What Does the Yield on Subordinated Bank Debt Measure?

What Does the Yield on Subordinated Bank Debt Measure? Urs W. Birchler and Diana Hancock ∗ December 18, 2003 Abstract Weprovideevidencethatabank’ssubordinated debtyield spread isnot, byitself,asufficientmeasureofdefaultrisk. Weuseamodelinwhichsubordinated debt is held by investors with superior knowledge (”informed investor hypothesis”). First, we show that in theory the yield spread on subordinated debt must compensate investors for expected loss plus give them an incentive not to prefer senior debt. Second we present strong empiricalevidenceinfavoroftheinformedinvestorhypothesisandofthe existenceoftheincentivepremiumpredictedbythemodel. Usingdataon the timing and pricing of public debt issues made by large U.S. banking organizations during the 1985-2002 period, we find that banks issue relativelymoresubordinateddebtingoodtimes,i.e. wheninformedinvestors havegoodnews. Spreadsatissuance(correctedforsampleselectionbias) react to (superior) private and to public information, in line with the comparativestaticsofthepostulatedincentivepremium. Interestingly,as the model predicts, the influence of sophisticated investors’ information onthesubordinatedyieldspreadbecameweakeraftertheintroductionof promptcorrectiveactionanddepositorpreferencereforms,whiletheinfluence of public risk perception grew stronger. Finally, our model explains anomaliesfromtheempiricalliteratureonsubordinateddebtspreadsand from market interviews (e.g. limited sensitivity to bank-specific risk and the ”ballooning” of spreads in bad times). We conclude that a bank’s subordinated yield spread conveys important information if interpreted together with its senior spread and with other banks’ subordinated yield spreads. Keywords: market discipline, subordinated debt, bank supervision. JEL-classifications: D8, G2, K2 For helpful comments we are indebted to Kathleen McDill, to Myron Kwast, to Matteo ∗ Facchinetti,toparticipantsintheBaselCommitteeresearchworkshopatBancad’Italia,Rome 2003,andtoseminarparticipantsatUniversitédeLausanne,atÖsterreichischeNationalbank, at Universität St. Gallen, and at the European Economic Association annual meeting in Lausanne (2001). Excellent research assistance was provided by Laura Kawano. The views expressed here are those of the authors and do not necessarily reflect those of the Board of Governors,theSwissNationalBank,ortheirrespective staff. Authors’ Information: Urs Birchler [Address: Swiss National Bank, Börsenstrasse 15, 8001 Zürich; e-mail: urs.birchler@snb.ch; Phone: +41 1 631 34 26; Fax: +41 1 631 81 09] and Diana Hancock [Address: MailStop 153,Board ofGovernors oftheFederalReserve System, Washington,DC 20551;e-mail: Diana.Hancock@frb.gov;Phone: (202)452-3019;Fax: (202) 452-5295]. 1

1 Introduction Over the last few years, economists have intensely debated the role of subordinated debt as a vehicle for improved market discipline in banking.1 Some authorsadvocatemandatoryissuanceofsubordinateddebtinstrumentsbybanks.2 Other proposals focus on the so-called yield spread, i.e., the difference between the yield on a subordinated instrument and the yield of a comparable risk-free bond (like government debt). This yield spread is generally considered to be a good measure of bank risk, as the holders of subordinated debt absorb the firstloss (aftershareholders)in theeventof abankfailure. Consequently, some authorshaveproposedaregulatoryceilingonbanks’subordinatedyieldspreads (Calomiris,1999)whileothershavearguedthattheywouldusesuchaspreadas an early warning signal and as a trigger for prompt corrective actions (Evanoff and Wall, 2000). Taking stock with respect to the ongoing debate, BoG/DoT (2000) concluded that supervisory authorities would continue, “as part of the supervisory process,tomonitorbothyieldsandissuancepatternsofindividualinstitutions” (BoG/DoT, 2000, p. iv). Hence, it is of crucial importance to know what observed yields and issuance patterns really foretell. Proponents,aswellasopponents,ofmandatorysubordinateddebtproposals seem to agree that, in theory at least, the subordinated debt spread measures bankriskdefinedastheexpectedlosstoinvestors(i.e.,theprobabilityofdefault times the loss given default).3 In practice, subordinated bond spreads may be distortedbyseveralfactors: individualinstrumentcharacteristics,poormarket liquidity, investors’ risk aversion, and fluctuations in the market price of risk.4 Intheabsenceofsuchdisturbances,however,formostparticipantsinthedebate the subordinated bond yield spread accurately measures a bank’s risk. In this paper, we try to show that the subordinated debt spread by itself is not a sufficient measure of a bank’s default risk, since it is only meaningful when it is used with both senior debt spreads and subordinated debt spreads at other banks. Our starting point is the observation that subordinated debt, by its very definition, does not exist alone, but only as one of at least a pair of debt instruments. It is thus important to first understand the reason for the existence of dual debt before any inferences about yields are made. There are several reasons why a bank may want to issue dual class debt, including reasons that build on the heterogeneity of either investors, or of banks (see, e.g. Diamond, 1993; Barclay and Smith, 1995; Winton, 1995; Birchler, 2000). 1Foranoverview,seeBoardofGovernorsoftheFederalReserveSystem(BoG)(1999)and BoardofGovernorsoftheFederalReserveSystemandUnitedStatesDepartmentofTreasury (BoG/DoT)(2000). 2For a textbook discussion of the incentive effects of subordinated debt, see Dewatripont and Tirole (1994),section 13.3.1. 3See,forexample,Campbelland Huisman (2003). 4Elton,Gruber,AgrawalandMann(2001)estimatethatlessthan25percentofcorporate spot spreads can be explained by expected default losses. They also demonstrate that the sensitivity to factors commonly used to explain risk premiums in common stocks explains between 66 percent and 85 percent of the spread in corporate and government rates that is not explained by the difference in promised and expected payments and taxes. By these calculations, 15 to 34 percent of a typical corporate bond spread is left unexplained. In addition, Hancock and Kwast (2000) report that subordinated debt spreads for large U.S. banking organizations are sensitive to systematic risk factors such as stock market excess returns. 2

In what follows we will focus on the heterogeneity of investors. Our model assumes that some investors have superior information.5 This assumption fits well with the different subordinated debt proposals: Subordinated debt can provide market discipline only if its potential holders know more about the quality of a bank than do other investors. The “informed investor hypothesis” also fits the data well: We find strong empirical support for this assumption in thepricing of publicdebt issues made bylargeU.S.bankingorganizations in the 1985-2002 period. In addition, our model allows for differences in risk appetite (i.e., the “risk aversion hypothesis”), for which we find less empirical support. And, an alternativehypothesis—the“informedbank,orsignallinghypothesis”—wherebanks signalearningsprospectsbychoosingtoissueeitherseniororsubordinateddebt (see Barclay and Smith, 1995) is rejected by bond market data for large U.S. banking organizations. Regardless of the reason that a firm wants to issue subordinated and senior debt, it must make both instruments sufficiently attractive for the respective investors. Subordinated debt must “beat” not only risk-free alternatives, e.g., a Treasury bond, but, also must “beat” in the eyes of some investors its own sister, senior debt. This is why its equilibrium yield spread contains not only a risk premium (to make subordinated debt attractive compared to the risk-free asset), but also an “incentive premium” (to compete with senior debt). Under the informed investor hypothesis the incentive premium can be thought of as an information rent earned by investors who have favorable information about the issuer.6 Under the risk aversion hypothesis, senior debt pays a higher risk premiumwhich, liketheincentivepremium, alsocarries forwardtotheyieldon subordinated debt, even though the latter may be held by risk neutral agents. We reject claims that subordinated debt spreads measure a bank’s risk as perceived by sophisticated investors. Instead, our results almost literally support views formulated by market participats that signals from subordinated debt spreads are jammed by movements in general market sentiment.7 Our model shows that this “noise” reflects completely rational behavior on the part of investors, notwithstanding the doubts of some observers.8 In addition, our model yields predictions on banks’ issuance policies for subordinated and for seniordebt. Accordingtotheinformedinvestorhypothesis,abankwouldissue subordinated debt upon receipt of good news and it would issue senior debt 5Diversityofinformation,andthusofopinion,probablyisafactorbehindtheexistenceof manyfinancialcontractsandinstitutions(AllenandGale,2000)andbehindsomephenomena ofassetpricing (Diether,Malloy and Scherbina,2002). 6Itisawellknownresultfromcontractingmodelsthatamongheterogeneousagentsthose with superiorinformation earn an information rent. 7Inresponsetoarequestforcommentsonthefeasibilityandappropriatenessofamandatory subordinated debt policy, a U.S. Bank Holding Company stated that: “(T)he practical issuesassociatedwithdistentanglingtheseparateinfluencesofmarketfactorsandofchanges intheriskprofileofafinancialinstitutiononitsdebtspreadswouldmakesuchdebtspreadsa poortoolforsupervisorymonitoringpurposes.” (BoG/DoT,2000,p. 78). Moreover,market participantshavenotedthat: “Spreadsneedtobeinterpretedwithgreatcare. Forexample, the generallevelof spreads is quite sensitive to cyclicalfluctuations. In good times,spreads tendtoberathernarrow,reflectingtheviewthatallbanksandbankholdingcompaniesarein goodshape. Inbadtimes,spreadsballoon,reflectingbroadskepticismregardingthefinancial health ofbanking institutions.” (BoG,1999,p. 16.) 8In 2002, a U.S. Bank Examiner lamented: “The time-series movements of observed subordinateddebtspreadsdonotreflectchangesinsupervisoryinformation. Italmostseems like stupid people buy and sellthesebonds.” 3

uponreceiptofunfavorablenews. Thisisbecausethebankwouldissuedebtof different priority status to separate investors with different, yet unobservable, beliefs on the probability of its failure. Our analysis of U.S. bond market data onthetimingofpublicdebtissuesbylargebankingorganizationssuggeststhat the health of the organization influences its bond issuance decisions, and these decisions are usually consistent with the informed investor hypothesis. OnereasonwefocusonU.S.bankingorganizationsoverthe1985-2002period is because two regulatory reforms — capital-based prompt corrective actions by bank supervisors (initiated by the Federal Deposit Insurance Corporation Improvement Act, FDICIA) and depositor preference rules (which established a clear priority for the distribution of (unsecured) claims realized from the liquidationorotherresolutionofanyinsureddepository)wereimplementedapproximately in the middle of the sample period. These reforms would, of course, influence investors’ perceptions about the likelihood of bank failures and their prospectivelosseswhenbankdefaultsoccur. Aswillbeseenbelow,theseperceptions importantly influence the magnitude of the incentive premium contained in subordinated debt yields. Ourpaperisstructuredasfollows: InSection2,weintroduceourtheoretical model of the subordinated debt spread. In Section 3 we present the empirical evidenceontherisk-sensitivityoftheissuancedecisionandofsubordinateddebt spreads. In section 4, we discuss our results. And, section 5 concludes. 2 The model 2.1 Assumptions There are several reasons why a bank may want to issue debt instruments that differ with respect to their rank in the event of an insolvency.9 A plausible explanation for the existence of dual class bankdebt is heterogeneity of investors. These individuals may differ in their preferences, wealth, sophistication, or information. In what follows, we will focus on differences in investor opinion assuming that some investors have superior information (i.e., the “informed investor hypothesis”). In addition, we allow for differences in risk taste (i.e., the “risk aversion hypothesis”). Unobservable heterogeneity of investors provides the basis for partial investor separation through the use of dual debt, i.e., of a senior and a subordinated instrument. We use a variation of the Birchler (2000) model for the use of dual debt instruments by a bank borrowing from investors having different information. In this model, a bank can invest in a single one-period asset. The asset has an observable but uncertain per dollar return y Y,Y with Y > R > Y, ∈ where R is the return on a riskfree asset. We denote the prior probability of © ª “success” by p = prob Y > 0.5. The bank has no funds of its own, i.e. it has strictly limited liability. In order to invest, the bank has to borrow from © ª a large number of small investors. We normalize their aggregate funds to one dollar. Thebankoffersinvestorscontractsona“take-it-or-leave-it”basis. After contractsareoffered,butbeforeinvestorshavedecidedtoinvestornot,thebank as well as some sophisticated investors get a signal on the project’s probability of success. The signal updates p to either q(> p) (good news) or (1 q) (bad − 9Forashortoverview ofthe literatureon debtpriorities,see Birchler(2000). 4

news). We denote the fraction of informed investors by h, of whom fraction u get a good signal, and (1 u) get a bad signal. Setting p=uq+(1 u)(1 q) − − − we assure that the different beliefs are consistent. We assume that all investors are risk-neutral, but we will relax this assumption below. At the time the bank offers contracts, it faces three potential types of investors: optimists (prob Y = q), uninformed (prob Y = p), and pessimists (prob Y = (1 q)). In expected terms the three groups have size hu (op- − © ª © ª timists), (1 h) (uninformed), and h(1 u) (pessimists), respectively. Even © ª − − after the signal has occurred, the bank cannot distinguish investors. Yet, by appropriate design of contracts, the bank can partly separate investors, i.e., it can attract some but not others.10 We restrict asset payoffs to make sure that: 1. The bank finds it worthwhile to borrow from uninformed investors and, a fortiori,fromoptimists—asufficientconditionisthatpY +(1 p)Y >R; − 2. The bank does not borrow from pessimists11 — a sufficient condition is (1 q)Y +qY <R; and, − 3. The bank cannot offer risk free contracts, even if all income in case of failure (Y) is promised to uninformed investors — a sufficient condition is: R > Y/(1 k), where k = (1 h)/[(1 h)+uh], the fraction of − − − uninformed in total depositors. Theseassumptionsalsoensurethatthebankissolventifrealizedy =Y and insolvent if realized y =Y. 2.2 Contracts Contracts can be written only on observable outcomes, not on (unobservable) types of investors. The bank can offer two contracts to separate the optimists from the uninformed investors. The contract menu can be written as C= c Y,Y ,c Y,Y , 1 2 where the two elements, c and c , represent senior (i.e., non-subordinated) 1 2 © ¡ ¢ ¡ ¢ª debt (c ) and junior (i.e., subordinated) debt (c ). It will turn out that unin- 1 2 formed investors prefer the senior contract, which is served first in the event of failure,whileoptimistspreferthesubordinatedcontract,whichpaysmorewhen the banks’ project succeeds and the bank does not fail. The bank thus borrows from uh informed and from 1 h uninformed investors. The fraction of in- − formedinvestorsintotallenders(asdistinctfromthefractionintotalpotential investors) thus is k =h/[(1 h)+uh]. − We will refer to success and failure payments promised under contract i as D (face value of debt) and M (minimum repayment). The two contracts thus i i are c = D ,M and c = D ,M . 1 1 1 2 2 2 { } { } 10Given any set of contracts, the bank, due to its limited liability, would borrow from all groups ofinvestors (as long as itpromises lessthan Y). 11Inourmodelpessimists’beliefsdonotinfluencedebtspreads;pessimistscanonlydecide nottobuydebt,buttheycannotsellbankdebtshort. Diether,MalloyandScherbina(2002) useamoreelaboratemodelofheterogeneousbeliefsinwhichthemorepessimisticexpectations arenotreflected in (stock)prices due to shortsellingconstraints. 5

The bank offers the contract menu that maximizes expected profit conditional on investors’ rational investment strategies and on the prior probability of success p. The bank thus solves max p Y kD (1 k)D +(1 p)[Y kM (1 k)M ] (1) 2 1 2 1 D1,M1,D2,M2 − − − − − − − £ ¤ subject to participation constraints for both groups of investors ((2) and (3)); to incentive, or self selection, constraints ensuring that each group prefers the “right” contract ((4) and (5)); as well as to limited liability, or wealth, constraints ((6) through (9)). pD +(1 p)M R 0 (2) 1 1 − − ≥ qD +(1 q)M R 0 (3) 2 2 − − ≥ pD +(1 p)M pD +(1 p)M (4) 1 1 2 2 − ≥ − qD +(1 q)M qD +(1 q)M (5) 2 2 1 1 − ≥ − Y kD (1 k)D 0 (6) 2 1 − − − ≥ Y kM (1 k)M 0 (7) 2 1 − − − ≥ D 0,M 0 (8) 1 1 ≥ ≥ D 0,M 0 (9) 2 2 ≥ ≥ Proposition 1 (The optimal contract). The banker offers dual debt consisting of a senior contract c = D ,M , and a subordinated contract, c = 1 1 1 2 { } D ,M with payoffs D in the event of success, and M in the event of the 2 2 { } banks’ failure, where 1 1 p D = R − M (10) 1 p − p 1 1 M = Y (11) 1 (1 k) − 1 q p D = R − M (12) 2 p − qp 1 M = 0 (13) 2 Proof. The marginal rates of substitution between success and failure income for the bank, optimists and uninformed investors are MRS = (1 p)/p, bank − − MRS = (1 q)/qandMRS > (1 p)/p,respectively. ThereforeMRS < g n g − − − − MRS <MRS . Of all three parties, uninformed investors thus attach the bank n highest relative weight to income in case of failure. Therefore, they get all of Y, plus a share of Y necessary to satisfy their participation constraint (2). Optimists get enough for (a) participation and (b) for not preferring c ; as (5) is 1 more restrictive than (3), (b) binds, hence the solution for D . 2 6

The two debt instruments are represented graphicallyin Figure2.1. Linesp andq areindifferencecurvesforuninformedinvestorsandforoptimists, respectively. As these indifference curves run through (R, R), they are the relevant participation constraints. The senior instrument, D ,M , just satisfies the 1 1 { } participation constraint of uninformed investors. These investors get all availablereturnsintheeventofabankfailure,whentotalreturnY isdividedamong the uninformed investors.12 Subordinated investors get nothing in the event of a failure.13 In the event of success, they must be paid D >D to satisfy their 2 1 incentiveor,self-selection,constraint,i.e.,theirindifferencecurve(q )thatruns 0 through c . In contrast, their participation constraint (q) does not bind. It 1 is obvious from Figure 2.1, that the self-selection constraint is more restrictive thanistheparticipationconstraint. SubordinateddebtthusnotonlypaysR/q, but an additional premium D R/q, to make it attractive relative to senior 2 − debt for the optimists. We will call this premium the incentive premium. The fact that the terms of subordinated debt are determined by the incentive constraint,ratherthanbytheparticipationconstraint,isthemaintheoreticalpoint of thispaper. Wewilluseittoexplaintheimpact ofchangesin differentmodel parameters on the subordinated debt yield spread. Senior investors might be risk averse instead of, or in addition to being uninformed. Risk aversion would change (2) into: p(D ρ)+(1 p)M R 0 (14) 1 1 − − − ≥ where ρ is the additional risk premium. From Figure 2.2, we can directly tell whattheimpactofriskaversiononthepartofuninformedinvestors(i.e.,senior investors) would be. Under risk aversion the uninformed investors’ participation constraint is not represented by the straight line p but by a concave line p. The optimal senior contract would thus be become c . Consequently, the 0 01 relevant incentive constraint becomes q . Subordinated debt holders thus get 00 the same additional risk premium as senior debt holders, even though they are risk neutral! Any factor that increases the risk premium on senior debt also increases the subordinated yield spread by the same amount. Theexistenceofanincentivepremiumaspartofthesubordinateddebtyield spread crucially depends on the assumption that the bank cannot distinguish investors by their type. If it could, it could segment markets, which would reduceor,intheextreme,eliminatetheincentivepremiumonsubordinateddebt. Collin-Dufresne et al. (2001) find in a study on yield spread changes that bond and equity markets might be somewhat segmented. One would also expect segmentation between trade credit and publicly held debt, as public investors cannotpretendtobesuppliersandthuspotentialtradecreditors. Somesegmentation between senior and subordinated debt markets thus cannot be excluded a priori. 12Thereare(1 k)uninformed,whilethefractionofinformedinvestorsiseitherk(aftera − good signal)or0(afterabad signal). 13This is fairly realistic. For most banks k is small. The fraction of subordinated debt to total assets (a proxy for k) for major banks in the US and in the EU is reported by Sironi (2001,p. 240)as2.42and 1.65 percent,respectively. 7

Figure 2.1 Senior and Subordinated Debt Yields when Investors are Risk Neutral q q' M p R Senior Debt Y/(1-k) Subordinated Debt D 0 R R/q R/p Risk Premium Incentive Premium Subordinated Debt Spread Figure 2.2 Senior and Subordinated Debt Yields when Investors are Risk Averse q q' q'' M p R Senior Debt Y/(1-k) p' Subordinated Debt D 0 R R/q R/p Risk Premium Incentive Premium Including RiskAversion Subordinated Debt Spread

2.3 Issuance decisions as a function of the signal Strictlyspeaking,inourmodelinvestorsdecideaboutthebank’sliabilitystructure. After a good signal, both, senior and subordinated debt are sold. After a bad signal, only senior debt is sold. Speaking of the bank’s issuance decision makes sense nevertheless. The bank designs and offers contracts; by doing so it implicitly decides under what circumstances it will issue what kind of debt. The composition of debt issued after the two possible signals, respectively, are represented in Table 2.1. Table 2.1 The Amount of Debt Issued as a Function of the Signal Signal Senior Subordinated good (1 h)u hu − bad (1 h)u 0 − Giventheoverlysimplisticmodel,thereareonlytwopossibleliabilitystructures, only one of which has subordinated debt. For the empirical part of the paper it is sufficient to remember that the model predicts that subordinated debt tends to be issued in good times, while senior debt is the main source of finance in bad times. 2.4 Equilibrium yield spreads The net contractual yield on a debt instrument is the difference between the contractual payment after success minus the original investment, expressed in percentage points of the original investment.14 As all payoffs are per dollar and the original investment is one dollar, the yields on the two instruments are (D 1)/1. Yield spreads relative to the the safe asset are (D R). For i i − − simplicity, we report equilibrium yield spreads for a risk neutral world, bearing in mind that risk aversion of uninformed investors would increase the required spreads on senior and on subordinated debt by ρ. From (10) to (13) we can derive the equilibrium yield spreads on the two instruments. These are: 1 p δ = D R= − (R M ) (15) 1 1 − p − 1 1 p q p δ = D R= − R − M (16) 2 2 − p − pq 1 Subordinated debt has the higher promised yield than senior debt, the difference being 1 q δ δ = − M (17) 2 − 1 q 1 This difference is the joint effect of two opposing forces: On the one hand, subordinated debt should pay more than senior debt when the bank is solvent, 14As usual in financial markets, we use “yield” in the sense of contractual, i.e., promised yield(intheeventofasuccess),whichisdistinctfromtheexpectedyield(whichincludesthe possibility of failure). 8

asitpaysnothingintheeventthatthebankfails. Ontheotherhand,subordinateddebtshouldpayless,sinceitisheldbyinvestorswithrelativelyoptimistic expectations. Obviously, the subordination effect is generally stronger than is the expectation effect. This is confirmed by comparison of the spreads on the two instruments to their actuarially fair values. In a risk neutral world, fair risk premia just provide the investor with an expected return of R, the relevant expectation being conditional on his state of knowledge. Fair premia (characterized by a bar) on the two instruments are thus: 1 p ¯δ = D R= − (R M ) (18) 1 1 − p − 1 1 q ¯δ = D R= − R (19) 2 2 − q The difference between the two fair spreads 1 p q p ¯δ =D D = − M − R<0 (20) 21 2 − 1 p 1 − pq is negative, as the holders of subordinated debt are relatively optimistic. Further, it is obvious that D =D , and hence ¯δ =δ as (18) is identical to (10). 1 1 1 1 Astheseniorcontractisdeterminedbytheparticipationconstraint,itjustpays thenecessaryriskpremium. Incontrast,(19)isnotidenticalto(12). Thisleads to the following proposition: Proposition 2 (The subordinated debt spread). The yield on subordinated debt exceeds the necessary risk premium, i.e. δ >δ =(1 q)R/q (21) 2 2 − Proof. The assertion follows directly from (16) and (19). The bank must promise holders of subordinated debt not just enough to make the subordinated contract (marginally) preferable to the risk free asset; it has to promise more than R/q to make optimists prefer the subordinated contract over the senior contract. The required incentive premium is equal to 1 q p q p 1 ˆδ =D R= − [R M ]= − R Y . (22) 2 2 − q pq − 1 pq − (1 k) · − ¸ Theincentivepremiumthusdependsontheprobabilitiesofdefaultpandq(i.e., probabilities of default as perceived by uninformed and optimistic investors, respectively) and on the expression in brackets which represents the potential loss on the senior contract, or the “loss given default” in the terminology of BCBS (2001). The equilibrium spread on the subordinated contract over the yield of the risk-free asset, δ , can thus be looked at from two sides. On the one hand, 2 it is the sum of the senior spread (δ = ¯δ ) plus the risk premium between 1 1 subordinated and senior debt. On the other hand, it is the sum of the fair subordinated spread, δ , plus the additional incentive premium, ˆδ , required to 2 2 induce optimists to prefer the subordinated contract; formally: δ +δ δ δ +ˆδ . (23) 1 21 2 2 2 ≡ ≡ 9

2.5 Sensitivity of debt yield spreads to model parameters Inthissection, welookatchangesindebtspreadsinresponsetochangesinthe parametersofourmodel. Forexample,thederivativesofthesubordinateddebt spreads with respect to model parameters p, q, and Y can be computed from (19),(22),and(16).15 Thesederivativesandthoseforseniordebtarepresented in Table 2.2. Table 2.2 has five rows for each of the spreads under consideration. The derivativessatisfythesamerelationastheparametersthemselvesgiveninequation(23): Thevaluesforthesubordinateddebtspreadδ (betweendoublelines) 2 are the sum of the two top rows for the senior spread δ and the differential 1 subordinated-senior spread δ ; or, at the same time, of the two bottom rows 21 for the fair subordinated spread ¯δ and the incentive premium ˆδ . 2 2 The columns for the different model parameters reveal several interesting properties of the subordinated debt spread: 1. Sensitivity to public risk perception (p): The column for p shows, that the subordinated spread, δ , rises when p falls, i.e., when the risk 2 perceivedbytheholdersofseniordebtincreases(whenthelinespandthus q in Figure 2.1 move counterclockwise around coordinates (R, R)). The 0 subordinateddebtspreadthusriseswithadeteriorationingeneralmarket sentiment, even though its holders (as well as the issuer) do not perceive anychangeinthequalityofthebank(i.e.,inq). The(negative)sensitivity of the subordinated yield spread with respect to p is particularly strong, when p is small and when Y and k are small (i.e., when the bankruptcy dividend to senior lenders is modest).16 Plus, sensitivity to p is strong when senior investors are risk averse, as can be seen in Figure 2.2. The subordinated spread is most sensitive to p when senior investors are risk averse, but do not expect to get much in the event that a bank fails. 2. Sensitivity to informed investors’ risk perception (q): Thecolumn for q shows that the subordinated spread, δ , rises when q falls, i.e., when 2 the risk perceived by subordinated yield holders increases (when lines q and thus q in Figure 2.1 rotate counterclockwise around (R,R)). How- 0 ever, the sensitivity of δ with respect to q is smaller than the fair risk 2 premium (based on the default probability perceived by holders of subordinated debt) would suggest. This is because the incentive premium, ˆδ , 2 moves in the opposite direction from the default probability perceived by informed investors. The sensitivity of the subordinated spread to q also depends on the senior bankruptcy dividend ( Y/(1 k)): If senior debt − holders expect a bankruptcy dividend that is large (that is, debt holders expect to receive close to the risk-free return, R), then the subordinated spreadisnotmuchabovethefairriskpremium,¯δ ,andreactsstronglyto 2 the default probability perceived by informed investors (i.e., q). In contrast, if senior debt holders expect a bankruptcy dividend that is small (i.e., close to zero), the incentive premium almost cancels the impact of 15RecallthatM1=R − (1 1 k) Y. 16RecallthatR −(1 1 k) Y =− R − M1 is the shortfallofthebankruptcy dividend compared totherisk-freereturn,−andthatweassumeforsimplicitythatk,thefractionofsubordinated debt,is constant. Itis alsorelatively smallatmostbanks. 10

Table 2.2 Derivatives of Spreads (in rows) with Respect to Model Parameters (in columns) p q pand q Y (cid:17) (cid:3) 1 R(cid:3) 1 Y 0 (cid:3) 1 R(cid:3) 1 Y (cid:3)1(cid:3)p 1 1 p2 (1(cid:3)k) p2 (1(cid:3)k) p (1(cid:3)k) (cid:17) 21 h 0 i (cid:3) q 1 2(1(cid:3) 1 k) Y (cid:3)h q 1 2(1(cid:3) 1 k) Y i (cid:3)1(cid:3) q q (1(cid:3) 1 k) (cid:17) (cid:3) 1 R(cid:3) 1 Y (cid:3) 1 1 Y (cid:3) 1 R+q2(cid:3)p2 1 Y (cid:3)q(cid:3)p 1 2 p2 (1(cid:3)k) q2(1(cid:3)k) p2 q2p2 (1(cid:3)k) pq (1(cid:3)k) ¯ (cid:17) 2 h 0 i (cid:3) q 1 2 R (cid:3) q 1 2 Rh i 0 ˆ (cid:17) (cid:3) 1 R(cid:3) 1 Y 1 R(cid:3) 1 Y (cid:3)q2(cid:3)p2 R(cid:3) 1 Y (cid:3)q(cid:3)p 1 2 p2 (1(cid:3)k) q2 (1(cid:3)k) q2p2 (1(cid:3)k) pq (1(cid:3)k) h i h i h i

fluctuations in the fair risk premium, and the subordinated yield spread becomes insensitive to the default probability perceived by informed investors (i.e., q), while remaining sensitive to p. 3. Sensitivity to senior lenders’ bankruptcy dividend ( Y/(1 k)): − The column for Y shows that a higher bankruptcy dividend for senior debt does reduce the senior spread, δ , and the subordinated spread, δ . 1 2 In contrast, the fair risk premium on subordinated debt (¯δ ) does not 2 react to changes inY, of course. But, via theincentivepremium(ˆδ ), the 2 subordinated spread depends on a payment subordinated investors will never get. It even reacts morestronglyto the senior bankruptcydividend thantheseniorspreaditself. TheseeffectscanalsobeshowninFigure2.1: Thehigherthebankruptcydividendforseniorlenders,thelowerthepayoff senior lenders need to get in the non-default state (D ), and the higher 1 the necessary incentive premium (as point (D , D ) moves to the left). 1 2 Again,thesubordinatedspreadismoresensitivetotheseniorbankruptcy dividendifseniorinvestorsareriskaverse: InFigure2.2thesubordinated spread falls by a larger amount with an increase in Y/(1 k) compared − to a corresponing movement in Figure 2.1. 4. Sensitivity to change in regulatory regimes: The implementation of regulatory reforms17 during the mid-1990s provides us with an opportunity to consider their potential effects — through simultaneous changes of several model parameters — on observed senior and subordinated debt spreads. Ontheonehand,pandq haveincreased,asseniorinvestorsmay believe that the probability of bank failure is lower in the post-FDICIA period because bank supervisors will undertake prompt correctiveactions when the financial condition of a banking organization deteriorates. This perception would result in a smaller incentive premium contained in subordinated debt spreads and in less risksensitivesubordinateddebt yields. On the other hand, Y/(1 k) has decreased as the lower liquidation − standing of senior debt investors, after the implementation of depositor preference rules, would reduce the bankruptcy dividend for senior debt (see Figure 2.3). As the bankruptcy dividend for senior debt falls, the incentive premium contained in subordinated debt spreads increases, while the subordinated spread becomes more sensitive to p and less sensitive to q. It is, of course, an empirical question whether these partly opposing forces would cancel one another out, or whether one is much greater than the other. 2.6 Quantitative importance of the incentive premium We have found that, under the informed investor hypothesis, the subordinated yield spread includes an incentive premium of yet unknown size. Before we examine the empirical evidence, we try to get a reasonable guess on the size of the incentive premium we should expect in practice. From (22) we know that the incentive premium depends on p, q, and on the bankruptcy dividend or its complement, the loss given default, for senior debt. For most banks in industrial countries, survival probabilities p and q, on 17Fordetailssee below. 11

Figure2.3 HowdidRegulatoryReformsAffecttheIncentivePremium? Pre-FDICIAPeriod q q' M p R Senior Debt Y/(1-k) SubordinatedDebt D 0 R R/q R/p RiskPremium IncentivePremium SubordinatedDebtSpread Post-FDICIAPeriod q q' M p R SeniorDebt Y*/(1-k) SubordinatedDebt D 0 R R/q R/p RiskPremium IncentivePremium SubordinatedDebtSpread

a five to ten year horizon are in the range of 90-99 percent, as implied by KMV data. The denominator in (22) thus can be roughly approximated by unity. The numerator, p q, measuring the difference of opinion between optimists − and uninformed investors, depends on the quality of information available to sophisticated investors and to the public, respectively. A few percentage points looks like the upper limit, but 1-2 percent may seem a reasonable guess. Data for loss given default are available from Moody’s (see Table 2.3). For the period 1988-1991 (which falls into the pre-FDICIA period) recovery rates fordefaulteddebt(inpercentofthenominalvalueofdebt)was26.5percenton senior debt and 4.9 percent on subordinated debt.18 Takentogether,thesefigureswouldsuggestthat,inthepre-FDICIAperiod, incentivepremiaonsubordinateddebtforbanksinindustrializedcountrieswere intherangeof75to150basispoints. Hence,incentivepremiaseemtohavesimilarsizesasliquiditypremia(seeCovitz,HancockandKwast,2001). Therefore, they should be large enough to leave their traces in the data. 3 An empirical test of model predictions 3.1 Representation of model parameters in the data To test our model we have to identify its parameters with observable variables. The key parameters in our model are success probabilities p and q, and the expected recovery rate on senior debt, Y/(1 k). − For expected recovery rates we have given some point estimates above. For success probabilities, however, we need data on individual banking organizations. More precisely, we need to identify the success probabilities, or rather, the corresponding failure probabilities perceived by informed investors and by uninformed investors, (1 q) and (1 p), respectively, with someempiricalrisk − − variables. Thus, our task is to split available risk proxies into information that is only known to relatively sophisticated investors and into information that is publicly known. 1. Publicly Available Risk Proxies Quarterly balance sheet and income statement data for each banking organization i at time t are publicly available from consolidated financial statements for U.S. bank holding companies (FR-Y-9C). These data are reported as of close of business on the last calendar day of the quarter. Risk proxies typically derived from these data include the ratio of nonaccruing loans to total assets (NATA ), the ratio of accruing loans past it due90daysormoretototalassets(PDTA ),theratioofotherrealestate it owned to total assets (OREO ), and the absolute value of the difference it between assets and liabilities maturing or repricing within one year as a proportion of equity value (AGAP ). it Alsopubliclyavailablearedataonthemarketvalueofcommonstock. Usingsuchdatatogetherwithbalancesheetinformation,theratiooftotal 18Our assumption that subordinated debtholders get nothing in case of failure is thus a reasonable approximation to reality. The data contained in Table 2.2 cover the entire 1985- 2002period,butthere werenolargebank failures during the post-FDICIA period. 12

Table 2.3 Comparison of Recovery Rates for Defaulted Senior and Subordinated Debt Senior Subordinated Face Market value of Face Market value of amount of defaulted debt, one amount of defaulted debt, one defaulted month after default Default defaulted month after default Issuer Name Cusip Default Date debt (recovery rate) Cusip Date debt (recovery rate) First City Bancorporation of Texas 319594AA 9-Sep-87 17.26 49.00% 319594AE 9-Sep-87 100 . 319594AB 9-Sep-87 50 45.00% 319594AE 15-Sep-92 10.2 . 319594AB 15-Sep-92 22.2 38.00% IFRB Corporation 449506AB 15-Mar-88 26 25.00% 449506AC 15-Mar-88 100 17.50% 458916AF 15-Mar-88 100 . MCorp 587541AB 21-Oct-88 50 35.00% 55267MAB 21-Oct-88 50 35.00% 55267MAD 21-Oct-88 25 30.00% 587541AF 21-Oct-88 100 33.00% 55267MAF 21-Oct-88 100 33.00% Bank of New England Corporation 063840AA 7-Jan-91 15 1.50% 063840AB 7-Jan-91 75 . 063840AC 7-Jan-91 150 8.00% 063840AD 7-Jan-91 200 1.50% 063840AE 7-Jan-91 250 17.50% First RepublicBank Corporation 336160AA 15-Mar-88 65 32.00% 336160AD 15-Mar-88 100 19.25% 336160AB 15-Mar-88 75 27.00% 760836AC 15-Mar-88 99.69 27.50% 336160AC 15-Mar-88 99.69 27.50% Southeast Banking Corporation 841338AC 19-Sep-91 57.1 46.00% 841338AD 19-Sep-91 99.2 4.00% 841338AA 19-Sep-91 12.1 . 841338AG 19-Sep-91 50 1.25% Southwest Bancshares, Inc. 844768AA 21-Oct-88 31.9 34.00% 55267MAE 21-Oct-88 31.9 34.00% 55267MAC 21-Oct-88 50 35.00% Texas American Bancshares, Inc. 882147AA 15-Sep-88 50 14.50% Weighted Average of Recovery Rates: Senior Debt 26.48% Subordinated Debt 4.89% Source: Moody's DRS Access Dataset

book liabilities to the sum of the market value of common stock and the book value of preferred stock (MKTLEV ) can be computed.19 it While all balance sheet risk proxies are derived from publicly availabledata,itisnotnecessarilythecasethatsuchinformationiscommonly known. Some risk proxies are highly correlated with widely known information on regional and sectoral business conditions and their magnitude can be easily inferred (e.g., OREO ). Other risk proxies may be so it broadly used that they are frequently reported in the press (e.g., MK- TLEV ), and therefore, would not requirea substantialinvestment tobe it usedbyaninvestor. Theremainingriskproxieswouldrequireaninvestor to be able (and willing) to read a balance sheet statement (e.g., NATA it and PDTA ). Such bank-specific risk measures are less likely to trickle it into the public domain. 2. Private Information BanksupervisorsregularlyexamineU.S.bankingorganizationsandassign confidentialratingsbasedontheseexaminations,balancesheetandincome statement information, and other publicly available information. Five areas of each banking organization are rated: bank subsidiaries, other nonbank subsidiaries, parent company, earnings and capital adequacy. Therefore, the composite supervisory rating is known by the acronym BOPEC. Banking organizations with a composite supervisory rating of 1 or 2 are considered the safest and most well-managed institutions by bank supervisors. And,bankingorganizationswithcompositesupervisoryratings of3,4,or5havemoderatetosubstantialdeficienciesthatwereuncovered during the examination process. From these ratings we constructed two indicator variables: (1) BOPEC2, equaled one if the composite supervisory rating equaled 2, and zero otherwise; and, (2) BOPEC345, equaled one if the composite supervisory rating equaled 3, 4, or 5, and zero otherwise. Supervisory ratings, BOPEC2 and BOPEC345, are the most private information among our explanatory variables. We assume that they are a reasonable statistic for the banks’ own knowledge of their riskiness. Althoughitcouldbearguedthatthesevariablesareproprietaryinformation of the bank, Krainer and Lopez (2002) have presented empirical evidence that demonstrates that balance sheet, income statement, and stock price information can be used to predict both supervisory ratings and their changes four quarters prior to their assignment. This finding suggests that these ratings are highly correlated with information that is available to sophisticated investors. 3. Business and Bond Market Conditions We used three quarterly measures of business and bond market conditions, which are publicly available. First, aggregate business conditions 19Data on the market value of common stock are available on the Center for Research in SecurityPrices(CRSP)tapethatispublishedbytheUniversityofChicagoGraduateSchool ofBusiness. 13

were proxied by the unemployment rate (UE). This measure was chosen because it is a lagging indicator of business conditions as are bank lendingactivities: Bothcommercialandindustrialloansoutstandingand the ratio of consumer installment credit to personal income are lagging indicators of business conditions.20 Second, the market risk premium was proxied by contemporaneous stock market excess returns (XR). In each quarter, this premium was measuredusingthequarterlyaverageof dailyexcess stockreturns (calculated as the difference between the daily value-weighted return on NYSE, Amex,andNasdaqstocksandtheoff-the-runonemonthTreasuryreturn). And third, bond market conditions were proxied by an implied stock volatilitymeasure(MKTVOL). Thismeasurewasbasedonreal-timeS&P 100 (OEX) index option bid/ask quotes supplied by the Chicago Board Options Exchange.21 3.2 The empirical model 3.2.1 The issuance decision Wepositthatdebtissuancedecisionsdependnotonlyonpubliclyavailablerisk proxies for each banking organization (NATA , PDTA , OREO , AGAP , it it it it and, MKTLEV ), private information (BOPEC2 and BOPEC345 ) as well it it it as business and bond market conditions (UE , XR ,and MKTVOL ), but also t t t on banking organization-specific characteristics such as recent debt issuance activities,banksize,andpotentialtaxbenefits.22 Recentdebtissuanceactivities were allowed to vary across seniority grades (g, where g = E for senior debt and g = U for subordinated debt) for each banking organization. For each senioritygrade,weconstructedanindicatorvariable,ISSUE ,thatequaled g,i,t 1 one when the banking organization issued debt grade g in the p−revious period, and zero otherwise.23 And, banking organization size was measured using the natural log of total assets, ln(ASSET ).24 it Tax shelter benefits for corporate debt imply that a banking organizations’ marginal tax rate will influence its choices with respect to debt or equity issuance. Asaproxyforthemarginaltaxratefacingeachbankingorganization, we used its foreign and domestic income taxes as a percentage of net income (AVGTAX ). Andtoaccountfordifferencesincapitalstructureacrossbanking it organizations, we used the ratio of book equity to book total assets (K/A ) in it the issuance decision model. Itisalsoexpectedthattheinteractionbetweenbankingorganization-specific risksandthemarketriskpremiumwouldalsoinfluencebondissuancedecisions. 20See The Conference Board,U.S.Composite Indexes. 21Impliedstockvolatilityisexogenousto,buthighlycorrelatedwith,bondmarketvolatility. 22To construct the issuance decision variables, ISSUEg,it, the CUSIP Masterfile was used toidentifyallseniorandsubordinateddebtissuesbylargeU.S.bankingorganizations. Then, foreachdebtissue,issuancedateswereassignedusingMoodys’,Fitch,Bloomberg,andWarga databases. 23More explicitly, ISSUEg,i,t 1 equals one if banking organization i issued debt type g in eitherquartert 2,orquarter−t 3,and zerootherwise. 24Thisproxyw − illalsodetectth − eriskreductiontypicallyachievedbygreaterdiversification atlarger firms. 14

Our issuance decision equations for both senior and subordinated debt are assumed to be linear in all of the variables, or:25 ISSUE =β +β NATA +β PDTA +β OREO +β AGAP g,i,t 0 1 it 2 it 3 it 4 it +β MKTLEV +β XR +β XR NATA +β XR PDTA 5 it 6 t 7 t • it 8 t • it +β XR OREO +β XR AGAP +β XR MKTLEV 9 t • it 10 t • it 11 t • it +β BOPEC2 +β BOPEC345 +β ISSUE 12 it 13 it 14 g,i,t 1 +β 15 ln(ASSET it )+β 16 AVGTAX it +β 17 K/Ait − +β MKTVOL +β UE . 18 t 19 t (24) These equations were estimated using standard latent variable techniques, which treat the decision to issue as a continuous unobserved variable that represents the probability that a banking organization issues debt type g. Each of these probit models was estimated using quarterly data for the largest U.S. banking organizations forthepre-FDICIAperiod(1985-1992) and forthepost- FDICIA period (1993-2002). To ascertain whether parameter estimates were statistically different across the two debt seniority grades in each period, we estimated a “stacked data issuance decision model.” That is, we created a combined data set for each banking organization that stacked its subordinated debt data set above its seniordebtdataset(foreachquarter)sothatthecombineddatasetcouldbeused toestimateanexpandedmodel. The“stackeddataissuancedecisionmodel”for each banking organization has as the dependent variable the stacked issuance decision indicator variables for subordinated debt (ISSUE ) and for senior U,t debt (ISSUE ). The explanatory variables included the original set of ex- E,t planatory variables (i.e., (NATA, PDTA, OREO, AGAP, and MKTLEV), XR, (XR NATA, XR PDTA, XR OREO, XR AGAP, and XR MKTLEV), (IS- • • • • • SUE , ln (ASSETS), AVGTAX, K/A, BOPEC2 and BOPEC345), (MKg,t 1 TVOL−and UE)) and each of those explanatory variables interacted with a senior grade indicator variable, I, that equaled one when g was senior and zero when g was subordinated. With this specification, the parameter estimates on the interacted explanatory variables were significant only when the individual parameter estimates for the original issuance decision model are statistically different when senior debt market data, rather than subordinated debt market data, are used. 3.2.2 Issuance spreads Issuance spreads are likely to depend on many of the same factors that influenceissuancedecisions,suchastheissuingbankingorganizations’risk(NATA , it PDTA , OREO ,AGAP ,, MKTLEV BOPEC2 and BOPEC345) the marit it it it ket risk premium (XR ), bond market conditions (MKTVOL ), the frequency t t of debt issuance (ISSUE ), and bank size (ln(ASSET )). In addition, g,i,t 1 it − 25For continuous right hand side variables, the average value for a two quarter interval wasused,and forbinary righthand sidevariables,the averageofthe appropriateunderlying variable over two quarters was used. The left hand side variable was set equal to one if the bank issues in a two quarter period and zero otherwise. To enhance the exogeneity of the righthand side variables,explanatory variables were lagged by one quarter. 15

observed issuance spreads will likely depend on the instrument characteristics of the bonds that were issued, such as call options, time to maturity, coupon frequency, and the amount issued.26 Because callable debt allows a firm to refinance it when the rate becomes lower, we included in our spread model an indicatorvariable,CALL,thatequaledonewhenanissuehadacalloption,and that equaled zero otherwise. To capture non-standard maturity effects on spreads, we used two indicator variables: (1)anindicatorvariableforbondsissuedwithamaturitylessthan10 years,MATLT10,and(2)anindicatorvariableforbondsissuedwithamaturity greaterthan20years,MATGT20. Eachoftheseindicatorvariablesequaledone for the specified maturity range, and zero otherwise. Itseemsreasonablethatcouponfrequencycouldaffectthetypesofinvestors willing to purchase an instrument. To capture this potential effect on debt spreads,weincludedtwoindicatorvariables,COUPON12 andCOUPON2,that equaled one when the coupon frequency is monthly and semi-annually, respectively, and that equaled zero otherwise. In addition, issuancespreadsmaydependon theamount of the debt instrumentsthatwasissued. Tocapturethispotentialeffect,weincludedtheissuance size of the debt instrument (ISSUESIZE). The issuance spread on each bond was calculated using derived bond yields computed by the Newton-Raphson iterative method from issuance prices and an interpolated Treasury yield of the same maturity.27 The term structure of Treasury interest rates was identified on each issuance date by using a smoothing spline of the forward rate curve that incorporated a “roughness” penalty determinedbygeneralizedcrossvalidation. Thisspliningtechniqueisdescribed in Fisher, Nychka, and Zervos (1995). Because issuance spreads are only observed for those banking organizations thatactuallychosetoissuedebt,itisimportanttouseasampleselectionmodel toanalyzesuchspreads. WeusedHeckman’stwo-stagemethodtoobtainconsistent estimates of our sample selection model: This method involved estimating the issuance decision equation described above (estimated with probit), and then using the inverse Mills ratio function of the probit residuals as an extra variable in a regression for the observed issuance spreads. The regression estimated for observed issuance spreads over Treasury securities with comparable maturities for our two debt seniority grades was: 26Instrument characteristics for each subordinated and senior bond were identified using Moody’sDefaultRiskService(DRS)database,FitchInvestmentSecuritiesDatabase,Warga andBloombergdatabases,aswellasmonthlyissuesofMergentBondRecord overtheJanuary 1984-December2002 period,inclusive. 27Issuance prices were obtained from various sources including Bloomberg and Moody’s DRS. Yields for floating rate bonds are calculated by contract (i.e., a pre-specified number of basis points above the LIBOR rate). The appropriate SWAP rate, from Bloomberg, on the issuance date was used to convert the yield on each floating rate bond to its fixed rate equivalent. 16

SPREAD =β +β NATA +β PDTA +β OREO +β AGAP g,i,t 0 1 it 2 it 3 it 4 it +β MKTLEV +β XR +β XR NATA 5 it 6 t 7 t • it +β XR PDTA +β XR OREO +β XR AGAP 8 t • it 9 t • it 10 t • it +β XR MKTLEV +β BOPEC2 +β BOPEC345 11 t • it 12 it 13 it +β ISSUE +β ln(ASSET )+β MKTVOL 14 g,i,t 1 15 it 16 t +β CALL −+β MATLT 10 +β MATGT 20 17 it 18 it 19 t +β COUPON12 +β COUPON2 +β ISSUESIZE 20 it 21 it 22 +β MILLSRATIO . 23 it (25) Aswasdonewiththeissuancedecisionmodel,weestimateda“stackeddata sampleselectionmodel”toascertainwhetherparameterestimateswerestatistically different across the two debt seniority grades in each period. This stacked model had as its dependent variable the stacked issuance spreads for subordinated debt (SPREAD ) and for senior debt (SPREAD ). The explanatory U,t E,t variables included the original set of explanatory variables (i.e., those indicated in equation (25)above)and each of thoseexplanatoryvariables interacted with a senior grade indicator variable, I, that equaled one when g was senior and zerowheng wassubordinated. Withthisspecification,theparameterestimates on the interacted explanatory variables were significant only when the parameter estimates for the original sample selection model for senior debt were statistically different from those for subordinated debt. 3.3 Regulatory regimes The key determinants of subordinated debt spreads — bank failure probabilities and senior debt recovery rates — are heavily influenced by regulatory regimes. Themostrelevantchangeswithinoursampleperiodaretheintroductionofthe Federal Deposit Insurance Corporation Improvement Act (FDICIA) in 1991, shortly followed by the enactment of depositor preference in 1993. To account for this regulatory reform, we split our sample accordingly and refer to the two subperiods as the pre-FDICIA (1985-1992) and post-FDICIA (1993-2002) periods. Bank failure rates in the pre-FDICIA period were generally higher than in thepost-FDICIAperiod. Thisislikelyduetolessfavorablebusinessconditions or to less timely corrective action by bank supervisors. Given a bank failure, investorsexpectedlowerrelativelossesinthisperiodduetofrequentresolutions through purchase-and-assumption. Thus, reported recovery rates of about 25 percentforseniordebtinthisperiodmaysomewhatunderestimatetruerecovery rates inclusive of purchase and assumption resolutions. Thepost-FDICIAperiodischaracterizedbypromptcorrectiveaction(PCA),28 higher barriers against bank bailouts, and depositor preference. These regulatory changes have different impacts on the variables of our model. Default frequencies most likely have been reduced by PCA and the resulting regulatory pressureonbanks toraisecapital beforetheybecomeinsolvent. Theimpacton recovery rates, once default does occur, is less favorable for investors. Under depositor preference, depositors (including the FDIC in receivership) are given a priority claim on a failing bank’s assets. The holders of non-deposit liabilities 28The capital-based policy ofpromptcorrectiveaction began in December1992. 17

only share what remains once depositor claims are paid off. Most of all, senior investorsseemtobeworseoffinabankfailurethantheytheywerepriortothe implementation of depositor preference. Subordinated investors have a weaker claimaswell,buttheycouldnotexpecttogetmuchevenbeforetheimplementation of depositor preference. With respect to risk characteristics, senior debt thus is similar to subordinated debt in the post-FDICIA period.29 3.4 Hypotheses 3.4.1 The issuance decision Expected Signs of the Parameter Estimates: Althoughthemagnitudeofparameterestimatesmayvaryacrosstheregulatory regimes, the expected signs of parameters in our issuance decision model are generally the same for both senior and subordinated debt. The expected signs of parameters for publicly available risk measures (NATA , PDTA ,OREO , it it it AGAP ,and, MKTLEV ) are negative because each of these measures reflect it it greaterriskand/oradeterioratingfinancialconditionforthebankingorganization. The expected sign of the parameters for stock market excess returns (XR) is negative because this measure rises when the market premium for risk rises. Moreover, the expected signs of parameters for risk measures interacted with stock market excess returns (XR NATA, XR PDTA, XR OREO, XR AGAP, • • • • and XR MKTLEV) are negative since a banking organization experiencing fi- • nancial distress may be much less likely to issue debt in a period when the market risk premium is high compared to when this premium is lower. In contrast, the expected signs of parameters for banking organizationspecific factors ((ISSUE ), ln(ASSET ), AVGTAX , K/A , BOPEC2 g,i,t 1 it it it it and BOPEC345 ) are posit−ive. It is conventional wisdom that issuance costs it are lower for firms that frequently issue debt.30 In addition, issuance costs may be lower for larger firms because major debt market investors typically specialize in gathering information on a small number of firms.31 Holding a banking organization’s capital structure (i.e., K/A ) constant, the higher its it tax rate, the more it benefits from being able to deduct the interest payments paidtobondholders. Atthesametime,thelessleveragedthefirmis, themore likelyitcanissuedebt,sobondissuanceactivitieswouldbepositivelycorrelated withK/A .Regardlessofthepotentialtaxbenefitsderivedfromdebtissuance, it bank supervisors may pressure a banking organizations’ management to raise regulatorycapital. U.S.bankingorganizationsaresubjecttoCapitalAdequacy Guidelines that allow some debt instruments (e.g., mandatory convertible debt securities andtermsubordinateddebt)tobeincluded insupplementarycapital (i.e., tier 2 capital). This means that not all debt instruments are equal from a supervisory perspective when a banking organization is under duress. Moreover, core capital elements (e.g., common stockholders equity) are preferred to any debt instrument in such circumstances. We would expect that banking 29We neglect that expected recovery rates on senior debt also depend on the fraction of subordinated and of senior debt to total liabilities as well as on the expected shortfall of assets with respectto liabilities. 30See Board ofGovernors ofthe FederalReserve System (1999,p. 46.) 31See Board ofGovernors ofthe FederalReserve System (1999,p. 47.) 18

organizations with a composite supervisory rating of 1 would be least likely to be under some pressure to improve their total regulatory capital. Andtheexpectedsignsofparametersforbondmarketconditionsandmacroeconomic conditions (MKTVOL and UE ) are negative. This is because bond t t market liquidity premiums tend to rise correspondingly with MKTVOL and t because bank lending activities are highly correlated with lagging indicators, such as unemployment. The Pre-FDICIA Period: Our “informed investor model” would imply that the bank borrows from uninformedinvestors(i.e.,issuesseniordebt)whenthereisbadnewswithrespectto the financial prospects of the firm. This suggests that the parameter estimates for accounting- and market-based risk variables would jointly be positive in the issuance decision model for senior debt. Incontrast,ourmodelwouldimplythatthebankborrowsfrominformedinvestors (i.e., issues subordinated debt) when there is good news with respect to the financial prospects of the banking organization. This suggests that the parameterestimatesforaccounting-andmarket-basedriskvariableswouldjointly be negative in the issuance decision model for subordinated debt. Moreover, in suchcircumstancesthebankingorganizationwouldnotborrowfromuninformed investors, i.e., it would not issue senior debt. We thus postulate that subordinated debt would be issued in times when private information contains good news, and senior debt would be issued in timeswhenprivateinformationcontainsbadnews. Ifthisistrue,ofcourse,the rival model where subordinated debt is issued when a bank has adverse private information (Barclay and Smith, 1995) is not plausible. The Post-FDICIA Period: In principle, prompt corrective action would reduce the frequency of bank defaults. Consequently, wewould expect issuancedecisionsforbothsubordinated and senior debt to be less risk sensitive in the post-FDICIA period compared to the pre-FDICIA period. At the same time depositor preference lowered the liquidation standing of senior creditors and made it more likely that such investors would incur losses in the event of a bank failure. In this case, we would expect that the risk sensitivity of issuance decisions would become similar for senior and subordinated debt. 3.4.2 Issuance spreads Expected Signs for the Parameter Estimates Intheissuancespreadregression,parameterestimatesonbankingorganizationspecific risk proxies are expected to be generally positive. Even uninformed investors should respond to risk proxies that are widely accessible and readily attainable (e.g., OREO and MKTLEV ). Thus, such risk proxies are exit it pected to be positively correlated with observed issuance spreads, regardless of the seniority of the debt instrument issued. In contrast, only informed investors should respond to risk proxies that are more private in nature (e.g., BOPEC2 and BOPEC345 ). Such risk proxies are expected to be positively it it correlated with observed subordinateddebt spreads, but not toinfluencesenior debt spreads. 19

Themarketpriceofriskisexpectedtobepositivelycorrelatedwithbanking organizaton debt spreads. If senior debt market investors are not risk averse, then we would expect the parameter on the market price of risk to be insignificant. But, if these investors are risk averse, the market price of risk parameter would be positive and significantly influence both senior and subordinated spreads. Similarly,investorriskaversionwouldinfluencethesignificanceofthe bank-specific risk measures interacted with the market price of risk. Positive parametersonthesevariables—separatelyortogetherasagroup—wouldimply that issuance spreads were relatively higher for riskier banking organizations in periods when the market price of risk was high. Frequentissuersandlargerbankingorganizationsareexpectedtohavelower issuance spreads than other banking organizations, holding all else constant. Thisisbecauseinvestorstypicallyfollowonlyahandfuloffirmsandtheyprefer to hold issues that are relatively liquid. Larger firms are able to issue debt in larger issue sizes more frequently, and thus their debt issues tend to be quite liquid. Spreads tend to increase in periods when bond market volatility rises.32 It is therefore expected that the parameter estimate for MKTVOL in the spread t regression would be positive. In addition, the observed issuance spreads for both senior and subordinated debtissuesareexpectedtobeinfluencedbythecharacteristicsoftheindividual instruments that are issued. First, consider a firm that issues two instruments simultaneously where one has a call option and the other does not. The instrument with the call option would be expected to have a larger spread than the one that does not. Thus, it is expected that the sign on CALL would be positive.33 It may be the case that bonds with non-standard maturities are less liquid than thosewithstandard maturities.34 This would suggest that theparameter estimatesoftheindicatorvariablesformaturitieslessthan10years(MATLT10) and for maturities greater than 20 years (MATGT20) would be expected to be positive. In contrast, smaller issues tend to rapidly get absorbed into investor portfolios, such issues tend to be less liquid in the secondary market.35 For this reason, smaller issues are more difficult and expensive to sell to institutional investors.36 Therefore, it is expected that issuance spreads are likely to be negatively correlated with issuance size (ISSUESIZE). 32See Covitz,Hancock,and Kwast,2001. 33A negativeorzerocoefficienton thecalloption indicatorvariablewouldimplythatdebt holders did notvaluethecalloption appropriately. 34Non-standard maturity instruments may be issued by banking organizations to match thedurationoftheirliabilitieswiththedurationoftheirassets,ortheseinstrumentsmaybe issued when an organization wants to attractfundsfrom smallretailinvestors. 35Hancock and Kwast (2001) present histograms of weekly subordinated debt spread discrepancies between Bloomberg and Interactive Data Corporation pricing data sources over the January 1997 to October 1999 period for bonds stratified by issuance size. The tightest distributionofspreaddiscrepanciesisforbondswithissuancesizesgreaterthan$300million. The next tightest distribution was for bonds with issuance sizes between $100 million and $300 million. And, the widest distribution was for bonds with issuance sizes less than $100 million. Thedecreaseddispersioninspreaddiscrepanciesforlargerissuessuggeststhatthere may be a positive correlation between the flow of trade in a particular bond and its amount outstanding atissuance. 36See Board ofGovernors ofthe FederalReserve System (1999,p. 46.) 20

Itseemsreasonablethathighercouponfrequencies(e.g.,monthlypayments) wouldattractsmaller“retail”investers, andtheresultinghigherdemandwould lower the issuance spread. Therefore, the expected sign for the parameter estimate of the monthly coupon payment indicator (COUPON12) would be negative. The Pre-FDICIA Period: Depositor preference rules were not taken for granted in this period, so senior debtholdersaredistinctfromsubordinateddebtholders. Inthiscase,wewould expectseniordebtholderstofocusonthepubliclyavailableandeasilyaccessible information. This is because senior debt holders have less of an incentive to become informed when they are likely to rank with (uninsured) depositors in this period. Senior spreads would also react to the market price of risk if senior investorsareriskaverse. If adifferenceininvestors’beliefs, notriskaversion, is themainreasonwhybanksissuedualclassdebt,wewouldexpectnosignificant influence of the market price of risk on spreads. Subordinated debt spreads, would also be sensitive to proxies of public risk (includingthemarketpriceofriskifseniorinvestorsareriskaverse),ashasbeen shown in our model.37 At the same time, subordinated debt holders would, of course,focusontheirprivateriskproxiesthatarequitecostlyforthemtoobtain. In their view, observed subordinated debt spreads would be unduly influenced by the publicly available information and not fully reflect their own private information. Thissuggeststhattheparameterestimatesontheinteractedpublic risk variables with the senior indicator in the stacked spread regression would not be significantly different from zero. But, the parameter estimates on the private risk variables would only be significant for subordinated debt spreads, since these instruments are purchased by sophisticated investors. In the spread regression for senior debt spreads, we would expect the parameter estimates on the private risk variables to not be statistically significant from zero. Our model also suggests that spreads are more sensitive to risk measures when both publicly and privately perceived risks are relatively high and when both increase at the same time. In terms of the spread regression, we would expectsignificanteffectsforthebankingorganization-specificriskproxiesinteracted with the market price of risk. The Post-FDICIA Period: After depositor preference legislation, all public debt holders are subordinated to depositors, even those that are “senior” debt holders. Depositor preference rules are thus expected to lower the recovery rate on senior debt in this period. Thisimpactofdepositorpreferencewouldlikelyincreasetheincentivepremium contained in subordinated debt spreads. Hence, such spreads would be more sensitive to the information used by senior investors, i.e., to publicly available and accessible information on banking organization risk, and less sensitive to private information gathered by informed investors. Atthesametime,thepremiumbetweensubordinatedandseniordebtwould become small as these instruments become more similar in their standing. We would expect differences in risk sensitivity between senior and subordinated debt to be lower in this period. 37Given pre-FDICIA recovery rates on senior debt of around 25 percent, we would expect the subordinated debtspread to be quite sensitive topublicly perceived bank risk. 21

These expectations are derived under the assumption that senior depositors stillrelyonthesamekindofpublicinformationinthisperiod. Inalonger-term perspective though, one should expect senior investors to become aware of the weaker standing of their claims under depositor preference. This might give these investors a stronger incentive to collect more private information on the quality of banking organizations. Both, senior and the subordinated spreads, wouldthusbecomelessresponsivetopublicinformationandmoreresponsiveto private information. Investors who do not find it worthwhile to collect private information would likely replace senior debt in their portfolio by instruments with higher rank in case of a bank failure such as certificates of deposit. 3.5 Findings 3.5.1 Trends in the data and distributions of spreads Before turning to our parameter estimates, it is instructive to take a brief look at the bond market data for large U.S. banking organizations over the 1985 to 2002 period, inclusive. First, it is readily apparent that the number of largebankingorganizationsthatissueseniorandsubordinateddebtinstruments varies considerably across time (Figure 3.1). Interestingly, it appears that fewerbankingorganizationsissueseniordebtduringperiodswhenmorebanking organizations issue subordinated debt. The number of public debt issues made by large U.S. banking organizations increased dramatically in the mid-1990s (Figure 3.2). In the late 1980s, it was not uncommon for such organizations to collectively issue less than 20 debt instruments in a year of each of the seniority grades. But, by the late 1990s, there were several years in which large U.S. banking organizations collectively issued more than 100 instruments in each seniority grade. With the increase in the number of debt issues made by large U.S. banking organizationsperannum,therewaslessstandardizationinthecontracttermsfor bothsubordinatedandseniorbonds. Table3.1presentsinformationonselected contract terms and the number of subordinated instruments issued for each year over the 1985-2002 period. In the mid-1980s, subordinated instruments typically were issued with a maturity in the 10 to 20 year range, with no call option, with semi-annual coupon payments, and with a fixed rate of interest. By the late-1990s, however, a higher proportion of subordinated instruments wereissuedwithmaturitieslessthan10years,withcalloptions, withafloating rate, and with coupons that were paid either monthly or quarterly. Similar patterns emerge from consideration of the information contained in the Table 3.2. Compared to the recent past, senior debt instruments issued by large banking organizations in the early years of the sample period were less likely to have a maturity of less than 10 years, and more likely to pay coupons semiannuallyatafixedrateofinterest. AcomparisonofTables3.1and3.2suggests that subordinated instruments tend to have longer maturities than those of seniorinstruments, andthat floatingratecontractsaremoreprevalentnearthe end of the sample period for both seniority grades. Figure 3.3 contains boxplots of observed issuance spreads for bonds in each senioritygradeforeachyearinthe1985:Q1to2002:Q4period. Theseboxplots are graphical representations of the center and width of spread distributions along with outliers. The height of each box is equal to the interquartile width, 22

Table 3.1 1 Characteristics of Subordinated Debt Instruments Issued by Large US Banking Organizations Annual Data, 1985-2002 Pre-FDICIA Period Post-FDICIA Period 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 Maturity Less than 10 years 9.09% 16.67% 12.50% 0.00% 20.00% 10.00% 44.74% 32.79% 16.36% 43.48% 38.74% 12.04% 3.13% 5.88% 7.27% 6.25% 5.38% 9.49% 10-20 years 63.64% 33.33% 84.38% 100.00% 70.00% 80.00% 50.00% 67.21% 81.82% 55.07% 53.15% 80.56% 79.17% 61.76% 87.27% 93.75% 59.14% 54.01% Greater than 20 years 27.27% 50.00% 3.13% 0.00% 10.00% 10.00% 5.26% 0.00% 1.82% 1.45% 8.11% 7.41% 17.71% 32.35% 5.45% 0.00% 35.48% 36.50% Call Option Yes 36.36% 100.00% 34.38% 28.57% 10.00% 0.00% 2.63% 0.00% 0.00% 62.32% 70.27% 48.15% 54.17% 50.00% 52.73% 9.38% 0.00% 0.00% No 63.64% 0.00% 65.63% 71.43% 90.00% 100.00% 97.37% 100.00% 100.00% 37.68% 29.73% 51.85% 45.83% 50.00% 47.27% 90.63% 100.00% 100.00% Coupon Frequency Monthly 9.09% 0.00% 9.38% 0.00% 0.00% 0.00% 0.00% 0.00% 0.00% 47.83% 46.85% 56.48% 56.25% 32.35% 38.18% 50.00% 26.88% 13.87% Semi-Annual 36.36% 100.00% 87.50% 85.71% 100.00% 100.00% 97.37% 96.72% 83.64% 50.72% 53.15% 41.67% 31.25% 60.29% 45.45% 50.00% 70.97% 83.94% Quarterly 54.55% 0.00% 3.13% 14.29% 0.00% 0.00% 0.00% 3.28% 16.36% 1.45% 0.00% 1.85% 5.21% 1.47% 10.91% 0.00% 2.15% 2.19% Zero Coupon 0.00% 0.00% 9.38% 0.00% 0.00% 0.00% 2.63% 0.00% 0.00% 0.00% 0.00% 0.00% 7.29% 5.88% 5.45% 0.00% 0.00% 0.00% Amount Issued (in millions of dollars) Maximum 200.00 250.00 300.00 300.00 400.00 200.00 750.00 500.00 600.00 500.00 443.40 500.00 800.00 750.00 1000.00 3000.00 3000.00 1000.00 Minimum 75.00 100.00 50.00 55.00 65.00 100.00 10.00 100.00 75.00 1.15 1.46 1.46 2.84 1.96 10.00 15.00 1.52 0.90 Mean 120.45 166.67 165.47 165.00 172.00 138.70 154.87 189.05 193.18 102.44 81.14 94.11 101.67 126.16 111.10 215.16 91.00 50.22 Median 100.00 150.00 175.00 150.00 150.00 118.50 100.00 200.00 200.00 100.00 25.00 25.00 25.00 25.00 25.00 25.00 26.02 8.22 TOTAL 1325.00 1000.00 5295.00 1155.00 3440.00 1387.00 5885.00 11532.00 10625.00 7068.50 9006.81 10164.24 9763.17 8579.18 6110.47 6885.00 8462.72 6879.68 Medium Term Notes2 9.09% 0.00% 12.50% 14.29% 0.00% 0.00% 15.79% 0.00% 0.00% 49.28% 63.06% 65.74% 77.08% 64.71% 80.00% 81.25% 91.40% 92.70% Floating Rate 63.64% 0.00% 0.00% 0.00% 0.00% 0.00% 0.00% 1.64% 0.00% 0.00% 0.00% 0.00% 3.13% 1.47% 1.82% 0.00% 2.15% 1.46% Total Number Issued per Annum 13 6 32 7 20 11 38 62 56 69 114 110 97 75 58 35 94 137 1 In each quarter, a banking organization was included in our sample only if it was in the "top 50" after all U.S. bank holding companies were ranked by total assets. 2 Medium term notes (MTN) are typically those that are $75 million USD or less outstanding at issuance. We also used Fitch's indicator for MTNs.

Table 3.2 1 Characteristics of Senior Debt Instruments Issued by Large US Banking Organizations Annual Data, 1985-2002 Pre-FDICIA Period Post-FDICIA Period 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 Maturity Less than 10 years 42.86% 73.33% 75.00% 100.00% 91.67% 100.00% 100.00% 100.00% 94.12% 100.00% 98.41% 90.43% 90.91% 92.03% 94.35% 97.80% 98.39% 89.29% 10-20 years 57.14% 23.33% 25.00% 0.00% 8.33% 0.00% 0.00% 0.00% 5.88% 0.00% 1.59% 8.70% 6.67% 2.90% 0.81% 1.10% 0.00% 10.71% Greater than 20 years 0.00% 3.33% 0.00% 0.00% 0.00% 0.00% 0.00% 0.00% 0.00% 0.00% 0.00% 0.87% 2.42% 5.07% 4.84% 1.10% 1.61% 0.00% Call Option Yes 42.86% 26.67% 10.00% 0.00% 8.33% 0.00% 0.00% 0.00% 11.76% 7.69% 4.76% 8.70% 6.67% 0.72% 0.81% 0.00% 0.00% 0.00% No 57.14% 73.33% 90.00% 100.00% 91.67% 100.00% 100.00% 100.00% 88.24% 92.31% 95.24% 91.30% 93.33% 99.28% 99.19% 100.00% 100.00% 100.00% Coupon Frequency Monthly 0.00% 3.33% 0.00% 4.17% 16.67% 0.00% 0.00% 0.00% 0.00% 3.85% 1.59% 5.22% 5.45% 7.25% 7.26% 25.27% 9.68% 29.76% Semi-Annual 57.14% 96.67% 100.00% 91.67% 58.33% 100.00% 100.00% 64.29% 70.59% 38.46% 44.44% 48.70% 51.52% 15.94% 25.81% 24.18% 25.81% 57.14% Quarterly 42.86% 0.00% 0.00% 4.17% 25.00% 0.00% 0.00% 35.71% 29.41% 57.69% 53.97% 46.09% 40.00% 72.46% 64.52% 50.55% 59.68% 9.52% Zero Coupon 0.00% 0.00% 0.00% 0.00% 0.00% 0.00% 10.00% 0.00% 0.00% 0.00% 0.00% 0.00% 1.82% 4.35% 1.61% 0.00% 1.61% 3.57% Amount Issued (in millions of dollars) Maximum 300.00 300.00 250.00 350.00 250.00 225.00 200.00 350.00 400.00 430.00 300.00 500.00 350.00 1375.00 1500.00 2500.00 2500.00 1500.00 Minimum 50.00 50.00 50.00 100.00 100.00 150.00 100.00 100.00 100.00 7.48 50.00 0.40 0.18 3.00 1.00 0.25 1.46 0.08 Mean 155.36 153.33 149.50 187.50 158.33 187.50 130.00 167.86 250.00 120.48 133.02 79.72 53.68 132.07 188.38 258.96 257.14 134.64 Median 137.50 150.00 137.50 200.00 150.00 187.38 125.00 125.00 250.00 100.00 100.00 50.00 20.00 75.00 90.00 100.00 58.04 4.67 TOTAL 2175.00 4600.00 2290.00 4500.00 1900.00 375.00 1300.00 2350.00 4250.00 3132.48 8380.00 9167.90 8856.86 18225.85 23547.17 24082.89 15942.48 11309.71 Medium Term Notes2 7.14% 13.33% 15.00% 0.00% 0.00% 0.00% 0.00% 0.00% 0.00% 34.62% 33.33% 68.70% 77.58% 54.35% 49.19% 40.66% 54.84% 71.43% Floating Rate 50.00% 13.33% 0.00% 8.33% 41.67% 0.00% 0.00% 35.71% 29.41% 61.54% 57.14% 56.52% 48.48% 80.43% 71.77% 75.82% 64.52% 11.90% Total Number Issued per Annum 20 40 24 24 14 2 10 16 18 34 64 118 167 138 124 94 62 84 1 In each quarter, a banking organization was included in our sample only if it was in the "top 50" after all U.S. bank holding companies were ranked by total assets. 2 Medium term notes (MTN) are defined as issues that are $75 million USD or less outstanding at issuance.

Figure 3.1 The Number of Large US Banking Organizations that Issued Senior and Subordinated Debt Numberof BankingOrganizations 24 24 Senior 22 22 Subordinated 20 20 18 18 16 16 14 14 12 12 10 10 8 8 6 6 4 4 2 2 0 0 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 Year (Quarterly Data) Figure 3.2 The Number of Senior and Subordinated Debt Instruments that were Issued by Large U.S. Banking Organizations NumberofIssues 180 180 160 Senior 160 Subordinated 140 140 120 120 100 100 80 80 60 60 40 40 20 20 0 0 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 Year (Annual Data) BusinesscyclepeaksandtroughsasidentifiedbytheNationalBureauofEconomicResearchareindicatedbyshading.

which is the difference between the third quartile and the first quartile of the data. This width widened in the period prior to 1992 for both senior and subordinated debt spreads and then narrowed considerably when the financial conditionoflargeU.S.bankingorganizationsimproved. Indeed,inthemiddleto late1990sthetopquartilesforseniorandforsubordinatedissuancespreadsare below the medians for such spreads (which are represented by bold horizontal lines in the interior of each box) that were observed in 1991. The brackets ([ ]) for each box plot are located at extreme values of the data for the year or at a distance equal to 1.5 times the interquartile distance from the center, whichever is less.38 In most years, the top bracket for subordinated issuance spreadsishigherthanthecorrespondingtopbracketforseniorissuancespreads withexceptionsoccurringin1990,1991,1996,1998,2001and2002. Throughout the 1992 to 1999 period, these upper brackets remained considerably below the medians for senior and subordinated issuance spreads observed in 1991. 3.5.2 The issuance decision The issuance decision model was estimated using quarterly data for the pre- FDICIA (1985:Q1-1992:Q4) and post-FDICIA (1993:Q1-2002:Q4) periods. Table 3.3 presents parameter estimates for the issuance decision probit models for subordinated and senior debt, equation (24). Dependent variables for the separate probit models are the issuance decision indicator variables for subordinated and senior debt, respectively. The left column of the table provides a short description for each of the explanatory variables. The left panel of the table presents estimates for the issuance decision model for subordinated debt during each of the time periods considered, and the right panel of the table presentsestimatesfortheissuancedecisionmodelforseniordebtduringeachof the time periods considered. And, Table 3.4 presents parameter estimates for the stacked data issuance decision model, where the dependent variable for the stacked data issuance decision model is the stacked issuance decision indicator variables for subordinated (ISSUE ) and for senior debt (ISSUE ). Pa- U,t E,t rameter estimates for the explanatory variables interacted with a senior grade indicator variable, I, in Table 3.4 are significant only when the individual parameter estimates for the original issuance decision models (in Table 3.3) are statistically different when senior debt market data, rather than subordinated debt market data, are used. Thus, parameter estimates in the stacked regressioncanbeusedtoinfera“directeffect”forsubordianteddebt(leftpanel)and an “additional effect” for senior debt (right panel) in each deposit insurance regime. Interestingly, there are some banking organization- or issue- specific factors thatinfluencetheprobabilityofissuanceindependentlyofthetypeofinstrument or the period considered. For example, ln(ASSETS) and ISSUE are always -1 positive. Thismeansthatrelativelylargebankingorganizationsamongthe50 largest,andorganizationsthathaveissuedinthepastsixmonths,aremorelikely toissuedebt. Inaddition,KAisalwaysnegativeforsubordinateddebt, andin the pre-FDICIA period for senior debt (though not significant). This suggests, albeit in the weakest manner, that higher capitalized bankingorganizations are 38FordatahavingaGaussiandistribution,approximately99.3percentofthedatafallinside the brackets. Horizontal dashes represent “unusually deviant data points” that are further than 1.5 times the interquartiledistance from the centerofthebox. 23

Table 3.3 Parameter Estimates for the Issuance Decision Model of Large U.S. Banking Organizations Dependent Variable / Deposit Insurance Regime Explanatory Variables Subordinated Issuance Senior Issuance Pre-FDICIA Expected Post-FDICIA Expected Pre-FDICIA Expected Post-FDICIA Expected 85:Q1-92:Q4 Sign? 93:Q1-02:Q4 Sign? 85:Q1-92:Q4 Sign? 93:Q1-02:Q4 Sign? Accounting- and Market- based Risk Measures The ratio of non-accruing loans to total assets (NATA) -4.389 X -15.983 X -21.964 X 24.391 (-0.69) (-1.23) (-2.95) (1.98) The ratio of accruing loans past due 90 days of more to total assets (PDTA) -40.005 X 6.576 53.655 -18.278 X (-1.47) (0.25) (2.04) (-0.68) The ratio of other real estate owned to total assets (OREO) 1.700 27.098 -10.454 X -48.539 X (0.12) (0.94) (-0.61) (-1.58) The absolute value of the difference between assets and liabilities maturing or repricing within one year as a proportion of equity value (AGAP) -0.0001 X -0.035 X 0.004 -0.030 X (-0.04) (-1.31) (0.72) (-0.94) The ratio of total book liabilities to the sum of the market value of common stock and the book value of preferred stock (MKTLEV) -0.0005 X 0.023 -0.029 X -0.021 X (-0.03) (0.82) (-2.33) (-0.66) Bank-Specific Risk Measures Interacted with the Stock Market Excess Returns The ratio of non-accruing loans to total assets interacted with the stock market excess return (NATA_M) -0.785 X -0.081 X -0.521 X 1.047 (-1.30) (-0.03) (-0.70) (0.46) The ratio of accruing loans past due 90 days of more to total assets interacted with the stock market excess return (PDTA_M) 6.995 -1.136 X -0.722 X 0.762 (2.20) (-0.26) (-0.21) (0.19) The ratio of other real estate owned to total assets interacted with the stock market excess return (OREO_M) -1.949 X -1.806 X -0.447 X 2.984 (-1.38) (-0.26) (-0.25) (0.44) The absolute value of the difference between assets and liabilities maturing or repricing within one year as a proportion of equity value interated with the stock market excess return (AGAP_M) -0.0001 X 0.001 0.001 -0.002 X (-0.15) (0.15) (0.67) (-0.42) The ratio of total book liabilities to the sum of the market value of common stock and the book value of preferred stock interacted with the stock market excess return (MKTLEV_M) 0.002 -0.002 X 0.002 0.003 (1.16) (-0.60) (1.11) (0.93)

Table 3.3 Continued Dependent Variable / Deposit Insurance Regime Explanatory Variables Subordinated Issuance Senior Issuance Pre-FDICIA Expected Post-FDICIA Expected Pre-FDICIA Expected Post-FDICIA Expected 85:Q1-92:Q4 Sign? 93:Q1-02:Q4 Sign? 85:Q1-92:Q4 Sign? 93:Q1-02:Q4 Sign? Other Banking Organization- Specific Factors The natural log of total assets (ln(ASSETS)) 0.574 X 0.552 X 0.821 X 0.297 X (7.29) (11.54) (8.52) (6.16) An indicator variable that equals one if the banking organization issued debt in the same seniority grade in the preceding 6 month period, and zero otherwise (ISSUE_-1) 0.395 X 0.733 X 0.727 X 1.585 X (3.31) (8.42) (5.95) (17.22) Foreign and domestic income taxes as a percentage of net income (AVGTAX) 0.0001 X -0.0003 -0.0001 -0.0002 (1.06) (-1.27) (-0.38) (-0.59) The ratio of book equity to book total assets (KA) -4.359 -1.838 -7.956 4.459 X (-0.66) (-0.57) (-1.17) (1.44) Business and Bond Market Conditions The unemployment rate (UE) -0.028 X 0.541 0.045 0.271 (-0.11) (2.86) (0.17) (1.34) Stock Market Excess Return (XR) -0.024 X -0.002 X -0.009 X -0.017 X (-1.41) (-0.11) (-0.48) (-0.81) The implied stock volatility measure calculated from option prices traded on the Chicago Board Option Exchange (MKTVOL) -0.051 X -0.004 X 0.002 0.010 (-2.58) (-0.21) (0.13) (0.57) Supervisory Pressure An indicator variable that equals one if the composite supervisory rating equals 2 (BOPEC2) -0.138 X 0.049 -0.025 X 0.209 (-1.25) (0.56) (-0.19) (2.24) An indicator variable that equals one if the composite supervisory rating equals 3, 4 or 5 (BOPEC345) -0.510 X 0.089 0.194 0.424 (-3.05) (0.27) (1.04) (1.42) Wald Tests Wald test statistic for "risk" coefficients jointly equalling zero 55.44 74.70 40.79 67.23 Critical value for the Wald test at the 5 percent confidence level 18.3 18.3 18.3 18.3 Goodness of Fit Measures Fraction of correct predictions for issuance decision 0.851 0.848 0.886 0.880 Number of Observations 1480 1933 1480 1933 R-Squared 0.21 0.31 0.21 0.35 Percent that issued debt of that seniority grade 16.89 21.83 13.11 17.43 Note: All specifications include a constant term which was significant at the 5% level. Year indicator variables, which were equal to one in a specific year of each panel, and zero otherwise were also included though these coefficient estimates are not reported here. Observed spread regressions are heteroskedastic-consistent. t-statistics are in parentheses.

Table 3.4 Parameter Estimates for the Stacked Data Issuance Decision Model for Large U.S. Banking Organizations Dependent Variable / Deposit Insurance Regime Pre-FDICIA Period Post-FDICIA Period Explanatory Variables Direct Effect for Expected Additional Effect for Expected Direct Effect for Expected Additional Effect for Expected Subordinated Debt Sign? Senior Debt Sign? Subordinated Debt Sign? Senior Debt Sign? Accounting- and Marketbased Risk Measures The ratio of non-accruing loans to total assets (NATA) -8.474 X -9.217 X -9.828 X 29.830 (-1.39) (-1.04) (-0.78) (1.71) The ratio of accruing loans past due 90 days of more to total assets (PDTA) -49.698 X 109.530 9.831 -36.950 X (-1.86) (3.03) (0.38) (-1.00) The ratio of other real estate owned to total assets (OREO) 14.103 -40.190 X 33.754 -99.285 X (1.01) (-1.89) (1.22) (-2.44) The absolute value of the difference between assets and liabilities maturing or repricing within one year as a proportion of equity value (AGAP) -0.001 X 0.004 -0.028 X -0.013 X (-0.27) (0.76) (-1.12) (-0.29) The ratio of total book liabilities to the sum of the market value of common stock and the book value of preferred stock (MKTLEV) 0.013 X -0.043 X 0.014 -0.030 X (1.03) (-2.67) (0.57) (-0.96) Accounting- and Market- based Risk Measures Interacted with the Stock Market Excess Returns The ratio of non-accruing loans to total assets interacted with the stock market excess return (NATA_M) -1.207 X 1.003 0.289 0.711 (-1.96) (1.07) (0.11) (0.20) The ratio of accruing loans past due 90 days of more to total assets interacted with the stock market excess return (PDTA_M) 7.899 -8.592 X -1.304 X 2.190 (2.36) (-1.84) (-0.31) (0.37) The ratio of other real estate owned to total assets interacted with the stock market excess return (OREO_M) -2.113 X 1.079 0.486 -1.431 X (-1.49) (0.48) (0.07) (-0.15) The absolute value of the difference between assets and liabilities maturing or repricing within one year as a proportion of equity value interated with the stock market excess return (AGAP_M) -0.00002 X 0.001 0.0004 -0.002 X (-0.02) (0.53) (0.12) (-0.44) The ratio of total book liabilities to the sum of the market value of common stock and the book value of preferred stock interacted with the stock market excess return (MKTLEV_M) 0.003 -0.002 X -0.001 X 0.003 (1.84) (-0.64) (-0.33) (0.72)

Table 3.4 Continued Pre-FDICIA Period Post-FDICIA Period Explanatory Variables Direct Effect for Expected Additional Effect for Expected Direct Effect for Expected Additional Effect for Expected Subordinated Debt Sign? Senior Debt Sign? Subordinated Debt Sign? Senior Debt Sign? Other Banking Organization- Specific Factors The natural log of total assets (ln(ASSETS)) 0.653 X 0.027 X 0.469 X -0.091 (10.50) (0.55) (12.40) (-2.58) An indicator variable that equals one if the banking organization issued SND in the preceding 6 month period, and zero otherwise (ISSUE_-1) 0.423 X 0.384 X 0.772 X 0.782 X (3.68) (2.34) (9.15) (6.33) Foreign and domestic income taxes as a percentage of net income (AVGTAX) 0.0001 X -0.0001 -0.0003 0.0001 X (1.01) (-0.82) (-1.23) (0.27) The ratio of book equity to book total assets (KA) 8.097 X -25.711 -4.227 X 11.216 X (1.44) (-3.55) (-1.42) (2.95) Business and Bond Market Conditions The unemployment rate (UE) -0.361 X 0.010 0.397 0.015 (-2.43) (0.11) (2.78) (0.19) Stock Market Excess Return (XR) -0.024 X 0.024 -0.006 X -0.006 X (-1.43) (0.96) (-0.34) (-0.21) The implied stock volatility measure calculated from option prices traded on the Chicago Board Option Exchange (MKTVOL) -0.047 X 0.062 -0.006 X 0.019 (-3.05) (3.78) (-0.43) (1.37) Supervisory Pressure An indicator variable that equals one if the composite supervisory rating equals 2 (BOPEC2) 0.007 -0.201 X 0.044 0.185 (0.06) (-1.30) (0.53) (1.53) An indicator variable that equals one if the composite supervisory rating equals 3, 4 or 5 (BOPEC345) -0.293 X 0.263 0.071 0.397 (-1.81) (1.16) (0.22) (0.93) Goodness of Fit Measures Fraction of correct predictions for issuance 0.870 0.865 decision Number of Observations 2960 3866 R-Squared 0.202 0.325 Percent that issued debt instrument 16.89 21.83 13.11 17.43 Note: All specifications include a constant term which was significant at the 5% level. Year indicator variables, which were equal to one in a specific year of each panel, and zero otherwise were also included though these coefficient estimates are not reported here. Observed spread regressions are heteroskedastic-consistent. t-statistics are in parentheses.

more likely to issue debt than are less well capitalized organizations. Inthepre-FDICIAperiod,atleastthreeoutoffiveofthebankingorganizationspecificvariableshavetheexpectedsignforeithersubordinatedorseniorissues. AWaldtest(Table3.5)forthejointimpactofthesefivevariablesyieldsa(nonsignificant) negative result for subordinated issues and a significant positive result for senior issues. This implies that banks with negative news tended to issue senior debt during this period. Interestingly, among the individual coefficients,themostprivatelyknown,NATAandPDTA,haveanegativesigninthe column for subordinated issues, while one of the publicly accessible variables, MKTLEV has a negative sign and is statistically significant in the column for senior issuance.39 These results suggest that banks’ preference for senior debt in times of relatively adverse information is stronger when bad news is private than when bad news is public. Moreover, the general preference for subordinated debt in good, and senior debt in bad times is confirmed by the proxies forpublicriskperception: parameter estimates forUE and MKTVOL areboth negative and the estimate for MKTVOL is significantly negative for subordinateddebtinthisperiod. And,thesignificantandnegativeparameterestimate for BOPEC345 during this period further confirms that the riskiest banks did nottakeadvantageofadverseinsiderinformationbyissuingsubordinateddebt. In contrast, the market price of risk (XR) is not significant, nor are the joint impactoftheinteractiontermswithbankingorganization-specificriskvariables (these interacted variables are marked by the suffix _M) significant.40 Taken together, these empirical findings mean that the “informed investor hypothesis” is well confirmed by the data in the pre-FDICIA period. In times of unfavorable information, banks do not, or cannot issue subordinated debt. Or,subordinateddebtmaybeefficientlypricedwhileseniordebtisunderpriced in bad times. In addition, we find scant evidence to support the “risk aversion hypothesis.” Moreover, the “signalling hypothesis,” which would predict more subordinatedissuesintimeswhenbankshavebadprivateinformation,isclearly rejected for this period. In the post-FDICIA period, four out of five of the banking organizationspecific risk variables have a sign that is consistent with the substitution effect between subordinated and senior debt that was detected in the pre-FDICIA period.41 However, in the post-FDICIA period the joint impact of these variables ceases to be significant in a Wald test (Table 3.5) and the direction of the impact even changes signs.42 In the post-FDICIA period, the effects of bond and business market con- 39In the stacked data model, Table 3.4, the senior debt indicator interacted risk variables forPDTA and MARKTLEV arestatistically significantin opposing directions. 40A Wald test forthe joint effect of the interacted terms yields a positive sign for subordinated debtand anegative sign forseniordebt,neitherof which is significant. 41In the stacked data models, only one of the parameter estimates for senior debt banking organization-specificriskvariables(OREO)wassignificantlydifferentfromsubordinateddebt banking organization-specificrisk variables in the post-FDICIA period. 42These findings were also confirmed using a multinomial logit regression model where bankingorganizationscouldchoseto(1)issuenopublicdebt,(2)toissueonlyseniordebt,(3) toissueonlysubordinateddebt,or(4)toissuebothseniorandsubordinateddebtinstruments. The joint marginal effects of the banking organization-specific risk proxies and of said risk proxiesinteractedwiththemarketpriceofriskwereofthesamesignsaswereobtainedusing the separate probit models in each period that are reported in Table 3.3. Therefore, the parameterestimatesforthelogitmodelsupporttheviewthatbankingorganizationssubstitute betweenseniorandsubordinateddebtissuanceactivitiesastheirfinancialconditionevolves. 24

Table 3.5 Hypotheses, Wald Test Statistics, Critical Values and Sign Tests for Joint Effects in the Issuance Decision Models Pre-FDICIA Period: 1985:Q1 - 1992:Q4 Post-FDICIA Period: 1993:Q1 - 2002:Q4 Hypothesis Test Critical Value for Joint Test Critical Value for Joint Statistic a 5 Percent Effect Statistic a 5 Percent Effect Confidence Level Confidence Level H1: The parameter estimates for accountingand market-based risk variables jointly 4.24 11.1 Negative 9.06 11.1 Positive equal zero in the model for subordinated debt. H2: The parameter estimates for accountingand market-based risk variables jointly 30.00 11.1 Positive 10.36 11.1 Negative equal zero in the model for senior debt.

ditions and of private information on debt issuance decisions are either fairly weak or quite unexpected. Although bond market volatility significantly reduced subordinated debt issuance activities in the pre-FDICIA period, the parameter estimate for MKTVOL was insignificant in the post-FDICIA period. Moreover, such activities were (unexpectedly) positively correlated with the unemployment rate (UE) in thepost-FDICIAperiod. Interestingly, the supervisoryindicatorvariablesdidnotinfluencesubordinateddebtissuanceactivities inthepost-FDICIAperiod,butBOPEC2 significantlyinfluencedseniordebtissuanceactivities,thoughthiseffectwasnotsubstantiallydifferentinthestacked issuance model (Table 3.4).43 Taken together, the issuance decision parameter estimates confirm our expectation that senior and subordinated debt would become more similar under recent the regulatory reforms that implemented depositor preference and capital-basedpromptcorrectiveactionsbybanksupervisors. Publiclyperceived risk does not appear to favor senior over subordinated debt any more. Only the most private bad information (represented by the BOPEC variables) still prompts banking organizations to prefer to issue senior debt. 3.5.3 Issuance spreads Table 3.6 presents parameter estimates for the sample selection models for subordinatedandseniordebtspreads,equation(25). Thesemodelswereestimated for the pre-FDICIA (1985:Q1- 1992:Q4) and post-FDICIA (1993:Q1-2002:Q4) periods using data on issuance spreads for all subordindated instruments, for onlysubordinatedinstrumentswithanissuancesizeofatleast$75millionUSD, and for all senior instruments issued by large U.S. banking organizations. As with Tables 3.3 and 3.4, the left column of the table provides a short description for each of the explanatory variables. The left panel of the table presents estimates for the sample selection model for subordinated debt during each of the time periods considered, and the right panel of the table presents estimates for the sample selection model for senior debt during each of the time periods considered. And, Table 3.7 presents parameter estimates for the stacked data sample selection models. The dependent variables for the stacked sample selection model are the stacked observed issuance spreads for subordinated (SPREAD ) and for senior debt (SPREAD ). Parameter estimates for U,i,t E,i,t theexplanatoryvariablesinteractedwithaseniorgradeindicatorvariable,I,in Table 3.7 are significant only when the individual parameter estimates for the original sample selection models (in Table 3.6) are statistically different when senior debt market data, rather than subordinated debt market data, are used. Thus, parameter estimates in the stacked regression can be used to infer a “direct effect” for subordianted debt and an “additional effect” for senior debt. Wald test statistics are presented in Table 3.8 for the joint impact of groups of risk varaibles on issuance spreads. Yieldspreadsinboththepre-andpost-FDICIAperiodswerestronglyinfluencedbyinstrument-specificcharacteristicssuchascalloptions,CALL,maturities(MATLT10,MATGT20),andcouponfrequencies(COUPON2). Although it appearsthat thelargest banks among the50 largest banks paidhighersenior spreadsinthepre-FDICIAperiodandlowerseniorspreadsinthepost-FDICIA 43ThelogitmodelparameterestimatesfortheBOPEC variableswerealsoconsistentwith the probitmodelparameterestimates thatare reported in Table3.3. 25

Table 3.6 Parameter Estimates for the Sample Selection Model of Large U.S. Banking Organizations Dependent Variable / Deposit Insurance Regime Explanatory Variables Subordinated Spreads Senior Spreads Pre-FDICIA Expected Pre-FDICIA Expected Post-FDICIA Expected Post-FDICIA Expected Pre-FDICIA Expected Post-FDICIA Expected 85:Q1-92:Q4 Sign? Issues gt 75 mn Sign? 93:Q1-02:Q4 Sign? Issues gt 75 mn Sign? 85:Q1-92:Q4 Sign? 93:Q1-02:Q4 Sign? Accounting- and Market- based Risk Measures The ratio of non-accruing loans to total assets (NATA) -7.075 -11.596 -4.985 -3.931 -23.634 9.565 X (-1.09) (-1.67) (-0.75) (-0.51) (-2.39) (1.07) The ratio of accruing loans past due 90 days of more to total assets (PDTA) 79.126 X 82.804 X 45.244 X 68.425 X 26.761 X 46.486 X (1.80) (1.81) (2.38) (2.54) (0.65) (2.75) The ratio of other real estate owned to total assets (OREO) 51.014 X 50.535 X 4.542 X 1.659 X 19.900 X -13.937 (2.87) (2.99) (0.43) (0.15) (0.69) (-0.89) The absolute value of the difference between assets and liabilities maturing or repricing within one year as a proportion of equity value (AGAP) -0.005 0.009 X -0.006 0.003 X 0.026 X 0.013 X (-0.31) (0.49) (-0.53) (0.20) (1.45) (0.49) The ratio of total book liabilities to the sum of the market value of common stock and the book value of preferred stock (MKTLEV) 0.007 X 0.004 X 0.044 X 0.045 X -0.043 0.001 X (0.36) (0.17) (3.62) (3.06) (-2.44) (0.05) Bank-Specific Risk Measures Interacted with the Stock Market Excess Returns The ratio of non-accruing loans to total assets interacted with the stock market excess return (NATA_M) 0.258 X -0.103 -3.310 -1.880 -0.603 -0.875 (0.41) (-0.15) (-3.18) (-1.49) (-0.69) (-0.55) The ratio of accruing loans past due 90 days of more to total assets interacted with the stock market excess return (PDTA_M) -2.079 -2.178 5.334 X 3.596 X -4.294 -1.152 (-0.48) (-0.54) (2.12) (0.73) (-0.65) (-0.79) The ratio of other real estate owned to total assets interacted with the stock market excess return (OREO_M) 3.287 X 2.727 X 4.278 X 2.351 X 5.704 X -4.039 (1.78) (1.49) (2.54) (1.03) (3.04) (-1.33) The absolute value of the difference between assets and liabilities maturing or repricing within one year as a proportion of equity value interated with the stock market excess return (AGAP_M) -0.002 0.0005 X 0.005 X 0.003 X 0.001 X -0.007 (-0.70) (0.17) (1.65) (0.85) (0.18) (-1.40) The ratio of total book liabilities to the sum of the market value of common stock and the book value of preferred stock interacted with the stock market excess return (MKTLEV_M) -0.0005 -0.002 0.001 X 0.002 X 0.003 X 0.002 X (-0.25) (-1.10) (0.66) (0.89) (1.37) (1.44)

Table 3.6 Continued Dependent Variable / Deposit Insurance Regime Explanatory Variables Subordinated Spreads Senior Spreads Pre-FDICIA Expected Pre-FDICIA Expected Post-FDICIA Expected Post-FDICIA Expected Pre-FDICIA Expected Post-FDICIA Expected 85:Q1-92:Q4 Sign? Issues gt 75 mn Sign? 93:Q1-02:Q4 Sign? Issues gt 75 mn Sign? 85:Q1-92:Q4 Sign? 93:Q1-02:Q4 Sign? Other Banking Organization- Specific Factors The natural log of total assets (ln(ASSETS)) -0.228 X -0.142 X 0.198 0.058 0.894 -0.084 X (-0.89) (-0.57) (1.19) (0.34) (3.20) (-2.20) An indicator variable that equals one if the banking organization issued debt in the same seniority grade in the preceding 6 month period, and zero otherwise (ISSUE_-1) -0.301 X -0.279 X 0.200 -0.028 X 0.363 -0.443 X (-1.86) (-1.73) (0.82) (-0.12) (0.91) (-1.67) Business and Bond Market Conditions Stock Market Excess Return (XR) 0.010 0.032 -0.018 X -0.009 X -0.038 X 0.0101 (0.45) (1.35) (-1.91) (-0.51) (-1.40) (0.80) The implied stock volatility measure calculated from option prices traded on the Chicago Board Option Exchange (MKTVOL) 0.077 X 0.088 X 0.013 X 0.025 X 0.015 X 0.021 X (4.34) (4.62) (1.67) (1.81) (1.59) (2.65) Supervisory Pressure An indicator variable that equals one if the composite supervisory rating equals 2 (BOPEC2) 0.368 X 0.450 X 0.083 X 0.022 X 0.192 X -0.102 (3.18) (3.73) (2.47) (0.45) (1.13) (-2.15) An indicator variable that equals one if the composite supervisory rating equals 3, 4 or 5 (BOPEC345) 0.7480 X 0.7046 X 0.0089 X -0.025 0.662 X -0.631 (3.69) (3.32) (0.12) (-0.20) (2.19) (-2.78) Instrument Characteristics An indicator that equals one when an issue has a call option (CALL) 0.317 X 0.373 X 0.090 X 0.188 X -0.1921 0.1154 X (1.13) (1.39) (2.85) (2.99) (-0.77) (1.12) An indicator that equals one when an issue has a maturity less than ten years (MATLT10) 0.123 X 0.157 X 0.143 X 0.131 X -0.256 0.077 X (1.39) (1.76) (5.30) (3.49) (-1.24) (1.66) An indicator that equals one when an issue has a maturity greater than twenty years (MATGT20) 0.950 X 0.992 X 0.126 X 0.125 X -- 0.348 X (3.55) (3.84) (3.97) (1.64) (3.13) An indicator that equals one when the coupon frequency is monthly (COUPON12) -- -- -0.118 -0.311 -- -0.052 (-1.70) (-2.01) (-1.04) An indicator that equals one when the coupon frequency is semi-annually (COUPON2) 0.657 X 0.741 X -0.171 -0.303 -0.248 -0.149 (3.26) (3.72) (-2.56) (-2.45) (-0.89) (-2.11) The dollar amount of the issue (ISSUESIZE) -0.001 X -0.001 X 0.0002 0.0003 0.001 0.0001 (-1.75) (-1.74) (2.60) (3.49) (0.76) (0.74) Wald Tests Wald test statistic for "risk" coefficients jointly equalling zero 30.83 32.70 77.45 42.49 44.16 19.02 Critical value for the Wald test at the 5 percent confidence level 18.3 18.3 18.3 18.3 18.3 18.3 Mills Inverse Ratio Mills inverse ratio coefficient -1.000 -0.831 0.647 0.240 1.368 -0.406 (-1.72) (-1.50) (1.35) (0.52) (2.33) (-1.67) Goodness of Fit Measures Number of Observations 158 151 735 232 104 712 R-Squared 0.74 0.75 0.61 0.69 0.81 0.249 Note: All specifications include a constant term which was significant at the 5% level. Year indicator variables, which were equal to one in a specific year of each panel, and zero otherwise were also included though these coefficient estimates are not reported here. Observed spread regressions are heteroskedastic-consistent. t-statistics are in parentheses.

Table 3.7 Parameter Estimates for the Stacked Data Sample Selection Model for Debt Spreads of Large U.S. Banking Organizations Dependent Variable / Deposit Insurance Regime Pre-FDICIA Period Post-FDICIA Period Explanatory Variables Direct Effect for Expected Additional Effect for Expected Direct Effect for Expected Additional Effect for Expected Subordinated Debt Sign? Senior Debt Sign? Subordinated Debt Sign? Senior Debt Sign? Accounting- and Marketbased Risk Measures The ratio of non-accruing loans to total assets (NATA) -9.209 -1.644 -10.300 44.070 X (-1.47) (-0.14) (-1.61) (3.69) The ratio of accruing loans past due 90 days of more to total assets (PDTA) 33.034 X -63.790 34.848 X 17.523 X (0.96) (-1.22) (1.82) (0.67) The ratio of other real estate owned to total assets (OREO) 51.083 X -32.388 10.737 X -48.416 (2.67) (-1.14) (1.14) (-2.80) The absolute value of the difference between assets and liabilities maturing or repricing within one year as a proportion of equity value (AGAP) -0.003 0.024 X -0.011 0.063 X (-0.16) (1.02) (-0.92) (2.08) The ratio of total book liabilities to the sum of the market value of common stock and the book value of preferred stock (MKTLEV) -0.003 -0.017 0.029 X -0.035 (-0.13) (-0.76) (3.25) (-2.16) Accounting- and Market- based Risk Measures Interacted with the Stock Market Excess Returns The ratio of non-accruing loans to total assets interacted with the stock market excess return (NATA_M) 0.152 X -0.051 -3.010 2.542 X (0.21) (-0.05) (-2.98) (1.41) The ratio of accruing loans past due 90 days of more to total assets interacted with the stock market excess return (PDTA_M) -2.374 -9.783 5.209 X -5.136 (-0.55) (-1.21) (2.08) (-1.81) The ratio of other real estate owned to total assets interacted with the stock market excess return (OREO_M) 3.718 X 4.002 X 4.166 X -7.373 (1.98) (1.68) (2.65) (-2.24) The absolute value of the difference between assets and liabilities maturing or repricing within one year as a proportion of equity value interated with the stock market excess return (AGAP_M) -0.001 -0.001 0.005 X -0.009 (-0.40) (-0.32) (1.48) (-1.68) The ratio of total book liabilities to the sum of the market value of common stock and the book value of preferred stock interacted with the stock market excess return (MKTLEV_M) -0.001 0.004 X 0.001 X 0.001 X (-0.71) (1.37) (1.00) (0.66)

Table 3.7 Continued Pre-FDICIA Period Post-FDICIA Period Explanatory Variables Direct Effect for Expected Additional Effect for Expected Direct Effect for Expected Additional Effect for Expected Subordinated Debt Sign? Senior Debt Sign? Subordinated Debt Sign? Senior Debt Sign? Other Banking Organization- Specific Factors The natural log of total assets (ln(ASSETS)) 0.242 0.078 0.051 -0.032 X (1.52) (0.34) (1.24) (-2.44) An indicator variable that equals one if the banking organization issued debt in the same seniority grade in the preceding 6 month period, and zero otherwise (ISSUE_-1) -0.130 X -0.036 X 0.014 0.217 (-1.02) (-0.12) (0.18) (1.31) Business and Bond Market Conditions Stock Market Excess Return (XR) 0.015 X -0.037 -0.018 0.017 X (0.66) (-1.09) (-2.24) (1.31) The implied stock volatility measure calculated from option prices traded on the Chicago Board Option Exchange (MKTVOL) 0.054 X -0.045 0.019 X 0.004 X (2.95) (-2.20) (4.18) (0.68) Supervisory Pressure An indicator variable that equals one if the composite supervisory rating equals 2 (BOPEC2) 0.403 X -0.245 0.035 X -0.033 (3.47) (-1.29) (0.91) (-0.50) An indicator variable that equals one if the composite supervisory rating equals 3, 4 or 5 (BOPEC345) 0.632 X -0.040 -0.028 -0.795 (3.26) (-0.12) (-0.37) (-3.32) Instrument Characteristics An indicator that equals one when an issue has a call option (CALL) 0.483 X -- 0.099 X 0.054 X (2.15) (3.25) (0.55) An indicator that equals one when an issue has a maturity less than ten years (MATLT10) 0.065 X 0.090 X 0.150 X -0.062 (0.69) (0.50) (5.20) (-1.12) An indicator that equals one when an issue has a maturity greater than twenty years (MATGT20) 1.012 X -- 0.084 X 0.202 X (3.43) (2.38) (1.80) An indicator that equals one when the coupon frequency is monthly (COUPON12) -- -- -0.131 0.066 X (-1.77) (0.63) An indicator that equals one when the coupon frequency is semi-annually (COUPON2) 0.588 X -0.968 -0.224 0.128 X (2.70) (-0.22) (-3.18) (1.26) The dollar amount of the issue (ISSUESIZE) -0.001 X 0.002 0.0003 -0.0002 X (-1.92) (1.81) (4.60) (-1.94) Mills inverse ratio coefficient 0.016 0.282 0.255 -0.005 (0.05) (0.66) (2.09) (-0.05) Number of Observations 260 1447 R-Squared 0.734 0.305 Note: All specifications include a constant term which was significant at the 5% level. Year indicator variables, which were equal to one in a specific year of each panel, and zero otherwise were also included though these coefficient estimates are not reported here. Observed spread regressions are heteroskedastic-consistent. t-statistics are in parentheses.

Table 3.8 Hypotheses, Wald Test Statistics, Critical Values and Sign Tests for Joint Effects in the Sample Selection Models Pre-FDICIA Period: 1986:Q2 - 1992:Q4 Post-FDICIA Period: 1993:Q1 - 1999:Q4 Hypothesis Test Critical Value for Joint Test Critical Value for Joint Statistic a 5 Percent Effect Statistic a 5 Percent Effect Confidence Level Confidence Level H1: The parameter estimates for accountingand market-based risk variables jointly equal zero in the model for subordinated debt spreads. 26.05 11.1 Positive 43.85 11.1 Positive H2: The parameter estimates for accountingand market-based risk variables jointly equal zero in the model for subordinated debt spreads for issues greater than $75 million USD. 28.53 11.1 Positive 29.94 11.1 Positive H3: The parameter estimates for accountingand market-based risk variables jointly equal zero in the model for senior debt spreads. 19.37 11.1 Positive 12.94 11.1 Positive

period, we find little influence of the size of individual issues (ISSUESIZE) on senior or subordinated spreads during either period.44 In the pre-FDICIA period, for each type of instrument three out of five banking organization-specific risk variables have the expected sign (Table 3.6). Only the parameter esimate for OREO in the subordinated debt spread model is significant and of the expected sign. However, the joint effect of the five risk variables together is strongly significant and of the expected sign for both types of instruments in a Wald test (Table 3.8). In addition, the parameter estimates for the most private information among the risk variables considered (i.e., the BOPECs) are positive for both instruments and strongly significant for subordinated debt. This means that supervisory ratings are to some degree correlated with knowledge of sophisticated investors. Poor supervisory ratings do significantly increase issuance spreads on subordinated debt. In contrast, the price of market risk (proxied by XR) has no significant impact in this period. The parameter estimate for XR is not significantly different from zero. Moreover, the interaction variables between XR and banking organization-specific risk variables are jointly insignificant for senior debt and for large and small subordinated debt issues.45 Taken together, the parameter estimates for the sample selection models in thepre-FDICIAperiodarelargelyinlinewiththeforecastsofourmodel. Both, subordinated and senior spreads, do react to risk variables. While the senior spread exclusively reflects public information, the subordinate spreads react more to specialized information of the respective investors as well as to public information. This is consistent with the existence of an incentive premium and withthe“informedinvestorhypothesis.” Atthesametime,wefindnoevidence of risk aversion on the part of senior bank debt investors. In the post-FDICIA period, three out of five of the banking organizationspecific risk variables have the expected sign in the column for subordinated debt,andfouroutoffiveofthesevariableshavetheexpectedsigninthecolumn forseniordebt(Table3.6). AWaldtest(Table3.8)revealsthatthejointimpact of the risk proxies remains positive for both debt spreads in the post-FDICIA period, but that the joint significance level has declined for senior spreads in the pre-FDICIA period. The parameter estimate on the private information variable, represented by the BOPEC2, remained positive and significant for subordinated debt during the post-FDICIA period. However, the magnitude of this estimate is only one fourth of the respective number for the pre-FDICIA period and the parameter estimate for BOPEC345 became both statistically and economically insignificant. Consistent with this result, the parameter estimate for XR became negativeand more statistically significant for subordinated spreads in the post-FDICIA period. Together these findings suggest that subordinated debt spreads became less sensitive to the private information held by subordinated investorsandmoresensitivetopublicriskperceptionafterregulatoryreforms.46 44The size of an issue may however have an impact on the sensitivity of its spread with respecttorisk parameters,see below. 45Waldteststatisticsforthejointeffectoftheinteractedriskvariableswiththemarketprice ofrisk were 641,5.14,and 10.18forallsubordinated debtissuancespreads,forsuborindated debt issuance spreads on issues of at least $75 million, and for senior debt issuance spreads, respectively. Each of these test statistics is below the 11.1 critical value for a 5 percent confidence level. 46Thisresultmayexplainwhypractitionerscomplainabout“ballooning”of[subordinated] 26

Theparameterestimatesforinteractiontermsofbankingorganization-specific risk with the market price of risk (XR) for the subordinated debt sample selection model are also interesting. Wald test statistics for the joint effect of these interactiontermssuggestthatthesubordinateddebtspreadsonrelativelylarge issues are sensitive to a simultaneous increase in bank risk and in the market price of risk, just as our model would predict.47 But Wald test statistics for this joint effect were insignificant for small subordinated issues of less than $75 million. Perhaps, this is because smaller issues, which are frequently tranches of medium-term-note programmes, seem to cater to unsophisticated investors, ratherthantoinvestors with special knowledgeon the issuing banks. Whileat first sight the parameter estimates on the interaction terms between the market price of risk and banking organization-specific risks appear to contradict our model, the exclusion of smaller subordinated debt issues demonstrates that these parameter estimates nicely confirm the “informed investor” hypothesis. The parameter estimates for the senior debt sample selection model are more puzzling. Even though the banking organization-specific risk variables arejointlysignificant(Table3.8),thenegativeandsignificantsignsoftheparameter estimates for BOPEC2 and BOPEC345 (Table 3.6) suggests that senior debtofbankingorganizationsinpoorfinancialconditionmaybechronicallyunderpriced. This may partly explain banking organizations’ preference to issue such debt when they have negative private information. It is also noteworthy that there was no sign of risk aversion of senior investors in this period either. 4 Discussion Our estimation of issuance decision models for subordinated and senior debt demonstrate that U.S. banking organizations issue subordinated debt upon receiptofgoodnewsandseniordebtuponreceiptofbadnews. Inthepre-FDICIA period,bothpublicandprivateinformationcontributetothiseffect. Inthepost- FDICIA period, public information has less impact on banking organizations’ choice among debt instruments, while private information remains influential, favoring subordinated (senior) debt when news is good (bad). These results are consistent with the “informed investor hypothesis” that claims that banking organizations would issue debt of different priority status to separate investors with different, yet unobservable, beliefs on the probability of bank failure. In contrast, we found no statistical evidence that there were differences in risk aversion across investor groups. Moreover, the “signalling hypothesis” (Barclay and Smith, 1995) is clearly rejected. The data simply do not support the view that banking organizations issue subordinated debt when they have adverse private information (i.e., when subordinated debt is most overvalued). Thus, U.S. banking organizations do not appear to exploit their informational advantage over investors by opportunistic choice of debt instruments. The lack of evidence to support opportunistic debt issuance does not necesspreads in bad times “reflecting broad skepticism regarding the financial health of banking institutions.” (BoG,1999,p.16) 47The Wald test statistic is 31.66 compared to a 11.1 (95 percent) critical level. And the sign test for the joint effects of the interacted bank-specific risk variables with the market price ofrisk waspositivein the post-FDICIA period. 27

sarilyimplythatbankingorganizationsbehave“idealistically.” Onereasonwhy banking organizations prefer to issue senior debt in bad times may be because senior debt issues are underpriced. Our empirical findings suggest that the senior spread does not react much to bad private, or specialist, information. In fact,thenegativeBOPECcoefficientsinthepost-FDICIAperiodimplythatthe seniorspreadmayevenfallwhenprivatenewsturnsbad. Itseemsthatbanking organizations with poor supervisory ratings have managed to issue senior debt cheaply in the post-FDICIA period.48 In contrast, publicly available information did affect the senior spread in both the pre- and post-FDICIA periods. The subordinated debt spread reacted to both public and specialist, or private, information in the pre-FDICIA period. In the post-FDICIA period, the subordinated spread became less sensitive to specialist, or private, information (proxied by supervisory ratings), but even more sensitive to public information (proxied by stock market excess returns). In the language of our model, this means that the incentive premium has become quantitatively more important. Because regulatory reforms implemented after the last recession have influenced the size of the incentive premium, it is important to recognize that it would be inappropriate to use information on subordinated debt spreads from thepre-FDICIAperiodtoeithersetaregulatoryceilingonbanks’subordinated debt yield (see Calomiris, 1999) or to set triggers for prompt corrective actions (seeEvanoffandWall, 2000and2001; LangandRobertson, 2000). Ontheone hand,theprobabilityoffailurehasbeenreducedbecauseoftheimplementation of prompt corrective actions by bank supervisors. Such reforms would likely have reduced the incentive premium. On the other hand, depositor preference rulessubordinatetheseniordebtinvestorstodepositorsandsuchreformswould boost the incentive premium. Because it is highly unlikely that these two effects would exactly cancel one another out, and because our empirical results suggest that the incentive premiums contained in subordinated debt spreads are different in the pre- and post-FDICIA periods, this implies that ceilings or triggers must be set using data from only the current regulatory regime. The stronger influence of public, rather than specialist, or private, information on subordinated debt spreads in the post-FDICIA period is likely a consequence of the implementation of depositor preference legislation. Depositorpreferencerulesstronglysubordinateseniordebttodepositsandthismeans thatexpectedrecoveryratesforseniordebtissuesarereduced. Accordingtoour model and to our empirical findings, the subordinated spread thus becomes responsivetopublicinformation,eventhoughtheinvestorswhobuysubordinated debt do not themselves believe the public information but hold more favorable, private views. These results explain why market practitioners complain about the “ballooning” of subordinated spreads in times of low general market confidence, or why they voice concerns about the difficulties with “disentangling the separate influences of market factors and of changes in the risk profile of a financial institution” on the subordinated debt spread.49 48One reason may be that senior investors’ beliefs move with the cycle. These investors may perceive strong credit growth at a banking organization as good news when, from a supervisory orprivateperspective,rapid loan growth may already raise somedoubts. 49BoG/DoT (2000),p.78. 28

5 Conclusions Our theoretical model buttressed by our empirical findings lead us to conclude that the yield spread on subordinated debt used in isolation has not been a straightforward measure of bank risk in either of the periods we considered. At a minimum, the subordinated yield spread does not reflect the best available information on a banking organizations’ risk (e.g., the information of sophisticated, informed investors). Rather, the relation of the spread to a banking organization’strueriskisblurredbyitssensitivitytopubliclyperceivedriskas, forexample,togeneralbondmarketvolatilityortostockmarketexcessreturns. Thisbehaviorofthesubordinatedspreadiswellinlinewithourmodel,building on the “informed investor hypothesis,” and with the existence of an incentive premium between the “fair” and the measured subordinated spreads. Paradoxically,thequalityofthesubordinateddebtspreadtomeasurebanking organizations’ risks as they are perceived by most sophisticated investors has deteriorated after the introduction of FDICIA or, more precisely, of depositor preference rules. With depositor preference rules, the risk characteristics of senior debt have become more similar to those of subordinated debt; at the sametime,thesubordinateddebtspreadhasbecome(even)moredependenton factors influencing the senior spread. Thedeteriorationoftheriskmeasurementqualityofthesubordinatedspread after the introduction of depositor preference, however, is likely to understate the longer term virtues of the reform. Once senior debtors realize that their claimsaresubordinatedtodepositors,seniorspreadsmaywellmorefullyreflect specialist information. Therefore, we expect that senior debt will be held by more sophisticated investors in the future. Forthesereasons,thequalityofthesubordinateddebtspreadasameasureof bankriskshouldnotbejudgedonthebasisofthefirstpost-FDICIAdecade. In thisperiod,depositorpreferencemadethesubordinatedspreadreactsimilarlyto theseniorspreadwithouthavingmadeseniorinvestorssufficientlyriskconscious yet. Still, our model and the empirical findings presented here suggest that the issuance spread on subordinated debt is not likely to ever reflect the best risk informationpresentinthemarket,i.e. theriskperceptionofinformedinvestors who buy subordinated debt. As long as subordinated debt coexists with some riskyseniordebtinstrument,whichisheldbylessinformed(ormoreriskaverse) investors, subordinated debt will pay an incentive premium. The incentive premiumdoesnotonlyremuneratesophisticatedinvestorsfortheperceivedrisk theytakebybuyingsubordinateddebt,butalsoforsomerisktheyknowtheyare not taking. Putting it differently, we might say that subordinated investors are not only remunerated for the risks they perceive in their — relatively optimistic — views, but also for the toil and trouble to become informed — and potentially optimistic — in the first place. In an equilibrium in the market for information, theincentivepremiumcontainedinthesubordinatedspreadcould thus beseen as a remuneration to become a sophisticated investor and a potential agent of market discipline. 29

References Allen, F., and D. Gale, 2000, Comparing Financial Systems, MIT Press: Cambridge, Mass., and London, England. Barclay, M. J. and C. W. Smith, Jr., 1995, “The Priority Structure of Corporate Liabilities,” The Journal of Finance, 24, 899-917. BaselCommitteeonBankSupervision(BCBS),2001,“TheNewBaselCapital Accord,” http://www.bis.org/publ/bcbsca03.pdf, January. Birchler, U.W., 2000, “Bankruptcy Priority for Bank Deposits: A Contract Theoretic Explanation,” Review of Financial Studies, 13, 813-839. Board of Governors of the Federal Reserve System, 1999, “Using Subordinated Debt as an Instrument of Market Discipline,” Staff Studies, no. 172, December. Board of Governors of the Federal Reserve System and United States Department of the Treasury, 2000, The Feasability and Desirability of Mandatory Subordinated Debt, Report to the Congress, December. Calomiris,C.W.,1999,“BuildinganIncentive-CompatibleSafetyNet,”Journal of Banking and Finance, 23, 1499-1519. Campbell,R.andR.Huisman,2003,“MeasuringCreditSpreadRisk,”Journal of Portfolio Management, 29, 121-127. Collin-Dufresne, P., R.S. Goldstein, and J.S. Martin, 2001, “The Determinants of Credit Spread Changes,” The Journal of Finance, 56, 2177-2207. Covitz, D. M., D. Hancock, and M. L. Kwast, 2001, “Mandatory Subordinated Debt: Would Banks Face More Market Discipline?” mimeo, Board of Governors of the Federal Reserve System, December. Dewatripont, M. and J. Tirole, 1994, The Prudential Regulation of Banks, The Walras-Pareto lectures, MIT Press: Cambridge, Mass. Diamond,D.W.,1993,“SeniorityandMaturityofDebtContracts,”Journal of Financial Economics, 33, 341-368. Diether,K.B.,C.J.Malloy,andA.Scherbina,2002,“DifferencesofOpinion andtheCrossSectionofStockReturns,”The Journal of Finance,5,2113-2141. Elton, E. J., M.J. Gruber, D.Agrawal and C. Mann, 2000, “Explainingthe Rate Spread on Corporate Bonds,” The Journal of Finance, 56, 247-277. Evanoff,D.D.andL.D.Wall,2000,“SubordinatedDebtasBankCapital: A Proposal for Regulatory Reform,” Federal Reserve Bank of Chicago, Economic Perspectives, Second Quarter, 40-53. Evanoff, D. D. and L.D. Wall, 2001, “Sub-debt Yield Spreads as BankRisk Measures,” Journal of Financial Services Research, 20, 121-145. Fisher, M., D. Nychka, and D. Zervos, 1995, “Fitting the Term Structure of Interest Rates with Smoothing Splines,” Finance and Economics Discussion Series,no. 1995/1,BoardofGovernorsoftheFederalReserveSystem,January. Hancock, D. and M. L. Kwast, 2001, “Using Subordinated Debt to Monitor Bank Holding Companies: Is it Feasible?,” Journal of Financial Services Research, 19, December. Krainer, J. and Lopez, J. A., 2002, “Using Equity Market Information to Monitor Banking Institutions,” Federal Reserve Bank of San Francisco Economic Letter, 2003-01, January 31. Lang, W. W. and D. D. Robertson, 2000, “A Retrospective Analysis of Subordinated Debt as a Trigger for Regulatory Intervention,” mimeo, Office of the Comptroller of the Currency, December. 30

Sironi, A., 2001, “An Analysis of European Banks SND Issues and its Implications fortheDesignof aMandatorySubordinated Debt Policy” Journal of Financial Services Research, 20, October, 233-66. Winton, A., 1995, “Costly State Verification and Multiple Investors: The Role of Seniority,” The Review of Financial Studies, 8, 91-123. 31