Fair Value Accounting for Financial Instruments: Some Implications for Bank Regulation
by Wayne R
Landsman
Monetary and Economic Department
August 2006
JEL Classification Numbers: E58, G15, M41
Keywords: Fair values, financial instruments, information
asymmetry
BIS Working Papers are written by
members of the Monetary and Economic Department of the Bank for International
Settlements, and from time to time by other economists, and are published by
the Bank. The views expressed in them are those of their authors and not
necessarily the views of the BIS.
Copies of publications are available from:
Bank for International Settlements
Press & Communications
CH-4002 Basel, Switzerland
E-mail: publications@bis.org
Fax: +41 61 280 9100 and +41 61 280 8100
This publication is available on the BIS website (www.bis.org).
© Bank for International Settlements 2006.
All rights reserved. Limited extracts may be reproduced or translated provided
the source is stated.
ISSN 1020-0959 (print)
ISSN 1682-7678 (online)
Foreword
On 11-12 November 2005, the BIS held a Workshop on
“Accounting, risk management and prudential regulation”, which brought together
a multi-disciplinary group of around 35 external participants including senior
accounting practitioners, standard setters, finance academics, supervisors and
central bank officials. The workshop programme is attached. This paper was
presented at the workshop. The views expressed are those of the author(s) and
not those of the BIS.
Abstract
I identify issues that bank regulators need to
consider if fair value accounting is used for determining bank regulatory
capital and when making regulatory decisions. In financial reporting, US and
international accounting standard setters have issued several disclosure and
measurement and recognition standards for financial instruments and all
indications are that both standard setters will mandate recognition of all
financial instruments at fair value. To help identify important issues for bank
regulators, I briefly review capital market studies that examine the usefulness
of fair value accounting to investors, and discuss marking-tomarket
implementation issues of determining financial instruments’ fair values. In
doing so, I identify several key issues. First, regulators need to consider how
to let managers reveal private information in their fair value estimates while
minimising strategic manipulation of model inputs to manage income and
regulatory capital. Second, regulators need to consider how best to minimise
measurement error in fair values to maximise their usefulness to investors and
creditors when making investment decisions, and to ensure bank managers have
incentives to select investments that maximise economic efficiency of the
banking system. Third, cross-country institutional differences are likely to
play an important role in determining the effectiveness of using mark-to-market
accounting for financial reporting and bank regulation.
Workshop on “Accounting, risk management and
prudential regulation”
Introduction
Accounting standards setters in many jurisdictions
around the world, including the United States, the United Kingdom, Australia,
and the European Union, have issued standards requiring recognition of balance
sheet amounts at fair value, and changes in their fair values in income. For
example, in the United States, the Financial Accounting Standards Board
requires recognition of some investment securities and derivatives at fair
value. In addition, as their accounting rules have evolved, many other balance
sheet amounts have been made subject to partial application of fair value rules
that depend on various ad hoc circumstances, including impairment (eg goodwill
and loans) and whether a derivative is used to hedge changes in fair value (eg
inventories, loans, and fixed lease payments). The Financial Accounting
Standards Board and the International Accounting Standards Board (hereafter
FASB and IASB) are jointly working on projects examining the feasibility of
mandating recognition of essentially all financial assets and liabilities at
fair value in the financial statements.
In the United States, fair value recognition of
financial assets and liabilities appears to enjoy the support the Securities
and Exchange Commission (hereafter SEC). In a recent report prepared for a
Congressional committee (SEC, 2005), the Office of the Chief Accountant of the
SEC states two primary benefits of requiring fair value accounting for
financial instruments. First, it would mitigate the use of accounting-motivated
transaction structures designed to exploit opportunities for earnings
management created by the current “mixedattribute” – part historical cost, part
fair values – accounting model. For example, it would eliminate the incentive
to use asset securitization as a means to recognise gains on sale of
receivables or loans. Second, fair value accounting for all financial
instruments would reduce the complexity of financial reporting arising from the
mixed attributed model. For example, with all financial instruments measured at
fair value, the hedge accounting model employed by the FASB’s derivatives
standard would all but be eliminated, making it unnecessary for investors to
study the choices made by management to determine what basis of accounting is
used for particular instruments, as well as the need for management to keep
extensive records of hedging relationships.
But, as noted in the SEC report, there are costs as
well associated with the application of fair value accounting. One key issue is
whether fair values of financial statement items can be measured reliably,
especially for those financial instruments for which active markets do not
readily exist (eg specialised receivables or privately placed loans). Both the
FASB and IASB state in their Concepts statements that they consider the
cost/benefit trade-off between relevance and reliability when assessing how
best to measure specific accounting amounts, and whether measurement is
sufficiently reliable for financial statement recognition. A cost to investors
of fair value measurement is that some or even many recognised financial
instruments might not be measured with sufficient precision to help them assess
adequately the firm’s financial position and earnings potential. This
reliability cost is compounded by the problem that in the absence of active
markets for a particular financial instrument, management must estimate its
fair value, which can be subject to discretion or manipulation.
Assessing the costs and benefits of fair value
accounting for financial reporting to investors and other financial statement
users in particular reporting regimes is difficult. Assessing the costs and
benefits of bank regulators mandating fair value accounting for financial
institutions for the purpose of assessing a bank’s regulatory capital is
perhaps even more challenging. The purpose of this paper is to provide some
preliminary views on the issues bank regulators face when assessing the costs
and benefits of using fair value for determining regulatory capital and making
other regulatory decisions. To this end, I begin by reviewing extant capital
market studies that examine the usefulness of fair value accounting to
investors. I then discuss implementation issues of determining financial
instruments’ fair values. In doing so, I again look to evidence from the
academic literature. Finally, I discuss marking-to-market implementation issues
that are of particular relevance to bank regulators as they consider the
effects of fair value measurement on bank earnings and capital, and the
attendant effects on real managerial decisions.[3]
Background of fair value accounting in standard setting
Definition of fair value
The FASB defines “fair value” as “the price at which
an asset or liability could be exchanged in a current transaction between
knowledgeable, unrelated willing parties” (FASB, 2004a).[4] As the FASB notes, “the objective of a fair
value measurement is to estimate an exchange price for the asset or liability
being measured in the absence of an actual transaction for that asset or
liability.” Implicit in this objective is the notion that fair value is well
defined so that an asset or liability’s exchange price fully captures its
value. That is, the price at which an asset can be exchanged between two
entities does not depend on the entities engaged in the exchange and this price
also equals the value-in-use to any entity. For example, the value of a swap
derivative to a bank equals the price at which it can purchase or sell that
derivative, and the swap's value does not depend on the existing assets and
liabilities on the bank’s balance sheet. For such a bank, Barth and Landsman
(1995) notes that this is a strong assumption to make particularly if many of
its assets and liabilities cannot readily be traded. I will return to the
implications of this problem when discussing implementation of
marking-to-market issues below.
Applications to standard setting
In the US, the FASB has issued several standards that
mandate disclosure or recognition of accounting amounts using fair values.
Among the most significant in terms of relevance to financial institutions are
those standards that explicitly relate to financial instruments. Two important
disclosure standards are Statement of Financial Accounting Standards (SFAS) no
107, Disclosures about fair value of
financial instruments (FASB, 1991) and SFAS no 119, Disclosure about derivative financial instruments and fair value of
financial instruments (FASB, 1994). SFAS no 107 requires disclosure of fair
estimates of all recognised assets and liabilities, and as such, was the first
standard that provided investors with estimates of the primary balance sheet
accounts of banks, including securities, loans, deposits, and longterm debt. In
addition, it was the first standard to provide a definition of fair value reflecting
the FASB’s objective of obtaining quoted market prices wherever possible. SFAS
no 119 requires disclosure of fair value estimates of derivative financial
instruments, including futures, forward, swap, and option contracts. It also
requires disclosure of estimates of holding gains and losses for instruments
that are held for trading purposes.
Among the most significant fair value recognition
standards the FASB has issued are SFAS no 115, Accounting for certain investments in debt and equity securities
(FASB, 1993), SFAS no 123 (revised), Share-based
payments (FASB, 2004), and SFAS no 133, Accounting
for derivative instruments and
hedging activities (FASB, 1998). SFAS no 115 requires recognition at fair
value investments in equity and debt securities classified as held for trading
or available-for-sale. Fair value changes for the former appear in income, and
fair value changes for the latter are included as a component of accumulated
other comprehensive income, ie, are excluded from income. Those debt securities
classified as held to maturity continue to be recognised at amortised cost.
SFAS no 123 (revised) requires the cost of employee stock options grants be
recognised in income using grant date fair value by amortising the cost during
the employee vesting or service period.[5] This requirement removed election of fair
value or intrinsic value cost measurement permitted under the original
recognition standard, SFAS no 123, Accounting
for Stock-based Compensation (FASB, 1995). Until recently, most firms
elected to measure the cost of employee stock options using intrinsic value.
However, for such firms, SFAS no 123 requires they disclose a pro forma income number computed using a
fair value cost for employee stock option grants, as well as key model inputs
they use to estimate fair values.
SFAS no 133 requires all freestanding derivatives be
recognised at fair value. However, SFAS no 133 retains elements of the existing
hedge accounting model. In particular, fair value changes in those derivatives
employed for purposes of hedging fair value risks (eg interest rate risk and
commodity price risk) are shown as a component of income, as are the changes in
fair value of the hedged balance sheet item (eg fixed rate loans and
inventories) or firm-commitments (ie forward contracts). If the so-called fair
value hedge is perfect, the effect on income of the hedging relationship is
zero. In contrast, fair value changes in those derivatives employed for
purposes of hedging cash flow risks (eg cash flows volatility resulting from
interest rate risk and commodity price risk) are shown as a component of
accumulated other comprehensive income because there is no recognised
off-setting change in fair value of an implicitly hedged balance sheet item or
anticipated transaction.[6]
Outside of the US, standards issued by the IASB are
often accepted or required as generally accepted accounting principles (GAAP)
in many countries. For example, the European Union generally requires member
country firms to issue financial statements prepared in accordance with IASB
GAAP beginning in 2005. IASB GAAP comprises standards issued by its predecessor
body, the International Accounting Standards Committee (IASC), as well as those
it has issued since its inception in 2001. The IASC issued two key fair value
standards, both of which have been adopted by the IASB, IAS 32: Financial Instruments: Disclosure and
Presentation (IASB, 2003a), IAS 39, Financial
Instruments: Recognition and Measurement (IASB, 2003b). The former standard
is primarily a disclosure standard, and is similar to its US GAAP counterparts,
SFAS no's 107 and 119. IAS 39 describes how particular financial assets and
liabilities are measured (ie amortised cost or fair value), and how changes in
their values are recognised in the financial statements. The scope of IAS 39
roughly encompasses accounting for investment securities and derivatives, which
are covered under SFAS no's 115 and 133, although there are some minor
differences between IAS and US GAAP.
The IASB has also issued a key fair value standard, International Financial Reporting Standard
2, Accounting for Share-based Payment (IASB, 2004). IFRS 2 is very similar to SFAS no 123 (revised) (FASB, 2004) in
requiring firms to recognise the cost of employee stock option grants using
grant date fair value.[7]
As part of their efforts to harmonise US and
international accounting standards, the IASB and FASB recently issued related
proposed or finished standards pertaining to disclosure of financial
instruments fair values, Exposure Draft:
Fair Value Measurements (FASB, 2004a) and International Financial Reporting Standard 7, Financial Instruments:
Disclosures (IASB, 2005). The US Exposure
Draft describes a hierarchy of preferred approaches to fair value
measurement for all assets and liabilities measured at fair value under other
FASB pronouncements, ranging from quoted market prices for the specific asset
or liability to use of models to estimate fair values.[8] Both the Exposure
Draft and IFRS 7 require
disclosure of fair value amounts at the end of each accounting period (year,
quarter), how the fair values are determined, and the effect on income arising
from each particular class of assets or liabilities (ie separate disclosure of
recognised and unrecognised gains and losses). IFRS 7 is more comprehensive than the Exposure Draft in that it requires disclosure of detailed
information for recognised financial instruments, both those measured at fair
value and those that are not, as well as qualitative information relating to
financial instruments’ liquidity, credit, and market risks.
Valuation techniques
As noted above, in its Exposure Draft: Fair value measurements, the FASB describes a
hierarchy of preferences for measurement of fair value. The preferred level 1 fair value estimates are those
based on quoted prices for identical assets and liabilities, and are most
applicable to those assets or liabilities that are actively traded (eg trading
investment securities). Level 2
estimates are those based on quoted market prices of similar or related assets
and liabilities. Level 3 estimates,
the least preferred, are those based on company estimates, and should only be
used if level 1 or 2 estimates are not available. With the emphasis on market
prices, the FASB emphasises that firms should base their estimates on market
prices as model inputs wherever possible (eg use of equity market volatility
estimates when employing the Black-Scholes valuation model to estimate the fair
value of employee stock options). Fair value estimates can be constructed using
entity-supplied inputs (eg discounted cash flow estimates) if other models
employing market inputs are not available.
Are fair values useful to investors? Evidence from research
US-based research
A natural question to ask is whether bank fair value
information is useful to investors. For example, when it was deliberating SFAS
no 107, the FASB was concerned with policy questions relating to the relevance
and reliability of disclosed amounts. Regarding relevance, the FASB was
interested in whether SFAS no 107 disclosures would be incrementally useful to
financial statement users relative to items already in financial statements,
including recognised book values and disclosed amounts. Regarding reliability,
the FASB was concerned with whether fair values estimates, especially those
relating to loans, would be too noisy to disclose.[9]
As Barth, Beaver, and Landsman (2001) note,
policy-based accounting research cannot directly address these questions, but
can provide evidence that helps standard setters assess relevance and
reliability questions. A common way to assess the so-called value relevance of
a recognised or disclosed accounting amount is to assess its incremental
association with share prices or share returns after controlling for other
accounting or market information. Several studies address the value relevance
of banks’ disclosed investment securities fair values before issuance of SFAS
no 115, mandating recognition of investment securities’ fair values and effects
of their changes on the balance sheet and the income statement. For a sample of
US banks with data from 1971–90, Barth (1994) finds that investment securities’
fair values are incrementally associated with bank share prices after
controlling for investment securities’ book values. When examined in an annual
returns context, the study finds mixed results for whether unrecognised
securities’ gains and losses provide incremental explanatory power relative to
other components of income. One leading candidate for the ambiguous finding is
that securities’ gains and losses estimates contain too much measurement error
relative to the true underlying changes in their market values.[10] Using essentially the same data base, Barth,
Landsman, and Wahlen (1995) confirm the Barth’s (1994) findings and lend
support to the measurement error explanation by showing that fair value-based
measures of net income are more volatile than historical cost-based measures,
but that the incremental volatility is not reflected in bank share prices. Of
particular interest to bank regulators, Barth, Landsman, and Wahlen (1995) also
find that banks violate regulatory capital requirements more frequently under
fair value than historical cost accounting, and that fair value regulatory
capital violations help predict future historical cost regulatory capital
violations, but share prices fail to reflect this increased regulatory risk.
Barth, Beaver, and Landsman (1996), Eccher, Ramesh
and Thiagarajan (1996) and Nelson (1996) use similar approaches to assess the
incremental value relevance of fair values of principal categories of banks
assets and liabilities disclosed under SFAS no 107 in 1992 and 1993, ie,
investment securities, loans, deposits, and long-term debt. Supporting the
findings of Barth (1994) using pre-SFAS no 107 data, all three studies find
investment securities fair values are incrementally informative relative to
their book values in explaining bank share prices. However, using a more
powerful research design that controls for the effects of potential omitted
variables, Barth, Beaver and Landsman (1996) also finds evidence that loans’
fair values are also incrementally informative relative to their book values in
explaining bank share prices. Barth, Beaver and Landsman (1996) also provide
additional evidence that loans’ fair values reflect information regarding
loans’ default and interest rate risk. Moreover, the study’s findings suggest
that investors appear to discount loans’ fair value estimates made by less
financially healthy banks (ie those banks with below sample median regulatory
capital), which is consistent with investors being able to see through attempts
by managers of less healthy banks to make their banks appear more healthy by
exercising discretion when estimating loans fair values.
Finally, Venkatachalam (1996) examines the value
relevance of banks’ derivatives disclosures provided under SFAS no 119 for a
sample of banks in 1993 and 1994. Findings from the study suggest that
derivatives’ fair value estimates explain cross-sectional variation in bank
share prices incremental to fair values of the primary on-balance accounts (ie
cash, investments, loans, deposits, and debt).
International research
Because Australian and UK GAAP permit upward asset
revaluations but, as with US GAAP, require downward revaluations in the case of
asset impairments, several studies examine the dimensions of value relevance of
revaluations in these countries. Most studies, including Easton, Eddey, and
Harris (1993), Barth and Clinch (1996), Barth and Clinch (1998), and Peasnell
and Lin (2000), focus on tangible fixed asset revaluations. However, Aboody,
Barth and Kasznik (1999) examine the association between asset revaluations for
financial, tangible, and intangible assets for a sample of Australian firms in
1991–95. Focusing on the financial assets, Aboody, Barth and Kasznik (1999)
find that revalued investments for financial firms as well as non-financial
firms are consistently significantly associated with share prices.
One interesting study of Danish banks, Bernard,
Merton and Palepu (1995), focuses on the impact of mark-to-market accounting on
regulatory capital as opposed to the value relevance of fair values for
investors. Denmark is an interesting research setting because Danish bank
regulators have used mark-to-market accounting to measure regulatory capital
for a long period of time. Bernard, Merton and Palepu (1995) find that although
there is evidence of earnings management, there is no reliable evidence that
mark-to-market numbers are managed to avoid regulatory capital constraints.
Moreover, Danish banks’ mark-to-market net equity book values are more reliable
estimates of their equity market values when compared to those of US banks,
thereby providing indirect evidence that fair value accounting could be
beneficial to US investors and depositors.[11]
US-based stock option research
As noted above, estimates of employee stock options
fair values have been required to be disclosed for several years under SFAS no
123. Several studies examine the value relevance of such disclosures, including
Bell, Landsman, Miller, and Yeh (2002), Aboody, Barth and Kasznik (2004), and
Landsman, Peasnell, Pope and Yeh (2005). Findings in Bell, Landsman, Miller and
Yeh (2002) differ somewhat from those in Aboody, Barth and Kasznik (2004),
although both studies provide evidence that employee option expense is value
relevant to investors. Landsman, Peasnell, Pope and Yeh (2005) provide
theoretical and empirical support for measuring the fair value of employee
stock option grants beyond grant date, with changes in fair value recognised in
income along with amortisation of grant date fair value.
Because quoted prices for employee stock options
typically are not available because of non-tradability provisions, the fair
value estimates are based on models that rely on inputs selected by reporting
firms. Aboody, Barth and Kasznik (2005) find evidence that firms select model
inputs so as to manage the pro forma income number disclosed in the employee
stock option footnote. This finding is potentially relevant to accounting
standard setters as well as bank regulators in that it is additional evidence
that managers facing incentives to manage earnings are likely to do so when
fair values must be estimated using entity-supplied estimates of values or
model inputs if quoted prices for assets or liabilities are not readily
available.[12] If managers have the incentive to use
discretion when estimating fair values of on and off-balance sheet asset and
liability amounts when such values are not recognised in the financial
statements, it is reasonable to assume the incentive will only increase if fair
value accounting is used for recognition of amounts on the balance sheet and in
the income statement.
Marking-to-market implementation issues
Marking-to-market financial instruments are
relatively easy if they are actively traded in liquid markets. The problem
becomes more complicated if active markets do not exist, particularly if the
financial instrument is a compound instrument comprising several embedded
optionlike features, values for which depend on inter-related default and price
risk characteristics. Moreover, Barth and Landsman (1995) makes the observation
that in the absence of active, liquid markets, fair value is not well defined
in the sense that an instrument’s acquisition price, selling price, and
value-in-use to the entity can differ from each other.[13] Stated another way, even if an instrument’s
acquisition or selling prices are observable, these prices can only, at best,
provide upper or lower bounds on its “fair value”. The FASB’s stated preference
for using an instrument’s selling price as its measure of fair value is
appropriate when fair value is well defined, but is somewhat arbitrary when it
is not.
In this section, I discuss issues relating to
implementation of fair value estimates when market prices for particular
financial instruments are not readily available by focusing on findings from
two related studies by Barth, Landsman and Rendleman (1998, 2000) on the use of
binomial option pricing models to estimate fair values for corporate debt and
its components.
Binomial option pricing of corporate debt
Barth, Landsman and Rendleman (1998) use a binomial
option pricing model to estimate the fair values of corporate debt and its
components, ie, conversion, call, put, and sinking fund features, to provide
evidence on the relevance and reliability of estimated fair values. A companion
study, Barth, Landsman and Rendleman (2000), describes details of how the
binomial model is implemented. The 1998 empirical study is based on data from
1990 for a sample of 120 publicly traded US firms that have corporate debt with
multiple embedded option features. The binomial model the study implements is
based on the models of Cox, Ross and Rubinstein (1979) and Rendleman and
Bartter (1979), and considers directly only default risk, but includes
information in the interest rate yield curve.
Findings from Barth, Landsman and Rendleman (1998)
reveal component value estimates are relevant in that they represent large
fractions of estimated total bond fair value. In addition, implementing a
fundamental components approach in which call options are classified as assets,
conversion options as equity, and put options as debt, indicates there are
material changes to recognised balance sheet accounts and debt-to-equity ratios
for sample firms.[14] The study also finds that estimates of
component fair values depend on whether a bond has multiple features. For
example, the value of conversion feature for a convertible, callable bond
depends on the value of the call feature and vice versa. In addition, because
components’ values are interdependent, the order in which components are
considered when estimating each bond’s total fair value can materially affect
each component’s estimated fair value. This issue is particularly important if
a fundamental components approach is used for separate recognition of bond
components as assets, liabilities, and equity.
However, additional evidence in Barth, Landsman and
Rendleman (1998) suggests model estimates of total bond value may lack
reliability. In particular, when the authors re-estimate bond fair values
excluding from the sample those bonds with available market prices (such bonds
comprise approximately half of sample bonds), estimated bond values for those
bonds that are not publicly traded differ significantly from value estimates
when all bonds are included in the estimation procedure. This finding suggests
that financial instruments’ fair value estimates are sensitive to whether
actual market price information from other instruments an entity has on its
balance sheet is available to be used as model inputs.
Barth, Landsman and Rendleman (1998) reaches
several conclusions regarding limitations to implementation of binomial option
pricing models for estimating bond fair values that generalise to all financial
instruments issued or held by an entity. First, the authors had to make several
educated guesses for values of model inputs (eg, conversion schedules and
equity volatility). In principle, managers of the reporting entities likely
have access to better information than financial statement users (including
academic researchers), and the authors suggest that fair value estimates could
improve if firms were required to disclose them. Second, models quickly become
too complex and difficult to implement if they are to consider all of the
dimensions of risk and value that can affect an instrument’s fair value. For
example, presently, few models consider both interest rate and default risk. In
addition, financial instruments’ fair values are interdependent. For example,
the fair value of one debt instrument issued by an entity is dependent upon
actions that holders of another debt instrument issued by that entity can take.
The model implemented by Barth, Landsman and Rendleman (1998) considers some
sources of bond value interdependence (eg debt priority) but basically ignores
the issue because of its complexity. The issue of financial instruments’ value
interdependence is another illustration of the issue raised by Barth and
Landsman (1995) that a financial instrument’s fair value may not be well
defined (eg its selling price may not equal its value-in-use to the entity).
Manipulation of model inputs
Having to rely on managers’ model estimates of
financial instruments’ fair values introduces the general problem of
informational asymmetry – ie, managers have private information regarding
appropriate values to select for model inputs as well the true underlying
economic value of a financial instrument to the firm. Informational asymmetry
creates two somewhat different problems, adverse selection and moral hazard. An
important implication of adverse selection is that the market will tend to
value apparently similar financial instruments held by two different firms
similarly when assessing their fair values and the values of the firms’
equities. Thus, for example, in the absence of credible and verifiable
information, two banks that are otherwise equivalent except one has a higher
quality loan portfolio than the other will have their stocks valued similarly
by the securities market. A solution to the adverse selection problem is to
permit managers of the bank with a higher quality loan portfolio to signal
their loans are of higher quality. For the signal to be credible, it must be
costly, but less costly for the bank with higher quality loans. This can be
achieved, for example, by permitting bank managers to disclose selectively
attributes about the loans’ fair values that would be too costly for bank
managers with low quality loans to disclose.
The problem of moral hazard is that managers will
tend to use their private information to their advantage by manipulating the
information that they disclose to the securities markets and regulators. In the
case of banks, this can lead to mispricing of their stocks and an inaccurate
portrayal of their capital ratios and their financial health to bank
regulators. As noted above, the findings in Aboody, Barth and Kasznik (2005),
which indicates that managers select model parameters to manage estimates of
disclosed employee stock option fair values, raise the broader question of
whether managers will behave similarly when selecting model parameters for fair
value estimates of other financial instruments, including those whose values
are recognised in the body of the financial statements. The Barth, Landsman and
Rendleman (1998) conclusion that managers can provide better estimates of bond
fair values because they have access to private information presumes implicitly
that managers apply their private information in a neutral fashion, ie, they do
not succumb to the temptation to manipulate bond fair value estimates for
private gain.
If fair value accounting for financial instruments
is generally applied for financial statement recognition and regulatory capital
determination, accounting standard setters as well as securities and bank
regulators face the challenge of determining how much latitude to give managers
when they estimate fair values, balancing the benefit of permitting managers to
reveal private information, thereby mitigating the adverse selection problem,
and the moral hazard cost of their exercising discretion to manipulate earnings
or capital ratios when selecting model parameters.
Marking-to-market: additional issues for bank regulators
I now turn to discussing additional issues that bank
regulators in particular need to consider if they are to require banks to
mark-to-market financial instruments when determining regulatory capital and
when assessing other dimensions of bank performance.
Fair values measurement error
The first obvious issue bank regulators face is that
fair value estimates of bank assets and liabilities (which are principally
financial instruments) are likely to contain measurement error. If the findings
in Barth, Landsman and Wahlen (1995) relating to investment securities are
generalised to other bank assets and liabilities, implementation of a full fair
value model for recognition of financial instruments at fair value could yield
unrecognised gains/losses that could cause earnings and regulatory capital to
be more volatile than earnings and regulatory capital based on the current
historical cost model. This would be expected to occur particularly if
measurement error in bank assets’ fair values – which is likely to be
positively correlated across assets – is not fully offset by measurement error
in bank liabilities’ fair values.
Of course not all earnings or regulatory capital
volatility arising from the application of fair value accounting is the result
of measurement error. Barth (2004) makes the observation that there are three
primary sources of “extra” volatility associated with fair value-based
accounting amounts relative to those determined under historical cost. The
first is true underlying economic volatility that is reflected by changes in
bank assets’ and liabilities’ fair value. The second is volatility induced by
measurement error in estimates of those fair value changes. The third, induced
volatility arising from using a mixed-attribute model would be less of a
concern if all instruments are recognised at fair value. The
relevance/reliability trade-off that accounting standard setters consider is
certainly applicable to bank regulators. A primary goal of regulators would
appear to develop a framework for measuring financial instruments’ fair values
– and changes in value – so as to maximise the ratio of (a) additional economic
volatility in bank earnings (or capital ratios) arising from using fair value
accounting instead of historical cost to (b) additional volatility arising from
measurement error in fair value estimates. As noted above, a significant
dimension to this problem is determining how much discretion to give bank
managers when they estimate fair values of their assets and liabilities.
Before leaving the discussion of measurement error,
it is important to note that although fair value estimates of bank assets and
liabilities likely contain measurement error relative to true economic values,
so do book value estimates. Casting the debate in terms of whether fair values
are “good” or “bad” is inappropriate. The more appropriate question to ask is
whether fair value-based financial statements improve information investors
receive relative to information provided by historical cost-based financial
statements, and whether regulation of bank capital will be more efficient under
one accounting system or the other.
Economic considerations
A natural question to ask is what the real economic
consequences will be of accounting standard setters and financial reporting and
bank regulators requiring mark-to-market accounting to measure bank performance
and financial condition. The desired outcomes are, of course, greater economic
and informational efficiency. However, as noted above, the extent to which
these goals are met depends on a variety of factors relating to how the model
is implemented (eg the amount of discretion managers are granted when selecting
fair value model inputs).
One notable implementation issue is whether real
economic decisions made by bank managers would improve. On the one hand,
managers would have less incentive to use accounting-motivated transaction
structures designed to exploit opportunities for income management arising from
the current mixed attribute accounting model. On the other hand, extra
volatility of fair value income and regulatory capital could cause bank
managers to apply a sub-optimal decision rule by selecting investments of lower
risk than would be the case if investment decisions were based solely on
economic considerations.
The effects on economic and informational efficiency
of requiring fair value accounting to measure bank performance and financial
condition are likely to vary considerably across countries, reflecting
differences in richness of securities markets, legal systems, bank and
securities markets regulatory enforcement, and a host of other institutional
features. The burgeoning “law and finance” literature (La Porta,
Lopez-de-Silanes, Shleifer and Vishny 1998) suggests that these differences are
likely to play an important role in determining the effectiveness of using fair
value accounting for financial reporting and bank regulation.
Concluding remarks
In this paper I identify issues that bank regulators
need to consider if they are to use fair value accounting for determining bank
regulatory capital and when making regulatory decisions. In the financial
reporting arena, the FASB and IASB have issued several disclosure and
measurement and recognition standards for financial instruments, and all
indications are that it’s only a matter of time before both standard setters
will mandate recognition of all financial instruments at fair value. To help
identify important issues for bank regulators, I briefly review capital market
studies that examine the usefulness of fair value accounting to investors, and
discuss marking-to-market implementation issues of determining financial
instruments’ fair values. In doing so, I identify several key issues. First,
regulators need to consider how to let managers reveal private information in
their fair value estimates while minimising strategic manipulation of model
inputs to manage income and regulatory capital. Second, they need to consider
more broadly how best to minimise measurement error in fair values so as to
maximise their usefulness to investors and creditors as they make their
investment decisions, and how best to ensure bank managers have incentives to
select those investments that maximise economic efficiency of the banking
system. Cross-country institutional differences are likely to play an important
role in determining the effectiveness of using mark-to-market accounting for
financial reporting and bank regulation.
References
Aboody, D, M E Barth and R Kasznik (1999):
“Revaluations of fixed assets and future firm performance”, Journal of Accounting and Economics 26,
pp 149–78.
——— (2004): “SFAS no 123 Stock-based employee
compensation and equity market values”, The
Accounting Review 79, pp 251–75.
——— (2005): “Do firms manage stock-based compensation
expense disclosed under SFAS 123?”, Working paper, Stanford University, Graduate
School of Business.
Ahmed, A S and C Takeda (1995): “Stock market
valuation of gains and losses on commercial banks’ investment securities: an
empirical analysis”, Journal of
Accounting and Economics 20, pp 207–25.
Barth, M E (1991): “Relative Measurement Errors Among
Alternative Pension Asset and Liability Measures”, The Accounting Review 66, pp 433–63.
——— (1994): “Fair value Accounting: Evidence from Investment Securities and the
Market Valuation of Banks”, The Accounting Review 69, pp 1–25.
——— (2004): “Fair values and financial statement
volatility”, in The Market Discipline
Across Countries and Industries, edited by Claudio Borio, William Curt
Hunter, George G Kaufman, and Kostas Tsatsaronis. Cambridge, MA: MIT Press.
Barth, M E, W H Beaver and W R Landsman (1996):
“Value-relevance of banks’ fair value disclosures under SFAS 107”, The Accounting Review 71, pp 513–37.
——— (2001): “The relevance of the value relevance
literature for accounting standard setting: another view”, Journal of Accounting and Economics 31, pp 77–104.
Barth, M E, and G Clinch (1996): “International
differences in accounting standards: evidence from UK, Australian, and Canadian
firms”, Contemporary Accounting Research,
Spring, pp 135–70.
——— (1998): “Revalued financial, tangible, and
intangible assets: associations with share prices and non market-based value
estimates”, Journal of Accounting
Research 36, pp 199– 233.
Barth, M E and W R Landsman (1995): “Fundamental
issues related to using fair value accounting for financial reporting”, Accounting Horizons 9, pp 97–107.
Barth, M E, W R Landsman and R J Rendleman, Jr
(1998): “Option pricing-based bond value estimates and a fundamental components
approach to account for corporate debt”, The
Accounting Review 73, pp 73–102.
——— (2000): “Implementation of an option
pricing-based bond valuation model for corporate debt and its components”, Accounting Horizons 14, pp 455–79.
Barth, M E, W R Landsman and J Wahlen (1995): “Fair
value accounting: effects on banks’ earnings volatility, regulatory capital,
and value of contractual cash flows”, Journal
of Banking and Finance, pp 577–605.
Bell, T B, W R Landsman, B L Miller and S Yeh (2002):
“The valuation implications of employee stock option accounting for profitable
computer software firms”, The Accounting
Review 77, pp 971–96.
Bernard, V L, R C Merton and K G Palepu (1995):
“Mark-to-market accounting for banks and thrifts: lessons from the Danish
experience”, Journal of Accounting
Research 33, pp 1–32.
Cox, J S, A Ross and M Rubinstein (1979): “Option
pricing: a simplified approach”, Journal
of Financial Economics 7, pp 51–84.
Eccher, A, K Ramesh and S R Thiagarajan (1996): “Fair
value disclosures by bank holding companies”, Journal of Accounting and Economics 22, pp 79–117.
Financial Accounting Standards
Board (1985): Statement of financial
accounting standards no 87, Employers’ accounting for pensions. FASB:
Norwalk, Connecticut.
——— (1990): Discussion memorandum, Distinguishing
between liability and equity instruments and accounting for instruments with
characteristics of both. FASB: Norwalk, Connecticut.
——— (1991): Statement of financial accounting standards no 107, Disclosures about fair value of financial
instruments. FASB: Norwalk, Connecticut.
——— (1993): Statement of financial accounting standards no 115, Accounting for certain investments in debt
and equity securities. FASB: Norwalk, Connecticut.
——— (1994): Statement of financial accounting standards no 119, Disclosure about
derivative financial instruments and fair value of financial instruments.
FASB: Norwalk, Connecticut.
——— (1995): Statement of financial accounting standards no 123, Accounting for
stockbased compensation. Norwalk, Connecticut.
——— (1998): Statement of financial accounting standards no 133, Accounting for derivative instruments and
hedging activities. FASB: Norwalk, Connecticut.
——— (2000): Proposed statement of financial accounting standards, Accounting for
financial instruments with characteristics of liabilities, equity, or both.
Norwalk, CT: FASB.
——— (2004a): Proposed Statement of Financial Accounting Standards, fair value
measurements. Norwalk, CT: FASB.
——— (2004b): Statement of financial accounting standards no 123 (revised), Share-based
payment. Norwalk, CT: FASB.
International Accounting Standards Board (2003a): International Accounting Standard 32:
Financial instruments: Disclosure and presentation, London, UK.
——— (2003b). International
Accounting Standard 39: Financial instruments: Recognition and measurement,
London, UK.
——— (2004): International Financial Reporting Standard 2, Accounting for
share-based payment, London, UK.
——— (2005). International Financial Reporting Standard 7 Financial Instruments: Disclosures, London, UK.
Landsman, W (1986): “An empirical investigation of
pension fund property rights”, The
Accounting Review 61, pp 662–691.
Landsman, W R, K Peasnell, P Pope and S Yeh (2005):
“The value relevance of alternative methods of accounting for employee stock
options”, Working paper, University of North Carolina.
La Porta R, F Lopez-de-Silanes, A Shleifer and A
Vishny (1998): “Law and finance”, Journal
of Political Economy, 106, pp 1113–55.
Nelson, K (1996): “Fair value accounting for
commercial banks: an empirical analysis of SFAS no 107”, The Accounting Review 71, pp 161–82.
Peasnell, K V and Y N Lin (2000): “Fixed asset
revaluation and equity depletion in the UK”, Journal of Business Finance and Accounting 27, pp 359–94.
Rendleman, R J and B J Bartter (1979): “Two-state
option pricing”, Journal of Finance
34, pp 1093–110.
United States Securities and Exchange Commission
(2005): Report and recommendations pursuant to section 401(c) of the
Sarbanes-Oxley act of 2002 on arrangements with offbalance sheet implications,
special purpose entities and transparency of filings by issuers.
Venkatachalam, M (1996): “Value-relevance of banks’
derivatives disclosures”, Journal of
Accounting and Economics 22, pp 327–55.
Relevance and reliability of fair values: discussion
of issues raised in “Fair value
accounting for financial instruments: some implications for bank regulation”
James O’Brien[15]
Introduction
In his paper, Professor Landsman reviews research on
both the relevance and reliability of reporting fair values for loans and other
financial instruments (Landsman (2005)). Accounting standard setters define
fair value as the amount that would be paid or received for the item being
valued in an arm’s length transaction between knowledgeable parties. This is a
market value definition and the standard setters have indicated that, if
available, a current market price for the item is said to be the best estimate
of its fair value. Relevance means that the fair value is capable of making a
difference to financial statement users’ decisions. Reliability means that the
reported fair value represents what it is purported to represent (Barth et al
(2001), p 80).
Professor Landsman concludes that the evidence on
fair value reporting supports its relevance. On reliability, he suggests there
is some uncertainty, using evidence from Barth, Landsman and Rendleman (1998)
based on testing a pricing model for corporate bonds. He further discusses
banks’ use of their private information in determining loan fair values and
consequences of model valuation errors on earnings volatility.
In my discussion, I first comment on issues
concerning fair value relevance tests and the standard setters’ relevance
criterion. I then consider the potential importance and reliability of models
for loan fair values. Here my comments expand on Professor Landsman’s
discussion of model reliability.
Relevance of reported fair values
By revealed preference, the accounting standard
setters view market exchange values as providing the best combination of
relevance and reliability. My conjecture of what underlies this view is the
following: regarding reliability, market exchange values are objective measures
of value and can be obtained from markets where the instruments are being
traded. When not directly observed, the standard setters believe that exchange
values can be closely approximated by reference to market information and the
use of valuation models. Historical costs are more reliable as a cost measure
but they lack the relevance of current market exchange values to the primary
users of financial statements. These users would be investors in claims to the
firm’s earnings, eg, equity claims and debt claims. Current market exchange
values for the firm’s assets better reflect the assets’ contribution to the
current market values of the claims on the firm’s earnings. Thus, fair values
provide investors, and others with aligned interests, with more useful
information than historical costs on what is determining the value of their
investments.
This greater usefulness of market value information
is hypothetical and the accounting literature has sought to evaluate it
primarily by using regressions of firms’ equity values on reported fair values,
with controls for historical costs and other variables. Professor Landsman
concludes that the literature provides support for fair value relevance, citing
findings of statistically significant coefficients (with the appropriate signs)
between equity values and reported fair values. While I think this literature
improves our understanding of how the market may be using the fair value
information, the equity value tests are subject to significant interpretation
difficulties. It also is not clear that the tests and the standard setters’
relevance criterion effectively address arguments opposing the adoption of full
fair value accounting. I comment on both issues.
Regarding statistical difficulties in interpreting
results of the equity value regressions, two problems appear frequently in the
literature. One is that the tests cannot distinguish between relevance and
reliability. Are weak results an indication that the market does not find fair
value relevant to its decisions or are the reported values not reliable? The
other is omitted variables. Equity values will be related to all the positions
in the balance sheet and it can be difficult to account for everything.
Moreover, assets that represent the core economic value of firms may not be
balance sheet items making it difficult to control for their influence on
equity values. Omitted on- and off-balance sheet assets or liabilities that are
correlated with the reported fair values will bias the estimated relation
between market equity values and the reported fair values.
To illustrate the interpretation difficulties of the
equity-fair value regressions, Eccher et al (1996) and Nelson (1996) get mixed
results on the significance of regression coefficients for loan fair values.
This might be attributable to a lack of reliability in the reported fair values
(eg Nissim (2003)). However, Barth, Beaver and Landsman (1996) get strong results
in testing loan fair value relevance. They attribute Eccher et al and Nelson’s
weak results to insufficient control variables in their regressions, not to
unreliable reported values.[16]
For securities fair values, reliability would seem to
be less of an issue since market values are more readily available. However, in
their 1996 study, Barth et al report mostly insignificant coefficients for
security investment fair values and mostly smaller coefficients than for the
loan fair values (tables 3 and 4). Also, Barth (1994) finds mixed results when
testing the significance of banks’ securities gains and losses on bank stock
returns. Barth (1994) suggests that this may reflect reporting errors in
securities gains and losses. However, Ahmed and Takeda (1995) argue that there
is an omitted variables bias and, after accounting for this, find securities
gains and losses significantly affect bank stock returns. Carroll et al (2002)
also take issue with the reporting error explanation of Barth (1994). They find
very strong support for securities gains and losses in explaining closed-end
mutual fund stock prices, which completely dominate historical costs. For
mutual funds, explicitly accounting for all of the firms’ assets might be an
easier task.
Another statistical issue is that tests are only for
significance (and correct sign) against a null hypothesis of a zero
coefficient. There should also be tests of a null hypothesis based on the
hypothetical coefficient value when the reported fair value is reliable and the
market is properly assessing its relevance, eg a coefficient of 1.0. A
rejection of this alternative null hypothesis is important in assessing the
consistency between the reported fair values and the null hypothesis and
potentially whether the market is correctly using the reported fair values.
The equity relevance tests and the relevance
criterion adopted by the accounting standard setters do not consider whether
the reported fair values are or will be used appropriately. In failing to do
so, they do not adequately address arguments against full financial fair value
accounting for banks. These arguments are often couched in terms of excess
volatility being introduced into bank earnings that include fair value gains
and losses on loans that are held to maturity. Implicit in the arguments is
that the market or other users of reported earnings will not correctly
interpret or react to the increase in reported earnings volatility due to the
inclusion of fair value gains and losses.
Freixas and Tsomocos (2004) and Plantin, Sapra and
Shin (2004) have developed formal models where fair value gains and losses will
create an excess volatility in reported earnings. In these models, the excess
volatility arises because the economic value of the bank is more stable (and
exceeds) the market exchange value of the loans. The two papers emphasise
different, but not incompatible, economic values of the bank. Freixas and
Tsomocos emphasise intertemporal income smoothing of earnings paid to the
ultimate claimants to the bank’s earnings; Plantin, Sapra and Shin emphasise
bank investment in borrower credit information and monitoring that produces
positive net present value in bank lending that cannot be properly valued in
(arm’s length) market transactions. In both papers, fair value gains and losses
generate reported earnings volatility that exceeds the volatility of payments
that go to the holders of claims on the bank. Nonetheless, banks’ will respond
to the higher volatility in reported earnings by undertaking either or both new
dividend policies and asset management policies that will be incompatible with
maximising their economic value in terms of intertemporal income smoothing or
providing value-added in investing in credit-risky assets.
Implicit in bank management responses to the
reporting of fair value gains and losses is that the users of earnings reports
will incorrectly interpret the increased earnings volatility as reflecting
volatility in the underlying economic value of the bank. The accounting
standard setters relevance criterion and the equity value regression tests of
relevance cannot address this misinterpretation issue or the broader issue of
relevance of the loan fair values for the economic value of the bank.
In should be noted that Freixas and Tsomocos (2004)
and Plantin, Sapra and Shin (2004) see some role for (accurate) market value
reporting by banks and hence a trade-off in the adopting a market value
accounting system. In particular, in Freixas and Tsomocos, reporting market
values is useful in identifying the current condition of the balance sheet and
can be effective in preventing moral hazard behaviour.
At the least, there seems to be agreement that fair
value reporting of bank assets that reflects the assets’ current credit
condition and market interest rates has substantial benefits in providing
objective and timely information on the bank’s financial condition. What has
not gotten much scrutiny, however, is the reliability of reported fair values
when market prices are not observed but must be estimated.
Fair value and model reliability for bank loans
In discussion papers on financial fair value,
accounting standard setters have set reliability hierarchy for different fair
value reporting methods. At the top of the hierarchy are observed market prices
of the instruments being valued. At the bottom is the use of models when market
prices are not available. The discussion papers seem to suggest that most often
market prices will be available for the exact item or a close substitute. The
modeling category is more of a residual.
The vast majority of bank loans, however, are not
traded and arm’s length market transactions prices generally will not be
available. Thus for most loans reported fair values will contain some mixture
of modeling and reliance on market prices. The amount of modeling and model
assumptions may be significant even where market prices are being used. For
illustration, consider the following hierarchy of commercial loan valuation
approaches based on three levels of market price availability:
1.
Valuation with the borrower’s debt market prices
2.
Valuation with the debt market prices of related
borrowers
3.
Valuation with models without debt market prices
In determining the fair values for loans in these
categories and the need for modeling, the bank’s full use of its information on
borrower credit worthiness is assumed. I ignore the issue of how or if the bank
might be able to actually sell loans in arms’ length transactions at their
internally calculated values (as raised in Plantin, Sapra and Shin (2004)). I
also ignore the issue of the bank’s incentives to report estimated values that
fully reflect its private information (as discussed in Professor Landsman’s
paper).
While the amount of modeling is lowest for Level 1,
it is still likely to be important in determining fair values. Modeling and
various model assumptions will be required to adjust the market prices for the
firm’s traded debt to account for differences between the traded debt and loan
contractual features. The different contractual features will produce
differences between the traded debt and loans’ periodic payments, expected
lives, and loss in the event of default. For example, large banks frequently
use credit default swaps (CDSs) on bonds issued by large corporations to hedge
or internally value loans made to the corporations. The CDSs will capture the
market’s assessment of the firms’ default likelihood but significant modeling
is required to account for other differences between the underlying bonds and
the loans. The differences will include embedded options often in loans but not
bonds, eg prepayment option, certain loan fees and periodic loan repricing
contingent on balance sheet measures of the borrowing firm’s condition.
Further, the loans are frequently part of a credit facility that includes a
line of credit. The line of credit exposes the bank to a contingent liability
whose value must be included in the valuation of the credit facility.[17]
The majority of bank loan obligors will not have
traded debt. For these obligors, a Level 2 fair value approach might be used by
making use of market prices or credit spreads of related borrowers. A likely
candidate will be generic credit spreads (or a term structure of credit
spreads) for bonds sorted by rating, industry, and possibly other criteria.
However, these generic credit spreads may be just the basic building blocks in
loan fair value model calculations. Modeling becomes more important because of
systematic differences between default probabilities embedded in the generic
bond credit spread data and the loan default probabilities, as well as the
differences between the contractual features of the bonds and bank loans.[18]
A level 3 approach will make little use of market
bond prices or credit spreads. It differs from a level 2 approach in that the
firm’s underlying default likelihood and loss in default is directly estimated,
rather than being fully or partly inferred from market credit spreads. Standard
models used for pricing corporate bonds will involve estimating the obligor
firm’s asset value and its asset volatility in determining default probability
and loss in the event of default. Other determinants will be the firm’s total
liabilities and the contractual features of the bonds (or loans) being valued.
Typically in corporate bond pricing, firm asset values and asset volatility are
estimated using the firm’s equity value and estimates of equity return
volatility. This approach is referred to as a structural model approach, while
the use of credit spreads as the basic building block is referred to as
reduced-form modeling.[19]
There is little evidence on the accuracy of loan
pricing using levels 1, 2, or 3 approaches. There is a good bit of evidence on
the accuracy of bond pricing models using a structural approach. Professor
Landsman discusses results from Barth, Landsman, and Rendleman (1998), who
developed and tested a structural bond pricing model. Here I add to Professor
Landsman’s discussion by presenting results from a recent extensive study by
Eom, Helwage and Huang (2004). They estimate and test 4 well-known bond pricing
models and some variants of the basic models (a total of 9 models are tested).
All the pricing models are structural models.
Eom et al, estimate parameters for the various
structural model using firms’ market equity values and equity return volatility
and make no use of the firm’s bond market prices other than to evaluate the
accuracy of the bond model prices. They limit their sample to bonds that should
be simplest to price: all bonds are senior and straight debt and all firms have
a simple capital structure. The bond prices also are traded quotes.
Table 1 presents some of the principal results in Eom
et al. The first three columns report statistics on pricing errors as a percent
of actual bond prices. The last two columns present percentage errors in
estimated credit spreads (the bond yield minus a comparable maturity Treasury
yield). The mean errors measure bias in the pricing models; the absolute mean
errors measure accuracy in terms of average size of (positive or negative)
errors, the standard deviations measure dispersion of the errors across the
different bonds. The last row in the Table presents the median values for the
error statistics across the 9 models. For brevity, I will only make some points
based on the results in table 1.
First, consider pricing model accuracy (col 2). The
last row indicates that for the median model, the average absolute pricing
error is a little less than 5 percent of the bond price. While not extreme,
this pricing error is still sizable given the selection of bonds that should be
easiest to value. Also important to note is that accuracy differs substantially
across models, ranging from about 3 to over 12 percent.
Second, consider bias (col 1). This is potentially
important in considering model accuracy at the portfolio level. For the median
model, the bias is fairly modest, less than 2 percent of the actual price.
Since the average absolute error is almost 5 percent, this suggests important
cancelling between positive and negative pricing errors across different bonds.
Nonetheless, for several variants of one model (CDG), the bias is over 10
percent of the actual bond prices. When Barth et al (1998) estimate their bond pricing model without using the
bond’s actual prices, the estimated prices also have a very large bias. A large
bias implies potentially large portfolio-wide errors if the model is used to
value a large part of the portfolio.
There is also another source of portfolio bias that
is not revealed by the cross-section pricing errors reported in table 1. Over
time, changes in market or economic conditions can produce correlated changes
in the valuation errors of individual bonds or loans and hence the entire
portfolio.
Third, consider credit spreads. By definition,
credit spreads are intended to reflect credit or default risk. The average
absolute credit spread errors (each expressed as a percent of the actual credit
spread) are shown in col 5. Average absolute errors are very large for all
models, with the absolute error being 125 percent for the median model. These
results indicate that structural models for credit risky debt cannot price the
credit risk, or at least cannot match the observed market spreads on credit
risky bonds. The results are consistent with earlier studies of structural
pricing models. An inability to price credit risk will assume progressively
greater importance for bonds or loans the lower the credit quality of the bonds
or loans.[20]
Barth et al (1998) found larger pricing errors (when
model prices were not used in model parameter estimation) than the median model
errors shown in Table 1. However, in contrast to the straight bonds studied by
Eom et al, the bonds studied by Barth et al include various conversion, call,
put, and sinking fund provisions. Bonds with embedded options and other
provisions may be more difficult to value. Barth et al also found that the bond
provisions account for a significant proportion of the bonds’ values,
suggesting their importance in bond pricing.21
These conclusions are limited to bonds (and to
structural models). Without formal study, it is difficult to say whether loan
valuation models will be more or less prone to error than bond valuation
models. One important feature of bank loans may make loan valuation
significantly easier. This is the much higher recovery rate on defaulted loans
than on bonds. This can significantly lower the loan’s credit risk and thus
make accurate valuation easier. However, the greater number and flexibility of
provisions in loans may make valuation more difficult and bond market prices
less applicable. There is also the important issue of the bank’s incentives in
judiciously making use of its information on borrower credit quality, which is
discussed by Professor Landsman.
Presently, all important issues on how banks will
determine loan fair values appear to be outstanding. These issues include the
extent of model use, the range of models and estimation methods that might be
employed, the likely accuracy of reported fair values, and the methods by which
reported values might be verified. Before adopting full financial fair value
reporting for banks, formal study of these issues would seem necessary.
Adopting full fair value accounting without such study risks the potential for
wide-spread abusive modeling practices or the imposition of heavy-handed rules
on how fair values are to be calculated.
References
Aguais, S, L Forest and D Rosen (2000): “Building a
credit risk valuation framework”, Algo
Research Quarterly, December, pp 21–46.
Ahmed, A S and C Takeda (1995): “Stock market
valuation of gains and losses on commercial banks’ investment securities”, Journal of Accounting and Economics,
September, pp 207–25.
Barth, M (1994): “Fair value accounting: evidence
from investment securities and the market valuation of banks”, The Accounting Review, January, pp 1–25.
Barth, M, W Beaver and W Landsman (1996):
“Value-relevance of banks’ fair value disclosures under SFAS no 107”, The Accounting Review, October, pp
513–37.
——— (2001): “The relevance of the value relevance
literature for financial accounting standard setting: another view”, Journal of Accounting and Economics, 31,
pp 77–104.
Barth, M, W Landsman and R Rendleman (1998): “Option
pricing-based bond value estimates and a fundamental components approach to
account for corporate debt”, The
Accounting Review, January, pp 73–102.
——— (2002): “Implementation of an option
pricing-based bond valuation model for corporate debt and its components, American Accounting Assoc Accounting
Horizons, December, pp 455–79.
Duffie, D and K Singleton (2003): Credit risk: pricing, measurement, and
management, Princeton, Princeton University Press.
Carroll, T, T Linsmeir and K Petroni (2002): “The
reliability of fair value VS. Historical cost information: Evidence closed-end
mutual funds”, manuscript.
Chava, S (2002): “Modeling loan commitments and
liquidity crisis: theory and estimation”, manuscript.
Eccher, E, K Ramesh and S Thiagarajan (1996): “Fair
value disclosures by bank holding companies”, Journal of Accounting and Economics, 22, pp 79–117.
Enria, A et al (2004): “Fair value accounting and
financial stability”, European Central Bank, Occasional Paper Series, no 13.
Eom, Y, J Helwege and J-Z Huang (2004): “Structural
models of corporate pricing: an emopirical analysis”, Review of Financial Studies, vol 17, no 2, pp 499–544
Freixas, X and D Tsomocos (2004): “Book vs fair value
accounting in banking and intertemporal smoothing”, manuscript, September.
Landsman, W (2005): “Fair value accounting for
financial instruments: some implications for bank regulation”, manuscript,
September.
Nelson, K (1996): “Fair value accounting for
commercial banks: an empirical analysis of SFAS 107”, The Accounting Review, 71, pp 161–82.
Nissim, D (2003): “Reliability of banks’ fair value
disclosure for loans”, Review of
Quantitative Finance and Accounting, 20, pp 355–84.
Plantin, G, H Sapra and H Shin (2004):
“Marking-to-market: panacea or pandora’s box?”, manuscript, December.
Table 1
Accuracy
of structural bond pricing models: Eom, Helwege, Huang (2004)
Pricing
model
|
|||||
1
mean
pricing
error
|
2
mean abs
pricing
error
|
3
std dev
of
pricing
error
|
4
spread error1
|
5
mean abs
credit
spread error1
|
|
Merton
|
1.69
|
3.67
|
4.94
|
-50.42
|
78.02
|
Geske (face recovery)
|
0.70
|
3.22
|
4.89
|
-29.57
|
66.93
|
Geske (firm recovery)
|
2.09
|
3.11
|
3.97
|
-52.92
|
65.73
|
Leland-Toft
|
-1.79
|
4.06
|
7.54
|
115.69
|
146.05
|
LS (1-day CMT)
|
-2.69
|
5.63
|
8.19
|
42.93
|
124.83
|
LS (1-month CMT)
|
-0.68
|
4.56
|
6.94
|
6.63
|
96.83
|
CDG (baseline)
|
-11.21
|
12.64
|
13.12
|
269.78
|
319.31
|
CDG (low κ )
|
-10.5
|
12.09
|
13.03
|
251.12
|
304.32
|
CDG (low μ)
|
-3.76
|
7.35
|
10.13
|
78.99
|
170.16
|
median values
|
-1.79
|
4.56
|
7.54
|
42.93
|
124.83
|
1 Spread refers to the credit spread (yield
minus risk-free rate). Error is expressed as a percent of the bond credit
spread.
[1] This paper has been prepared for presentation and discussion at the
Workshop on Accounting Risk
Management and Prudential Regulation, Bank of International
Settlement, Basel, Switzerland, 11–12
November 2005. I thank Mary Barth, Bill Beaver, Brad Cornell, and Bruce Miller
for helpful comments.
[2] Associate Dean, PhD
Programme, KPMG Professor of Accounting, University of North Carolina, Kenan
Flagler Business School, CB 3490 McColl Building, Chapel Hill, NC 27599-3490,
USA.
[3] “Marking-to-market” and
“fair values” are often used as synonyms. Use of the former implies the
existence of active markets with determinable market prices. As described
below, “fair value” can have multiple meanings and does not necessarily depend
on the existence of active markets. Moreover, even if market prices exist, the
instrument’s value to the entity need not equal its quoted market price.
[4] The IASB defines fair value similarly.
[5] Although SFAS No 123
(revised) requires the cost of option grants be recognised at fair value, it is
not entirely a fair value standard, in that the amortisation of the cost of
option grants is based on the grant date fair value ie the historical cost of
the grants. As discussed below, Landsman, Peasnell, Pope and Yeh (2005)
advocate also recognising in income changes in fair value of option grants.
[6] The FASB has issued
several other standards with elements of fair value recognition or disclosure.
For example, SFAS no 87, Employers’
Accounting for Pensions (FASB, 1985) requires footnote disclosure of the
fair value of pension plan assets and the pension obligation associated with
defined benefit plans. However, the standard requires balance sheet recognition
of only the net of the unrecognised asset, liability, and equity amounts. The
SEC report (SEC, 2005) recommends that pension assets and liabilities be
recognised at fair value in the body of the financial statements. Evidence in
Landsman (1986) and Barth (1991) is consistent with equity prices reflecting
pension asset and liability fair values. See the literature review on pricing
effects of financial instruments’ fair values in the next section.
[7] The comment in footnote 5 relating to SFAS no 123 (revised) applies
also to IFRS 2.
[8] The IASB adopts a similar hierarchy in IAS 39.
[9] Bank regulators are also
interested in these and related questions. As discussed below, some US-based
studies address the effects of fair values on regulatory capital.
[10] Another equally plausible
explanation is that investment securities’ fair value gains and losses are
naturally hedged by fair value changes of other balance sheet amounts, which
are not included in the estimating equations. Ahmed and Takeda (1995), which
includes other on-balance sheet net assets in the estimating equations, provide
support for this explanation with evidence of incremental explanatory power for
unrecognised securities gains and losses in explaining banks’ stock returns.
[11] Bernard, Merton and
Palepu (1995) cautions that drawing inferences from the Danish experience with
fair value accounting for banks regarding the benefits of requiring fair value
accounting for US banks is subject to many caveats. These include differences
in the relative size of the US and Danish banking sectors, as well as relative
differences in US and Danish banking regulatory systems.
[12] See also the discussion
above of the Barth, Beaver and Landsman (1996) findings relating to loans fair
values estimates by banks with lower regulatory capital.
[13]
Note that neither the FASB nor IASB considers value-in-use as a candidate for
fair value if it differs from the other two prices.
[14] See FASB (1990, 2000) for
a description of the fundamental components approach to accounting for complex
financial instruments. In addition to the FASB, several other standard setters
have considered separating compound financial instruments into components,
including the CICA (Section 3860 of the CICA Handbook, “Financial
Instruments—Disclosure and Presentation”), the AASB (AASB Accounting Standard
1033, Presentation and Disclosure of Financial Instruments), and the IASB
(IASB, 2003a). Each of these standard setters representing Canada, Australia,
and the international community concludes issuers of a compound financial
instrument should present the liability components and equity components of
that financial instrument separately.
[15] Senior Economist,
Division of Research and Statistics, Risk Analysis Section, Board of Governors
of the Federal Reserve Bank, 20th Street and Constitution Avenue, Washington,
DC 20551, USA.
[16] Nonetheless, there
appears to be a significant omitted variables issue in Barth et al’s (1996)
estimated loan fair value coefficients, as well as in the other studies.
Unbiased estimates of the loan value coefficients in the equity regressions
(such as specified in Bart et al’s (1996) theoretical equation (4)) require
controlling for the value of deposit insurance under fixed-rate deposit
insurance systems. The equity regression equations use bank liabilities as an
explanatory variable but a substantial fraction of these liabilities are
(explicitly or implicitly) insured deposits. As such, the specifications omit
the value of the deposit insurance. This value will be negatively correlated
with the value of the bank’s assets, which will create a negative bias in the
estimated asset value coefficients relative to the hypothetical values in the
authors’ specifications.
[17] See Chava (2002) for a
loan commitment valuation model using contingent claim pricing methods.
[18] Aguais, Forest, and Rosen
(2000) give a detailed presentation of constructing valuations of corporate
loans, including loan commitments, that would ultimately make use of generic
market credit spreads.
[19]
For an extensive review of structural and reduced-form bond pricing models, see
Duffie and Singleton (2003).
[20]
The large errors in the model credit spreads as a percent of the actual credit
spreads could reflect errors of only a few basis points for high-grade bonds
with small spreads. However, this explanation of the large percentage spread
errors does not appear to be the case. In further graphic results, Eom et al
show that the errors in estimated spread levels increase dramatically with the
spread level in going from high-grade to junk bond status. 21 Their results on the added difficulties in
valuing complex instruments are somewhat ambiguous. In their estimations that
use the bond’s actual price to estimate model parameters, model accuracy for
straight bonds was not better than that for the full set of bonds that included
those with various provisions. However, they presented further evidence that
suggest difficulties in estimating the values of the individual provisions in
the bonds.
No comments