07E MATURITY NWMKT NWMKT NWMKT * NWMKT R-squared Adjusted R-squared In the second analyss 127 RMBSs are compared to 141 CDOs. The estmated equaton s the same as n the prevous sectons equaton (1), however the varable RMBS_CDO s defned n a way smlar to RMBS_ABS : t s equal to zero f the bond s a RMBS, whle t s equal to 1 f the securty s a CDO. Table 6 shows that the estmated coeffcent for RMBS_CDO s postve and statstcally sgnfcant, ndcatng that CDOs have hgher 20

21 spreads than RMBSs. Moreover, snce RMBSCDO_R1-2-3 are not sgnfcant, meanng that the dstncton between the ratng class s not mportant. Only NWMKT3 and NWMKT5 are sgnfcant n the group and the coeffcents do not suggest a lnear trend or relaton between tme and spread. Table 6: Estmated coeffcents for equaton (1) for RMBSs and CDOs. The regresson has SPREAD (n bass ponts) as dependent varable, whle the followng varables are ntroduced as explanatory varables; only RMBSs and CDOs are ncluded n the analyss. CORP_ABS dscrmnates RMBSs from classcal ABSs; t s equal to 0 f the securty s ssued by a frm and t has value 1 f the note s ssued by a securtzaton transacton. T1 s equal to 1 f the bond s dexed to 3-months Eurbor; 0 otherwse. T2 s equal to 1 f the securty s dexed to 3-months Lbor; 0 otherwse. R1 dentfes AA-rated bonds; R2 represents A-rated securtes and R3 ndcates BBB-rated bonds. CORPABS_R1 s equal to CORP_ABS*R1; CORPABS_R2 s equal to CORP_ABS*R2; CORPABS_R3 s equal to CORP_ABS*R3. NWMKT s a group of dummes whch dscrmnate securtes consderng the ssung quarter of the year. NWMKT2 dentfes bonds ssued n the thrd quarter of 2001; NWMKT3 represents bonds ssued n the forth quarter of 2001; NWMKT4 dentfes securtes ssued n the frst quarter of 2002; NWMKT5 ndcates bonds ssued n the second quarter of SIZE s the sze of the ssue expressed as thousands of Euro. MATURITY s defed as follows: for corporate bonds t represents the fracton of year whch elapses from the ssue date to the maturty date; for ABS t represents the Weghted Average Lfe (WAL). Varable Coeffcent Std. Error t-statstc Prob. C RMBS_CDO T T R R R RMBSCDO_R RMBSCDO_R RMBSCDO_R SIZE -7.81E E MATURITY NWMKT NWMKT NWMKT NWMKT R-squared Adjusted R-squared Concluson 21

22 In ths paper credt spread on corporate bonds and on Asset-Backed Securtes are compared. A descrptve analyss shows that ABSs have hgher spread than corporate bonds; we called excess spread the dfference between spread on ABSs and spread on corporate bonds. Further analyss was mplemented, usng lnear regresson method. However the presence of excess spread s confrmed only for low rated securtes, whle for the categores AA and A Asset-Backed Securtes do not seem to have statstcally sgnfcant hgher spreads than bonds ssued by frms. Ths means that the market requres the same remuneraton to nvest n a bond of a sngle frm and n a bond on a portfolo wth the same ratng, f ths ratng s hgh. On the contrary, the market requres a dfferent remuneraton f the ratng s low. The reasons for ths are many and n partcular: () the dffculty of evaluatng a securtzaton transacton; () the presence of nformatonal problems and () the lack of experence of the market. In partcular, the asymmetres of nformaton generated by a portfolo are dfferent than the one generated by a sngle frm. In the paper we consdered both () RMBS and () CABS and ths phenomenon s confrmed for the retal as for the commercal portfolos. A potental nterpretaton of ths result s that the market do not consder as dfferent assets Retal MBS or Commercal ABS f the ratng s hgh. However, our analyss shows that the market requres a hgher return on Commercal ABS wth respect to RMBS when the ratng s low. The reason of ths result could be the hgher development of RMBS market and the hgher expertse on rsk measurement of retal portfolos. Moreover we argued that part of the excess spread could be due to the fact that ABSs are a new product n the European market, so nvestors could requre a hgher spread because they are not confdent wth ths knd of product. The requred spread should lower as nvestors gan famlarty wth ABSs; a smlar argument s found by Mars and Segal (2002). However the relaton between spread and tme s not supported by emprcal evdence provded n ths paper. Even f we regress spread separately for each ABSs type, we fnd no evdence of a tme dependence for spread. To be more precse, most of the varables that should capture the tmefactor for the sub-sample CDOs are sgnfcant, but a clear relaton between tme and spread cannot be found. So we must conclude that evdence does not support the dea that requred spread on ABSs gets lower as the learnng process s gong to complete, n other words we are not able to assert that nvestors ask for lower spreads on ABSs when they become more famlar wth ths product. 22

23 References ANNAERT J., DE CEUSTER M. (1999) Modellng European Credt Spreads Unverstet Antwerpen UFSIA BNP PARIBAS (2002) European ABS 2001 Forth Quarter Revew, Report BNP PARIBAS (2002) European ABS 2001 Second Quarter Revew, Report BNP PARIBAS (2002) European ABS 2001 Thrd Quarter Revew, Report BNP PARIBAS (2002) European ABS 2002 Frst Quarter Revew, Report BNP PARIBAS (2002) European ABS 2002 Second Quarter Revew, Report COLLIN-DUFRESNE P., GOLDSTEIN R., MARTIN S. (2001) The Determnants of Credt Spread Changes, Journal of Fnance, 56 ELTON E., GRUBER M., AGRAWAL D., MANN C. (2001) Explanng the Rate Spread on Corporate Bonds, Journal of Fnance, 56 HE J., HU W., LANG L. (2000) Credt Spread Curves and Credt Ratngs Workng Paper, Chnese Unversty of Hong Kong HUANG J., HUANG M. (2002) How much of the Corporate-Treasury Yeld Spread s due to Credt Rsk? A New Calbraton Approach, Workng Paper Pennsylvana State Unversty JACKSON P., PERRAUDIN W. (1999) The Nature of Credt Rsk: the Effect of Maturty, Type of Oblgor, and Country of Domcle Fnancal Stablty Revew (November) MARIS B., SEGAL W. (2002) Analyss of Yeld Spreads on Commercal Mortgage-Backed Securtes Journal Real Estate Research, vol. 23 ROTHBERG J., NOTHAFT F., GABRIEL S. (1989) On the Determnants of Yeld Spreads Between Mortgage Pass-Through and Treasury Securtes Journal of Real Estate Fnance and Economcs SARIG O., WARGA A. (1989) Some emprcal Estmates of the Rsk Structure of Interest Rates Journal of Fnance, 44(5) STANDARD AND POOR S (2002) Italan Banks Embrace Securtzaton downloadable from the nternet ste 23