Premiums in Private versus Public Bank Branch Sales

Premiums in Privateversus Public Bank Branch Sales --- James A. Berkovec FederalHome Loan Mortgage Corporation John J. Mingo* Board of Governors FederalReserve System XuechunZhang AmericanUniverity Draft: June, 1997 Abstract Thispaper isthe firstto directlyestimatethe determinantsof differencesinpremiums receivedbypublicandprivate sellersinthe market for bank branches(deposit bases). Deposit premiumsreceivedinprivate sector transactions exceededthose receivedbythe FDIC andthe RTC, evenafier controllingfor known characteristics ofthe transactions and afier corrections for possiblesampleselectionbias. The observed differentialdisappearedby 1992,suggestingimprovedmarket efficiencyand/or the impactofFDICIA (1991), which mandated “least-cost” resolution procedures for failedinstitutions. Additionally,the evidencesuggeststhat bankbranches are IndependentValue Objectswhose auctionsalways result in“unintended”transfers ofvalueto the winningbidders. Thisresult, whileconsistent with previous literature that found positiveCumulativeAbnormalReturns to the winners of auctionsfor the branches offailedbanks, neverthelesssuggeststhat not allofthe positive CARScanbe dueto market ineticiency. *corresponding author. Address: MailStop 153,FederalReseNe Board, 20th & ConstitutionAve., Washington,DC 20551; Telephone:202-452-3866; Fax: 202-452-5295; e-mail:mljjmOO@frb.gov. The authors wishto thank AllenBerger and StephenRhoades for commentson an earlierdrafi.

Premiums inPrivateversus Public Bank Branch Sales I. Introduction and Summary During the period 1989through 1992,atthe peak ofthe U.S. bankingcrisis,the FDIC andthe RTC auctioned offthebranchbuildingsand associated depositbases of 625 failed banks and 653 failedthrifis, respectively. Observers inthe popular financialpress (e.g., AmericanBanker, various) routinelyviewedthese government auctionsas“fire-sales,” generating saleprices, or “premiums,”farbelow what would havebeenreceived inprivate sector transactions. Generally,the “low” priceswere thought to bethe result oflegaland politicalpressures on the federalagenciesto accomplishquick salesto acceptablebidders. There isa substantialacademicliteraturethat hasexaminedfailedbank andthrifi auctions (see SectionIII below.) Thisliteraturethough hasnot directlyexaminedwhether the government salesoffailedbanks andthrifis haverealizedreturns lower than would have been achievedbyprivate sellers. Rather, manystudieshaveused event studymethodology to test the related question ofwhether the winningbidders inthe auctions show abnormal stock returns. Overallthe conclusionsofthe studieshavebeen mixedregarding whether the FDIC/RTC transactions do, infact, giveawayvalueto winningbidders. In thispaper, we directlyexaminethe differenceinpricesreceivedbypublicand private sellers,by comparingthe pricesreceived inprivate sector branch salesandthose receivedintheFDIC/RTC transactions, incaseswhere the private andpublicsector transactionswere structured in similarfashion. Our analysisdoes findthat the “deposit premiums”(prices)receivedbythe FDIC andRTC from the auctionsis substantiallylower on averagethanthose receivedbyprivate sellersof branches. Further, thisgap inprices remainsafier controllingfor the known characteristicsofthe transactions, and afier sample selectioncorrections areused to attempt to control for differencesinthe “unobservable” characteristicsbetween the publicand private sales.The results do suggestthat the differencesbetween publicand privatepricesfelloverthe sampleperiod with onlya small and statisticallyinsignificantgap remainingin 1992. The post-1991 result suggeststhat changesinagencyauction procedures, mandatedbytheFederal Deposit Insurance

Corporation Improvement Act of 1991,mayhavecontributed to improved salepricesin public-sectortransactions (see SectionIV below). The obviousnext question afier findingthat FDIC/RTC premiumsare (or were) low iswhy? Onepotential explanationisthat buyersperceive(correctly or not) that the failed institutionsbeingauctionedbythe FDIC/RTC are inferiorto the privatebranchesbeing sold, for reasonsthat are not correlated with the transaction characteristicsinour data sample. Another potentialexplanationisthat the auctionprocedures employedbythe FDIC are inefficientrelativeto the negotiated transactions used byprivatebranch sellers.This inefficiencycouldpossiblybe caused by suboptimalexpenditureson marketing and information-disseminationbythe FDIC/RTC. The low expenditures(relativeto private sector saleexpenses)mayhaveledto highercosts ofinformationcollectionbypotential bidders or higheruncertainty inbidders’ determinationsofvalue,thus leadingto lower clearingprices. Utiortunately, the data do not permittests ofthe influence,ifany,of sales/marketingexpensesor the differentialimpactof sealed-bidauctionsversus negotiated transactions. The remainder ofthe paper isorganized asfollows. SectionII provides abrief review ofthe form ofFDIC/RTC transactions visavisthose ofthe private sector. The sectionalso describesthe database. SectionIII brieflyreviewsthe literature, includingthe relevanceof auction models. This sectionjustifiesthe use of a directpricingmodelintesting for the existenceofunintendedsubsidiesto winners offailedbankbranch auctions. The empiricalresults are presented in SectionIV. SectionV provides someconclusions. II. PublicandPrivate Branch Sales;the Database. Thetwo federal agencieschargedwith actingasreceiversfor failedbanks andthrifis -- theFederal Deposit Insurance Corporation andthe Resolution Trust Corporation -- have severalfailure-resolutionmethods attheir disposal.l Theseincludetotal asset purchases and 1 TheRTCnolongerexists. TheFDICisthereceiverforanyfailedbankorthrift. 2

total liabilityassumptionsbythebuyer (so-called“wholebank” deals), aswellasvariations of more limitedtransactions inwhichthe buyer assumesonlythe depositliabilities(and/or someother, limitednon-depositliabilities),andpurchases somelimitednumberofthe failed institution’sassets (generally,bankbuildingsand other fixedassets, andreadily marked-to-market assets). Whilecomplicatedwhole-bank deals,involvingcomplex government support of asset returns, receivedmuch ofthe research attention,the vast majorityofgovernment dealswere structured in simplefashion,similarto private sector branch saletransactions (describedbelow). Banks andthrifis ofienbuy and selleach others’ branchesinawelldocumented market. For example,duringthe periodunder observation (1989-1992), there were 197 privatebankbranch sales,and 320 privatethrifibranch sales. In good times, the branch sale market isat leastas robust as duringcrisisperiods -- from 1993through Q1,1996 there were 619total bank andthrifi branch sales. Institutions selltheirbranches, and associated deposit bases, for severalreasons, includingto raisetheir capitalratios andto improvetheir operating efficiency. For example,Bank A mayfindaparticularbranch to be a loss operation, dueto the locationandmixofbusiness. However, BankB maybewillingto pay a positivepurchase pricefor the samebranch,because oftheway itfitsinto Bank B’s branch system. Somelargebanksroutinelyconsiderselling(buying)groups ofbranches, whenever the calculatedrates ofreturn on the associated depositbases, on a sustainedbasis, arebelow (above) internalhurdlerates.’ Inthetypicalprivatebranch sale/purchase,the buyer assumesallor most ofthe depositliabilities(the customer accounts) associated with the branch. Assets to balancethe assumedliabilitiesconsist offixedbranch assets (buildings,etc.), readilymarketableassets suchasmortgage loans, and cash -- asinthe majorityof government-conducted branch sales. Generally,the buyerreceivesfromthe sellerfewer assets (at marketvalue)than the deposit liabilitiesassumed,the differencebeingtermed the “depositpremium”andrepresenting the 2 SeeBerger,Leusner,andMingo(1997)foracompletediscussionofbranchefficiencyandpricing.

seller’sprofit. The differencesbetween publicsector andprivate sector deposit premiums,if any,are the focus ofthe present study. In previous studies,the existenceofinefficienciesinpublicauctionsoffailedbanks or thrifis generallyhasbeen inferredthrough the results of event studies. In these studies, “whole-bank”transactions generallyhavebeenmixedtogether withtransactions that maybe --- more narrowly characterized as “branchsales.” The existence(absence) of cumulative abnormalequityreturns to the winningbidder hasbeen interpreted as meaningthat the winnerdoes (does not) obtainanunintended subsidybyacquiringthe depositbase ofthe failedbank or thrifi. Also, some studieshaveexaminedthe efficiencydifferencesacross types ofFDIC transactions, but not between FDIC (RTC) branch salesandthose ofthe private sector. Finally,afew studieshavelooked atthe influenceon FDIC deposit premiums ofthe numberofbidders;but these studiesdidnot compareFDIC premiumswith private sector premiumsor, with two exceptions,otherwise measure cumulativeabnormalequity returns to winningbidders (see SectionIII following). We directlycompare publicsector deposit premiumswith those achievedinprivate salesbyusing a commerciallyavailabledatabasethat tracks thesetransactions for participantsinthe mergers/acquisitionsarena.3 The databaseprovidesan estimateofthe depositpremiumineachprivate andpublicsale,alongwith other variablesdescribingthe sizeand structure ofthe transaction. III. Previous Literature andthe PricingModel. Event studiesbased on FDIC failedbank auctionshavefound mixedresultswith regard to whether winningbiddersreceive aneffectivesubsidy. James andWeir (1987), and Bertin, Ghazanfari,and Torabzadeh (1989) found positivecumulativeabnormalreturns (CARS)inthe equityprices ofwinningbiddersinFDIC failedbank auctions. Pettway and Trifis (1985) found negativeCARSafier aninitialpositiveCARS(with the negativeCARS 3 SNLSecurities,Inc.,BankM&ADataSource@.

outweighingthe initialpositiveCARS). Davies (1991) found no relationbetween winning bidsand stock pricemovementsofwinningbidders. Billett,Coburn, and O’Keefe (1995) found positiveCARS,whichneverthelessdecreased afier 1991.4 Studiesof failedthrifi acquisitionsgenerallyfound positiveCARS(see Cole, Eisenbeis,andMcKenzie(1994), Balbirer,Jud, andLindahl(1992), and Gupta, LeCompte, andMisra (1993), althoughthe results of Gosnell,Hudgins, andMacDonald were mixed.5 Note that theFSLIC studiesdealtprimarilywith complexassistancetransactions inthe period before the RTC failedthrifibranch salesbeganto be structured like,and couldbe compared directlyto, those ofthe private sector. Each ofthe event studies suffersfrom the problemassociated generallywith event studiesinthis context -- conclusionsdrawn from stock pricemovementsrest on the assumptionofinformation symmetry(i.e., stock market participantsunderstand the nature of an assistedFDIC saletransaction). However, whileprivate sector transactions are analyzed intensivelyinthe financialpress, the detailsofthe governmenttransactions are ofien never widelyknown, especiallyinthe case ofthe larger, “wholebank” dealsinvolvingcomplex asset-valuesuppofis and associated tax benefits. The effects ofthe transactions maywork theirway into equityprices onlyafier longlagsbeyondthe horizon of most event studies. In addition,the event studiesgenerallyinvolvedlow numberof observations (TheDavies study andtheBilletet aliastudyhad the most observations, 39 and 69, respectively). Withinthese previous studies,severalreasons are givenwhythe publictransactions mightentailunintended subsidiesto the winners, hencelower prices,than would be observed inprivate sales.GFirst the knowledgethat the FDIC/RTC generally“must sell”the failed 4 TheFederalDepositInsuranceCorporationImprovementActof1991effectivelymandatedchangesto theFDIC’Sfailureresolutionpolicies,whichhadtheeffectofincreasingthenumberofFDICtransactions thatwerestructuredasprivatesectorbranchsales. 5ThefailedthriftstudiesgenerallydealtwithFSLICdeals(theFederalSavingsandLoanInsurance Corporationwasthereceiveroffailedthrifispriorto 1989whentheRTCtookover). 6Notdiscussedherearestudiesthatlookattherelativeefficienciesofdifferingmodesoffailureresolution. SeeBovenziandMurton(1988)andBovenziandMuldoon(1990)fordetaileddiscussionsof resolutionmodes. Presumably,inefficientmodes,arbitrarilychosenbytheFDICorRTC,couldimpart unintendedsubsidiestowinningbidders. However,BerkovecandLiang(1993)provideempiricalevidence 5

institutions,acceptingthe highestnonnegativedepositpremium,couldinfluencethebidding strategies ofpotentialbuyers.7 In contrast, private sellerscan choose not to sellifnobidis sufficientlyhigh. Second,government rulesmayreduce the competitivenessofthe auction byrequiring that the winningbiddermustbe “acceptable”to the governmentfrom aprudential standpoint.gThismayactto eliminatepotentialbidderswith low capitalratios or other undesirablecharacteristics, drivingdown clearingprices. Thisargumentismade notwithstandingthe existenceof“forbearance”inwhichlow capitalbanks are nevertheless allowedto bidinfailedbank auctions. Further, the usualregulatory scrutinyofprivate sales mayalso act similarlyinlimitingthe pool ofpotentialbuyers. In addition,potentialwaiver of antitrust concerns for failedbank salescould evenincreasethe numberof potentialbuyers abovethose who could participateinprivate sales. Studiesfocused onlyon gainsfromwinningbidsmayalso be subjectto a selection biasby ignoringthe costs of submittinglosingbids. Manywinningbiddersbid on more than onetransaction. Therefore, winningbidsmightbe structured to cover the costs offailed bids. Event studies,focusingonlyon the effectson stock prices ofwinningbids, should therefore seeeitherpositiveeffects(to offsetthe unseen negativeeffectsoflostbids) or no effects (ifthe stock market totally discountsthe effectsof allbids,winningor losing). Auctiontheory also suggeststhat the nature ofthe objectbeingauctioned could affect pricesand CARS. Common-value(CV) auctions,where the valueofthe auctioned objectis thatthemajorityofFDICcostdifferencesacrosstransactiontypescanbeattributedtotheselectionof banksintoresolutionmethodsbasedonobservedorunobservedqualitydifferences,leavinglittletobe explainedbyrelativeeficiency ofresolutionmodes. 7 Alternatively,liquidationofthefailedbank’sdepositseffectivelyentailsanegative“premium.” Thatis, thenoninterestcoststotheFDICofmailingcheckstoeachdepositorarethoughttobehighenoughsothat liquidationisconsideredonlywhentherearenopositivebidsforthedepositbase. Technically,theFDICdoesnot“havetosell.” Legally,itcancreate“bridgebanks”intowhichto placethefailedbank’sdeposits. However,longte~museofbridgebanksisnotconsideredavalidoption politically. 8InthepastFDICruleslimitedbiddingtofirmsinthesamegeographicmarketasthefailedbank,butthese ruleswereremovedlongbeforetheperiodunderconsiderationhere. 6

assumedto bethe sameto allbiddersbut isunknown atthe timeofthe auction, are subjectto the“winner’scurse”. Failureofbiddersto discountbids sufficientlyto account for winner’s cursewillcausepricesto exceedvalues,therefore resultinginnegativeCARS. Alternatively, inindependentprivatevalue(IPV) auctions, eachbidderknows preciselythevalueofthe objectto him-- but thisvaluediffersacross bidders andthe other bidders’values are unknownto eachbidder. Optimally,the IPV auctionbidder attempts to capture someofthe surplusbetween hisown valuationofthe objectandthat ofthe nexthighestbidder, by biddinglessthan the object’strue valueto him.Thisunderbidding,limitedbythe uncertainty aboutthe other bids, resultsin again(positiveCAR) for the winningbidder. Gilberto and Varaiya(1989), ina studyusinginformationonthe numberofbidders,9 concludethat either failedbankbranches represent independentprivatevalueobjects,not commonvalue objects, ~ bidders are systematicallyignoringwinner’scurse.10 Neither CV or IPV objectsare inconsistentwith government auctionsthat realize prices equalto those achievablebyprivate sellers. However, ifdepositbases~ IPV objects, event studiesshowingpositiveCARSdo not necessarilyimplythat FDIC/RTC auctionswere conducted inefficiently. Further for IPV items,winningbids shouldbe determinedbythe characteristicsofthe winningbidder aswellasthe characteristics ofthe branch depositbasebeing sold(includingthe characteristicsofthe markets inwhichthe bidders andthe branchesreside). Bidders’valuationswillreflecthow the acquiredbranches fitinto their own branch networks, andhow the acquisitionsmightleadto scale,scope, or x-eficiencies, etc. Unintended subsidiesstemmingfrom auctioninefficiencies,ifthey exist, could onlybe detected by direct comparison ofFDIC/RTC premiumswith those received in similarprivate sector transactions. It by directlyviewingthe differencesinprivate sector and publicsector deposit premiums,we findthe publicsector premiumsto be lower (higher)ceterzsparzbust,he next 9 SeealsoHirschhorn(1985)andBillett,Cobum,andO’Keefe(1995). ‘0Formoredetailonauctionmodels,seeRileyandSamuelson,(1981). 7

question is“why?” One possibleexplanationfor low pricesinFDIC/RTC salesisunderinvestmentbygovernment sellersinmarketingandinformationprior to the auction. Experiencewith private and publicsector transactions suggeststhat private sellersspend significantlymore than doesthe FDIC/RTC inboth marketingthe transaction andin providingdue diligenceinformation.ll Suchinformationincludespre-paid independent appraisalsofbuildingvalues. In the absenceof suchappraisals,eachbiddermust separately payfor buildingappraisals(whichreduces, dollarfor dollar,the effectivebidpremium) or, facinguncertaintyover the carryingvalueoffixedassets,bid lower to accommodate the uncertainty.Lower publicsector branch depositpremiumscould reflect suboptimal marketingandinformationexpendituresonthe part ofthe FDIC/RTC. IV. EmpiricalResults. The dataused inthisanalysiscoversboth private andpublicsalesofbanks andthrifis form 1989to 1992. It iscrucialto the private sector/publicsector comparisonto use FDIC/RTC transactions that were structured in similarfashionto privatebranch sales. Thus, “wholebank”assistedtransactions involvingassetvalue supports or putback optionswere notused for comparison purposes. Only“purchase& assumption”transactions (without putback options) and/or “deposittransfers” were used. Also,FDIC/RTC transactions with total deposits above $200 millionwere excluded,sincelargetransactionswere more likelyto be structured with significantasset-value supports (butwhere the existenceofthe supports would not necessarilybe detected withinthe transaction database).12Even ifno asset-value 11Oneoftheauthorsservedasafinancialadvisoronseveralmajorprivateandpublicbranch saletransactions.Aninformalsurveyofotheradvisorsprovidedsupportfortheviewthatprivate sellersspendmoreonpre-saleanalysis,commoninformationgathering,andmarketing. 12Transactionsalsowereexcludediftherewerenovalidpremiums,ifnobuyer(orsellerin thecaseofprivatesales)couldbeproperlyidentifiedi,fnogeographicmarketcouldbe identifiedo,rifkeyfinancialdata,suchasthatnecessaryforthesampleselectionbiascorrection, wereunavailable.

supports were inplace,the transfer to the buyer of alargeportion ofthe failedinstitution’s financialassets couldresult inimplicitdiscounts(premiums)to the extentthe transferring valueofthe financialassetswas substantiallybelow (above)theirtrue market value. To minimizethispossibilitywe also excludedpublictransactions inwhich75 percent or more of the failedinstitution’sassets were transferred to the buyerofthe depositbase. Theresulting observations,for which allexplanatoryvariablescouldbe calculated,numbered 134FDIC deals,271 RTC deals, 51privatebankbranch deals,and 64 privatethrifibranch dealsover the four year period.13 Table 1provides abasicdescriptionofthe data, comparingthe differenttypes oftransactions. As shown inthe table, private salesare larger on averagethan the publicsalesoffailedinstitutions. Bank andthrifi sales,though, are ofvery similarsizes for each ofthe two transaction types. Deposit premiums(the differencebetween the deposits assumedbythe buyer andthe valueofthe balancingassets, dividedbythe acquired deposits) averageabout 1.7percent and are noticeablyhigherfor private salesthan for publicsales. For banks,the average deposit premiuminprivate salesisover 2.5 timesthat for publicsales,with the differencesomewhat higherfor thrifi deposit sales. The empiricalanalysiswilldeterminehow much ofthis differenceindepositpremiumsbetween the publicandprivate salescanbe explainedbythe characteristicsofthe depositbasebeing sold,the sellingfirm,andthe buyingfirm. The residualunexplaineddifferenceiseitherdueto unobserved differencesinthese characteristicsor dueto inherentdifferencesinthe salespractices ofthe publicsellers. Table 1. Data Description TransactionT.v~e Public Private All InstitutionType Bank Thrifi Bank Thrift All NumberofObservations 134 271 51 64 520 DepositPremium(Mean) 1.21 1.35 3.24 4.40 1.68 DepositSize($Millions,mean) 38.4 47.4 119.4 114.6 60.4 NumberofBranchesSold(mean) 1.7 2.1 4.9 4.4 2.5 13Insomecases,transactionsinvolved2ormorebuyers. Ineachcasewhereanidentifiablebuyeris recorded,thisistreatedasaseparateobservation. 9

A singleregression modelfor banks andthrifis combinedwas used to modeldeposit premiumsandto estimatethe average premiumdifferentialassociatedwith agovernment assistedtransaction. Means and standard deviationsfor the 25 independentvariablesinthe regression modelare shownintable 2. More detailedvariabledefinitionsare shown inthe Appendix. Coefficientestimatesfromthe regression modelare shownintable 3. Table2. VariableMeans Variable Description SampleMean SD BANK 1ifbankbranchesbeingsold .36 .48 RATEDIFF (tbill-cdrate) [3month] .52 .29 STOCK Nasdaqbankindex 354.15 61.14 TGLAM Mill’sratio*Govt. -.99 .75 BGLAM Mill’sratio*Govt*BANK -.43 .80 JLAM Mill’sratio -.12 .16 MARKDEP Marketdeposits($billion) 20.09 37.7 POPOFF populatiofianking office(100s) 11.72 16.9 MIDWEST midwestdummy .26 .44 BANK*MW BANK*midwestdummy .052 .22 SOUTHEAST southeastdummy .22 .42 SAMEMSA 1ifbuyer/sellerinsameMSA .37 .48 BANKSAME BANK*S~MSA .10 .31 BUYEQUITY buyer’sequity/assets .075 .03 BANKBUY bankbuyingthrifibranches .52 .50 CORE coredeposits/totaldeposits .81 .16 GROWDEP depositgrowthrate -.024 .15 D1990 1990sale .28 .45 D1991 1991sale .34 .47 D1992 1992sale .24 .43 Govt89 Governmentsalein 1989 .13 .34 Govt90 Governmentsalein 1990 .22 .41 Govt91 Governmentsalein 1991 .23 .43 Govt92 Governmentsalein 1992 .20 .40 --. .- - NumberoiObservations= 520 Theprimaryvariablesofinterest inthe regression are the dummy variables(Govt89, Govt90, Govt91, Govt92) indicatingthat the observationwas agovernment (public)sale duringeachyear. The results indicatethat in allyearsthe deposit premiumsreceived in publicsaleswere, on average, lower than in ‘comparable’private sales. Thisstatement, of course, assumesthat the control variablesinthe modeldo an adequatejob of capturingthe relevantdifferencesbetween the characteristics ofpublicandprivatetransactions. The 10

controlvariablesthat are includedinthe modelcanbegrouped into 4 basiccategories. First there arevariablesthat describethe deposit packagebeing soldandthe attractiveness ofthe market inwhichthe deposits are located. Second, sincebuyercharacteristicsmayindicate value,there are variablesthat describethe buyer. Third,there arevariablesthat are associatedwith the timingofthe transaction, andfinallythere are variablesdesignedto measure and control for possibleselectionbias. Aprimarydeterminantofthe valueofthe depositpackagebeing soldisthe shareof core deposits, measured by CORE. Core deposits, definedas depositsinaccountsunder $100,000, are more valuablethan other depositsbecausethey are more likelyto remainwith the branch for a longerperiod oftime andbecausethey areless sensitiveto interest rates. As expected, apositivecoe~lcient for CORE does indicatea higherpremium(price) ispaid as the percentage of core depositsrises. The other major characteristicofthe deposits inthe modelisthe growth rate ofthe deposits, GROWDEP, measured over the lastyear before the transaction. Higher growth indeposits over the recent past shouldindicateabetter franchise that willresult in continuedgrowth indepositsinthe fiture. Thepositivecoefficienton GROWDEP supports this. A dummyvariable,BANK, isalsoincludedto indicatethat the depositsbeing soldare from abank (rather than athrifi). As shown,bank deposits command higherpremiums,cete]’isparibus,than those ofthrifts.14 Severalvariablesare entered to describethe attractiveness ofthe market area where the depositsbeing soldare located. MARKDEP measuresthe sizeofthe total depositsinthe market area; depositstend to be worth lessinareaswith larger quantitiesof deposits (perhaps because customers are more price consciousinlargemarkets). POPOFF isdefinedasthe populationper bankingofice inthe localmarket andrepresents ameasure ofthe degree of competitioninthe area. Highervaluesfor POPOFF reflectlesscompetition and/or more opportunityfor depositgrowth for the branchesthat are includedinthe sale. Regional dummyvariablesare alsoincludedinthe modelto control for differencesinaverageprices receivedindifferentparts ofthe country. Estimates indicatethat premiumswere higherin 14Thesizeofthetransaction(depositsornumberofbranchessold)doesnotstatisticallytiect premiums. 11

the SOUTHEAST and forthrifis inthe MIDWEST. Banks inthe midwest,though, are estimatedto commandlower premiumsrelativeto other banks, asthe negativeBANK*MW coefficientoutweighsthe MIDWEST effect. Characteristicsofthe buyerincludedinthe modelareBUYEQUITY, BANKBUY, SAMEMSA, andB~SAME. BUYEQUITY measuresthe equityratio ofthe buyingfirm; buyerswith more equitypayhigherpremiumson average. Thisresult probablyreflectsthe factthat deposit acquisitionsare accounted for ascash acquisitionsrather than poolingof interests. Therefore, regulatory capitalratios (especiallytangibleequityratios) necessarily declineas a result ofa depositpurchase. Institutionswith higherequityratios thus can “afford”branch purchases (from a regulatory viewpoint)more than other banks or thrifis. BANKBUY isa dummyvariableindicatingwhen there isabank or bank holdingcompany buyer and athriftbranchesbeing sold. Bank holdingcompanyor bankwinners ofthe thrifi branches paid 54bp more for thebranchesthan didthrifiwinners, inboth private and public transactions. Possibly,the banksviewedthe thrifi acquisitionsasprovidingdiversification, or the operating efficiencyofthe banksgenerallyexceededthose ofthe thrifis, sothat bank winnerscouldbe expectedto reap greater x-efficiencies(bytransformingthrifibranchesinto better petiormers at anyoutput level)than couldthe thrifiwinners. It isalso possiblethe bankwinners intendedto “cross-sell”the thrifi customers an array of “bank-like”products. Thevariables SAMEMSA andB~SAME (for bank sellers)indicatewhether the buyer and sellerare located inthe samemarket. The significantnegativecoefficienton BANKSAME showsthat winnerspaid less(more) for bankbranchesifthe brancheswere in the same(different)market asthe winner. Thiseffectisnot evidentfor thrifi sellersas SAMEMSA has a small(not significant)positivesign. Overall,the fact that buyer characteristicsare significantdeterminantsofpremiumspaid inbranch salesis supportive of a conclusionthat bankbranchesare IPV type objectsthat vary invaluebetween potential bidders. Deposit premiaare expectedto vary over timewith the valueof additionaldepositsto financialinstitutions. Consequentlythe modelincludestwo timevaryingvariables, RATEDIFF and STOCK, alongwith dummyvariablesfor the differentyears inthe 12

estimationsample. RATEDIFF measuresthe differencebetween the 3 montht-billrate and the average 3 month CD rate; the negativesignindicatesthat premiumsincreaseas CD rates increaserelativeto t-billrates. Thisisconsistentwith higherCDrates indicatingperiods of strong demandfor additionaldeposits. STOCK represents the averagevalueofthe Nasdaq bank stock index;highervaluesofthe indexare associatedwith lower average deposit premiums.During periods ofhighbank stock value,bankscan more “cheaply”(interms of the impacton projected earningsper share)implementacquisitionsbybuyingentirebanks in a swap of shares(pooling ofinterests), thereby loweringrelativedemandfor cash acquisitionssuchas depositpurchases. The estimated coefficientsonthe time dummyvariables(D1990, D1991, D1992) indicatethat average depositpremiumsfallsharplyduringthe sampleperiod, by 3.71 percentage points from 1989to 1992,for reasonsthat are not explainedbythe independent variablesinthe model. Onepossibleexplanationisthat, overthe period observed, banks and thrifiswere under pressure from regulators andthe markets to increasetheir capitalratios and thus generaldemandfor deposit acquisitions(whichalwaysresult inalowered capitalratio) fellduringthe period. Thefinalset ofvariablesincludedinthe modelarethe selectionbiascorrection variables. Threevariablesare included:TGLAM, BGLAM, andJLAM. Allthree are variants of standard Mill’sRatio sampleselectioncorrections (Heckman, 1979)that estimate the averagevalueofthe unobservablecomponent ofthe premiumthat couldbe causedbythe “selection”of an observation intothe estimation sample.15In our sample,there isconcern about severaldifferentwaysthat sampleselectioncouldbiasresults. First, allofthe publicsalesinvolvefailedinstitutions,onesthat were taken overby the governmentbecause ofinsolvency. An obviousconcern isthat the deposits of failed institutionsare systematicallylessvaluablethan the deposits ofinstitutionsthat do not fail becausethe event offailuremightbe partly determinedbythe valueofthe depositbase of the depository. Although conceptuallythis selectiondoes occur, the effectisalleviated 15TheformulafortheMill’sratioandtheunderlyingprobitmodelsareshownintheAppendix. 13

becausethe failureof a depository isalwayscausedbythe poor performance ofthe assets (e.g. defaultedloans) rather than poor qualityliabilities,andthe degreeto whichthe deposit valuesoffailedinstitutionsisbiasedisthus uncertain. The modeluses two variablesto measure and correct for thispotentialbias. Onevariable,TGLAM, represents the sample selectioneffectfor failedthrifis, andthe other, BGLAM, represents the sampleselection effectfor failedbanks.lG The signsofthe thrifi andbank selectioneffects differandneither ofthe coefficientson these variablesisstatisticallysignificant. A secondtype of selectioneffectfor failedbankscouldbe causedbythe process used to resolvefailedbanks andthrifis. As discussedearlier,the FDIC andRTC used several methodsfor resolvingfailedbanks,but onlythe “pure” salesof depositbases are includedin our analysis. Thus, our data sampleexcludesobservationswhere banks andthrifis were sold via“wholebank”transactions or “purchase and assumption”transactions involving substantialamounts oftransferred loans. Rather, inour sample,the mainasset typesused to balancethe transferred depositswere branchbuildings,mortgages and other assets transferred at current market values, and cash. As showninBerkovec andLiang,FDIC procedures for selectingfailuremethods during 1987to 1990didresult in a selectionof lower-valued institutionsthat soldinthe deposit-onlysales. However, itisunclearwhether these effectscanbe expected to result inanyempiricallyimportant effects on observed deposit premiumsinthis analysis,sinceasset quality,not deposit quality,was stillthe predominant selectioncriteriaused bytheFDIC andtheRTC. Further, afier 1990,fewer of the failedbank resolutionstended to includeloan saleswith deposits. In anyevent, JLAM is as attempt to measurethe selectionbiascausedbythe criteriaused to selectour estimation samplefrom theuniverseof 1989to 1992failedbanktransactions.17 JLAM does not have anysignificantimpacton depositpremiums. 16TGLAMistheMill’sratiocalculatedfromanestimatedthriftfailuremodelandisenteredonlyforfailed thrifis. BGLAMiscalculatedfi-omabankfailuremodelandisenteredonlyforfailedbanks.Variousformsof selectioneffectsweretried,includingtheuseofestimatedfailureprobabilitiesfornon-failedsellers.Nonewere foundtobesignificantorhaveimpo~lanteffectsontheGOVTcoefficients. 17JLAMistheMill’sRatiocalculatedfromaprobitmodelwherefailedbank(thrifi)transactionsareeitherinor outoftheestimationsample. 14

Selectioneffectscould also occur for the private salesinour sample. Mer all, private salesshouldoccur onlywhen thebuyervaluesabranch more highlythan the seller. A positiveselectionbiaswould result ifbranch salesoccurred when buyersidenti~ branches (of another institution)that are especiallyvaluableto them and areforced to paypremium pricesto inducesales.Alternately,private salescould disproportionatelyconsistof distress saleswhere sellersare sellingassets to quicklyraise capital,a situationthat couldleadto a negativeselectioneffect on observed premiums. In the course ofthis analysis,various efforts were madeto identifi availablecharacteristicsthat are associatedwith privatebranch saleswith the intentof creating selectioncorrection terms for the private sales;these attempts were not successful. Table3.Regression Results Dependent Variable= deposit premium* 100 Variable Description ParameterEstimate TStatistic Constant 5.52 2.66 BANK 1ifbankbranchesbeingsold 1.44 3.70 RATEDIFF (tbill-cdrate) [3month] -1.34 -2.00 STOCK Nasdaqbankindex -.0056 -1.77 TGLAM Mill’sratio*Govt -.190 0.86 BGLAM Mill’sratio*Govt*BANK .267 1.23 JLAM Mill’sratio -.031 -0.05 MARKDEP Marketdeposits($billion) -.014 -2.37 POPOFF populatiotianking office(100s) .031 2.30 MIDWEST midwestdummy 1.11 4.65 BANK*MW bank*midwestdummy -1.72 -4.13 SOUTHEAST southeastdummy .56 2.85 SAMEMSA 1ifbuyer/sellerinsameMSA .17 0.81 BANKSAME BANK*S~MSA -1.00 -3.04 BUYEQUITY buyer’sequity/assets 5.43 2.13 BANKBUY bankbuyingthriftbranches .54 2.22 CORE coredeposits/totaldeposits 1.73 3.06 GROWDEP depositgrowthrate .96 1.97 D1990 1990sale -2.55 -3.02 D1991 1991sale -3.42 -3.85 D1992 1992sale -3.71 -4.19 Govt89 Governmentsalein 1989 -3.18 -5.26 Govt90 Governmentsalein 1990 -2.38 -5.59 Govt91 Governmentsalein1991 -1.43 -3.68 Govt92 Governmentsalein 1992 -.78 -1.59 NumberofObservations= 520 R2= .30 15

The GOVT variablesestimatethe remainingeffecton depositpremiaof government salesversus private salesbyyear, afier controllingfor the other observed factors that influencedepositprices. These estimatesindicatethat there isa substantialnegativeaverage effectofgovernment saleson deposit premiumsin 1989-1991. The estimated sizeofthe effectdeclinesover time substantially;the estimated premiumdifferentialin 1991isless than halfthat of 1989. Another 50percent declineinmagnitudeisobserved for the 1992 estimate,whichisalsonegativebut not statisticallysignificantat conventionallevels. Overall,these results couldindicatethat the market discountappliedto deposits offailed banksthat are soldbythe governmentwere substantialbut havedeclinedasthe market gainedmore experiencewith failedinstitutionsandthe sellingprocedures employedbythe FDIC andRTC. Also, the disappearanceindifferentialby 1992suggeststhat the 1991 FederalDeposit Insurance Corporation Improvement Act of 1991,whichmandated “leastcost” solutionsto failedbanks andthri~s, mayhavecontributedto improved saleseficiency bythe government agencies. Aninterestingcomparisonisto examinehow muchofthe observed average differentialinpremiumsremainsafier the modelcontrolsfor the effectofthe independent variables. Thiscomparisonis shownintable4. Obviously,very littleofthe observed premiumdifferentialisexplainedbythe model. In fact, for the firsttwo years, the estimated government effectislarger thanthe observed meanpremiumdifferential. Themodel does seemto explaina smallportion ofthe observed differentialin 1991and 192,but the vast majorityofthe public-privatepremiumdifferentialcannotbe explainedbythe independent variablesthat describecharacteristicsofthe depositpackage, characteristics ofthe buyer, time, or possibleselectionbias. Table4. Percentage points of observed depositpremiumdifferentialexplainedbymodel. I 1989 1990 1991 1992 I AveragePrivatePrem. 4.09 3.09 2.97 2.51 (Yo) AveragePublicPrem. 1.34 1.13 1.16 1.65 (Yo) ObservedPrivate-Public 2.75 1.95 1.81 0.86 EstimatedResidualEffect 3.19 2.38 1.42 0.78 16

V. Conclusions Using a commerciallyavailabledatabase,this studyisthe firstto directlycompare depositpremiums(purchase prices) infailedbank or failedthrifi deposit saleswith premiums received inprivate deposittransactions. Three empiricalresults are noteworthy: During the firstthree ofthe four years under observation (1989-1992), FDIC and l RTC salesof depositbasesbrought significantlylower premiumsthan similarly structured private sector transactions, other thingsequal(includingcorrections for possiblesampleselectionbias). The public-privatepremiumdifferentialdeclinedover the four-year period until, l in 1992,itwas eliminated(inthe senseof statisticalsignificance). Independentvariablespeculiarto the winningbidderwere statisticallysignificant l in determiningthe levelofthewinningbid (these variablesincludedthe buyer’s leverageratio, whether thebuyerwas abank or athrifi, and (for bankbuyers) whether the buyerwas located ina differentmarket thanthe depositbase being sold). Takentogether, andinthe context ofthe previousliterature on this subject,these empiricalresults suggestthe followingconclusions: Q Consistentwith popularbelief,duringthe period 1989through 1991,eitherthe two government agencieswere relativelyinefficientatthe process of conducting “plainvanilla”deposit sales,~ the market was slowto assimilateinformation regardingthe existenceof apremiumdifferential. Over the four year span, however, the agenciesand/or the market learnedtheir lessons. The disappearance ofapremiumdifferentialin 1992mayalsobe attributed to the passage of FDICIA in 1991. The Act required the FDIC to pursue “least-cost” resolution procedures and mayhaveinducedthe agenciesto more activelyseek marketdetermined solutionto the dispositionofthe depositbases offailedinstitutions. l “Unintendedvalue” accrued to the winningbidders ofthe failed-bankandfailed- 17

thrifi depositbases duringthe period 1989through 1992onlv~artiallvas a result ofthe agencyand/or market inefficiencies. The statisticalimportance ofbuyer characteristicssuggeststhat depositbases are “independentprivatevalue” objects, and auctiontheory suggeststhat auctionsof suchIPV objects always entailwinningbidderspayinglessthan the true valueofthe objectto the bidder. Thus,to the extent previousevent studiesfound cumulativeabnormalequity returns to the winningbidders ingovernment deals,these CARSwere only partiallydueto agencyor market inefficiencies. We can onlyspeculateover the proximate causes ofthe perceived agencyor market inefficiencies– perhaps the FDIC andtheRTC didindeedfeelpoliticalpressure to accomplishquick saleswithout benefitof optimalmarketingexpenditures. Whatever the reasons, this study suggeststhat pricesdid adjustover time, and, duringthe nextbanking crisis,the disposalofthe depositbases ofthe failedinstitutionsmightbe accomplishedwith fewer misallocativeeffects. References 1)AmericanBanker,1991,“RTCRigidityClobbersTaxpayers,”November 20, pp. 4-13. 2) Balbirer, SheldonD., G. Donald Jud, andFrederick W. Lindahl, 1992,“Regulation, Competition,andAbnormalReturns intheMarket for FailedThrifis,”JournalofFinancial Economics,31, pp. 107-131. 3)Berger, AllenN., John H. Leusner, and JohnJ. Mingo, 1997,“TheEfficiencyofBank Branches,”JournalofMonetaryEconomics,forthcoming,Vol. 40,No. 1,August, 1997. 4) Berkovec, James A. andJ. NellieLiang, 1993,“SelectioninFailedBank AuctionPrices: AnEconometric Model ofFDIC Resolutions,”Finance&EconomicsDiscussionSeries, FederalReserve, 93-40. 5)Befiin, WilliamJ., Farrokh Ghazanfari,andKhalilM. Torabzadeh, 1989,“FailedBank Acquisitionsand SuccessfulBidders’Returns,” FinancialManagement(,Summer), pp. 93- 100. 6)Billett,Matthew T, JaneF. Coburn, andJohnP. O’Keefe, 1995,“Acquirer Gainsin 18

FDIC-Assisted Bank Mergers: TheInfluenceofBidder CompetitionandFDIC Resolution Policies,”FDIC, mimeograph. 7) Bovenzi,JohnF. andMaureen Muldoon, 1990,“Failure-ResolutionMethods andPolicy Considerations,”FDICBankingReview,3,Fall,pp. 1-11. 8)Bovenzi,John F. andArthur J. Murton, 1988,“ResolutionCosts ofBank Failures,”FDIC BankingReview,1,Fall,pp. 1-13. 9) Cole,RebelA., Robert A. Eisenbeis,andJoseph A. McKenzie, 1994, “Asymmetric-Itiormation andPrincipal-AgentProblems as Sources of Valuein FSLIC-Assisted Acquisitionsof InsolventThrifts,”JournalofFinancialServicesResearch, 8,PP.5-28. 10)Davies, SallyM., 1992,“FDIC FailedBank Auctions,”FederalReserve Board, mimeograph. 11)Gilberto, S.MichaelandNikhilP. Varaiya, 1989,“TheWinner’sCurse andBidder CompetitioninAcquisitions:Evidencefrom FailedBank Auctions,”JournalofFinance, XLIV, no. 1,March, pp. 59-75. 12)Gosnell,ThomasF., SylviaC.Hudgins, andJohn A. MacDonald, 1993,“The AcquisitionofFailingThrifis:Returns to Acquirers,”FinancialManagementW, inter, pp. 58-68. 13)Gupta, Atul, RichardL. B. LeCompte, andLalatenduMisra, 1993,“FSLIC Assistance andthe Wealth Effects of SavingsandLoan Acquisitions,”JournalofMonetaryEconomics, 31,PP.117-128. 14)Hirschhorn, Eric, 1985,“BiddingLevelsinPurchase andAssumptionAuctions,” Proceedingsof theConferenceonBankStructureandCompetitionF,ederal ReserveBank of Chicago,May, pp. 369-388. 15)James, Christopher andPeggy Weir, 1987,“AnAnalysisofFDIC FailedBank Auctions,”JournalofMonetaryEconomics,20,pp. 141-153. 16)Pettway, RichardH. and JackW. Trifis, 1985,“Do Banks OverbidWhen Acquiring FailedBanks?”FinancialManagement,14,pp. 5-15. 17)Riley,John G. andWilliamF. Samuelson, 1981,“OptimalAuctions,” American EconomicReview,71,June, pp. 381-392. 19

Appendix A. Variabledefinitions: BANK = 1ifbankbranchesbeing sold;Oifthrifibranches. RATEDIFF = Differencebetween dailyaverage3-month T-billand CD rates for the calendarmonth ofthe announcementdate. STOCK = NASDAQ Bank Stock Index onthe announcementdate. MARKDEP = Market deposits ($billion)as of mid-yearprior to announcementdate. POPOFF = Population per bankingofficeinMSA duringcensusyear prior to announcement. MIDWEST = 1ifbranches soldare inMidwest, Ootherwise;MW includesthe states of IA, IL, ~, KS, KY, MI, MN, MO, ND, NE, OH, SD, andWI. SOUTHEAST = 1ifbranches soldare in Southeast, Ootherwise; SE includesstates of AL, AR, FL, GA, MS, NC, SC, TN, VA, andWV. SAMEMSA = 1ifbuyer and“seller”havebranchesinthe sameMSA prior to announcement;Ootherwise. BUYEQUITY = Thebuyer’stangibleequityto total asset ratio; quarter-end prior to announcement. CORE = Deposits inaccounts lessthan $100,000 dividedbytotal domestic deposits, quarter end prior to announcement. Ratio calculatedfor the soldbranches, when possible;otherwise, for “seller”inthe aggregate. GROWDEP = Percentage growth rate ofbranches, proxiedbygrowth rate of “seller’s”domestic deposits duringyear prior to announcement. JLAM = Mill’sratio. BGLAM = Mill’sratio*Govt (where Govt = 1ifFDIC/RTC transaction in anyyear, Ootherwise). TGLAM = Mill’sratio*Govt*BANK. 20

B. Probit Models for CalculationofMill’sRatios. The probit modelunderlyingthe Mill’sratio calculationisasfollows: P = f( CAMEL88, DELTCAM, AST88, GROUP, GAS) where P —— 1iffailureoccurred duringthe observationperiod, Ootherwise. CAMEL88 = Institution’s composite“CAMEL” supervisoryrating during 1988. DELTCAM = Changeinthe CAMEL ratio from 1987to 1988. AST88 = Total assets in 1988,as ofthe examinationdate; in$millions. GROUP = 1ifinstitution’stotal assets exceed $500millionat 1988examdate. GAS = GROUP*AST88. The probit modelisestimated separatelyfor banksandfor thrifis. Parameter estimatesand standard errors are shownbelow for each ofthetwo probit equations. Bank FailurePrediction Model. Variable Estimate Std.Error INTERCEPT 3.87 .094 CAMEL88 -.714 .027 DELTCAM .216 .047 GROUP -.523 .133 AST88 -.0017 .0003 GAS .00171 .00034 ThrifiFailurePrediction Model. Variable Estimate Std. Error ~TERCEPT 5.19 .208 CAMEL88 -1.11 .056 DELTCAM -.132 .120 GROUP -.335 .188 AST88 -1.157 .712 GAS 1.21 .713 21