Market value and accounting implications of off-balance-sheet items

Market value and accounting implications of

o -balance-sheet items

Ray J.Pfei er,Jr.

1

Department of Accounting and Information Systems,School of Management,University of

Massachusetts Amherst,Amherst,MA 01003-4915,USA

Abstract

This paper examines the value relevance of a previously unexplored o -balance-sheet asset:originated mortgage servicing rights.Also,it provides evidence regarding an as-sertion that mortgage bank managers manage their reported earnings by swapping portfolios of originated servicing rights with other servicers.The evidence suggests that despite the absence of originated servicing rights in the balance sheet,they are priced by investors.Regarding the earnings management assertion,the evidence suggests that ?rms facing net losses before recognizing gains on sales of servicing are more likely to recognize discretionary gains,ceteris paribus.This work makes several contributions:(1)It provides evidence about investors'valuation of o -balance-sheet items in a setting where the sample ?rms are homogeneous and the item in question is of substantial economic importance to the ?rms;(2)It provides evidence regarding an o -balance-sheet item that has not been studied previously;(3)It provides a test of earnings management in a setting that does not require the estimation of discretionary accruals;and (4)It investigates an issue of current interest to accounting policy makers and to ?rms in the mortgage banking industry.ó1998Elsevier Science Inc.All rights reserved.

1.Introduction

This paper examines the extent to which o -balance-sheet items are reˉected in ?rms'security prices.In an e cient market,security prices reˉect all value-relevant information conveyed by ?rms'?nancial reports,both the information on the face of the ?nancial statements and in the accompanying

disclosures.

Journal of Accounting and Public Policy 17(1998)

185±207

1

Tel.:1-4135455653;fax:1-4135453858;e-mail:pfei er@https://www.sodocs.net/doc/ac5843749.html,.

0278-4254/98/$19.00ó1998Elsevier Science Inc.All rights reserved.PII:S 0278-4254(98)10005-4

186R.J.Pfei er,Jr./Journal of Accounting and Public Policy17(1998)185±207 However,substantial evidence in the literature suggests that equity markets may be less than perfectly e cient(Bernard and Thomas,1990,p.307;Ou and Penman,1989,p.296;Abarbanell and Bushee,1998,p.19).If this is so,it is likely that the degree of e ciency with respect to certain?nancial information items is related to the quantity and quality(for example,precision)of the in-formation presented by?rms regarding those items.A priori,items measured using Generally Accepted Accounting Principles(GAAP)and presented on the face of the statements may be more e ciently reˉected in prices than o -bal-ance-sheet items,for which less(and less complete)information is provided, usually in the notes to the statements.

A large body of recent work can be characterized as testing this conjecture (for example,Amir and Lev,1996;Barth,1994;Landsman,1986).In general, studies have found evidence that suggests investors do look beyond the face of the statements to information contained in the notes to statements.For ex-ample,Barth(1994,p.12)found that disclosed investment securities fair values are relevant to investors,and Landsman(1986),p.678?nds that disclosed pension assets and liabilities are priced by investors.However,in large-sample studies,it is often di cult to control for unmeasured,value-relevant attributes other than the disclosures of direct interest.Also,while evidence has been presented regarding several speci?c items,such as pension assets and liabilities (Barth et al.,1992;Landsman,1986)post-retirement bene?ts other than pen-sions(Amir,1993),and investments in research and development(Sougiannis, 1994),other o -balance-sheet items remain uninvestigated.

My study focuses on one such uninvestigated o -balance-sheet item:origi-nated mortgage servicing rights.Mortgage servicing rights are contractual rights(separate from the associated mortgages)to receive compensation for performing primarily collection-related duties associated with mortgage loans. These servicing rights,which relate to mortgages originated and then sold by the holder of the rights,are not recorded as assets in the balance sheet under Statement of Financial Accounting Standards No.65(``SFAS65'')(FASB 1982),even though they clearly have economic value.2Servicing rights o er several advantages as a vehicle for the study of the value relevance of o -balance-sheet items.First,for the?rm-years included in my study,servicing rights are a very signi?cant operating asset.Second,the?rms investigated, mortgage banking companies,are a homogeneous sample.Third,mortgage banks tend not to have signi?cant o -balance-sheet items other than servicing rights.Fourth,servicing rights accounting received recent attention of the Financial Accounting Standards Board(FASB)that resulted in the issuance of SFAS122,``Accounting for Mortgage Servicing Rights''(FASB,1995).

2Purchased servicing rights are recognized in the balance sheet as assets under SFAS65(FASB, 1982).

R.J.Pfei er,Jr./Journal of Accounting and Public Policy17(1998)185±207187 Proponents of this new standard,in particular the Mortgage Bankers Asso-ciation of America,argued that security prices fail to reˉect the value of o -balance-sheet servicing rights(Carlton,1993).My research provides evidence that some of that conventional wisdom may have been unfounded. Accounting for mortgage servicing rights under SFAS65(FASB,1982) created unusual?nancial statement e ects.For example,originated servicing rights were immediately expensed when the underlying loans were sold,thus distorting earnings.Also,although originated servicing rights are a primary source of income for some?rms,these assets were excluded from the balance sheet,thus understating total assets.Mortgage bank managers and the?nan-cial press claimed that as a result of this accounting treatment,equity values did not properly reˉect the value of o -balance-sheet servicing rights.3Fur-thermore,it was reported that several mortgage bank managers publicly as-serted that they engaged in otherwise undesirable transactions±including swapping portfolios of servicing rights with other servicers to bring them onto the balance sheet±to alter their?nancial statements for the e ects of servicing rights accounting.4

My study investigates these assertions by posing the following two ques-tions:(1)Do stock prices reˉect estimates of o -balance-sheet assets derived from information available in the notes to the?nancial statements?(2)Is there evidence to support the assertion that?rms enter into transactions designed to manage their reported income and?nancial position in response to perceived undesirable e ects of GAAP for certain transactions?Using a sample of all publicly traded?rms for which mortgage banking is the primary line of busi-ness,I?rst use?nancial statement information and disclosures in the notes to the statements to construct estimates of the value of each?rm's unrecorded servicing rights portfolio.These estimates are included along with recorded balance sheet items as explanatory variables for?rms'market equity values.

A priori,it is not clear which of the extant valuation models in the ac-counting literature is most consistent with observed stock prices.Thus,I also consider an income-statement-based valuation model.Speci?cally,using in-formation in the?nancial statements of adjoining years to estimate the degree of income understatement,?rm market values are regressed on this estimated understatement together with reported income to determine whether reported or adjusted income is more consistent with the income number reˉected in ?rms'stock prices.

To investigate the second question,whether there is evidence consistent with managers manipulating income to undo the?nancial statement e ects of ser-vicing rights accounting,I regressed the amount of recognized gains from sales

3See,for example,the comments in``Long One,Short Another''(1993,p.4).

4See Carlton(1993).

188R.J.Pfei er,Jr./Journal of Accounting and Public Policy17(1998)185±207

of servicing rights(a largely discretionary action by managers)on the pre-gain level of income,along with appropriate control variables.If the assertions are true,I expected to?nd that all else equal,the lower the level of income before gain recognition,the larger will be the discretionary gains recognized by managers.

I found the following:From a balance-sheet valuation perspective,there is evidence consistent with the interpretation that o -balance-sheet servicing is valued,though di erently from on-balance-sheet assets.Tests using an income-statement valuation perspective reinforce the balance-sheet results;the evidence suggests that the estimated income adjustment is signi?cantly associated with equity market values.Stated equivalently,adjusted income is more highly as-sociated with equity values than is reported income.Regarding tests of income manipulation,I found that the level of a?rm's earnings and the?rm's cash position before gain recognition are both signi?cantly associated with the magnitude of gains recognized.This is consistent with earnings management but also with the interpretation that real?nancing considerations are a contributing motivation for managers to recognize discretionary gains on sales of servicing.

2.Background and hypothesis development

2.1.Mortgage servicing rights

Relative to commercial banks,it is disadvantageous for mortgage banks to fund loans,because mortgage banks do not have deposits as a source of low cost?nancing(Imho ,1992,p.5).Mortgage banks'comparative advantage is in originating and servicing mortgages.Within30±60days after originating or purchasing a loan,mortgage banks typically sell all of their loans(see,for example,Arbor National Holdings,Inc.,1994;p.11).Mortgage banks can either sell whole loans or they can form pools of loans and sell(securitize)those loans as mortgage-backed securities.5The mortgage-backed securities can then be sold in the market to receive cash to fund additional new loan originations.

A mortgage loan can be seen as a bundle of discrete assets including(1)a right to receive principal payments,(2)a right to receive interest payments,and (3)a right to receive compensation for performing servicing activities(Eades et al.,1990,p.1).When a mortgage bank sells loans,it has a choice to sell all rights to the loans or to retain selected components.The decision is ultimately

5Several government-sponsored agencies facilitate the securitization of mortgage pools,for example,the Federal Housing Administration,the Veterans Administration,the Federal National Mortgage Association(``Fannie Mae''),and the Federal Home Loan Mortgage Corporation (``Freddie Mac'')(Arbor National Holdings,Inc.,1994,p.9).

R.J.Pfei er,Jr./Journal of Accounting and Public Policy17(1998)185±207189 based on the characteristics of the components as an investment project and the?nancial attributes(for example,capacity and liquidity)of the?rm at the time of the decision.When the?rm chooses to retain the servicing rights on loans that are sold,the servicing rights are known as originated mortgage servicing rights(Arbor National Holdings,Inc.,1994,p.26).

Mortgage banks also acquire servicing rights in other ways.One way is to purchase loans from other lenders,strip the servicing rights,and then sell the loans as described above.Also,there is a market for portfolios of servicing rights,and thus mortgage banks can purchase servicing rights from other servicers.Servicing rights acquired using these two methods are known col-lectively as purchased mortgage servicing rights(Countrywide Credit In-dustries,Inc.,1993,p.31).

2.2.Accounting for mortgage servicing rights

When a mortgage bank sells loans but retains the related servicing rights,the proceeds reˉect only that portion of the total loan value that was transferred (the rights to principal and interest payments).Before the sale,if the mortgage rate paid by the borrower is determined in a competitive market,the recorded amount of the loan on the books of the seller reˉects all of the rights related to the loan,including the servicing rights.Therefore,other things equal(including interest rates in the economy during the short period between origination and sale of the loans),the proceeds received from selling the loan will be less than the book value of the loan.The di erence represents the value of the servicing rights retained(among other things).Under SFAS65(FASB,1982,para.14),?rms are prohibited from recognizing this di erence as an asset,and thus it is recorded as a loss on the sale of the related loans.Mortgage banks,in e ect, expense the acquisition cost of the servicing rights at the time of the sale of the related loan;servicing revenues,however,are recognized as earned over the lives of the related loans.This practice causes a mismatch of revenues and expenses in the income statement and results in all originated servicing rights being o -balance-sheet assets.

One added complication is the nature of the loan sale itself.There are ef-fectively two prices involved in the sale.The buyer pays a price for the right to the principal and interest payments;the buyer also agrees to pay a periodic fee to the seller for performing the servicing.These two prices permit an in?nite number of possible sale prices.Since the acquisition cost of the servicing rights is expensed at the time of the sale,without any restriction,it would be possible to structure the contract such that the seller could manage the size of the ac-counting loss.SFAS65stipulates that any gain or loss recognized on the sale of loans with servicing retained must be adjusted to recognize a normal servicing fee rate over the expected life of the loans(FASB,1982,para11).The ad-justment to the gain or loss is capitalized and is usually referred to as an excess

190R.J.Pfei er,Jr./Journal of Accounting and Public Policy17(1998)185±207 servicing asset(Arbor National Holdings,Inc.,1994,p.F-9).This provision eliminates some of the latitude that?rms have with respect to the amount of gain or loss recognized.However,to the extent that the loan purchaser is ˉexible as to how the contract is structured,the provision also creates an op-portunity for loan servicers to bring some of the otherwise o -balance-sheet servicing into the balance sheet.6

Economically,servicing rights are an asset of the?rm whether originated or purchased.Prevailing accounting treatment of servicing rights causes several problems with mortgage bank?nancial statements:

·Although servicing revenues comprise a substantial portion of the total rev-enue of mortgage banks,assets that generate such revenue are largely absent from the balance sheet.

·When a loan is sold with servicing retained,the value of the servicing rights is treated as an accounting loss.Thus income is understated relative to econom-ic income in the period that the servicing rights are originated and is overstat-ed each subsequent year until the underlying loans mature or are pre-paid.·Firms often sell portfolios of servicing rights to other servicers.Because most of these rights are o -balance-sheet and thus have an accounting basis of zero,each dollar of servicing sold generates a dollar of pre-tax pro?ts±an overstatement of the economic gain on the transaction,and an opportunity for management of reported income.

·In the securitization transaction,the agency that guarantees or insures the mortgage-backed securities bears some of the credit losses on the underlying mortgages in exchange for a share in the future cashˉow stream of the pool. This share is e ectively a payment from the loan originator to the guarantor to purchase the guarantee.This guarantee is an asset to the originator,since it represents the originator's right to be reimbursed for any credit losses on the pool,including some costs of foreclosure.However,like the related ser-vicing rights,this asset is absent from the balance sheet(Arbor National Holdings,Inc.,1994,p.9)

·To avoid some of these problems,many mortgage banks trade servicing rights portfolios,(Carlton,1993).When a mortgage bank sells its originated servicing rights,it recognizes a gain equal to their value.By using the pro-ceeds of the sale to purchase an equal amount of servicing rights from anoth-er?rm,it can record the new servicing rights as purchased servicing.These

6In practice,transactions settle within a fairly narrow range of normal servicing fees.The contractual servicing fee is bounded from above by restrictions placed on the size of the loss by SFAS65's required capitalization of excess servicing fees,and bounded from below by the fact that a loan investor would not want e ectively to pay the servicer in advance for all of the future servicing,as this would provide an incentive for the servicer to shirk in later years of the contract when the servicing fees fall short of servicing costs.Ultimately the sales contract is settled based upon the characteristics of the marginal loan buyer and seller.

R.J.Pfei er,Jr./Journal of Accounting and Public Policy17(1998)185±207191 transactions e ectively undo the o -balance-sheet treatment of SFAS65,but they are not costless.7

My paper attempts to investigate the extent to which market prices reˉect information about values of servicing rights portfolios despite these short-comings of mortgage banks'?nancial statements.

2.3.Mortgage bank equity valuation models

2.3.1.A balance sheet valuation perspective

Several recent studies have modeled the value of?rms'common equity as being based upon the net asset value of the?rm(e.g.Barth,1993,p.6).From this perspective,

MVE MVA MVL Y 1 where MVE is the market value of the?rm's equity,MVA is the market value of its assets(on-and o -balance sheet),and MVL is the market value of its(on-and o -balance-sheet)liabilities.This perspective assumes that market values of assets and liabilities are linearly and additively related to equity value. One of the objectives of my paper is to assess to what extent o -balance-sheet mortgage servicing rights are reˉected in equity values.This suggests decomposing MVA to isolate o -balance-sheet servicing:

MVA OBSMSR OTHASSETS Y

where OBSMSR is the value of o -balance-sheet mortgage servicing rights,and OTHASSETS is the value of all assets other than OBSMSR.Furthermore, under the nulls of market e ciency and Eq.(1)as representative of equity value,the weights on OBSMSR,OTHASSETS,and MVL are implicitly1,1, andà1,respectively.In empirically estimating these weights in a cross-sec-tional regression context,di erences in the relation across?rms,di erences in the measurement accuracy for the independent variables,and di erences in market-perceived risk of the?rm are all expected to cause the estimated weights to di er from theoretical weights.Together with the decomposition of MVA,this leads to the following,empirical version of Eq.(1)

MVE it a0 a1OTHASSETS it a2OBSMSR it a3MVL it e it Y 3

7Despite the presence of a somewhat active secondary market in servicing rights,the transactions costs(including the bid/ask spread,commissions paid to market makers,costs of evaluating the quality of the purchased servicing rights,and the personnel resources used in executing the transactions)are likely to be non-trivial.Furthermore,the information set held by a mortgage bank about its own originations is likely to be more extensive than that available from another party's servicing rights.This potentially makes originated servicing rights more valuable(per dollar of principal serviced)than those received in a swap.

where,under the(joint)null,a0equals zero,a1 a2 1Y and a3 à1.The?rst question addressed in this paper focuses on a2.Previously mentioned assertions by mortgage bank managers and the?nancial press suggest that servicing rights accounting causes under-appreciation of the value of o -balance-sheet servicing.This implies an expectation that a2`a1,and possibly a2 0.

2.3.2.An income statement valuation perspective

A priori,it is not clear which of the extant valuation models in the ac-counting literature is most consistent with observed stock prices.Moreover, di erences across speci?cations in the measurement accuracy of explanatory variables can lead to di erent inferences(Barth,1993,p.8).Thus as a test of robustness,I also consider a permanent-income-based model of value(Miller and Modigliani,1966;p.336).

In a setting of certainty,the value of any asset is described by the following equation:

i

r

Y 4

where is the value of the asset,i is the permanent income generated by the asset,and r is the discount rate±that rate that adjusts a perpetual cashˉow stream for the time value of money.In an uncertain setting,r will also include adjustment of i for risk,and i represents expected permanent income.Prior studies have used this model to represent?rms'equity values as a function of accounting earnings(Miller and Modigliani,1966,p.353;Barth et al.,1992, p.30).

Certainly accounting income and permanent income are unequal.However, in this setting,we know one speci?c cause of the di erence between accounting income and permanent income:the amount of originated mortgage servicing rights generated in the period,which are incorrectly expensed in recording the sale of the underlying loans.Adjusting accounting income for this known dis-tortion should improve the ability of accounting income to proxy for permanent income.Consider the following characterization of permanent income(i t):

i t t ADJ t t Y 5 where t is year t reported accounting earnings, ADJ t is the amount of in-correctly expensed originated mortgage servicing rights generated,and t is the sum of all di erences between accounting income and permanent income other than ADJ t X8

Recall that some mortgage banks are alleged to be swapping portfolios of originated servicing rights to manage their?nancial reports.If so,the

8 t also includes the di erence between expected permanent income and the ex post proxy for permanent income,as expected permanent income is unobservable.

192R.J.Pfei er,Jr./Journal of Accounting and Public Policy17(1998)185±207

component of income that results from such transactions(the gain on sales of servicing rights)is likely to be valued di erently by investors.In particular, managed components of earnings tend to have lower valuation coe cients than unmanaged components because they are viewed by investors as transi-tory.This motivates a further modi?cation of Eq.(5)

i t pre-gain

t

MSRGAIN t ADJ t t Y 6

where pre-gain

t is accounting earnings excluding reported gains from sales of

originated servicing MSRGAIN t .Substituting Eq.(6)into Eq.(4)and al-lowing b i to represent the inverse of each income component's distinct discount rate yields the income-statement-based valuation model used in my paper:

MVE t b0 b1 pre-gain

t

b2MSRGAIN t b3 ADJ t l t X 7 If mortgage banks'income statements are incomplete,that is,they incor-rectly treat newly generated originated servicing rights as expenses,and if in-vestors recognize this in valuing mortgage bank equity,then b3b0.If investors view gains on sales of originated servicing as managed components of income,then b2`b1.

2.3.3.Model of determinants of discretionary gain recognition

Based on discussions with mortgage bank managers and analysts and on other sources of information about mortgage banking,I sense that the in-centives for selling servicing rights include the following:(1)cashˉow needs±the proceeds from such sales may satisfy liquidity demands;(2)tax position, since gains on servicing are taxable;(3)net income prior to the contemplated gain recognition(mortgage bank analysts assert that managers use servicing sales to augment operating income when it is perceived as too low);and(4)the prevailing market price for servicing rights±servicing sales are more appealing if at a given time prices are relatively favorable.9The exact form of the relation among these factors and the magnitude of gains on sales of servicing rights is not known,but there is reason to expect that it is nonlinear(see footnote9). Accordingly,to accommodate such non-linearity I included the above factors in a model categorically rather than using(continuous)magnitudes of the variables.This approach assigns each?rm-year in the sample to a category based on its earnings,tax position,and cash position(prevailing market prices of servicing rights are unavailable,and thus that factor is excluded from the model).The earnings dummy variable is set equal to one(zero)if earnings

9Note that(3)implies a kinked,nonlinear relation between net income before gains and gains taken;that is,when net income is considered too low there is a negative relation between sales of servicing and net income,but at a certain point±where net income is no longer too low±this source of motivation for selling servicing no longer exists.

R.J.Pfei er,Jr./Journal of Accounting and Public Policy17(1998)185±207193

194R.J.Pfei er,Jr./Journal of Accounting and Public Policy17(1998)185±207 before sales of servicing are negative(positive).Arguments made by mortgage bank managers and in the?nancial press suggest that?rms with negative earnings have the strongest incentive to recognize gains.The tax dummy variable is set equal to one if the?rm is facing relatively larger positive mar-ginal tax rates.Firms that report net operating loss carry forwards in a given year are presumed to have smaller marginal tax rates.Firms facing larger marginal tax rates will have a disincentive to recognize gains,since those gains would be taxed.The cash dummy variable is set equal to one if the?rm's cash and cash equivalents before sales of servicing as a fraction of total liabilities is below the annual cross-sectional median.I assume that?rms below the median cash/debt ratio are those with the greatest incentive to sell servicing for li-quidity reasons.

In addition,cash-and tax-related incentives to sell servicing are expected to be conditional on the earnings management incentives of the?rm.For ex-ample,the incentive to recognize gains for liquidity reasons may be much stronger if such a decision also makes sense from an earnings perspective. Similarly,in isolation,the taxability of servicing sale gains is a clear disin-centive,although?rms may be willing to recognize such gains when there are strong earnings-related incentives.To investigate these conjectures,I also ex-amine the conditional relation between the cash-and tax-related incentives for ?rms with positive and negative pre-gain earnings.

The three dummy variables are used in the following model:

MSRGAIN it c0 c1EARNINGS it c2TAX it

c3CASH it c4EARNINGS itáTAX it

c5EARNINGS itáCASH it l it Y 8 where MSRGAIN is the reported gain on sales of servicing rights scaled by total revenue(a size scaler),EARNINGS is equal to one for all?rm-years where reported earnings are negative and zero otherwise,TAX is equal to one for?rm-years with no operating loss carry-forwards and zero otherwise,and CASH is equal to one for?rm-years where pre-gain cash scaled by total lia-bilities is below the yearly cross-sectional median.Predicted signs are c1b0Y c2`0Y c3b0Y c4b0Y and c5b0.

3.Empirical tests

3.1.Measuring key variables

Eqs.(3)and(7)include two servicing-related variables,OBSMSR(the o -balance-sheet servicing asset)and ADJ(the servicing-related income

R.J.Pfei er,Jr./Journal of Accounting and Public Policy17(1998)185±207195 adjustment),that are not disclosed by?rms and consequently must be estimated.Two di culties arise when measuring OBSMSR.First,servic-ing rights are inherently di cult to value,because their values depend on assumptions about the future behavior of borrowers and the future paths of inˉation and interest rates in the economy.Second,?rms typically disclose the total principal balance of mortgages serviced but not the esti-mated value of servicing rights nor the details necessary to make an inde-pendent assessment.Furthermore,this disclosure does not identify the on-versus o -balance-sheet breakdown of the total principal serviced.Therefore, to obtain a measure of the o -balance-sheet servicing value,I model OBSMSR as OBSMSR t cfáPRINCIPAL tàCAPMSR t,where PRINCI-PAL is the amount disclosed as total principal serviced by?rms,cf is a factor to convert dollars of principal serviced to dollars of servicing value,and CAPMSR is the total on-balance-sheet servicing(purchased servicing plus excess servicing).The conversion factor is arbitrarily set equal to0.02(200 basis points)during the period encompassed by the study based on repre-sentations by mortgage bank managers and market makers.10Varying the conversion factor does not impact the conclusions drawn regarding the im-portance of this variable in the model.OTHASSETS and MVL are de?ned as the total recorded value of on-balance-sheet assets,and liabilities and pre-ferred stock,respectively.

The general approach to measuring ADJ is to recognize that the distor-tion of income is equal to the net of additions to and amortization of originated servicing rights.This can be estimated if all other activity in the?nancial statements related to servicing rights can be identi?ed or estimated using the ?nancial statements and notes.The empirical measure of ADJ is obtained using the following expression:

ADJ t PRINCIPAL tàPRINCIPAL tà1 ácf

à CAPMSR tàCAPMSR tà1 MSRGAIN t Y 9

10As an illustration of the estimation of OBSMSR for a sample?rm-year,Margaretten Financial Corporation(1994),p.32reported in note14to its1993consolidated?nancial statements that``the unpaid principal balance of mortgage loans serviced by the Company as of December31,1993... was approximately615X6billion''.The consolidated balance sheet for1993showed net purchased mortgage servicing rights(net of accumulated amortization)equal to6124X6million.Thus for this ?rm year,OBSMSR would be equal to0.02 615Y600Y000Y000 à6124Y600Y000 6187Y400Y000. Based on my review of the?nancial statements of all sample?rms,Margaretten's disclosure is typical of the disclosures made by sample?rms.

where MSRGAIN is the reported gain on sales of mortgage servicing rights,and other variables are as de?ned above.11,12The remaining variable in Eq.(7), pre-gain t is measured as reported income from continuing operations less re-ported gains on sales of servicing rights.

As discussed in the Appendix,the measurement of ADJ is only possible when one assumes that all servicing rights sales are sales of originated servicing.However,as discussed in Section 2,gains on servicing rights sales are taxable,and thus since gains on capitalized servicing are smaller (they are net of the recorded cost of the servicing),managers of ?rms facing positive marginal tax rates might choose to sell capitalized servicing.To test for this possibility,I estimate Eq.(7)separately for ?rms expected to be facing zero (i.e.very small)marginal tax rates and those with positive marginal tax rates (based on the existence of disclosed net operating loss carry-forwards).If ?rms facing posi-tive marginal tax rates are selling capitalized servicing,it would cause mea-surement error in ADJ for those ?rms to be greater than for ?rms facing zero marginal tax rates.The e ect of the added measurement error would be to increase ADJ because MSRGAIN would be overstated (see Eq.(9)).Thus,all else equal,the coe cient on ADJ in the positive tax subsample would be smaller (or at least less precise)than in the zero marginal tax rate subsample.The results (not reported)indicate that the adjustment coe cient is signi-?cant in both subsamples;however the magnitude of the coe cient is actually larger for positive marginal tax rate ?rms (0.901compared to 0.490for zero marginal tax rate ?rms).Thus the assumption used in constructing ADJ does not appear to cause incorrect inferences.

11

To illustrate,Margaretten Financial Corporation reported in note 14to its 1993?nancial statements that the principal balance of mortgages serviced is 615.6billion as of December 31,1993and 614.4billion as of December 31,1992.The 1993and 1992consolidated balance sheets show (capitalized)purchased mortgage servicing rights of 6124X 6million and 6156X 3million,respectively.And the 1993consolidated statement of income shows revenue from sales of servicing rights of 654X 2million.Thus for this ?rm-year, ADJ would equal (615.6billion A 614.4billion)á0.02A (6124.6million A 6156.3million)+654.2million 6109million.Details on deriving ADJ,the adjustment to accounting income,are located in the appendix.12

ADJ is likely measured with error.To the extent that the error is random,this reduces the chances of ?nding signi?cance for this income adjustment.A priori,there is no clear reason to expect that the error is nonrandom.While the measurement techniques are admittedly crude,they are consistent with techniques used by some analysts in the mortgage banking industry.Furthermore,one mortgage bank manager,in a telephone conversation with me,indicated that analyses similar to the one used in this paper were used at his mortgage bank to adjust net income for purposes of bonus computations for mortgage bank o cers.

196R.J.Pfei er,Jr./Journal of Accounting and Public Policy 17(1998)185±207

R.J.Pfei er,Jr./Journal of Accounting and Public Policy17(1998)185±207197 3.2.Data and sample selection

The sample consists of all publicly traded mortgage banks for whom annual reports and stock price data are available for the period1983±1994.Financial statement data for the study are hand-collected from annual reports.Security price data are obtained from COMPUSTAT,or when COMPUSTAT did not have prices,from Standard and Poor's Daily Stock Price Record.

3.3.Econometric issues

Both models are estimated in undeˉated form.This potentially leads to two scale-related problems:scale bias and heteroskedastic disturbances.Following Barth and Kallapur(1996),p.529,I address scale bias by including a proxy for size in each model.Given that a substantial operating asset,originated mort-gage servicing,is o -balance-sheet,traditionally used measures of size such as total assets or book value of equity are inappropriate.Though some servicing is o -balance-sheet,all servicing revenuesˉow through the income statement.I therefore use total revenue(REVENUE)as a size proxy.Heteroskedasticity is addressed by using White-corrected t-statistics,again following procedures described in Barth and Kallapur(1996),p.529.

Models are estimated on a pooled sample of?rm-years to increase the number of available observations.Given the industry concentration of the sample and the use of some?rms in several years,it is highly likely that dis-turbances will be correlated across observations.This biases traditional t-sta-tistics.To alleviate this problem,I use bootstrapping and report statistical signi?cance based on empirical distributions of the estimated coe cients.13 Finally,when attempting to?nd value relevance for one o -balance-sheet item,it is di cult or sometimes impossible to identify and measure all other relevant o -balance-sheet items.Also,unobservable di erences between mar-ket and book values of balance-sheet items can confound inferences.In this setting,however,omitted o -balance-sheet items such as unrecorded pensions or lease obligations are not expected to be a problem.Based on my review of all sample?rms'annual reports,few mortgage banks have de?ned bene?t pension plans,and the numbers of employees are small relative to average?rms.Thus the magnitudes of unrecorded pension assets and liabilities(and other post-employment liabilities other than pensions)are small or nonexistent for most ?rms.Likewise,leased assets are not a substantial portion of total assets for ?rms in this sample.Nevertheless,in Section4I discuss additional analyses aimed at eliminating these potential sources of ambiguity in the results.

13Given the small sample size,results are likely to be sensitive to extreme observations; accordingly observations where studentized residuals exceed two are deleted in all analyses.

198R.J.Pfei er,Jr./Journal of Accounting and Public Policy17(1998)185±207

4.Results

Table1shows descriptive statistics for variables used in the three regression models.Note that there is substantial variation in the sizes of sample?rms in both the balance-sheet and income-statement tests.Median estimated o -bal-ance-sheet servicing value is approximately9%of median total recorded assets. The range(not reported in the table)is from one-half of one percent to nearly 100%.The median estimated income adjustment, ADJ,is approximately16 million,nearly three times median net income(not reported in the table). Clearly the servicing rights items can be of economic signi?cance to many ?rms.Also,the median gain is large relative to pre-gain income;this illustrates the degree of discretion over reported earnings that o -balance-sheet servicing provides to managers.

Table1

Descriptive statistics for variables used in regression models(6millions)

Variable n"x r Median Balance-sheet-based valuation model

MVE65185.4337.347.1 OTHASSETS651085.62006.1343.4 LIABS65969.51841.9298.9 OBSMSR6581.2123.030.3 REVENUE65113.1149.849.8

Income-statement-based valuation model

MVE37136.1230.556.5

pre-gain37 2.829.20.2 MSRGAIN3712.719.8 5.1

ADJ3736.644.615.7 REVENUE3791.5102.449.8

Discretionary gain recognition model

MSRGAIN450.0430.0560.018

pre-gain45A0.0640.5760.067 CASHNEEDS450.0970.2230.011 MVE is the market value of common equity for?rm i on the last day of?scal year t.OTH-ASSETS is the book value of all recorded assets.LIABS is the book value of liabilities and all non-common equity.OBSMSR is the estimated value of o -balance-sheet servicing rights.REVENUE is total revenues as reported in the income statement. pre-gain is reported accounting earnings before extraordinary items and discontinued operations and before reported gains on sales of mortgage servicing rights. ADJ is the estimated adjustment to earnings to correct for the e ects of accounting for servicing rights.MSRGAIN is the ratio of recognized gains on sales of servicing rights to total revenue;and CASHNEEDS is the ratio of cash and cash equivalents to total lia-bilities.

R.J.Pfei er,Jr./Journal of Accounting and Public Policy17(1998)185±207199 Results for the balance-sheet valuation model,presented in Table2,indicate that estimated o -balance-sheet servicing appears to be priced.Its coe cient (0.659)is signi?cantly positive(according to all three measures of signi?cance, OLS t,White-corrected t,and bootstrap p values)and indistinguishable from the coe cient on recorded balance sheet assets(v2from test of equality of coe cients using White-corrected covariance matrix 0X32Y p 0X5718).This result suggests that despite the absence of recognition of the value of originated servicing,investors apparently use information from the notes to the state-ments(such as has been done in estimating OBSMSR±see footnotes10and 11),from?nancial analysts,or other sources of information to infer the con-tribution of such assets to equity value.14

As discussed above,in the absence of measurement error or speci?cation error,the expected coe cients on OTHASSETS and LIABS are1.0andà1X0, respectively.Table2shows that empirically,the OTHASSETS(LIABS)co-e cient di ers signi?cantly from1.0(à1X0),and the coe cient on LIABS di ers signi?cantly in magnitude from that on OTHASSETS.This is a com-mon?nding in studies that use balance sheet valuation models(e.g.Barth et al.,1993,p.25report coe cients on assets and liabilities ranging from0.42to 0.84and fromà0X43toà0X89respectively).Nonetheless,it suggests cautious interpretation of the results.The income-statement-based tests that follow help to assess the robustness of the inferences from the balance-sheet-based tests. To investigate the possibility that other o -balance-sheet items are causing incorrect inferences,I test for three speci?c items suggested to be important based on prior research:pension assets and liabilities,operating leases,and di erences between the market values and book values of?nancial instruments. Consistent with measures used by Barth et al.(1993),p.15,I add total pension assets and the projected pension bene?t obligation for each?rm year to Eq.(3).Of the25?rms in the estimation sample,only four?rms have de?ned bene?t pension plans.Median plan assets and projected bene?t obligation as a

14If one assumes that OBSMSR perfectly measures the value of servicing rights and that the balance sheet valuation model is an appropriate representation of market prices,then the coe cient on OBSMSR should equal one.The fact that the coe cient di ers from one has several alternative explanations,including the possibility that investors mis-assess the implications of o -balance-sheet servicing for?rms'equity values as asserted by proponents of SFAS122(FASB,1995).However, the strict assumptions required to make this inference are unlikely to hold in practice,and thus other explanations,such as error in measuring OBSMSR cannot conclusively be ruled out.The inference from the tests in Table2is that information about o -balance-sheet servicing rights found in the notes to the statements is value-relevant.I do not draw any inference regarding whether investors value o -balance-sheet servicing appropriately.Furthermore,that aside,a ?nding that the information is value-relevant does not preclude bene?ts from disclosure such as mandated by SFAS122.

fraction of total market value of equity for those ?rm-years with pension plans are 0.090and 0.067,respectively.

The relevant o -balance-sheet facet of operating leases is the net of the unrecognized right to use productive assets under the leases and the unrecog-nized obligation to make future rent payments.Empirically,this is di cult to measure,because under GAAP ?rms only disclose the aggregate future mini-mum lease payments under noncancelable operating leases.Nevertheless,I collected such disclosures for all ?rm-years and added this variable to Eq.(3).Median total future minimum lease payments as a fraction of market value of equity is 0.110.As a fraction of total recorded liabilities,the median total future minimum lease payments is 0.016.

I also collected ?rms'disclosures of fair values of ?nancial instruments under SFAS 107(FASB,1991).Di erences between fair values and recorded amounts may also be considered o -balance-sheet items.Accordingly,I de?ned two variables that capture this di erence,one for ?nancial instrument assets and one for liabilities,and added these variables to Eq.(3).The median asset

Table 2

Estimation of the balance-sheet-based valuation model

a

a 0

a 1

a 2

a 3a 4 a

j A 25.240

0.559A 0.4770.6590.120OLS t a j A 2.25?13.21?A 10.48? 4.77?0.86White t a j A 2.87?9.07?A 6.99? 4.38?0.78Mean a

j b A 25.193?

0.558?

A 0.477?

0.661?

0.118

n 65:25distinct ?rms,10years. 2 0X 96;overall F-test:p 337X 85 p 0X 0001 Additional Tests c

White's test:v 2 43X 608 p 0X 0001

a 1 1X 0X Reject;v 2 51X 326 p 0X 0000 a 2 à1X 0X Reject;v 2 59X 422 p 0X 0000 a 1 àa 2X Reject;v 2 76X 257 p 0X 0000

a 3 a 1X

Fail to reject,v 2 0X 320 p 0X 5718

MVE is the market value of common equity for ?rm i on the last day of ?scal year t .OTH-ASSETS is the book value of all recorded assets.LIABS is the book value of liabilities and all non-common equity.OBSMSR is the estimated value of o -balance-sheet servicing rights.REVENUE is total revenues as reported in the income statement.The sample consists of all publicly traded mortgage banks for which complete data are available and for which the estimated amount of o -balance-sheet servicing rights is positive.Additionally,six observations are deleted because they have a studentized residual exceeding 2.0in absolute value.?

Signi?cant at a T 0X 05using a two-tailed test.For bootstrap-generated estimates,?indicates that a 95%con?dence interval around the mean of the empirical parameter estimates does not include zero.a

Model:MVE it a 0 a 1OTHASSETS it a 2LIABS it a 3OBSMSR it a 4REVENUE it e it .b

Mean is from a sample of 1000sets of coe cients generated by a bootstrapping procedure.c

Coe cient tests are chi-square tests based on the estimated asymptotic covariance matrix of the estimates under the hypothesis of heteroskedasticity.

200R.J.Pfei er,Jr./Journal of Accounting and Public Policy 17(1998)185±207

R.J.Pfei er,Jr./Journal of Accounting and Public Policy17(1998)185±207201 and liability di erences as a fraction of market value of equity for the11?rm-years for which SFAS107disclosures were present are0.038and0.033,re-spectively.

Overall,the inferences regarding the value-relevance of o -balance-sheet servicing rights are una ected by inclusion of these additional variables.The results of the tests(not reported)are as follows:(1)Neither the pension asset nor the pension liability variables are signi?cant when added to Eq.(3).The coe cient on OBSMSR declines slightly to0.578but remains signi?cant (White-corrected t 2X69).(2)Adding the lease obligation variable to Eq.(3) has very little e ect on OBSMSR's coe cient;it declines slightly to0.592but remains signi?cant(White-corrected t 4X61).The lease obligation variable itself is not signi?cant.(3)The fair value di erence for assets is signi?cant when added to Eq.(3),but the fair value di erence for liabilities is not.The coe -cient on OBSMSR increases slightly to0.765and remains signi?cant(White-corrected t 4X20).Finally,in a model that includes all candidate omitted variables(pension asset and liability,lease obligations,and fair value di er-ences for?nancial instruments),the fair value di erence for assets is the only additional variable that is signi?cant.In this model,the coe cient on OBSMSR declines to0.288(White-corrected t 1X49)which is marginally signi?cant p`0X071,one-tailed test).Note that four of the?ve additional variables are not incrementally correlated with market value of equity,and thus their inclusion adds considerable noise to the model and depresses the signi?cance of all remaining variables.In sum,the value-relevance of o -bal-ance-sheet servicing rights information appears to be robust to other measured o -balance-sheet items.

Table3shows the results of estimating the income-statement-based valua-tion model with the correction to reported income.The statistically signi?cant, positive coe cient for the earnings adjustment variable indicates that income as adjusted is more highly associated with market equity value than reported income.The adjustment coe cient is signi?cantly smaller than that on the rest of income.This would be expected if the adjustment were less permanent than other sources of income or if the former were measured with more measure-ment error.However,given that the adjustment is simply adding back a loss that economically should not have been recognized,there is no reason to be-lieve that it would be of a di erent character than the rest of income.

The estimated gain coe cient is signi?cantly smaller than that on the rest of income.In most settings,gains and losses are non-repeating events and thus are expected to be viewed by market participants as transitory and to be given lower valuation multiples than other sources of income.In this setting,however,the nature of servicing rights accounting guarantees that?rms will be able to rec-ognize gains on a continuing basis as long as they continue to originate and sell new mortgage loans.On the other hand,to the extent that the timing of gain recognition is discretionary,investors may assign lower multiples to gains.For

this reason,the ?nding that the coe cients are lower than that on the rest of income is plausible.It also reinforces the motivation for the ?nal set of tests ±the investigation of determinants of the magnitude of recognized gains.

Table 4focuses on the factors that potentially inˉuence the recognition of gains on servicing rights sales.The hypothesis suggested by the assertions made in the ?nancial press is that ?rms with lower net income (especially losses)swap portfolios with other servicers and recognize gains to increase their reported income.The evidence only weakly supports this claim,however.Across all sample observations,the coe cient on EARNINGS is positive,but only sig-ni?cant according to the White-corrected t .It is not signi?cant using the OLS t or the bootstrap metric.The strongest factor appears instead to be the liquidity motivation.The coe cient estimate on CASH is positive and signi?cant using all three signi?cance measures.On the overall sample,the tax motivation ap-pears to be unrelated to the level of gain.

When the data are partitioned based on positive/negative pre-gain earnings,the evidence suggests that cash needs motivations become much stronger for

Table 3

Estimation of the income-statement-based valuation model

a

b 0

b 1b 2

b 3b 4 b j

A 17.075

6.250 3.1880.9980.643OLS t b j A 1.8110.89? 2.59? 3.06? 3.85?White t b j A 2.93? 6.29? 1.90 2.91? 2.88?Mean b j b

A 16.857?

6.266?

3.176?

1.000?

0.638?

n 37X 16distinct firms Y 8years Y 2 0X 97Y

White v 2 25X 197 p 0X 0327 Y overall F-test X p 281X 554 p 0X 0001 Additional tests c

b 1 b 2Reject;v 2 11X 528Y p 0X 0070b 1 b 3

Reject;v 2 19X 598Y p 0X 0000

MVE is the market value of common equity for ?rm I on the last day of ?scal year t . pre-gain is reported accounting earnings before extraordinary items and discontinued operations and gains on sales of servicing rights (MSRGAIN). ADJ is the estimated adjustment to earnings to correct for the e ects of accounting for servicing rights.REVENUE is total revenue as reported in the income statement.The sample consists of all publicly traded mortgage banks for which complete data are available and for which accounting earnings and ADJ are positive.Additionally,three obser-vations are deleted because they have a studentized residual exceeding 2.0in absolute value.?

Signi?cant at a T 0X 05using two-tailed test.For bootstrap-generated estimates,?indicates that a 95%con?dence interval around the mean of the empirical parameter estimates does not include zero.

a

Model:MVE it b 0 b 1 pre-gain it b 2MSRGAIN it b 3 ADJ it b 4REVENUE it e it X b

Mean is from a sample of 1000sets of coe cients generated by a bootstrapping procedure.c

Coe cient tests are chi-square tests based on the estimated asymptotic covariance matrix of the estimates under the hypothesis of heteroskedasticity.

202R.J.Pfei er,Jr./Journal of Accounting and Public Policy 17(1998)185±207

?rms with negative pre-gain earnings.The cash needs variable remains positive,and statistical signi?cance increases c 3 c 5 0X 145Y v 2 85X 949Y p 0X 0000 .The tax variable goes from being insigni?cant to negative and signi?cant c 2 c 4 à0X 051Y v 2 11X 113Y p 0X 0009 .It is not clear why the tax dis-incentive would be stronger for ?rms with negative earnings.5.Summary and conclusions

In this paper,I examine the value relevance of a previously uninvestigated o -balance-sheet asset:originated mortgage servicing rights.Also,I provide evidence regarding an assertion that mortgage bank managers manage their reported earnings by swapping portfolios of originated servicing rights with other servicers.

This work (1)extends the literature by providing evidence about investors'valuation of o -balance-sheet items in a setting where the sample ?rms are

Table 4

Estimation of the discretionary gain recognition model

a

c 0

c 1c 2c 3c 4c 5 c

j 0.004

0.0300.0140.041A 0.0650.104OLS t c j 0.29 1.690.94 3.07?A 1.98 3.87?White t c j 0.44 2.08?0.78 3.16?A 2.79? 5.08?Mean c

j b 0.006

0.047

0.008

0.048?

A 0.058

0.060

n 45X 16distinct firms Y 8years Y 2 0X 61Y Overall F-test X p 12X 148 p 0X 0001 Additional tests c

White's test Reject;v 2 14X 355 p 0X 0452 c 2 c 4 0Reject;v 2 11X 113 p 0X 0009 c 3 c 5 0

Reject;v 2 85X 949 p 0X 0000

MSRGAIN is the ratio of recognized gains on sales of servicing rights to total revenue;EARNINGS is equal to one if earnings before extraordinary items and discontinued operations and before recognition of gains is negative and zero otherwise;TAX indicates those ?rms that have net operating loss carryforwards in a given year TAX 1 ;CASH indicates those ?rms who have below-median cash and cash equivalents relative to total liabilities in a given year CASH 1 .The sample consists of all publicly traded mortgage banks for which complete data are available.Additionally,?ve observations are deleted because they have a studentized residual exceeding 2.0in absolute value.?

Signi?cant at a T 0X 05using a two-tailed test.For bootstrap-generated estimates,?indicates that a 95%con?dence interval around the mean of the empirical parameter estimates does not include zero.a

Model:MSRGAIN it c 0 c 1EARNINGS it c 2TAX it c 3CASH it c 4EARNINGS it á

TAX it c 5EARNINGS it áCASH it l it .

b

Mean is from a sample of 1000sets of coe cients generated by a bootstrapping procedure.c

Coe cient tests are chi-square tests based on the estimated asymptotic covariance matrix of the estimates under the hypothesis of heteroskedasticity.

R.J.Pfei er,Jr./Journal of Accounting and Public Policy 17(1998)185±207

203

204R.J.Pfei er,Jr./Journal of Accounting and Public Policy17(1998)185±207 highly homogeneous and the item in question is of substantial economic im-portance to the?rms;(2)provides evidence regarding an o -balance-sheet item that has not been studied previously;(3)provides a test of earnings manage-ment in a setting that does not require the estimation of discretionary accruals; and(4)investigates an item of current interest to accounting policy makers and to?rms in the mortgage banking industry.

The results suggest that mortgage servicing rights are priced even when they are not reported in?nancial statements.This conclusion is supported from both a balance-sheet-based and income-statement-based valuation perspective. Also,the evidence suggests that liquidity is a stronger motivation for entering into sales of servicing rights than an earnings-management-based incentive. Such?ndings contradict assertions in the?nancial press regarding the pricing of mortgage bank equity and the behavior of mortgage bank managers.

Acknowledgements

I wish to thank Wayne Landsman,my dissertation chair,and the other members of my committee:Mary Barth,Robert Eisenbeis,John Hand,and James Wahlen,for their guidance and helpful comments and suggestions.I also bene?tted from the contributions of Bill Baber,Ben Branch,Julie Collins, David Downs,Pieter Elgers,Greg Geisler,Susan Machuga,Krishna Palepu, Gary Patterson,Pamela Pfei er,several anonymous referees,and accounting workshop participants at the University of Massachusetts,Michigan State University,University of New Hampshire,University of Tennessee,and Washington University.I am also grateful for the?nancial support of the Arthur Andersen&Co.,SC Foundation.

Appendix A

A.1.Details on the measurement of eht

In the period when servicing rights are separated from underlying mortgages as part of the securitization or sale of the mortgages,income is understated by the value of the servicing rights retained,because SFAS65(FASB,1982;para 14)prohibits recognition of the servicing rights as an asset.In all subsequent periods until the underlying loans mature,they are repaid,or the servicing rights themselves are sold,income is overstated by the amount of amortization that would have been recorded had the servicing rights been recognized (servicing rights are amortized over the estimated life of the underlying mortgage).The net di erence between the overstatement and understatement in a given period is the object of interest in the income-statement-based tests,as