Financial contracting and re-rating experience, the cases of make whole, claw back and other wise ordinary callable bonds. Frank S.

Transcription

1 Fnancal contractng and re-ratng experence, the cases of make whole, claw back and other wse ordnary callable bonds 1 Frank S. Sknner 1 *, Unversty of Surrey Dmtros Gounopoulos Unversty of Surrey Abstract The lterature states that ordnary callable bonds help resolve several agency ssues whle make whole and claw back bonds are supposed to resolve the undernvestment problem. The justfcaton for refnements of the classc call opton s not clear as frstly, the classc call opton seems able to handle the undernvestment problem anyway and secondly, the exstng emprcal evdence does not examne whether make whole and claw back bondholders actually beneft from the resoluton of agency problems. We fnd evdence that frms do select make whole and claw back provsons to better confront the undernvestment problem for below nvestment grade bonds. Keywords: Callable bonds, Claw back call provsons, whole make call provsons JEL classfcaton: G24, G32, G38, K12, K22 1 Frank Sknner s Professor of Fnance at the Unversty of Surrey School of Management, Unversty of Surrey, Guldford, Surrey, GU2 7XH, Unted Kngdom e-mal: f.sknner@surrey.ac.uk, tel , Fax Dmtros Gounopoulos s Lecturer of Fnance at the Unversty of Surrey. * We thank Panayots Andreou, Smone Varrotto and Peggy Huang for many helpful comments. We gratefully acknowledge Campbell Harvey. We would lke to thank meetng partcpants of the Multnatonal Fnance Assocaton 2009, the Fnancal Management Assocaton 2009 and European Fnancal Management Assocaton 2009 for helpful comments on earler versons of ths paper. Any errors are the authors responsblty. Correspondng Author: Frank Sknner.

2 Fnancal contractng and re-ratng experence, the cases of make whole, claw back and other wse ordnary callable bonds 2 1. Introducton In recent years many bond ndenture agreements nclude two new provsons, the make whole and the claw back provson that refne the crcumstances upon whch a bond can be redeemed pror to maturty. The make whole call prce represents the present value of all coupon and prncpal repayments. The call prce s determned by a dscount rate set as the yeld on a smlar maturty Treasury bond plus a fxed spread. Ths mples that the frm has lttle ncentve to refnance ts debt due to a fall n the level of nterest rates. Smlarly, the claw back provson reduces the ncentve to refnance n order to save on nterest costs by allowng the frm to call debt only from the proceeds of an equty ssue. We seek to understand the purpose of these new refnements of the ordnary call provson. Goyal et al. (1998) and Nayar and Stock (2008) fnd evdence that make whole and claw back bonds are justfed by allevatng agency problems such as the undernvestment problem that s often present for frms wth rsky prospects and debt n ts captal structure. Yet whle ths work shows that stockholders can beneft from the use of these provson t s not clear whether bondholders do. For example Nayar and Stock (2008) fnd that the stock of frms that ssue make whole bonds enjoy a sgnfcant favorable prce reacton to the ssue, have superor post ssue returns and analysts forecast hgher post ssue growth for frms ssung make whole bonds. Yet the lterature s slent on whether bondholders beneft. The stuaton s made more puzzlng because Banko and Zhou (2010) fnd that the classc call opton s used to resolve a combnaton of asymmetrc nformaton and undernvestment problems. Therefore the justfcaton for refnements of the classc call opton s not clear as frstly, the classc call opton seems able to handle the undernvestment problem anyway and secondly, the

3 3 exstng emprcal evdence does not examne whether make whole and claw back bondholders actually beneft from the resoluton of agency problems. Ths work s of nterest because we now understand that f a bond contract term appears to beneft shareholders ex post, t can well be the case that bondholders can antcpate these benefts and exproprate them n the ntal terms of the bond contract. For example, as t s well understood from Kraus (1973) that potental nterest cost savngs s not a vald reason for callable bonds snce any potental expropraton of wealth by re-ssung lower coupon bonds s antcpated by bondholders through hgher ntal coupon rates and call prces. One queston that we address s whether make whole and claw back provsons genunely mprove callable bond contracts by better handlng the undernvestment problem or can they be seen as an attempt to explot bondholders n a zero sum game? Our nsght s that f the make whole and claw back bonds represent a genune mprovement n dealng wth the undernvestment problem then the subsequent re-ratng of make whole and claw back bonds should be domnated by more upgrades and/or less downgrades than ordnary callable bonds. If the zero sum game explanaton holds true, then subsequent re-ratng experence should be no dfferent, or ndeed worse, that ordnary callable bonds so there s no evdent reason to refne the classc call opton contract. The undernvestment problem occurs because postve NPV projects look too rsky for debt holders. Whle the frm knows ther projects are good they have dffculty n convncng skeptcal bondholders. As a result the frm must pay hgher nterest costs than they should gven the qualty of ther projects. If the projects prove themselves bondholder s beneft at the expense of stockholders as they wll contnue to enjoy hgh coupon rates even though rsk s now revealed to be modest. Ths reduces the benefts that flows to the shareholders and weakens the nventve to nvest. Meanwhle the asymmetrc nformaton problem occurs because the frm s unable to reveal the value of future growth opportuntes wthout moral hazard so bondholders undervalue the frm.

4 4 To resolve the undernvestment problem the frm could ssue a callable bond. When the qualty of the projects s revealed the frm can exercse the call feature thereby replacng the hgh coupon callable bond wth a lower coupon bond. If n fact make whole and claw back bonds are used by frms to resolve the undernvestment problem then the frm s stock prce and growth prospects should mprove (Nayar and Stock (2008)), the ssung frm s credt rsk should decrease and the ratngs of make whole and claw back bonds should ncrease. If n fact make whole and claw back bond re-ratng experence s superor to ordnary callable bond re-ratng experence then ths ndcates that they are justfed by refnng the classc call opton contract to better handle the undernvestment problems faced by frms that use them. An dea of the complexty of the ssue can be gleaned from Table 1 that reports the salent quanttatve detals of the call provsons of our sample. 2 Table 1 show that the prce for nvokng a call provson of any type generally ncreases as the credt ratng decreases. Table 1 also hghlghts that the call protecton perod for ordnary calls s substantal, approxmately three and a half years on average, whereas for make whole bonds t s vrtually non exstent, usually less than three months. Meanwhle the claw back provson s operatve upon ssue. Clearly nvestors are offered much more protecton from ordnary call provsons than they are for claw back and make whole provsons. 3 On the other hand controllng for credt ratng the average call prce for ordnary call provsons s modest whereas substantal premums are offered for the exercse of a make whole or a claw back provson. Therefore t s unclear f bondholders vew claw back and make whole bond provsons more favorably than ordnary call provsons because whle evdently claw back and make whole provsons requre a shorter protecton perod, they do requre more compensaton f exercsed. 2 The sample sze of Table I s often less than the correspondng sample szes n Tables II, III and IV as detaled later because detals of some of the call provsons are mssng from the Mergent fxed ncome database. 3 Interestngly ths hnts at the crcumstances upon whch the claw back and make whole bonds can be better suted to resolve the undernvestment problem. A frm that has an undervalued project that wll shortly be revealed as good when the frst results are n would desre a type of call provson than can be mmedately exercsed.

5 5 << Table 1 about here >> We fnd nterestng dfferences n the use of make whole and claw back provson bonds. Make whole bonds are used for all credt grades whereas claw back bonds are used almost exclusvely for bonds rated below nvestment grade. Moreover ordnary callable bonds are most frequently ssued by fnancal frms whereas make whole and claw back bonds are most frequently ssued by ndustral frms. Importantly, after applyng controls for credt ratng, ndustry, foregn status, bond covenants (put, snkng fund, securty level and converson features), prvate placement and exchange lstng status, maturty, bond market condtons and ssung frm characterstcs (sze, leverage, lqudty, growth and proftablty), we fnd that below nvestment grade make whole and claw back bonds have a sgnfcantly hgher lkelhood to upgrade and below nvestment grade claw back bonds have a sgnfcantly lower lkelhood to downgrade than ordnary callable bonds. In other words, we fnd evdence that make whole and claw back call provsons genunely confront the undernvestment problem better than ordnary callable bonds for the ratng grades where agency problems tend to be most sever. 2. Agency Cost Explanatons of Callable Bonds The lterature suggests that three types of call provsons can resolve the undernvestment problem, the classc call, make whole and claw back call provsons. A classcal call opton empowers the ssuer to take advantage of bondholders by repayng the debt n advance when market yelds declne. When nterest rates decrease, the classcal call opton settlement amount s less than what the far value of a debt would have been absent the call opton. Followng Kraus (1973) fnance has largely rejected nterest cost savngs as an explanaton for call provsons snce n an effcent market gans to shareholders va refnancng at lower nterest rates would be antcpated and

6 6 exproprated by bondholders n the terms of the ntal call provson. Recently Banko and Zhou (2010) fnd that ordnary call provsons are useful for hedgng nterest rate uncertanty but only for nvestment grade bonds. Instead, n the man, call provsons are supposed to exst because of agency problems such as the undernvestment, asymmetrc nformaton and asset substtuton problems. More to the pont, to resolve the undernvestment problem, classc callable bonds can be called once the value of the frm s projects s revealed. Bode and Taggart (1978) state that even when nterest rates are such that bondholders are ndfferent between callable and non-callable bonds, the frm s rsky projects can gve shareholders a defnte preference for callable bonds. If ths explanaton holds then all else equal callable bonds should experence more upgrade and less downgrade credt events as the good results of the frm s projects are subsequently revealed. Thatcher (1985), Ksh and Lvngston (1992), Boreko and Lombardo (2008), Joron and Zhang (2010) and Esdorfer (2011) all suggest that agency explanatons as a whole can explan the use of call provsons. However, Crabbe and Helwege (1994) fnd that the confoundng effects of maturty, default rsk and varyng trends n the popularty of call provsons make t dffcult to emprcally verfy f any of the specfc agency problems can explan the use of ordnary call provsons. Banko and Zhou (2010) reply to Crabbe and Helwege (1994) fndng that ordnary call provsons are useful n dealng wth a combnaton of undernvestment and asymmetrc nformaton problems rather than each agency problem n solaton. Addtonally several authors provde explanatons why frms can employ a sub optmal call polcy. Mauer (1993) fnds that frms can delay calls due to transactons costs, Longstaff and Tuckman (1994) and Mtto and Zhang (2010) explans that wealth transfers resultng from temporary captal structure changes can cause a sub optmal delay and Kng and Mauer (2000) observe that frms can employ a sub-optmal call polcy for a varety of reasons. In any event Crabbe and Helwege (1994) do not fnd any evdence that callable bonds reduce the undernvestment problem alone. Specfcally when controllng for credt

7 7 rsk, maturty and trends n the use of call provsons they fnd that callable bonds are no more lkely to upgrade or downgrade than non-callable bonds. Wth a make-whole call provson, the call prce s not determned by a prce schedule. Instead as Colln-Dufresne and Goldsten (2001), Colln-Dufresne et al. (2001) and Gottesman and Roberts (2007) note make whole provsons have a call prce determned by dscountng the bond s remanng contractual cash flows at a specfed spread over a smlar maturty Treasury rate. Therefore the call prce floats nversely wth Treasury rates. If exercsed, the make-whole call prce s calculated as the maxmum of par value or the present value of the bond s remanng payments. A prmary beneft of the make-whole call provson relatve to the classcal call provson s that the floatng call prce vrtually elmnates the ncentve for the frm to call when nterest rates decrease. Thus, make whole bondholders face lttle rsk that ther ncome can be exproprated due to adverse nterest rate movements. Make-whole call provsons have become qute common n corporate debt over the past ten years. 4 Mann and Powers (2003b) suggest that the make-whole call s useful for frms that antcpate a need for fnancal flexblty wthout any dependence on low economy-wde nterest rates. Powers and Tsyplakov (2004) use a structural model to examne whether make-whole call provsons are farly prced ntally and conclude that make-whole call provsons are too expensve for the benefts that they provde. Power and Tsyplakov (2008) report that make-whole call provsons are sgnfcantly overprced on average across a 10-year sample but ncremental yelds are decreasng n the later part of the sample. Nayer and Stock (2008) fnd clear evdence that stockholders do n fact beneft from ssues of make whole bonds and suggest that make whole bonds allevate agency costs wthout the use of expensve ordnary call provsons. Survey evdence n Mann and Powers (2003b) and Powers and Sarkar (2006) fnd that corporate executves beleve make-whole call provsons offer tangble 4 Approxmately 20% of the ssues n the Merrll Lynch 1-5 Year Government Corporate Index have make-whole call provson.

8 8 benefts to the frm n the form of ncreased fnancal flexblty. Thus, frms are wllng to pay a premum n order to ncorporate make-whole call provsons. However, Mann and Powers (2003a) and Power and Tsyplakov (2004) see lttle opportunty for arbtrage opportuntes due to the ncompleteness of the corporate bond market. Wthout arbtrage opportuntes, there s no obvous mechansm that wll drve down the ncremental offer yeld of make-whole call provsons. Make whole provsons can allevate the undernvestment problem because t separates the ncentve to call a bond to acheve nterest cost savngs from callng a bond to avod the undernvestment problem. Specfcally the frm can call the entre bond ssue only at a call prce determned by a yeld that s set at a predetermned spread above a smlar maturty treasury yeld. Ths means that as nterest rates decrease, the call prce ncreases so there s lttle room for coupon cost savngs due to nterest rate changes. The frm s free to take on a rsky project n the assurance that f the project subsequently proves tself, the frm can call the bond at prces consstent wth the general level of nterest rates and so avod expropraton of the benefts of the now proven project by the bondholders. Once the qualty of the project s revealed, stock returns mprove and stock analysts growth estmates are upgraded (Nayar and Stock 2008) so t s not unreasonable to suspect that ratng agences would upgrade the credt qualty of the frm s bonds. Accordngly make whole bonds should experence unusually postve re-ratng experence f frms use them to fund rsk projects subject to the undernvestment problem. Accordng to Frdson (1993) claw back provsons appeared n hgh yeld offerngs n These clauses allow ssuers to partally call a bond ssue wth captal rased from equty fnancngs despte ordnary call lmtatons. Bonds wth claw back provsons are also thought to allevate the under nvestment problem. Generally, asymmetrc nformaton problems become less severe when a new project commences va equty nvestment. Ths means the nvestment decson can cause a wealth transfer from shareholders to bondholders as bondholders beneft from a reducton n credt rsk but stll enjoy a hgh coupon rate set when nformaton asymmetres were hgh. Ths can lead to

9 9 the undernvestment problem because managers wll wsh to avod ths wealth transfer and so wll be reluctant to accept proftable projects requrng new equty fnancng. Claw back provsons make t possble for ssuers to mtgate wealth transfers that result from a reducton n nformaton asymmetres surroundng equty offerngs as the frm can repurchase a porton of old hgh coupon bond ssues at relatvely low prces. Therefore bonds that contan claw back call provsons should on average experence unusually good re-ratng experence as frms wth claw back provsons should ssue equty, thereby reducng relance of debt n the frms captal structure, to fnance valuable projects. Goyal et al. (1998) argue that claw back provsons are desgned to soften problems of undernvestment that are a result of changes n a frm s nformaton envronment when the frm ssues new equty. Consstent wth ths hypothess they fnd that frms most lkely to suffer from the undernvestment problem, specfcally unregulated and prvate frms wth more ntangble and less lqud assets are the most lkely frms to ssue claw back bonds. Smlarly Danels et al. (2009) fnd that smaller frms wth lower ratngs and lower proft favor claw back bonds, precsely the knd of frm that wll suffer most from the undernvestment problem. They also fnd that claw back bonds have a hgher offerng yeld spread and therefore are more costly to ssuers than ordnary callable bonds. In summary we fnd that there s no emprcal evdence that ordnary callable bonds deal specfcally wth the undernvestment problem. However, we fnd that the theoretcal justfcaton for refnng the call provson va make whole and claw back clauses s to releve the undernvestment dsncentve. Moreover, the lterature clams that these bonds do help resolve the undernvestment problem as frms that ssue make whole and claw back bonds are frms that are lkely to face undernvestment and after ssue enjoy mprovng growth prospects and stock prces. If ndeed make whole and claw back provsons genunely allevate the undernvestment dsncentve then bondholders as well as shareholders should beneft. Therefore the re-ratng experence of

10 10 ordnary callable bonds, that s callable bonds wthout a make whole or claw back provson, can act as a control sample to detect whether the make whole and claw back provson bonds are dealng more effectvely wth the undernvestment problem than ordnary callable bonds. Specfcally, bonds that contan make whole or claw back provsons should on average experence more upgrades and/or less downgrades than bonds employng ordnary call provsons as subsequent to ssue, frms wth make whole or claw back provsons should on average have more future nvestment opportuntes that subsequently prove to be valuable to justfy the mprovng growth prospects, hgher stock prces and hgher credt ratngs. 3. Data selecton We use the Mergent Inc s Fxed Investment Securtes Database FISD. The FISD conssts of detaled cross sectonal nformaton on ssue characterstcs of all bonds that the Natonal Assocaton of Insurance Commssoners had on ther books as of January 1, 1995, and all bonds that they bought up to and ncludng May 27, Each of the approxmately 100,000 bond ssues s dentfed by the ISIN number and ncludes nformaton on the maturty date, offerng date, ratng date, ratng, ratng type, offerng amount, ndustry code and type of call provson. From the FISD, we select all bonds that belong to the ndustral, fnancal, and utlty ndustres whle we elmnate Treasury bonds. Therefore our sample contans corporate bonds only. On examnng these corporate bonds for ratng type we fnd that Duff and Phelps do not rate many bonds wthn each ratng category. Moreover, vrtually all bonds rated by Duff and Phelps are also rated by one of the other manstream ratng agences, so we decde to neglect Duff and Phelps ratngs. However, we consder all Standard and Poor s, Moodys and Ftch rated bonds because they rate a large number of bonds n all ndustry categores. Of these we only keep those wth a ratng date wthn one year of the offerng date to ensure that the bond under study has the same ratng t had on the date t was offered. Moreover, we cut off the sample selecton after May 27, 2007 to

11 11 assure that all bonds n the sample have at least one year of agng so that bond ratng agences have the opportunty to adjust the ratngs for the mpact of new nvestment. To report the characterstcs of the sample by ratng we convert Standard and Poors, Moodys and Ftch letter ratngs nto numercal equvalents accordng to Table 2. Notce that all ratng agences have an almost dentcal ratng system wth eght broad ratng categores, sx of whch are sub dvded nto three shades of ratngs. At the lower end there appears to be a mnor devaton where Standard and Poors has one lower ratng D and Ftch has two addtonal lower ratngs of DD and DDD than Moodys so that n total Moodys has 21, Standard and Poors 22 and Ftch 24 ratngs. However ths devaton s mnor as very few bonds have a ratng of D, DD or DDD wthn one year of ssue so we smply assgn the same numercal ratng of one to Moodys ratng of C, Standard and Poors ratngs of C and D, and Ftch s ratngs of C, D, DD and DDD. Only very few bonds have no ratng that we have coded as NR. <<Table 2 about here>> Often the ratng agences dsagree as to the ratng of a bond usually by one shade of credt ratng. However, the pont of ths paper s to determne whether the type of fnancal contract s able to genunely resolve the undernvestment problem by examnng the change n a gven bond s ratng over tme rather than level of the orgnal ratng at the date of ssue. Therefore we can avod ssues regardng the ratng level when aggregatng the sample of Standard and Poors, Moodys and Ftch rated bonds, as we do not have to average the ratngs of the three agences. Instead, we take the most recent after ssue ratng of Standard and Poors, Moodys or Ftch. In the case of tes we prortze Standard and Poors, then Moodys and fnally Ftch as these are the rankngs of the most common ratng agences on the FISD.

12 12 From ths ntal selecton of bonds we select three sub samples, the make whole, claw back and ordnary callable bond sub samples. The make whole sub sample conssts of bonds from the above selecton that have make whole call provson but do not contan a claw back provson. Smlarly the claw back sub sample conssts of bonds that contan a claw back provson but do not contan a make whole provson. Fnally ordnary callable bonds have call provsons but do not contan make whole or claw back provsons. We also delete all ordnary callable bonds of frms that have make whole or claw back bonds. Ths helps ensure that the otherwse callable bond sample can act as a control sample snce make whole and claw back provsons are supposed to resolve the undernvestment problem for the frm. Banko and Zhou (2010) and Crabbe and Helwege (1994) note that the use of call provsons vary through tme so to ensure that there s no dfference n the trend n the use of make whole, claw back and otherwse callable bonds we plot the number of bonds by offerng year and call provson type n Fgure 1. Ths fgure shows that pror to the 1995 offerng year the database s domnated by otherwse callable bonds except for a sngle spke of make whole bonds n In contrast, from 1995 onwards a farly large number of bonds of each type are ssued n each year. As the dstrbuton of offerng dates by call provson type s so dfferent pror to 1995 than n subsequent offerng years we thnk that the use of bonds ssued pror to 1995 wll produce a trend bas to our results so we decded to nclude only bonds that were offered n 1995 or later n our fnal sample. <<Fgure 1 about here>> These selecton procedures leave a total sample of 16,612 callable bonds consstng of 5,263 make whole, 3,699 claw back and 7,650 ordnary callable bonds. Tables 3, 4 and 5 report the detals of the make whole, claw back and otherwse callable sub samples.

13 <<Table 3, 4 and 5 about here>> 13 We make three observatons concernng our sample as reported n Tables 3, 4 and 5. Frst examnng the sub samples of bonds by ndustry, we note that whle the make whole and claw back bonds are domnated by the ndustral category, the ordnary callable bonds are domnated by the fnancal ndustry. Moreover there are few claw back provson bonds n the utlty ndustry. In contrast all types of callable bonds are well represented n the ndustral category. Snce utlty bonds are subject to hgh regulatory rsk that s n large part absent n ndustral bonds and the rsks assocated wth fnancal bonds are dfferent than the rsks assocated wth ndustral bonds, we must fnd a way to control for ndustry effects so that we are confdent that the results wll be due to the use of the call provsons rather than dfferences n ndustry categores. Second we examne the sample by credt ratng. Wth the excepton of fnancal bonds, make whole bonds are hgher rated and claw back bonds are lower rated than ordnary callable bonds. Specfcally, for non-fnancal bonds the average claw back provson bond has a ratng of B and the average make whole call provson bond has a ratng of BBB+, slghtly more or less than one whole credt ratng category lower and hgher respectvely than the typcally BBB- rated ordnary callable ndustral bond. It s gratfyng to note that except for the very hgh ratngs, AA+ and AA, and for the very low ratngs CCC- and below, the granularty of the ordnary callable ndustral bond data s very fne. Ths means that when we refne the sample by broad credt ratngs we are able to fnd a reasonable sample sze, at least 35 or more for the ordnary callable ndustral bonds, and 139 or more for all ordnary callable bonds to act as a control sample all broad categores of credt ratngs. Thrd, we examne the sub samples by maturty. We note that ordnary callable bonds and make whole bonds have an average maturty of 15 years and 13 years respectvely, whch s hgher than 9 years average maturty for the claw back provson bonds. However, Bal and Sknner (2006) note that the average maturty of corporate bonds typcally declnes wth credt ratng and evdently

14 14 much of ths dfference n average maturty s accounted for by the dfferences n average credt ratng. For example, examnng the average maturty for the B ratng, otherwse callable and make whole bonds typcally have a much lower average maturty of 8 to 11 years that s much closer to the 9 years average maturty for the claw back provson B rated bonds. Overall the sample that we select appears capable of provdng the data necessary to statstcally test to see whether the reratng experence of make whole and claw back provson bonds are dfferent than ordnary callable bonds. 4. Emprcal Results Relevng the undernvestment problem should not only mprove earnngs prospects for the frm but also mprove the credt ratng of ther bonds, as more valuable projects wll be accepted. Wder economc events can swamp the postve re-ratng effect, but stll f these provsons are resolvng the undernvestment problem then bonds employng these provsons f not enjoyng more upgrade events then they should at least enjoy less downgrade events than otherwse smlar ordnary callable bonds. The lterature clearly specfes that call behavor s nfluenced by bond provsons (e.g. Kng and Mauer 2000, Mtchell 1991) and lstng, bond market and frm effects (e.g. Banko and Zhou 2010, Nayar and Stock 2008, Crabbe and Helwege 1994, Thatcher 1985). Therefore we must control for all of these confoundng nfluences. The FISD contans varables that ndcate the presence of the full range of bond covenants ncludng put, converson and snkng fund features. Addtonally the securty level s specfed n the FISD that we convert to a seven pont numercal scale as specfed n Table 6 rankng from the hghest securty level of seven for senor secured bonds to one for the junor subordnate bonds. There are also ndcator varables for exchange lstng and for prvate placement of the bond ssue. As bond market and frm level data s not avalable from the FISD we employ two addtonal sources of nformaton. Bond market nformaton s

15 15 collected from the Federal Reserve Bank of New York, Table H6. We also employ the Q database provded by Captal IQ. The Q database contans fnancal statement nformaton that can be lnked to the FISD bond nformaton va the nne-dgt CUSIP numbers. 5 We are faced wth the task of determnng whether or not the upgrade and downgrade frequences for make whole and claw back bonds are dfferent than the upgrade and downgrade frequences for ordnary callable bonds. Ths zero, one specfcaton of the nvestgaton naturally leads to a probt model the full specfcaton of whch s as follows. P(YU 1) F(Constant CB MW UTL OCutl FIN OCfn YANKEE PUT SF SL Convert Prvate Lst MAT LEVEL ROA(0) ROA( 1) ROA( 2) ) SLOPE SIZE DA CR CAPrev (1) Where YU = 1 f there s an upgrade and 0 f no change n ratng wthn one year of the ratng date CB = 1 f the bond s a claw back bond, 0 otherwse MW = 1 f the bond s a make whole bond, 0 otherwse UTL = 1 f the bond s a utlty bond, 0 otherwse OCutl = 1 f the bond s an ordnary callable utlty ndustry bond, 0 otherwse FIN = 1 f the bond s a fnance ndustry bond, 0 otherwse OCfn = 1 f the bond s an ordnary callable fnancal ndustry bond, 0 otherwse YANKEE = 1 f the bond s a Yankee bond, 0 otherwse PUT = 1 f the bond s a putable bond, 0 otherwse SF = 1 f the bond s a snkng fund bond, 0 otherwse SL = 7 to 1 from the hghest to lowest securty level as specfed n Table 6 Convert = 1 f the bond s a convertble bond, 0 otherwse 5 In performng the match of the Q data wth the FISD database we gratefully acknowledge expert help from the staff of Q data. All of the subsequent matches made by CUSIPS were double checked by matchng company names.

16 Prvate = 1 f the bond s a prvate placement bond, 0 otherwse 16 Lst = 1 f the bond s exchange lsted, 0 otherwse MAT s the at ssue maturty of the bond measured n years LEVEL s the level of the Treasury term structure as proxed by the one year Treasury nterest rate SLOPE s the slope of the Treasury term structure as proxed by the dfference between the ten-year and one-year Treasury nterest rate SIZE s the log of ssung frm s assets DA s the debt to assets rato of the ssung frm CR s the current rato of the ssung frm CAPrev s the rato of captal expendture to revenue of the ssung frm ROA(0) s the ssue year return on assets of the ssung frm ROA(-1) s the pror year return on assets of the ssung frm ROA(-2) s the two years pror return on assets of the ssung frm We wll run two sets of probt regressons, the frst set above examnes the upgrade experence and the second set examnes the downgrade experence. Specfcally, the second set of probt regressons wll be the same as above except that the dependent varable wll be YD where one represents a downgrade and zero f there s no ratng change wthn one year of the ratng date. These regressons are done n sets that control for the broad ratng category of the bond as well as ts nvestment grade (BBB- and above) and below nvestment grade (BB+ and below) status. Industry dummes FIN and UTL control for ndustry effects. By ncludng dummy varables for both the call provson type and for the ndustry category, we are also mplctly assumng that the upgrade and downgrade lkelhood of the types of call provsons are the same across ndustres. Ths may not hold n the case of ordnary callable bonds snce Table 5 reveals that ordnary callable bonds have a larger concentraton (76.29%) n the fnance ndustry than for make whole (Table 2,

17 %) and claw back (Table 3, 4.49%) provson bonds. Moreover, Table 3 reports that claw back provson bonds are almost exclusvely ndustral (92.73%) wth few fnancal (4.49%) and even fewer utlty (2.78%) bonds. It s possble that fnance and utlty bonds have more (or less) stable ratngs than ndustral bonds and the dfference n ndustry profle of ordnary callable bonds can tant the nterpretaton of the call provson MW and CB coeffcents. 6 Therefore we nclude nteracton varables OCfn and OCutl for ordnary callable fnancal and utlty bonds respectvely and as a result nterpret the CB and MW coeffcents as the dfferental lkelhood of a ratng change relatve to ordnary callable ndustral bonds. Eght varables control for the partcular characterstcs of a gven bond ssue. We nclude the YANKEE dummy varable because foregn bonds can have a dfferent re-ratng experence than domestc bonds snce they are more exposed to nternatonal factors than domestc bonds. Put, snkng fund, securty level and converson feature varables are ncluded because changes n market condtons effect the value of these covenants and therefore could ether confound the nfluence of market condtons on the lkelhood of an up or downgrade or drectly nfluence ratng agency ratng decsons. 7 We nclude prvate placement and exchange lstng status as control varables because the presence of make whole and claw back provsons, rather than allowng the stockholders to capture the full value of +NPV projects, could be used by weak credt frms to access fnancng not otherwse avalable. In fact Goyal et al (1998) fnds that many frms that ssue claw back bonds are prvate frms wth low credt ratngs, just the sort of frm that can have restrctve access to the publc debt market. Fnally Barnea et al. (1980) argue that short-term bonds are equally capable of resolvng agency problems and nformaton asymmetres as callable bonds so ncludng short-term 6 Indeed evdently t does, as the nteracton varables OCfn and OCutl are frequently sgnfcant at the 1% level. Typcally the nteracton varables report a dfferent re-ratng experence than the ndustral category. For example, our results show that fnance bonds have a hgher lkelhood of an upgrade than ndustral bonds. Wthout controllng for ths nteracton effect, we would be comparng claw back bonds that are bascally ndustral bonds wth ordnary callable bonds that are typcally fnancal bonds and lkely concludng that claw backs upgrade less frequently than ordnary callable bonds when n fact t s the case that ndustral bonds upgrade less frequently than fnancal bonds. 7 For example, the decson not to downgrade s a decson that can be nfluenced by a strong securty level provson durng a poor economc clmate.

18 18 bonds wthout any control for maturty effects can confound the evdence of whether make whole and claw back provsons can resolve the undernvestment problem. Rather than arbtrarly cull our sample for short term bonds we choose nstead to nclude bonds of all maturtes and nclude the ntal maturty of the bonds MAT as a control varable. The current level and slope of the term structure of nterest rates are summary statstcs of the current state of the overall bond market that can nfluence the lkelhood that a gven bond can up or down grade. For nstance, as the level of the term structure LEVEL rses new projects become more expensve and are less lkely to be accepted so the lkelhood of an upgrade can decrease. Addtonally, Estrella and Hardouvels (1991) and Estrella and Mshkn (1997) fnd that an ncrease n the slope of the term structure foreshadow mprovements n real economc actvty whle Estrella and Mshkn (1998) fnd that decrease n the slope of the term structure foreshadows recessons. Therefore as the slope of the term structure ncreases the lkelhood that a gven bond wll upgrade can ncrease, as economc condtons are more lkely to mprove. We proxy the level of the term structure as the one year Treasury nterest rate and the slope of the term structure as dfference between the ten year and the one year Treasury nterest rate. We nclude seven frm level varables as ether control varables or because they can be an alteratve explanaton for the change n ratng of the frm s debt. Frm sze s ncluded as a control varable because Goyal et al. (1998) show that the presence of claw back provsons s related to frm sze. Bonds can be re-rated for many reasons that are related to the general condton of the frm such as the frm s debt burden, lqudty, growth and proftablty. Therefore we nclude proxes for the ssung frm s debt burden as the rato of debt to assets DA, lqudty as the current rato CR, growth as the rato of captal expendtures to revenue CAPrev and proftablty as the frm s return on assets ROA. Moreover a track record of proftablty rather than a sngle year s result could cause a ratng change so we nclude the ROA of the ssue year and two years pror to the ssue year.

19 19 In summary, equaton (1) says that at date t, the probablty that a gven callable bond wll upgrade (YU t = 1), or mantan ts ratng (YU t = 0) relatve to the base case (ordnary callable ndustral bonds) s a functon F of a constant, ts call provson (make whole, claw back), ndustry category (fnancal, utlty), whether the bond s a foregn bond, the presence of covenants (put, snkng fund, securty level and converson features), ts ssue status (exchange lsted, prvate placement), the maturty of the bond, the level and slope of the term structure and the characterstcs of the ssung frm (sze, debt to asset rato, current rato, captal expendtures to revenue rato and current and past proftablty). The above equaton s estmated va a maxmum lkelhood probt regresson rather than OLS snce the dependent varable s dchotomous, only beng able to take the value of one or zero. The standard errors are corrected for heteroscedastcty. Table 7 reports the result of (1) for upgrades and Table 8 reports the results of the correspondng regresson for downgrades. << Tables 7 and 8 about here>> Lookng at the last column, the below nvestment grade category of Tables 7 and 8, we see that claw back bonds are more lkely to upgrade and are less lkely to downgrade than ordnary callable bonds. Moreover make whole bonds are more lkely to upgrade than ordnary callable ndustral bonds. All of these results are statstcally sgnfcant. Clearly there s strong evdence that below nvestment grade claw back and make whole bonds have an exceptonally good re-ratng experence that perssts even when we control for ndustry, bond market condton, ssue and frm characterstcs. In contrast, we are unable to relably estmate the re-ratng experence of nvestment grade claw back bonds as Table 4 reveals that there are so very few bonds of ths type. Addtonally, we see that nvestment grade make whole bonds are more lkely to downgrade than ndustral ordnary

20 20 callable bonds. Clearly we have no evdence that for the nvestment grade, claw back and make whole bonds have an unusually good re-ratng experence. These conclusons are renforced when we examne the re-ratng experence by broad ratng category. For claw back bonds, we fnd that B rated bonds, whch represents more than 76% of all claw back bonds, are more lkely to upgrade than ordnary callable ndustral bonds. Moreover BB rated bonds are less lkely to downgrade. Both of these results are hghly sgnfcant so surely are not due to mere chance. We see that for make whole below nvestment grade bonds, B bonds are more lkely to upgrade. In contrast, the overall downgrade experence for below nvestment grade make whole bonds s neutral because whle BB bonds are less lkely to downgrade, B rated make whole bonds are more lkely to downgrade. Fnally we note that A rated make whole bonds are more lkely to downgrade and are less lkely to upgrade than ordnary callable ndustral bonds but as we report above, when aggregated by nvestment grade category the overall effect s neutral. Overall we fnd clear evdence that claw back and make whole bonds do n fact appear to enjoy an exceptonally good re-ratng experence but ths effect s concentrated n the below nvestment grade category, just where we would expect agency problems to be most prevalent. We note that these fndngs are robust to frm and ssue characterstcs as well as ndustry effects and bond market condtons. 8 Next we examne ndustry effects. Generally speakng, utlty bonds UTL seem to have more stable ratngs than ordnary callable bonds. Below nvestment grade utltes are less lkely to downgrade and for utltes that are also ordnary callable OCutl they are less lkely to upgrade f they are nvestment grade and are less lkely to downgrade f they are below nvestment grade. 8 We conduct a varety of robustness checks. Frst, we examne the re-ratng experence of make whole, claw back and ordnary callable bonds for the ndustry category only. Second, we exclude zero coupon and varable rate bonds. We prefer to nclude zero and varable rate bonds as reported n the text because ths ncreases sample sze by 1,763 and t s not clear f the fact that a bond s zero or varable rate t wll affect the re-ratng experence. Thrd to check on the stablty of the coeffcents we splt our sample nto two equal szed data sets by assgnng all even observatons n one set and all odd observatons n another data set. These results also clearly show that claw back and make whole bonds have a sgnfcantly better re-ratng experence for the below nvestment grades. For make whole bonds the nvestment grade experence proves agan to be no dfferent than ordnary callable bonds. These results are not ncluded for the sake of brevty, but are avalable from the correspondng author upon request.

21 21 Fnancal bonds FIN appear to have a good re-ratng experence that s moderated when we examne the performance of fnancal bonds that are also ordnary callable. Specfcally, below nvestment grade fnancal bonds FIN are more lkely to upgrade and nvestment grade FIN bonds are less lkely to downgrade than ordnary callable ndustral bonds. However, when a FIN bond s also an ordnary callable bond OCfn ths good re-ratng experence s moderated as these bonds are more lkely to downgrade and less lkely to upgrade, the only excepton s for nvestment grade OCfn where they are more lkely to upgrade. Except for below nvestment grade bonds, Yankee bonds do not have a dfferent re-ratng experence than domestc bonds. For AA and BBB broad ratng categores Yankee bonds are ether less lkely to upgrade or downgrade and B rated bonds are more lkely to upgrade than domestc bonds. Aggregatng by nvestment and below nvestment grades however, the nvestment grade experence of Yankee bonds s no dfferent than the experence of domestc bonds but the good up gradng experence of below nvestment grade bonds persst. In most cases bond covenants appear as sweeteners to entce bondholder s to nvest n them because most types of bonds wth covenants tend to have sgnfcantly poorer re-ratng experence when compared to bonds wthout these covenants. Convertble nvestment grade bonds are less lkely to upgrade and are more lkely to downgrade than nonconvertble bonds. Snkng fund bonds are less lkely to upgrade and below nvestment grade snkng fund bonds are more lkely to downgrade than bonds wthout snkng funds. For the nvestment grades, the hgher the level the securty the less lkely the bond wll upgrade and the more lkely the bond wll downgrade. For below nvestment grades, the more securty the more unstable the ratng as below nvestment grade bonds wth a hgh level of securty are more lkely to upgrade as well as downgrade. Only put features are an excepton where the put feature s assocated wth an exceptonally good re-ratng experence. Even here however ths good re-ratng experence apply only for the below nvestment grade as these bonds are more lkely to upgrade than bonds wthout put features.

22 22 We fnd that the ratngs of ssues that are prvately placed are more stable, beng less lkely to upgrade and downgrade for all ratng categores. Ths result s remarkably strong and sgnfcant. Meanwhle exchange lsted bonds that are also below nvestment grade are more lkely to upgrade and less lkely to downgrade. Investment grade exchange lsted bonds however, have more unstable ratngs beng more lkely to upgrade as well as downgrade than bonds that are not exchange lsted. Barnea et al. (1980) argue that short-term bonds are equally capable of resolvng agency problems as callable bonds so we are also nterested to see f the maturty of callable bonds has an effect on the re-ratng experence. Interestngly we do fnd a maturty effect where nvestment grade bonds are more lkely to upgrade and below nvestment grade bonds are less lkely to upgrade wth maturty. Ths s consstent wth Barnea et al. (1980) snce relatvely weak credts do not ssue short term bonds and tend to rely on longer term bonds and hence the relatvely poor re-ratng experence of below nvestment grade longer term bonds. We examne the nfluence of bond market condtons on callable bond re-ratng experence. We do not fnd much evdence that the level or the slope of the term structure affects the re-ratng experence of bonds when aggregatng the results by nvestment or below nvestment grade status. The sole excepton s that below nvestment grade bonds are less lkely to downgrade as the level rses. These results suggest that bond-ratng agences make re-ratng decsons based on longerterm fundamental factors rather than current bond market condtons, a result that s consstent wth Amato and Furfne (2004). Lookng at frm level control varables, we fnd that the larger the frm the more lkely bond ssues are to upgrade or downgrade suggestng perhaps that larger frms are more actvely montored by ratng agences. Hgher debt burdens foreshadow poor re-ratng experence as the hgher the total debt to asset rato DA the less lkely a bond wll upgrade and the more lkely a bond wll downgrade. Hgher proftablty on the other hand foreshadows a good re-ratng experence as the hgher the current ROA the more lkely a bond wll upgrade and the less lkely a bond wll

23 23 downgrade. There appears to be a tmng element to the relaton between proftablty and re-ratng experence as pror years return on assets ROA n nversely related to a good re-ratng experence. For nstance, the hgher the proftablty one year pror to ssue ROA(-1) the more lkely a bond wll downgrade and for below nvestment grade bonds, the less lkely they wll upgrade. The effect of hgher growth on the re-ratng experence appears related to the orgnal credt ratng. For below nvestment grade bonds the hgher CAPrev the less lkely the bond wll downgrade but for nvestment grade bonds the hgher CAPrev the less lkely to the bond wll upgrade. Fnally shortterm lqudty when aggregated by nvestment vs. below nvestment grade s not related to future reratng experence suggestng that re-ratng s related to longer-term structural problems. 5. Summary and Conclusons To date t s not clear whether the make whole and claw back provsons are economcally justfed. We know from Banko and Zhou (2010) that ordnary calls can allevate the combnaton of asymmetrc nformaton and undernvestment problems. Other emprcal suggests that recent refnements to the call provson, specfcally the make whole and claw back call provson, are explaned as methods to resolve the undernvestment problem that s already at least partly resolved by ordnary calls. Specfcally stockholders appear to beneft from hgher earnngs and growth subsequent to the ssue of make whole provson bonds (Nayar and Stock 2008) and frms that ssue claw back bonds are more lkely to have problems wth undernvestment (Goyal et al.1998). However f these types of callable bonds do n fact allevate the undernvestment problem, then bondholders should also beneft, otherwse the supposed beneft to the stockholders can represent a repatraton of value already extracted by bondholders wth no mprovement n the value of the frm. We examne the re-ratng experence of callable bonds to see f the re-ratng experence of make whole and claw back provson bonds s dfferent from the re-ratng experence of ordnary callable bonds. We suggest that f these provsons genunely confront the undernvestment problem n a

24 24 way that s superor to ordnary call provsons then make whole and claw back provson bonds should share n the benefts from the resoluton of the undernvestment problem. Specfcally as the frm wll be encouraged to accept projects leadng to hgher growth and earnngs, make whole and claw back provson bonds wll be more lkely to upgrade and/or less lkely to downgrade than otherwse smlar ordnary callable bonds that do not employ these provsons. We fnd evdence that supports ths theory for claw back and make whole bonds rated below nvestment grade, just where agency problems tend to be most severe. Specfcally when we control for orgnal ratng, ndustry effects, foregn bond status, bond covenants, lstng and maturty effects, economc condtons n the bond market and frm effects such as sze, debt burden, lqudty, growth potental and proftablty, we fnd a sgnfcantly hgher lkelhood that below nvestment grade claw back and make whole bonds wll upgrade and a sgnfcant lower lkelhood that below nvestment grade claw back bonds wll downgrade than ordnary callable bonds. Clearly ths suggests that the make whole and claw back refnements to the ordnary call provsons are justfed as methods to deal more effectvely wth the under nvestment problem at least for bonds that are orgnally rated below nvestment grade. Moreover we fnd ndustry effects where fnancal bonds have a better re-ratng experence and utlty bonds have a more stable re-ratng experence than ndustral bonds. Many bond market covenants, specfcally the converson and snkng fund features and the securty provson, have a poor re-ratng experence suggestng that these sorts of covenants are used as sweeteners to compensate nvestors for buyng rsky bonds. Several control proxes suggest that certan types of bonds attract more attenton from the ratng agences. Specfcally larger frms that lst ther bonds on an exchange tend to have a more unstable re-ratng experence havng both a greater lkelhood of upgrades and downgrades, as opposed to smaller frms that prvately place ther bonds that have a stable re-ratng experence wth a lower lkelhood of an upgrade or a downgrade. Ratng agences do not appear to re-rate based on short-term factors such as bond market condtons and the frm s