Are Securitizations in Substance Sales or Secured ......par le marché (c’est-à-dire, le bêta du MÉDAF), que les passifs ﬁgurant au bilan. Cette observation est conforme au

Are Securitizations in Substance Sales or Secured Borrowings? Capital-Market Evidence*

FLORA F. NIU, Wilfrid Laurier University

GORDON D. RICHARDSON, University of Toronto

AbstractTwo standard-setting approaches have emerged globally to guide the choice of accountingfor securitizations: the control and components approach (SFAS No. 125 and SFAS No. 140)and the risks and rewards transfer approach (IAS No. 39). A lack of consensus about derecog-nition accounting is a major impediment to achieving convergence in global standards thatmust be resolved. Thus, both SFAS No. 140 and IAS No. 39 will be reexamined, and evidencepertinent to the debate is timely and important. In this study, we present evidence consistentwith the view of credit-rating analysts, who view many securitizations as, in substance,secured borrowings. Specifically, for a sample of originators applying sale accounting guid-ance in SFAS No. 125 /140 during the period 1997–2003, we show that off-balance-sheetdebt related to securitizations has, on average, the same risk-relevance for explaining marketmeasures of risk (that is, CAPM beta) as on-balance-sheet debt. We also find that, in a returnsand earnings association framework, the pricing multiple on securitization gains declines asthe amount of off-balance-sheet debt increases, implying that investors take off-balance-sheetdebt into account when assessing the valuation-relevance of such gains. For those who advo-cate the control and components approach to securitization accounting, our results suggestthat, at least for frequent securitizers, the put option arising from implicit recourse is a“missing piece” that is not currently accounted for when calculating securitization gains.Our results challenge the extant measurement standards in SFAS No. 140.

Contemporary Accounting Research Vol. 23 No. 4 (Winter 2006) pp. 1105–33 © CAAA

* Accepted by Michel Magnan. This paper is partially based on Flora Niu’s dissertation at the Univer-sity of Waterloo, supervised by Gordon Richardson. We are grateful to other dissertation committeemembers for their generous input: Phelim Boyle, Duane Kennedy, and Tony Wirjanto. We thankSati Bandyopadhyay, Donal Byard, Alan Douglas, Ken Klassen, Tom Linsmeier, Michel Magnan,James Moore, Kathy Petroni, Stephen Ryan, Catherine Shakespeare, Katherine Schipper, MichaelStein, Dan Thornton, Ken Vetzal, two anonymous referees, workshop participants at the followinguniversities: Baruch, Chinese University of Hong Kong, Concordia, McMaster, New York, Oregon,Simon Fraser, Toronto, University of Technology at Sydney, Waterloo, Wilfrid Laurier, and York;and attendants at the 2003 annual meetings of American Accounting Association (AAA), theCanadian Academic Accounting Association, the 2004 annual meeting of the Australia and NewZealand Accounting and Finance Association, 2005 AAA/FASB (Financial Accounting StandardsBoard) Financial Reporting Issues Conference, and the 2006 Journal of Contemporary Account-ing and Economics Symposium for helpful comments. We thank FitchRatings for allowing us toaccess data and for their overall support of the project. We especially thank Tricia O’Malley of theInternational Accounting Standards Board, Peter Martin of the Canadian Institute of CharteredAccountants, and Gale Kelly of KPMG LLP for their detailed comments on an earlier version andinsightful discussions regarding securitization accounting. Gordon Richardson is grateful toKPMG for the generous financial support.

1106 Contemporary Accounting Research

Keywords Control and components approach; Earnings multiple; Financial assets;Securitization

JEL Descriptors G320, M410

La titrisation est-elle par essence une vente ou un emprunt avec nantissement ? Observation du marché financier

CondenséLa titrisation des actifs consiste dans le regroupement d’actifs — prêts hypothécaires, créancesclients, soldes de carte de crédit ou autres créances — et le financement de ces actifs au moyende titres négociables que l’on vend à des investisseurs. Les analystes de crédit envisagentgénéralement la titrisation sous l’angle de l’emprunt avec nantissement et disposent d’algo-rithmes permettant d’ajuster le bilan de l’initiateur afin d’y réintégrer les actifs transférés etles passifs hors bilan qui s’y rattachent. Les analystes ont critiqué l’application aux opérationsde titrisation du traitement comptable réservé aux ventes, invoquant le fait que la plusgrande partie si ce n’est la totalité du risque lié aux actifs transférés incombe toujours àl’initiateur. Ce traitement comptable est le nœud du débat chez les normalisateurs du mondeentier.

À l’échelle internationale, deux axes de normalisation ont fait surface pour guider lechoix de la comptabilisation des opérations de titrisation comme ventes ou commeemprunts avec nantissement. Ces deux méthodes peuvent être ainsi décrites : 1) les actifssont sortis du bilan au motif du transfert du contrôle, le traitement comptable réservé auxventes étant appliqué aux éléments transférés (méthode du contrôle et des composantes);2) les actifs sont sortis du bilan au motif du transfert des risques et des avantages, le traitementcomptable réservé aux ventes étant appliqué aux actifs à l’égard desquels l’entité a transféréla quasi-totalité des risques et des avantages (méthode des risques et des avantages).

La norme en vigueur aux États-Unis, le FAS 140, s’appuie sur le transfert du contrôle(méthode du contrôle et des composantes) et exige l’adoption d’une méthode de comptabili-sation séparée des composantes si le critère du transfert du contrôle est respecté. La normecomptable internationale en vigueur, l’IAS 39, s’appuie principalement, quant à elle, surune analyse du transfert des risques et des avantages. Le consortium international de norma-lisation comptable (qui regroupe l’IASB, le FASB et d’autres organismes nationaux de nor-malisation importants) estime que l’absence de consensus relativement à la sortie du bilanest un obstacle majeur aux visées de convergence des normes mondiales, obstacle qui doitêtre éliminé. Le FAS 140 de même que l’IAS 39 seront donc réexaminés, ce qui fait que lesdonnées pertinentes aux questions à débattre viennent à point nommé.

Les auteurs recueillent des données qui confirment l’hypothèse selon laquelle lamoyenne des initiateurs continuent d’assumer la plus grande partie si ce n’est la totalité desrisques liés au transfert des créances. Plus précisément, pour un échantillon d’initiateurs quiappliquent les recommandations des FAS 125 et 140 relatives à l’application aux opérationsde titrisation du traitement comptable réservé aux ventes, au cours de la période 1997–2003,les auteurs montrent que les passifs hors bilan liés à la titrisation ont, en moyenne, la mêmepertinence relativement au risque, lorsqu’il s’agit d’expliquer les mesures du risque utilisées

CAR Vol. 23 No. 4 (Winter 2006)

Are Securitizations in Substance Sales or Secured Borrowings? 1107

par le marché (c’est-à-dire, le bêta du MÉDAF), que les passifs figurant au bilan. Cetteobservation est conforme au point de vue des analystes de crédit qui considèrent maintesopérations de titrisation comme des emprunts avec nantissement, par essence.

Les auteurs présentent également d’autres faits confirmant les constatations précédentes :dans un cadre associant rendements et bénéfices, le coefficient de capitalisation des gainsrésultant de la titrisation diminue à mesure que le montant des passifs hors bilan augmente, cequi suppose que les investisseurs tiennent compte des passifs hors bilan lorsqu’ils apprécientla pertinence de ces gains aux fins de l’évaluation. Il semble que l’option de vente découlantde la garantie implicite (c’est-à-dire l’obligation d’absorber les pertes à concurrence duplein montant des actifs transférés, par souci de préservation de la réputation) soit un« élément manquant » que ne comptabilisent pas les « titriseurs » assidus de l’échantillondes auteurs, lorsque ces derniers calculent les gains résultant de la titrisation.

Les auteurs d’études apparentées concurrentes sur la comptabilisation des opérationsde titrisation faisant appel aux échantillons du FAS 125 — Shakespeare (2004) et Karaoglu(2005) — ainsi que les auteurs d’une étude faisant appel à l’échantillon du FAS 140 —Dechow, Myers et Shakespeare (2005) — formulent l’hypothèse, confirmée par les donnéesqu’ils observent, que les initiateurs utilisent le pouvoir discrétionnaire accordé par les FAS 125et 140 pour évaluer les intérêts conservés de manière opportuniste afin d’atteindre desobjectifs de gestion du résultat, ce qui met en relief les problèmes soulevés par la méthodedu contrôle et des composantes qui s’appuie lourdement sur la fiabilité des estimations de lajuste valeur des composantes vendues et conservées. Les données que recueillent ceschercheurs, lorsqu’elles sont conjuguées aux observations des auteurs de la présente étuderelativement aux passifs hors bilan, remettent en question les normes de mesure quisubsistent dans le FAS 140.

1. Introduction

If a company retains the subordinated piece of a transaction, or retains a levelof recourse close to the expected level of loss, essentially all the economic riskremains with the seller. There is no rating benefit that is deserved becausethere is no significant transfer of risk and there is no point analyzing such acompany differently from the way it would be analyzed if it had kept thereceivables on the balance sheet. Another serious concern is “moral recourse”,the reality that companies feel that they must bail out a troubled securitizationalthough there is no legal requirement for them to do so. Companies thatdepend on securitizations as a funding source may be especially prone to tak-ing such actions. In many situations, this expectation undermines the notion ofsecuritization as a risk transfer of securitization as a risk transfer mechanism.

(Standard and Poor’s [S&P] 2001, 106)

Asset securitization involves pooling groups of assets, such as mortgages andtrade or credit card receivables, and financing them with securities that are sold toinvestors. As the above quotation implies, credit analysts generally adopt a securedborrowing view of securitizations and have algorithms for adjusting the balancesheet of the originator to bring the transferred assets and related off-balance-sheet



(OBS) debt back on to the balance sheet. The analysts have criticized sale account-ing treatment for securitization deals, arguing that most, or all, of the risk of thetransferred assets remains with the originator. This accounting is the focus ofdebate among standard-setters around the globe.

Two standard-setting approaches have emerged globally to guide the choicebetween sale accounting and secured borrowing accounting for securitizations.The two approaches are characterized as follows: assets are derecognized on thebasis of (a) transfers of control, with sale accounting applied to transferred compo-nents (control and components approach),1 or (b) transfer of risks and rewards, withsale accounting applied to assets for which the entity has transferred substantiallyall risks and rewards (risks and rewards transfer approach).

The current U.S. standard, Statement of Financial Accounting Standards(SFAS) No. 140 (Financial Accounting Standards Board [FASB] 2001), is based ona transfer of control (control and components approach) and adopts separate compo-nents accounting if a control transfer test is met. In contrast, the current interna-tional standard, International Accounting Standard (IAS) No. 39 (InternationalAccounting Standards Board [IASB] 1998), is based primarily on an analysis ofrisks and rewards transfer. The international accounting standard-setting partner-ship (IASB, FASB, and other major national standard-setters) has identified a lackof consensus about derecognition accounting as a major impediment to achievingconvergence in global standards that must be resolved. Thus, both SFAS No. 140and IAS No. 39 will be reexamined, and evidence that is pertinent to the issues tobe debated is timely.2

In this study, we present evidence consistent with the notion that originators,on average, retain most, if not all, of the risks related to the transfer of receivables.Specifically, for a sample of originators applying the sale accounting guidance inSFAS No. 125 /140 during the period 1997–2003, we show that OBS debt relatedto securitizations has, on average, the same risk-relevance for explaining marketmeasures of risk (that is, capital asset pricing model [CAPM] beta) as on-balance-sheet debt.3 This is consistent with the view of credit-rating analysts, who viewmany securitizations as, in substance, secured borrowings.

We also present additional evidence supporting the above findings: in a returnsand earnings association framework, the pricing multiple for securitization gainsdeclines as the amount of OBS debt increases, implying that investors take OBSdebt into account when assessing the valuation-relevance of such gains. It wouldappear that the put option arising from implicit recourse (that is, the obligation toabsorb losses up to the full amount of transferred assets, in order to protect reputa-tion) is a “missing piece” that is not accounted for by the frequent securitizers inour sample when they calculate securitization gains.

Related concurrent studies on securitization accounting using SFAS No. 125samples by Shakespeare 2004 and Karaoglu 2005, and one study using an SFASNo. 140 sample by Dechow, Myers, and Shakespeare 2005, hypothesize and findevidence consistent with originators using the discretion afforded by SFASNo. 125 / 140 to value retained interests in an opportunistic manner to achieveearnings-management objectives. These studies thus point to problems with the



control and components approach that relies heavily on reliable fair value estimatesof components sold and retained. Their evidence, when combined with our OBSdebt evidence, challenges the extant measurement standards in SFAS No. 140.

The rest of the paper proceeds as follows. Section 2 provides some institu-tional background information. Section 3 reviews the related literature. Section 4develops the hypotheses. Empirical models and variable measurement are describedin section 5. Section 6 describes the sample and section 7 discusses the results.Finally, section 8 concludes and discusses limitations of the study.

2. Background

The securitization process: A simple example

The mechanics of asset securitization can be complex. Typically, the originatorsells the pool of assets to a special-purpose entity (SPE). The SPE uses the pro-ceeds from the securities it has issued, backed by the pool of assets, to pay for thepurchase of those assets from the originator. Securities marketed in this manner arereferred to as asset-backed securities.4

Many scholars believe that the rapid growth of securitizations is due to itsnumerous benefits. For example, originators can achieve better asset portfolio diver-sification, a higher level of liquidity, and sale accounting results in OBS financing.One benefit of this form of financing is that the SPE can typically raise financing ata lower cost of capital, compared with the originator, due to lower expected bank-ruptcy costs (see Gorton and Souleles 2005). Typically, most of the SPE’s debt isissued to public investors who require that the senior securities be highly rated(generally AAA or AA). In order to achieve these high ratings for the senior securi-ties, the SPE must obtain credit enhancements that insulate senior securities fromthe risk of default on the underlying financial assets. Such credit enhancements areusually provided by the originators in the form of cash collateral accounts, reservefunds, commitments to purchase assets in default, recourse provisions, or holdingthe most junior securities issued by the SPE.

Asset securitization has expanded dramatically over the last decade, and isnow being applied by a variety of firms, from financial institutions to manufactur-ing firms. The issuance of asset-backed securities totaled $1.10 trillion in 2005, upfrom $100 billion in 1995 (Bond Market Association 2006). Clearly, these are eco-nomically important developments and the accounting issues are drawing a fairamount of attention from various regulatory bodies.

Figure 1 illustrates how the balance sheet of the originator can vary dependingon whether sale or secured borrowing accounting is used. Under sale accounting, the$100 of transferred receivables are removed from the balance sheet (for simplicity,we assume a zero gain on sale) and replaced by consideration consisting of the sub-ordinated tranche or class ($30) plus cash ($70). The book debt-to-equity ratio inthis instance is zero, implying no financial leverage. The OBS debt in this instanceis $70, which is the $100 of transferred receivables minus the $30 of subordinateddebt securities held by the originator, which provides credit support to the seniorsecurities holders and takes the first loss position. As explained below, the $70 is,



in effect, the amount of implicit liability to public investors who purchased seniordebt securities. Under secured borrowing accounting, the $70 is reflected as securedborrowing and the $100 of receivables remains on the originator’s balance sheet.The book debt-to-equity ratio in this instance is 70/200 � 35 percent. While sim-plistic, this example illustrates that the book debt-to-equity ratio under securedborrowing accounting will be higher than under sale accounting. The actual differ-ence between the two methods for an originator is typically far more dramatic (seeS&P 2001). The primary empirical question we pose is the following: Is saleaccounting appropriate if a measure of financial leverage, based on the secured bor-rowing accounting, better explains equity beta for our sample of originators?

Finally, Figure 1 illustrates another point of importance to our paper. Theexplicit recourse in our example is the first loss position of $30. Why then wouldthere be a liability for the remaining $70? Gorton and Souleles (2005) refer to the$70 as involving implicit recourse: if the first $30 is wiped out by default losses,holders of the Tranche A securities issued by the SPE will look to the originator toabsorb any additional losses, up to the full $100. The originator will feel compelledto do so, if it intends to go back to investors with subsequent securitization deals.

Accounting for transfers of financial assets under SFAS No. 125/140

In this paper, we examine our research questions by using a combined SFASNo. 125 /140 sample. The FASB adopted the control and components approachwhen it issued SFAS No. 125 in 1996. In 2001, SFAS No. 125 was replaced bySFAS No. 140, which expands disclosure requirements but carries over most ofSFAS No. 125’s measurement provisions. Upon completion of a transfer, SFASNo. 125 /140 requires the transferred assets to be decomposed into separate com-ponents: retained and sold. In its simplest form, the fair values of parts of thetransferred assets sold and retained are estimated, and the relative fair values areused to prorate the original carrying value between the parts sold and the partsretained. The gain on sale is the difference between the net cash proceeds and thecarrying value of the parts of the assets sold.5

Critics of SFAS No. 125 /140 have expressed concerns over the reliability ofthis decomposition process and whether investors can depend on the reported bal-ance sheet and income statement information to assess the substance of securitiza-tion firms’ risk and earnings. First of all, like other accounting standards (e.g.,SFAS No. 115) that rely on fair value estimates, SFAS No. 125 /140 requires theestimation of fair values of retained interests (for example, the $30 subordinatedsecurities in Figure 1) that absorb the loss first when there is a default risk. Becausefair value estimates involve uncertain future events (for example, borrower defaultrate, prepayment rate, and appropriate discount rates), the subjectivity in the estima-tion process provides originators with discretionary opportunities to overestimatethe fair value of retained interests, resulting in an underestimate of the carrying valueallocated to the components sold and thus an overstatement of the securitization-related gains. As a consequence, subsequent downward earnings restatements maybe required during unanticipated business environments (Scism 1998).



Perhaps the most serious challenge to SFAS No. 125 /140 accounting arisesfrom implicit recourse. For example, Higgins and Mason (2004) document sizableevents of voluntary support of credit card securitizations by commercial banks inthe mid-1990s. The following reflects the concerns of rating analysts (Fitch Inc.1999, 5):

The most dramatic risk is the moral imperative to protect a securitization fromdefault. Although not legally required, there is a pattern of behaviour thatstrongly suggests finance companies and other entities will take whatevermeasures available to support their transactions and ensure continued marketsuccess.


Figure 1 Securitization structure

Originator’s initial balance sheet

Receivables $200 Owner’s equity $200

Originator’s balance sheet aftertransferring $100 of receivables to SPE

(sale accounting)

Receivables

Retained interestCash

$100

$30$70

$200Owner’s equity $200

$200

Book debt-to-equity ratio: nil

Originator’s balance sheet aftertransferring $100 of receivables to SPE

(secured borrowing accounting)

Receivables

Cash

$200

$70$270

Debt

Owner’s equity

$70

$200$270

Book debt-to-equity ratio: 70/200 � 35%

SPE’s balance sheet after$100 of receivables transferred to it

Receivables $100

$100

Debt, Tranche A

Debt, Tranche B

$70

$30$100

Borrowers

Loan originator(e.g., finance company)

Special-purpose entity(e.g., trust)

Investors

Securities:Tranche A

to investors,Tranche B

to originator

$

$

Sale ofreceivables

Retained interest(Tranche B)

$ Receivables


Implicit recourse, if it exists, negates the SFAS No. 125 /140 premise that theoriginator has surrendered control to the SPE. However, SFAS No. 125 /140 con-siders explicit (that is, contractual) recourse, but not implicit recourse, in tests ofwhether control has been transferred. It is the extent to which the market acts as ifimplicit recourse exists that drives the risk-relevance of OBS debt when investorsestimate CAPM beta. As elaborated on in section 4, securitization gains on sale areoverstated if implicit recourse exists but is not accounted for.

3. Literature review

The formal theory and evidence supporting the existence of implicit recourseappears in Gorton and Souleles 2005. They use commitment on the part of theoriginator as a mechanism to resolve the adverse selection problem in a two-persongame involving the originator and the SPE investor. The commitment takes theform of an expectation that the originator will subsidize the SPE investor for anydefault losses related to the transferred receivables. What sustains the commitmentis the repeated context of the game and the threat by the SPE investor to not takethe originator’s debt in the future if the originator deviates from the implicit con-tract. Gorton and Souleles stress that the understanding must not be a formalcontract, because a formal contract would violate accounting and regulator rulesallowing sale accounting.

One important implication of the above theory is that SPE investors care aboutthe bankruptcy risk of the originator because the originator must stay in business inorder to honor its implicit guarantee. They test this idea empirically with a sampleof credit card securitizations and show that the yields on SPE debt reflect the bondratings of the originator. This represents indirect evidence consistent with the exist-ence of implicit recourse.

This study complements the findings from several other studies examining thecapital market implications of accounting for securitizations per SFAS No. 125 orSFAS No. 140 requirements by focusing on the leverage impact of OBS debt arisingfrom sale accounting. Using SFAS No. 125 data for the period 1997–99, Shake-speare (2004) examines whether managers use their discretion to meet earningstargets either by managing the securitization volume or by managing the assump-tions used to estimate the fair values of the retained interests. She finds evidence ofearnings management to meet or beat targets with fair value estimates relating tothe retained interests, but not with the securitization volume. Karaoglu (2005)examines regulatory filings between 1997 and 2000 and finds evidence that banksuse gains from securitization to manage both regulatory capital and accountingearnings. Dechow et al. (2005) examine the gain on sale reported in the incomestatement, the retained interest reported on the balance sheet, and the adversechange disclosures under SFAS No. 140 reported in the notes. Like Shakespeare2004, the authors conclude that originators manage assumptions used to estimatethe fair values of retained interests in order to achieve earnings management objec-tives. Specifically, larger gains on sale are recorded when presecuritization earn-ings are low. Their results suggest that firms with stronger corporate governancetend to report smaller gains.



4. Hypotheses

Assessment of OBS debt from securitizations

The implicit recourse argument supports our primary empirical hypotheses —namely, that investors take OBS liabilities related to securitization into accountwhen assessing the financial leverage and systematic equity risk of the firm.Returning to our simple example discussed in section 2, it is the net of the transferredreceivables minus any retained interests on the balance sheet of the originator thatrepresents the originator’s implicit recourse liability to the holders of senior SPEdebt ($100 receivables � $30 retained interests � $70 implicit recourse obligation).To the extent that sale accounting results in unrecorded financial leverage, the fol-lowing hypothesis is implied (in alternative form):

HYPOTHESIS 1. Measures of OBS risk are positively associated with systematicequity risk.

Consistent with the view that originators retain most, if not all, of the risksassociated with transferred assets, the economic substance of many securitizationsis in fact secured borrowing (see Ryan 2002, 168). If investors view OBS liabilitiesto be equivalent to on-balance-sheet leverage, the following hypothesis is implied(stated in the null form):

HYPOTHESIS 2. There is no difference in the risk-relevance of OBS liabilitiesand debt on the balance sheet.

The interaction of gain on sales with perceived OBS debt from securitization

Because accounting numbers recognized on the balance sheet articulate with theincome statement, if OBS risk involvement is important in explaining how investorsassess equity risk, the perceived risk involvement likely affects investors’ perceptionof the valuation-relevance of securitization-related earnings. The next hypothesis,therefore, predicts the interaction between gains on sale and perceived OBS risk.

As noted by Shakespeare 2004 and Dechow et al. 2005, aggressive estimatesregarding the fair value of retained interests have a direct impact on the valuation-relevance of securitization gains. Prior to the adoption of SFAS No. 140, disclosureof managerial estimates used to determine the estimated fair value of retainedinterests was not required. In 1997, several securitizing firms announced lossesresulting from downward adjustments to previously recorded retained interests.The adjustments occurred because securitized assets were prepaid more quickly thanthe seller’s original estimates. The losses added to the concern of equity analystsregarding securitization gains (Scism 1998; Fabozzi 1998, 129).

We predict that the valuation-relevance of securitization-related earnings isperceived to be lower when firms have more OBS risk. This follows directly fromthe theory of Gorton and Souleles 2005 discussed in section 3, because firms withmore OBS risk are more likely to be frequent securitizers concerned about theirreputation with SPE investors. Thus, the put option arising from implicit recourseis a “missing piece” that is not accounted for when calculating the securitization



gains of frequent securitizers.6 Hence, our next hypothesis (in the alternative form)is as follows:

HYPOTHESIS 3. The association between securitization gains and stock returnsis lower for firms with more OBS risk from securitization.

5. Models

The model used for testing the OBS debt hypotheses

We consider several possible alternative measures of equity risk, the dependentvariable in our risk-relevance model. There is evidence that beta is an appropriate,though noisy, measure of systemic risk. For example, Kothari and Shanken (1995)demonstrate a positive relation between beta and average returns in a cross-sectionalsetting, although Fama and French (1992) show that measures of systematic equityrisk are unable to fully explain expected returns. We opt for CAPM beta as ourdependent variable because the Fama and French risk factors lack theoreticalsupport. Moreover, the extant accounting risk-relevance literature also uses CAPMbeta. The following model was originated by Hamada 1972 and was furtherdeveloped and implemented by Bowman 1979, Dhaliwal 1986, and Kimmel andWarfield 1995.

We begin with the well-known result in Hamada 1972:

�L � [1 � (1 � � )(DL/SL)]�U (1),

where �L and �U are the CAPM betas of a levered and unlevered firm, respec-tively, � is the corporate tax rate, and DL/SL denotes the debt-to-equity leverageratio. Equation (1) states that the equity beta of a levered firm is equal to the equitybeta of an unlevered firm times one, plus the leverage ratio adjusted for the interestrate tax shelter.

Bowman (1979) shows that, given certain assumptions, one can use account-ing beta, �O, as a proxy for �U. Accounting beta is the covariability of a firm’saccounting earnings before interest and taxes with the corresponding earnings ofthe market portfolio. Bowman is clear that �O measures �U with noise, in part dueto the error with which accounting income proxies for economic income (that is,stock return). Substituting �O for �U, Bowman (1980, 250) derives the followingmultiplicative empirical specification from (1):

� � �0 � �1�O � �2�O (1 � � )(D/S) � � (2).

Absent measurement error in accounting beta, the predicted coefficients for �1and �2 are unity. However, measurement error in �O will cause the slope coeffi-cients to be biased downward toward zero. This is problematic, especially for studieslike ours, in which we examine various candidates for the debt-to-equity ratio. Toavoid having the variable of interest (D/S) being interacted with a variable knownto contain (potentially serious) measurement error, Bowman (1980, 250) proposesthe following additive empirical specification as a modification of (2):



� � 0 � 1�O � 2(1 � � )(D/S) � (3).

The additive specification has the advantage that the variable measured witherror, �O, is isolated and only 1 is biased downward. This permits reliable infer-ences regarding the estimated coefficient of interest, D / S. Equation (3) is ourprimary empirical specification. We report robustness checks using the multiplica-tive empirical specification. Because our primary interest is the OBS debt relatedto securitization transactions, the total debt, D, can be further decomposed into on-balance-sheet liabilities (DBS) and OBS liabilities related to securitization (DOBS)measuring the implicit recourse liability to SPE debtholders. Thus, we have:

D � DBS � DOBS (4).

Substituting (4) into (3) results in our primary empirical specification for testingour OBS debt hypotheses:7

� � 0 � 1�O � 2(1 � � )(DBS /S) � 3(DOBS /S) � (5).

In (5), 3 is used to test Hypothesis 1. If most or all risk remains with the origina-tor through retained interests and implicit recourse, and investors view the economicsubstance of securitization to be that of secured borrowing, then we would predictthat 3 � 0 and 2 � 3, which are implied by Hypothesis 1 and Hypothesis 2.

Market beta in (5) is estimated on the basis of a market model regression ofdaily firm returns on daily market portfolio returns. Center for Research in Secu-rity Prices (CRSP) daily return data for the fiscal year corresponding to the periodfor which the financial data are reported are used in the estimation period. The CRSPvalue-weighted index for the corresponding period is used to measure the return onthe market portfolio. To avoid potential bias in estimating betas for smaller firmsdue to nonsynchronous trading for daily returns, market model parameters areadjusted by using the Scholes and Williams 1977 one lag and one lead model.

Following prior studies (e.g., Dhaliwal 1986; Kimmel and Warfield 1995), wedefine operating beta as follows:

�O, it � COV(OEit, OEmt)/ VAR(OEmt) (6),

where �O, it is the operating beta for firm i in period t, OEit is the operating earn-ings of firm i in period t, and OEmt is the operating earnings for a market portfolioin period t.

Firm i’s operating earnings (OEit) are defined as operating income beforedepreciation, interest, and tax expense, deflated by the beginning-of-period totalassets. The corresponding market measure (OEmt) is similarly constructed as theS&P 500 value-weighted operating income deflated by the beginning-of-periodtotal assets. A firm’s operating beta, �O, is then estimated for each period by run-ning a time-series regression of operating earnings on market operating earnings.In order to allow sufficient observations in the estimation regression, we use quar-terly data up to the year of interest to calculate operating betas. At least 10, and up



to 56, quarters of data during the period 1983 to 2002 are used to estimate theoperating beta for each firm-year observation.

On-balance-sheet leverage (DBS) is measured by total (current and long-term)debt net of the interest tax shield as of the fiscal year-end. Following Dhaliwal1986 and Kimmel and Warfield 1995, we measure the tax rate as total income taxesdivided by pre-tax income.8 We define credit enhancements to refer to retainedinterests (such as the subordinated securities shown in Figure 1), and other formsof contractual recourse. We measure DOBS as the outstanding principal amount offinancial assets securitized after subtracting the credit enhancements outstandingas of the fiscal year-end, disclosed in firms’ annual reports.9

The model used for testing the earnings hypothesis

Following prior studies (e.g., Easton and Harris 1991), we use the followingreturns–earnings association model to test our third hypothesis:

Rit � � 0 � �1Eit � �2 Eit � � it (7),

where

Rit � raw annual stock return for firm i in year t;

Eit � earnings for firm i in year t; and

Eit � annual change in earnings for firm i from year t � 1 to year t.

To examine the valuation-relevance of securitization-related earnings, wedecompose total earnings into the net of tax amount of securitization-related earn-ings (GOS) and other earnings (NI_GOS) as follows (the firm and year subscriptsare dropped for simplicity):

R � � 0 � �1(NI_GOS) � � 2 (NI_GOS) � �3GOS � �4 (GOS) � � (8).

All variables are deflated by the market value of equity at the start of the fiscalyear. The coefficients on GOS and GOS measure the association between returnsand the level and changes in securitization-related earnings. Allowing OBS debt tobe interacted with GOS results in the following regression equation:

R � � 0 � �1(NI_GOS) � � 2 (NI_GOS) � �3GOS � �4 (GOS)� �5(GOS*DOBSDUM) � �6( GOS*DOBSDUM) � � (9).

In (9), DOBSDUM is a dummy variable, which is 1 if DOBS is above the samplemedian and 0 otherwise. Hypothesis 3 predicts that the association between securi-tization gains and stock returns is lower for firms with more OBS risk. Thus, ourpredicted interaction effects are that �5 � 0 and �6 � 0.



6. Sample

SFAS No. 125 became effective for fiscal years beginning after December 15, 1996,and SFAS No. 140 became effective beginning 2001. The sample selection involvesidentifying as many U.S. firms as possible that pursued securitization activities andapplied SFAS No. 125/140 accounting during any year from 1997 to 2003.

Our sample consists of 535 firm-year observations for the 1997–2003 period,generated by 103 firms, with a complete panel of data available for 41 firms. To beincluded, the firm had to be a listed company with data on COMPUSTAT and CRSP,and the firm had to disclose sufficient securitization information in order for us tocalculate implicit and contractual recourse.10 We use the maximum number ofobservations available for each test, and the number of observations thus variesacross tests.

Table 1 shows the sample distribution by industry. Similar to the sample ofDechow et al. 2005, the intensity of securitization-related transactions in the tradi-tional financial sector is stronger than in the other sectors, with roughly 35 percentof our sample consisting of banks, savings and loan companies, and insurancecompanies (compared with 33 percent for Dechow et al. 2005). Consistent with therapid spread of securitizations, our sample consists of many other industries inaddition to financial sector. The implicit recourse argument of Gorton and Souleles2005 assumes that originators engage in securitization deals on an on-going basis.Our sample of originators satisfies this assumption. Of our 103 sample firms, allbut 8 are in the sample for three or more years during 1997–2003. In the sensitivityanalysis section, we also partition our sample between traditional financial institu-tions and other firms to ascertain whether our empirical results differ by industry.

7. Results

Results from testing the OBS debt hypothesis

Table 2, panel A provides descriptive statistics for the primary variables used totest the OBS debt hypothesis.11 Panel A indicates that the mean market beta for thesample is 0.99. It also indicates that securitization appears to be an economicallysignificant event, as shown by a mean DOBS of 4.3, implying that the outstandingamount of transferred receivables minus the related credit enhancements represents4.3 times the market value of equity of originators, on average. This implies thatthe mean book debt-to-equity ratio goes from 5.9 (Table 2, panel A) using saleaccounting to 10.2 using secured borrowing accounting, representing a substantialdifference between the two methods as claimed by debt-rating agencies. Of particu-lar interest in panel A is the ratio (CRD) of dollar amounts of outstanding creditenhancements provided by an originating firm to the dollar amounts of outstandingfinancial assets securitized by that firm. The mean ratio of CRD is 18.2 percent,implying that the dollar amounts of credit enhancements such as retained interestsrepresent almost 20 percent of the dollar amounts of transferred receivables. Asnoted by Ryan 2002 (168), a ratio of this magnitude for our sample suggests that,on average, retained interests and other credit enhancements represent a large por-tion of the principal of the transferred receivables, implying that secured borrowing




TABLE 1Sample distribution by industry

6162 Mortgage bankers and loan correspondents 82 15.36021 National commercial banks 66 12.36022 State commercial banks 54 10.16141 Personal credit institutions 42 7.96798 Real estate investment trusts 33 6.26035 Savings institutions, federally chartered 30 5.66159 Miscellaneous business credit institutions 21 3.96172 Finance lessors 14 2.66036 Savings institutions, not federally chartered 13 2.42451 Mobile homes 13 2.46321 Accident and health insurance 13 2.46153 Short-term business credit institutions, except

agricultural 11 2.16531 Real estate agents and managers 11 2.15311 Department stores 7 1.36199 Finance — services 7 1.33711 Motor vehicles and passenger car bodies 7 1.35731 Radio, television, and consumer electronic stores 7 1.31531 Operative builders 7 1.32911 Petroleum refining 7 1.33350 Rolling drawing and extruding of nonferrous metals 7 1.34813 Telephone communications (no radiotelephone) 7 1.35040 Wholesale — professional and commercial equipment

and supplies 7 1.36020 Financial institutions 7 1.37363 Help supply services 7 1.36361 Title insurance 6 1.14911 Electric services 5 0.96311 Life insurance 4 0.76726 Unit investment trusts 4 0.71381 Drilling oil and gas wells 4 0.75712 Furniture stores 4 0.76794 Patent owners and lessors 4 0.73531 Construction machinery and equipment 3 0.63523 Farm machinery and equipment 3 0.65500 Auto dealers, gas stations 3 0.63724 Aircraft engine, engine parts 3 0.65084 Wholesale — industrial machinery and equipment 3 0.66282 Investment advice 3 0.66736 Other 3 0.67389 Services — business services, NEC 3 0.6Total 535 100%

SICcode SIC name n

Frequency(%)


accounting captures substance better than sale accounting. Panel B of Table 2 pre-sents a correlation matrix of variables used to test the OBS debt hypothesis. Asexpected, there is a positive association between �O and market beta at the 1 percentlevel. DBS and DOBS are also positively associated with the market beta. Diagnosticsin panel B indicate that collinearity is not a serious concern, because all the VIFstatistics are below the threshold level of 10.

In this paper, we estimate our regression models by running annual regressionsand reporting both mean regression coefficients and the cross-temporal t-statistic(Z1 and Z2 test), following Barth, Kasznik, and Foster 1998.12 As a check, we alsoprovide the estimation results based on panel regression methodology (Kmenta1986) using a complete panel of firms over the sample period. A complete paneldesign allows each sample firm to serve as its own control, thereby eliminating anydifferences that might result from temporal variation in sample composition.13

Table 3 reports the mean regression coefficients of tests of the market percep-tion of OBS risk related with securitization and the corresponding Z1 and Z2 tests.The results from the annual regressions indicate that both the operating risk proxy(�O) and the on-balance-sheet leverage measure (DBS) explain a significant amountof the variation in equity risk, consistent with prior studies (Dhaliwal 1986; Rosett2001). We examine our OBS debt hypothesis by focusing on the estimated coeffi-cient on DOBS. The results reveal that 3 is significantly positive, suggesting thatour OBS leverage measure, DOBS, captures risk-relevant information, incrementalto on-balance-sheet leverage and operating risk measures. The mean coefficient forDOBS is positive and significant (0.012; Z1 � 3.007; Z2 � 3.014). Therefore, theevidence is consistent with Hypothesis 1. In addition, there is no significant differ-ence between the coefficients on DBS and DOBS (t � 0.501), implying that investorsperceive the risk relevance of OBS debt to be, on average, similar to on-balance-sheet leverage, a view consistent with that of analysts. Thus the evidence is consistentwith Hypothesis 2.

The findings from the panel regression based on 41 panel firms in Table 3 aresimilar to those obtained from annual regressions. Specifically, there is a significantassociation between the proxy for the OBS risk related to securitization (DOBS) andthe equity beta. As well, there is no significant difference between the coefficienton DBS and on DOBS, suggesting that OBS leverage affects systematic risk in amanner similar to on-balance-sheet leverage. The result is consistent with the claimof Mian and Smith 1994, among others, that there is very little actual transfer ofcredit risk when originators transfer receivables.

Results from testing the earnings hypothesis

Table 4 presents descriptive statistics for the variables used in the returns andearnings association tests. Panel A indicates that the median deflated securitizationgain is 0.026. The median gain represents 36 percent (0.026/0.072) of net incomebefore securitizations, suggesting that securitization gains are a significant compo-nent of total income for such companies. It is apparent from panel A that NI_GOSis negative for at least 25 percent of the sample firm-years. Untabulated analysesreveal that NI_GOS is negative for 208 of the 549 firm-years, but net income is


TA
BL

E 2

Des

crip

tive

stat

istic

s of

var

iabl

es u

sed

to te

st th

e O

BS

debt

hyp

othe

sis

Pan

el A

: D

escr

iptiv

e st

atis

tics

�53

50.

993

1.01

10.

961

�1.

682

0.64

11.

495

3.26

0�

O53

51.

816

2.24

90.

907

0.19

30.

893

2.39

63.

261

DB

S53

55.

973

3.90

26.

699

0.14

71.

584

8.18

532

.600

DO

BS

535

4.34

80.

820

8.48

90.

000

0.00

13.

949

37.0

00C

RD

535

0.18

20.

063

0.24

50.

000

0.00

10.

226

0.98

0

Pan

el B

: C

orre

latio

n m

atri

x of

var

iabl

es

�1

0.42

4*0.

394*

0.44

4*

�O

0.30

1*1

0.27

2†0.

320*

DB

S0.

375*

0.25

3†1

0.33

3*

DO

BS

0.45

1*0.

197†

0.19

2†1

VIF

1.49

31.

266

1.24

21.

321

(The

tabl

e is

con

tinue

d on

the

next

pag

e.)

nM

ean

Med

ian

s.d.

Min

imum

Q1

Q3

Max

imum

��

OD

BS

DO

BS



negative for only 95 firm-years, suggesting that securitization gains are used bymanagers to smooth income (an observation also made by Shakespeare 2004).

Panel B of Table 4 reports the correlations of variables used in the returns andearnings regression. Panel B shows some strong correlations between the compo-nents of earnings and their changes, although diagnostics in panel B indicate thatcollinearity is not a serious concern, because the largest VIF is below the thresholdlevel. As expected, the correlation between returns and the interaction of securiti-zation gains and OBS debt (GOS*DOBSDUM) is significantly negative. There is anegative association between NI_GOS and GOS (�0.446), suggesting, once again,that GOS is used for income smoothing.

Our earnings hypothesis predicts that the association between securitizationgains and stock returns is lower for firms with more OBS risk from securitization.Our main results testing the interaction between securitization gains and OBS riskappear in Table 5. Panel A presents the mean coefficients and cross-temporal t-statisticsfrom annual regressions based on 549 firm-year observations, and panel B providesthe results from panel regressions based on 38 firms.

For the main effects model, the mean coefficients for both NI_GOS and (NI_GOS) in the separate year regressions have the expected signs and are statis-tically significant, with magnitudes comparable to prior studies (e.g., Easton andHarris 1991). The coefficient for GOS is significantly positive (0.191; Z1 � 2.693;Z2 � 1.789), suggesting that the current levels of gain on sale from securitizationare relevant for explaining returns. Unreported results indicate that the coefficient

TABLE 2 (Continued)

Notes:

In panel B, Pearson (Spearman) correlations are reported above (below) the diagonal. Variance inflation index (VIF) was used to test the collinearity among variables. A rule of thumb is that collinearity is a serious problem if VIF � 10.

The variables are defined as follows:

� � Market betas estimated by using the Scholes-Williams 1977 method, where the CRSP value-weighted market index is used.

�O � Operating betas, estimated each quarter by running a time-series regression of quarterly operating earnings on market operating earnings using available past data.

DBS � Total on-balance-sheet leverage, net of interest tax shield, deflated by fiscal year-end market value of equity.

DOBS � Outstanding dollar amount of financial assets securitized as of the fiscal year-end, after subtracting securitizing firms’ dollar amount of credit enhancements, which include retained interests and other forms of contractual recourse as of the fiscal year-end, deflated by fiscal year-end market value of equity.

* Significant at the 0.01 level (two-tailed).

† Significant at the 0.05 level (two-tailed).



on NI_GOS is significantly greater than that of GOS, suggesting that investors per-ceive the valuation-relevance of securitization gains to be less than that of earningsexcluding such gains. Panel A of Table 5 indicates that, although the coefficient on GOS*DOBSDUM is statistically insignificant, the coefficient on the interactionterm, GOS*DOBSDUM , is negative and significant (�0.432, Z1 � �2.785;Z2 � �2.195), as predicted by Hypothesis 3. In addition, a test of the linearrestriction that �3 � �5 � 0 cannot be rejected, indicating that securitization gainshave zero valuation-relevance for firms with DOBS above the sample median.14

Panel B of Table 5 reports the corresponding panel regression results. Consis-tent with the results from the annual regressions, the unreported t-tests indicate that


TABLE 3Tests of the OBS debt hypothesis: Regression estimates from annual regressions and panel regression

This panel presents the mean coefficient estimates and cross-temporal t-statistics from the annual regressions. The last two columns also report the panel regression estimates.

� � 0 � 1�O � 2(DBS /S) � 3(DOBS /S) �

Intercept ? 0.023 0.219 0.117 0.551 7.056�O � 0.535 7.548* 4.088* 0.103 2.806*

DBS � 0.024 4.850* 3.826* 0.031 6.391*

DOBS � 0.012 3.007* 3.014* 0.033 9.930*

H0: t-statistic t-statisticDBS � DOBS 0.501 �0.44Mean adj. R 2 24.5% Buse R2 45.8%

Notes:

The Z1-statistic, which assumes residual independence, is

,

where tj is the White’s 1980 t-statistic for year j, kj is degrees of freedom for year j, and T is number of years. The Z2-statistic, which accounts for cross-sectional and temporal residual dependence, is defined as mean t-statistic/(s.d. of t-statistics/ ).

Variables are as defined in Table 2.

* Significant at the 0.01 level (one-tailed).

Mean coefficient estimates from annual regressions

(n � 535)

Panel regression estimates

(41 panel firms)

Variable SignMean

coefficient Z1-statistic Z2-statistic Coefficient t-statistic

1 T/( ) t j k j k j 2�( )�[ ]1⁄2

�{ }j 1�

T

�

T 1�


TAB

LE

4D

escr

iptiv

e st

atis

tics

of v

aria

bles

use

d to

test

the

retu

rns

and

earn

ings

ass

ocia

tion

Pan

el A

: D

escr

iptiv

e st

atis

tics

R54

90.

019

�0.

018

0.52

5�

1.32

2�

0.33

60.

309

2.05

6N

I_G

OS

549

�0.

115

0.07

20.

768

�4.

721

�0.

151

0.20

2 1

.789

(N

I_G

OS

)54

9�

0.00

40.

015

0.80

9�

4.01

0�

0.10

80.

102

4.0

12G

OS

549

0.24

20.

026

0.61

3�

0.70

10.

000

0.01

94.

191

G

OS

549

0.00

30.

001

0.61

0�

4.11

2�

0.01

20.

053

2.93

1G

OS*

DO

BSD

UM

549

0.15

70.

000

0.46

0�

0.69

70.

000

0.10

6 4

.187

G

OS*

DO

BSD

UM

549

0.00

60.

000

0.36

7�

3.23

20.

000

0.00

7 2

.930

Pan

el B

: C

orre

latio

n m

atri

x of

var

iabl

es

R1

0.32

3*0.

235*

�0.

079‡

0.04

6�

0.13

8†�

0.06

3N

I_G

OS

0.33

7*1

0.34

8*�

0.66

3*0.

001

�0.

501*

�0.

117‡

(N

I_G

OS

)0.

291*

0.51

3*1

�0.

119‡

�0.

460*

�0.

028

�0.

325*

GO

S�

0.16

3†�

0.44

6*�

0.23

1*1

0.22

2*0.

703*

0.20

7†

G

OS

0.05

0�

0.13

7†�

0.34

5*0.

490*

10.

025

0.59

7*

GO

S*D

OB

SDU

M�

0.16

7†�

0.28

1*�

0.09

1‡0.

639*

0.25

0*1

0.2

79*

G

OS*

DO

BSD

UM

0.01

5�

0.08

3‡�

0.17

4†0.

334*

0.71

3*0.

450*

1V

IF1.

193

2.73

51.

763

3.57

92.

457

2.42

11.

857

(The

tabl

e is

con

tinue

d on

the

next

pag

e.)

nM

ean

Med

ian

s.d.

Min

imum

Q1

Q3

Max

imum

RN

I_G

OS

(N

I_G

OS

)G

OS

G

OS

GO

S*D

OB

SDU

M

G

OS*

DO

BSD

UM



the coefficient on NI_GOS is significantly greater than that of GOS. In addition, wefind a negative association between GOS*DOBSDUM and stock return (�0.207;t � �2.144), as expected. Once again, the linear restriction that �3 � �5 � 0 cannotbe rejected. On the other hand, the coefficient on GOS for firms with DOBS below thesample median is 0.306, an estimate close to the coefficient on NI_GOS (0.328).

Overall, the results support Hypothesis 3, suggesting that investors assignlower valuation multiples to securitization-related earnings for firms with a higherperceived level of OBS risk. These findings are consistent with the criticism ofanalysts related to the low valuation-relevance of securitization gains, especially asthe volume of securitization business increases. To be fair to sale accounting, itshould be noted that, for firms with DOBS below the sample median, the valuation-relevance of securitization gains is nonzero and in fact is similar to the valuation-relevance of earnings excluding such gains. Thus, the low valuation-relevance ofsecuritization gains is evident only for firms with substantial OBS risk. This isplausible, because these are the firms where the problems related to moral recourseare most pronounced and implicit put options are not accounted for when calculat-ing securitization gains.

TABLE 4 (Continued)

Notes:

In panel B, Pearson (Spearman) correlations are reported above (below) the diagonal. Variance inflation index (VIF) was used to test the collinearity among variables. A rule of thumb is that collinearity is a serious problem if VIF � 10.

The variables are defined as follows:

R � The return on a share over the 12 months extending from three months after the last fiscal year-end to three months after the current fiscal year-end.

NI_GOS � Pre-tax income other than securitization-related income, net of taxes, deflated by the market value of equity at the start of the year.

(NI_GOS) � Year-to-year change of NI_GOS.

GOS � Securitization-related gain, deflated by the market value of equity at the start of the year.

GOS � Year-to-year change of GOS.

DOBSDUM � Indicator, coded as 1 if DOBS (defined in Table 2) is above the sample median, and 0 otherwise.

GOS*DOBSDUM , GOS*DOBSDUM � Interactive variables.

* Significant at the 0.01 level (two-tailed).

† Significant at the 0.05 level (two-tailed).

‡ Significant at the 0.10 level (two-tailed).



TABLE 5Tests of the returns and earnings association: Regression estimates from annual regressions and panel regressions

R � � 0 � �1(NI_GOS) � � 2 (NI_GOS) � �3GOS � � 4 (GOS) � � 5(GOS*DOBSDUM)� � 6( GOS*DOBSDUM) � �

Panel A: The mean coefficient estimates and cross-temporal t-statistics from the annual regressions

Intercept ? 0.056 0.608 0.142 0.058 0.158 0.035NI_GOS � 0.290 5.403* 2.948* 0.305 5.518* 3.216*

(NI_GOS) � 0.069 2.951* 1.314‡ 0.043 2.963* 1.388‡

GOS � 0.191 2.693* 1.789† 0.251 3.112* 2.178†

GOS � 0.038 1.053 0.549 �0.379 0.492 0.301GOS*DOBSDUM � �0.432 �2.785* �2.195†

GOS*DOBSDUM � 0.429 1.665 1.303

Mean adj. R 2 18.9% 19.2%

Panel B: Panel regression estimates

Intercept ? 0.045 0.051(2.133) (2.390)

NI_GOS � 0.336 0.328(5.504)* (4.906)†

(NI_GOS) � 0.024 0.031(0.370) (0.442)

GOS � 0.294 0.306(6.282)* (4.967)*

GOS � �0.046 �0.082(�0.560) (�0.948)

GOS*DOBSDUM � �0.207(�2.144)†

GOS*DOBSDUM � 0.145(1.050)

Buse R2 18.9% 21.3%

(The table is continued on the next page.)

Mean coefficients from annual regressions (n � 549)

Mean coefficients from annual regressions (n � 549)

Variable SignMean

coefficientZ1-

statisticZ2-

statisticMean

coefficientZ1-

statisticZ2-

statistic


(38 panel firms)


(38 panel firms)

Variable Sign Coefficient Coefficient



Alternative model specifications and sensitivity analysis

We conduct the following sensitivity checks on our OBS model: using the multipli-cative model specification; partitioning our sample between financial institutions andother firms; and controlling for other risk-relevant factors documented in prior stud-ies. We also conduct additional tests on our returns and earnings association model:partitioning our sample between financial institutions and other firms; controlling forother factors documented in prior returns and earnings studies; examining whetherthe lower earnings multiple is due to a higher discount rate; and dropping observa-tions with negative net income and negative income before securitization gains.

OBS debt analysis

Using the multiplicative version of the Hamada 1972 model As discussed insection 4, to minimize potential measurement errors related with operating beta,our main results are based on the additive version of Hamada 1972 suggested byBowman 1980. We also estimate the following multiplicative model using bothannual regression and panel regressions:

� � �0 � �1�O � �2�O(1 � � )DBS � � 3�ODOBS � � (10).

The untabulated results indicate that our main inferences are not sensitive tousing the multiplicative version of the model: there is a significant associationbetween OBS risk and market beta (the mean coefficient estimate of �3 in (10) is0.004, Z1 � 1.83; Z2 � 1.50), and there is no statistically significant differencebetween the coefficient estimates for �2 and �3 in (10) for both our annual regres-sions (t � 0.53) and panel regression (t � 0.23).

Following Kimmel and Warfield 1995, we also estimate the multiplicativeversion of the model after taking a log transform of all right-hand-side variables toreduce the influence of variables measured with error. Although the coefficientson operating beta, leverage, and OBS debt are larger in magnitude than the

TABLE 5 (Continued)

Notes:

The Z1-statistic, which assumes residual independence, is

,

where tj is the White’s 1980 t-statistic for year j, kj is degrees of freedom for year j, and T is number of years. The Z2-statistic, which accounts for cross-sectional and temporal residual dependence is defined as mean t-statistic/(s.d. of t-statistics/ ).

Variables are as defined in Table 4.* Significant at the 0.01 level (one-tailed).† Significant at the 0.05 level (one-tailed).‡ Significant at the 0.10 level (one-tailed).

1 T/( ) t j k j k j 2�( )�[ ]1⁄2

�{ }j 1�

T

�

T 1�



corresponding coefficients in the case when no such transformation is performed,our main results are qualitatively similar.

Controlling for financial institutions To minimize the concerns that our resultsare driven by financial institutions, we partition our sample and repeat our tests.15

We estimate the panel regression reported in Table 3 after allowing both the inter-cept and slope coefficients to differ for financial institutions. Our main inferencesregarding Hypothesis 1 and Hypothesis 2 are not sensitive to controlling for financialinstitutions. Referring to (5), we find that, after controlling for financial institu-tions, the estimated 3 � 0 and the estimated 3 is insignificantly different from 2.

Controlling for other risk-relevant factors Our OBS risk inferences assumethat other risk-related factors are adequately controlled for. We include proxies forfirm size (the natural logarithm of total assets) and dividend payout (cash divi-dends divided by income before extraordinary items) in the OBS model reported inTable 3, because previous research (e.g., Beaver, Kettler, and Scholes 1970) docu-ments, both theoretically and empirically, that firm size and dividend payout ratioare relevant in investors’ risk assessments. Untabulated results indicate that theinclusion of these two additional variables does not significantly change the resultsreported in Table 3.

Sensitivity tests for our returns and earnings association

Controlling for financial institutions We estimate the panel regression reportedin Table 5 after once again allowing both the intercept and slope coefficients to differfor financial institutions. Our results are qualitatively similar to those reported inTable 5.

Controlling for other factors as documented in prior earnings and returns studies Previous returns and earnings association studies in the literature haveidentified several returns–earnings association determinants, such as systematicrisk (e.g., Collins and Kothari 1989; Lipe 1990); growth (e.g., Collins and Kothari1989; Martikainen 1997); persistence (e.g., Collins and Kothari 1989; Lipe 1990);and size (e.g., Collins, Kothari, and Rayburn 1987). To ensure that our Table 5results are not confounded by the omission of these variables, we partition ourfirms on the basis of the risk indicator variable (DOBSDUM) and analyze firm char-acteristics such as market beta, size (natural logarithm of total assets at year-end),growth (market-to-book ratio at year-end), and persistence (absolute value ofNI_GOS, GOS, and their changes) for the two DOBSDUM partitions.16 We conductt-tests on means and z-tests on medians (untabulated) and find that there is no sig-nificant difference among size, growth, and persistence across these two groups,although we find that mean market beta in the high DOBSDUM partition (1.17, untab-ulated) is significantly higher than the corresponding metric in the low DOBSDUMpartition (0.84). This is to be expected, given our results reported in Table 3.

Ascertaining whether our Table 5 inferences are driven by discount ratesTo further examine whether our returns and earnings results reported in Table 5 aredriven by lower discount rates investors use to assess firm value rather than by the



lower valuation-relevance of securitization gains, we augment the empirical modelreported in Table 5 by interacting DOBSDUM with both the level and the changes ofNI_GOS. If the discount rate effect dominates, we would expect lower earningsmultiples on earnings levels and changes that exclude securitization gains.17 Theuntabulated results indicate that the coefficients on the two additional interactivevariables are not significantly different from zero, while our main hypothesizedresults reported in Table 5 remain qualitatively the same.

Dropping observations with negative net income and negative income before securitization gains Previous research indicates (e.g., Hayn 1995) that the associ-ation between earnings and returns differs between profits and losses. To examinewhether our results in Table 5 are driven by negative earnings observations, weeliminate observations with negative net income for the year and repeat our regres-sion analysis reported in Table 5.18 We conducted the analysis for both annualregressions and the panel regression. Although the results (untabulated) are gener-ally weaker, they are qualitatively similar to those reported in Table 5. To furtherexamine whether our results are sensitive to excluding observations with negativeearnings excluding securitization gains, we also drop observations with nega-tive income before securitization gains, and reestimate our annual regressions andthe panel regression.19 Once again, our results are not materially affected by thisadditional robustness check.

In summary, as expected, we find that OBS debt related to securitizations has,on average, the same risk-relevance for explaining market measures of risk as on-balance-sheet debt. In addition, the valuation-relevance of securitization gainsdeclines as the amount of OBS debt increases. Our inferences do not appear to besensitive to alternative model specifications and research design choices.20

8. Conclusions

Two standard-setting approaches have emerged globally to guide the choice betweensale accounting and secured borrowing accounting for securitizations: (a) the con-trol and components approach, by which assets are recognized and derecognizedon the basis of the transfer of control over the assets, and (b) the risks and rewardstransfer approach, by which assets are derecognized on the basis of transfer of risksand rewards, with sale accounting applied to assets for which the entity has trans-ferred substantially all risks and rewards.

The international accounting standard-setting partnership (IASB, FASB, andother major national standard-setters) has identified that these differences must beresolved as one of the steps to achieving common global standards. Thus, bothSFAS No. 140 and IAS No. 39 are reexamined, and we present timely evidence thatis pertinent to the issues to be debated.

In this study, we present evidence for U.S. firms consistent with the notionthat, on average, originators retain most, if not all, of the risks related to the transferof receivables. Specifically, for a sample of originators applying the guidance inSFAS No. 125 /140 during the period 1997–2003, we show that OBS debt relatedto securitizations has, on average, the same risk-relevance for explaining market



measures of risk (that is, CAPM beta) as on-balance-sheet debt. This is consistentwith the view of credit-rating analysts, who view many securitizations as, in sub-stance, secured borrowings. In addition, our results are consistent with the theoryand evidence in Gorton and Souleles 2005 that implicit recourse exists for theentire amount of transferred receivables, especially for frequent securitizers con-cerned about their reputation with investors.

We also present additional evidence consistent with the above findings. Wefind that the pricing multiple for securitization gains declines as the amount ofOBS debt increases, implying lower valuation-relevance of gains on sale as OBSrisk increases. Our findings are robust to several sensitivity checks, alternativemodel specifications, and research design variations such as annual cross-sectionalregressions and panel regressions, and controlling for other relevant factors as doc-umented in prior studies.

Our findings have implications for standard-setters. For those who advocatethe risks and rewards transfer approach to accounting for securitizations, ourresults suggest that implicit recourse must be considered an additional source ofretained risk when evaluating risks and rewards transfer. For those who advocatethe control and components approach, our results suggest that, at least for frequentsecuritizers, the put option arising from implicit recourse is a “missing piece” thatis not accounted for when calculating securitization gains. This missing piece is onepotential explanation for why securitization gains have zero valuation-relevancefor our sample firms with substantial OBS debt. An estimate of the fair value of theimplicit put option represents a challenge to those who advocate the control andcomponents approach.

An important caveat applies to our empirical results. Our OBS debt risk-relevance results employ cross-sectional methodology, and the results hold, onaverage, for our entire sample. However, no single firm in our sample looks likethe average firm, and our results do not necessarily hold for every firm in our sam-ple. Given the substantial variation across firms in recourse and the extent of OBSrisk apparent in our sample, we are careful not to overstate the case for securedborrowing accounting as the only appropriate accounting choice for each andevery firm. There is an important role for supplementary disclosures to accompanyany measurement approach adopted by a securitizing firm, so that investors canevaluate financial risk and quality of securitization gains on a firm-by-firm basis.

Several concurrent research studies on securitization accounting (e.g., Shake-speare 2004; Karaoglu 2005; Dechow et al. 2005) hypothesize and find evidenceconsistent with originators using the discretion afforded by SFAS No. 125 /140 tovalue retained interests in an opportunistic manner to achieve earnings-managementobjectives. These studies thus point to problems with the control and componentsapproach that relies heavily on reliable fair value estimates of components sold andretained. Our results, when combined with those of concurrent SFAS No. 125 /140studies, challenge the extant measurement standards in SFAS No. 140. The debatesurrounding SFAS No. 140 accounting will, we suspect, continue for some time tocome.



Endnotes1. Derecognition refers to the removal of financial assets or liabilities from the balance

sheet.2. The boards (IASB and FASB) recently directed staff to begin a research project to

develop an approach to derecognition that would be an improvement to both IAS No. 39 and SFAS No. 140 (IASB 2005).

3. SFAS No. 140 became effective in 2001. Although SFAS No. 140 expanded disclosure requirements, it carried over most of SFAS No. 125’s measurement provisions.

4. For a detailed discussion of the technical and legal aspects of securitization structures, see Obay 2000.

5. The securitization gains represent the present value of the interest spreads expected to be earned by the originator on the receivables. See Dechow et al. 2005.

6. As additional support for Hypothesis 3, the potential for future losses to wipe out present securitization gains increases as the OBS book of business increases. As noted by Ryan 2002 (167), the unreported financial leverage arising from sale accounting for securitizations is analogous to an OBS derivative. Small variances of actual default rates from estimates can result in substantial write-downs of retained interests (for example, the $30 in Figure 1) for the originator. See Dechow et al. 2005 for a numerical example.

7. As discussed by Gorton and Souleles 2005, the originator does not enjoy any tax interest shield related to DOBS. Thus, (1 � � ) is not multiplied by DOBS in (5).

8. Following prior studies (e.g., Plesko 2003), we set the tax rate to be zero if either tax expense or pre-tax income is negative.

9. As discussed in section 2, the entire amount of assets transferred to SPEs, net of the amount of credit support explicitly specified according to the structure of securitization deals, represents OBS leverage that the originator is implicitly responsible for. Other forms of contractual recourse include, for example, cash set aside by the originator to absorb credit losses.

10. We acknowledge that, like other empirical studies, our sample is based on firms who disclose securitization information. To the extent that firms with more OBS risk opt to disclose less information, this sampling criterion biases against finding our hypothesized results.

11. To reduce the impact of extreme observations, all variables used in the study are winsorized at the 2.5th and 97.5th percentiles.

12. The Z1 statistic, which assumes residual independence, is

,

where tj is the White’s 1980 t-statistic for year j, kj is degrees of freedom for year j, and T is number of years. The Z2 statistic, which accounts for cross-sectional and temporal residual dependence, is defined as: mean t-statistic/(s.d. of t-statistics/ ); thus, more efficient estimates can be obtained.

13. The pooling technique introduced by Kmenta 1986 employs a set of assumptions about the disturbance covariance matrix that gives a cross-sectionally heteroscedastic and

1 T/( ) t j k j k j 2�( )�[ ]1⁄2

�{ }j 1�

T

�

T 1�



time-wise autoregressive model. Then, an estimator is obtained by a generalized least squares (GLS) procedure.

14. Specifically, the p-values for a test that �3 � � 5 � 0 are insignificant for our panel regression (Table 5, panel B) and for 6 out of 7 annual regressions (Table 5, panel A).

15. Following Dechow et al. 2005, we define financial institutions as banks, savings and loans companies, and insurance companies (SIC codes of 6021, 6022, 6035, 6036, 6361, and 6321).

16. We use absolute value of NI_GOS and GOS and their changes to measure persistence because prior studies indicate that large magnitude earnings innovations have less persistence (see Freeman and Tse 1992).

17. Specifically, we estimate the following model using both annual regressions and panel regression: R � � 0 � �1(NI_GOS) � � 2 (NI_GOS) � � 3GOS � �4 (GOS) � � 5(GOS*DOBSDUM ) � �6( GOS*DOBSDUM) � � 7(NI_GOS)*DOBSDUM � � 8 (NI_GOS)*DOBSDUM � � .

18. We drop 95 observations with negative net income when we estimate the annual regressions, and we drop 23 panel firms when we estimate the panel regression.

19. We drop 208 observations with negative NI_GOS for the year when we estimate the annual regressions and drop 30 panel firms when we estimate the panel regression.

20. Our results are also robust to separate estimation across the SFAS No. 125 and SFAS No. 140 periods. Specifically, when we estimate the Table 3 and Table 5 models separately for the periods (1997–2000, 2001–3), our key inferences remain unchanged.

ReferencesBarth, M., R. Kasznik, and G. Foster. 1998. Brand values and capital market valuation.

Review of Accounting Studies 3 (1–2): 41–68.Beaver, W. H., P. Kettler, and M. Scholes. 1970. The association between market determined

and accounting determined risk measures. The Accounting Review 45 (4): 654–82.Bond Market Association. 2006. Research Quarterly (February): 7.Bowman, R. 1979. The theoretical relationship between systematic risk and financial

(accounting) variables. The Journal of Finance 34 (3): 617–30.Bowman, R. 1980. The importance of a market-value measurement of debt in assessing

leverage. Journal of Accounting Research 18 (1): 242–54.Collins, D. W., and S. P. Kothari. 1989. An analysis of international and cross-sectional

determinants of earnings response coefficients. Journal of Accounting and Economics 11 (2–3): 143–81.

Collins, D. W., S. P. Kothari, and J. D. Rayburn. 1987. Firm size and information content of price with respect to earnings. Journal of Accounting and Economics 9 (2): 111–38.

Dechow, M., L. Myers, and C. Shakespeare. 2005. Reporting a rosy future today: The role of corporate governance in the reporting of gains from asset securitizations. Working paper, University of Michigan.

Dhaliwal, D. S. 1986. Measurement of financial leverage in the presence of unfunded pension obligations. The Accounting Review 61 (4): 651–61.

Easton, P., and T. Harris. 1991. Earnings as an explanatory variable for returns. Journal of Accounting Research 29 (1): 19–36.



Fabozzi, F. 1998. Issuer perspectives on securitization. New Hope, PA: Frank J. Fabozzi Associates.

Fama, E., and K. French. 1992. The cross-section of expected stock returns. The Journal of Finance 47 (2): 427–66.

Financial Accounting Standards Board (FASB). 1993. Statement of Financial Accounting Standards No. 115: Accounting for certain investments in debt and equity securities. Norwalk, CT: FASB.

Financial Accounting Standards Board (FASB). 1996. Statement of Financial Accounting Standards No. 125: Accounting for transfers and servicing of financial assets and extinguishments of liabilities. Norwalk, CT: FASB.

Financial Accounting Standards Board (FASB). 2001. Statement of Financial Accounting Standards No. 140: Accounting for transfers and servicing of financial assets and extinguishments of liabilities. Norwalk, CT: FASB.

Fitch Inc. 1999. Implications of securitzation for finance companies. Research report. New York: Fitch Inc.

Freeman, R., and S. Tse. 1992. A nonlinear model of security price responses to accounting earnings. Journal of Accounting Research 30 (2): 185–209.

Gorton, G., and N. S. Souleles. 2005. Special purpose vehicles and securitization. Working paper, University of Pennsylvania and NBER.

Hamada, R. S. 1972. The effect of the firm’s capital structure on the systematic risk of common stocks. The Journal of Finance 27 (2): 435–52.

Hayn, C. 1995. The information content of losses. Journal of Accounting and Economics 20 (2): 125–53.

Higgins, E., and J. R. Mason. 2004. What is the value of recourse to asset backed securities? A study of credit card bank ABS rescues. Journal of Banking and Finance 28 (4): 857–74.

International Accounting Standards Board (IASB). 1998. International Accounting Standard (IAS) No. 39: Financial instruments: Recognition and measurement. London: IASB.

International Accounting Standards Board (IASB). 2005. IASB Update (April): 1.Karaoglu, N. E. 2005. Regulatory capital and earnings management in banks: The case of

loan sales and securitizations. Working paper, University of Southern California.Kimmel, P., and T. Warfield. 1995. The usefulness of hybrid security classifications:

Evidence from redeemable preferred stock. The Accounting Review 70 (1): 151–67.Kmenta, J. 1986. Elements of econometrics, 2nd ed. New York: Macmillan.Kothari, S. P., and J. Shanken. 1995. In defense of beta. Journal of Applied Corporate

Finance 8 (1): 53–8.Lipe, R. C. 1990. The relation between stock returns and accounting earnings given

alternative information. The Accounting Review 65 (1): 49–71.Martikainen, M. 1997. Accounting losses and earnings response coefficients: The impact of

leverage and growth opportunities. Journal of Business Finance and Accounting 24 (2): 277–91.

Mian, S. L., and C. W. Smith. 1994. Extending trade credit and financing receivables. Journal of Applied Corporate Finance 7 (1): 75–84.

Obay, L. 2000. Financial innovation in the banking industry: The case of asset securitization. New York and London: Garland Publishing.



Plesko, G. A. 2003. An evaluation of alternative measures of corporate tax rates. Journal of Accounting and Economics 35 (2): 201–26.

Rosett, O. 2001. Equity risk and the labour stock: The case of union contracts. Journal of Accounting Research 39 (2): 337–65.

Ryan, S. 2002. Financial instruments and institutions: Accounting and disclosure rules. New York: Wiley.

Scholes, M., and J. T. Williams. 1977. Estimating betas from nonsynchronous data. Journal of Financial Economics 5 (3): 309–27.

Scism, L. 1998. Some analysts seek accounting-rule change that would slash profits of subprime lenders. Wall Street Journal, September 29, 1.

Shakespeare, C. 2004. Do managers use securitization volume and fair value estimates to hit earnings targets? Working paper, University of Michigan.

Standard & Poor’s (S&P). 2001. Standard and Poor’s corporate ratings criteria. New York: Standard & Poor’s.

White, H. 1980. A heteroskedasticity-consistent covariance matrix estimator and a direct test for heteroskedasticity. Econometrica 48 (4): 817–38.


Documents

Are Securitizations in Substance Sales or Secured ......par le marché (c’est-à-dire, le bêta du MÉDAF), que les passifs ﬁgurant au bilan. Cette observation est conforme au