+MODEL ARTICLE IN PRESS InternationalReviewofEconomicsandFinancexx(2006)xxx–xxx www.elsevier.com/locate/econbase Earnings management behaviors under different economic environments: Evidence from Japanese banks ⁎ Sumit Agarwal a,1, Souphala Chomsisengphet b, , Chunlin Liu c,2, S. Ghon Rhee d,3 aBankofAmerica,MailStopMD9-978-03-02,1101WoottenParkway,Rockville,MD20852,UnitedStates bOfficeofComptrolleroftheCurrency,RiskAnalysisDivision,250EStreet,SW,Washington,DC20219,UnitedStates c UniversityofNevada-Reno,CollegeofBusinessAdministration,Reno,NV89557,UnitedStates d UniversityofHawaii,DepartmentofFinance,CollegeofBusinessAdministration,2404MaileWay,Honolulu,HI96822,UnitedStates Received20April2005;receivedinrevisedform6May2005;accepted2August2005 Abstract ThispaperinvestigatesJapanesebanks'earningsmanagementbehaviorunderthreedistincteconomicenvironments:(1)high- growthwithassetpricebubbleeconomy(1985–1990);(2)stagnantgrowthwithfinancialdistresseconomy(1991–1996);and(3) severerecessionwithcreditcruncheconomy(1997–1999).Usingbankbalancesheetinformationof78Japanesebanks,wefind thatearningsmanagementbehaviorbyJapanesebanksdifferconsiderablyacrossthethreeperiods.Ourresultsindicatethatbanks usedsecuritygainsasameanstomanageearningsthroughoutallthreeperiods.Wealsofindthatbanksusedloanlossprovisionsto manageearnings;however,thisbehaviorisonlyprevalentduringthefirsttwoperiods.Duetothefactthatbanksfacedrecord-high non-performingloansduringthelattersevererecessionperiod,banksonaveragemayhavebeenrestrainedfromusingloanloss provisions to smooth income and/or to replenish regulatory capital. Consistent with previous studies, we find that the Japanese bankssignificantly lowered their lending with increasedprovisions. ©2006Elsevier Inc.All rights reserved. JELclassification:F34;F42;G28 Keywords:Loanlossprovision;Securitygains;Earningsmanagement 1. Introduction Earningsmanagementoccurswhenmanagersusetheirdiscretionarypowerinthefinancialreportingprocessandin structuringtransactions.Bysmoothingearningsovertime,managersconveyprivateinformationtostakeholdersabout theunderlyingeconomicperformanceofthecompanyorattempttoinfluencecontractualoutcomesthatdependonthe ⁎ Correspondingauthor.Tel.:+12028745386. E-mailaddresses:[email protected](S.Agarwal),[email protected](S.Chomsisengphet), [email protected](C.Liu),[email protected](S.GhonRhee). 1 Tel.:+13015173227. 2 Tel.:+17757846993. 3 Tel.:+18089562535. 1059-0560/$-seefrontmatter©2006ElsevierInc.Allrightsreserved. doi:10.1016/j.iref.2005.08.003 REVECO-00322;NoofPages15 ARTICLE IN PRESS 2 S.Agarwaletal./InternationalReviewofEconomicsandFinancexx(2006)xxx–xxx reported accounting numbers.4 Most empirical studies on bank earnings management practices focus on bank managers' discretionary actions to adjust the timing and size of transactions and accruals in their financial reports. Therearethreemainincentivesforbankmanagerstorecognizeandrecordcertainfinancialtransactions:(i)toachieve regulatorycapitalrequirements;(ii)todefertaxpaymentonearnedincome;and(iii)todisseminateenhancedearnings informationto investors. In the case of Japanese banks, only two studies assess earnings management practices. In the first study, Genay (1998)examinestherelationshipbetweentheperformanceofJapanesebanksandtheirfinancialcharacteristicsduring the 1991–1997 period. The author reports that Japanese banks increased their loan loss provisions when their core earningsas well as thereturns onthe stock marketare high, and attributesthese behaviorsto income smoothing. Recently, Shrieves and Dahl (2003)—henceforth SD—investigate the discretionary accounting practices of Japanese banks operating under the financial constraints imposed by the Basle Accord guidelines. The authors find that,duringthe1989–1996period,surplusregulatorycapitalplaysasignificantlypositiveroleinthelendingdecisions ofJapanesebanks.TheyalsofindthatJapanesebanksusedrealizedsecuritiesgainsandloanlossprovisionstosmooth income,andcapital-constrainedbanks,inparticular,usedearningsmanagementtoreplenishregulatorycapitalduring thisperiodoffinancialduress.Theauthorsthenarguethatbanksusediscretionaryaccountingpracticestoincreasetheir regulatory capital levels without reducing therisk of insolvency—a concept known asregulatory-capitalarbitrage. In this paper, we extend the SD study to investigate whether similar earnings management behaviors of Japanese bankscanbeobservedacrossothertimeperiodsunderextremelydifferenteconomicenvironments.Ourmotivationto addressthisquestionstemsfromthefactsincethemid1980s,Japanesebanksexperiencedupturnsanddownturnsin the economy as well as some structural changes in the function of its financial intermediaries. With bank lending, provisioning, and profitability being typically pro-cyclical (see e.g., Berger & Udell, 2003; Borio, Furfine, & Lowe, 2004), we ask whether earnings management behaviors similar to those documented in the SD study can also be observed in an earlier period of the late 1980s, when the Japanese economy enjoyed high economic growth with escalating asset price bubbles and absent the BIS regulatory capital requirement that allowed the Japanese banks to enjoyhigherassetvaluesandprofits,orinalaterperiodofthelate1990s,markedbysevereeconomicrecession,credit crunch and weakened banking system as the accumulation of non-performing loans further deteriorated the balance sheets of Japanesebanks? Tothisend,wefollowtheSDmethodologyandempiricallyassesstheroleofloanlossprovisionsandrealizedgains from securitiesportfolioson theearningsmanagement practices of78 Japanesebanksovera 15-year period (1985– 1999).BasedonanumberofeconomicandfinancialfactorsthatchangedtheprofitabilityoftheJapanesebankover thissampleperiod(whichwediscussinthenextsection),webreakoursampledataintothreesub-periods:(1)high- growthwithassetpricebubblesera(1985–1990);(2)stagnantgrowthwithfinancialdistressperiod(1991–1996);and (3)severerecessionwithcreditcrunchperiod(1997–1999).Forourbroadsampleof78Japanesebanks,weestimatea simultaneousequationmodelofinvestmentandfinancialdecisions,whichexplicitlyincorporatestheendogeneityof accountingdiscretion with respect tosecuritygains andloan loss provisions. Our findings reveal that Japanese banks on average realized gains from securities in order to offset the negative impactofloanlossprovisionsandtherebyengagedinincomesmoothingthroughoutallthreeperiods.However,the propensityofusinggainstosmoothincomewassmallerforbankswithnegativenondiscretionaryincomeduringthe latter periods of financial distress. On the other hand, we detect that Japanese banks used loan loss provisions as a means of managing earnings only during the high-growth and stagnant periods. During the third period of severe recession and more intensive capital constraints, however, banks on average did not use loan loss provisions to smooth income in the face of declining return on investment. This behavior is consistent with the notion that Japanese banks needed to increase their provisions-to-assets ratio to reflect the accumulating non-performing loans, and not necessarily to manage earnings. On the other hand, banks with negative nondiscretionary income continued 4 Manymotivationsforearningsmanagementhavebeenexaminedintheliteratureincluding:(i)realizationofearningsthresholds[Burgstahler& Dichev,1997;Degeorge,Patel,&Zeckhauser,1999];(ii)initialpublicofferingsandseasonedequityofferings[Rangan,1998;Teoh,Welch,& Wong,1998a,1998b];(iii)incomesmoothing[Beaver&Engel,1996;Greenwalt&Sinkey,1988];(iv)regulatorycapitalplanning[Kim&Kross, 1998;Scholes,Wilson,&Wolfson,1990;Shrieves&Dahl,2002];(v)book-taxreportingdifferences[Mills&Newberry,2001;Phillips,Pincus,& Rego,2002];(vi)debtcovenantrestraints[DeFond&Jiambalvo,1994];(vii)bonusplansandcompensation[DeAngelo,1986;Gaver,Gaver,& Austin, 1995: Holthausen, Larcker, & Sloan, 1995; Pourciau, 1993]; (viii) alteration of risk perceptions and earnings information to investors [Barth, Elliot, & Finn, 1999; Bhattacharya, Daouk, & Welker, 2002]; (ix) government investigation [Bonner, Palmrose, & Young, 1998]; (x) managementofbaddebts[McNichols&Wilson,1988]. ARTICLE IN PRESS S.Agarwaletal./InternationalReviewofEconomicsandFinancexx(2006)xxx–xxx 3 touseprovisionsasameanstosmoothincome,despitetheneedforthesebankstoincreasetheirloanlossprovisions given the enormous share of non-performing loans during the period of severe economic recession and capital constraints. Moreover, we find that the lending activities of Japanese banks differed significantly across the three periods. Relativetothesecondorthirdtroubledperiod,banklendingduringthehighgrowthperiodwashinderedbythehigher realized net dividend earnings and lower loan loss provisions on average. On the other hand, lending (particularly lending of the larger banks) deteriorated considerably following macroeconomic contraction during the stagnant growth period, and significantly increased provisions during the severe recession period. Overall, our findings are consistentwiththeobservationsthatJapanesebankssignificantlyloweredtheirlendingduringthelattertwoperiods followingthe bursting ofthe asset price bubble (Hall, 1993;Horiuchi &Shimizu, 1995). Therestofthepaperisorganizedasfollows.ThenextsectiondescribestheevolutionoftheJapanesebanksfrom 1985tothepresent.Section3presentsthedataandthesimultaneous-equationmethodologythatisusedtomodelthe determinants oflending,securitiesgains,loan lossprovisionsanddividends.Section4presentsdescriptivestatistics and theempirical results. Section 5 provides some concluding remarks. 2. Three economic environments TheJapanesebankingindustryhastraditionallybeenoneofthemostheavilyregulatedindustries.5Beforebanking deregulation, the Ministry of Finance (the internal regulator) traditionally favored more confidentiality and only required limited disclosure of the true financial conditions and accounting practices of a bank's operations. This indirectly allowed banks to conceal financial difficulties. However, with increasing bank deregulation and with the industryhavingtotransitiontoanewregulatorycapitalregimeundertheBasleAccordIintheearly1990s,regulators alsorequiredgreaterdisclosuresthathavemadeiteasiertodeterminethetruevalueofabank'sloanportfolio.Banks withpubliclytradedequityshareswereforcedtocomplywithsometimes-conflictingregulations.Thesechangestothe bankingsystemhaveeitherrequiredorencouragedbankstodisclosetheirequitypositionsintheirfinancialstatements, enabling investors to make more informed decisions. Deregulation has also led to a loss of much of the banks' protection from competition, both domestic and international. Consequently, banks faced competition across their traditionalbusinesslines,leadingbankstodiversifytheirloanportfolios.Allofthesefactorshavecontributedtothe difficulties faced by banks. Below we describe economic and regulatory changes faced by the Japanese banks from 1985 to 1999 and conclude that an examination of the differing earnings management behaviors of Japanese banks under thethree economic environments iswarranted. 2.1. High-growth with asset price bubble economy (1985–1990) Japanesebanksexpandedvigorouslyandenjoyedhighprofitsduringthe1980s.Themainreasonsfortheirsuccess included strong macroeconomic performance of the Japanese economy, a rapidly rising stock market, low domestic interest rates, a relatively strong yen, and a vibrant real estate market. In July 1988, The Banker had the following quote,“Inayearwhichwastheworst-everformanyUSandEuropeanbanks,Japan'sbanksturnedinhigherprofits, increasedtheircapitaland took ina largershare ofworldlendingand capital-market business”(p. 109). Atthesametime,theJapaneseeconomyalsoexperiencedthestrengtheningof“assetpricebubbles”(seee.g.,Okina & Shiratsuka, 2002, 2004; Shiratsuka, 2003). During this period, the asset portfolios of Japanese banks were significantly inflated by the substantial rise in land prices as well as other assets as banks significantly shifted from manufacturingandwholesalesectorstofinance/insurance,realestate,andothersectorsthroughoutthesecondhalfof 1980s(see,e.g.,Tsuru,2001b).Furthermore,withfinancialderegulationpoliciesthatresultedinstructuralchangesin corporatefinanceduringthelate1980s,largefirmscreatedmanysmall-andmedium-sizedsubsidiaries.TheJapanese banks' lending to these subsidiary firms rose substantially (e.g., banks lending to property developers via non-bank financecompaniesknownasjusen)from45%in1985toalmost57%in1990,withtheproportionofloanstothereal estate industry rising from 7% in1985 to12% in 1990(Hoshi,2001). 5 TheJapanesebankingsystemhasbeengoingthroughaverygradualderegulationprocessthatstartedintheearly1980sandhastakenaround25 yearsforitscompletion[Hoshi&Kashyap,2001]. ARTICLE IN PRESS 4 S.Agarwaletal./InternationalReviewofEconomicsandFinancexx(2006)xxx–xxx Moreover, Hoshi (2001) points out that because banks lacked intrinsic information about these new client firms, banksdemandedcollateralinordertoprovideloans.Realestatewasadesirablechoiceforcollateralprimarilybecause the value of real estate had not declined throughout the entire postwar period. In fact, real estate asset prices were experiencing significant increase. Hence, both Ueda (2000) and Hoshi (2001) maintain that Japanese banks shifted towardsportfoliosthatcontainedhighersharesofloanscollateralizedbyrealestate.Unfortunatelyduetothesurgein the value of the collateral, banks did not view it necessary to monitor the real estate related loans (Baba & Hisada, 2002).Giventheriseincollaterallendingpatterns,thebanks'assetportfolioswereextremelyvulnerabletotheriskof assetpricedeclineaswellasperformancedeteriorationoftheconcentratedindustries(seee.g.,Shimizu&Shiratsuka, 2000).Expost,Ueda(2000)empiricallydocumentsthatthelargeincreaseinlandpriceinflationduringthelate1980s and the decline between 1991 and 1996 positively affected the bad loan ratio in 1996. Hoshi (2001) finds that the increaseinrealestateloansduringthelate1980spositivelyandsignificantlycontributedtotheriseintheratioofnon- performance loans in 1998. With stable economic growth, high demand for credit, and increasing asset prices to support collateral lending portfolio, the banking sector grew vibrantly with respect to assets, lending, and profits during the late 1980s. There are numerous reasons why, even during a period of rapid expansion, Japanese banks may have wanted to manage theirearnings.Banksmaydesiretoreducetheirincometaxburdenbysmoothingprofitsovertime.Intheabsenceof the Basle Accord, regulatory capital did not constrain overall bank capital; therefore, banks could finance their lending activities without being constrained by regulatory capital. Thus, banks could easily maintain growth in lending activities as well as stable dividend policy for investors. In addition, banks may also want to display a constantincomestreaminordertogainconfidencefromthepublicand/orinvestors.Thus,itcanbeconjecturedthat bank managers may have reduced the realization of gains from sales of securities by increasing their loan loss provisions as well as dividends in order to smooth their earnings during this period of prosperity, financial growth, and high profits. Alternatively,giventheeffectsofmacroeconomicdevelopmentsandstockmarketperformanceonriskprovisions and earnings, banks may behave in two ways. On the one hand, banks may behave pro-cyclically, e.g., reducing provisionsduringeconomicupswings.Itisplausiblethatbanks,generallyfacedwithveryfewnon-performingloans duringthegoodtimes,arewillingtoreducethelevelofprovisioning.Hence,theremayexistaninverserelationship betweeneconomicgrowthandtheloanlossprovisionsforJapanesebanksduringhighgrowthperiods.Ontheother hand, banks may behave counter-cyclically, e.g., may not reduce and/or may even increase risk provisioning during high-growth periods. The reason is that banks may anticipate a cyclical economy, in which an expansionary period wouldbefollowedbyaneconomicdownturn,andshouldpositionthemforthepossibilitythattheirdebtorsmayhave difficulties repayingtheirloans. 2.2. Stagnant growth with financial distress economy (1991–1996) By 1991, the asset price bubbles began to burst, with mild annual deflation at 1%.6 Shortly thereafter, real estate pricesdeclinedsignificantly.Thefallinrealestatevalueserodedthecapitalbaseofthebanks'balancesheet,andthus significantly contributed to the problems faced by Japanese banks. Between 1990 and 1994, the Nikkei 225 Index declinedby50%.BecauseJapanesebanksheldlargepositionsincommonstock,theysufferedasignificantdeclinein their capital7, compounding thebanking industry'sbalance sheet problem. Economic growth also slowed down dramatically (industrial production growth rate over this 6-year period averaged−0.34%)andtheinstabilityofthefinancialsystemintensifiedwhen,inadditiontothefactthatthecollateral used to support the loans declined in value, the banks experienced a significant increase in the number of non- performingloansandsufferedlargelossesinthevalueoftheirownsecuritiesholdings.8DuringtheJapaneseeconomic downturn, many of the small- to medium-sized firms were unable to make their loan payments. Japan's Ministry of 6 Ithasbeenarguedthatmonetarypolicyfailureswerethemaincauseoflargeassetpricefluctuations(seee.g.,Hoshi,2001). 7 AccordingtoFukao(2001),a10%dropinstockpriceindexreduces20%ofthenetcapitalsincethemarketvalueofstocksheldbybanksis abouttwiceasmuchastheirnetcapital. 8 Moreover,thiswasintandemwiththeoveralldeclineintheJapaneseeconomy.Between1990and1994theaveragegrowthratewas1.5% comparedto5.5%fortheprevious4years.Moreover,monetarypolicytoolswerealsoineffectiveeventhoughtheBankofJapanreducedthe discountrateseventimesinaperiodof3yearsfrom6%to1.75%.Allthesemacroeconomiceventsfurtherledtothebankloanproblems. ARTICLE IN PRESS S.Agarwaletal./InternationalReviewofEconomicsandFinancexx(2006)xxx–xxx 5 Finance (MOF) inspected the jusen companies and found that 40% of their loans were non-performing (Milhaupt, 1999). In fact, all jusen companies were insolvent, with 75% assets non-performing, by mid-1996 (Milhaupt, 1999; Mori,Shiratsuka,&Taguchi,2001).Thus,thecollapseoftheassetpricebubblestimulatedthenon-performingloan problems and substantially weakenedfinancial intermediation. AccordingtotheBankofJapan(BOJ),loanlossprovisionsandloanwrite-offswereatrecordhighat¥13trillionin FY1995and¥7.6trillioninFY1996.Despitethesignificantincreaseinloanlossprovisions,operatingprofitsremained highin1995becausebanksrealizednetgainsonbond-sellingoperations.In1996,however,thebanksoperatingprofits declined5%.Despitetheincreaseinnon-performingloansandtheincreaseinloanlossprovisions,ithasbeenargued and empirically found that rather than writing off bad loans, the banks refinanced the bad loans—this is known as forbearance lending (Kobayashi, Saita, &Sekin,2002). During this post bubble period, the BIS regulatory capital requirements emerged in 1993, further reducing the banks'abilitytolend.Ifcapitalconstrainsbanklending,thenitisexpectedthatlendingwillvarypositivelywithcapital ratios, creating a new incentive for Japanese banks to manage their earnings. Through the use of discretionary accounting practices, banks could maintain or increase their lending, and still meet capital reserve requirements. Specifically, banks that were near the BIS capital-constraint limit had an incentive to alter the appearance of their capital position by increasing equity and net income. As pointed outin Hoshi and Kashyap (2001), despite the long traditionofholdingcompanyshares,thebanksstartedtosellsomeoftheirstockpositionstorealizegainsandsupport their weakened balance sheets.9 According to the capital-arbitrage view (see, e.g., Shrieves and Dahl, 2003), banks could increase their equity position and thereby increase or at least maintain their lending activities by reducing dividendsandincreasingthediscretionarycomponentsofincome(e.g.,increaserealizedsecuritygainsandreduceloan loss provisions). 2.3. Severe recession with credit crunch economy (1997–1999) Thefailuresofmanylargefinancialinstitutions[includingthe10thlargestbank(Hokkaido-TakushokuBank),the 4th largest securities firm (Yamaichi Securities)] as well as the East Asian Crisis hampered the stability of the financial system, and exerted negative economic growth for five consecutive quarters since the 4th quarter of 1997 (Hayakawa&Maeda,2000;Morietal.,2001).Inthispost-bubbleeconomy,therecessioninthewordsofHayakawa andMaeda(2000),“….turnedouttobetheworsteverexperiencedbyJapaninthepostwarera.”(p.1).Inturn,there was a substantial tightening in bank lending of private banks, which resulted in a significant reduction in business fixedinvestment,particularlyinfixedinvestmentofsmallandfinanciallyconstrainedfirms.Hence,thisthirdperiod was marked by a credit crunch (Hayakawa & Maeda, 2000; Motonishi & Yoshkikawa, 1999; Schaede, 2004; Woo, 1999). Japanesebanksenduredacontinuingdeteriorationoftheirbalancesheetsandanintensifyingbankingdistress.In particular,banksfacedextendedeconomicrecession,mountingbadloans,andrelativelyweakcapitalreservepositions that continued to adversely impact their lending positions. Equally important, because a significant portion of their loans used real estate as collateral, Japanese banks should have engaged in prudent risk management practices by increasingtheirloanlossprovisionstoreflectmountingbadloansaswellasadecline inthecollateralvalueoftheir loans.Forexample, theratioof bad loans to total loans increased to6% or 6.5%in thelaterperiod from 3% or less duringtheearlierperiodoffinancialdistress.In1998and1997,netoperatinglosses(definedasthedifferencebetween gross profits and loan losses) peaked at −¥8.3 trillion and −¥7.9 trillion, respectively. This contrasts sharply from positive profits of¥2.5 trillion inboth1991 and 1992.10 Between December 1997 and March 1998, supervisory and regulatory reforms were initiated for the banking industry to govern bank risk regulation, failed bank resolutions, and bank recapitalizations (Daigo, Yonetani, & Marumo,1998;Milhaupt,1999;Tsuru,2001a).About¥7trillionofpublicfundswereinjectedtostabilizethefinancial system, with ¥1.8 trillion to recapitalize 21 banks (Mori et al., 2001). In addition to the public funds, the Japanese governmentalsorevisedaccountingrulesbyallowingbankstoreevaluatetheirpropertyassetsatthemarketpriceand to use the cost method in valuing their equity portfolio (Daigo et al., 1998). As a result, “the use of cost method accountingforstockseasedthepainofbankssufferingfromsharpdropofstockprices,andthereevaluationofproperty 9 Sincemostoftheseshareholdingweredecadesoldandthushadaverylowbasepricecomparedeventotheirpost-peakcurrentmarketprice. 10 SeeFukao(2001). ARTICLE IN PRESS 6 S.Agarwaletal./InternationalReviewofEconomicsandFinancexx(2006)xxx–xxx Table1 Macrosummarystatistics Period1: Period2: Period3: T-testformeandifference 1985–1990 1991–1996 1997–1999 betweenperiods Mean Stddev Mean Stddev Mean Stddev Periods Periods Periods 1and2 1and3 2and3 STOCKS(%change) 0.2095 0.1395 −0.0482 0.1003 −0.1360 0.0018 −29.74 −33.48 −11.83 STOCKS_BANK(%change) 0.1437 0.3630 −0.0800 0.1889 −0.0866 0.2274 −10.84 −7.90 −0.36 LAND(%change) 0.0475 0.0332 −0.0332 0.0188 −0.0413 0.0054 −41.90 −35.89 −5.66 Prime 3.7049 1.1297 2.5002 1.7321 0.5000 0.0000 −11.57 −38.36 −15.61 INDPROD(%change) 0.0443 0.0299 −0.0034 0.0340 −0.0083 0.0129 −20.95 −22.85 −1.88 assets allowed banks to count unrealized gains of property assets as Tier II capital” (Daigo et al., 1998, p. 1). Accordingly,theuseofcostmethodaccountingallowedbankstoincreaseTier1capital,whiletheabilitytorevaluate property assets at the market price allowed banks to increase Tier 2 capital. Hence, we expect a more intensive patternofregulatorycapitalarbitragethanthatobservedintheSD'sstudy,particularlyforthefinanciallyconstrained banks. 2.4. Summary statistics across three sub-periods Table1givessummarystatisticsofmacroeconomicandfinancialcharacteristicsenduredbyJapanesebanksacross three unique time periods. In addition, this table provides t-statistics testing the difference in the mean value of the macroeconomicandfinancialcharacteristicsbetweenthethreeperiods.AscanbeseeninTable1,domesticproduction, asmeasuredbyindustrialproduction(INDPROD),increasedonaverageatanannualrateof4.4%duringthe1985– 1990period,ascomparedtogrowthratesof−0.34%inthe1991–1996periodand−0.83%inthe1997–1999periods. Thestockmarketindex(STOCKS)increasedonaverage20.9%duringthe1985–1990period,whiletheindexreturned −4.8%onaverageduringthe1991–1996periodand−13.6%during1997–1999period.Thelandpriceindex(LAND) declinedfroma4.8%annualgrowthrateinthehigh-growthperiodtoa−3.3%and−4.1%annualgrowthrateduring the last two periods, respectively. Meanwhile, average growth in stock prices of Japanese banks (STOCKS_BANK) roseattherateof14.7%duringtheeconomicandstockmarketupswings,butthendeclinedonaverage8.0%and8.7% duringthetwolatterperiods,respectively.Overall,thet-statisticsrevealthattheJapanesebanksoperatedunderthree distinct macroeconomicenvironments. Table 2 presents the financial characteristics of large city banks and smaller regional banks. These two panels reveal the financial struggles by Japanese banks in the last two periods as a result of the collapse of the asset price bubbleandnon-performingloanproblems.Citybanks'loanlosswriteoffjumpeddrasticallyfromanaverageof¥525 millionduringthehighgrowthperiodtoanaverageof¥9.6billionduringtheeconomicstagnancyandtoanaverage of ¥25.7 billion during the recession period. Consistently, city banks' income on average fell sharply, exhibiting negative net income during the most recent period. Similar trends are observed for the regional banks. Furthermore, thet-statisticsrevealthattheJapanesebanks'profitabilityinthelattertwoperiodssignificantlydifferedfromthefirst period. Summary statistics are also provided for the main variables in our model. For example, the growth in lending activitiesasapercentageoftotalassets(dLOANS)forthecitybanksonaveragedeceleratedfroma6.3%growthrate during the 1985–1990 period to only 1.2% during the 1991–1996 period and −1.1% during the 1997–1999 period. Furthermore,loanlossprovisionsasapercentageoftotalassetsincreasedfromanaverageof0.04%inthefirstperiod to about 0.28% in the second period and 0.86% in the third period. Equally important, average gains from sales of securitiesasashareoftotalassetsdecreasedfrom0.29%oftotalassetsduringthehigh-growthperiodto0.37%during thestagnantperiodandto0.31%duringtheperiodofmoresevererecessionperiod.11Finally,netdividendsdeclined 11 Gainsfromsalesofsecuritiesonaveragewerehigherinthelattertwoperiodsdespitethedeclineinthemarketvalueofoverallequitiesof thebanks,asmeasuredbydeclineinthestockmarketindex.Thisinconsistentobservationisduetothefactthatequityinvestmentswereheldat cost,whichwasstillbelowmarketvalue.Andwhentheyweresoldandrepurchasedbackatmarketvalue,Japanesebanksrealizedtheirlatentgains, increasedthebookvalueoftheirequityinvestments,andtherebyincreasedtheirTier1Capital. Table2 Banksummarystatistics Period1:1985–1990 Period2:1991–1996 Period3:1997–1999 T-testformeandifference betweenperiods Mean Stddev Mean Stddev Mean Stddev Periods Periods Periods 1and2 1and3 2and3 City N 396 392 183 S banks dLOANS 0.0634 0.0438 0.0115 0.0399 −0.0109 0.0341 −17.36 −20.25 −6.56 .A GAINS 0.0029 0.0019 0.0037 0.0038 0.0031 0.0035 3.78 0.87 −1.86 ga r PROV 0.0004 0.0004 0.0028 0.0043 0.0086 0.0129 10.87 12.64 8.01 w a NETDIV 0.0006 0.0001 0.0005 0.0001 0.0005 0.0002 −9.56 −3.12 2.76 le ASSETS(Yen,mills.) 9,738,464.01 13,891,765.31 11,656,399.45 15,854,721.45 11,279,643.93 16,269,502.56 1.81 1.17 −0.26 ta A CASH(Yen,mills.) 163,967.94 2,971,044.66 1,234,839.56 2,119,419.11 588,373.61 1,223,499.58 −2.01 −4.45 −3.84 l./ INVSTMT(Yen,mills.) 1,404,238.32 1,572,385.17 1,772,506.35 1,983,494.28 1,959,481.82 2,556,532.47 2.89 3.21 0.96 Inte R TLIAB 9,471,061.54 13,535,069.18 11,245,730.12 15,335,606.88 10,842,559.44 15,683,789.61 1.72 1.08 −0.29 rn T a INnEcoTm_IeN_tCaOxME(Yen,mills.) 2235,,641100..2407 3347,,842867..9608 −112,668117..3352 6128,,446293..9833 −−44690,795.360.91 5186,42,0475.11.544 −−67..5351 −−78..4458 −−54..0916 tiona IC NONDISC_INCOME(Yen,mills.) 17,797.81 31,150.52 −15,269.62 82,823.33 −13,108.90 124,867.36 −7.43 −4.63 0.25 lR L e ROI 0.0028 0.0023 0.0007 0.0043 0.0016 0.0069 −8.74 −3.29 1.89 vie E Loanlosswrite-off(Yen,mills.) 525.58 1106.87 9663.78 40,123.13 25,751.37 63,288.53 4.53 7.94 3.69 w o LLNLRAESSSERVE(Yen,mills.) 309.6,329769.99 507.0,187107.75 100.697,63958.64 107.087,90686.48 204.678,99483.48 305.086,39981.33 78..4348 118..2886 62..1577 fEco IN n RSRVRAT 0.0041 0.0015 0.0081 0.0070 0.0221 0.0231 11.31 15.49 10.99 o m CAP 0.0312 0.0075 0.0382 0.0072 0.0402 0.0201 13.50 7.86 1.73 ic P s Regional N 132 130 61 a R banks dLOANS 0.0593 0.0471 0.0202 0.0306 0.0203 0.1642 −7.95 −2.52 0.01 nd GAINS 0.0027 0.0024 0.0026 0.0306 0.0017 0.1642 −0.45 −2.61 −1.89 Fin E PROV 0.0009 0.0019 0.0023 0.0030 0.0108 0.0024 3.86 10.20 7.94 an S NETDIV 0.0008 0.0013 0.0005 0.0005 0.0005 0.0005 −2.24 −1.81 −0.40 cex S ASSETS(Yen,mills.) 1,320,214.99 634,069.33 1,695,153.12 716,603.14 1,956,262.07 1,039,053.32 4.49 5.24 2.02 x CASH(Yen,mills.) 93,001.22 164,177.52 101,886.78 124,624.17 93,878.95 99,006.43 0.49 0.04 −0.44 (20 0 INVSTMT(Yen,mills.) 253,510.16 147,837.12 312,872.48 260,952.35 385,751.98 386,036.57 2.27 3.44 1.53 6 ) TLIAB 1,251,824.92 508,248.22 1,592,505.68 583,149.96 1,855,162.62 959,312.49 5.04 5.71 2.34 xx Income_tax 8401.05 25,479.41 2746.73 2525.26 42.03 15,539.57 −2.52 −2.37 −1.94 xx– NET_INCOME(Yen,mills.) 7362.92 22,260.96 −578.32 18,462.34 −10,016.20 31,028.61 −3.14 −4.43 −2.62 xx NONDISC_INCOME(Yen,mills.) 13,488.86 48,183.72 1155.61 20,705.27 −261.78 29,267.65 −2.68 −2.06 −0.38 ROI 0.01 0.0146 0.0010 0.0103 0.0005 0.0099 −3.22 −2.66 −0.28 Loan-losswrite-off(Yen,mills.) 187.63 227.10 518.48 861.95 1230.11 4588.04 4.26 2.61 1.71 LNASS 0.65 0.1444 0.6913 0.1629 0.7016 0.1614 2.34 2.36 0.41 LLRESERVE(Yen,mills.) 6151.83 436,6359 11,607.16 9622.48 377.37.90 24,485.19 5.92 14.38 10.57 RSRVRAT 0.01 0.0045 0.0079 0.0069 0.0235 0.0174 3.46 11.26 8.94 CAP 0.04 0.0345 0.0482 0.0436 0.0473 0.0632 1.98 1.25 −0.10 7 ARTICLE IN PRESS 8 S.Agarwaletal./InternationalReviewofEconomicsandFinancexx(2006)xxx–xxx slightly from 0.054% in the first period to 0.048% and 0.040% during the two periods of financial distress. Similar trends are again observed for the regional banks. 3. Data and methodology 3.1. Data Our bank-level data are obtained from the Pacific-Basin Capital Market Research Center (PACAP) Database- Japan, which is jointly created and maintained by the University of Hawaii and University of Rhode Island in collaboration with the Daiwa Institute of Research and the Tokyo Keizai Inc. We compile a large data set, with annual income statements and balance sheet data for 78 Japanese banks over a 15-year study period, 1985–1999. Moreover, our sample also includes city and regional banks that are listed on the Tokyo Stock Exchange (TSE).12 Ourinitialsampleincludes13citybanksand87regionalbanks.WeexcludebanksthatceasetobelistedontheTSE for reasonsoffailure or nationalization.Mergers areaccountedfor bytreating the merged bankasonefor theentire sample period. We also exclude banks with no information on their financial statements. Our final sample has 78 banks thatarecomparabletoSD'ssample.Theyreport thatthenumberofbanks intheirfinalsampleisbetween 67 and 79 varying by year. 3.2. Methodology WefollowtheSDmethodologyandestimateasimultaneousequationmodeltoanalyzethediscretionaryaccounting practices of Japanese banks on four decision variables: lending, securities gains, loan loss provision, and dividends. This model assumes that Japanese banks periodically and simultaneously adjust their decisions to achieve their objective of income smoothing and/or capital arbitrage. Specifically, the equations for securities gains and loan loss provisionofferempiricalevidenceonhowJapanesebanksusediscretionaryaccountingtosmoothreportedearnings, while the equations for lending and dividends reflect how banks make decisions regarding investment and financial activities, respectively.13 These four financing decisions are functions of financial bank attributes as well as other exogenous variable. dLOANSit ¼þa0aþ6CaA1RPEMGIiDtþit−1a2þAaS7SCEATPSHit−I1it−þ1þa3La8NRAOSISitiþt−1aþ9ðRa4OINI⁎DNREOGDÞiittþþaa150CGAAPILNOSsiti−t1 þa PROV þa NETDIV þe ð1Þ 11 it 12 it it GAINSit ¼bþ0bþ6CbA1RPELGOiitt−þ1þb2bA7SCSAEPTMSiIt−D1itþ−1bþ3LbN8CAASPSHit−I1it−þ1þb3SbT9ROOCIKitiþtþb1b05ðPRROIIM⁎NEiEtGÞit þb dLOANS þb PROV þa NETDIV þn ð2Þ 11 it 12 it 13 it it PROVit ¼þd0dþ6CdA1RPEMGIiDtþit−1d2þAdS7SCEATPSHit−I1it−þ1þd3Rd8SRROVIRitAþTdit9−ð1RþOId⁎4NLEAGNÞDititþþdd105dCLAOPALNOSit−si1t þd GAINS þd NETDIV þw ð3Þ 11 it 12 it it NETDIVit ¼g0þg1REGitþg2ASSETSit−1þg3N⁎ETDIVit−1þg4CAPLOit−1þg5CAPMIDit−1 þg6CAPHIit−1þg7ROIitþg8ðROI NEGÞitþg9dLOANSsitþg10GAINSit þd NETDIV þx ð4Þ 10 it it 12 Longtermcreditbanksandtrustbanksareexcludedfromoursamplefortworeasons.First,wewanttomakeourresultscomparabletoSD (2003)inwhichtheseinstitutionsareexcludedfromtheirsample.Second,aspointedoutinSD(2003),thereasonfornotincludinglong-termcredit banksandtrustbanksistoreduceheterogeneityduetospecializationinlendingactivitiesoftheseinstitutions. 13 Severalotherstudieshaveusedasimilarmodelspecification,likeGreenawaltandSinkey(1988)andMoyer(1990).However,ourspecification isbasedontheworkbyShrievesandDahl(2002),wheretheyalsomodelasimultaneousmodelwithfourequations. ARTICLE IN PRESS S.Agarwaletal./InternationalReviewofEconomicsandFinancexx(2006)xxx–xxx 9 Table3 Expectedsignsofkeyvariablesunderincomesmoothingandcapitalarbitrage dLOANS GAINS PROV NETDIV ROI CAPLO CAPMID CAPHI dLOANS NA + − − + + + + GAINS + NA + + − − − − PROV − + NA − + + + + NETDIV − + − NA + + + + WeusethesamedefinitionsasinShrievesandDahl(2003)forallvariablesinthesimultaneousequationsystem.14 Theonlydeviationfromtheirdefinitionisinourmeasureofcapitalratios.Insteadofusingsurplusregulatorycapital ratios(BIS)asinShrievesandDahl'sstudy,weusetheoverallequitycapitalratios(CAP)15becauseofthedifficultyin measuringrisk-weightedcapitalduringthetransitionperiod,duringwhichthebankingregulatordidnotfullyenforce the reported BIS ratios in Japan. We are comforted by Shrieves and Dahl's acknowledgement that the use of CAP measures leads to similar conclusions. The endogenous and exogenous variables as well as bank characteristics are definedas follows.16 Thefourendogenousvariablesaredefinedasfollows.dLOANSisa1-yearchangeinabank'stotalloansdividedby thebeginningoftheyeartotalassets,GAINSisthegainsorlossesonasaleofsecurities17dividedbythebeginningof theyeartotalassets,PROVisthebank'sprovisionforloanlossesdividedbythebeginningoftheyeartotalassets,and NETDIVisthe bank's cash dividends (net of stock issuance) divided bythe beginningof theyear total assets. Thefourendogenousvariablesareregressedsimultaneouslyonasetofbankfinancialcharacteristics.Thevariable REGisadummyvariablethatequalsoneifabankisaregionalbankandzeroifitisacitybank.ASSETS_t−1islogof total assets lagged by 1 year, while the LNASS_t−1 is the 1-year lag of total loans divided by total assets and the RSRVRAT_t−1 is the 1-year lag of loan loss reserves divided by total assets. CAPi_t−1 is the 1-year lag of equity capitaldividedbytotalassetsinthei=lowest,middle,andhighestquartiles.ROIisthenondiscretionaryearnings18 dividedbytotalassets,whileNEGisadummythatequalsoneifabankhasnegativenondiscretionaryincome,andzero otherwise. Wealsocontrolforasetofexogenousvariables.ThevariableSTOCKSisa2-yearaveragepercentagechangeinthe TokyoStockExchangepriceindex.PRIMEisa2-yearaveragechangeinthelong-termprimelendingrate,LANDisa 2-yearaveragepercentagechangeinthelandpriceindex.AndINDPRODisa2-yearaveragechangeintheindustrial productionindex. Sinceourfocusisonincomesmoothingandcapitalarbitragebehavior,wesummarizeinTable3,theexpectedsigns ofkeycoefficientsineachequation.Inthelendingequation(dLOANS),ifregulatorycapitalconstrainsbanklending, then we expect (i) the three capital-to-assets ratios (CAPLO, CAPMID, CAPHI), return on investment (ROI), and gains/lossesonsaleofsecurities(GAINS)tobepositivelyrelatedtodLOANS;and(ii)loanlossprovisions(PROV) andnetdividendpayouts(NETDIV)tobenegativelyrelatedtodLOANS.Intheequationforgains/lossesonsaleof securities (GAINS) equation, if income smoothing and capital arbitrage behavior exists, we expect: (i) lending 14 WeconductedtwoformaltestsfortheTwo-StageLeastSquare(2SLS)model.ThefirsttestexamineswhethertheparametersfromOLSand 2SLSmethodsareequal,astheywouldbeintheabsenceofsimultaneityormodelmisspecification.ThespecificationtestofHausman(1978) rejectsthehypothesisofequivalenceoftheparametersineachofthefourequations,supportingtheassumptionofendogeneityinlending,security gains,loanlossprovision,andnetdividends.Thesecondtestformodelmisspecificationconsistsofaregressionoftheresidualsfromeachequation inthe2SLSmodelonalltheinstrumentsandcomputingtestsofthehypothesisthatthecoefficientsineachequationarejointlyzero;theresults showthatnoneofthetestsweresignificantateventhe10%level. 15 Totalequitycapitaldividedbytotalassets. 16 Someofourvariabledefinitionspansintodifferenteras;however,webelievethisisnotaproblem.Thereasonisbecausetheextenttowhich banksmanagetheirearningsinthecurrentyearisdependentontheirbalancesheetlastyear.Thereisnoreasontobelievethecausalitywilltotally break down as the economic environment changes. In other word, the lag variables serve as controls for firm specific characteristics. Even if external conditions may change, those constraints may still affect a bank's behavior. It is important to remember that earnings management behaviorsareinfluencedbybothexternalandinternalfactors.Hence,theexternalfactoriseconomicenvironmentsandlagvariablesareproxyfor internalfactors.Weonlyexaminehowabankrespondstothechangesoftheexternalfactor. 17 Includesgainsonforeignexchangesold,tradingaccountsecuritiestransactions,maturityofsecurities,nationalandotherbondsredeemedless lossonsecuritiessold. 18 Non-discretionaryearningsiscomposedofincomelessexpenses,provisionsforloanlosses,otherrevenue/expenses(calculatedbyPCAPto include reversal of special/specific reserves, provision of special/specific reserves, special profit, non-operating income, special loss, and non- operatingexpenses)andgains/lossesonsalesofsecurities. ARTICLE IN PRESS 10 S.Agarwaletal./InternationalReviewofEconomicsandFinancexx(2006)xxx–xxx (dLOANS),loanlossprovisions(PROV),aswellasnetdividendpayouts(NETDIV)tobedirectlyrelatedtoGAINS; and (ii) the three capital-to-assets (CAPLO, CAPMID,CAPHI)and thereturnoninvestment (ROI) tobe negatively related to GAINS. Likewise, if income smoothingand capital arbitrage behaviors exist,then: (i)the threecapital-to- assets ratios (CAPLO, CAPMID, CAPHI), the return on investment (ROI) and gains/losses from sale of securities (GAINS) should positively impact loan loss provisions (PROV); (ii) lending (dLOANS) as well as net dividend payouts(NETDIV)shouldbeinverselyrelatedtoPROV.Andfinally,CAPLO,CAPMID,CAPHI,ROI,andGAINS (dLOANS and PROV) should positively (negatively)affect netdividendpayouts (NETDIV). 4. Empirical results We first estimate the four equations using the same period as in the SD study (1989–1996). While we do not however formally include the table of results in the paper,19 our results overall are fairly consistent with the SD's evidence of earnings management by Japanese banks. On the other hand, when we estimate the model for three different periods: high-growth period (1985–1990); economic stagnant period (1991–1996); and severe recession period(1997–1999),thereareinterestingsimilaritiesanddifferencesbetweenourresultsandthoseofSD.Hence,the breakdownoftimeperiodsshouldenableustocompareandcontrastJapanesebanks'decisionsinlending,gainsfrom sales of securities, loan loss provisions, and net dividend across the three distinct economic environments. Table 4 presents the empirical results of each financial decision variable for each of the three periods. Belowwe discuss the results. 4.1. Bank lending ThefirstequationinTable4showstheempiricalresultsforvariousfactorsthatdeterminebank-lendingdecisions (dLOANS).Overall,lendingbehaviorsdiffersignificantlyacrossthethreeperiods.Specifically,ourresultsshowthat nondiscretionaryincomeasapercentageofassets(ROI)issignificantlyandpositivelyrelatedtolendingonlyduring highgrowthperiod.Also,anincreaseinnetdividendpayout(NETDIV)significantlyreducesbanklendingonlyduring high growth period. Furthermore, loan loss provisions as a percentage of assets (PROV) has an inverse relationship with banks' loan growth only during the high-growth and the severe recession periods, but not during the period of economicstagnant.Ontheotherhand,therealizationofsecuritygainssignificantlyincreasesbanklendingduringthe high-growthperiod,whilesignificantlyreducesbanklendingduringthestagnatedeconomy.Theseresultsimplythat bank lending was constrained by discretionary accounting factors duringthe high growthperiod. Inaddition,wefindthatlendingissignificantlycurtailed(1)bythelowerdemandforloans(measuredbythegrowth ofindustrialproduction)inthesecondperiodofeconomicstagnation;(2)forlargerbanks(asmeasuredbyassets)inthe second period; and (3) by the significant increase in loan loss provisions in the third period. On average, lending activitiesforthesmallerregionalbankswererelativelyhigherthanthoseofthelargercitybanksduringthelattertwo troubled periods. All three ranges of capital ratios for (CAPLO, CAPMID, CAPHI) are surprisingly negative and insignificant in determiningbank-lendingactivitiesacrossallthreesub-periods,suggestingthatlendingbybankswerenotconstrained bythelackofcapital.Theimplicationofthisfindingisthatincreaseinbanks'capital-to-assetsratioscouldhavebeen aimed to improve their capital position rather than to expand their loan portfolio. Similarly, Horiuchi and Shimizu (1995) find that Japanese banks that increased their equity capital had a reduction in lending. Likewise, Hall (1993) argues that in order to maintain adequate levels of capital during 1990–1993, Japanese banks reduced their loan portfoliosas a way toreduce the percent of risky assets intheirportfolios. 4.2. Security gains ThesecondequationinTable4illustratestheimportantdeterminantsofbankdecisionstorealizegainsfromsalesof securities across the high-growth, financial distress, and banking crisis periods. Overall, our empirical evidence suggeststhatbanksconsistentlyusedgainsasameanstomanagetheirearningsthroughoutallthreeperiods.Whilethe degreeofusingrealizedgainstosmoothincomeforbankswithnegativenondiscretionaryearningsisstrongerduring 19 ThetableofresultsfortheSDtimeperiodisavailableuponrequest.