article summary- fair value

THE ACCOUNTING REVIEWVol. 85, No. 4
2010
pp. 1375–1410
American Accounting Association
DOI: 10.2308/accr.2010.85.4.1375
Chang Joon Song
Sungkyunkwan University
Wayne B. Thomas
Han Yi
The University of Oklahoma
ABSTRACT: Statement of Financial Accounting Standards No. 157 共FAS No. 157兲, Fair
Value Measurements, prioritizes the source of information used in fair value measurements into three levels: 共1兲 Level 1 共observable inputs from quoted prices in active
markets兲, 共2兲 Level 2 共indirectly observable inputs from quoted prices of comparable
items in active markets, identical items in inactive markets, or other market-related
information兲, and 共3兲 Level 3 共unobservable, firm-generated inputs兲. Using quarterly
reports of banking firms in 2008, we find that the value relevance of Level 1 and Level
2 fair values is greater than the value relevance of Level 3 fair values. In addition, we
find evidence that the value relevance of fair values 共especially Level 3 fair values兲 is
greater for firms with strong corporate governance. Overall, our results support the
relevance of fair value measurements under FAS No. 157, but weaker corporate governance mechanisms may reduce the relevance of these measures.
Keywords: FAS No. 157; fair value; fair value hierarchy; value relevance; corporate
governance.
Data Availability: Data are publicly available from the sources identified in the study.
We are grateful for helpful comments received from Matt Crow, Vicki Dickinson, Karen Hennes, Leslie Hodder, OleKristian Hope, William Kinney, Mark Kohlbeck, Christian Leuz, Tom Linsmeier, Bharat Sarath, participants of the 2008
Conference on Financial Economics and Accounting, the 2009 International Conference on Assurance and Governance
共University of Florida兲, the 2009 Spring Research Conference on Corporate Governance 共The University of Texas at
Austin兲, the 2009 AAA Annual Meeting, the FASB Office Hours, and two anonymous reviewers. We also appreciate the
research assistance of Ken Bills. Professor Song gratefully acknowledges support from Virginia Polytechnic Institute and
State University.
Editor’s note: Accepted by Mark Trombley.
Submitted: February 2009
Accepted: November 2009
Published Online: June 2010
1375
Downloaded from http://meridian.allenpress.com/accounting-review/article-pdf/85/4/1375/1349625/accr_2010_85_4_1375.pdf by Ronald Tsang on 23 October 2021
Value Relevance of FAS No. 157 Fair Value
Hierarchy Information and the Impact
of Corporate Governance Mechanisms
1376
Song, Thomas, and Yi
I. INTRODUCTION
n response to users’ requests for a comprehensive framework for measuring and disclosing fair
value, the Financial Accounting Standards Board 共FASB兲 issued Statement of Financial Accounting Standards No. 157, Fair Value Measurements 共FAS No. 157; FASB 2006兲 in 2006,
effective for fiscal years beginning after November 15, 2007. FAS No. 157 does not increase the
use of fair value measurements, but instead provides a uniform definition of fair value, establishes
a framework for measuring fair value, and expands disclosure about fair value measurements.
More specifically and of particular interest to our study, FAS No. 157 requires fair value assets and
liabilities to be disclosed by levels, where levels are based on inputs used to measure fair values:
共1兲 Level 1 共observable inputs from quoted prices in active markets兲, 共2兲 Level 2 共indirectly
observable inputs from quoted prices of comparable items in active markets, identical items in
inactive markets, or other market-related information兲, and 共3兲 Level 3 共unobservable, firmgenerated inputs兲.1
First, we test the value relevance of fair value measures for each of the three disclosure levels.
A long-standing debate of fair value accounting has been centered on the trade-off between
relevance and reliability. Proponents of fair value accounting argue that fair value information has
greater relevance, more accurately reflects real volatility, and simplifies financial reporting 共e.g.,
hedge accounting兲. In contrast, opponents of fair value accounting argue that fair value measurements are less verifiable by investors, subject to greater estimation error by management, and
prone to greater managerial manipulation. These shortcomings create information asymmetry
between investors and managers that can be a serious threat to the reliability of fair values
共Landsman 2007; Penman 2007兲. Because these problems are expected to become more severe as
fair value inputs become less observable by investors, we are particularly interested in testing the
value relevance of Level 1 fair values versus Level 3 fair values. The reliability and therefore
potential information asymmetry problems associated with Level 2 fair values potentially fall
between those of Level 1 and Level 3 fair values.
As our second test, we examine whether the value relevance of fair values depends on firms’
corporate governance mechanisms. The motivation for this test arises from the greater subjectivity
on the part of management in measuring and reporting fair values. Although in some instances
managers may use their private information to credibly report fair values 共e.g., Barth et al. 1998兲,
prior studies also provide evidence that managers may manipulate inputs for fair values for their
own interests 共e.g., Aboody et al. 2006; Bartov et al. 2007兲. For firms with weaker corporate
governance mechanisms, information asymmetry problems associated with fair values may be
greater, leading to more severe moral hazard problems, and therefore lower value relevance of
these disclosures. On the whole, Level 1 fair value reporting is likely to suffer the least from
information asymmetry, and one might expect corporate governance mechanisms to have the least
impact on the valuation of these fair values. Instead, corporate governance should probably play a
larger role in the value relevance of Level 3 fair values where information asymmetry is likely the
highest.
Using a sample of quarterly reports by banking firms in 2008, we provide the following
results. First, we find evidence that fair value measurements of Level 1, Level 2, and Level 3
I
The FASB Staff Position 共FSP兲 FAS No. 157-4 共FASB 2009兲 was issued in April 2009 to clarify the application of FAS
No. 157. It provides additional guidance on estimating fair values when the volume and level of activity for asset or
liability have significantly decreased. Specifically, FSP FAS No. 157-4 provides a list of factors that a company should
evaluate in order to determine whether a significant decrease has occurred in the volume and level of activity for the
asset or liability in relation to normal market activity for the asset or liability. When estimating fair value, the company
should place 共1兲 more weight on transactions that the company concludes are orderly; 共2兲 less weight on transactions for
which the company does not have sufficient information to conclude whether the transaction is orderly; and 共3兲 little, if
any, weight on transactions that the company concludes are not orderly.
The Accounting Review
American Accounting Association
July 2010
Downloaded from http://meridian.allenpress.com/accounting-review/article-pdf/85/4/1375/1349625/accr_2010_85_4_1375.pdf by Ronald Tsang on 23 October 2021
1
Value Relevance of FAS No. 157 Fair Value Hierarchy Information
1377
2
These theoretical predictions are based on implicit assumptions that 共1兲 the valuation model is properly specified and 共2兲
markets are efficient. Therefore, we do not claim that estimated coefficients must be 1 or ⫺1. We merely use these
theoretical values as benchmarks for our statistical testing.
The Accounting Review
July 2010
American Accounting Association
Downloaded from http://meridian.allenpress.com/accounting-review/article-pdf/85/4/1375/1349625/accr_2010_85_4_1375.pdf by Ronald Tsang on 23 October 2021
inputs are value-relevant. We also find that the valuation coefficient of Level 3 assets is significantly less positive than those of Level 1 and Level 2 assets. The lower valuation of Level 3 assets
is consistent with investors decreasing the weight they place on less reliable fair value measurements 共Maines and Wahlen 2006兲. The valuation coefficient of Level 3 liabilities is significantly
more negative than those of Level 1 and Level 2 liabilities, consistent with less reliable amounts
for Level 3 liabilities. We additionally test whether the coefficient on each Level is significantly
different from its theoretically predicted value of 1 for assets and ⫺1 for liabilities.2 In particular,
if investors perceive reliability concerns for Level 3 assets 共liabilities兲, the coefficient on these fair
values could be less than 1 共⫺1兲. Our results are consistent with this prediction. In order to
mitigate concerns that firm size may be correlated with the types of assets held by banks and that
the strength of a bank’s capital ratio could be correlated with managers’ choice of valuation levels,
we examine whether our results hold in subsamples partitioned by firm size and Tier 1 capital
ratio. We find that our results continue to hold in all subsamples.
For our second test, we examine whether the value relevance of fair values varies across six
individual governance mechanisms 共i.e., board independence, audit committee financial expertise,
the frequency of annual audit committee meetings, the percent of shares held by institutional
investors, the auditor’s office size, and no material control weakness problem under Sections 302
and 404 of the Sarbanes-Oxley Act兲, as well as a factor score based on these mechanisms. We find
that governance has a significant impact on Level 2 and Level 3 fair values. As the strength of
corporate governance increases, investors’ valuation of these fair value assets and liabilities increases toward the theoretically predicted coefficient values of 1 and ⫺1. These results highlight
the importance of corporate governance mechanisms in mitigating the information asymmetry
problem associated with Level 2 and Level 3 inputs. Consistent with the notion that Level 1 fair
value information suffers the least from information asymmetry, we find little impact of corporate
governance on these fair values.
Finally, we provide two additional analyses. First, we find that the ability of Level information
provided by FAS No. 157 interacted with existing asset and liability Type information has a
greater explanatory power for firm value than does Type information only. This difference suggests that valuation of Level information is not simply a reordering of valuation by Type information that existed prior to FAS No. 157 共i.e., FAS No. 157 provides new information兲. Second,
we consider how conclusions may have changed as the economic crisis worsened and market
illiquidity became more severe in 2008. Although we find no evidence of a decline in the valuation
of fair values during the first three quarters of 2008, we do find some evidence that the valuation
of Level 3 assets and liabilities actually moved closer to their predicted values of 1 and ⫺1. In
addition, we find evidence that the impact of corporate governance for valuation of fair value
assets remains significant over time.
Overall, we conclude that the fair value hierarchy required by FAS No. 157 provides useful
information to investors and the strength of corporate governance appears to mitigate the information asymmetry problem arising from relatively less reliable fair value inputs. These results
contribute to the literature on fair value accounting. We provide early evidence of the value
relevance of new disclosures under FAS No. 157. Prior to FAS No. 157, direct tests of the
association between the reliability of fair value information and equity prices were more difficult.
Using the fair value hierarchy under FAS No. 157, we provide direct evidence of the value
relevance of more reliable 共Level 1兲 versus less reliable 共Level 3兲 information.
1378
Song, Thomas, and Yi
II. LITERATURE REVIEW AND HYPOTHESES DEVELOPMENT FAS NO. 157
In recent years, U.S. Generally Accepted Accounting Principles 共GAAP兲 have evolved toward
greater use of fair values for reporting assets and liabilities. However, prior to FAS No. 157,
neither a single coherent definition for fair value nor detailed guidance for applying the fair value
measurement existed. In June 2003, the FASB added the fair value measurement project to its
agenda to address these issues. In September 2006, the FASB released FAS No. 157, Fair Value
Measurements. FAS No. 157 does not expand the use of fair value measurements, but provides a
coherent framework for applying fair value measurements and enhances disclosures about the
nature and source of fair value measurements to increase consistency and comparability.
The FASB has concluded that fair value information is relevant 共e.g., see Paragraph 217 of
FAS No. 133 or Paragraph C2 of FAS No. 157兲.5 To the extent that established, active markets for
3
4
5
We acknowledge that different samples, time periods, and variable definitions across these studies do not allow us to
make direct comparisons.
Other differences between our study and Kolev 共2009兲 and Goh et al. 共2009兲 include: 共1兲 we have a larger sample size
and/or longer sample period, 共2兲 we consider the impact of a larger set of corporate governance mechanisms on the
pricing of fair value assets under FAS No. 157, and 共3兲 we consider the incremental pricing effect of expanded
disclosure format under FAS No. 157. However, it should be noted these studies provide additional tests beyond ours
that will contribute to the literature. For example, Kolev 共2009兲 uses a changes specification, allowing tests of the value
relevance of Level 3 realized and unrealized gains/losses versus net purchases and transfers.
Claims have been made that fair value may not be the most relevant measurement attribute. For example, one claim is
that fair value may misrepresent management’s intent 共e.g., intent to hold/owe an asset/a liability to its maturity兲.
However, fair value measurements allow shareholders to assess whether holding an asset or owing a liability is appropriate by providing timely value change information. See Ryan 共2007, Ch. 6兲 for other claims and related discussions.
Also see Ramesh and Graziano 共2004兲 for various discussions around the benefits and costs of historical cost versus fair
value accounting.
The Accounting Review
American Accounting Association
July 2010
Downloaded from http://meridian.allenpress.com/accounting-review/article-pdf/85/4/1375/1349625/accr_2010_85_4_1375.pdf by Ronald Tsang on 23 October 2021
Two recent working papers 共Kolev 2009; Goh et al. 2009兲 provide tests of value relevance
similar to ours. Although we find valuations of Level 1 and Level 2 assets and liabilities close to
1 and ⫺1, these papers find valuations of Level 1 and Level 2 net assets significantly less than 1.3
Furthermore, Goh et al. 共2009兲 document that investors value Level 2 net assets less than Level 1
net assets but do not value Level 2 and Level 3 net assets differently. In contrast, we show that
Level 1 and Level 2 assets are valued similarly, while Level 3 assets are valued the least. Goh et
al. 共2009兲 also document that the value relevance of net fair value assets decreases over the first
three quarters of 2008, whereas we find that the value relevance of fair values does not decrease
over this period.4 Finding evidence that the value relevance of fair values does not decrease as
markets become less liquid may be particularly important to standard-setters who are interested in
the market’s perception of the reliability of fair values during an economic crisis.
We also contribute to the literature by examining directly the association between the strength
of corporate governance and value relevance of fair values. Standard-setters understand that information asymmetry would be higher for Level 3 assets and for this reason they required firms to
provide additional disclosures for these items 共e.g., the inputs used to measure fair value and the
effect of the measurements on earnings or changes in net assets兲. Presumably these additional
disclosures would reduce or even eliminate the information asymmetry problem. We find evidence
consistent with the information asymmetry problem continuing to exist, but the strength of corporate governance appears to ameliorate this problem. These results highlight the importance of
corporate governance for the value relevance of accounting information, especially for information
that is potentially less reliable.
The next section describes FAS No. 157, reviews relevant prior research, and develops hypotheses. Section III outlines the sample selection process and provides descriptive statistics.
Section IV presents empirical results. The study concludes with a discussion of the results.
Value Relevance of FAS No. 157 Fair Value Hierarchy Information
1379
Value Relevance of Fair Value Hierarchy
Accounting information is considered to be value-relevant when it has the predicted association with market value of equity 共Barth et al. 2001兲. If a significant association is found, then it is
assumed that the accounting information of research interest is relevant to investors and reliable
enough to be reflected in share prices. In this section, we discuss literature related to the value
relevance of fair value measures. We also detail why reliability concerns could affect the value
relevance of fair values. This discussion helps in formulating an expectation regarding the differential value relevance across Levels information under FAS No. 157.
Several studies document that fair values of investment securities of banks and propertyliability insurers are value-relevant 共Barth 1994; Petroni and Wahlen 1995; Barth et al. 1996;
Eccher et al. 1996; Nelson 1996; Carroll et al. 2003兲. Petroni and Wahlen 共1995兲 find that fair
values for equities and Treasury securities are value-relevant, but fair values of municipal and
corporate bonds are not, suggesting securities actively traded in the market are more reliably
associated with market value of equity. In contrast, using a sample of closed-end mutual funds,
Carroll et al. 共2003兲 find that fair values of thinly traded securities are value-relevant.
When no established market is available 共i.e., fair values are less reliable兲, the evidence of
value relevance of other financial instruments is mixed. For example, Nelson 共1996兲 finds that fair
values of loans, deposits, and long-term debt are not value-relevant. In contrast, Barth et al. 共1996兲
find that fair values of loans are value-relevant, whereas Eccher et al. 共1996兲 find the value
relevance of fair values of loans only in limited settings. Finally, Venkatachalam 共1996兲 examines
the value relevance of fair values of derivatives and finds that these fair values are positively
associated with market values of equity.
Another stream of fair value accounting research examines how managerial opportunism
affects the reliability of fair value measurements. Beaver and Venkatachalam 共2003兲 partition U.S.
bank loan fair values into three categories 共i.e., non-discretionary, discretionary, and noisy components兲 and show that the pricing coefficient of the discretionary loan component is negative if
The Accounting Review
July 2010
American Accounting Association
Downloaded from http://meridian.allenpress.com/accounting-review/article-pdf/85/4/1375/1349625/accr_2010_85_4_1375.pdf by Ronald Tsang on 23 October 2021
assets/liabilities are in place, fair value is likely to be the most relevant measurement attribute.
However, fair value measurements in the absence of observed prices might be unreliable due to
intrinsic measurement error 共noise兲 and management-induced error 共bias兲. FAS No. 157 establishes a fair value hierarchy that prioritizes the inputs used to measure fair values into three broad
levels, considering the relative reliability of the inputs to fair value measurements. Level 1 inputs
are quoted prices in active markets, and a quoted price is a single primary basis for the fair value
measurement. Therefore, the information asymmetry between preparers and users is very low for
Level 1 inputs. Level 2 inputs include either observable prices in active market for comparable
assets and liabilities or observable market prices in inactive markets for identical assets and
liabilities. Level 2 inputs could also include prices corroborated by market-based measures 共e.g.,
correlation with the yield curve兲. Level 3 inputs are not observable from the market and reflect
management’s assessment of the assumptions that market participants would use in pricing the
asset or liability.
The key distinction between Level 2 inputs and Level 3 inputs is whether inputs are observable. This distinction is important for verifiability of fair values. Therefore, the Level 3 inputs are
subject to the highest degree of information asymmetry between preparers and users. FAS No. 157
enhances fair value disclosure by requiring firms to provide fair value measurements by input
levels in the hierarchy, enabling users to assess the relative reliability of fair value measurements.
The Appendix provides an example of FAS No. 157 fair value measurements disclosures from a
quarterly report filed by Wells Fargo & Company.
1380
Song, Thomas, and Yi
6
7
Nissim 共2003兲 examines whether the U.S. bank managers strategically overstate bank loans, finding that the estimated
degree of overstatement systematically varies with regulatory capital requirements and the change in the rate of credit
losses. In contrast, Bernard et al. 共1995兲 examine whether Danish bank managers manage fair value estimates to meet
the mark-to-market based regulatory capital requirement in Denmark, but find no such evidence.
In contrast, Hodder et al. 共2006兲 find that a large proportion of firms use value-increasing discretion in their valuation
model inputs. Note that FAS No. 157 does not apply to stock option compensation measurements in FAS No. 123共R兲.
However, these studies are relevant, showing managerial discretion over input values in option value models.
The Accounting Review
American Accounting Association
July 2010
Downloaded from http://meridian.allenpress.com/accounting-review/article-pdf/85/4/1375/1349625/accr_2010_85_4_1375.pdf by Ronald Tsang on 23 October 2021
the managerial intent for the discretion is likely opportunistic.6 Studies also examine whether
managers use private information for inputs to option valuation models to help or hurt information
users in the U.S. setting. For example, in the case of stock option expense reporting under FAS
No. 123 共FASB 2004兲, Aboody et al. 共2006兲 and Bartov et al. 共2007兲 show that managerial
incentives induce managers’ discretion over model assumptions 共e.g., the expected option life and
stock volatility input assumptions兲, lowering option values.7
In addition to financial assets, several studies support the value relevance of fair value nonfinancial assets 共e.g., Barth and Clinch 1998; Aboody et al. 1999; Easton et al. 1993; Dietrich et
al. 2000兲. Some of these studies also examine whether the value relevance of fair value measurements varies with the reliability of the information. For example, Dietrich et al. 共2000兲 find the
reliability of fair value estimates is an increasing function of the presence of monitoring of
external 共as opposed to internal兲 appraisers. Similarly, Muller and Riedl 共2002兲 provide evidence
that market participants appear to find external appraisals to be more reliable. However, Barth and
Clinch 共1998兲 find no difference in value relevance between internal and external appraisals.
In summary, the literature generally suggests that fair value measurements are value-relevant,
and that the value relevance of fair value measurements varies with the source of information.
Although the source of fair value measurements is assumed to be associated with the reliability of
information, direct archival tests of the association between the reliability of fair value information
and equity prices were difficult prior to FAS No. 157. FAS No. 157 now explicitly categorizes the
inputs to fair value measurements and allows us to assess whether the value relevance of less
reliable fair values 共i.e., Level 3兲 is different from that of more reliable fair values 共i.e., Level 1兲.
Regarding investors and how they weigh reported fair values in their pricing of the firm, two
factors are important to consider: the discount adjustment and the cash flow adjustment. Investors,
to the extent that they perceive greater uncertainty of fair values, will adjust upward the discount
rate applied to those reported amounts, resulting in less than a dollar-for-dollar valuation. Furthermore, investors, to the extent that they perceive reported assets 共liabilities兲 to be biased upward
共downward兲, either intentionally or unintentionally, will adjust downward for a cash flow effect.
Again, this adjustment results in less than a dollar-for-dollar valuation of reported amounts. Below
we discuss why investors’ downward valuation adjustment is likely to be greater for reported
Level 3 fair values than for reported Level 1 and Level 2 fair values. Given that our tests of value
relevance estimate investors’ adjustment to reported fair values 共i.e., valuation coefficient on
reported discounted cash flows兲, we do not specify whether the adjustment is more likely to be the
consequence of a discount effect versus a cash flow effect. Rather, we contend that both are likely
contributing factors.
Level 3 fair values are less observable and are naturally subject to greater information asymmetry between investors and management. In addition, the more subjective nature of Level 3 fair
values makes them prone to greater estimation error by management 共i.e., noise兲. Both information
asymmetry and estimation error inherent to the production of specific accounting information
increase investors’ adverse selection, liquidity risk, and information-processing costs, all of which
Value Relevance of FAS No. 157 Fair Value Hierarchy Information
1381
This concept relates to estimation risk 共Barry and Brown 1985兲 and investors’ assessment of the extent to which a firm’s
cash flows are correlated with other firms’ cash flows 共Lambert et al. 2007兲. The extent to which a firm’s 共assessed兲 cash
flows are non-diversifiable positively impacts its cost of capital. These conjectures are supported empirically 共e.g., Leuz
and Verrecchia 2000; Francis et al. 2004; Hope et al. 2009兲.
9
Another type of bias is unintentional optimism by managers 共e.g., psychological upward bias兲 共Martin et al. 2006; Roll
1986兲.
10
Several studies find evidence consistent with higher quality accounting information facilitating investors’ ability to
monitor managers’ decisions and therefore improving firm performance 共Lombardo and Pagano 2002; Bens and Monahan 2004; Kanodia et al. 2004; Biddle and Hilary 2006; Hope and Thomas 2008兲.
8
The Accounting Review
July 2010
American Accounting Association
Downloaded from http://meridian.allenpress.com/accounting-review/article-pdf/85/4/1375/1349625/accr_2010_85_4_1375.pdf by Ronald Tsang on 23 October 2021
increase a firm’s cost of capital 共Diamond and Verrecchia 1991; Baiman and Verrecchia 1996兲.8
These effects hold even without the presence of moral hazard. As cost of capital increases, the
value of a firm’s assets decreases. From an investor’s perspective, accounting amounts that are less
reliable 共e.g., Level 3 fair values兲 are assigned a higher cost of capital and, therefore, are valued
less than are more reliable accounting amounts 共e.g., Level 1 fair values兲. Thus, we expect
investors’ downward adjustment to Level 3 fair values to be greater because of a greater discount
effect.
We now turn our discussion to cash flow effects. By cash flow effect, we mean that investors
perceive management’s estimate of future cash flows to differ systematically from realized future
cash flows 共i.e., bias兲. When certain accounting information is highly subjective in nature, and
managers are allowed to exercise a high degree of discretion over it, managers may be more likely
to generate intentional biases in their estimations 共e.g., Aboody et al. 2006; Bartov et al. 2007兲.9
To the extent that these biases are expected on average, investors likely adjust such estimates in
valuing the firm. Specifically, if investors are concerned about possible overstatement of Level 3
fair value assets and understatement of Level 3 fair value liabilities, then they will adjust their
valuation of management-reported Level 3 assets and liabilities to less than 1 and ⫺1, respectively.
In addition, less reliable accounting information reduces the ability of investors to monitor
managerial behavior, potentially reducing the firm’s operating performance and future cash flows.
Bushman and Smith 共2001兲 discuss the important role of financial accounting information as a
mechanism to discipline managerial behavior. As the quality of financial information deteriorates,
investors lose their ability to link the activities of the manager to firm performance. In other words,
without the disciplining mechanism afforded by reliable financial accounting information, managers are held less accountable for their actions and therefore operate the firm less efficiently or
extract private benefits directly, both of which are detrimental to firm value.10 Level 3 fair values
are less observable, making it more difficult 共compared to Level 1 fair values兲 for investors to link
their performance to managerial decisions, reducing the efficiency of these activities.
Because of the issues discussed above, investors are likely to decrease the weight they place
on less reliable Level 3 fair value measurements in their equity-pricing decisions 共Maines and
Wahlen 2006兲. Level 1 fair values are expected to suffer less from these issues, as these amounts
can be easily verified by investors. Thus, we expect the pricing coefficient of Level 1 fair value
measurements to be greater than that of Level 3 fair value measurements. Regarding Level 2 fair
values, these measurements potentially represent the middle ground of reliability. Level 2 fair
values are based on observable market inputs, but in some cases it may be more difficult for
investors to observe directly how bank managers adapt those inputs to generate reported fair
values. Because observability of market inputs for Level 2 fair values is meant to indicate increased reliability of fair value measurements, we expect the valuation of Level 2 fair values to be
greater than that of Level 3 fair values, but we do not make a prediction for the difference in
valuation of Level 1 versus Level 2 fair values. This leads to our first hypothesis 共stated in
alternative form兲:
1382
Song, Thomas, and Yi
H1: The value relevance of Level 1 and Level 2 fair values is greater than the value relevance of Level 3 fair values.
The Effect of Corporate Governance on the Value Relevance of Fair Values
In the previous section, we discuss how intrinsic estimation error 共noise兲 and managementinduced error 共bias兲 inherent in fair value measurements might reduce the ability of investors to
rely on these amounts. In this section, we consider the extent to which stronger corporate governance mechanisms can mitigate these problems. We do so by testing whether the value relevance
of fair values varies with the strength of a firm’s corporate governance.
Given that measurement errors are likely to be more severe for inputs without observable
prices 共Level 3 and perhaps Level 2兲 than for inputs directly observable in active markets 共Level
1兲, we expect corporate governance to be more effective at mitigating problems associated with
Level 3 fair values.
Bhat 共2009兲 provides evidence that over the 2003–2005 period market participants perceived
fair value gains and losses of banks with strong corporate governance as more relevant and
reliable. She interprets these results to mean that corporate governance aids market participants in
evaluating the quality of fair value estimates. Aboody et al. 共2006兲 find firms with stronger
governance are less likely to understate option value estimates. Penman 共2007兲 discusses the
importance of both the competence and independence of monitors, as well as the effectiveness of
internal control systems in minimizing biases in Level 3 fair value estimates.
To the extent that strong corporate governance mechanisms reduce reliability concerns relating to estimation error and management induced bias in Level 3 fair values, investors are more
likely to view Level 3 fair values as relevant, which leads to our second hypothesis 共stated in
alternative form兲:
H2: Corporate governance has a greater impact on the value relevance of Level 3 fair values
than on the value relevance of Level 1 or Level 2 fair values.
III. SAMPLE SELECTION AND DESCRIPTIVE STATISTICS
Sample Selection
To maximize the power of our value relevance test, we focus on the banking industry where
firms have significant amounts of fair value assets and liabilities. Table 1 delineates the sample
selection process. The sample firms were initially identified from Compustat Bank Fundamentals
Quarterly Research File. To be included in the sample, firms need to provide the first quarter FAS
No. 157 fair value hierarchy disclosure after November 15, 2007 共n ⫽ 522兲. To avoid the confounding effect of prices from different macroeconomic events, we focus on firms that end the first
quarter on March 31, 2008, eliminating 10 non-March 31 quarter firms 共n ⫽ 512兲. We further
require firms to have price information in the Center for Research in Security Prices 共CRSP兲
database 共n ⫽ 452兲. To avoid the effect from extreme outliers, we follow Belsley et al. 共1980兲 and
Fox 共1991兲, eliminating 21 observations that have studentized residuals greater than 2 in the
estimation of Equation 共1兲 below. This procedure yields an initial sample of 431 firms.
The Accounting Review
American Accounting Association
July 2010
Downloaded from http://meridian.allenpress.com/accounting-review/article-pdf/85/4/1375/1349625/accr_2010_85_4_1375.pdf by Ronald Tsang on 23 October 2021
Although not explicitly stated as part of our hypothesis, we also test whether valuation
coefficients differ from their theoretically predicted values. That is, if reported fair values accurately represent their underlying economic value, then investors are expected to assign dollar-fordollar value to these amounts. Easily verifiable in most instances, one would predict Level 1 and
Level 2 assets 共liabilities兲 to have valuation coefficients close to 1 共⫺1兲. Furthermore, consistent
with arguments above to support H1, the more subjective, less reliable Level 3 assets 共liabilities兲
would have a valuation coefficient less than 1 共⫺1兲.
Value Relevance of FAS No. 157 Fair Value Hierarchy Information
1383
TABLE 1
Sample Selection
Process
522
共10兲
共60兲
共21兲
Sample for the Test of H1 共Table 3兲
Sample for the Additional Type Test 共Table 7兲
Less: Firms that do not have proxy statements
or have incomplete information on governance
variables
431
431
共26兲
419
410
1,260
431
Sample for the Test of H2 共Table 6兲
405
398
392
1,195
This table delineates the sample selection process. The sample firms were initially identified from Compustat Bank
Fundamentals Quarterly Research File. To be included in the sample, firms should provide the first quarter FAS No. 157
fair value hierarchy disclosure after November 15, 2007 共n ⫽ 522兲. To avoid the confounding effect of prices from
different macroeconomic events, we focus on firms that end the first quarter on March 31, 2008, eliminating 10 non-March
31-quarter firms 共n ⫽ 512兲. We further require firms to have price information in the Center for Research in Security Prices
共CRSP兲 database 共n ⫽ 452兲. To avoid the effect from extreme outliers, we follow Belsley et al. 共1980兲 and Fox 共1991兲,
eliminating 21 observations that have studentized residuals greater than 2 in the estimation of Equation 共1兲. This procedure
yields an initial sample of 431 firms for H1. Hypothesis 2 requires the sample firms to have valid governance variables. We
further eliminate 26 firms due to missing proxy statements or incomplete proxy statements to determine the value of
corporate governance variables of interest in this study. An additional analysis 共Table 7兲 is performed based on the first
quarter observations only 共n ⫽ 431兲 due to manual data collection of fair value asset/liability Type information.
We test H1 for the first three quarters of 2008. Some firms fall from the sample. Therefore the
number of firms in the second and third quarters is reduced, resulting in a total of 1,260 firmquarter observations for tests of H1.11 As discussed in more detail below, tests of H2 require
hand-collected corporate governance data. Thus, we collect corporate governance data only for the
first quarter of 2008 共i.e., using the initial 431 firms兲 and assume that these governance variables
remain constant in 2008. For tests of H2, missing or incomplete proxy statement data eliminate 65
bank-quarter observations for 26 firms from these tests 共n ⫽ 1,195兲.
Descriptive Statistics
Panel A of Table 2 provides descriptive statistics on the relative size of fair value assets and
liabilities from 1,260 firm-quarters. Compared to total assets and total liabilities, the mean total
11
In sensitivity tests discussed below, we rerun all tests using a constant sample of firms 共i.e., firms that have complete
data for the first three quarters of 2008兲. None of our inferences are affected when using this slightly reduced sample.
The Accounting Review
July 2010
American Accounting Association
Downloaded from http://meridian.allenpress.com/accounting-review/article-pdf/85/4/1375/1349625/accr_2010_85_4_1375.pdf by Ronald Tsang on 23 October 2021
Firms that report the first quarter FAS No. 157
information in Compustat Bank Fundamentals
Quarterly after November 15, 2007.
Less: Non-March 31 1st Quarter Ending Firms
Less: Firms that do not have price information
in the CRSP database
Less: Outliers that have studentized residual
greater than 2 in the estimation of Equation 共1兲,
following Belsley et al. 共1980兲 and Fox 共1991兲
# of Firms
# of Firms
# of Firms
(1st quarter) (2nd quarter) (3rd quarter) Total
Std. Dev.
25th
Percentile
50th
Percentile
75th
Percentile
FVA/Total Assets
FVA1/Total Assets
FVA2/Total Assets
FVA3/Total Assets
FVL/Total Liabilities
FVL1/Total Liabilities
FVL2/Total Liabilities
FVL3/Total Liabilities
9.89%
3.32%
9.94%
1.51%
2.57%
0.24%
2.36%
0.20%
7.93%
0.00%
5.93%
0.00%
0.00%
0.00%
0.00%
0.00%
13.81%
0.04%
12.32%
0.00%
0.00%
0.00%
0.00%
0.00%
19.96%
0.57%
18.23%
0.33%
0.01%
0.00%
0.00%
0.00%
Panel B: Per Share Value of Price, Non-Fair Value, Fair Value Assets and Liabilities, and Income
n
(Firm25th
Variable
Quarters)
Mean
Std. Dev.
Percentile
50th
Percentile
75th
Percentile
PRICE
NFVA/Share
FVA1/Share
FVA2/Share
FVA3/Share
NFVL/Share
FVL1/Share
FVL2/Share
FVL12/Share
FVL3/Share
NI/Share
11.625
121.797
0.061
16.352
0.000
131.275
0.000
0.000
0.000
0.000
0.163
17.960
163.279
0.815
27.250
0.502
171.936
0.000
0.002
0.004
0.000
0.326
1,260
1,260
1,260
1,260
1,260
1,260
1,260
1,260
1,260
1,260
1,260
1,260
1,260
1,260
1,260
1,260
1,260
1,260
1,260
14.97%
1.15%
13.31%
0.51%
0.37%
0.03%
0.32%
0.03%
14.024
133.606
1.842
21.150
0.889
142.619
0.071
0.783
0.854
0.044
0.008
9.798
73.340
6.209
21.321
2.712
77.589
0.701
7.552
8.178
0.380
0.915
7.705
85.004
0.000
7.630
0.000
89.420
0.000
0.000
0.000
0.000
0.032
July 2010
(continued on next page)
Song, Thomas, and Yi
Panel A: Relative Size of Fair Value Assets and Liabilities
n
(FirmVariable
Quarters)
Mean
Downloaded from http://meridian.allenpress.com/accounting-review/article-pdf/85/4/1375/1349625/accr_2010_85_4_1375.pdf by Ronald Tsang on 23 October 2021
1384
The Accounting Review
American Accounting Association
TABLE 2
Descriptive Statistics
Level 1
Level 2
Level 3
INVSEC
LOAN
AOTHER
Freq.
%
Freq.
%
Freq.
%
Freq.
%
260/431
390/431
122/431
60.32%
90.49%
28.31%
7/431
65/431
44/431
1.62%
15.08%
10.21%
11/431
96/431
24/431
2.55%
22.27%
5.57%
20/431
35/431
51/431
4.64%
8.12%
11.83%
Liability
Level 1
Level 2
Level 3
ADEV
TRDL
LTDEBT
LDEV
LOTHER
Freq.
%
Freq.
%
Freq.
%
Freq.
%
6/431
7/431
1/431
1.39%
1.62%
0.23%
5/431
27/431
6/431
1.16%
6.26%
1.39%
10/431
85/431
17/431
2.32%
19.72%
3.94%
11/431
16/431
8/431
2.55%
3.71%
1.86%
Panel D: Relative Frequency and Size of Fair Value Hierarchy Information within Each Fair Value Asset Type (n ⴝ Firms from the First Quarter
2008 that Have Non-Zero Asset Types)
Asset Types
ADEV
LOAN
AOTHER
Relative
Freq.
Relative
Size
Relative
Freq.
Relative
Size
Relative
Freq.
Relative
Size
Relative
Freq.
Relative
Size
260/424
390/424
122/424
15.7%
82.2%
2.1%
7/108
65/108
44/108
1.2%
90.3%
8.5%
11/97
96/97
24/97
2.8%
62.0%
35.2%
20/84
35/84
51/84
1.2%
90.3%
8.5%
Panel A provides descriptive statistics on the relative size of fair value assets and liabilities by Level. FVA 共FVL兲 indicates total fair value of assets 共liabilities兲. FVA1 共FVL1兲 indicates
fair value of Level 1 assets 共liabilities兲. Similarly, FVA2, FVA3, FVL2, and FVL3 are fair values of Level 2 and Level 3 assets and liabilities.
Panel B provides descriptive statistics on share price, non-fair value assets 共NFVA兲, non-fair value liabilities 共NFVL兲, fair value assets and liabilities, and net income 共all amounts per
share兲. FVA1 共FVL1兲 indicates fair value of Level 1 assets 共liabilities兲. Similarly, FVA2, FVA3, FVL2, and FVL3 are fair values of Level 2 and Level 3 assets and liabilities. FVL12
is the sum of FVL1 and FVL2. NI is income before extraordinary items for common shareholders.
Panel C provides the frequency 共i.e., the number of non-zero observations兲 and the percentage 共i.e., the frequency divided by 431 observations from the first quarter of 2008兲 of
reported fair value assets and liabilities by both asset/liability types and the fair value hierarchy. We classify fair value assets and liabilities into the following categories: 共1兲 INVSEC
(continued on next page)
1385
July 2010
American Accounting Association
Level 1
Level 2
Level 3
INVSEC
Downloaded from http://meridian.allenpress.com/accounting-review/article-pdf/85/4/1375/1349625/accr_2010_85_4_1375.pdf by Ronald Tsang on 23 October 2021
Assets
Value Relevance of FAS No. 157 Fair Value Hierarchy Information
The Accounting Review
Panel C: Frequency of Non-Zero Reported Fair Value Assets/Liabilities by the Fair Value Hierarchy (n ⴝ 431 Firms from the First Quarter 2008)
Song, Thomas, and Yi
July 2010
Downloaded from http://meridian.allenpress.com/accounting-review/article-pdf/85/4/1375/1349625/accr_2010_85_4_1375.pdf by Ronald Tsang on 23 October 2021
1386
The Accounting Review
American Accounting Association
⫽ investment securities including available for sale securities and trading securities; 共2兲 ADEV ⫽ derivative assets; 共3兲 LOAN ⫽ loan; 共4兲 AOTHER ⫽ all other asset items including
mortgage service rights, federal funds sold, assets-backed security, and other investments; 共5兲 TRDL ⫽ trading liabilities; 共6兲 LTDEBT ⫽ debts; 共7兲 LDEV ⫽ derivative liabilities;
and 共8兲 LOTHER ⫽ all other liability items.
Panel D provides the relative frequency and the relative size 共i.e., dollar amount兲 of fair value hierarchy information for each fair value asset type based on the non-zero observations.
For example, there are 424 non-zero investment security observations out of 431 firms from the first quarter of 2008. Within 424 firm observations, 260, 390, and 122 firms have
Level 1, Level 2, and Level 3 investment securities, respectively, and the relative sizes 共i.e., dollar amount兲 of these assets are 15.7%, 82.2% and 2.1%, respectively. We classify fair
value assets into the following categories: 共1兲 INVSEC ⫽ investment securities including available for sale securities and trading securities; 共2兲 ADEV ⫽ derivative assets; 共3兲 LOAN
⫽ loan; and 共4兲 AOTHER ⫽ all other asset items including mortgage service rights, federal funds sold, assets-backed security, and other investments.
Value Relevance of FAS No. 157 Fair Value Hierarchy Information
1387
IV. RESULTS
Value Relevance of Fair Value Hierarchy (H1)
To test the value relevance of FAS No. 157’s fair value hierarchy, we estimate the
association between share prices and fair values of assets and liabilities per share using a modified
Ohlson 共1995兲 model, which has been extensively employed in the literature. Barth and Clinch
共2009兲 provide evidence that share-deflated specifications 共as opposed to equity book value12
The relative size of fair value assets and non-fair value assets reported in this study is comparable to those reported in
Bech and Rice 共2009兲. They collect balance sheet data from the Consolidated Reports of Condition and Income for
ensured domestic commercial banks and non-deposit trust companies. They report that in 2008 共our sample period兲 81
percent of banks’ assets consist of the following: loans and leases 共57 percent兲, noninterest-earning assets 共15 percent兲,
federal funds transactions 共6 percent兲, and other 共3 percent兲. These amounts are typically carried at historical cost or
carrying amount. The remaining 19 percent of assets are classified as investments. Of this amount, any securities not
classified as held-to-maturity would be reported at fair value, requiring classification into Levels 1–3. Thus, the portion
of fair value assets for our sample 共15 percent兲 is similar to that of all banking institutions in the U.S.
The Accounting Review
July 2010
American Accounting Association
Downloaded from http://meridian.allenpress.com/accounting-review/article-pdf/85/4/1375/1349625/accr_2010_85_4_1375.pdf by Ronald Tsang on 23 October 2021
fair value assets and liabilities are about 15 percent and 0.4 percent, respectively.12 The fair value
amounts under Level 2 inputs account for most fair values.
Panel B of Table 2 presents descriptive statistics of variables used to test for the value
relevance of Level information under FAS No. 157. All variables are per-share numbers. The mean
share price 共PRC兲 is 14.024. The mean of non-fair value assets 共NFVA兲 is 133.606, and the means
of fair value assets using Level 1 inputs 共FVA1兲, Level 2 inputs 共FVA2兲, and Level 3 inputs 共FVA3兲
are 1.842, 21.150, and 0.889, respectively. Similarly, the mean of non-fair value liabilities 共NFVL兲
is 142.619, whereas the means of fair value liabilities using Level 1 inputs 共FVL1兲, Level 2 inputs
共FVL2兲, and Level 3 inputs 共FVL3兲 are 0.071, 0.783, and 0.044, respectively.
In Panel C of Table 2 we provide a matrix of the reporting frequency of fair value Levels by
hand-collected Type information from the first quarter 10-Qs of our 431 sample firms. For Type
information, we classify fair value assets and liabilities into the following categories: 共1兲 investment securities 共INVSEC兲 including trading securities and available-for-sale securities; 共2兲 derivative assets 共ADEV兲; 共3兲 loans 共LOAN兲; 共4兲 other assets 共AOTHER兲 including mortgage service
rights, federal funds sold, asset-backed securities, and other investments; 共5兲 trading liabilities
共TRDL兲; 共6兲 long-term debt 共LTDEBT兲; 共7兲 derivative liabilities 共LDEV兲; and 共8兲 other liabilities
共LOTHER兲. Panel C shows, for example, that 260 out of 431 firms 共60.32 percent兲 report Level 1
investment securities. Similarly, 390 firms 共90.49 percent兲 and 122 firms 共28.31 percent兲 report
Level 2 and Level 3 investment securities, respectively. Across the four asset types, a reasonable
number of firms disclose fair values at all three levels, with just a few exceptions. The average
frequency across the 12 cells is 21.75 percent. However, for liabilities, the average frequency
across the 12 cells is only 3.85 percent, and only one cell 共Level 2 derivative liabilities, LDEV兲 has
a reasonable number of observations. In sum, both the amount and the frequency of fair values are
much greater for assets than for liabilities. Nevertheless, the fact that all asset/liability types are
represented in all levels of FAS No. 157’s hierarchy provides evidence of the potential usefulness
of these expanded disclosures.
Whereas Panel C of Table 2 shows how many firms report nonzero values for each type and
level, Panel D provides the relative dollar-value size of fair value hierarchy information within
each fair value asset type. For example, out of 424 nonzero INVSEC observations from the first
quarter of 2008, 260, 390, and 122 firms have Level 1, Level 2, and Level 3 investment securities,
and the relative sizes of these assets are 15.7 percent, 82.2 percent, and 2.1 percent of total
investment securities, respectively. The relative size is qualitatively similar to relative frequency
reported in Panel C.
1388
Song, Thomas, and Yi
deflated, lagged price-deflated, returns, or equity market value-deflated specifications兲 perform the
best in reducing scale effects in the modified Ohlson 共1995兲 model. We partition book value into
non-fair value assets and liabilities and each of the fair value levels.13
PRCit = ␣0 + ␣1NFVAit + ␣2FVA1it + ␣3FVA2it + ␣4FVA3it + ␣5NFVLit + ␣6FVL12it
共1兲
Due to the low frequency of fair value liability reporting in our sample 共see Table 2兲, we
combine Level 1 and Level 2 liabilities 共FVL12兲. The dependent variable, PRC, is per share price
measured on the 10-Q filing month-end for firm i in quarter t. Other variables are as defined
previously on a per share basis.
In estimating Equation 共1兲, we pool observations from the first three quarters of 2008 for each
firm. Residuals could be correlated across quarters or across firms, so we correct standard errors
and related t-statistics based on two dimensions 共i.e., firms and quarters兲 following Petersen
共2009兲.14 Petersen 共2009兲 shows that this standard error adjustment 共or clustering by two dimensions兲 produces less-biased standard errors.
Hypothesis 1 examines whether the value relevance of Level 1 and Level 2 fair values is
greater than the value relevance of Level 3 fair values. In Table 3 we report results of Equation 共1兲
and test whether the valuation coefficient is 0 共Column A兲, whether it is different from 1 for assets
共Column B兲, and whether it is different from ⫺1 for liabilities 共Column C兲. Before reporting tests
of H1, we first note that estimated coefficients for both non-fair value assets/liabilities and fair
value assets/liabilities are different from 0, indicating their value relevance.15 In addition, we find
that the coefficients on FVA1 and FVA2 are not different from their theoretically predicted value of
1. In contrast, the coefficient on FVA3 is significantly less than 1, meaning that investors place less
weight on Level 3 fair value assets relative to Level 1 and Level 2. At the bottom of Table 3, we
provide tests of whether the coefficients across Levels are equal. We find that the coefficients on
Level 1 and Level 2 assets are not significantly different, and these coefficients are both significantly greater than the coefficient on Level 3 assets. These results support H1.
As for liabilities, the coefficient on FVL12 is not statistically different from its predicted value
of ⫺1. For FVL3, we find the valuation coefficient to be significantly less than ⫺1 共or greater than
1 in absolute terms兲 and less than the coefficient on FVL12. The coefficient’s magnitude is
consistent with investors’ perception that Level 3 fair value liabilities are understated.
In summary, Level 1 and Level 2 fair value estimates are value-relevant and valuation coefficients are not statistically different from their theoretical values of 1 or ⫺1. For Level 3 fair
13
We believe that a price model rather than a returns model is appropriate for testing our hypotheses because our research
question is to determine whether fair value hierarchy is reflected in firm value 共i.e., value relevance兲, rather than testing
whether fair value hierarchy is reflected in changes in firm value over a specific period of time 共see Barth et al. 2001兲.
14
As an example of how failure to adjust standard errors in this manner can affect inferences in accounting research, see
Gow et al. 共2010兲. We used cluster2 command in STATA provided by Mitchell Petersen. Our results are robust to
alternative panel data model specifications such as the firm fixed effect model.
15
Based on discussion in Barth and Clinch 共2009兲, our theoretical prediction is that the coefficient on non-fair value assets
共liabilities兲 would be between 0 and 1 共⫺1兲. We note that our observed coefficients are similar to those reported in
earlier research 共e.g., Barth 共1994兲 who reports a coefficient of 0.79 for historical cost book value in a regression similar
to ours兲 and to those reported in more recent studies 共e.g., Kolev 共2009兲 reports coefficients on book value of equity
between 0.709 and 0.748兲.
The Accounting Review
American Accounting Association
July 2010
Downloaded from http://meridian.allenpress.com/accounting-review/article-pdf/85/4/1375/1349625/accr_2010_85_4_1375.pdf by Ronald Tsang on 23 October 2021
+ ␣7FVL3it + ␤1NIit + ␧it
(A)
Independent
Variables
Intercept
NFVA
FVA1
FVA2
FVA3
NFVL
FVL12
FVL3
NI
n
Adj. R2
Coeff.
1.632
0.801
0.968
0.972
0.683
⫺0.818
⫺1.006
⫺2.185
2.488
1260
56.53%
(B)
Robust
Std. Err.
t-stat
Coeff. ⴝ 0
0.602
0.080
0.102
0.098
0.112
0.089
0.134
0.291
1.018
2.710
9.990
9.490
9.900
6.110
⫺9.170
⫺7.520
⫺7.500
2.440
p-value
F-stat
Coeff. ⴝ 1
0.007***
0.000***
0.000***
0.000***
0.000***
0.000***
0.000***
0.000***
0.015**
6.120
0.100
0.080
8.040
(C)
p-value
F-stat
Coeff. ⴝ ⴚ1
p-value
0.014**
0.751
0.773
0.005***
F-stat
p-value
Test of FVA1 ⫽ FVA2
Test of FVA1 ⫽ FVA3
Test of FVA2 ⫽ FVA3
Test of FVL12 ⫽ FVL3
0.010
17.250
18.630
0.933
0.000***
0.000***
0.041**
0.966
0.000***
F-stat
p-value
11.160
0.001**
*, **, *** Indicate statistical significance at the 0.10, 0.05, and 0.01 levels 共two-tailed兲, respectively.
This table provides the result of OLS regression of share price on non-fair value and fair value assets and liabilities. The sample includes 1,260 firm-quarters of 431 distinct firms
from the first three quarters of 2008. NFVA is non-fair value assets per share. FVA1, FVA2, and FVA3 are fair value assets per share from Levels 1, 2, and 3 inputs, respectively. NFVL
is non-fair value liabilities per share. FVL12 and FVL3 are fair value liabilities per share from combined Levels 1 and 2 and Level 3 inputs, respectively. NI is net income before
extraordinary items per share. This table provides three sets of test statistics. Column 共A兲 provides t-statistics testing whether the coefficient estimates are different from 0. Column
共B兲 provides F-statistics testing whether the coefficient estimates of each Level of fair value assets are different from 1. Column 共C兲 provides F-statistics testing whether the
coefficient estimates of each Level of fair value liabilities are different from ⫺1. Standard errors are adjusted for two dimensions 共both firms and quarters兲 following Petersen 共2009兲
and Wooldridge 共2002兲.
1389
July 2010
American Accounting Association
Coefficient Comparisons
4.170
0.000
16.540
Downloaded from http://meridian.allenpress.com/accounting-review/article-pdf/85/4/1375/1349625/accr_2010_85_4_1375.pdf by Ronald Tsang on 23 October 2021
Dependent Variable ⴝ Share Price
Value Relevance of FAS No. 157 Fair Value Hierarchy Information
The Accounting Review
TABLE 3
Value Relevance of Fair Values Hierarchy of FAS No. 157
1390
Song, Thomas, and Yi
The Effect of Corporate Governance on the Value Relevance of Fair Values (H2)
Results so far indicate that FAS No. 157 Level information is value-relevant. Given the high
level of information asymmetry between managers and investors, especially relating to reliability
of Level 3 and perhaps Level 2 inputs, H2 tests whether investors place differential weights on fair
values across Levels, based on the firm’s corporate governance mechanisms. Based on prior
literature, we use six measures of corporate governance mechanisms germane to the reliability of
Level 3 fair value estimates. The six governance measures include 共1兲 board independence
共BDIND兲18 measured by the number of independent board members divided by the number of total
board members, 共2兲 audit committee financial expertise 共ACFE兲 19 measured by the number of
audit committee members with financial expertise divided by the number of total audit committee
16
Recall from Table 1 that the number of firms in the sample decreases from 431 in the first quarter to 410 by the third
quarter for tests of H1. We repeat tests using only the 410 firms that have data for all three quarters 共i.e., 1,230
firm-quarter observations兲. We find results nearly identical to those reported in Table 3. Coefficients for FVA1, FVA2,
and FVL12 共0.958, 0.962, and ⫺1.003, respectively兲 are not significantly different from 1 or ⫺1, the coefficient on
FVA3 is less than 1 共0.611兲 and significant at the 0.01 level, and the coefficient on FVL3 is less than ⫺1 共⫺1.983兲 and
significant at the 0.01 level.
17
We also conduct tests by including firm size and capital ratio as separate independent variables in a pooled regression
共untabulated兲. These variables are included in alternate forms 共indicator, deciles, or continuous兲 and either interacted or
non-interacted with main test variables. Under all specifications, we continue to reach the same conclusion; levels
valuations are not driven by differences in firm size or capital ratio. We also consider a growth factor 共i.e., the asset
growth兲 and do not find any evidence that the control for the asset growth alters our conclusions.
18
Dechow et al. 共1996兲 and Beasley 共1996兲 show that firms that are investigated by the SEC had boards with significantly
lower percentages of outside members. Similarly, both Klein 共2002兲 and Peasnell et al. 共2006兲 find a negative relation
between board independence and earnings management. Ajinkya et al. 共2005兲 provide evidence that the percentage of
outside directors on the board is negatively associated with optimistic biases in management forecasts.
19
Section 407 of SOX required the SEC to issue a rule requiring public companies to disclose whether their audit
committee has at least one financial expert. Additionally, if companies do not have financial experts on their audit
committees, then they are required to disclose that fact. Recent studies provide evidence that audit committees with
more financial expertise and activities are less likely to manage earnings and to issue restatements 共Abbott and Parker
2000; Agrawal and Chadha 2005; Bédard et al. 2004兲.
The Accounting Review
American Accounting Association
July 2010
Downloaded from http://meridian.allenpress.com/accounting-review/article-pdf/85/4/1375/1349625/accr_2010_85_4_1375.pdf by Ronald Tsang on 23 October 2021
values, valuation coefficients suggest that investors place less weight on these less reliable
amounts.16
As a robustness test, we consider that results in Table 3 could be confounded by certain bank
characteristics. For example, suppose larger banks tend to report Level 1 and Level 2 fair values,
whereas smaller banks tend to report Level 3 fair values. The differences in value relevance
reported in Table 3 could reflect differences in bank characteristics rather than differences in fair
value levels. For similar reasons, we may also consider that differences in capital ratio are correlated with managers’ choice of valuation levels. If this is the case, and investors value banks’
assets according to bank characteristics, then the results in Table 3 could be confounded.
To examine these issues, we repeat the analyses with subsamples partitioned by these bank
characteristics in Table 4 共i.e., large versus small banks in Panel A and high versus low Tier 1
capital ratio banks in Panel B兲. We find that results for H1 generally hold for all subsamples.
Specifically, we find across subsamples that 共1兲 all of fair value hierarchies are value-relevant, 共2兲
Level 3 asset valuation is less than that of Level 1 or Level 2 assets except for the low capital ratio
subsample, 共3兲 Level 3 liability valuation is more negative than those of Level 1 and Level 2
liabilities, 共4兲 Level 1 and Level 2 asset valuations are not significantly different from 1, 共5兲 Level
3 asset valuation is less than 1, 共6兲 Level 1 and Level 2 liability valuations are not significantly
different from ⫺1, and 共7兲 Level 3 liability valuation is less than ⫺1.17 We also note that the
valuation coefficients for small banks tend to be lower than those for large banks. These results are
consistent with investors perceiving fair value estimates by managers at small banks as being less
reliable.
Independent
Variables
Coeff.
Intercept
NFVA
FVA1
FVA2
FVA3
NFVL
FVL12
FVL3
NI
n
Adj. R2
2.579
0.778
1.007
0.997
0.681
⫺0.798
⫺1.054
⫺2.507
3.016
631
57.32%
Robust
Std. Err.
t-stat
Coeff. ⴝ 0
0.934
0.114
0.141
0.130
0.145
0.125
0.169
0.622
0.828
2.760
6.810
7.150
7.660
4.680
⫺6.390
⫺6.230
⫺4.030
3.640
Small Banks
p-value
Coeff.
0.006***
0.000***
0.000***
0.000***
0.000***
0.000***
0.000***
0.000***
0.000***
2.019
0.770
0.891
0.877
0.568
⫺0.785
⫺1.029
⫺1.623
1.467
629
52.66%
Robust
Std. Err.
t-stat
Coeff. ⴝ 0
p-value
0.706
0.099
0.105
0.094
0.109
0.101
0.093
0.489
0.856
2.860
7.780
8.480
9.340
5.230
⫺7.790
⫺11.010
⫺3.320
1.710
0.004***
0.000***
0.000***
0.000***
0.000***
0.000***
0.000***
0.001***
0.087*
Large Banks
Small Banks
F-stat
p-value
F-stat
p-value
Test of FVA1 ⫽ FVA2
Test of FVA2 ⫽ FVA3
Test of FVA1 ⫽ FVA3
Test of FVL12 ⫽ FVL3
0.010
7.750
7.040
3.990
0.918
0.006***
0.008***
0.046**
0.150
28.240
36.560
1.460
0.703
0.000***
0.000***
0.227**
(continued on next page)
1391
July 2010
American Accounting Association
Coefficient Comparisons
Downloaded from http://meridian.allenpress.com/accounting-review/article-pdf/85/4/1375/1349625/accr_2010_85_4_1375.pdf by Ronald Tsang on 23 October 2021
Panel A: Large versus Small Banks (Dependent Variable ⴝ Share Price)
Large Banks
Value Relevance of FAS No. 157 Fair Value Hierarchy Information
The Accounting Review
TABLE 4
Value Relevance of Fair Values Hierarchy of FAS No. 157 for Large versus Small Banks and for Banks with High versus Low Capital
Ratios
Independent
Variables
Coeff.
Intercept
NFVA
FVA1
FVA2
FVA3
NFVL
FVL12
FVL3
NI
n
Adj. R2
1.742
0.797
0.851
0.905
0.489
⫺0.802
⫺1.031
⫺2.164
3.781
576
53.58%
Robust
Std. Err.
t-stat
Coeff. ⴝ 0
0.908
0.110
0.118
0.127
0.132
0.119
0.119
0.446
1.670
1.920
7.270
7.180
7.120
3.700
⫺6.760
⫺8.690
⫺4.850
2.260
Banks with Low Tier 1 Capital Ratio
p-value
Coeff.
0.056*
0.000***
0.000***
0.000***
0.000***
0.000***
0.000***
0.000***
0.024**
0.966
0.794
1.162
1.031
0.824
⫺0.816
⫺1.108
⫺2.382
2.030
572
62.50%
Robust
Std. Err.
t-stat
Coeff. ⴝ 0
0.898
0.122
0.172
0.133
0.156
0.132
0.173
0.655
0.787
1.080
6.510
6.760
7.740
5.290
⫺6.180
⫺6.420
⫺3.630
2.580
p-value
0.282
0.000***
0.000***
0.000***
0.000***
0.000***
0.000***
0.000***
0.010***
Banks with High Tier
1 Capital Ratio
Banks with Low Tier 1
Capital Ratio
Coefficient Comparisons
F-stat
p-value
F-stat
p-value
Test of FVA1 ⫽ FVA2
Test of FVA2 ⫽ FVA3
Test of FVA1 ⫽ FVA3
Test of FVL12 ⫽ FVL3
3.510
18.600
15.930
6.780
0.061*
0.000***
0.000***
0.009**
1.410
1.480
5.120
2.890
0.235
0.225
0.024**
0.089*
*,**,*** Indicate statistical significance at the 0.10, 0.05, and 0.01 levels 共two-tailed兲, respectively.
July 2010
Song, Thomas, and Yi
Panel A provides the regression results for subsamples partitioned by the asset size of banks. Firms are classified as either large or small banks based on the median value of total
assets in each quarter. See notes for Table 3 for variable definitions and the estimation methodology. The sum of observations 共n ⫽ 631 ⫹ 629 ⫽ 1,260兲 is the same as the total
number of observations reported in Table 3.
Panel B provides the regression results for subsamples partitioned by the Tier 1 capital ratio. Firms are classified to either high or low Tier 1 capital ratio banks based on the median
value of Tier 1 capital ratio in each quarter. See notes for Table 3 for variable definitions and the estimation methodology. Due to missing values of Tier 1 capital ratio in Compustat,
the sum of observations 共n ⫽ 576 ⫹ 572 ⫽ 1,148兲 is lower than the total number of observations 共n ⫽ 1,260兲 reported in Table 3.
Downloaded from http://meridian.allenpress.com/accounting-review/article-pdf/85/4/1375/1349625/accr_2010_85_4_1375.pdf by Ronald Tsang on 23 October 2021
Banks with High Tier 1 Capital Ratio
1392
The Accounting Review
American Accounting Association
Panel B: High versus Low Tier 1 Capital Ratio Banks (Dependent Variable ⴝ Share Price)
Value Relevance of FAS No. 157 Fair Value Hierarchy Information
1393
PRCit = ␣0 + ␣1NFVAit + ␣2FVA1it + ␣3FVA2it + ␣4FVA3it + ␣2aFVA1it ⴱ GOVRANKi
+ ␣3aFVA2it ⴱ GOVRANKi + ␣4aFVA3it ⴱ GOVRANKi + ␣5NFVLit + ␣6FVL12it
+ ␣7FVL3it + ␤1NIit + ␤2GOVRANKi + ␧it
共2兲
Equation 共2兲 is essentially identical to Equation 共1兲 with the exception that all levels of fair
Institutional investors appear to play a monitoring role over corporate financial reporting. Chung et al. 共2002兲 find a
negative association between earnings management and the presence of institutional investors. Bushee 共1998兲 and
Bange and DeBondt 共1998兲 find that the presence of institutional investors with a long-term investment horizon reduces
the level of earnings management through R&D activities. Ajinkya et al. 共2005兲 provide evidence that institutional
ownership has a favorable effect on the likelihood of forecast occurrence as well as the frequency of issuance.
21
Previous studies use a dichotomous variable 共e.g., Big N auditors兲 to measure the quality of the auditor. Recent studies
共Francis and Yu 2009; Choi et al. 2010兲 further show that even within brand name auditors, the size of a specific
engagement office matters for the audit quality and audit fees. Thus, following Francis and Yu 共2009兲, we calculate the
auditor office size of our sample firms based on their 2007 audit fee revenues to proxy for the quality of audit provided
for our sample firms.
22
Ashbaugh-Skaife et al. 共2008兲 and Doyle et al. 共2007兲 document a negative association between disclosed control
weaknesses and earnings quality. Similarly, Feng et al. 共2009兲 show that ineffective internal controls are likely to induce
less accurate management earnings forecasts.
20
The Accounting Review
July 2010
American Accounting Association
Downloaded from http://meridian.allenpress.com/accounting-review/article-pdf/85/4/1375/1349625/accr_2010_85_4_1375.pdf by Ronald Tsang on 23 October 2021
members, 共3兲 the frequency of annual audit committee meetings 共ACMEET兲, 共4兲 total percent of
shares held by institutional investors 共INSTHOLDPCT兲20 calculated from 13-F filings, 共5兲 size of
audit engagement office 共AUDITOFFICESIZE兲,21 and 共6兲 no material control weaknesses problem
共NOMCW兲22 under Sarbanes-Oxley Act 共SOX兲 302 or 404. Prior studies indicate that firms with
independent boards, highly financially literate audit committees, active audit committees, the
presence of institutional investors, audits by auditors from larger offices, and no material control
weaknesses are less likely to engage in financial reporting biases.
Panel A of Table 5 provides the descriptive statistics for the six attributes of corporate governance. The mean value of BDIND is 0.78, reflecting the improved board independence after
SOX. The mean of ACFE is 0.32, indicating that one-third of the audit committee members are
classified as financial experts under SOX. Audit committees in our sample meet eight times
annually. On average, 30 percent of shares of sample firms are held by institutional investors, and
the average log of audit engagement office audit fee revenues is 15.04. Finally, 82 percent of our
sample firms do not report any material control weakness problem 共NOMCW兲 prior to 2007. Panel
B of Table 5 provides the correlation coefficients among corporate governance variables.
To reduce the random measurement error of individual governance variables and to parsimoniously summarize the underlying latent construct of governance quality, we create a standardized
governance score 共GOVSCORE兲 based on the principal-component factor analysis of six aforementioned governance variables. Archival accounting research often uses this methodology to
summarize investor characteristics 共e.g., Bonner et al. 2003兲 or firm characteristics 共e.g., Baik et
al. 2009兲. The factor loadings for the varimax orthogonal rotation are shown in the first column of
Panel C of Table 5, representing how individual governance variables are weighted for 共i.e.,
correlated with兲 GOVSCORE. Consistent with our intuition, all governance variables are positively loaded in generating GOVSCORE. The factor analysis generates a factor with the eigenvalue
of 1.878, which accounts for about 39 percent of the total variations in the original variables. The
second column provides the Kaiser-Meyer-Olkin measure of sampling adequacy 共Kaiser 1974兲.
The mean of this statistic 共0.62兲 is greater than 0.5, indicating that GOVSCORE well captures the
underlying common factor of six individual variables 共Stewart 1981兲. Panel D of Table 5 shows
the distribution of GOVSCORE. Due to this standardization, the mean and the standard deviation
of GOVSCORE are 0 and 1, respectively. Based on GOVSCORE, we create the decile rank
共GOVRANK兲 from 0 to 9, and then scale by 9. To test H2, we estimate Equation 共2兲.
Panel A: Descriptive Statistics
共1兲
共2兲
共3兲
共4兲
共5兲
共6兲
Variable
n
Mean
Std.
Dev.
25th
Percentile
50th
Percentile
75th
Percentile
BDIND
ACFE
ACMEET
INSTHOLDPCT
AUDITOFFICESIZE
NOMCW
405
405
405
405
405
405
0.78
0.32
7.99
0.30
15.04
0.82
0.12
0.21
3.74
0.24
1.86
0.38
0.71
0.20
5.00
0.09
13.76
1.00
0.80
0.25
8.00
0.24
14.55
1.00
0.88
0.33
10.00
0.43
16.26
1.00
Panel B: Correlations
ACFE
ACMEET
BDIND
1
ACFE
⫺0.109
0.029
⫺0.017
0.739
0.031
0.538
⫺0.013
0.792
0.005
0.922
⫺0.048
0.337
1
⫺0.061
0.222
0.094
0.060
1
ACMEET
INSTHOLDPCT
AUDITOFFICESIZE
NOMCW
0.099
0.047
0.246