Contents lists available at SciVerse ScienceDirect

Journal of Financial Economics

Journal of Financial Economics 106 (2012) 614–634

0304-40

http://d

$ We

Adrian,

Jacob G

Laux, C

Masulis

Douglas

Wolf W

Bank of

nationa

Booth S

the Uni

of the E

and Fin

Landon

and con

They shn Corr

E-m

journal homepage: www.elsevier.com/locate/jfec

Bank valuation and accounting discretion during a financial crisis$

Harry Huizinga a,b,n, Luc Laeven b,c

a Tilburg University, Netherlandsb CEPR, United Kingdomc International Monetary Fund, United States

a r t i c l e i n f o

Article history:

Received 23 March 2010

Received in revised form

28 September 2011

Accepted 6 December 2011Available online 28 June 2012

Keywords:

Managerial discretion

Regulatory forbearance

Accounting

Banking

Financial crisis

5X/$ - see front matter & 2012 Elsevier B.V.

x.doi.org/10.1016/j.jfineco.2012.06.008

thank Bill Schwert (the Editor), an anonym

Paolo Angelini, Alexander Bleck, Stijn Clae

oldfield, Peter Howitt, Rocco Huang, Edwa

hristian Leuz, Ross Levine, Tom Linsmeier

, Marco Pagano, Thomas Philippon, Lev Rat

Skinner, Hans Stoll, Rene Stulz, Kenneth Sul

agner, Sweder van Wijnbergen, and seminar

Canada, the Federal Deposit Insurance Corp

l Monetary Fund, Brown University, the Un

chool of Business, Frankfurt University, Geor

versity of Amsterdam, Vanderbilt University, t

CB-CFS Network, and the CEPR-EBC Conferen

ancial Regulation for comments or sugge

i for excellent research assistance. The findin

clusions expressed in this paper are entirely t

ould not be attributed to the International M

esponding author at: Tilburg University, Net

ail address: h.p.huizinga@uvt.nl (H. Huizinga)

a b s t r a c t

This paper shows that banks overstate the value of distressed assets and their

regulatory capital during the US mortgage crisis. Real estate-related assets are over-

valued in banks’ balance sheets, especially those of bigger banks, compared to the

market value of these assets. Banks with large exposure to mortgage-backed securities

also provision less for bad loans. Furthermore, distressed banks use discretion over the

classification of mortgage-backed securities to inflate their books. Our results indicate

that banks’ balance sheets offer a distorted view of the financial health of the banks and

provide suggestive evidence of regulatory forbearance and noncompliance with

accounting rules.

& 2012 Elsevier B.V. All rights reserved.

1. Introduction

Distressed asset markets provide financial firms withincentives to use managerial discretion over financialreporting to increase earnings and preserve book value.Bank regulators can, in principle, impose regulatory

All rights reserved.

ous referee, Tobias

ssens, Falko Fecht,

rd Kane, Christian

, Joe Mason, Ron

novski, Amit Seru,

livan, Harald Uhlig,

participants at the

oration, the Inter-

iversity of Chicago

getown University,

he 12th Conference

ce on Procyclicality

stions, and Mattia

gs, interpretations,

hose of the authors.

onetary Fund.

herlands.

.

discipline on banks to adjust their capital downward.However, during a financial crisis when bank distress iswidespread, regulatory forbearance is often applied toavoid disruptions from bank failures to the real economyand the financial system. As a consequence, discretionover accounting rules combined with regulatory forbear-ance leads banks to understate balance sheet stresses andto overstate regulatory capital.

During the recent financial crisis, large differenceshave arisen between market and book values of the assetsof US banks. By end-2008, 60% of US bank holdingcompanies had a market-to-book ratio of assets belowone, compared to only 8% of banks at the end of 2001.During this period, the market values of some bank assets,such as mortgage-backed securities (MBS), declined shar-ply, in part due to information asymmetries about thequality of these assets (Gorton, 2009; Diamond and Rajan,2011). The average ratio of Tier 1 capital to bank assets,however, declined only slightly from 12% to 11% over thisperiod. The market value of bank equity thus has droppedprecipitously against a backdrop of virtually constantbook capital, suggesting that declines in reported bankcapital understate the deterioration of bank assets.

H. Huizinga, L. Laeven / Journal of Financial Economics 106 (2012) 614–634 615

We show that banks have systematically understatedthe impact of the impairment of real estate-related assetson bank capital in their financial accounts since the onsetof the recent financial crisis. Their financial accountsoverstate the value of real estate related-assets as com-pared to stock market valuations, understate loan impair-ment, and are based on valuation techniques that yieldrelatively high asset values at a time of declining marketvalues. We take this as evidence of systematic use ofaccounting discretion by banks.

Some discretion over financial reporting is unavoid-able, as financial reporting systems in part are mechan-isms for firms to reveal asymmetric information toinvestors and other outside parties. This is particularlyrelevant for banks that collect valuable private informa-tion on their bank–borrower relationships (James, 1987;Gonzalez and James, 2007; Hale and Santos, 2009).Managers generally have incentives to overstate assetsand smooth earnings even in normal times (Collins et al.,1995; Leuz et al., 2003; Hutton et al., 2009), althoughenhanced corporate disclosure and transparency is gen-erally considered to boost corporate valuation (Karpoffet al., 2008). Banks are also subject to minimum capitalrequirements and, therefore, have an additional incentiveto overstate asset values in downturns to maintainregulatory capital requirements.

Misreporting will be limited if regulators enforce strictcompliance with accounting rules. During financial crisesregulators may fail to do this for two reasons. First,regulators could find it hard to act on individual bankswhen there is only evidence of general misreporting bybanks (as provided in this paper). Second, regulators couldtemporarily allow noncompliance of accounting rules toprevent failures of systemically important banks, thuseffectively applying regulatory forbearance, with concomi-tant risks for tax payers. The lingering uncertainty aboutasset values may serve regulators well by casting doubtabout the degree to which they apply forbearance. There isample evidence of regulatory forbearance in previouscrises in the US and elsewhere (Kane, 1989; Kroszner andStrahan, 1996; Caballero et al., 2008; Skinner, 2008; Brownand Dinc, 2011). Moreover, when banks fail, large unac-counted losses are often revealed, resulting in low recoveryvalues of failed banks (James, 1991).

We offer three pieces of evidence that banks overstatetheir asset values and regulatory capital in the currentfinancial crisis. First, we estimate large market discountsimplicit in stock prices on real estate-related assets suchas mortgage loans and MBS, and find that these discountsare more pronounced for bigger banks. To estimateimplicit market discounts on bank assets, we empiricallyrelate the market-to-book value of overall bank assets tospecific bank asset exposures using quarterly accountingdata on US bank holding companies for the period 2001–2008.

We find significant discounts on banks’ real estateloans starting in 2008, averaging about 17%. As theaverage bank in 2008 holds about 53% of its assets inthe form of real estate loans, the implicit discount in loanvalues goes a long way toward explaining the depressedstate of bank share prices in 2008. We further find that

investors started discounting banks’ holdings of MBS in2008. For that year, we find an average discount on theseassets of 14%, while the average MBS exposure amountedto 10% of assets. These discounts are primarily present forprivate label MBS, which is not surprising given that theyare not implicitly backed by government guarantees andtherefore carry higher default risk. These sizeable dis-counts suggest that banks have used discretion overdetermining the book values of real estate loans andMBS to limit reported impairment of these assets. Whenwe split the sample into large and small banks, we findthat increases in estimated discounts on real estate loansand MBS in 2008 are larger for bigger banks. This resultsuggests that regulators allow large banks more discre-tion over asset valuation as part of regulatory forbearanceof banks that are considered too big to fail.

Second, we consider bank behavior regarding theirloan loss provisioning and loan charge-offs in the currentfinancial crisis to assess whether banks preserve regula-tory capital by holding back on loan loss provisioning.We find that banks with large exposure to MBS reportedsignificantly lower loan loss provisions in 2008. Thissuggests that weakened banks manipulated their loanloss provisioning to manage their regulatory capitalduring the recent crisis.

Third, we examine banks’ choices regarding the classi-fication of the MBS on their books to analyze whetherbanks take advantage of valuation differences betweendifferent accounting methods. In principle, banks canaugment the book value of their assets by reclassifyingavailable-for-sale MBS as held-to-maturity when the fairvalue of these MBS is less than their amortized cost.Indeed, we find that the share of nonguaranteed MBSthat are held-to-maturity increased substantially in 2008,when the fair value of these MBS tended to be less thantheir amortized cost. At the same time, banks with highexposures to real estate loans and with large gapsbetween the amortized cost and the fair value of theirMBS report larger exposures to held-to-maturity MBS in2008. These findings suggest that banks use discretionover MBS classification.

Taken together, the evidence of this paper provides acomprehensive picture of banks’ accounting discretionduring a financial crisis. Our finding that real estate-related assets, and in particular MBS, are discounted bythe market relative to book valuation in 2008 is consistentwith bank discretion over book valuation, even if it couldalternatively reflect that banks with high MBS experi-enced some bad luck in the choice of their assets resultingin lower stock market valuation. Our evidence on discre-tion over accounting for loan impairment and MBS classi-fication, however, suggests that the estimated discountson real estate-related assets in 2008 at least in part reflectaccounting discretion. The discount results thereforeprobably reflect that regulators, while being aware ofthe deterioration in asset values at banks, allowed espe-cially large banks to use accounting discretion to over-state the value of the assets in their books at a time offinancial crisis. Overall, discretion over financial reportingenables banks to soften the impact of the crisis on thebook valuation of assets and regulatory capital.

H. Huizinga, L. Laeven / Journal of Financial Economics 106 (2012) 614–634616

Such discretion delivers highly inaccurate financialinformation, especially during financial crises when assetsbecome distressed, with potential real consequences forthe allocation of capital in the economy (Peek andRosengren, 2000; Calomiris and Mason, 2003; Kedia andPhilippon, 2009). Financial misreporting potentially alsoimpedes market discipline by bank debt holders, althoughmarket discipline of banks is generally found to beeffective (Flannery and Sorescu, 1996; Calomiris andWilson, 2004). In 2009, the US Treasury conducted aseries of stress tests of major US banks that revealedcapital shortfalls at several banks, even though they weremeeting minimum capital requirements. This is testimonyto these banks’ financial reports providing a misleadingpicture of the health of the concerned banks.

Our paper relates to an emerging literature on thecauses and effects of the recent financial crisis. This workshows that house price appreciation (Demyanyk and VanHemert, 2011), asset securitization (Keys et al., 2010;Mian and Sufi, 2009; Loutskina and Strahan, 2009), anda deterioration in lending standards (Dell’Ariccia et al.,forthcoming) helped fuel a crisis in US mortgage markets,with bank capital being eroded as the real estate bubbleburst. Goh et al. (2009) study the market pricing of bankassets and find that securities that are valued withrelatively much discretion were discounted the most. Thisfinding is consistent with our evidence that the marketattaches discounts to real estate-related assets that arevalued with considerable discretion. Recent work alsoinvestigates concerns about the potential procyclical nat-ure of fair value accounting, which could magnify fluctua-tions in bank lending and economic activity. Laux andLeuz (2010) find little evidence that such effects are theresult of fair value accounting.

We are not the first to show that banks use discre-tionary accounting to manage earnings and capital, espe-cially during distress times. Earlier work has largelyfocused on the role of loan loss provisioning (Moyer,1990; Ahmed et al., 1999; Laeven and Majnoni, 2003;Shrieves and Dahl, 2003) or deferred taxes (Skinner, 2008)in managing earnings and capital.

Our paper extends the literature on accounting discre-tion by banks in three main ways. First, we provide a newtest of discretion regarding loan loss provisioning byshowing that such provisioning is lower at banks thatheld considerable MBS in 2008. MBS were not an impor-tant asset category during previous banking crises in theUS. Secondly, we estimate market discounts applied toMBS that are partly reported at historical cost and partlyat fair value. A key result of this paper is that the marketeven applied discounts to MBS that were reported at fairvalue. Thirdly, we examine whether banks reclassify theirMBS using more favorable valuation methods so as toboost book valuation. We are able to do this as banks arerequired to report the historical cost as well as fair valueof their MBS regardless of whether they are actually heldat historical cost or fair value. This enables us to computethe share of MBS held either way using the same account-ing convention for all MBS.

The paper continues as follows. Section 2 outlines ourmain hypotheses and the empirical tests that follow.

Section 3 describes the data used in the empirical analy-sis. Section 4 presents empirical evidence on marketdiscounts of real estate-related assets, relative to bookvalues, during the present financial crisis. Section 5examines the use of bank discretion regarding loan lossprovisioning and loan charge-offs, and the classification ofMBS into different valuation categories. Section 6 pro-vides evidence of market discounts across fair valueaccounting techniques and of banks’ stock price reactionto amendments of fair value accounting rules. Section 7concludes.

2. Empirical methodology

2.1. Testing for accounting discretion at banks

Banks generally have an incentive to use discretionover accounting rules to understate balance sheet stressesand to overstate bank capital. This incentive is particu-larly strong at times of financial crisis when balancesheets are under pressure. Moreover, during financialcrises, when bank distress is widespread, regulators gen-erally apply regulatory forbearance to avoid disruptionsfrom bank failures to the real economy and the financialsystem, further enhancing banks’ ability to overstateregulatory bank capital. In this paper, we use the 2008US financial crisis as a shock to the value of bank assets toprovide evidence of the importance of accounting discre-tion at banks. We perform three types of empiricalanalysis to test whether banks overstated their assetvalues and regulatory capital in the recent financial crisis.

First, we estimate market discounts implicit in stockprices on real estate-related assets such as mortgage loansand MBS. Our primary focus is on real estate-relatedassets because these assets constitute a large fraction ofthe total assets of the average bank, and because recentdeclines in US real estate prices have raised doubts aboutthe underlying value of these assets. If these real estateassets command discounts, this would be consistent withbanks having used discretion in determining the bookvalues of real estate loans and MBS to limit reportedimpairment of these assets, although such findings arealso consistent with other explanations such as bankswith large exposures to real estate assets having had badluck. We therefore complement our analysis of marketdiscounts with two additional analyses to provide furtherevidence of banks’ use of accounting discretion.

Second, we analyze bank behavior regarding their loanloss provisioning and loan charge-offs in the current finan-cial crisis. Banks can preserve regulatory capital by holdingback on loan loss provisioning. They have considerableleeway in setting the loan loss provisioning for bad loansand in realizing loan losses in the form of charge-offs,potentially providing room for managing their regulatorycapital. We focus on discretion over the disclosure of loanimpairment, as loans are by far the largest asset category,amounting to 71% of bank assets, on average, in 2008. Wetest whether banks with larger exposures to MBS—andtherefore larger anticipated losses—reported significantlylower loan loss provisions in 2008. This would suggest that

H. Huizinga, L. Laeven / Journal of Financial Economics 106 (2012) 614–634 617

weakened banks manipulated their loan loss provisioning tomanage their regulatory capital during the crisis.

Third, we examine banks’ choices regarding the classi-fication of MBS as either held-to-maturity or available-for-sale. In 2008, the fair value of especially nonguaran-teed MBS tended to be less than their amortized cost. Thisimplies that banks could augment the book value of assetsby classifying these MBS as held-to-maturity. We there-fore examine whether the share of nonguaranteed MBSthat are held-to-maturity increased significantly in 2008.Additionally, the incentive to classify MBS as held-to-maturity should increase in the difference between theamortized cost and fair value of a bank’s MBS portfolio.We therefore also analyze whether the share of MBS intotal assets kept as held-to-maturity is significantlyrelated to the gap between the amortized cost and fairvalue of MBS exposure, and whether this relation betweenthe reporting of MBS as held-to-maturity and the corre-sponding valuation gain is stronger at a time of financialcrisis in 2008 and for distressed banks with lowvaluations.

Taken together, these three pieces of empirical analy-sis allow us to analyze the degree of discretion at banksover setting the book values of assets higher than valuesimplicit in stock prices and over limiting asset impair-ment. We now describe in more detail the methodology ofeach of these three tests.

2.2. Relation between Tobin’s q and market discounts

We empirically relate the market-to-book value ofassets (or Tobin’s q) to banks’ asset exposures to provideestimates of market discounts on main bank balancesheet items such as real estate loans and MBS. Lowervalues of q for banks suggest that the valuation of bankassets implicit in the stock market has declined more thanthe book values of these assets. By relating q to banks’exposure to various asset classes, we can infer marketdiscounts on these assets relative to their book values.1 Toimplement this, let MV be the market value of the bank.Further, let Ai be the book value of asset i and let Li be thebook value of liability i. If we assume there are operatingmarkets for a bank’s assets and liabilities, then we canstate a bank’s market value as follows:

MV ¼X

i

vai Ai�

X

i

vliLi, ð1Þ

where vai is the market value of asset i and vl

i is the marketvalue of liability i.

We can now define q as the market value of the equityof the bank plus the book value of all liabilities divided by

1 In similar fashion, Sachs and Huizinga (1987) estimate discounts

on third world debt on the books of US banks at the time of the

international debt crisis of the 1980s. A related literature, starting with

Lang and Stulz (1994) and including Laeven and Levine (2007), has

studied discounts in q arising from corporate diversification. In that

literature, discounts are computed for each business unit of a conglom-

erate with respect to the value of comparable stand-alone firms, while

here, we compute discounts for different assets and liabilities of the

same bank.

the book value of all assets as follows:

q¼MVþ

PiLiP

iAi:

Substituting for MV from Eq. (1) into the expression forq, we get:

q¼ 1�X

i

dai aiþ

X

i

dlili, ð2Þ

where dai ¼ 1�va

i , dli ¼ 1�vl

i, ai ¼ Ai=P

iAi, and li ¼ Li=P

iAi.

The to-be-estimated coefficients dai and dl

i are thediscounts implicit in the bank’s stock price of its assetsand liabilities relative to book values. The variables ai andli are the current book values of particular assets andliabilities relative to the book value of all assets.2 InEq. (2), q potentially differs from one if the stockmarket-implied valuation of at least one balance sheetitem differs from its book value.

To estimate implicit market discounts on bank assets,we empirically relate the market-to-book value of assetsto banks’ asset exposures using quarterly accounting dataon US bank holding companies for the period 2001–2008.It is especially interesting to assess whether the valuationof balance sheet items implicit in stock prices differs frombook values at a time of financial crisis. Therefore, theemphasis of the subsequent empirical work will be on theyear 2008, one year into the recession and what isgenerally considered the start of the US mortgage defaultcrisis (e.g., Mian and Sufi, 2009).

2.3. Loan loss provisioning and charge-offs

Loan loss provisioning, in principle, should mirrorexpected future loan losses, but in practice, banks haveconsiderable discretion over setting loan loss provisioningrates. To examine potential discretion, we regress loanloss provisioning rates on measures of bank distress totest whether distressed banks have systematically heldback on their loan loss provisioning during the recentfinancial crisis to preserve bank capital. We focus onaccounting for the impairment of loans as the predomi-nance of loans in the average bank portfolio renders bankcapital very sensitive to loan performance. We also runregressions of loan charge-off rates on measures of bankdistress to analyze whether distressed banks report lowerloan loss realizations. Loan charge-offs reduce loan lossallowances, with no immediate implications for regula-tory capital, but current charge-offs could trigger a needfor future loan loss provisioning that will reduce regula-tory capital. To proxy for potential bank distress, we useinformation on banks’ MBS exposure and q. The idea isthat banks that need to absorb large losses arising fromexposure to MBS could lower their provisioning standardsin an effort to preserve regulatory capital. Similarly, theincentive to hold back on loan loss provisioning should beparticularly pronounced for banks with distressed market

2 Book values of assets could already reflect some recognition of

impairment. Estimated discounts then reflect the difference between the

market perception of any asset impairment and the accounting treat-

ment of this impairment.

H. Huizinga, L. Laeven / Journal of Financial Economics 106 (2012) 614–634618

values. The regressions include time fixed effects and firmfixed effects to control for time-invariant bank character-istics, and all explanatory variables are lagged one quar-terly period to mitigate concerns about reverse causality.

2.4. Asset reclassification

Reclassification of previously acquired securities byaccounting method generally affects the overall bookvalue of securities. Specifically, book value rises if avail-able-for-sale securities are reclassified as held-to-matur-ity at a time when amortized cost exceeds fair value,giving banks an incentive to reclassify securities as held-to-maturity to the extent possible to boost the book valueof assets.

According to Financial Accounting Standard (FAS) 115,banks have the option to classify securities as held-to-maturity or available-for-sale. Securities are to be classi-fied as held-to-maturity and carried at amortized cost, ifmanagement has the intention to hold them until matur-ity. Otherwise, securities are available-for-sale and carriedat fair value. This classification is to be made on the dateof purchase of the security, and it is irreversible inprinciple. On the purchase date, amortized cost and fairvalue should be essentially the same and hence novaluation advantage can be obtained by classifying secu-rities either way.

Banks can achieve some reclassification of previouslyacquired securities in compliance with FAS 115 by sellingand buying equivalent securities that are categorizeddifferently within the same reporting period. Additionally,there are special circumstances under which banks arepermitted to transfer securities out of the held-to-matur-ity category without violating US Generally AcceptedAccounting Standards (GAAP), including most notablywhen there is evidence of a significant deterioration increditworthiness (revealed, for example, by a downgradeof the bank’s credit rating).3 In the exceptional case ofCitigroup, regulators in 2008 publicly approved a straightreclassification of part of the bank’s MBS portfolio despiteFAS 115. Regulators could have tacitly approved reclassi-fications at other banks as well, and some banks may havereclassified their MBS in violation of FAS 115 withoutregulatory approval.

Although we have no direct evidence of MBS reclassi-fication by banks, apart from the exceptional caseapproved publicly by regulators, we can examine whetherchanges in MBS classifications over time are consistentwith a reclassification incentive. To this end, we analyzewhether the share of MBS kept as held-to-maturity issignificantly related to the gap between the amortizedcost and fair value of MBS exposure, especially in 2008.Specifically, we regress the share of held-to-maturity MBSin total assets on a variable, MBS difference, which equalsthe difference between MBS at amortized cost and MBS atfair value scaled by total assets. This variable measures

3 Other specific circumstances that can justify such a transfer relate

to certain changes in tax law, certain business combinations or disposi-

tions, and certain changes in statutory or regulatory capital require-

ments. See paragraphs 25—6 of the US GAAP for further details.

the valuation gain to be achieved if a bank’s entire MBSportfolio is classified as held-to-maturity rather thanavailable-for-sale, and is thus a direct proxy for a reclas-sification incentive. The regression model also includesthe shares of real estate loans and non-real estate loans intotal assets, and a variable, MBS amortized, that equalsthe share of all MBS valued at amortized cost in totalassets. These variables capture a bank’s asset portfolioconsiderations. For instance, a bank with a large realestate loan portfolio could decide to hold fewer MBS. Inaddition, these loan and MBS exposure variables repre-sent pressures to classify MBS as held-to-maturity, atleast during 2008 when banks with large real estaterelated exposures faced mounting losses as house pricescollapsed and amortized cost of MBS exceeded fair value.

To assess whether MBS classification behavior changedduring 2008, we include interactions between each expla-natory variable and a dummy variable that takes a valueof one for the year 2008 and zero otherwise. A change inclassification behavior during 2008 relative to earlieryears would be consistent with banks using their discre-tion over financial reporting to overstate the book value oftheir assets in the face of a severe financial crisis. More-over, the incentive to reclassify MBS as held-to-maturityshould be particularly pronounced for distressed banks.To assess whether bank distress influences the classifica-tion of MBS, we include interactions between each expla-natory variable and a low-valuation dummy variable thattakes a value of one if the q is less than one. Regressionsare based on the complete sample of banks and includetime and firm fixed effects. To mitigate concerns aboutreverse causality, we lag all explanatory variables onequarterly period.

3. Data and summary statistics

This section discusses the bank asset and liabilitycategories that enter the subsequent empirical work onmarket discounts, loan loss provisioning, and asset reclas-sification. Our sample consists of US bank holding com-panies that are stock exchange listed. These companiesreport a range of accounting data to the Federal ReserveSystem by way of the Report on condition and income(Call report). We use quarterly data from these Callreports from the end of 2001 till the end of 2008. Thiscovers a full business cycle as defined by the NationalBureau of Economic Research from the end of the pre-vious recession in November 2001 until the next reces-sion that started in December 2007. Although we use theterms bank holding company and bank interchangeably,strictly speaking, our unit of analysis is a bank holdingcompany.

Taking stock market data from Datastream, we com-pute the market value of a bank’s assets as the marketvalue of common equity plus the book value of preferredequity and liabilities. Tobin’s q is then constructed as theratio of the market value and the book value of a bank’sassets. The data show a substantial divergence betweenmarket and book values of bank assets during the finan-cial crisis of 2008. This divergence is reflected in themarket value of bank assets relative to their book value.

H. Huizinga, L. Laeven / Journal of Financial Economics 106 (2012) 614–634 619

Fig. 1 reports the average q per quarter over the sampleperiod. The average q has fallen from 1.06 in the finalquarter of 2001 to 1.00 in the final quarter of 2008,indicating that over this period the market value of bankassets has declined more than its book value. The declinein average q has been accompanied by an increase in theshare of banks with a q of less than one, which increasedfrom 8.0% at the end of 2001 to 59.6% at the end of 2008(Fig. 1). During most of this period, the share of bankswith a q of less than one has been smaller than in 2001and 2008 reflecting an upswing of the business cycle. Infact, the share of banks with a q of less than one reached alow of 0.3% during the second quarter of 2004.

Against a background of declining equity prices, banks’regulatory capital ratios have remained surprisinglystable throughout the sample period. Fig. 2 shows thedevelopment of the Tier 1 capital ratio and the share of

0.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.94

0.96

0.98

1.00

1.02

1.04

1.06

1.08

1.10

1.12

2001

Q4

2002

Q4

2003

Q4

2004

Q4

2005

Q4

2006

Q4

2007

Q4

2008

Q4

Sha

re o

f ban

ks w

ith q

<1

Tobi

n's q

Quarter

Tobin's q (left-hand side axis(LHS))

Share of banks with q<1 (right-hand side axis (RHS))

Fig. 1. Tobin’s q and share of banks with qo1. Tobin’s q is the ratio of

market value to book value of assets. Share of banks with qo1 is the

fraction of banks with Tobin’s q less than one. Median values across

banks using quarterly data from Call reports and Datastream from the

fourth quarter in 2001 through the fourth quarter in 2008.

0.84

0.85

0.86

0.87

0.88

0.89

0.08

0.09

0.10

0.11

0.12

0.13

2001

Q4

2002

Q4

2003

Q4

2004

Q4

2005

Q4

2006

Q4

2007

Q4

2008

Q4

Rat

io o

f Tie

r 1 c

apita

l to

tota

l cap

ital

Rat

io o

f Tie

r 1 c

apita

l to

tota

l ass

ets

Quarter

Tier 1 capital to total assets (LHS)

Tier 1 capital in total capital (RHS)

Fig. 2. Capitalization and composition of bank regulatory capital. Tier 1

capital to total assets is the ratio of Tier 1 capital to total risk-weighted

assets. Tier 1 capital in total capital is the ratio of Tier 1 capital to total

regulatory capital. Median values across banks using quarterly data from

Call reports from the fourth quarter in 2001 through the fourth quarter

in 2008.

Tier 1 capital in total bank capital. Tier 1 capital (consist-ing primarily of common stock, retained earnings, anddisclosed reserves) represents the core component ofcapital for banks and is regarded by regulators as a keymeasure of a bank’s financial strength. While leverageincreased for some banks, consistent with findings byAdrian and Shin (2010), the average Tier 1 capital ratiodecreased only modestly from 12.0% in 2001 to 11.2% in2008. The composition of capital also changed onlymodestly over the sample period, with the share of Tier1 capital in total capital shrinking from 88.1% in 2001 to86.4% in 2008. This suggests that, although some banksmay have looked for less traditional, non-core sources ofcapital, such as subordinated debt or perpetual stock, toboost capital and increase assets, most banks did so whilealso increasing Tier 1 capital and maintaining excessregulatory capital.

Our main interest is in banks’ exposure to the realestate market, which comes primarily in the form of realestate loans and MBS. To reflect banks’ direct exposure toreal estate loans, we construct the ratio of real estateloans to overall assets. From 2001 to 2008, the averageasset share of real estate loans has increased substantiallyfrom 45.5% to 52.9% (Fig. 3). Thus, about half of theaverage bank’s assets consist of real estate loans by2008. Banks’ exposure to MBS in turn is measured bythe share of MBS in all assets. This MBS share increasedonly slightly from 9.9% in 2001 to 10.3% at the endof 2008.

While there has been a move toward fair valueaccounting of bank assets, most assets, including mort-gage loans held for investment, are still reported based onhistorical cost. The book value of MBS reflects differentaccounting conventions depending on whether thesesecurities are held-to-maturity or available-for-sale. MBSclassified as held-to-maturity are carried at amortizedcost. The amortized cost may be adjusted periodically foramortization and capitalized interest, and could also

0.02

0.04

0.06

0.08

0.10

0.12

0.14

0.42

0.44

0.46

0.48

0.50

0.52

0.54

0.56

2001

Q4

2002

Q4

2003

Q4

2004

Q4

2005

Q4

2006

Q4

2007

Q4

2008

Q4

Mor

tgag

e-ba

cked

sec

uriti

es (f

ract

ion

of to

tal a

sset

s)

Rea

l est

ate

loan

s (fr

actio

n of

tota

l ass

ets)

Quarter

Real estate loans (LHS)

Mortgage-backed securities (RHS)

Fig. 3. Real estate loans and mortgage-backed securities. Real estate

loans is the ratio of real estate loans to total assets. Mortgage-backed

securities is the ratio of MBS to total assets. Securities are valued at

amortized cost if held-to-maturity and at fair value if available-for-sale.

Median values across banks using quarterly data from Call reports from

the fourth quarter in 2001 through the fourth quarter in 2008.

0.05

0.06

0.07

0.08

0.09

0.10

0.11

0.12

2001

Q4

2002

Q4

2003

Q4

2004

Q4

2005

Q4

2006

Q4

2007

Q4

2008

Q4

Frac

tion

of h

eld-

to-m

atur

ity M

BS

Quarter

Guaranteed MBSNonguaranteed MBS

Fig. 4. Share of mortgage-backed securities that is held-to-maturity.

Guaranteed MBS is the fraction of guaranteed MBS that is held-to-

maturity. Nonguaranteed MBS is the fraction of nonguaranteed MBS that

is held-to-maturity. Median values across banks using quarterly data

from Call reports from the fourth quarter in 2001 through the fourth

quarter in 2008.

0.80

0.85

0.90

0.95

1.00

1.05

2001

Q4

2002

Q4

2003

Q4

2004

Q4

2005

Q4

2006

Q4

2007

Q4

2008

Q4

Rat

io o

f fai

r val

ue to

am

ortiz

ed c

ost o

f MB

S

Quarter

Guaranteed MBS

Nonguaranteed MBS

Fig. 5. Fair value of mortgage-backed securities relative to amortized

cost. Guaranteed MBS is the fair value of guaranteed MBS to the

amortized value of guaranteed MBS. Nonguaranteed MBS is the fair

value of nonguaranteed MBS to the amortized value of nonguaranteed

MBS. Median values across banks using quarterly data from Call reports

from the fourth quarter in 2001 through the fourth quarter in 2008.

H. Huizinga, L. Laeven / Journal of Financial Economics 106 (2012) 614–634620

reflect previous loan loss provisioning. However, theseadjustments to amortized cost are likely to be relativelysmall so that amortized cost is relatively close to origina-tion values. MBS classified as available-for-sale, in con-trast, are to be carried on the books at fair value.

Fair value reflects observed market values (of eitherthe underlying asset-level 1 assets, or of comparableasset-level 2 assets) or otherwise the outcome of a bank’sown valuation models (level 3 assets). Assessments of fairvalue can differ across banks because the determinationof fair value in practice leaves banks with significantdiscretion.4 At any rate, at a time of declining asset values,one expects fair values to be less than amortized cost.

Interestingly, banks report in their Call report filingsboth the amortized cost and fair value of MBS regardlessof whether these are held-to-maturity or available-for-sale. Thus, for MBS that are carried at amortized cost, wealso know the assessed fair value, while for MBS carried atfair value, we also know the reported amortized cost. Thisenables us to compute a bank’s share of MBS on a singleaccounting basis. Specifically, we can compute the shareof MBS that is held-to-maturity using amortized costs forall MBS.

We also have information about the fraction of MBS thatis guaranteed or issued by US government agencies such asthe Federal National Mortgage Association (FNMA), theFederal Home Loan Mortgage Corporation (FHLMC), andthe Government National Mortgage Association (GNMA).5

Fig. 4 shows that for most of the sample period the share ofnonguaranteed MBS classified as held-to-maturity exceededthe analogous share of guaranteed securities. Moreover,during 2008, the share of nonguaranteed MBS labeled

4 Goh et al. (2009) show that market discounts tend to be greatest

for level 3 assets, where banks have most discretion.5 Note that these guarantees tend to cover underlying repayment of

interest and principle, but not valuation risk stemming from interest

rate changes or mortgage prepayment.

held-to-maturity rose strongly from 8.3% to 11.4%. Duringthat year, the share of guaranteed MBS that is held-to-maturity, instead, fell from 6.2% to 5.7%.

Fig. 5 reports the mean ratio of fair value to amortizedcost over the sample period separately for guaranteed andnonguaranteed MBS. This ratio is fairly close to one forguaranteed MBS throughout the sample period. For non-guaranteed MBS, however, fair values relative to amor-tized cost declined from about one in 2001 to 86.6%, onaverage, at end-2008. These accounting valuations wouldhave given banks an incentive to reclassify nonguaranteedMBS as held-to-maturity to the extent possible so as toboost the book value of assets. We indeed find that thefraction of nonguaranteed MBS that is reported as held-to-maturity increased during 2008 (as seen in Fig. 4).

Summary statistics of the main variables in 2008 areprovided in Table 1. We exclude banks with q exceedingits 99th percentile (equal to q greater than 1.5) as theseare not ordinary banks that carry primarily financialassets. The mean ratio of real estate loans to assets is53.3%, while the mean ratio of non-real estate loans toassets is 18.0%, and the share of real estate loans in totalloans amounts to 74.2%, on average. The average ratio ofMBS (using amortized cost to value held-to-maturitysecurities and fair values for securities available-for-sale)to assets is 9.7% and the ratio of non-MBS securities toassets is 7.3%. MBS can be split into MBS held-to-maturityat 0.8% of assets, and MBS available-for-sale at 8.9% ofassets. Held-to-maturity MBS can again be split intoguaranteed and nonguaranteed securities equivalent to0.7% and 0.1% of assets, respectively. Guaranteed andnonguaranteed MBS that are available-for-sale amountto 8.1% and 0.8% of assets. Low valuation is a dummyvariable that equals one in a given quarter if a bank’s q isless than one, and zero otherwise. By the end of 2008, 60%of US banks had a value of q of less than one. Trading isdefined as trading assets relative to total assets. Trading

Table 1Summary statistics of the main regression variables for 2008.

This table reports the summary statistics of the main regression variables for a sample of US bank holding companies that are stock-listed for the year

2008 based on quarterly data. We refer to the Appendix for variable definitions and data sources.

Variable Obs. Mean Std. dev. Min. Max.

Tobin’s q 1,114 1.0138 0.0552 0.8976 1.3280

Real estate loans 1,114 0.5328 0.1361 0.0561 0.8638

Non-real estate loans 1,114 0.1800 0.0941 0.0000 0.6116

MBS 1,114 0.0965 0.0729 0.0000 0.4664

Non-MBS securities 1,114 0.0729 0.0591 0.0000 0.4367

MBS, held 1,114 0.0080 0.0316 0.0000 0.3594

MBS, for sale 1,114 0.0885 0.0666 0.0000 0.4009

MBS, held, guaranteed 1,114 0.0068 0.0284 0.0000 0.3577

MBS, held, nonguaranteed 1,114 0.0012 0.0114 0.0000 0.2006

MBS, for sale, guaranteed 1,114 0.0810 0.0623 0.0000 0.4009

MBS, for sale, nonguaranteed 1,114 0.0075 0.0165 0.0000 0.1592

Low valuation 1,114 0.4560 0.4983 0.0000 1.0000

Trading 1,114 0.0052 0.0255 0.0000 0.2996

Tier 1 1,114 0.1076 0.0254 0.0000 0.2086

Share of Tier 1 1,114 0.8634 0.0810 0.5000 1.0000

Loan loss provisioning 1,114 0.7541 1.0431 0.0000 14.3493

Net charge-offs 1,114 0.4919 0.7583 �0.0936 9.0577

Share of real estate loans 1,114 0.7421 0.1412 0.0993 1.0000

MBS amortized 1,114 0.0967 0.0729 0.0000 0.4661

MBS difference 1,083 0.0003 0.0028 �0.0091 0.0509

6 The estimation model implicitly sets the discount on excluded

asset categories to zero. Asset categories excluded from the regression

are cash-like assets (including cash, federal funds sold, and government

securities), amounting to 15% of total assets, and non-cash-like assets

(including fixed assets), amounting to the remainder of 3% of total

assets. Thus, with cash-like assets carrying a discount of close to zero

and constituting the majority of excluded assets, the implicit assump-

tion of a discount of zero on excluded asset categories appears to be

reasonable.7 We only consider the market valuation of MBS as implicit in share

prices. Empirical models of direct pricing of MBS are offered by Dunn

and Singleton (1983) and Boudoukh et al. (1997).

H. Huizinga, L. Laeven / Journal of Financial Economics 106 (2012) 614–634 621

assets are obtained from Schedule HC-B of the Call reportand are to be reported only by bank holding companieswith average trading assets of $2 million or more in any ofthe four preceding quarters. Trading assets, which includesome MBS, are carried at fair value and held in the bank’strading book. A detailed split-up of trading assets is onlyavailable for domestic offices of bank holding companiesand is not reported. On average, trading assets onlyamount to 0.5% of assets, because only large banks tendto have trading assets. Tier 1 denotes the Tier 1 capitalratio, which equals 10.8%, on average. The mean share ofTier 1 capital in bank capital equals 86.3%. Loan lossprovisioning is loan loss provisions as a percentage ofthe book value of loans. The mean loan loss provisioningrate is 0.8%. Net charge-offs is the difference between loancharge-offs and loan recoveries. The mean net loancharge-off rate is 0.5% of loans. Thus, loan loss provision-ing exceeded net loan charge-offs, as expectations ofadditional future loan losses surpassed loan write-offs.The average loan loss provisioning rate increased sharplyduring the crisis from less than 0.1% in the first quarter of2007 to 1.5% in the fourth quarter of 2008.

4. Tobin’s q and market discounts

This section provides the results of regressions of q toobtain implicit market valuations of main balance sheetitems. The focus is on the real estate components of loansand securities. This emphasis is justified by the fact thatthe real estate components of loans and securitiestogether comprise, on average, 63% of bank assets in2008, and by the fact that real estate assets have sufferedfrom house price declines during the recent financialcrisis. Nevertheless, we include several other balancesheet items in the analysis as well.

To start, Table 2 reports regressions of q that includethe real estate loans and MBS variables with data for

2008.6 The regressions also control for exposure to non-real estate loans, trading assets, and the composition ofbank capital. All regressions include US state fixed effectsand time fixed effects (except the quarterly regressions)to control for systematic differences across US states andquarterly periods, such as housing and labor marketconditions, or the monetary policy stance. Standard errorsare corrected for clustering at the firm level. The realestate loans variable enters regression 1 with a coefficientof �0.173 that is significant at the 1% level implying thatthe implicit market discount of real estate loans relativeto book value is 17.3%. The MBS variable similarly enterswith a coefficient of �0.136 that is significant at the 5%level so that MBS appear to be discounted 13.6%.7

The economic effects of these estimates are substan-tial. A market discount of 17.3% applied to the averagebank’s real estate loan portfolio (equal to 53.3% of assetsin 2008 according to Table 1) causes q to drop by 0.092,which about mirrors the slide of q from its peak of 1.10 in2004 to 1.00 at the end of 2008 seen in Fig. 1. The drop inbanks’ market values during 2008 thus seems to bemostly due to the deterioration of banks’ real estate loans.Our estimation further implies that a one standard devia-tion increase in real estate loans would reduce q byanother 2.4 percentage points, and that a one standard

Table 2Tobin’s q and real estate-related assets during 2008.

The dependent variable is Tobin’s q. Non-real estate loans is ratio of non-real estate loans to assets. Real estate loans is ratio of real estate loans to

assets. MBS is ratio of MBS to assets. MBS, held is ratio of MBS that are held-to-maturity to assets. MBS, for sale is ratio of MBS that are available-for-sale

to assets. Trading is ratio of assets in trading account to total assets. Share of Tier 1 is ratio of Tier 1 capital in total capital. Leverage is ratio of liabilities to

assets. Risk-weighted assets is ratio of risk-weighted assets to total assets. Liquid assets is ratio of holdings of cash and US securities to assets. Fixed

assets is ratio of premises and fixed assets to total assets. Sample consists of US bank holding companies that are stock-listed. Sample in columns 1, 2, and

7 includes observations for all four quarters during the year 2008. Sample in column 3 is the first quarter of 2008. Sample in column 4 is the second

quarter of 2008. Sample in column 5 is the third quarter of 2008. Sample in column 6 is the fourth quarter of 2008. All regressions include state fixed

effects; regressions in columns 1, 2 and 7 also include quarterly period fixed effects (not reported). Standard errors in columns 1, 2, and 7 are corrected

for clustering at the bank level. Columns 3–6 report White’s heteroskedasticity-consistent standard errors. Standard errors are in parentheses. n, nn, Andnnn denote significance at the 10%, 5%, and 1% levels, respectively.

2008 2008 2008Q1 2008Q2 2008Q3 2008Q4 2008

Variables (1) (2) (3) (4) (5) (6) (7)

Non-real estate loans �0.042 �0.049 �0.046 �0.059 �0.056 �0.041 0.017

(0.058) (0.056) (0.064) (0.073) (0.081) (0.052) (0.075)

Real estate loans �0.173nnn�0.176nnn

�0.169nnn�0.180nnn

�0.223nnn�0.142nnn

�0.132nn

(0.052) (0.051) (0.056) (0.067) (0.074) (0.047) (0.065)

MBS �0.136nn

(0.063)

MBS, held �0.219nnn�0.209nn

�0.237nn�0.282nnn

�0.170nn�0.271nnn

(0.077) (0.097) (0.096) (0.097) (0.082) (0.079)

MBS, for sale �0.119n�0.165nn

�0.127 �0.109 �0.099 �0.136n

(0.068) (0.077) (0.086) (0.092) (0.063) (0.074)

Trading �0.285nnn�0.288nnn

�0.271nnn�0.269nn

�0.326nn�0.286nnn

�0.310nnn

(0.088) (0.087) (0.101) (0.106) (0.147) (0.093) (0.099)

Share of Tier 1 0.103nnn 0.105nnn 0.083nn 0.122nnn 0.150nnn 0.093nnn 0.102nnn

(0.030) (0.030) (0.041) (0.040) (0.049) (0.031) (0.031)

Leverage 0.278nn

(0.108)

Risk-weighted assets �0.090n

(0.054)

Liquid assets 0.092

(0.161)

Fixed assets �0.258

(0.316)

Constant 1.011nnn 1.012nnn 1.031nnn 0.999nnn 0.987nnn 0.959nnn 0.812nnn

(0.043) (0.043) (0.057) (0.058) (0.072) (0.039) (0.119)

Time fixed effects Y Y N N N N Y

State fixed effects Y Y Y Y Y Y Y

Observations 1,114 1,114 286 279 277 272 1,114

R-squared 0.377 0.380 0.343 0.389 0.418 0.327 0.397

H. Huizinga, L. Laeven / Journal of Financial Economics 106 (2012) 614–634622

deviation increase in MBS would reduce q by another 1.0percentage points. These are substantial effects given thestandard deviation of q of 5.5%.

Trading enters the regression with a coefficient of�0.285 that is significant at the 1% level. The economiceffect of this result is small given that trading assets, onaverage, comprise only 0.5% of total assets in 2008. Thecomposition of equity capital can also influence bankvalue, especially during 2008 as markets reassessed thesuperior value of Tier 1 capital relative to Tier 2 capital,partly in response to stricter capital requirements pro-posed by regulators. We find that Tier 1, denoting theshare of Tier 1 capital in total capital, enters with apositive coefficient of 0.103 that is significant at the 1%level. This suggests that a one standard deviation increaseof 8% in the share of Tier 1 capital in total capital increasesbank value by 0.8 percentage points, which is not irrele-vant given a standard deviation of q of 5.5%.

In regression 2, we replace the MBS variable with twoseparate variables, MBS, held and MBS, for sale thatrepresent the parts of MBS that are held-to-maturity(and carried at amortized cost) and available-for-sale

(and carried at fair value). To the extent that fair valuesreflect market prices, we expect the discount to be smallerfor available-for-sale MBS. The MBS, held variable obtainsa coefficient of �0.219 that is significant at 1%, while theMBS, for sale variable enters with a coefficient of �0.119that is significant at 10%. Thus, MBS classified as held-to-maturity appear to be discounted significantly at 21.9%,while the MBS available-for-sale tend to have a smallerdiscount of 11.9%, on average. Thus, the gap betweenimplicit market prices and accounting values appears tobe largest for MBS classified as held-to-maturity.

In the next four regressions in Table 2, we re-estimateregression 2 separately for each of the four quarters in2008. As we lose the time-series dimension, these reg-ressions do not include time fixed effects. We find that thediscounts on the real estate-related variables are fairly stableover these four quarters. Interestingly, the discounts on realestate loans and MBS that are held-to-maturity both reach apeak of 22.3% and 28.2%, respectively, during the thirdquarter of 2008, though the difference is not statisticallysignificant. In separate regressions, we further split theMBS variables into their guaranteed and nonguaranteed

H. Huizinga, L. Laeven / Journal of Financial Economics 106 (2012) 614–634 623

parts and obtain that the nonguaranteed MBS com-manded the highest discount during the crisis period(not reported).8

In regression 7 we re-estimate regression 2 afterincluding additional control variables. Leverage, mea-sured as the ratio of total liabilities to total assets,captures leverage on a non-risk-adjusted basis. Risk-weighted assets, measured as the ratio of risk-weightedassets to total assets, controls for the riskiness of thebank’s on-balance sheet and off-balance sheet assets.Liquid assets, measured as the ratio of cash holdings andholdings of US Treasury securities to total assets, controlsfor the liquidity position of the bank. Fixed assets,measured as the ratio of premises and fixed assets tototal assets, controls for the fraction of tangible assets. Wefind that the results on the real estate loans and MBSvariables are robust to the inclusion of these additionalcontrol variables. In unreported regressions, we find thatthe results are also robust to the inclusion of the ratio oftotal deposits to total assets and the ratio of bank capitalto assets.

The evidence indicates sizeable market discounts onreal estate-related assets relative to book value for USbank holding companies in 2008. As we have data startingin 2001, we next analyze whether such discounts existedbefore 2008. For this purpose, we re-estimate regression 2of Table 2 with data for each of the years 2001–2007. Theresults are reported in Table 3.

Throughout the period 2001–2007, none of the realestate asset categories is estimated with a significantdiscount. Thus, real estate loans and MBS asset categoriesare not estimated with significant discounts until 2008,suggesting that the deterioration of the implicit marketvalue of real estate assets relative to book value wassudden rather than gradual.9

Regression 8 in Table 3 uses data for the period 2001–2008 and includes interactions between the explanatoryvariables and an indicator variable that takes a value ofone for year 2008 observations to see whether the 2008discounts are significantly different from those in earlieryears. We indeed find that q is significantly lower onaccount of exposure to real estate loans and held-to-maturity MBS in 2008 as compared to earlier years. Realestate loans obtain a discount of 14.7% compared toearlier years, and held-to-maturity MBS obtain a discountof 9.0% compared to earlier years. Both results aresignificant at the 1% level. Available-for-sale MBS, on theother hand, do not obtain a discount relative to earlier

8 We also estimated regression 2 separately for each of the four

quarters in 2007 and did not find that the market already discounted

MBS exposure in 2007 (not reported).9 We thus find asymmetric deviations of book values from market

values over the cycle, with market discounts for real estate-related

assets in 2008, but no market premiums prior to 2008. This suggests that

banks mark most of their assets to market values during good times (the

pre-2008 period). This asymmetric pattern for banks is different from

that for nonfinancial firms that are dependent on the real estate sector,

such as firms in the construction and real estate industries, for which the

median Tobin’s q (based on Compustat quarterly data) shows an

increase during the 2001–2006 period followed by a subsequent decline

during the years 2007 and 2008.

years, suggesting that reductions in book value ofthese assets have kept pace with market perceptions ofimpairment.

The discount on real estate loans in 2008 relative toprior years of 14.7% is larger than the analogous discounton non-real estate loans of 9.8%, consistent with theexpectation that US house price declines affect loanimpairment of mortgage loans the most.10 All the same,non-real estate loans also obtain a significant discount in2008, consistent with expected loan impairment andaccounting discretion on the valuation of such impair-ment for non-real estate loans as well. In principle, thereis no reason why accounting discretion over non-realestate loans should be any different than discretion overreal estate loans. The overall market discount on non-realestate loans in 2008 (obtained by summing up thecoefficients on Non-real estate loans and Non-real estateloans�2008), however, is smaller (in absolute terms)than the analogous discount on real estate loans. To beprecise, the estimated discount in 2008 on non-real estateloans is �0.046 compared to a discount of �0.090 on realestate loans, and an F-test indicates that these discountsare significantly different at the 1% level. In other words,the estimated discount on non-real estate loans is onlyabout half that on real estate loans. The economic effectfor real estate loans is also markedly larger than that fornon-real estate loans: a one standard deviation increase inthe ratio of non-real estate loans implies a reduction inTobin’s q in 2008 of 0.004, while a one standard deviationincrease in the ratio of real estate loans implies a reduc-tion in Tobin’s q in 2008 of 0.012. Thus, the economiceffect for real estate loans is almost three times as large asthat for non-real estate loans.

One concern is that our results are driven by anovershooting in asset prices, meaning a temporary devia-tion in value from fundamental value. However, ourmeasure of firm value is based on equity prices, whichreflect the consensus view of many market participants.The stock market continued to be liquid throughout 2008,even if markets for some MBS became illiquid. Still,asymmetric information regarding underlying asset qual-ity appears to have led to fire sales in MBS markets(Diamond and Rajan, 2011). The estimated market dis-counts for MBS implicit in equity prices therefore arelikely to reflect a combination of book overvaluation ofthese assets and market assessment of the likelihood thatbanks may need to sell these assets in a fire sale.

We continue to find implicit discounts on real estate-related assets in the final quarter of 2008, following themassive government interventions in October 2008. Lauxand Leuz (2010) argue that these interventions shouldhave reduced the likelihood of distressed sales of banks’assets into illiquid markets. The implicit discounts on realestate related-assets at this time thus are likely to reflectbook overvaluation rather than the chance that theseassets have to be sold in a fire sale. Furthermore,

10 When splitting the non-real estate loan variable in its two main

components—commercial and industrial (C&I) loans and consumer

loans—we find that the market discount on non-real estate loans in

2008 is mainly driven by C&I loans.

Table 3Tobin’s q and real estate related assets during 2001–2008.

The dependent variable is Tobin’s q. Non-real estate loans is ratio of non-real estate loans to assets. Real estate loans is ratio of real estate loans to

assets. MBS, held is ratio of held-to-maturity MBS to assets. MBS, for sale is ratio of available-for-sale MBS to assets. Trading is ratio of trading account

assets to total assets. Share of Tier 1 is ratio of Tier 1 capital in total capital. 2008 denotes observations from year 2008. Sample consists of US bank

holding companies that are stock-listed. Regressions in columns 1 to 7 include state fixed effects and quarterly period fixed effects (not reported), with

standard errors corrected for clustering at the bank level. Regression in column 8 includes firm fixed effects. Sample consists of US bank holding

companies that are stock-listed. Data are based on quarterly observations over the period 2001–2008. Standard errors are in parentheses. n, nn, And nnn

denote significance at the 10%, 5%, and 1% levels, respectively.

2001 2002 2003 2004 2005 2006 2007 2001–2008

Variables (1) (2) (3) (4) (5) (6) (7) (8)

Non-real estate loans 0.068 0.091 0.002 0.0927n 0.104nn 0.084n 0.042 0.052nnn

(0.078) (0.059) (0.049) (0.051) (0.046) (0.050) (0.054) (0.013)

Real estate loans �0.011 0.065 0.023 0.026 0.023 �0.014 �0.063 0.057nnn

(0.058) (0.045) (0.040) (0.041) (0.039) (0.045) (0.049) (0.010)

MBS, held 0.066 0.150 0.063 0.024 0.002 �0.029 �0.052 �0.0068

(0.121) (0.106) (0.100) (0.070) (0.062) (0.070) (0.080) (0.025)

MBS, for sale 0.043 0.066 0.013 0.053 0.027 �0.000 �0.068 0.045nnn

(0.075) (0.058) (0.051) (0.050) (0.048) (0.056) (0.062) (0.012)

Trading 0.198 0.117 0.038 �0.005 �0.209n�0.193n

�0.139n 0.140nnn

(0.263) (0.202) (0.161) (0.127) (0.110) (0.105) (0.075) (0.046)

Share of Tier 1 �0.037 �0.038 �0.011 0.030 �0.039 0.011 0.009 0.065nnn

(0.068) (0.056) (0.041) (0.037) (0.037) (0.042) (0.033) (0.010)

Non-real estate loansn2008 �0.098nnn

(0.017)

Real estate loansn2008 �0.147nnn

(0.014)

MBS, heldn2008 �0.090nnn

(0.034)

MBS, for salen2008 �0.021

(0.019)

Tradingn2008 �0.134nnn

(0.048)

Share of Tier 1n2008 0.127nnn

(0.014)

Constant 1.087nnn 1.017nnn 1.046nnn 1.016nnn 1.068nnn 1.045nnn 1.057nnn 0.964nnn

(0.087) (0.069) (0.049) (0.052) (0.050) (0.055) (0.048) (0.012)

Time fixed effects Y Y Y Y Y Y Y Y

State fixed effects Y Y Y Y Y Y Y N

Firm fixed effects N N N N N N N Y

Observations 287 1,189 1,247 1,264 1,287 1,167 1,155 8,710

R2 0.206 0.183 0.190 0.245 0.310 0.318 0.362 0.463

H. Huizinga, L. Laeven / Journal of Financial Economics 106 (2012) 614–634624

empirical evidence on the pricing of investment-gradecredit risk during the financial crisis provided by Covalet al. (2009) casts doubt on the premise that prices incredit markets were systematically distorted. We there-fore conclude that the book values of real estate-relatedassets on the books of banks were inflated in 2008, thoughwe do not rule out that part of the estimated discountsreflect potential fire sales.

Accounting discretion on the part of banks can beexplained by banks’ desire to overstate asset valuesduring downturns in the presence of regulatory forbear-ance. Such forbearance by regulators can be motivated bypolitical considerations to allow banks to continue to lendand support the real economy during downturns despitetheir weakened capital positions, as well as by careerrepercussions for the regulators themselves if they closedown potentially viable banks. Such considerations aremore pronounced for large banks—giving rise to the toobig to fail problem. If regulatory pressure on large banks isless pronounced, then we should find that increases in themarket discounts on real estate assets in 2008 are morepronounced for such banks. In Table 4, we separately

estimate market discounts for small and large banks(using the median value of assets in a given quarter as acutoff). We find that the estimated differential marketdiscounts on real estate-related assets in 2008 are morepronounced for larger banks, consistent with regulatoryforbearance and too big to fail considerations (columns 1and 2). These results are confirmed when we use aninteraction term between a dummy variable denotinglarge banks and the real estate loan variable (column 3,Table 4). Our interpretation of these results is thatregulators are aware of the deterioration in asset valuesbut that they allow banks, and especially large banks, touse accounting discretion to overstate the value of theassets in their books at a time of financial crisis.

Next, we distinguish between guaranteed MBS andnonguaranteed (private label) MBS. Our market discountresults should be more pronounced for private label MBSbecause they are not protected against default risk, unlikethe guaranteed MBS that are backed by indirect govern-ment guarantees (see Keys et al., 2009). Moreover, themarket for private label MBS experienced an illiquidityfreeze starting in mid-2007 and throughout 2008, while

Table 4Tobin’s q, bank size, and guarantees on mortgage-backed securities during 2001–2008.

The dependent variable is Tobin’s q. Non-real estate loans is ratio of non-real estate loans to assets. Real estate loans is ratio of real estate loans to

assets. MBS, held is the ratio of held-to-maturity MBS to assets. MBS, for sale is the ratio of available-for-sale MBS to assets. MBS, nonguaranteed is the

ratio of nonguaranteed MBS to assets, MBS, guaranteed is the ratio of guaranteed MBS to assets, with held-to-maturity MBS valued at amortized cost and

available-for-sale MBS valued at fair value. Trading is ratio of trading account assets to total assets. Share of Tier 1 is ratio of Tier 1 capital in total capital.

Large bank is a dummy variable that takes a value of one if total assets are above the sample median in a given quarter. 2008 denotes observations from

year 2008. Sample consists of US bank holding companies that are stock-listed. Regressions in columns 1 and 2 split the sample based on whether banks

have total assets below the sample median in a given quarter (small banks) or above the sample median (large banks). Regressions include firm and

quarterly period fixed effects (not reported). Data are based on quarterly observations over the period 2001–2008. Standard errors are in parentheses. n,nn, And nnn denote significance at the 10%, 5%, and 1% levels, respectively.

(1) (2) (3) (4)

Variables Small banks Large banks Interaction with large bank MBS guarantees

Non-real estate loans �0.012 0.160nnn 0.051nnn 0.046nnn

(0.017) (0.021) (0.013) (0.014)

Real estate loans �0.003 0.081nnn 0.054nnn 0.059nnn

(0.013) (0.016) (0.010) (0.010)

Real estate loansnLarge bank �0.001

(0.004)

MBS, held 0.041 0.022 �0.008

(0.043) (0.031) (0.025)

MBS, for sale �0.078nnn 0.136nnn 0.039nnn

(0.016) (0.017) (0.012)

MBS, nonguaranteed 0.070nn

(0.031)

MBS, guaranteed 0.036nnn

(0.012)

Trading �0.104 0.315nnn 0.136nnn

(0.072) (0.062) (0.046)

Share of Tier 1 0.051nnn 0.087nnn 0.067nnn

(0.014) (0.014) (0.010)

Non-real estate loansn2008 �0.100nnn�0.112nnn

�0.090nnn�0.108nnn

(0.023) (0.025) (0.017) (0.016)

Real estate loansn2008 �0.110nnn�0.213nnn

�0.138nnn�0.130nnn

(0.018) (0.020) (0.014) (0.012)

Real estate loansnLarge bankn2008 �0.030nnn

(0.004)

MBS, heldn2008 �0.041 �0.0900nn�0.072nn

(0.050) (0.045) (0.034)

MBS, for salen2008 0.058nn�0.0439n

�0.006

(0.026) (0.025) (0.019)

MBS, nonguaranteedn2008 �0.118nn

(0.055)

MBS, guaranteedn2008 0.016

(0.018)

Tradingn2008 �0.130 �0.169nnn�0.148nnn

(0.111) (0.056) (0.048)

Share of Tier 1n2008 0.016 0.159nnn 0.100nnn

(0.024) (0.019) (0.015)

Observations 4,356 4,354 8,710 8,710

R2 0.461 0.524 0.466 0.449

Number of banks 197 192 338 338

H. Huizinga, L. Laeven / Journal of Financial Economics 106 (2012) 614–634 625

the market for guaranteed MBS remained liquid until latein the crisis. We therefore ran a discount regression withseparate variables for guaranteed MBS and nonguaran-teed MBS (column 4, Table 4). We find, as expected, thatthe discount effect on MBS during the crisis period isdriven by nonguaranteed MBS for which the market valuewas most depressed, on account of lack of protectionagainst default risk and/or an illiquidity freeze. In parti-cular, we find a negative and significant coefficient for aninteraction term of MBS, nonguaranteed with a dummyvariable for 2008 in a regression of Tobin’s q. This resultsuggests that banks applied more accounting discretion tothe valuation of private label MBS, consistent with ourfindings thus far.

5. Discretion regarding asset impairment disclosure andasset classification

Our evidence on discounts on real estate-relatedassets implicit in bank stock prices suggests that theinformation disclosure on the impairment of these assetshas not kept up with market perceptions of this impair-ment. In this section, we show that this does not merelyreflect a passive and irresponsive financial reportingsystem, but it also results from the active use by banksof discretion over their financial reporting to preventdeclines in the book value of assets. We consider discre-tion over loan impairment and MBS classification inturn.

H. Huizinga, L. Laeven / Journal of Financial Economics 106 (2012) 614–634626

5.1. Discretion over provisioning for loan losses

In this section, we examine whether distressed bankshave systematically held back on their loan loss provisioningduring the present financial crisis to preserve bank capital.In addition, we consider whether distressed banks havedifferent loan loss realizations in the form of loan charge-offs.

We first report regressions that test whether loan lossprovisioning behavior of banks changed significantly dur-ing the crisis year 2008 when asset prices were depressed.In regression 1 of Table 5, the loan loss provisioningvariable is related to the share of real estate loans in totalloans and the share of MBS in total assets. We also includeinteraction terms between these two real estate exposurevariables and a dummy variable that denotes whether theobservation is for the year 2008 or not.

Table 5Loan loss provisions during 2001–2008.

The dependent variable is loan loss provisions as a percentage of total loans.

MBS is ratio of MBS to assets. MBS, held is the ratio of held-to-maturity MBS

nonguaranteed is the ratio of nonguaranteed MBS to total assets with held-to-m

at fair value. Similarly, MBS, guaranteed is ratio of guaranteed MBS to total asset

share in 2006 is share of bonus payments in total compensation and CEO owne

2006 data. Explanatory variables (apart from CEO variables) are lagged one quar

of US bank holding companies that are stock-listed. Regressions include firm

observations over the period 2001–2008. Standard errors are in parentheses. n,

Variables (1)

Share of real estate loans (t�1) �0.618nnn�

(0.104) (

MBS (t�1) �0.144

(0.124)

MBS, held (t�1) �

(

MBS, for sale (t�1) �

(

MBS, nonguaranteed (t�1)

MBS, guaranteed (t�1)

Share of real estate loans (t�1)n2008 0.265nnn 0

(0.101) (

MBS (t�1)n2008 �1.738nnn

(0.197)

MBS, held (t�1)n2008 �

(

MBS, for sale (t�1)n2008 �

(

Low valuation (t�1)

Share of real estate loans (t�1)nLow valuation (t�1)

MBS (t�1)nLow valuation (t�1)

MBS, held (t�1)nLow valuation (t�1)

MBS, for sale (t�1)nLow valuation (t�1)

MBS, nonguaranteed (t�1)n2008

MBS, guaranteed (t�1)n2008

CEO bonus share in 2006n2008

CEO ownership in 2006n2008

Observations 8,325

R2 0.357

We expect loan loss provisioning to be more positivelyrelated to the share of real estate loans in 2008, as theseloans have been particularly affected by recent houseprice declines. We indeed find that loan loss provisioningrate of banks with a large share of real estate loans wassignificantly higher during the year 2008. However, therewas a significant negative relation between the share ofreal estate loans and loan loss provisioning in earlieryears, possibly because banks were anticipating contin-ued appreciation of real estate prices.

Banks that need to absorb large losses arising fromexposure to MBS could lower their provisioning standardsin an effort to preserve regulatory capital. In line withthis, we find that the loan loss provisioning rate of bankswith large MBS exposure was significantly lower duringthe year 2008, suggesting that banks with potential large

Share of real estate loans is the fraction of real estate loans in total loans.

to assets. MBS, for sale is ratio of available-for-sale MBS to assets. MBS,

aturity MBS valued at amortized cost and available-for-sale MBS valued

s. Low valuation denotes banks with Tobin’s q of less than one. CEO bonus

rship in 2006 is fraction of shares held by the CEO, both measured using

terly period. 2008 denotes observations from year 2008. Sample consists

and quarter fixed effects (not reported). Data are based on quarterlynn, And nnn denote significance at the 10%, 5%, and 1% levels, respectively.

(2) (3) (4) (5) (6)

0.629nnn�0.618nnn

�0.631nnn�0.615nnn

�0.386

0.104) (0.103) (0.103) (0.104) (0.255)

�0.139 �0.214

(0.124) (0.266)

0.041 0.039

0.306) (0.300)

0.173 �0.191

0.127) (0.127)

�0.153

(0.129)

�0.554

(0.376)

.318nnn 0.268nnn 1.992nnn

0.102) (0.101) (0.200)

�4.892nnn

(0.517)

3.356nnn

0.455)

1.380nnn

0.216)

0.492nnn 0.381nnn

(0.099) (0.101)

�0.019 0.097

(0.123) (0.125)

�0.986nnn

(0.225)

�3.563nnn

(0.558)

�0.474n

(0.247)

�2.315nnn

(0.211)

3.444nnn

(0.735)

0.119nnn

(0.013)

0.084nnn

(0.007)

8,325 8,325 8,325 8,325 2,181

0.358 0.382 0.384 0.362 0.515

H. Huizinga, L. Laeven / Journal of Financial Economics 106 (2012) 614–634 627

losses on MBS were holding back on their loan lossprovisioning in that year. The MBS exposure variableobtains a statistically significant coefficient that is �1.74lower in 2008 than in earlier years.

The economic effect of this result is substantial. A onestandard deviation increase in MBS implies a reduction inthe loan loss provisioning rate in 2008 compared toearlier years of 0.13 percentage points. This is a substan-tial effect compared to the standard deviation of the loanloss provisioning rate of 1.0%. For a bank with the meanloan to assets ratio of 0.71, the implied reduction in loanloss provisioning equals 0.1% of assets. The average bankhad a standard deviation of the Tier 1 capital to assetsratio of 1.8% in 2008, and hence, a capital saving of 0.1% ofassets is substantial.

In regression 2, we find that the loan loss provisioningrate is more negatively related to MBS that are held-to-maturity than to MBS that are available-for-sale in 2008as compared to earlier years. Perhaps this reflects that theamortized cost of MBS exceeds fair value in 2008, whichimplies relatively large future write-downs of MBS thatare held-to-maturity.

Next, we analyze whether loan loss provisioning beha-vior has been significantly different for low-valuationbanks. The incentive to hold back on loan loss provisioningshould be particularly pronounced for banks with dis-tressed market values. To test this, regressions 3 and 4 inTable 5 include interaction terms between the real estateexposure variables and a dummy variable that takes avalue of one for banks with a value of q of less than one.Regression 3 reports a coefficient for the MBS variable thatis more negative at �0.986 and statistically significant at1% for low-valuation banks, while regression 4 shows thatthe result is particularly pronounced for held-to-maturityMBS. Thus, we find that the negative relation betweenloan loss provisioning and a bank’s MBS exposure issignificantly stronger for banks with a low valuation, asmeasured by a value of q of less than one. This suggeststhat distressed banks with large exposure to MBS wereparticularly holding back on their loan provisioning.

Next, we split the MBS exposure variable in its guar-anteed and nonguaranteed parts (column 5 of Table 5).We find that the lower level of provisioning during thecrisis is driven by exposure to nonguaranteed MBS, whichare not protected against default risk and for which themarket was illiquid during 2008.

We also tested whether loan loss provisioning beha-vior depends on the size of the bank. We find thatprovisioning by large banks is relatively more sensitiveto MBS exposure in 2008 (not reported), consistent with alarger discounting of real estate and non-real estate loansof large banks as shown in Table 4.

Finally, we consider whether managerial compensa-tion and ownership can be shown to affect provisioningbehavior.11 Shareholders may be more interested in riskystrategies than bank managers, who could lose their jobs

11 An emerging literature investigates the impact of managerial

compensation and ownership on the performance of banks during the

crisis (see, e.g., Fahlenbrach and Stulz, 2011; Keys et al., 2009; Ellul and

Yerramilli, 2010).

if a bank fails. Bank managers whose incentives are wellaligned with those of bank shareholders thus can beexpected to choose relatively risky bank asset portfolios.12

During a financial crisis, banks with risky loan portfoliosare the ones that need to provision relatively much forloan losses. Thus, banks in which managers pre-crisis (in2006) had incentives to take more risk can be expected toneed to provision more for loan losses during the crisis (in2008). We test this hypothesis, focusing on the behaviorof the chief executive officer (CEO) of each bank.

As proxies for the alignment of CEO incentives withshareholder interests, we use two variables: CEO bonusshare, which is the share of bonus payments in total CEOcompensation, and CEO ownership, which is the fractionof shares in the firm held by the CEO. Both variables areconstructed using data from Compustat’s ExecuCompdatabase for the year 2006. We extend the provisioningmodel in column 1 of Table 5 to include these two CEO-specific variables interacted with a 2008 dummy variable.The sample is limited to the period 2006–2008, becausethere is a break in the coverage and definition of Execu-Comp data starting in 2006. The sample is further reducedbecause the ExecuComp database only covers Stan-dard&Poor (S&P) 1500 firms. The results are presentedin column 6 of Table 5. Both CEO variables (interactedwith the 2008 dummy variable) enter the regression withpositive coefficients that are significant at 1%. This isevidence that bank managers whose incentives werealigned more with shareholder interests pre-crisis endedup provisioning more for loan losses during the crisis.

Distressed banks could also be slow in recognizinglosses on their real estate loan portfolio in the form ofwrite-downs or charge-offs.13 To analyze this, regressions1–5 of Table 6 take as dependent variable the ratio of netcharge-offs to loans (where net charge-offs are the differ-ence between charge-offs and recoveries) expressed inpercentages. Otherwise, these regressions are similar toregressions 1–5 of Table 5. Consistent with the earlierresults, we find in regression 1 of Table 6 that the ratio ofnet charge-offs to loans is negatively and significantlyrelated to the MBS variable in 2008 compared to earlieryears. The results of regressions 2–4 in Table 6 similarlyare analogous to those in regressions 2–4 of Table 5.When splitting the MBS exposure variable in its guaran-teed and nonguaranteed parts (column 5 of Table 6), wefind that charge-offs in 2008 are more pronounced whenbanks hold more nonguaranteed MBS, contrary to theprovisioning regression 5 of Table 5. One reason for thisdifference could be that banks face less discretion inaccounting for loan charge-offs than for loss provisioning.Another potential reason is that provisioning is deductedfrom profits and therefore has an immediate negativeimpact on bank capital, while loan charge-offs arededucted from loan loss reserves built up over time.

12 Laeven and Levine (2009) find that bank risk-taking varies

positively with the comparative power of shareholders vis- �a-vis bank

managers.13 Loan write-downs include write-downs arising from the transfer

of loans to a held-for-sale account.

Table 6Net loan charge-offs during 2001–2008.

The dependent variable is net loan charge-offs as a percentage of total loans. Share of real estate loans is the fraction of real estate loans in total loans.

MBS is ratio of MBS to assets. MBS, held is ratio of held-to-maturity MBS to assets. MBS, for sale is ratio of available-for-sale MBS to assets. MBS,

nonguaranteed is ratio of nonguaranteed MBS to total assets, with held-to-maturity MBS valued at amortized cost and available-for-sale MBS valued at

fair value. MBS, guaranteed is ratio of guaranteed MBS, with held-to-maturity MBS valued at amortized cost and available-for-sale MBS valued at fair

value. Low valuation denotes banks with Tobin’s q of less than one. Explanatory variables are lagged one quarterly period. 2008 denotes observations

from year 2008. Sample consists of US bank holding companies that are stock-listed. Regressions include firm and quarterly period fixed effects (not

reported). Data are based on quarterly observations over the period 2001–2008. Standard errors are in parentheses. n, nn, And nnn denote significance at

the 10%, 5%, and 1% levels, respectively.

Variables (1) (2) (3) (4) (5)

Share of real estate loans (t�1) �0.883nnn�0.886nnn

�0.932nnn�0.937nnn

�0.880nnn

(0.088) (0.088) (0.088) (0.088) (0.087)

MBS (t�1) �0.104 �0.081

(0.105) (0.105)

MBS, held (t�1) �0.097 �0.030

(0.258) (0.255)

MBS, for sale (t�1) �0.110 �0.099

(0.107) (0.108)

MBS, nonguaranteed (t�1) �0.404

(0.317)

MBS, guaranteed (t�1) �0.114

(0.109)

Share of real estate loans (t�1)n2008 0.486nnn 0.502nnn 0.488nnn

(0.085) (0.086) (0.085)

MBS (t�1)n2008 �1.191nnn

(0.166)

MBS, held (t�1)n2008 �1.685nnn

(0.384)

MBS, for sale (t�1)n2008 �1.083nnn

(0.183)

Low valuation (t�1) �0.036 �0.079

(0.084) (0.086)

Share of real estate loans (t�1)nLow valuation (t�1) 0.491nnn 0.536nnn

(0.104) (0.106)

MBS (t�1)nLow valuation (t�1) �0.803nnn

(0.191)

MBS, held (t�1)nLow valuation (t�1) �1.793nnn

(0.475)

MBS, for sale (t�1)nLow valuation (t�1) �0.606nnn

(0.210)

MBS, nonguaranteed (t�1)n2008 2.997nnn

(0.620)

MBS, guaranteed (t�1)n2008 �1.658nnn

(0.178)

Observations 8,325 8,325 8,325 8,325 8,325

R2 0.256 0.256 0.274 0.275 0.260

H. Huizinga, L. Laeven / Journal of Financial Economics 106 (2012) 614–634628

We also estimated loan loss provisioning and write-offregressions separately for each year (not reported) andfind that there is hardly any impact of MBS exposureon provisioning and write-offs up until 2008, whenthe market values for MBS fell and banks with largeexposures to held-to-maturity MBS started to provisionsignificantly less. Overall, we find evidence that low-valuation banks with large MBS exposures hold back ontheir loan loss provisioning and charge-offs at a time offinancial crisis.

5.2. Classification of mortgage-backed securities

Reclassification of previously acquired securities gen-erally affects the overall book value of securities. Specifi-cally, book value rises if available-for-sale securities arereclassified as held-to-maturity at a time when amortizedcost exceeds fair value. In 2008, the mean ratio of fairvalue to amortized cost for nonguaranteed MBS was

0.925, against a mean ratio of fair value to amortized costfor guaranteed MBS of 1.005 (Fig. 5). These accountingvaluations gave banks an incentive to reclassify nonguar-anteed MBS as held-to-maturity to the extent possible toboost the book value of assets. Indeed, the fraction ofnonguaranteed MBS that is held-to-maturity increasedfrom 7.6% at end-2007 to 11.4% at end-2008, consistentwith banks having incentives during 2008 to classify alarger fraction of their MBS as held-to-maturity (Fig. 4).

Table 7 reports regressions of the share of held-to-maturity MBS in total assets. In regression 1, we obtain anegative and significant relation between the share ofheld-to-maturity MBS and the loan variables. We also findthat banks with a larger share of MBS in total assets,represented by the MBS amortized variable, tend to holdmore MBS as held-to-maturity relative to total assets. Theestimated coefficient of 0.071 on the MBS amortizedvariable can be interpreted as the marginal propensityto hold MBS as held-to-maturity. The MBS difference

Table 7Share of held-to-maturity mortgage-backed securities during 2001–2008.

The dependent variable is the ratio of held-to-maturity MBS to total assets. Non-real estate loans is ratio of non-real

estate loans to assets. Real estate loans is ratio of real estate loans to assets. MBS amortized is total MBS at amortized cost

to total assets. MBS difference is the difference between MBS at amortized cost and MBS at fair value scaled by total

assets. Low valuation denotes banks with Tobin’s q of less than one. 2008 denotes observations from year 2008. Sample

consists of US bank holding companies that are stock-listed. Regressions include firm and quarterly period fixed effects

(not reported). Explanatory variables are lagged one quarterly period. Data are based on quarterly observations. Standard

errors are in parentheses. n, nn, And nnn denote significance at the 10%, 5%, and 1% levels, respectively.

Variables (1) (2) (3)

Non-real estate loans (t�1) �0.019nnn�0.017nnn

�0.016nnn

(0.006) (0.006) (0.006)

Real estate loans (t�1) �0.029nnn�0.030nnn

�0.029nnn

(0.005) (0.005) (0.005)

MBS amortized (t�1) 0.071nnn 0.071nnn 0.071nnn

(0.005) (0.005) (0.005)

MBS difference (t�1) 0.411nnn 0.298nnn 0.330nnn

(0.098) (0.104) (0.100)

Non-real estate loans (t�1)n2008 0.009

(0.008)

Real estate loans (t�1)n2008 0.015nn

(0.006)

MBS amortized (t�1)n2008 �0.021nn

(0.008)

MBS difference (t�1)n2008 1.086nnn

(0.291)

Low valuation (t�1) 0.000

(0.001)

Non-real estate loans (t�1)nLow valuation (t�1) �0.005

(0.005)

Real estate loans (t�1)nLow valuation (t�1) 0.005nn

(0.002)

MBS amortized (t�1)nLow valuation (t�1) �0.009

(0.008)

MBS difference (t�1)nLow valuation (t�1) 0.935nnn

(0.286)

Observations 8,350 8,350 8,350

R2 0.062 0.067 0.064

H. Huizinga, L. Laeven / Journal of Financial Economics 106 (2012) 614–634 629

variable is estimated with a positive coefficient that issignificant at the 1% level, consistent with a valuationincentive affecting MBS classification.

In regression 2, we add interactions between eachexplanatory variable and a dummy variable for year 2008observations to assess whether MBS classification behaviorchanged during 2008. The interacted real estate loansvariable obtains a coefficient that is positive and statisti-cally significant at the 5% level, suggesting that banks withlarge exposures to real estate loans had an incentive tocompensate for imminent losses arising from rapidly fall-ing house prices by reclassifying MBS. The interacted MBSamortized variable, instead, obtains a negative and signifi-cant coefficient. Thus, banks’ marginal propensity to holdMBS as held-to-maturity in 2008 was lower than before,perhaps because additions to MBS portfolios were primar-ily classified as available-for-sale in 2008. The interactedMBS difference variable obtains a positive coefficient thatis significant at the 1% level, suggesting that banks actedmore strongly on a reclassification incentive to boost assetvalues in 2008 than before.

The implied effects are economically significant. Speci-fically, regression 2 indicates that a one standard deviationincrease in the MBS difference variable implies an increasein the share of held-to-maturity MBS of 0.3 percentagepoints in 2008 as compared to earlier years. Similarly, a

one standard deviation increase in the real estate loansvariable implies an increase in the share of held-to-maturity MBS of 0.2 percentage points in 2008 relativeto earlier years. These are substantial effects compared tothe standard deviation of this MBS variable of 3.2%.

Regression 3 adds interactions between each explana-tory variable and the low-valuation dummy variable toassess whether bank distress influences the classificationof MBS. We find that the effects of the real estate loansand MBS difference variables on the share of held-to-maturity MBS are significantly more pronounced for low-valuation banks than for high-valuation banks. Thus,especially low-valuation banks with large real estate loanexposures and facing large differences between amortizedcost and fair value of MBS reported more MBS as held-to-maturity. We find no significant differential effect of theMBS amortized variable for low-valuation banks, suggest-ing that the propensity to report MBS as held-to-maturityis not different for distressed banks.

In sum, we find evidence that banks with large realestate loan exposures reported more of their MBS as held-to-maturity in 2008. In addition, we find that banksresponded to the reclassification incentive provided bythe difference between amortized cost and fair value ofMBS throughout the sample period, but more strongly in2008. The changed classification behavior in 2008 relative

Table 8Tobin’s q and fair value of net assets by valuation technique.

Dependent variable is Tobin’s q. Net level 1 assets is the fair value of assets minus liabilities valued using level 1 valuation techniques, scaled by total

assets. Net level 2 assets is the fair value of assets minus liabilities valued using level 2 valuation techniques, scaled by total assets. Net level 3 assets is

the fair value of assets minus liabilities valued using level 3 valuation techniques, scaled by total assets. MBS is ratio of MBS to assets. MBS, held is ratio of

held-to-maturity MBS to assets. MBS, for sale is ratio of available-for-sale MBS to assets. Sample consists of US bank holding companies that are stock-

listed. Regressions include state fixed effects and quarterly period fixed effects (not reported), with standard errors corrected for clustering at the bank

level. Regressions are reported separately for the years 2007 and 2008. Data are based on quarterly observations over the period 2007–2008. Standard

errors are in parentheses. n, nn, And nnn denote significance at the 10%, 5%, and 1% levels, respectively.

(1) (2) (3) (4) (5) (6)

Variables 2007 2008 2007 2008 2007 2008

Net level 1 assets 0.200 �0.257 0.208 �0.231 0.209 �0.221

(0.213) (0.543) (0.210) (0.550) (0.209) (0.554)

Net level 2 assets 0.027 0.215 0.012 0.215 0.010 0.211

(0.112) (0.146) (0.122) (0.149) (0.171) (0.145)

Net level 3 assets �0.322 �0.882nn�0.847 �1.119n

�0.846 �1.150n

(0.493) (0.413) (0.730) (0.619) (0.731) (0.629)

Net level 3 assetsnMBS 7.797 3.699

(10.11) (8.752)

Net level 3 assetsnMBS, held 11.07 �28.91

(127.9) (93.86)

Net level 3 assetsnMBS, for sale 7.738 4.638

(9.863) (8.580)

Constant 1.013nnn 1.067nnn 1.013nnn 1.067nnn 1.013nnn 1.067nnn

(0.002) (0.002) (0.002) (0.002) (0.002) (0.002)

Observations 570 647 570 647 570 647

R2 0.402 0.399 0.402 0.399 0.402 0.400

15 In practice, there is only limited variation in fair value accounting

techniques across banks, as 90% of fair value assets in 2008 were valued

as level 2 assets.16 Goh et al. (2009) estimate market discounts by relating a bank’s

stock price to net assets by valuation technique without considering

interactions with individual asset categories.17 Since Schedule HC-Q does not have separate information on level

H. Huizinga, L. Laeven / Journal of Financial Economics 106 (2012) 614–634630

to earlier years suggests that banks use their discretionover financial reporting to overstate the book value oftheir assets in the face of a severe financial crisis.

6. Market discounts across fair value accountingtechniques and the valuation of rule changes

We now consider two extensions of our analysis of theuse of accounting discretion by banks during the financialcrisis. First, we extend the analysis of Section 4 to estimatemarket discounts for net assets reported at fair value asbased on market prices (level 1), the prices of comparableassets (level 2), or a bank’s own valuation models (level 3).Second, we estimate banks’ stock price reaction to amend-ments to fair value accounting rules announced on Octo-ber 10, 2008 that appear to provide banks with morediscretion in determining the fair value of securities.14

6.1. Market valuation across fair valuation techniques

A breakdown of fair values of assets and liabilities byvaluation technique (levels 1–3) is available from Sche-dule HC-Q of the Call report. In this subsection, weestimate market discounts of net assets reported at fairvalue by valuation technique, and consider whether thesediscounts reflect a bank’s exposure to MBS. A direct

14 In a Web Appendix to this paper, we extend the analysis of loan

loss provisioning in Section 5.1 to test whether the tendency of banks

with high MBS exposures to reduce their loan loss provisioning in 2008

reflects their use of loan loss provisioning to smooth earnings in the pre-

2007 period. We find that banks with a tendency to apply discretionary

provisioning before 2007 provisioned less during the financial crisis of

2008, especially if they have significant MBS exposure.

estimation of market discounts of MBS by valuationtechnique is not possible, as the fair value accounting ofMBS is not broken down by valuation technique.15 How-ever, we can include an interaction term of a measure offair value accounting by valuation technique with bankMBS exposure to see whether market discounting of fairvalue assets by valuation technique systematicallydepends on MBS exposure.16

Table 8 reports market discount regression resultsusing q as dependent variable and the fair value of netassets, i.e., assets minus liabilities, by valuation technique(levels 1–3) scaled by total assets as explanatory vari-ables.17 These regressions do not include the loans andMBS variables seen in Table 2, as these balance sheetitems are also partially included in the net assets variablesrepresenting fair value valuation techniques. The numberof observations drops markedly to a sample of 181 banksthat file a Schedule HC-Q.18

1 assets for 2007, we compute the amount of level 1 assets in 2007 as

the difference between total assets reported at fair value and the sum of

levels 2 and 3 assets.18 Schedule HC-Q reporting financial assets and liabilities measured

at fair value is only filed by bank holding companies that have adopted

Financial Accounting Standards Board Statement no. 157 on fair value

measurements, and have elected to account for financial instruments

under a fair value option or have significant trading assets, with average

trading assets of $2 million or more in any of the four preceding

H. Huizinga, L. Laeven / Journal of Financial Economics 106 (2012) 614–634 631

The valuation of level 3 assets leaves relatively muchroom for accounting discretion. We therefore expect thatthe market applies larger discounts during the crisis year2008 to level 3 assets, when the market for many level 3assets became illiquid and it became relatively attractiveto use managerial discretion in the valuation of theseassets. Accordingly, in regression 2 we estimate a steepmarket discount of 88% for level 3 net assets during theyear 2008 that is significant at 5%, while the estimateddiscounts for levels 1 and 2 assets are statistically insig-nificant in that year. In contrast, in regression 1 we do notfind a significant discounting of fair value assets by anyaccounting technique during 2007. The steep discountingof level 3 net assets in 2008 is likely to reflect acombination of overvaluation of these assets on banks’balance sheets, and possibly market illiquidity resultingfrom thin trading and information asymmetry regardingunderlying asset quality.

We would like to know whether the apparent discountapplied to banks’ level 3 assets in 2008 is particularlypronounced for their real estate-related assets such asMBS that are the focus of this study. To check this,regression 4 includes an interaction term of the Net level3 assets and MBS variables. The estimated coefficient forthis interaction term should be negative, if MBS arerelatively overvalued on banks’ books and if a larger bankexposure to MBS leads to a larger share of MBS in the level3 category. The estimated coefficient is seen to be positivebut not statistically significant to suggest that the marketdiscounting applied to level 3 net assets in 2008 is notmaterially affected by MBS exposure. Similarly, we obtaininsignificant estimated coefficients on interaction termsof the Net level 3 assets variable with the MBS, held andMBS, for sale variables for 2008 in regression 6.

We estimated regressions analogous to Table 8 usingloan loss provisioning or charge-offs as the dependentvariable, but we did not find that these aspects ofaccounting for bad loans depended on the distributionof levels 1–3 assets in either 2007 or 2008 (not reported).This is not very surprising as the results in Table 5indicate that provisioning depends on exposure to realestate assets, while the choice of fair value accountingtechnique could be largely unrelated to a bank’s exposureto real estate-related assets such as MBS.

6.2. Banks’ stock price reaction to amendments of fair value

accounting rules

On October 10, 2008, the Financial Accounting Stan-dards Board (FASB) clarified rules for determining the fairvalue of a financial instrument applying FAS 157 whenthe market for that financial asset is not active.19 Theclarification made explicit that the use of a bank’s ownassumptions about future cash flows and risk-adjusted

(footnote continued)

quarters. See Ryan (2008) for a discussion of the application of FASB no.

157 during the subprime crisis.19 These rules, issued under Final Staff Position on FAS 157-3, were

effective upon issuance, including prior periods for which financial

statements have not been issued.

discount rates is acceptable when relevant observableinputs into value calculation are not available. Also, itwas made clear that broker quotes may be appropriateinput when measuring fair value. These announced inter-pretations of FAS 157 were seen to provide banks withmore discretion in determining the fair value of securitiesand to enable them to limit markdowns in the face ofilliquid securities markets during the US mortgage defaultcrisis. In this subsection, we assess how this change inaccounting rules that increased accounting discretionaffected the valuation of banks by studying the immediatestock price reaction to the announcement of this rulechange.

The impact of the announced accounting rule changeon bank stock prices is theoretically ambiguous. Theimplied enhanced discretion per se may be valued posi-tively or negatively by investors, depending on whetherinvestors think these particular rule changes will lead tomore or less informative bank accounting. The rulechange, purposely, was made at a time of great stressfor banks, and its application at this time is likely to leadto a revaluation of bank assets. Such a revaluation shouldbe valued positively by bank investors, as it made it easierfor a bank to maintain capital solvency.

We use a standard event study methodology to com-pute the average price effect on bank shares of theannouncement of a change in accounting rules. Also, weassess whether the share prices of different types of banksreacted differently to the announcement. Average cumu-lative abnormal returns are reported for the entire sampleand for different subsamples of banks, with sample splitsbased on relevant bank characteristics. Specifically, weconsider sample splits by the quarterly sample medians ofthe bank’s total assets, the fraction of MBS in total assets,and the potential maximum revaluation gain that wouldaccrue to the bank if the rule changes would allow thebank to value all its MBS at amortized cost. These samplesplits are meant to capture the potential of differentgroups of banks to benefit from asset revaluation. Weuse third quarter 2008 Call report data to construct thebank-specific variables, while daily total equity returndata are obtained from Datastream.

Table 9 reports the event study results. Cumulativeabnormal returns are based on a market model withestimation window [t�250, t�30], where t denotesOctober 10, 2008, and time is counted in trading days.We use the total return on the S&P 500 as a proxy for thedaily market return. We report results for two differentevent windows. Panel A reports results using an eventwindow of (t�3, tþ2], while Panel B reports results usingan event window of (t�1, t]. Using such a short eventwindow of a single day is acceptable given the high stockmarket volatility around the time of this event. Tomitigate concerns that returns from illiquid firms aredriving the result, we exclude from the sample observa-tions of firms with more than 100 zero returns over theestimation window or a zero return on the event date.

The average cumulative abnormal return (CAR) is largeand significantly different from zero at 1% on the eventday itself, but the average CAR across all banks is muchlower and statistically significant at 10% if we extend the

Table 9Event study of new FASB rules on fair value accounting of illiquid assets (FAS 157) announced on October 10, 2008.

This table reports average cumulative abnormal returns (CARs) for the overall sample and for different subsamples of firms. CARs are based on a market

model with estimation window of [t�250, t�30], where t denotes October 10, 2008, and time is counted in trading days. Panel A reports results using an

event window of (t�3, tþ2], where t denotes October 10, 2008, and time is counted in trading days, while Panel B reports results using an event window

of (t�1, t]. Sample consists of US bank holding companies that are stock-listed. Observations from firms with more than 100 zero returns over the

estimation window or a zero return on the event date are excluded from the sample. Large (small) denotes firms with total assets above (below) the

quarterly sample median. High (Low) share of MBS denotes firms with mortgage-backed securities as a fraction of total assets above (below) the

quarterly sample median. High (Low) revaluation gain denotes firms with the difference between available-for-sale MBS at amortized value and

available-for-sale MBS at fair value as a fraction of total assets above (below) the quarterly sample median. Standard errors of the average CARs are

reported in parentheses. We also report the p-value of a t-test of coefficient difference between subsamples. Standard errors are in parentheses. nnn,nn,

And n denote significance at the 1%, 5% and 10% level, respectively.

Panel A: Event window is October 8, 2008 until October 12, 2008

(1) (2) (3) (4) (5) (6) (7)

All firms Large Small High share

of MBS

Low share

of MBS

High revaluation

gain

Low revaluation

gain

CAR 0.0128n 0.0260nnn�0.0005 0.0140 0.0116 0.0090 0.0168n

(0.0070) (0.0092) (0.0105) (0.0087) (0.0111) (0.0100) (0.0099)

Observations 270 136 134 136 134 136 134

Test of coefficient difference – 0.059n 0.865 0.581

Panel B: Event window is October 10, 2008

(1) (2) (3) (4) (5) (6) (7)

All firms Large Small High share of MBS Low share of MBS High revaluation gain Low revaluation gain

CAR 0.0761nnn 0.1290nnn 0.0225nn 0.0903nnn 0.0616nnn 0.0914nnn 0.0605nnn

(0.0074) (0.0079) (0.0107) (0.0094) (0.0113) (0.0102) (0.0104)

Observations 270 136 134 136 134 136 134

Test of coefficient difference – 0.000nnn 0.051n 0.035nn

H. Huizinga, L. Laeven / Journal of Financial Economics 106 (2012) 614–634632

event window. The reason is that October 10, 2008 wasthe only day that week that the stock market experiencedpositive returns in what otherwise was a rapidly fallingmarket, especially for bank stocks.

The sample splits reveal a number of interestingdifferences in the valuation effect across different typesof banks. The CAR of large banks is consistently higher andeconomically large. An explanation for this result is thatlarger US banks tend to have a larger fraction of hard-to-value assets, including off-balance sheet items, and thesebanks thus tend to benefit most from the enhancedcapability to revalue assets. Next, the share price of bankswith a large fraction of MBS also reacts positively andsignificantly to the relaxation of fair value accounting inPanel B, reflecting that banks with sizeable MBS similarlyhave considerable scope for revaluation. Finally, we findthat the share prices of banks that stand to potentially gainthe most from a revaluation of MBS at amortized costreact most positively and significantly to the relaxation offair value accounting in Panel B. These results suggest thatthe market recognized the accounting rule change inOctober 2008, and valued the associated enhancedaccounting discretion in valuing distressed MBS positively.

7. Conclusions

In 2008, the market value of bank assets was lowerthan their book value for the majority of US banks. This isprima facie evidence that the book value of US banks wasinflated. Consistent with this, we find that the market

discounts the value of banks’ real estate loans and MBSrelative to book values. The discrepancy between marketand book values suggests that banks have been slow toadjust book values to reflect market expectations aboutfuture asset losses.

A large market discount on real estate loans goes along way toward explaining the US banking problems of2008, as real estate loans constitute just over half of theaverage bank’s assets. We further find a larger discountfor held-to-maturity MBS (carried at amortized cost) thanfor available-for-sale MBS (carried at fair value), suggest-ing that fair values recognize the impairment of MBS to agreater extent and more quickly than amortized costs do.

The slowness of book values to reflect changes inmarket expectations does not merely reflect the rigidityof financial reporting rules, but it in part reflects theactive use by banks of accounting discretion to preventbook value deterioration. Specifically, banks with largeMBS exposures systematically report lower loan lossprovisioning rates in 2008 to inflate asset values and bookcapital. At the same time, banks with large real estateexposures tend to classify more of their MBS as held-to-maturity, to be able to carry these assets at higheramortized cost. The amount of MBS classified as held-to-maturity further reflects the valuation gap betweenamortized cost and fair value, as evidence that banksclassify MBS to boost asset values.

Our finding that distressed banks tend to exploit theirdiscretion over loan loss provisioning and classification ofMBS to boost their book value should be reason for

Table A1Variable definitions and data sources.

Variable Definition Source

Tobin’s q Ratio of market value of common equity plus book value of preferred

equity and liabilities to book value of assets

Call report and Datastream

Real estate loans Ratio of real estate loans to assets Call report

Non-real estate loans Ratio of non-real estate loans to assets Call report

MBS Ratio of MBS to assets. Held-to-maturity securities are at amortized cost

and available-for-sale securities are at fair value

Call report

Non-MBS securities Ratio of non-MBS securities to assets. Securities held-to-maturity are at

amortized cost and securities available-for-sale at fair value

Call report

MBS, held Ratio of MBS that are held-to-maturity to assets Call report

MBS, for sale Ratio of MBS that are available-for-sale to assets Call report

MBS, held, nonguaranteed Ratio of nonguaranteed MBS that are held-to-maturity to assets Call report

MBS, held, guaranteed Ratio of MBS that are held-to-maturity and issued or guaranteed by FNMA,

FHLMC, or GNMA to assets

Call report

MBS, for sale, nonguaranteed Ratio of nonguaranteed MBS that are available-for-sale to assets Call report

MBS, for sale, guaranteed Ratio of MBS that are available-for-sale and issued or guaranteed by FNMA,

FHLMC, or GNMA to assets

Call report

Low valuation Dummy variable that equals one if Tobin’s q is less than one, and zero

otherwise

Call report

Trading Ratio of assets in trading account to total assets Call report

Tier 1 Tier 1 capital ratio Call report

Share of Tier 1 Ratio of Tier 1 capital in total capital Call report

Leverage Ratio of liabilities to assets Call report

Risk-weighted assets Ratio of risk-weighted assets to assets Call report

Liquid assets Ratio of holdings of cash and US Treasury securities to assets Call report

Fixed assets Ratio of premises and fixed assets to assets Call report

Loan loss provisioning Ratio of loan loss provisioning to loans in % Call report

Net charge-offs Ratio of loan charge-offs minus recoveries to loans in % Call report

Share of real estate loans Share of real estate loans in total loans Call report

CEO bonus share Share of bonus payments in total compensation of the chief executive

officer

Compustat’s ExecuComp

CEO ownership Fraction of shares in the firm held by the chief executive officer Compustat’s ExecuComp

MBS amortized Ratio of MBS to assets. Both held-to-maturity and available-for-sale MBS

are at amortized cost

Call report

MBS difference Difference between MBS valued at amortized cost and MBS valued at fair

value divided by total assets

Call report

H. Huizinga, L. Laeven / Journal of Financial Economics 106 (2012) 614–634 633

concern, as it implies that the discretion implicit incurrent financial reporting rules leads to systematicbiases in valuations on bank balance sheets. Accountingdiscretion enables banks with impaired asset portfolios tosatisfy capital adequacy requirements, but makes it diffi-cult to assess the true health of the affected banks. Insetting loan loss provisions, banks no doubt make use ofprivate information about the prospects of loan repay-ment. This makes some discretion over loan loss provi-sioning generally beneficial in that private informationabout loan quality is revealed. Bank classification of MBS,in contrast, does not serve the purpose of revealingprivate information about MBS quality, while it carriesthe cost of enabling banks to alter the book value of theirassets and their regulatory capital. Our empirical evidencesuggests that changes in classification behavior are moti-vated to boost bank asset values. This suggests that inpractice, FAS 115 is frequently violated, with or withoutregulatory approval, and that regulators have covertlyextended classification options to at least some US banksto boost regulatory capital.

The rationale for reclassifications would disappear if allsecurities are carried at fair value. More generally, repla-cing the mixed attribute model of accounting with a modelbased entirely on fair value accounting will mitigateincentives for accounting arbitrage and could serve to

improve the information value of financial reports, evenif fair value calculations are subject to discretion. Similarly,a more forward-looking approach to provisioning for badloans on an expected loss basis could improve the informa-tion content of banks’ financial reports.

No regulatory system for banks will be perfect. Valua-tion models can be misused or misinterpreted. But rea-sonable and auditable methods exist today to incorporateinformation embedded in market prices when disclosingthe value of financial assets and associated expectedlosses. For example, Hart and Zingales (2011) propose touse observable market prices on credit default swaps toidentify bank vulnerabilities. More reliable financialreports are beneficial to regulatory and market disciplineand could potentially have helped to avoid some of thelosses that banks currently face.

Appendix

See Table A1.

References

Adrian, T., Shin, H., 2010. Liquidity and leverage. Journal of FinancialIntermediation 19, 418–437.

H. Huizinga, L. Laeven / Journal of Financial Economics 106 (2012) 614–634634

Ahmed, A., Takeda, C., Thomas, S., 1999. Bank loan loss provisions: areexamination of capital management, earnings management andsignaling effect. Journal of Accounting and Economics 28, 1–25.

Boudoukh, J., Richardson, M., Stanton, R., Whitelaw, R., 1997. Pricingmortgage-backed securities in a multifactor interest rate environ-ment: a multivariate density estimation approach. Review of FinancialStudies 10, 405–446.

Brown, C., Dinc, S., 2011. Too many to fail? Evidence of regulatoryforbearance when the banking sector is weak. Review of FinancialStudies 24, 1378–1405.

Caballero, R., Hoshi, T., Kashyap, A., 2008. Zombie lending and depressedrestructuring in Japan. American Economic Review 98, 1943–1977.

Calomiris, C., Mason, J., 2003. Consequences of U.S. bank distress duringthe depression. American Economic Review 93, 937–947.

Calomiris, C., Wilson, B., 2004. Bank capital and portfolio management:the 1930s ‘‘capital crunch’’ and the scramble to shed risk. Journal ofBusiness 77, 421–456.

Collins, J., Shackelford, D., Wahlen, J., 1995. Bank differences in thecoordination of regulatory capital, earnings, and taxes. Journal ofAccounting Research 33, 263–291.

Coval, J., Jurek, J., Stafford, E., 2009. The pricing of investment gradecredit risk during the financial crisis. Unpublished working paper.Harvard Business School and Princeton University.

Dell’Ariccia, G., Igan, D., Laeven, L., 2012. Credit booms and lendingstandards: evidence from the subprime mortgage market. Journal ofMoney, Credit, and Banking 44, 367–384.

Demyanyk, Y., Van Hemert, O., 2011. Understanding the subprimemortgage crisis. Review of Financial Studies 24, 1848–1880.

Diamond, D., Rajan, R., 2011. Fear of fire sales, illiquidity seeking, and thecredit freeze. Quarterly Journal of Economics 126, 557–591.

Dunn, K., Singleton, K., 1983. An empirical analysis of the pricing ofmortgage-backed securities. Journal of Finance 38, 613–623.

Ellul, A., Yerramilli, V., 2010. Stronger risk controls, lower risk: evidencefrom U.S. bank holding companies. Unpublished working paper.Indiana University at Bloomington and University of Houston.

Fahlenbrach, R., Stulz, R., 2011. Bank CEO incentives and the credit crisis.Journal of Financial Economics 99, 11–26.

Flannery, M., Sorescu, S., 1996. Evidence of bank market discipline insubordinated debenture yields: 1983–1991. Journal of Finance 51,1347–1377.

Goh, B., Ng, J., Yong, K., 2009. Market pricing of banks’ fair value assetsreported under SFAS 157 during the 2008 economic crisis. Unpub-lished working paper. Singapore Management University and MIT.

Gonzalez, L., James, C., 2007. Banks and bubbles: how good are banks atspotting winners? Journal of Financial Economics 86, 40–70.

Gorton, G., 2009. Information, liquidity, and the (ongoing) panic of 2007.American Economic Review 99, 567–572.

Hale, G., Santos, J., 2009. Do banks price their information monopoly?Journal of Financial Economics 93, 185–206.

Hart, O., Zingales, L., 2011. A new capital regulation for large financialinstitutions. American Law and Economics Review 13, 453–490.

Hutton, A., Marcus, A., Tehranian, H., 2009. Opaque financial reports, R2,and crash risk. Journal of Financial Economics 94, 67–86.

James, C., 1987. Some evidence on the uniqueness of bank loans. Journalof Financial Economics 19, 217–235.

James, C., 1991. The losses realized in bank failures. Journal of Finance46, 1223–1242.

Kane, E., 1989. The S&L Insurance Mess: How Did It Happen?. UrbanInstitute, Washington, DC.

Karpoff, J., Lee, S., Martin, G., 2008. The cost to firms of cooking thebooks. Journal of Financial and Quantitative Analysis 43, 581–612.

Kedia, S., Philippon, T., 2009. The economics of fraudulent accounting.Review of Financial Studies 22, 2169–2199.

Keys, B., Mukherjee, T., Seru, A., Vig, V., 2009. Financial regulation andsecuritization: evidence from subprime loans. Journal of MonetaryEconomics 56, 700–720.

Keys, B., Mukherjee, T., Seru, A., Vig, V., 2010. Did securitization lead tolax screening? Evidence from subprime loans. Quarterly Journal ofEconomics 125, 307–362.

Kroszner, R., Strahan, P., 1996. Regulatory incentives and the thrift crisis:dividends, mutual-to-stock conversions, and financial distress. Journalof Finance 51, 1285–1319.

Laeven, L., Levine, R., 2007. Is there a diversification discount in financialconglomerates? Journal of Financial Economics 85, 331–367.

Laeven, L., Levine, R., 2009. Bank governance, regulation and risk taking.Journal of Financial Economics 93, 259–275.

Laeven, L., Majnoni, G., 2003. Loan loss provisioning and economicslowdowns: too much, too late. Journal of Financial Intermediation12, 178–197.

Lang, L., Stulz, R., 1994. Tobin’s q, corporate diversification, and firmperformance. Journal of Political Economy 102, 1248–1280.

Laux, C., Leuz, C., 2010. Did fair-value accounting contribute to thefinancial crisis? Journal of Economic Perspectives 24, 93–118.

Leuz, C., Nanda, D., Wysocki, P., 2003. Investor protection and earningsmanagement: an international comparison. Journal of FinancialEconomics 69, 505–527.

Loutskina, E., Strahan, P., 2009. Securitization and the declining impactof bank finance on loan supply: evidence from mortgage origina-tions. Journal of Finance 64, 861–889.

Mian, A., Sufi, A., 2009. The consequences of mortgage credit expansion:evidence from the U.S. mortgage default crisis. Quarterly Journal ofEconomics 124, 1449–1496.

Moyer, S., 1990. Capital adequacy ratio regulations and accountingchoices in commercial banks. Journal of Accounting and Economics13, 123–154.

Peek, J., Rosengren, E., 2000. Collateral damage: evidence from theJapanese banking crisis on real activity in the United States. AmericanEconomic Review 90, 30–45.

Ryan, S., 2008. Accounting in and for the subprime crisis. AccountingReview 83, 1605–1638.

Sachs, J., Huizinga, H., 1987. U.S. commercial banks and the developingcountry debt crisis. Brookings Papers on Economic Activity 2,555–606.

Shrieves, R., Dahl, D., 2003. Discretionary accounting and the behavior ofJapanese banks under financial duress. Journal of Banking andFinance 27, 1219–1243.

Skinner, D., 2008. The rise of deferred tax assets in Japan: the case of themajor Japanese banks. Journal of Accounting and Economics 46,218–239.