Preserving Amortized Cost within a Fair-Value-Accounting Framework Reclassification of Gains and Losses on Available-for-Sale Securities upon Realization

115 requires firms to record available-for-sale “AFS” securities on the balance sheet at fair value, with accumulated unrealized gains and losses “AUGL” recorded in accumulated other co

Preserving Amortized Cost within a Fair-Value-Accounting Framework:

Reclassification of Gains and Losses on Available-for-Sale Securities upon Realization

University of California, Berkeley

First draft: December 2008Current draft: October 2010(Revision in Process, for Columbia Burton Conference Participants Only:

Please Do Not Cite without Permission)

† Universite de Lausanne, Faculty of Business and Economics, Quartier UNIL-Dorigny,Baliment Internef Bureau 595, CH-1015 Lausanne, Switzerland, (41)216923367 ‡ Stern School

of Business, Kaufman Management Center, 44 West 4th Street, New York, NY 10012.(01)2129980020 * 545 Student Services Building #1900, Berkeley, CA 94720, USA,(01)5106424789 Comments and suggestions from Anne Cristine dArcy, Jonathan Glover,Pierre Liang, Jack Stecher, Danqing Yang and seminar participants at Carnegie MellonUniversity, Chinese University of Hong Kong, and University of Lausanne are gratefullyacknowledged We thank Jialu Shan and Joseph Cadora for research and editorial assistance

Preserving Amortized Cost within a Fair-Value-Accounting Framework:

Reclassification of Gains and Losses on Available-for-Sale Securities upon Realization

ABSTRACT: SFAS No 115 requires firms to record available-for-sale (“AFS”) securities on

the balance sheet at fair value, with accumulated unrealized gains and losses (“AUGL”) recorded

in accumulated other comprehensive income (“AOCI”), a component of owners’ equity Firms reclassify AUGL to net income when they realize gains and losses either economically through sale of AFS securities or for accounting purposes through transfer of the securities to trading or other-than-temporary impairment write-downs We refer to the amount of this reclassification each period as “RECLASS.”

For a sample of 200 large U.S commercial banks from 1998-2006, we examine the incremental value relevance of RECLASS beyond AUGL and other components of book value of equity and comprehensive income We find that the market value of equity is significantly positively associated with RECLASS, with the coefficient on RECLASS being closer to the coefficient on the relatively permanent net income before extraordinary items and discontinued operations than

to the much lower coefficients on the remaining more transitory components of comprehensive income We also find that the coefficient on RECLASS is much higher than the coefficients on AUGL and other components of book value This result obtains despite also finding that the coefficient on AUGL is significantly positive in a pure balance-sheet model and higher than the coefficient on the remainder of book value in a combined balance-sheet/comprehensive-income model, consistent with investors placing at least a normal amount of credence in AUGL

We conduct three analyses investigating possible explanations for the incremental value

relevance of RECLASS First, we consider the possibility that unrealized gains and losses are unreliable, as opponents of fair value accounting often allege Contradicting this possibility, we

find that the coefficient on RECLASS is higher and more significant for banks that hold more

liquid securities Second, we consider the possibility that RECLASS interacts with or is an indicator of future bank growth Consistent with this possibility, we find that the incremental value relevance of RECLASS is greater for higher growth banks Third, and further supporting this possibility, we find that RECLASS is significantly positively associated with one-year-aheadcomprehensive income, controlling for other components of current book value and

comprehensive income, more so for banks holding liquid securities and growing faster

Overall, our findings suggest that the value relevance of RECLASS is primarily attributable tothe importance of the realization of realized gains and losses as an indicator of bank growthrather than to the limitations of fair value accounting for AFS securities At a minimum, thesefindings suggest that the FASB should continue to require information about realized gains andlosses, an amortized cost accounting construct, within the fair value accounting framework forAFS securities

Key words: Available-for-sale securities; Reclassification; Fair value accounting; Realization

principle

1 INTRODUCTION

In this paper, we examine the incremental value-relevance of realized gains and lossesbeyond unrealized gains and losses and other components of book value and comprehensiveincome for commercial banks’ available for sale (“AFS”) securities We focus on AFS securitiesbecause they are reported at fair value on the balance sheet, but amortized cost information aboutrealized gains and losses is preserved and reported on the income statement through the use of

“dirty surplus” accounting described below The financial reporting for AFS securities contrastswith the typical accounting for financial instruments under current U.S GAAP, in which one offair value and amortized cost information is reported only in footnote disclosures or not at all.Prior research generally shows that investors react more strongly to recognized than disclosedinformation, either because they do not have the ability or inclination to evaluate the manydisclosures in financial reports or because they deem recognized amounts more reliable(Schipper 2007) Hence, AFS securities constitute a relatively unambiguous setting in which totest for the incremental value relevance of information about the fair values and amortized costs

of financial instruments

Statement of Financial Accounting Standards (“SFAS”) No 115, Accounting for Certain

Investments in Debt and Equity Securities, requires a hybrid

fair-value-on-the-balance-sheet/amortized-cost-on-the-income-statement approach to accounting for AFS securities Specifically, firms record AFS securities on the balance sheet at fair value, with accumulated unrealized gains and losses (“AUGL”) recorded in accumulated other

comprehensive income (“AOCI”), a component of owners’ equity distinct from retained

earnings This is dirty surplus accounting because changes in owners’ equity occur without corresponding changes on net income Subsequently, firms reclassify AUGL to net income

when gains and losses are realized economically through sale of AFS securities or for accountingpurposes through transfer of the securities to trading or other-than-temporary (“OTT”)

impairment write-downs We refer to the amount of this reclassification each period as

“RECLASS.”

Advocates of fair value accounting often criticize this accounting for AFS securities aspolitically motivated and convoluted, particularly for liquid securities for which fair value is themost relevant, comprehensive, and timely measure of the value of the securities.1 However, thisaccounting has the desirable feature of preserving certain aspects of amortized cost accountingfor AFS securities—particularly the realization of gains and losses reported on the incomestatement via RECLASS—within a primarily fair value accounting framework Even if fairvalues are well measured, amortized costs may be incrementally value relevant beyond fairvalues for various reasons The FASB recognizes this fact in its May 2010 Exposure Draft,

Accounting for Financial Instruments and Revisions to the Accounting for Derivative Instruments and Hedging Activities (“the ED”), in which it observes that amortized costs may

have incremental value relevance due to their verifiability, association with contractual cashflows, or correspondence with the firm’s business strategy Consistent with this observation, inthe ED the FASB proposes dual presentation on the balance sheet of the amortized costs and fairvalues of many financial instruments

The FASB’s proposal is consistent with at least two positions expressed by opponents offair value accounting First, most of these opponents question the reliability of fair values (e.g.,Wallison 2008, Forbes 2009) In our view, reliability is a minor concern for most of banks’ AFSsecurities, which are dominated by federal governmental, government-sponsored agency (e.g.,

1 For example, in testimony before the Committee on Banking, Housing, and Urban Affairs, U.S Senate, September

10, 1990, former Securities and Exchange Commission (SEC) Chairman Richard Breeden advocated a move to fair value accounting, stating that market-based information is more relevant than cost-based information.

Fannie Mae), and other liquid securities, although it is a significant concern for some of these(e.g., structured asset-backed) securities More interestingly, some opponents of fair valueaccounting point out that realization of gains and losses is important for various purposes, such

as contracting and stewardship assessment (Watts 1993 and Holthausen and Watts 2001), capitalregulation (Moyer 1990 and Ahmed and Takeda 1995), other aspects of firms’ businessstrategies (Nissim and Penman 2008), and managerial signaling of their private information(Abdel-Khalik 2008 and Ronen 2008)

In this study, we provide evidence regarding the benefit of preserving amortized cost information within a fair value accounting framework in the specific context of the value

relevance of RECLASS RECLASS constitutes a somewhat more limited and less visible preservation of amortized costs in a fair value accounting framework than the FASB’s proposal

in the ED to require dual presentation on the balance sheet of the amortized costs and fair values

of many financial instruments However, our examination of the value relevance of RECLASS ismeaningful because this variable embodies the realization principle that is central to amortized cost accounting

We hand collected RECLASS for the 200 largest publicly traded U.S commercial banks(based on total assets in 1998) for the period 1998-2006 We limit our sample to banks becausetotal (realized) gains and losses on AFS securities often constitute significant portions of theirowners’ equity (net income) Our sample period necessarily begins in fiscal year 1998, when

FAS No 130, Reporting Comprehensive Income, first required firms to disclose RECLASS in a

Our primary findings are as follows First, we find that banks’ market values aresignificantly positively associated with AUGL in a pure balance sheet model and higher than thecoefficient on the remainder of book value in a combined balance-sheet/comprehensive-incomemodel, consistent with investors placing at least a normal amount of credence in AUGL Thissuggests that the incremental value relevance of RECLASS that we document in this paper is notsolely attributable to RECLASS remedying the unreliability of the fair value accounting-basedAUGL

Second, we find that investors ascribe considerable incremental value relevance toRECLASS, controlling for the other components of book value and comprehensive income.Specifically, we find that the market value of equity is significantly positively associated withRECLASS, with the coefficient on RECLASS being closer to the coefficient on the relativelypermanent net income before extraordinary items and discontinued operations (NIBEX) than tothe much lower coefficients on the more transitory components of comprehensive income Thecoefficient on RECLASS is also much higher than the coefficients on AUGL and othercomponents of book value These findings imply that stock investors ascribe considerable

significance to the realization of gains and losses recorded in RECLASS These findings obtain

even though RECLASS effectively just reclassifies one component of owners’ equity, AOCI, toanother, retained earnings Our remaining primary findings pertain to analyses that we conduct

to investigate possible explanations for the significance of the realization of gains and lossesrecorded in RECLASS

Third, we consider the possibility that unrealized gains and losses are unreliable, asopponents of fair value accounting often allege This possibility is inconsistent with ouraforementioned findings for AUGL and with the high liquidity of most AFS securities Directly

contradicting this possibility, we find that the incremental value relevance of RECLASS is

stronger for banks that hold more liquid securities This implies that this incremental value relevance is not primarily attributable to the lack of verifiability of fair values

Fourth, we consider the possibility that RECLASS interacts with or is an indicator offuture bank growth Consistent with this possibility, we find this incremental value relevance islarger for higher growth banks This suggests RECLASS conveys incremental information aboutfuture valuation attributes, which have greater valuation consequences for higher growth firms

Fifth, consistent with this suggestion, we find that RECLASS is significantly positivelyassociated with year-ahead comprehensive income, controlling for the other components of bookvalue and comprehensive income Consistent with the valuation model results, this association isstronger for banks holding more liquid securities and growing faster

Sixth, we consider the possibility that investors undervalue unrealized gains and losses,say because they fixate on reported net income We conduct regression analyses controlling forFama and French’s (1992, 1993) three factors and stock return momentum (Jegadeesh andTitman 1993) as well as portfolio analyses grouping banks into terciles each year based on theamount of unrealized gains and losses Both analyses yield weak statistical evidence that bankswith higher unrealized gains and losses experience economically modest higher future excessreturns In the portfolio analysis, we find that the return drift is stronger when excess returns forthe tercile observations are value-weighted rather than equally weighted, implying that themarket under-reaction is stronger for larger banks While there are various possible explanationsfor this finding, we conjecture it may be attributable to the greater difficulty that investors face inevaluating unrealized gains and losses for larger banks, given their more diverse holdings of AFSsecurities and greater overall complexity Given the statistically weak and economically modest

return drift, investor mispricing appears to explain at most a small portion of the value-relevance

of RECLASS

To our best of our knowledge, our paper is the first to document the value relevance ofRECLASS or the reclassification of any other component of AOCI Our results haveimplications for the ongoing debate about the relative and incremental usefulness of fair valueversus amortized cost accounting for financial instruments Our findings collectively support theFASB’s view expressed in the ED that it is incrementally value relevant to preserve amortizedcost information based on the realization principle within a fair value accounting framework

2. Prior Research

Our study is primarily related to two areas of prior research: (1) studies on theincremental and relative value relevance of the fair values versus amortized costs of financialinstruments and (2) studies on the incremental and relative value relevance of othercomprehensive income versus net income We discuss these two literatures in turn

Advocates of fair value accounting generally claim that the fair values of financialinstruments have higher or incremental value relevance compared to the amortized costs of theinstruments Numerous empirical studies have tested this claim, typically using disclosed fair

values under SFAS No 107, Disclosures about Fair Values of Financial Instruments While

generally supportive of this claim, the results of these studies vary somewhat depending the typeand liquidity of the financial instruments considered, the type of firms involved, as well asaspects of the research design such as the use of levels versus first differences valuation models.These studies often but not always find that market values are significantly incrementallyassociated with fair values controlling for amortized costs, but not vice-versa

The most related prior study to ours, Barth (1994), examines the value relevance ofdisclosures about the fair values of banks’ investment securities.3 Barth estimates levels and firstdifferences models in annual cross-sectional regressions and pooled regressions with fixedeffects In her levels model, the market value of equity is regressed on the book value of equityand the fair value and amortized costs of marketable securities The levels model estimationyields a highly significantly positive coefficient on the fair value of marketable securities and aninsignificant or negative significant coefficient on the amortized cost of marketable securities.Barth concludes that the fair values of marketable securities provide significant explanatorypower beyond amortized costs, but not vice versa In her first differences model, returns isregressed on net income before securities gains and losses (alternatively, the change in that netincome measure) and realized gains and losses and total gains and losses The first differencesmodel estimation yields a negative coefficient on realized gains and losses and a positivecoefficient on total gains and losses that usually is insignificant except for large and firmsholding liquid securities Barth interprets the weaker result for the income statement variables inthe first differences model as attributable to greater noise in these variables

Barth’s (1994) results suggest that RECLASS should have little value relevance, partlybecause it is an amortized cost number and partly because of the aforementioned noise issues.However, our study differs from Barth (1994) on two important dimensions First, we useamortized costs, fair values, and gains and losses—in particular RECLASS—that are recognized

in financial statements rather than simply disclosed One explanation for Barth’s weak results forgains and losses is investors put less weight on disclosures rather than recognized amounts.Second, our levels valuation models include considerably more extensive breakdowns of book

3 Barth (1994) hand collected the fair values of marketable securities for a sample of banks from 1970-1990, a period entirely prior to the issuance of SFAS No 107 During this period, banks appear to have disclosed the fair values of marketable securities under industry GAAP or practice

value, net income, and other comprehensive income, consistent with subsequently developedcombined balance sheet and income statement valuation models (e.g., Ohlson 1995)

Barth, Beaver, and Landsman (1996), Nelson (1996), and Eccher, Ramesh, andThiagarajan (1996) examine the incremental value relevance of the fair values of essentially allbanks’ financial instruments—such as loans, deposits, and debt—which they disclose underSFAS No 107 The results of the three studies differ somewhat due to their differing modelspecifications This is particularly true for loans, for which only Barth, Beaver, and Landsmanfind fair values to be incrementally value relevant By and large, however, these studies find thatthe fair value of most financial instruments are incrementally value relevant beyond theamortized costs of those instruments

Studies investigating the value relevance of comprehensive income and the components

of other comprehensive income items generally find that comprehensive income and its

components are less value relevant prior to effective date of SFAS No 130 than afterwards, consistent with the greater salience to investors of amounts recognized in financial statements rather than simply disclosed For a sample prior to the effective date of SFAS No 130,

Dhaliwal, Subramanyam, and Trezevant (1999) find that comprehensive income does not have a stronger association with stock returns than net income, except for financial firms They also find that the AFS securities adjustment is the only component of other comprehensive income that improves the association between income and returns, again primarily for financial firms O’Hanlon and Pope (1999) report similarly negative results for other comprehensive income items for a sample of U.K firms

For samples after the effective date of SFAS No 130, Biddle and Choi (2006) find that comprehensive income dominates other income measures in explaining equity returns

Chambers, Linsmeier, Shakespeare, and Sougiannis (2007) find that other comprehensive

income is priced approximately dollar-for-dollar by investors

3 HYPOTHESIS DEVELOPMENT AND RESEARCH DESIGN

3.1 RECLASS and the Mechanics of Recyling Gains and Losses

Under SFAS No 115, firms initially record aftertax unrealized gains and losses on AFSsecurities in accumulated other comprehensive income (AOCI), a component of owners’ equity.Accumulated unrealized gains and losses (AUGL) remain in AOCI until one of the followingevents occurs involving the securities:

a. Sale,

b. Transfer to trading, or

c. Other-than-temporary (OTT) impairment write-downs

When one of these events occurs, the related AUGL is reclassified to net income (with pretaxrealized gains and losses and tax effects recorded on separate line items on the income statement)and thence to retained earnings We refer to the aftertax reclassification as RECLASS.RECLASS changes the components but not the total amounts of owners’ equity andcomprehensive income As a result, RECLASS often is referred to as “recycling” amountspreviously recognized on the balance sheet onto the income statement

The following equations illustrate the mechanics of this recycling All amounts areaftertax in these equations Net income (NI) during a period equals net income before realizedgains and losses (NIBRGL) plus RECLASS

(1)

The change in retained earnings (RE) during a period equals NI minus dividends (DIV) duringthe period, which equals NIBRGL plus RECLASS minus DIV during the period:

ΔREREt = NIt – DIVt = NIBRGLt + RECLASSt – DIVt (2)

Hence, RE increases with RECLASS

The change in AUGL during a period equals the unrealized gain or loss (UGL) during theperiod UGL equals the total (i.e., unrealized plus realized) gain or loss (TGL) minus RECLASSduring the period

ΔREAUGLt = UGLt = TGLt - RECLASSt

(3)

Hence, AUGL decreases with RECLASS

The change in owners’ equity (OE) equals ΔRERE plus ΔREAOCI during the period ΔRERE isgiven in equation (2) and ΔREAOCI equals ΔREAUGL plus other comprehensive income fromsources other than AFS securities (OCIother) during the period, yielding

ΔREOEt = ΔREREt + ΔREAOCIt

= NIBRGLt + RECLASSt - DIVt + TGLt - RECLASSt + OCIother

3.1. Valuation Model and Related Hypotheses

Because Barth (1994) finds considerable noise in first differences valuations models, ourprimary analyses employ levels valuation models.4 We estimate several variants of the followingmodel that includes both balance sheet and comprehensive income statement information:

(5)

where MVt denotes the market value of owners’ equity, BVt denotes the book value of owners’ equity, and CIt denotes comprehensive income Ohlson (1995) derives a similar model from three assumptions: (1) dividend discounting with a constant cost of equity r, (2) clean surplus accounting, and (3) a first order autoregressive process for abnormal comprehensive income.5 Ohlson shows that perfectly transitory (e.g., fair value accounting based) abnormal

comprehensive income yields a pure balance sheet model with β1=1 and β2=0 He shows that perfectly permanent abnormal comprehensive income yields a pure income statement model with

β1=0 and, assuming a constant rate of dividend payout of comprehensive income λ, β2=(1+r)/r-λ.6

Our regression models begin with equation (5) and decompose both BV and CI intocomponents Specifically, we decompose BV into the aftertax book value of AFS securities(BVafs) plus the aftertax book value of other net assets (BVother) We further decompose BVafs

4 We did estimate a returns model derived directly from equation (6) below as the change in MV plus dividends, with all model variables deflated by beginning price Consistent with Barth’s (1994) findings, this returns model yields similar but weaker results than our levels valuation model For example, we obtain a less positive and significant coefficient on ΔRERECLASS in this model—3.76 significant at the 5% level—than for the corresponding coefficient on RECLASS in our primary levels valuation model reported in column III of Table 3—8.10 significant

at the 1% level

5 Equation (7) in Ohlson (1995) differs in four primary respects from our equation (5) First, his comprehensive income term is a capitalization of comprehensive income reduced by dividends Second, the weights on the book value and income terms are inversely related Third, he allows for non-accounting information Fourth, his equation does not include an intercept

6 If dividend payout is not constant, then dividends should be incorporated in Ohlson’s (1995) capitalized

comprehensive income term However, prior research by Hand and Landsman (1998) finds that including dividends

in the Ohlson model empirically appears to capture dividends signaling future income rather than dividend payout

into the amortized cost of AFS securities (COST) plus AUGL We decompose CI into netincome before extraordinary items and discontinued operations (NIBEX), extraordinary itemsand discontinued operations (EX), and other comprehensive income (OCI) We furtherdecompose NIBEX into RECLASS and NIBEXother We further decompose OCI into ΔREAUGLand OCIother

Incorporating these variable decompositions into equation (5) yields our most extensive valuation model:

MVt = α + β1COSTtafs + β2AUGLt + β3BVother

t + β4RECLASSt + β5NIBEXother

We develop five sets of null and alternative hypotheses to investigate whether and whyRECLASS is value relevant The first pertains to whether RECLASS is incrementally valuerelevant beyond AUGL and the other explanatory variables in equation (6) The null hypothesis(H1N) is that RECLASS is not value relevant because the fair value of those securities and thecash received from selling the securities conveys all relevant information, leaving no room forRECLASS to convey incremental information The alternative hypothesis (H1A) is thatRECLASS conveys incremental information due to the importance of realization RECLASS

appears both directly as a separate variable (its effect on NI) as indirectly as a decrease inΔREAUGLt (its effect on OCI) in equation (6) Hence, we take both the direct and indirect effects

of RECLASS into account and state these hypotheses as the restrictions on the “total” coefficient

β4- β7 on RECLASS:7

H1N: β4- β7=0

H1A: β4-β7>0

3.2.Liquidity of AFS Securities

Our second set of hypotheses examines one possible reason for the value relevance ofRECLASS Opponents of fair value accounting often allege that fair values are unreliable Ifthis lack of reliability is the reason for the positive association between MV and RECLASS, thenthis association should be stronger for banks that hold less liquid securities for which fair valueslikely are measured less reliably We again state null and corresponding alternative hypotheses

as restrictions on the total coefficient on RECLASS:

H2N: β4-β7 does not vary with the liquidity of banks’ AFS securities

H2A: β4-β7 is higher for banks holding less liquid AFS securities

7 RECLASS also affects AUGL and, depending on the type of realization of gains and losses, one of COST or

BV other in equation (6) For example, for economic realizations of gains, RECLASS also appears as an increase in

BV other (its effect on cash) and a decrease in AUGL (its effect on AFS securities) However, unlike RECLASS’s direct and indirect effects discussed in the text, these two additional effects would exist in a clean surplus fair value accounting system that does not report RECLASS or otherwise preserve amortized cost information about realized gains and losses We conduct our tests using the total coefficient β 4 -β 7 because it better corresponds to our focus in this paper However, we have also conducted all tests using the alternative total coefficient (β 3 +β 4 )-(β 2 +β 7 ), which yields the same conclusions as for our tests using β 4 -β 7 , albeit with slightly lower significance levels (e.g., 5% instead of 1% in our primary tests reported in column III of Table 3)

3.3 Growth

Our third set of hypotheses examines the possibility that RECLASS is value relevantbecause the realization of gains and losses informs about bank growth For example, this couldoccur if bank managers realize gains and losses to signal future income Any such effect would

be enhanced to the extent that future income is more important for higher growth banks Weagain state null and corresponding alternative hypotheses in terms of the total coefficient onRECLASS:

H3N: β4-β7 does not vary with banks’ growth

H3A: β4-β7 is higher for higher growth banks

3.4.Prediction of Future Comprehensive Income

To investigate the aforementioned possibility that realized gains and losses signal futureincome, we regress year-ahead comprehensive income on the same explanatory variables inequation (6)

CIt+1 = δ + γ1COSTtafs + γ 2AUGLt + γ3BVother

t + γ4RECLASSt + γ5NIBEXother

H5N: TGL is not associated with future excess returns.

H5A: TGL is positively associated with future excess returns

4. SAMPLE, DATA, and DESCRIPTIVE STATISTICS

We obtain most accounting and market value data from Compustat and stock return datafrom CRSP We hand collected AUGL and RECLASS from disclosures of the components of(accumulated) other comprehensive income in banks’ annual reports In the Appendix weprovide several examples of these disclosures

SFAS No 130 required firms to report other comprehensive income items for fiscal yearsbeginning after December 15, 1997 For this reason, our sample period begins in 1998, and it

covers the nine fiscal years through 2006 Because of the time required to hand collect AUGLand RECLASS, we restrict our sample to the 200 largest U.S commercial banks based on totalassets in 1998 We also require the sample banks to be traded on NYSE, AMEX or NASDAQand to have all necessary data on the variables in equation (6) The market capitalization of the

200 banks ranges from $50 million to $230 billion, indicating that our sample selection based onsize is not excessively restrictive Our final sample contains 1,033 bank-year observations withcomplete data on the equation (6) variables

Recall that RECLASS results from three different events: sales of AFS securities, transfer

of the securities to trading, and OTT impairment write-downs of the securities The first eventinvolves economic realization and the second and third events involve realization for accountingpurposes only Analysis of our hand-collected disclosures indicates that only 11 of the samplereclassifications were due to AFS securities being transferred to trading or other-than-temporaryimpairment write-downs, i.e., realization for accounting purposes only.8 The low frequency oftransfers to trading likely results from the proviso in paragraph 15 of SFAS No 115 that suchtransfers should be “rare.” The low frequency of OTT impairment write-downs likely resultsfrom our sample period preceding the financial crisis and from the fact firms generally do notrecord impairments for declines in value driven by movements in interest rates unless they havedecided to sell the affected securities Hence, almost all of the variation in RECLASS reflectsactual sales of securities, i.e., economic realization

Table 1, Panel A reports descriptive statistics for the variables AUGL, TGL, andRECLASS for our sample banks The mean AUGL is 10 cents per share, with a sizeable standarddeviation of 70 cents per share The likely reason why AUGL is positive on average is that

8 Possibly this low frequency reflects nondisclosure by banks due to the immateriality of some realizations for accounting purposes only, because SFAS No 115’s provisions need not be applied to immaterial items

interest rates decreased significantly and fairly steadily for almost two decades prior to oursample period Equity prices also rose considerably, albeit less steadily, over this period

The mean of RECLASS is 3 cents per share, with a standard deviation of 16 cents per share The mean ratio of the absolute value of RECLASS to the absolute value of net income equals 6%, indicating that the realization of gains and losses on AFS securities typically has a sizeable effect on net income The mean of TGL is -2 cent per share, with a standard deviation of

57 cents per share The mean ratio of the absolute value of TGL to the absolute value of net income equals 26%, over four times larger than the corresponding ratio involving RECLASS Hence, total gains and losses are far more variable than realized gains and losses

Table 1, Panel B reconciles the means of the beginning and ending balances of AUGL,TGL and RECLASS for the sample observations for which all four variables are available, asexpressed in equation (3) above To maintain consistency, all variables are all deflated by theend of year number of shares outstanding for the purposes of constructing this panel MeanAUGL decreases by 6 cents per share per year, from 13 cents per share to 7 cents per share Thisdecrease is attributable to mean TGL of negative 3 cents per share and mean RECLASS of 3cents per share The opposite signs of mean TGL and mean RECLASS indicate the disconnectbetween total and realized gains and losses

Table 2 reports two descriptive analyses regressing RECLASS on current NIBEXother andeither current AUGL (Panel A) or TGL for the current and three prior years (Panel B) Thepurpose of these analyses is to provide insight into two issues: (1) whether banks use RECLASS

to manage income—income smoothing would yield a negative association of RECLASS withNIBEXother, while big baths would yield the opposite—and (2) the strength and lag structure ofthe relationship between realized and unrealized gains and losses

The two panels of Table 2 provide consistent results There is evidence of a low level ofincome smoothing through realization of gains and losses, with a small negative coefficient of-0.01 on NIBEXother that is significant at the 1% level in Panel A and 10% level in Panel B.There is also evidence of gradual realization of gains and losses, with statistically significant butrelatively low positive associations between RECLASS and AUGL (coefficient=0.05, significant

at the 1% level) and between RECLASS and current and lagged TGL (coefficients smoothlydeclining from 0.12 for current TGL to 0.05 for three-year-lagged TGL, all significant at the 5%level or better)

5 VALUE RELEVANCE OF RECLASS

Table 3 presents the results of estimating equation (6) To mitigate heteroscedasticity, wedeflate all variables by number of shares outstanding In untabulated robustness tests, wealternatively deflate by total assets, which yields substantively the same results To incorporateclustering of observations by time period and firm, we include year fixed effects and report tstatistics that incorporate firm-level clustering of observations (Petersen 2009)

As benchmarks for the estimation of the full equation (6) in column III of Table 3, wefirst discuss the estimation of two nested versions of the equation: a pure balance sheet model(column I) and a pure comprehensive income statement model (column II) We maintain thesame decompositions of BV or CI in these nested models as in the full equation

The results of estimating the pure balance sheet model reported in column I of Table 3indicate that MV is significantly positively associated with AUGL at the 5% level Moreover,the coefficient of 3.25 on AUGL is over twice as high as the coefficients on COST and BVother; infact, this coefficient is too high for transitory gains and losses As discussed below, this

coefficient decreases to a normal level once RECLASS and the remaining comprehensiveincome variables are added to the model These results are consistent with investors placingcredence in unrealized gains and losses.

The results of estimating the pure comprehensive income statement model reported incolumn II of Table 3 indicate that MV is significantly positively associated with RECLASS Thetotal coefficient on RECLASS is β4-β7=9.88-0.42=9.46, significant at the 1% level The totalcoefficient is closer to the coefficient of 13.34 on the relatively permanent NIBEXother than it is tothe coefficient on the relatively transitory comprehensive income components: 2.05 on EX, 0.42

on ΔREAUGL, and 1.49 on OCIother These results are consistent with investors puttingconsiderably greater valuation weight on realized than unrealized gains and losses This likely isattributable in part RECLASS having greater persistence than ΔREAUGL, as suggested by theresults reported in Table 2

The results of estimating the full equation (6) reported in Column III of Table 3 indicatethat the value relevance of RECLASS diminishes only slightly when AUGL and other balancesheet variables are added to the model In contrast, the coefficient on AUGL becomesinsignificant once RECLASS and other comprehensive income statement variables are added tothe model Specifically, the total coefficient on RECLASS is β4-β7=7.84-0.56=7.28, significant

at the 1% level This total coefficient is somewhat closer to the coefficient of 10.84 on therelatively permanent NIBEXother than to the coefficients on the relatively transitorycomprehensive income components: 1.82 on EX, 0.56 on ΔREAUGL, and 1.29 on OCIother Theseresults are consistent with investors deeming RECLASS to have considerable incremental valuerelevance beyond AUGL and the other equation (6) variables

6. DETERMINANTS OF THE VALUE RELEVANCE OF RECLASS: LIQUIDITY AND GROWTH

In this section, we examine two possible reasons for the value relevance of RECLASS.First, we consider the possibility that unrealized gains and losses are unreliable, as opponents offair value accounting often allege This possibility is inconsistent with our findings discussedabove that AUGL is value relevant in a pure balance sheet model and also with the high liquidity

of most AFS securities Second, we consider the possibility that RECLASS interacts with or is

an indicator of future bank growth, rather than simply a reclassification of previously unrealizedbut recognized gains and losses

6.1 Liquidity

To test the liquidity explanation for the value relevance of RECLASS, each year we estimate the full equation (6), as in column III of Table 3, for two equal-sized subsamples formedbased on a proxy for the liquidity of AFS securities Following Barth (1994), we use the

percentage of AFS securities that are U.S Treasuries as our proxy In untabulated robustness tests, we include broader measures of liquid securities, with substantively the same results

The results are reported in Table 4 The coefficient on RECLASS is higher and more significant for the high liquidity subsample Specifically, the total coefficient on RECLASS is

β4-β7=11.41-3.77=7.64 in the high liquidity subsample compared to 3.18+1.19=4.37 in the low liquidity subsample, with the difference significant at 5% level These results are consistent with hypothesis H2N that the value relevance of RECLASS does not result from the unreliability

of unrealized gains and losses recorded in AUGL