Accountng dscreton of banks durng a fnancal crss Harry Huznga * (Tlburg Unversty and CEPR) and Luc Laeven (Internatonal Monetary Fund and CEPR) November 6, 2009 Abstract: Ths paper shows that banks use accountng dscreton to overstate the value of dstressed assets. Banks balance sheets overvalue real estate-related assets compared to the market value of these assets durng the US mortgage crss. Banks wth large exposure to mortgage-backed securtes also provson less for bad loans. Furthermore, dstressed banks use dscreton n the classfcaton of mortgage-backed securtes to nflate ther books. Our results ndcate that banks balance sheets offer a dstorted vew of the fnancal health of the banks. Key words: fnancal regulaton, accountng standards, corporate dsclosure, asymmetrc nformaton JEL Classfcaton: G14, G21 * Huznga s Professor of Economcs at CentER, Tlburg Unversty, and Research Fellow at CEPR; Laeven s Deputy Dvson Chef, Research Department at the Internatonal Monetary Fund, Professor of Fnance at CentER, Tlburg Unversty, and Research Fellow at CEPR. We would lke to thank Alexander Bleck, Stjn Claessens, Rocco Huang, Chrstan Laux, Chrstan Leuz, Joe Mason, Lev Ratnovsk, Douglas Sknner, Kenneth Sullvan, Wolf Wagner, and semnar partcpants at the Bank of Canada, the Federal Depost Insurance Corporaton, the Internatonal Monetary Fund, and the Unversty of Chcago Booth School of Busness for comments or suggestons, and Matta Landon for excellent research assstance. The fndngs, nterpretatons, and conclusons expressed n ths paper are entrely those of the authors. They should not be attrbuted to the Internatonal Monetary Fund. Contact nformaton: Harry Huznga: H.P.Huznga@uvt.nl; Luc Laeven: llaeven@mf.org.
1. Introducton At tmes of fnancal crss when asset markets become dstressed, large dfferences between book and market values of bank assets may arse, especally when assets are carred at values based on hstorcal cost. Such dfferences gve rse to ncentves for banks to use accountng dscreton to preserve the book value of the bank, for example, by usng advantageous valuaton technques and makng favorable assessments of asset mparment. As a consequence, dscreton n accountng rules causes banks to understate underlyng balance sheet stresses and to overstate regulatory captal. Durng the recent fnancal crss, large dfferences have arsen between the market values and book values of the assets of U.S. banks. Specfcally, by end-2008, 60% of U.S. bank holdng companes had a market-to-book value rato of assets below one, compared to only 8% of banks at the end of 2001. Durng ths perod, the market values of some bank assets, such as mortgage-backed securtes (MBS), declned sharply, as nformaton asymmetres about the qualty of these assets led to a collapse of the MBS market (see Gorton, 2008, Brunnermeer, 2009, Damond and Rajan, 2009, and Krshnamurthy, 2009). The average rato of Ter 1 captal to bank assets, however, has declned only slghtly from 12% to 11% over ths perod. The market value of bank equty thus has dropped precptously aganst a backdrop of vrtually constant book captal. Ths suggests that values of bank assets on bank balance sheets are generally overstated by year-end 2008, and t rases doubts about the relevance and relablty of bank s accountng nformaton - the two man crtera by whch accountng systems are generally evaluated - at a tme of fnancal crss. Ths paper shows that banks have systematcally understated the mparment of real estate related assets n ther publc accounts snce the onset of the current fnancal crss. We provde 1
the frst evaluaton of such behavor, and offer three peces of compellng evdence to support our thess that banks use accountng dscreton to overstate the book value of captal. Frst, we estmate large market dscounts mplct n stock prces on real estate related assets as such mortgage loans and MBS. To estmate mplct market dscounts on bank assets, we emprcally relate Tobn s q, computed as the market-to-book value of assets, to banks asset exposures usng quarterly accountng data on U.S. bank holdng companes for the perod 2001 to 2008. Our prmary focus s on real estate related assets, as these assets consttute a large fracton of the total assets of the average bank, and as recent declnes n U.S. real estate prces have rased doubts about the underlyng value of these assets. We fnd sgnfcant dscounts on banks real estate loans startng n 2008, averagng about 17% on average. As the average bank holdng company n 2008 holds about 53% of ts assets n the form of real estate loans, the mplct dscount n loan values goes a long way towards explanng the current depressed state of bank share prces. We further fnd that nvestors started dscountng banks holdngs of MBS n 2008. For that year, we fnd an average dscount on these assets of 14%, whle the average MBS exposure amounted to 10% of assets. The market dscount on MBS that are avalable-for-sale (and carred at far value) s estmated to be 12%, aganst a larger dscount of 22% for MBS that are held-to-maturty (and carred at values based on hstorcal cost). These varous dscounts suggest that banks have used ther dscreton n determnng the book values of real estate and MBS to lmt book-value mparment of these assets. Second, ths paper consders bank behavor regardng ther loan loss provsonng and loan charge-offs n the current fnancal crss. A bank that holds back on ts loan loss provsonng mantans hgh book values of ts loans and preserves regulatory captal, potentally 2
explanng large mplct dscounts on ts loan portfolo. In practce, banks have consderable leeway n settng the loan loss provsonng for bad loans and n realzng loan losses n the form of charge-offs, potentally provdng room for manpulaton. We focus on the use of dscreton regardng the accountng for loan mparment, as loans are by far the largest asset category, amountng to 71 percent of assets on average n 2008. We fnd that banks wth large exposure to MBS reported sgnfcantly lower loan loss provsons n 2008. Ths ndeed suggests that weakened banks manpulate ther loan loss provsonng to manage ther regulatory captal durng the present crss. Specfcally, we fnd that a one standard devaton ncrease n MBS exposure (relatve to assets) leads the average bank to reduce loan loss provsonng n 2008 by an amount that s equvalent to 0.8 percent of Ter 1 captal. Ths s a sgnfcant savng n regulatory captal. For troubled banks, wth aboveaverage real estate related exposure and below-average Ter 1 captal, the captal savng effect would be even stronger. Thrd, we examne banks choces regardng the classfcaton of MBS as ether held-tomaturty or avalable-for-sale. In 2008, the far value of especally non-guaranteed MBS tended to be less than ther amortzed cost. Ths mples that banks could augment the book value of assets by classfyng MBS as held-to-maturty. Indeed, we show that the share of MBS that are held-to-maturty ncreased substantally n 2008. The ncentve to classfy MBS as held-tomaturty rather than avalable-for-sale should ncrease n the dfference between the amortzed cost and far value of a bank s MBS portfolo. Consstent wth ths, we fnd that the share of MBS kept as held-to-maturty durng 2008 s sgnfcantly related to the gap between the amortzed cost and far value of MBS exposures. 3
Taken together, the evdence of ths paper shows that banks have consderable dscreton n settng the book value of assets hgher than values mplct n stock prces and to lmt asset mparment. Overall, accountng dscreton enables banks to soften the mpact of the crss on the book valuaton of assets durng the present fnancal crss. Some accountng dscreton s unavodable, as accountng systems n part are mechansms for frms to reveal asymmetrc nformaton to nvestors and other outsde partes (Shackelford et al., 2008). Accountng systems wth consderable dscreton, however, may delver hghly naccurate accountng nformaton at a tme of great turmol, such as the present fnancal crss, wth potental real consequences for the allocaton of captal n the economy (Keda and Phlppon, 2006, and Leuz and Wysock, 2008). Inaccurate accountng nformaton can be especally harmful n the case of banks, as t may lead to regulatory forbearance wth concomtant rsks for tax payers. There s ndeed ample evdence of such regulatory forbearance n prevous crss epsodes n the U.S., Japan, and elsewhere (see, for nstance, Kane, 1989, Kroszner and Strahan, 1996, Barth et al., 2006, Caballero et al., 2008, Sknner, 2008, and Brown and Dnc, 2009). 1 In the present crss, the fnancal statements of banks appear to overstate the book value of assets to the pont of becomng msleadng gudes to nvestors and regulators alke. 2 Thus, the present crss can be seen as a stress test of the accountng framework that reveals that book valuaton need not always reflect the best estmate of asset value, especally at a tme of sharp declnes n market values. Our paper relates to a large lterature on the costs and benefts of earnngs management of frms (see, e.g., Leuz et al., 2003, and Hutton et al., 2008), and the management of loan loss 1 For example, Sknner (2008) shows that Japanese bank regulators allowed banks to use deferred tax accountng to bolster ther banks' regulatory captal levels when ther economc crcumstances deterorated durng the Japanese bankng crss. 2 The outcomes of stress tests of major U.S. banks conducted by the U.S. Treasury n 2009, whch calculated captal shortfalls at several major banks, are testmony to the fact that publcly avalable accountng nformaton at the tme provded an nadequate pcture of the health of the concerned banks. 4
provsonng n partcular (see, e.g., Moyers, 1990, Beatty et al., 1995, Collns et al., 1995, and Ahmed et al., 1999). A theoretcal lterature outlnes that managers of frms may have ncentves to smooth reported accountng ncomes ether to smooth ther own compensaton, to ncrease ther job securty, or to ncrease frm valuaton by nvestors (see, e.g., Trueman and Ttman, 1988, Fudenberg and Trole, 1995, Sankar and Subramanyam, 2000). There s also work on the costs and benefts of enhanced corporate dsclosure and accountng transparency (see Leuz and Wysock, 2008, for a revew). For example, Karpoff et al. (2008) usng frm-level nformaton on legal enforcement actons show that fnancal msrepresentaton has reputatonal consequences for frms and depresses frm valuaton. A related lterature revewed by Barth et al. (2001) and Holthausen and Watts (2001) asks whether accountng nformaton s value relevant n the sense that t conforms to the nformaton that bank shareholders use to prce bank shares. Barth et al. (1996) and Eccher et al. (1996) fnd that far value estmates of loan portfolos and securtes help to explan bank share prces beyond amortzed cost. There s also recent work on the market prcng of bank assets reported under dfferent far valuaton technques (e.g., Kolev, 2009, Goh et al., 2009, and Song et al., 2009). Ths work shows that level 3 assets, whch banks value wth relatvely much dscreton, tend to be dscounted the most. Ths fndng s consstent wth our evdence that the market attaches dscounts to real estate related assets that are valued wth consderable dscreton. Recent work also nvestgates concerns about the potental procyclcal nature of far value accountng, whch could magnfy fluctuatons n bank lendng and economc actvty (see IMF, 5
2009, and Heaton et al., 2009). Laux and Leuz (2009) fnd lttle evdence that such effects are the result of far value accountng. 3 Our paper s part of an emergng lterature on the causes and effects of the 2007 U.S. fnancal crss. Ths work shows that house prce apprecaton (e.g., Demyanyk and Van Hemert, 2008) and asset securtzaton (e.g., Keys et al., 2008; Man and Suf, 2008; Loutskna and Strahan, 2009), combned wth a more general deteroraton of lendng standards by banks (e.g., Dell Arcca et al., 2008), helped fuel a crss n U.S. mortgage markets, wth bank captal beng eroded as the asset prce bubble n real estate markets burst startng n 2007. The paper contnues as follows. Secton 2 presents emprcal evdence on market dscounts of real estate related assets, relatve to book values, durng the present fnancal crss. Secton 3 examnes the use of bank dscreton regardng loan loss provsonng and loan chargeoffs, and the classfcaton of MBS nto dfferent accountng categores. Secton 4 concludes. 2. Tobn s q and market dscounts 2.1 Dvergence between market value and book value of bank assets In ths study, we consder U.S. bank holdng companes that are stock exchange lsted. These companes report a range of accountng data to the Federal Reserve System by way of the Report on condton and ncome (Call report). We use quarterly data from these Call reports from the fnal quarter of 2001 tll the end of 2008. Ths covers a full busness cycle as defned by the Natonal Bureau of Economc Research (NBER) from the prevous recesson whch ended n November 2001 untl the current ongong recesson whch started n December 2007. 3 A common vew s that bank regulaton should target any undesrable credt procyclcalty drectly, for nstance by prescrbng cyclcal captal requrements (for a more detaled dscusson of ths debate, see Laeven and Majnon, 2003, Kashyap and Sten, 2004, and Repullo and Suarez, 2008). 6
Takng stock market data from Datastream, we use the market value of common equty plus the book value of preferred equty and labltes as a proxy for the market value of a bank s assets. Tobn s q s then constructed as the rato of ths proxy for the market value of bank assets and the book value of assets. Fgure 1 reports the average Tobn s q per quarter over our sample perod. The mean value of q has declned from 1.06 n the fnal quarter of 2001 to 1.00 n the fnal quarter of 2008. Ths suggests that over ths perod, the market value of bank assets has declned more than ts book value. We defne a zombe bank as a bank wth a q of less than one. 4 The declne of the average q has been accompaned by an ncrease of the share of banks that are zombe banks. As presented n Fgure 1, the share of zombe banks has ncreased from 8.0% at the end of 2001 to 59.6% at the end of 2008. Durng ths perod, the share of zombe banks has tended to be smaller than n 2001 and 2008 reflectng an upswng of the busness cycle. In fact, the share of zombe banks reached a low of 0.3% durng the second quarter of 2004. Aganst a background of declnes n many banks equty prces, banks regulatory captal, as measured by the rato of Ter 1 captal to total rsk-weghted assets, has remaned surprsngly stable throughout the sample perod. Fgure 2 shows the development of the Ter 1 captal rato and the share of Ter 1 captal n total bank captal. Whle leverage ncreased for some banks, consstent wth fndngs by Adran and Shn (2008), the average Ter 1 captal rato decreased only modestly from 12.0% n 2001 to 11.2% n 2008. 5 The composton of captal also changed only modestly over the sample perod, wth the share of Ter 1 captal n total captal shrnkng from 88.1% n 2001 to 86.4% n 2008. Ths suggests that, although some banks may have looked 4 The term zombe bank has frequently been used n the context of Japan durng the 1990 s bankng crss when Japanese banks contnued to lend to unproftable borrowers (e.g., Caballero et al., 2008). 5 Ter 1 captal represents the core component of captal for banks and s regarded as the key measure of a bank s fnancal strength from a regulator s pont of vew. Ter 1 captal conssts prmarly of common stock, retaned earnngs, and dsclosed reserves. 7
for less tradtonal, non-core sources of captal, such as subordnated debt or perpetual stock, to boost captal and ncrease assets, most banks contnued to do so whle ncreasng Ter 1 captal and mantanng excess regulatory captal. 2.2 Relatonshp between Tobn s q and market dscounts Lower values of Tobn s q for banks suggest that the valuaton of bank assets mplct n the stock market has declned more that the book values of these assets. In fact, by relatng Tobn s q to banks exposure to varous asset classes, we can nfer market dscounts on these assets relatve to ther book values. 6 To mplement ths, let MV be the market value of the bank. Further, let A be the accountng value of asset and let L be the accountng value of lablty. If we assume there are operatng markets for a bank s assets and labltes, then we can state a bank s market value as follows: MV a v A = v L (1) l where a v s the market value of asset and l v s the market value of lablty. 7 We can now defne q as the market value of the equty of the bank plus the book value of all labltes dvded by the book value of all assets as follows: q = MV + A L Substtutng for MV from (1) nto the expresson for q, we get: 6 In smlar fashon, Sachs and Huznga (1987) estmate dscounts on thrd world debt on the books of U.S. banks at the tme of the nternatonal debt crss of the 1980s. A related lterature, startng wth Lang and Stulz (1994) and ncludng Laeven and Levne (2007), has studed dscounts n Tobn s q arsng from corporate dversfcaton. In that lterature, dscounts are computed for each busness unt of a conglomerate wth respect to the value of comparable stand-alone frms, whle here we compute dscounts for dfferent assets and labltes of the same bank. 7 In eq. (1), we gnore that market value may depend on the co-exstence of certan assets and labltes as dscussed n, for nstance, DeYoung and Yom (2008). 8
a + q = 1 d a d l (2) l where d a = 1 v, a d l = 1 v, l a = A A and l = L A. The coeffcents a d and l d, that are to be estmated, are the dscounts mplct n the bank s stock prce of a bank s assets and labltes relatve to book values. The varables a and l are the current book values of partcular assets and labltes relatve to the book value of all assets. 8 In eq. (2), Tobn s q wll dffer from 1 f at the valuaton of at least one balance sheet tems mplct n the stock market dffers from ts book value. 2.3 Data on bank balance sheet tems Ths secton dscusses the bank asset and lablty categores that enter the subsequent emprcal work on market dscounts. Of man nterest n ths analyss wll be banks exposure to the real estate market, whch comes n the form or real estate loans and MBS. To reflect banks drect exposure to real estate loans, we construct the rato of real estate loans to overall assets. From 2001 to 2008 the asset share of real estate loans has ncreased substantally from 45.5% to 52.9% for the average bank holdng company as reflected n Fgure 3. Thus, about half of the average bank s assets consst of real estate loans by 2008. Banks exposure to MBS n turn s measured by the share of MBS n all assets. Ths MBS share ncreased only slghtly from 9.9% n 2001 to 10.3% at the end of 2008. 8 Current book values of, say, real estate loans could already reflect some loan loss provsonng. Estmated dscounts on bank assets then reflect the dfference between market percepton of asset mparment and the recognton of ths mparment through reported loan loss provsonng (rather than the dfference between market value and orgnaton value). Put dfferently, the estmated dscount reflects the dfference between market percepton of any asset mparment and the accountng treatment of ths mparment. 9
Whle there has been a move towards far value accountng of bank assets, most assets of the average bank, ncludng mortgage loans held for nvestment, are stll reported based on hstorcal cost. 9 The book value of MBS reflects dfferent accountng conventons dependng on whether these securtes are held-to-maturty or avalable-for-sale. MBS classfed as held-tomaturty are carred at amortzed cost. Ths amortzed cost may be adjusted perodcally for amortzaton and captalzed nterest, and t may also reflect prevous loan loss provsonng. However, these adjustments to amortzed cost are lkely to be relatvely small so that amortzed cost s relatvely close to orgnaton values. MBS classfed as avalable-for-sale, n contrast, are to be carred on the books at far value. Accordng to exstng account rules (FAS 159), securtes are to be classfed as held-to-maturty and carred at amortzed cost, f management has the ntenton to hold them untl maturty, whle otherwse they are avalable-for-sale and carred at far value. 10 Far value s meant to reflect observed market values (of ether the underlyng asset level 1 assets or a comparable asset level 2 assets) or otherwse reflect the outcome of a bank s own valuaton models (level 3 assets). 11 Agan, banks assessments of far value may dffer across bankng nsttutons as the determnaton of far value n practce leaves banks wth 9 The majorty of (real estate) loans are carred at hstorcal cost, as loans held for sale, that are reported at the lower of hstorcal cost and far value, consttute only a small fracton of less than 1% of total assets for the average bank. 10 Ths classfcaton s to be made on the date of purchase of the securty and t s n prncple rreversble. On the purchase date, amortzed cost and far value should be essentally the same and hence no valuaton advantage can be obtaned by classfyng securtes ether way. 11 A breakdown of far value assets by valuaton technque (level 1 to 3) s n prncple avalable from Schedule HC- Q of the Call report. We do not use ths nformaton n our analyss, because these assets are reported for only one of the three far valuaton technques, makng t dffcult to draw any nference based on a drect comparson of the amount of assets reported n each category. Furthermore, the level 1 to 3 assets are not broken down separately for real-estate related assets, whch are the prmary focus of our study, and the majorty of these assets are valued as level 2 assets (about 90 percent of far value assets n 2008) so that there s not much varaton n far valuaton technque. 10
sgnfcant dscreton. 12 At any rate, at a tme of declnng asset values, one expects far values to be less than amortzed cost. Interestngly, banks report n ther Call report flngs both the amortzed cost and far value of MBS regardless of whether these are held-to-maturty or avalable-for-sale. Thus, for MBS that are carred at amortzed cost we also know the assessed far value, whle for MBS carred at far value we also know the reported amortzed cost. Ths enables us to compute a bank s share of MBS that are held-to-maturty (rather than avalable-for-sale) on a sngle accountng bass. Specfcally, we can compute the share of MBS that s held-to-maturty usng amortzed costs for all MBS. MBS can be broken down nto those that do and do not beneft from some explct or mplct offcal guarantee. Guaranteed MBS are those that are guaranteed or ssued by U.S. government agences such as the Federal Natonal Mortgage Assocaton (FNMA), the Federal Home Loan Mortgage Corporaton (FHLMC), and the Government Natonal Mortgage Assocaton (GNMA), more generally known as Fanne Mae, Fredde Mac, and Gnne Mae, respectvely. 13 Fgure 4 shows that for most of the sample perod the share of non-guaranteed MBS classfed as held-to-maturty exceeded the analogous share of guaranteed securtes. Moreover, durng 2008 the share of non-guaranteed MBS labeled held-to-maturty rose strongly from 8.3% to 11.4%. Durng that year, the share of guaranteed MBS that s held-to-maturty, nstead, fell from 6.2% to 5.7%. Fgure 5 reports the mean rato of far value to amortzed cost as reported by dfferent banks over the sample perod separately for guaranteed and non-guaranteed MBS (regardless of 12 Indeed, work by Kolev (2009), Goh et al. (2009), and Song et al. (2009) shows that market dscounts tend to be greatest for level 3 assets, where banks have most dscreton. 13 Note that these guarantees tend to cover underlyng repayment of nterest and prncple, but not valuaton rsk stemmng from nterest rate changes or mortgage prepayment. 11
whether these securtes are actually classfed as held-to-maturty or avalable-for-sale). Ths rato s farly close to one for guaranteed MBS throughout the sample perod. For non-guaranteed MBS, however, far values relatve to amortzed cost declned from about one n 2001 to 86.6% on average at end-2008. These accountng valuatons would have gven banks an ncentve to reclassfy non-guaranteed MBS as held-to-maturty to the extent possble so as to boost the book value of assets. We ndeed fnd that the fracton of non-guaranteed MBS that s reported as heldto-maturty ncreased durng 2008 (as seen n Fgure 4). Though straghtforward reclassfcatons are not allowed under exstng accountng rules, n the exceptonal case of Ctgroup, regulators have publcly approved a straght reclassfcaton of part of the bank s MBS portfolo. Regulators may have tactly approved reclassfcatons at other banks, and potentally some banks reclassfed ther MBS n volaton of FAS 159 wthout regulatory approval, though we have no drect evdence of such MBS reclassfcaton by banks apart from the exceptonal case approved publcly by regulators. It s especally nterestng to assess whether the valuaton of balance sheet tems mplct n stock prces dffers from book values at a tme of fnancal crss. Therefore, the emphass of the subsequent emprcal work wll be on the year 2008, one year nto the recesson and what s generally consdered the start of the U.S. mortgage default crss (see for example Dell Arcca et al., 2008, and Man and Suf, 2008), when delnquences on mortgage loans ncreased sharply. Summary statstcs for the man varables n 2008 are provded n Table 1. We exclude banks wth Tobn s q exceedng ts 99 th percentle (amountng to a Tobn s q greater than 1.5) as these are not ordnary banks that carry prmarly fnancal assets. The mean rato of real estate loans to assets s 53.3%, whle the mean rato of non-real estate loans to assets s 18.0%. The average rato of MBS (usng amortzed cost to value held-to-maturty securtes and far values 12
for securtes avalable-for-sale) to assets s 9.6% and the rato of non-mbs securtes to assets s 7.3%. MBS can be splt nto MBS held-to-maturty at 0.8% of assets, and MBS avalable-for-sale at 8.9% of assets. MBS that are held-to-maturty can agan be splt nto guaranteed and nonguaranteed securtes equvalent to 0.7% and 0.1% of assets, respectvely. Guaranteed and nonguaranteed MBS that are avalable-for-sale n turn amount to 8.1% and 0.8% of assets. Low valuaton s a dummy varable that equals one n a gven quarter f a bank s q s less than one, and zero otherwse. By the end of 2008, 60% of U.S. banks had a value of q of less than one. Several addtonal asset categores are consdered as well. Tradng s defned as tradng assets relatve to total assets (obtaned from Schedule HC-B of the Call report). Tradng assets, whch nclude some MBS, are carred at far value and held n the bank s tradng book. 14 A detaled splt-up of tradng assets s only avalable for the domestc offces of bank holdng companes and s not reported. On average, tradng assets only amount to a share of 0.5% of assets, because only large banks tend to have such assets. As a bank lablty varable, Deposts s defned as total deposts dvded by total assets, and t amounts to 72.2% of assets on average. Data on deposts are obtaned from Schedule HC- E of the Call report fles. Bank captal, beng the sum of Ter 1 and Ter 2 captal, s composed mostly of Ter 1 captal, amountng to 86.3% of captal on average. Also n Table 1, loan loss provsonng s calculated as loan loss provsons dvded by the book value of all loans. The mean loan loss provsonng rate s 0.8%. Net charge-offs, n turn, s the rato of the dfference between loan charge-offs and loan recoveres to the book value of loans. The mean net loan charge-off rate s 0.5%. Thus, loan loss provsonng exceeded net loan 14 Tradng assets are to be reported only by bank holdng companes wth average tradng assets of $2 mllon or more n any of the four precedng quarters. 13
charge-offs, as expectatons of loan losses surpassed actual loan wrte-offs. Fnally, the share of real estate loans s the rato of real estate loans to total loans wth a mean value of 74.2%. The loan loss provsonng rate ncreased sharply as the crss unfolded from less than 0.1% on average durng the frst quarter of 2007 to 1.5% on average durng the fourth quarter of 2008. 2.4 Estmaton of market dscounts Ths secton provdes the results of regressons of Tobn s q to obtan mplct market valuatons of man balance sheet tems. The focus s on the real estate components of loans and securtes. Ths emphass s justfed by the fact that the real estate components of loans and securtes together comprse on average 63% of bank assets n 2008, and by the fact that real estate assets have suffered from house prce declnes durng the recent fnancal crss. Nevertheless, we nclude several other balance sheet tems n the analyss as well. To start, Table 2 reports regressons of q that nclude the real estate loans and MBS varables wth data for 2008. 15 The regressons also control for exposure to non-real estate loans, tradng assets, and the composton of bank captal. All regressons nclude U.S. state fxed effects and tme fxed effects to control for systematc dfferences across U.S. states and quarterly perods, such as housng and labor market condtons, or the monetary polcy stance. Standard errors are corrected for clusterng at the frm level. The real estate loans varable enters wth a coeffcent of -0.173 that s sgnfcant at the 1% level mplyng that the mplct market dscount of real estate loans relatve to book value s 17.3%. The MBS varable smlarly enters wth a coeffcent of -0.136 that s sgnfcant at the 5% level so that MBS appear to be 15 The estmaton model mplctly sets the dscount on excluded asset categores to zero. Asset categores excluded from the regresson are cash-lke assets (ncludng cash, federal funds sold, and government securtes) amountng to 15% of total assets, and non-cash lke assets (ncludng other securtes, and fxed assets) amountng to the remander of 3% of total assets. Thus, wth cash-lke assets carryng a dscount of close to zero and consttutng the majorty of excluded assets, the mplct assumpton of a dscount of zero on excluded asset categores appears to be reasonable. 14
dscounted 13.6%. 16 The economc effects are substantal. The results mply that a one standard devaton ncrease n real estate loans would reduce Tobn s q by another 2.4 percentage ponts, and that a one standard devaton ncrease n MBS would reduce Tobn s q by another 1.0 percentage ponts. These are substantal effects gven the standard devaton of Tobn s q of 5.5%. In unreported regressons, we splt the MBS varable nto ts guaranteed and nonguaranteed parts yeldng that non-guaranteed MBS are dscounted the most. Tradng, denotng the rato of tradng assets to total assets, enters the regresson wth a coeffcent of -0.285 that s sgnfcant at the 1% level. The economc effect of ths result s small gven that tradng assets on average comprse only 0.5% of total assets n 2008. The composton of equty captal can also nfluence bank value, especally durng 2008 as markets reassessed the superor value of Ter 1 captal relatve to Ter 2 captal, partly n response to strcter captal requrements proposed by regulators. We ndeed fnd that Ter 1, denotng the share of Ter 1 captal n total captal, enters wth a postve coeffcent of 0.103 that s sgnfcant at the 1% level. Ths suggests that a one standard devaton ncrease of 8% n the share of Ter 1 captal n total captal ncreases bank value by 0.8 percentage ponts, whch s not rrelevant gven a standard devaton of q of 5.5%. In unreported regressons, we fnd that the results are robust to ncludng addtonal lablty varables, ncludng the rato of total deposts to total assets and the rato of bank captal to assets. In regresson 2, we replace the MBS varable wth two separate varables, MBS, held and MBS, for sale that represent the parts of MBS that are held-to-maturty (and carred at amortzed 16 We only consder the market valuaton of MBS as mplct n share prces. Emprcal models of the drect prcng of MBS are offered by Dunn and Sngleton (1983), Boudoukh et al. (1997), and Schwartz and Torous (1989). 15
cost) and avalable-for-sale (and carred at far value). To the extent that far values reflect market prces, we expect the dscount to be smaller for avalable-for-sale MBS. The MBS, held varable obtans a coeffcent of -0.219 that s sgnfcant at 1%, whle the MBS, for sale varable enters wth a coeffcent of -0.119 that s sgnfcant at 10%. Thus, MBS classfed as held-tomaturty appear to be dscounted sgnfcantly at 21.9%, whle the MBS avalable-for-sale tend to have a smaller dscount of 11.9% on average. Thus, the gap between mplct market prces and accountng values appears to be largest for MBS classfed as held-to-maturty. Fnally, n the last four regressons n Table 2 we re-estmate regresson 2 separately for each of the four quarters n 2008. As we lose the tme seres dmenson, these regressons do not nclude tme fxed effects. We fnd that the dscounts on the real estate related varables are farly stable over these four quarters. Interestngly, the dscounts on real estate loans and MBS that are held-to-maturty both reach a peak of 22.3% and 22.8%, respectvely, durng the thrd quarter of 2008, though the dfference s not statstcally sgnfcant. We obtan a smlar pattern when we further splt the MBS, held and MBS, for sale varables nto ther guaranteed and non-guaranteed parts (not reported). The evdence ndcates szeable market dscounts on real-estate related assets relatve to book values for U.S. bank holdng companes n 2008. As we have data startng n 2001, we next analyze whether such dscounts exsted before 2008. For ths purpose, we re-estmate regresson 2 of Table 2 wth data for each of the years 2001 to 2007. The results are reported n Table 3. Throughout the perod 2001 to 2007, none of the real estate asset categores (real estate loans and MBS varables) s estmated wth a sgnfcant dscount. Thus, real estate loans and MBS asset categores are estmated wth sgnfcant dscounts not untl 2008, suggestng that the 16
deteroraton of the mplct market value of real estate assets relatve to book value was sudden rather than gradual. The fnal regresson n Table 3 uses data for the perod 2001 to 2008 and ncludes nteractons between the explanatory varables and an ndcator varable that takes a value of one for year 2008 observatons to show that the 2008 dscounts reported thus far are sgnfcantly dfferent from those n earler years. We fnd that Tobn s q s sgnfcantly lower on account of exposure to real estate loans are held-to-maturty MBS n 2008 as compared to earler years. Real estate loans obtan a dscount of 14.7% compared to earler years, and held-to-maturty MBS obtan a dscount of 9.0% compared to earler years. Both results are sgnfcant at the 1% level. The dscount of real estate loans s also larger than the dscount on non-real estate loans, consstent wth the expectaton that U.S. house prce declnes affect loan mparment of mortgage loans the most. Avalable-for-sale MBS do not obtan a dscount relatve to earler years, suggestng that wrte-downs of these assets have kept pace wth market perceptons of mparment. One concern s that our results are drven by an overshootng n asset prces, meanng a temporary devaton n value from fundamental value. However, our measure of frm value s based on equty prces, whch reflect the consensus vew of many fnancal market partcpants. Whle fre sales and llqudty may have led to overshootng n some asset markets, notably the market for dervatves on mortgage-backed securtes, stock markets contnued to be lqud throughout 2008. Also, we do not obtan a sgnfcant dscount on avalable-for-sale MBS that are reported at far value n 2008 compared to earler years, suggestng that wrte-downs of these assets have kept pace wth market perceptons of mparment. Moreover, banks market-to-book values contnued to declne n the fourth quarter of 2008 and we stll fnd sgnfcant mplct 17
dscounts on real estate related assets n ths quarter, followng the massve government nterventons n October 2008. Laux and Leuz (2009) explan that these nterventons should have reduced the lkelhood of dstressed sales of banks assets nto llqud markets. Fnally, emprcal evdence on the prcng of nvestment-grade credt rsk durng the fnancal crss provded by Coval et al. (2009) casts doubt on the premse that prces n credt markets were systematcally dstorted. We therefore assert that stock market prces offer the best avalable nformaton on the value of banks, and conclude that the accountng values of real estate related assets on the books of banks were nflated n 2008. 3. Accountng dscreton regardng asset mparment and asset classfcaton Our evdence on dscounts on real estate related assets mplct n bank stock prces suggests that the accountng for the mparment of these assets has not kept up wth market perceptons of ths mparment. In ths secton, we show that ths does not merely reflect a passve and rresponsve accountng system, but t also results from the actve use by banks of accountng dscreton to prevent declnes n the book value of assets. We consder accountng dscreton n the areas of accountng for loan mparment and MBS classfcaton n turn. 3.1 Accountng dscreton regardng bad loans Loan loss provsonng n prncple should mrror expected future loan losses, but n practce banks have consderable dscreton n settng loan loss provsonng rates. In ths secton, we examne whether dstressed banks have systematcally held back on ther loan loss provsonng durng the present fnancal crss n order to preserve bank captal. 17 We focus on accountng for the mparment of loans as the sze of the average bank s loan portfolo, at 71% of 17 No breakdown of loss provsonng for real estate loans and other loan categores s avalable from the Call report. 18
assets, renders bank captal very senstve to loan performance. 18 In addton, we consder whether dstressed banks have dfferent loan loss realzatons n the form of loan charge-offs. Loan charge-offs reduce loan loss allowances, wth no mmedate mplcatons for regulatory captal, but present charge-offs may trgger a need for future loan loss provsonng whch wll reduce regulatory captal. To capture loan loss provsonng, we construct the rato of loan loss provsons to total loans, expressed n percentages. We obtan data on loan charge-offs and provsons from Schedule HI-B of the Call report fles. To proxy for potental bank dstress, we use nformaton on banks MBS exposure and Tobn s q. We frst report regressons that test whether loan loss provsonng behavor of banks changed sgnfcantly durng the crss year 2008 when asset prces were depressed. The regressons are based on the entre sample perod and nclude tme fxed effects and frm fxed effects to control for tme-nvarant bank characterstcs. All explanatory varables are lagged one quarterly perod. In regresson 1 of Table 4, the loan loss provsonng varable s related to the share of real estate loans n total loans and the share of MBS n total assets. We also nclude nteracton terms between these two real estate exposure varables and a dummy varable that denotes whether the observaton s for the year 2008 or not. We expect loan loss provsonng to be more postvely related to the share of real estate loans n 2008, as these loans have been partcularly affected by recent house prce declnes. We ndeed fnd that loan loss provsonng rate of banks wth a large share of real estate loans was sgnfcantly hgher durng the year 2008. However, there s no sgnfcant relatonshp between 18 Laux and Leuz (2009) report that loans accounted for roughly three quarters of the balance sheets for the 31 bank holdng companes that faled and were sezed by U.S. bank regulators between January 2007 and July 2009. 19
the share of real estate loans and loan loss provsonng n earler years, possbly because banks were antcpatng contnued apprecaton of real estate prces. Banks that need to absorb large losses arsng from exposure to MBS may lower ther provsonng standards n an effort to preserve regulatory captal. In lne wth ths, we fnd that the loan loss provsonng rate of banks wth large MBS exposure was sgnfcantly lower durng the year 2008, suggestng that banks wth potental large losses on MBS were holdng back on ther loan loss provsonng n that year. The MBS exposure varable obtans a negatve coeffcent that s -1.738 lower n 2008 than n earler years, and t s statstcally sgnfcant at the 1% level. The economc effect of ths result s substantal. A one standard devaton ncrease n MBS mples a reducton n the loan loss provsonng rate n 2008 compared to earler years of 0.13 percentage ponts. Ths s a substantal effect compared to the standard devaton of the loan loss provsonng rate of 1.0%. For a bank wth the mean loan to assets rato of 0.71, the mpled reducton n loan loss provsonng s equvalent to 0.09 percent of assets. The average bank had a Ter 1 captal rato of 10.8% n 2008, and hence a captal savng of 0.09 percent of assets amounts to 0.8 percent of Ter 1 captal Next, we analyze whether loan loss provsonng behavor has been sgnfcantly dfferent for low valuaton banks. The ncentve to hold back on loan loss provsonng should be partcularly pronounced for banks wth dstressed market values. To test ths, regressons 3 and 4 n Table 4 nclude nteracton terms between the real estate exposure varables and a dummy varable that takes a value of one for banks wth Tobn s q value of less than one. Regresson 3 reports a coeffcent for the MBS varable that s more negatve at -0.986 and statstcally sgnfcant at 1% for low-valuaton banks, whle regresson 4 shows that the result s partcularly 20
pronounced for held-to-maturty MBS. Thus, we fnd that the negatve relatonshp between loan loss provsonng and a bank s MBS exposure s sgnfcantly stronger for banks wth a low valuaton, as measured by a value of Tobn s q of less than one. Ths suggests that dstressed banks wth large exposure to MBS were partcularly holdng back on ther loan provsonng. Dstressed banks also may be slow n recognzng losses on ther real estate loan portfolo n the form of wrte-downs or charge-offs. 19 To analyze ths, regressons 5 to 8 take as dependent varable the rato of net charge-offs to loans (where net charge-offs are the dfference between charge-offs and recoveres) expressed n percentages. Otherwse, these regressons are smlar to regressons 1 to 4. Consstent wth the earler results, we fnd n regresson 5 that the rato of net charge-offs to loans s negatvely and sgnfcantly related to the MBS varable n 2008 compared to earler years. The results of regressons 6 to 8 smlarly are analogous to those n regressons 2 to 4. In sum, we fnd evdence that low-valuaton banks wth large MBS exposures hold back on ther loan loss provsonng and charge-offs at a tme of fnancal crss. 3.2 Classfcaton of mortgage-backed securtes Accordng to FAS 159, banks have the opton to classfy securtes as held-to-maturty or avalable-for-sale. Securtes are to be classfed as held-to-maturty and carred at amortzed cost, f management has the ntenton to hold them untl maturty. Otherwse, securtes are avalable-for-sale and carred at far value. Ths classfcaton s to be made on the date of purchase of the securty and t s n prncple rreversble. On the purchase date, amortzed cost and far value should be essentally the same and hence no valuaton advantage can be obtaned by classfyng securtes ether way. 19 Loan wrte-downs nclude wrte-downs arsng from transfers of loans to a held-for-sale account. 21
Reclassfcaton of prevously acqured securtes generally does affect the overall book value of securtes. Specfcally, overall book value rses f avalable-for-sale securtes are reclassfed as held-to-maturty at a tme when amortzed cost exceeds far value. In 2008, the mean rato of far value to amortzed cost for non-guaranteed MBS was 0.925, aganst a mean rato of far value to amortzed cost for guaranteed MBS of 1.005 (see Fgure 5). These accountng valuatons gave banks an ncentve to reclassfy non-guaranteed MBS as held-tomaturty to the extent possble so as to boost the book value of assets. Indeed, the fracton of non-guaranteed MBS that s held-to-maturty ncreased from 7.6% at end-2007 to 11.4% at end- 2008, consstent wth the noton that banks had ncentves durng the year 2008 to classfy a larger fracton of ther MBS as held-to-maturty (see Fgure 4). Bank can acheve some reclassfcaton of prevously acqured securtes n complance wth FAS 159 by sellng and buyng equvalent securtes that are categorzed dfferently wthn the same reportng perod. In the exceptonal case of Ctgroup, regulators have publcly approved a straght reclassfcaton of part of the bank s MBS portfolo. Regulators may have tactly approved reclassfcatons at other banks, and potentally some banks reclassfed ther MBS n volaton of FAS 159 wthout regulatory approval. We have no drect evdence of MBS reclassfcaton by banks, apart from the exceptonal case approved publcly by regulators. However, we can examne whether changes n MBS classfcatons over tme are consstent wth a reclassfcaton ncentve. In ths secton, we fnd evdence that banks wth large MBS exposure classfed a larger share of ther non-guaranteed MBS as held-to-maturty as the fnancal crss developed over our sample perod of 2001-2008. Ths evdence s consstent wth a reclassfcaton ncentve. 22
Table 5 reports regressons of the shares of MBS that are held-to-maturty. In the calculaton of these shares, the MBS that are actually avalable-for-sale are also valued at amortzed cost. Banks may have an ncentve to reclassfy MBS both when the exposure to MBS s large and when the prce dfferental between MBS valued at amortzed cost or far value s large. To capture both these quantty and prce effects, we nclude as explanatory varables MBS, amortzed, whch s the fracton of MBS at amortzed value to total assets and MBS, amortzed mnus far value, whch s the dfference between MBS at amortzed cost and MBS at far value scaled by MBS at amortzed cost. To assess whether classfcaton of MBS changed durng 2008, we nclude nteractons between each explanatory varable and a dummy varable that takes a value of one for the year 2008 and zero otherwse. Regressons are based on the complete sample of banks and nclude tme and frm fxed effects. Regresson 1 relates the share of MBS that s held-to-maturty to contemporaneous rghthand-sde varables, whle n regressons 2 the explanatory varables are lagged one quarterly perod. In both regressons, we fnd negatve and sgnfcant relatonshps between the share of MBS that s held-to-maturty and the two loan and the MBS, amortzed varables. Apparently, banks that nvest more n relatvely rsky assets,.e., loans and MBS, make more use of far value accountng, perhaps reflectng ncreased usefulness of far value accountng n the face of bank asset rsk. In regresson 2, the MBS, amortzed far value varable s nsgnfcant, suggestng that there was no pressure to classfy MBS so as to boost asset values n the pre-2008 perod. Both regressons further suggest sgnfcantly more postve relatonshps between the share of MBS that s held-to-maturty and the real estate related varables (real estate loans, MBS, amortzed, and MBS, amortzed mnus far value) n 2008 than n earler years, wth the excepton that the MBS, amortzed varable nteracted wth the 2008 fxed effect s not sgnfcant n regresson 2 23
The mpled effects are economcally sgnfcant. Specfcally, regresson 2 ndcates that a one standard devaton ncrease n the MBS, amortzed mnus far value varable mples an ncrease n the share of MBS that s held-to-maturty of 1.3 percentage ponts. Smlarly, a one standard devaton ncrease n the Real estate loans varable mples an ncrease n the share of MBS that s held-to-maturty of 1.6 percentage ponts. These are substantal effects compared to the standard devaton of the share of MBS that s held-to-maturty varable of 18.7%. In sum, we fnd evdence that banks wth large real estate related exposure tend to report a relatvely large share of MBS as held-to-maturty n 2008. The changed classfcaton behavor n 2008 relatve to earler years appears to reflect the applcaton of accountng dscreton n the face of a severe fnancal crss. 4. Conclusons In 2008, the majorty of U.S. banks were zombe banks as evdenced by market values of bank assets beng lower than ther book values. Ths s prma face evdence that the book value of banks balance sheets s nflated. Consstent wth ths, we fnd that the market dscounts the value of banks real estate loans and MBS relatve to book values. The apparent dscrepancy between market and book values of bank assets suggests that banks have been slow to adjust the asset book values to conform to market expectatons about future declnes n asset performance. We fnd a larger dscount for held-to-maturty MBS (that are carred at amortzed cost) than for avalable-for-sale MBS (that are carred at far value), suggestng that far values recognze the mparment of MBS to a greater extent and more quckly than amortzed costs do. 24
The slowness of book values to reflect market declnes does not merely reflect the rgdty of accountng rules, but t n part reflects the actve use by banks of accountng dscreton to prevent book value deteroraton. Specfcally, we demonstrate that banks wth large exposures to MBS systematcally report relatvely low loan loss provsonng rates n 2008 so as to nflate asset values and book captal. At the same tme, these banks tend to classfy a relatvely large share of ther MBS as held-to-maturty n 2008, to be able to carry these assets at amortzed costs. Our fndng that dstressed banks tend to explot ther dscreton n loan loss provsonng and classfcaton of MBS to boost ther accountng value should be reason for concern, as t mples that the dscreton mplct n current accountng rules leads to systematc bases n valuatons on bank balance sheets. Accountng dscreton enables banks wth mpared asset portfolos to satsfy captal adequacy requrements, but t makes t dffcult to assess the true health of the affected banks. In settng loan loss provsons, banks no doubt make use of prvate nformaton about the prospects of loan repayment. Ths makes bank dscreton regardng loan loss provsonng generally benefcal n that prvate nformaton about loan qualty s revealed, and dscreton regardng loan loss provsonng may well be unavodable. Bank classfcaton of MBS, n contrast, does not serve the purpose of revealng prvate nformaton about MBS qualty. Banks can use such nformaton, to the extent that they have t, n determnng amortzed cost and far value of ther MBS assets. Classfcaton behavor, however, carres the cost of enablng banks to alter the overall book value of ther assets and ther regulatory captal, wthout the beneft of revealng any nformaton about MBS qualty not already contaned n accountng values. Ths may underle the statutory rule aganst securty reclassfcaton contaned n FAS 159. Our 25