usefulness of fair values for predicting banks’ future ...€¦ · financial reporting (barth...
TRANSCRIPT
Usefulness of fair values for predicting banks’ future earnings:
Evidence from other comprehensive income and its components
Brian Bratten
Assistant Professor
Von Allmen School of Accountancy, University of Kentucky
Email: [email protected]
Monika Causholli
Associate Professor
Von Allmen School of Accountancy, University of Kentucky
Email: [email protected]
Urooj Khan
Assistant Professor
Graduate School of Business, Columbia University
Email: [email protected]
Tel: (212) 851 5866
Fax: (212) 316 9219
November 2015
Acknowledgements
We thank B.W. Baer, Gauri Bhat, Patricia Dechow (editor), two anonymous referees, Mark
Evans, Fabrizio Ferri, Leslie Hodder, Sharon Katz, Yuri Loktionov, Doron Nissim, Jeff Payne,
Stephen Penman, Robert Ramsay, Shiva Rajgopal, Miguel Duro Rivas, Joshua Ronen, Ethan
Rouen, Gil Sadka, Dan Stone, Abhishek Varma, Mohan Venkatachalam, Dushyantkumar Vyas,
Dave Ziebart, and seminar participants at Columbia Business School, University of Kentucky,
University of Western Ontario, the 2012 American Accounting Association Annual Meeting, the
2012 Annual Congress of the European Accounting Association, the 23rd annual Conference on
Financial Economics and Accounting, and the 2014 PwC Young Scholars Symposium at the
University of Illinois for helpful comments and suggestions.
Usefulness of fair values for predicting banks’ future earnings:
Evidence from other comprehensive income and its components
Abstract
This paper examines whether fair value adjustments included in other comprehensive income
(OCI) predict future bank performance. It also examines whether the reliability of these estimates
affects their predictive value. Using a sample of bank holding companies, we find that fair value
adjustments included in OCI can predict earnings both one and two years ahead. However, not
all fair value-related unrealized gains and losses included in OCI have similar implications.
While net unrealized gains and losses on available-for-sale securities are positively associated
with future earnings, net unrealized gains and losses on derivative contracts classified as cash
flow hedges are negatively associated with future earnings. We also find that reliable
measurement of fair values enhances predictive value. Finally, we show that fair value
adjustments recorded in OCI during the 2007–2009 financial crisis predicted future profitability,
contradicting criticism that fair value accounting forced banks to record excessive downward
adjustments.
JEL classification: G21, M41, M48
Keywords : Earnings; Other Comprehensive Income; Fair value; Predictability.
- 1 -
1. Introduction
We examine whether fair value adjustments embedded in other comprehensive income
(OCI) are useful for predicting future performance in banks. We also examine whether attributes
related to the reliability of fair value adjustments affect their predictive value. Our study is
motivated by the objective of financial reporting as stated in the Financial Accounting Standard
Board’s (FASB) Conceptual Framework, which is to provide decision-useful information that
assists in the prediction of future performance (FASB 2010).
Controversy has surrounded fair value accounting as financial statement users, preparers,
regulators, and others disagree about the extent to which fair values contain information useful to
decision-making. More recently, disagreement about the usefulness of fair values has intensified.
Critics argue that fair value accounting contributed to the 2007–2009 financial crisis by forcing
banks to record unjustified downward adjustments to assets’ fair values, leading to fire sales and
contagion (Bhat et al. 2011; Bowen and Khan 2014; Khan 2015). However, despite the criticism,
the FASB and the International Accounting Standards Board (IASB) believe that the fair value
measurement basis meets the conceptual framework criteria better than other measurement bases,
and both boards have issued standards that increasingly require firms to use fair values in
financial reporting (Barth 2006, 2014).
By definition, fair values are intended to summarize the present value of expected future
cash flows, and proponents of fair value assert that fair values matter for decision-making since
they incorporate current economic conditions and reflect up-to-date expectations (Barth 2014).
On the other hand, critics argue that changes in fair values are transitory and driven by short-
term market movements that have little to do with changes in expectations about future outcomes
(Chisnall 2001). Fair value estimates are also criticized for being more volatile than those based
- 2 -
on historical or amortized cost (Barth et al.1995; Hodder et al. 2006) and being subject to greater
measurement error, which hurts their reliability and predictive value (Landsman 2007).
Given this continuing debate, our research examines the ability of fair value estimates to
predict future performance. Under the mixed-attribute model prescribed by GAAP, some
changes in fair value estimates of assets and liabilities are included in net income, and other
changes are included in OCI. For example, SFAS 115 and SFAS 133 require inclusion in OCI of
unrealized gains and losses on available-for-sale securities and derivative contracts classified as
cash flow hedges. Therefore our first research question investigates whether fair value estimates
embedded in OCI can be used to predict future performance in banks. Our second research
question investigates whether differences in the reliability of these fair value estimates affect
their predictive value.
There are several reasons why we expect fair value estimates included in OCI to predict
future performance. First, fair value estimates are value relevant (e.g., Barth 1994; Petroni and
Wahlen 1995; Barth et al. 1996; Eccher et al. 1996; Nelson 1996; Venkatachalam 1996; Park et
al. 1999; Barth 2006). Second, unrealized fair value gains and losses included in OCI can predict
future performance due to timing of asset sales. For example, a positive association between
unrealized and realized gains can occur when managers sell assets that previously experienced an
increase in fair value (e.g., Park et al. 1999; Evans et al. 2014). Finally, changes in the fair values
of certain derivative instruments (e.g., derivative contracts classified as cash flow hedges) imply
changes in the prices of the underlying hedged items, which can have implications for future firm
performance (Campbell 2015).
To test our predictions, we focus on banks because a large proportion of banks’ assets are
financial assets, many of which are reported at fair value and were among the earliest assets to be
- 3 -
subjected to fair value standards. Also, detailed fair value information for banks’ assets and
liabilities is available through their regulatory filings, allowing us to construct a panel of U.S.
bank holding companies (“banks”), both publicly traded and privately held, from 2001 through
2013.
In our tests, we focus on the relation between contemporaneous total OCI, OCI
components, and one- and two-year-ahead bank performance measures. We measure future
performance using two variants of future earnings: (i) pre-tax earnings and (ii) pre-tax earnings
before the provision for loan and lease losses. Our second measure of earnings excludes the
largest accrual in banks and therefore mitigates managerial discretion, resulting in a less biased
measure of performance (Beatty et al 2002).1 We focus on earnings as a measure of performance
rather than future cash flows because earnings is accounting’s summary measure of performance
(Dechow 1994) and is used for a wide range of purposes, including in executive compensation
plans, debt contracts, and initial public offering prospectuses. Moreover, earnings better predict
future cash flows than do current cash flows (Dechow et al. 1998).
To address our second question, we partition our sample along two dimensions: (i)
market-wide liquidity and (ii) the proportion of investment securities guaranteed by the U.S.
government or its agencies. We expect the predictive ability of fair value estimates to be
enhanced when market liquidity and the proportion of investment securities guaranteed by the
U.S. government or its agencies is high, conditions that should result in more reliable fair values.
In additional analyses, we investigate the criticism that fair value accounting forced banks to
record excessive downward adjustments during the 2007–2009 financial crisis by testing the
1 Unlike for nonfinancial firms, estimation of cash flows is problematic in financial firms. As a result, prior studies
have used earnings before taxes and provision for loan losses as a proxy for banks’ cash flows (e.g., Wahlen 1994;
Liu et al. 1997; Kanagaretnam et al. 2014; Altamuro and Beatty 2010; Bischof et al. 2012). In additional tests, we
also perform tests using residual earnings as an alternative measure of performance.
- 4 -
extent to which fair value adjustments recorded in OCI during the crisis predicted one- and two-
year-ahead performance of banks.
Our findings can be summarized as follows. After controlling for current period net
income, fair value-oriented OCI is incrementally associated with one- and two-year-ahead bank
performance. Decomposing OCI into its components, we find that the net unrealized gains and
losses from available-for-sale securities are positively and significantly associated with future
earnings, while the net unrealized gains and losses from derivatives are negatively related to
future earnings.
Analyses that distinguish between the reliability of fair values provide two additional
insights. First, we find that each dollar of OCI translates into a higher amount of earnings in the
future during periods of higher market-wide liquidity. Second, we find that OCI predicts bank
earnings only for those banks that hold a higher proportion of securities guaranteed by the U.S.
government or its agencies in their investment portfolio. In summary, these findings suggest that
the reliability with which fair values are measured affects their ability to predict future
performance. When we restrict our analysis to the financial crisis years, we continue to find an
association between fair value adjustments and future earnings, suggesting that fair value
adjustments during the crisis predicted future performance. This evidence contradicts claims that
declines in assets’ fair values during the crisis were unrelated to the deterioration in the
underlying fundamentals. Finally, our results are robust to a host of other tests, including
alternative measures of performance, additional controls for banks’ business models, and the
inclusion of bank-fixed effects.
Our study makes several contributions. First, it informs the ongoing debate about the
merits of a fair value accounting-based reporting system. This debate intensified during the
- 5 -
financial crisis, leading some critics to charge fair value accounting with exacerbating the crisis
because falling market prices required banks to excessively mark down asset values. Many
critics, especially bankers, argued that price movements during the crisis were only temporary
and were not consistent with the actual economic value of the assets (ABA 2008). Several of
these claims are unsubstantiated with empirical evidence. We show that, contrary to this
criticism, fair value adjustments included in OCI during the crisis predicted future bank
performance.
Second, our results complement value relevance research—which has documented an
incremental association between share prices and fair value disclosures related to investment
securities and derivative contracts—by providing direct evidence on whether fair values improve
the prediction of firm performance. Furthermore, our study directly informs on a feature of
accounting estimates desired by standard setters: the ability to predict future outcomes.
Third, we extend the literature on the predictive value of OCI in three important ways.
There is limited and mixed evidence in the literature regarding the predictive ability of OCI for
operating performance. For example, some studies have found that OCI and its components are
value relevant (Barth 1994; Kanagaretnam et al. 2009; Campbell 2015), while others find that
they are not (Dhaliwal et al. 1999) or that they are value relevant but have limited predictive
value (Jones and Smith 2011). By showing that different OCI components have differing
implications for future performance, our results suggest the importance of considering OCI
components individually when making predictions about future performance. Also, departing
from most of these studies, we rely on a sample comprised exclusively of banks for our analyses.
Given the extensive use of fair value estimates in financial reporting for banks, we believe banks
- 6 -
are best suited to investigate the properties of fair value of estimates.2 Finally, in contrast to
studies that have examined the association between fair value adjustments and future
performance of only one type of financial instrument (e.g., Park et al. 1999; Evans et al. 2014;
Campbell 2015), we use a more comprehensive measure of fair values adjustments.
Our evidence is also relevant to U.S. bank regulators, who have recently adopted the
Basel III Capital Framework recommendations and amended the regulatory capital calculations
to include components of accumulated other comprehensive income (FDIC 2013).3 Critics have
argued that unrealized gains and losses in accumulated other comprehensive income should not
be included in regulatory capital, as these changes are temporary and driven by movements in
interest rates as opposed to changes in credit risks (ABA 2012). Our evidence contradicts this
criticism by showing that certain unrealized gains and losses included in OCI relate to future
bank performance.
The remainder of the paper is organized as follows. Section 2 reviews the literature,
provides institutional background, and presents the hypotheses. Section 3 describes our research
design. Section 4 discusses our sample and the results of our analyses. Section 5 concludes.
2. Literature review and hypotheses development
2.1. Background
An objective of the Conceptual Framework is to provide decision-useful information that
would assist in the assessments of the amounts, timing, and uncertainty of future cash flows and
the prediction of returns on economic resources (FASB 2010). Standard setters such as the FASB
2 For example, Jones and Smith (2011) only have 26 financial firms in their sample, and in this mixed sample, they
find that unrealized gains and losses on available-for-sale securities do not predict future cash flows (p. 2066). 3 Before the adoption of Basel III recommendations, only unrealized gains and losses on equity securities classified
as available-for-sale and foreign currency translation adjustments were included in the calculation of regulatory
capital.
- 7 -
in the United States believe that incorporating more current information in the financial reports is
a means to achieve this objective. As a result, financial reporting rules in the United States have
become increasingly fair value-based as the FASB has issued several standards requiring
recognition and disclosure of fair values; yet, as discussed in more detail below, regulators,
academics, and practitioners have continued to debate the merits of fair values.4
In theory, fair values represent the most current expectations and changes in expectations
about future performance. Therefore, to the extent that fair values can be measured reliably, fair
value estimates should be useful in predicting future performance (Barth 2000, 2007, 2014; CFA
Institute 2005). However, fair value estimates are more volatile than those determined by other
measurement bases, and changes in fair values are often driven by short-term market movements
that reverse over time. This suggests that fair value estimates are measured with less reliability
and are not likely to predict future outcomes.
Barth (2006) notes that incorporating estimates of the future into the financial statements
affects the nature of reported income and the predictability of future income. Under a full fair
value measurement system, all changes in the fair values of net assets would be incorporated in
net income. This measurement system would result in a comprehensive measure of income—
one that includes all changes in net assets—and would likely result in future income that is more
difficult to predict.5 However, in the current mixed-attribute model, changes in the fair values of
only some assets and liabilities are recognized in earnings, some fair value changes are ignored,
and other values affect net assets and comprehensive income but not contemporaneous net
4 Fair value is defined as “the price that would be received to sell an asset or paid to transfer a liability in an orderly
transaction between market participants at the measurement date” (FASB 2006). 5 Under a full fair value reporting system, earnings would be equal to changes in net assets and thus would be
uninformative about future income because changes in fair value follow a random walk and are unpredictable
(Storey and Storey, 1998; Schipper and Vincent 2003; Nissim and Penman 2008). Comprehensive income is not a
full fair value measure, as it does not include fair value changes in instruments such as held-to-maturity securities,
loans, financial liabilities, and nonterm deposits (Hodder et al. 2006).
- 8 -
income. Currently, four types of gains and losses are included in OCI: unrealized gains and
losses related to changes in the fair values of investment securities classified as available for sale
(per SFAS 115), unrealized gains and losses related to changes in the fair values of derivative
contracts classified as cash flow hedges (per SFAS 133), foreign currency translation
adjustments (per SFAS 52), and additional minimum pension liability adjustments (per SFAS
87). Thus, under the mixed-attribute model, many of the fair value adjustments for a typical bank
bypass net income and are included in OCI. Therefore, to investigate whether a fair value-based
accounting system better meets the objectives of financial reporting specified in the Conceptual
Framework, we examine whether fair value-based aggregate OCI and its individual components
help predict future performance in banks.
2.2. The predictive ability of fair value information in OCI
The fair-value-related adjustments included in OCI can predict future performance for
several reasons. Ohlson (1999) shows analytically that, while changes of fair values follow a
random walk and do not predict future fair value changes, they could matter for predicting future
performance. For example, to the extent that unrealized gains and losses accumulate over time
before the asset is sold, as is the case under the current mixed-attribute model, the amount of the
unrealized gains and losses can be associated with future firm performance. A few studies that
focus on specific assets find evidence that fair value adjustments related to those assets are
associated with future cash flows and earnings generated specifically from those assets. For
example, using a sample of U.K. firms, Aboody et al. (1999) show that firms’ upward
revaluations of fixed assets relate significantly to changes in future operating income and future
cash from operations. Park et al. (1999) and Evans et al. (2014) find that accumulated fair value
adjustments for investment securities are associated with future income realized from these
- 9 -
instruments. Similarly, Petroni and Wahlen (1997) show that bond investment fair values are
positively associated with future reported interest income on those investments.
Changes in the fair values of certain derivative contracts can also be informative about
future profitability. Currently, under SFAS 133, firms must record derivative contracts classified
as cash flow hedges at fair value on the balance sheet on a recurring basis, and the related
unrealized gains and losses are included in OCI. When the hedged transaction occurs, the
unrealized gains/losses are reclassified into earnings, and the underlying hedged item also affects
earnings. An unrealized gain (loss) on a derivative contract classified as a cash flow hedge in a
given period suggests that the price of the underlying hedged item has moved in an unfavorable
(favorable) direction and will result in lower (higher) future profitability after the expiration of
the hedge.6 In a sample of nonfinancial firms, Makar et al. (2013) and Campbell (2015) report
that unrealized gains and losses on derivative contracts classified a cash flow hedges are
inversely associated with future profitability and cash flows.
The value relevance literature also documents that fair value adjustments—many of
which are included in OCI—are incrementally relevant to amortized cost estimates, offering
some indirect evidence that equity investors believe fair values matter for the prediction of future
performance.7
Without focusing specifically on the predictive ability of fair values, a related literature
has examined the usefulness of OCI and its components to predict performance. This literature
has largely relied on two approaches to evaluate the usefulness of OCI: (i) examination of the
association between OCI and stock prices and returns or (ii) direct examination of the ability of
6 Firms typically hedge on a rolling basis. At the expiration of the hedge, the firm can buy new hedges. However,
new hedges will only protect the firm against future price changes, not the current price change. 7 See, for example, Barth (1994), Petroni and Wahlen (1995), Barth et al. (1996), Eccher et al. (1996), Nelson
(1996), Venkatachalam (1996), and Park et al. (1999).
- 10 -
OCI to predict earnings and cash flows. A few studies find that OCI and its components are
value relevant (Barth 1994; Kanagaretnam et al. 2009), while others find contrary evidence
(Dhaliwal et al. 1999). Dhaliwal et al. (1999) find that net income is more strongly associated
with returns than comprehensive income. They fail to find any evidence of comprehensive
income being more strongly associated with future earnings or cash flows. In contrast,
Kanagaretnam et al. (2009) report that comprehensive income is more strongly associated with
one-year-ahead cash flows than net income. However, Kanagaretnam et al. (2009) also find that
net income better predicts future net income relative to comprehensive income and conclude that
components of OCI are poor predictors of profitability due to their transitory nature. It is difficult
to draw definite conclusions based on these studies because the differences in the results could
also be due to differences in their methodology and data (Chambers et al. 2007).
Using a sample of nonbanks, Jones and Smith (2011) find that OCI has predictive value
for one-year-ahead earnings but not for longer horizon earnings. Also, OCI predicts cash flows in
some settings but not consistently. We extend their study and this literature by examining the
predictive ability of OCI and its components in banks. Based on the arguments above, we expect
that fair value estimates included in OCI predict future performance and state our first hypothesis
in the alternative form:
Hypothesis 1: Fair value estimates embedded in OCI are predictive of future performance
2.3. The reliability of fair value estimates
The Conceptual Framework lists reliability as one of the two fundamental attributes of
financial information, and many critics have expressed concern that fair values lack reliability
(Barth 2014).8 Landsman (2007) asserts that measurement error and management bias embedded
8 We use the more familiar term “reliability” to capture the construct currently referred to in the Conceptual
Framework as “representational faithfulness”.
- 11 -
in fair values can inhibit their ability to predict operating performance. Similarly, Barth (2007)
suggests that one of the concerns relating to fair value estimates is the effect of management
incentives, especially on the reliability of estimates for which input prices are not observable.
Prior value relevance studies provide indirect evidence that fair value estimates are less likely to
predict future performance (as captured in share prices) in the presence of measurement error and
bias. While Barth et al. (1996) find that fair value estimates of loans and long-term debt explain
bank share prices beyond related book values, other studies have shown that the fair values of
loans contain substantial measurement error and bias, which in turn reduces their value relevance
(e.g., Nelson 1996; Eccher et al. 1996; Park et al. 1999; Beaver and Venkatachalam 2003;
Nissim 2003). In the same spirit, a few studies investigate the value relevance of fair value
estimates based on the source of information used to calculate fair values as mandated by SFAS
157 (Kolev 2008; Goh et al. 2009; Song et al. 2010). They find that the value relevance of level 1
and level 2 fair value estimates (directly or indirectly observable inputs) is greater than that of
level 3 estimates (unobservable firm-generated inputs). In summary, prior research finds that fair
value estimates are less value relevant when they cannot be measured reliably, implying that
equityholders believe fair value estimates’ predictive value is reduced in such circumstances.
We examine whether the ability of fair value adjustments embedded in OCI to predict
performance varies with the reliability with which fair values of the underlying assets can be
measured. We identify two factors that might affect the reliability of fair values estimates:
market-wide liquidity and the proportion of the bank’s investment securities guaranteed by the
U.S. government or its agencies.
Critics of fair value accounting have expressed concerns that fair value estimates are less
relevant, less reliable, or both during periods of market illiquidity. When asset markets are
- 12 -
illiquid, volume and level of activity in asset markets significantly decreases and asset prices
reflect the amount of liquidity available in the market rather than the future earnings power of
assets (Allen and Carletti 2008). Thus we expect that fair value estimates embedded in OCI are
likely to have lower predictive value during periods of higher market illiquidity.
Fair value adjustments of securities issued or guaranteed by the U.S. government and its
agencies (agency securities) are likely to be more reliable because agency securities trade in
larger and more liquid markets compared to non-agency securities. Consistent with this
conjecture, unrealized gains and losses of investment securities are more strongly associated with
stock prices for banks with a higher proportion of agency securities (Barth 1994). Also, Bhat et
al. (2011) report that agency securities are subject to significantly lower uncertainty with regard
to default during financial crises, leading to lower price variability. Based on these arguments,
we expect that banks that hold a greater amount of securities issued or guaranteed by the U.S.
government will have a larger percentage of their securities measured more reliably, which in
turn will affect the reliability of fair value adjustments related to these assets included in OCI.
To summarize, we expect that the predictive value of fair value estimates included in OCI
is affected by the reliability with which fair values can be estimated. We state below our
expectation in the alternative form:
Hypothesis 2: The reliability of fair value estimates embedded in OCI enhances their ability to
predict future performance.
3. Research Design and Sample
3.1. The predictive ability of fair value information
Our sample is comprised of all banks, both publicly traded and privately held, that have
FR Y-9C report data available on the Bank Holding Companies Database maintained by the
- 13 -
Federal Reserve Bank of Chicago during 2001–2013. The Bank Holding Companies (BHC)
Database collects financial data included in FR Y-9C reports filed by BHCs. The FR Y-9C
reports contain information from the balance sheet and income statement and risk-based capital
measures, as well as other reporting schedules. The FR Y-9C report is filed by all BHCs with
total consolidated assets of $500 million or more. In addition, BHCs meeting certain criteria may
be required to file this report. We use 2013 data only to calculate our dependent variable, which
is measured in periods t+1 or t+2.
We combine the FR Y-9C data with data for the components of OCI (which are not
reported in the FR Y-9C) from SNL Financial, which is available for the same period but for
only approximately 44% of the observations in our full sample. In our tests, we compare the
incremental ability of OCI to predict future earnings after controlling for contemporaneous net
income. Compared to net income, OCI is more fair value-oriented because it includes several fair
value adjustments that bypass the income statement but are included in shareholders’ equity
directly (e.g., net unrealized fair value gains and losses on available-for-sale securities and net
unrealized gains and losses on derivative contracts classified as cash flow hedges). If the fair
value information included in these OCI items aids in the prediction of future performance, we
expect that OCI and its components individually will be incrementally predictive of future
performance measures after controlling for current earnings.
3.2. The reliability of fair value estimates
To test the second hypothesis, we partition our sample based on (i) market-wide liquidity,
and (ii) the proportion of investment securities guaranteed by the U.S. government or its
agencies.
- 14 -
Our proxy for market-wide liquidity is based on Amihud’s (2002) measure of price
response associated with each dollar of trading volume. Specifically, for all stocks listed on
NYSE with positive daily volume, we calculate the ratio of daily absolute return to trading
volume. Next, after eliminating the top and bottom 1 percent observations to remove outliers, we
calculate a market-wide annual liquidity measure, Liquid_Annual, as the daily market-cap-
weighted average of all daily measures. Higher values of Liquid_Annual represent greater
market-wide illiquidity. We partition our sample period into high liquidity years and low
liquidity years based on the median of Liquid_Annual. The years in which Liquid_Annual is
above (below) the median are classified as low liquidity years (high liquidity years).9 If fair value
estimates are less reliable during periods of low market-wide liquidity, we expect to find that fair
value estimates are less predictive of future performance in the low liquidity sample years.
Next, we partition our sample banks based on their holdings of securities issued or
guaranteed by the U.S. government and its agencies. The proportion of U.S. guaranteed
securities is measured as the sum of the amortized cost of investment securities issued or
guaranteed by the U.S. government or its agencies divided by the amortized cost of total
investment securities held by a bank. We run our analyses separately for the high and low public
bank subsamples based on the median proportion of U.S. guaranteed securities.10
9 Arguably, Liquid_Annual can be criticized for only capturing the liquidity of equity markets, as it is estimated
using data of stocks listed on the NYSE, whereas banks trade assets and financial instruments in markets other than
the equity markets. Chordia et al. (2005) document that liquidity co-varies across asset markets with the shocks to
spreads in one market increasing the spreads in the other markets. Thus, we expect that equity market liquidity co-
varies with liquidity in other asset markets. 10 Banks whose holdings of agency securities as a proportion of their total investment securities are above (below)
the sample median are assigned to the high (low) subsample. Only banks with assets in excess of $1 billion are
required to provide the breakdown of their agency securities. This severely limits the data available for this test for
privately held banks, so we perform this analysis only for publicly traded banks.
- 15 -
3.3. Empirical models
We test for the association between OCI and our two measures of earnings as proxies of
bank performance by following the approach of Altamuro and Beatty (2010) and Kanagaretnam
et al. (2014). Specifically, we augment their models by including OCI and its components and
estimate the following models:
Pre-tax ROAt+1 (t+2) = α +β1Pre-tax ROAt+ β2Log(Assetst) +
β3[Pre-tax ROAt* Log(Assets t)] + β4PtOCI t + fixed-year effects + εt+1 (t+2); (1)
Pre-tax EBPt+1 (t+2) = α +β1Pre-tax ROAt+ β2Log(Assetst) +
β3[Pre-tax ROAt* Log(Assetst)] + β4PtOCI t + fixed-year effecs + εt+1(t+2); (2)
where PtOCI, our variable of interest, is pre-tax other comprehensive income, calculated as the
reported after-tax other comprehensive income (BHCKB511) divided by one minus the
maximum statutory corporate tax rate (35% for all years in our sample), scaled by lagged total
assets.11 Comprehensive income is a more fair value-based reported income number than net
income because several fair value adjustments are included in OCI but not included in net
income. Thus the coefficient on PtOCI captures the incremental value of OCI in predicting
future earnings after controlling for current pre-tax earnings. Following Wahlen (1994), Liu et
al. (1997), Kanagaretnam et al. (2014), Altamuro and Beatty (2010), and Bischof et al. (2012),
Pre-tax ROA is income before taxes (BHCK4301) divided by lagged total assets, whereas pre-tax
earnings before the provision for loan and lease losses, Pre-tax EBP, is calculated as income
before taxes plus the provision for loan losses (BHCK4230), scaled by lagged total assets.12 Pre-
tax ROA and Pre-tax EBP are measured one and two years ahead, denoted by subscripts t+1 and
t+2, respectively. Log(Assets) is the natural log of total assets (BHCK2170). We measure our
11 BHCK item number mnemonics reported in parentheses indicate data items taken from Federal Reserve Board’s
bank holding company database, representing financial data reported on form FR-Y9C. All variables are also
defined in the Appendix. 12 We estimate our measures of performance on a pre-tax basis to avoid confounding effects of tax avoidance on the
relation between fair value adjustments and future operating performance.
- 16 -
independent variables at time t to investigate whether current fair value adjustments predict
performance one year and two years in the future. Standard errors are clustered by bank.
Empirical models (3) and (4) below examine the association between individual OCI
components and future performance. We decompose OCI into three components representing the
ratio of i) net unrealized gains and losses related to changes in the fair values of investment
securities classified as available for sale (PtOCI-AFS), ii) net unrealized gains and losses related
to changes in the fair values of derivative contracts classified as cash flow hedges (PtOCI-
Derivatives), and iii) all other adjustments included in OCI (e.g., foreign currency translation
adjustments and additional minimum pension liability adjustments) (PtOCI-Other), to total
assets.
Pre-tax ROAt+1 (t+2) = α +β1Pre-tax ROAt+ β2Log(Assetst) +
β3[Pre-tax ROAt* Log(Assets t)] + β4PtOCI-AFS t + β5PtOCI-Derivativest +
β6PtOCI-Othert + fixed-year effects + εt+1(t+2); (3)
Pre-tax EBPt+1 (t+2) = α +β1Pre-tax ROAt+ β2Log(Assetst) +
β3[Pre-tax ROAt* Log(Assetst)] + β4PtOCI-AFS t + β5PtOCI-Derivativest +
β6PtOCI-Othert + fixed-year effects + εt+1(t+2). (4)
4. Results
4.1. Descriptive Statistics
Table 1 presents the descriptive statistics for our samples.13 The average (median) bank
has assets of $5.66 ($0.688) billion. Pre-tax ROA is approximately 1.11%, whereas Pre-tax EBP
are about 1.57% of total assets on average. The mean (median) pre-tax other comprehensive
income, PtOCI, is 0.02% (0.00%), suggesting that, on average, fair value adjustments that are
excluded from net income but included in comprehensive income are income increasing.14 In
13 Continuous variables have been winsorized at 1% and 99% to reduce the influence of outliers. 14 Untabulated correlations reveal that pre-tax other comprehensive income (PtOCIt) is positively and significantly
correlated with Pre-tax EBPt+1 and Pre-tax ROAt+1, providing univariate evidence that fair value estimates included
in other comprehensive income predict future earnings.
- 17 -
terms of OCI components, the largest is unrealized gains and losses from available-for-sale
securities, PtOCI – AFS, with a mean (median) of 0.04% (0.02%).
4.2. Fair value accounting and predictability of pre-tax earnings
In this section, we report the results of tests that examine the association between OCI
and future performance measures. The results of equation (1), which estimates the predictive
value of PtOCI for future pre-tax ROA, are shown in Table 2, Panel A, whereas those of equation
(3), which estimates the predictive value of OCI components, are shown in Panel B. The primary
coefficients of interest for hypothesis 1 are PtOCIt in Panel A and PtOCI-AFSt and PtOCI-
Derivativest in Panel B. Results are presented for the prediction of one-year-ahead (left column)
and two-year-ahead (right column) Pre-tax ROA. The regression models include year fixed
effects, and t statistics are based on standard errors clustered by bank. In all of our analyses, the
statistical significance of the coefficients is based on one-tailed p-values for variables for which
we have a predicted sign and two-tailed p-values otherwise. The adjusted R-squared of our
model ranges between 39% (for models using future pre-tax ROA in year t+2) and 54% (for
models using future pre-tax ROA in year t+1), which is comparable to prior research (e.g.,
Altamuro and Beatty 2010). In Panel A, we find that current-year earnings predict future
earnings (the coefficients on Pre-tax ROAt are positive and statistically significant) and larger
banks have relatively higher future earnings (the coefficients on Log(Assetst) are positive and
statistically significant). More importantly, consistent with hypothesis 1, we find that pre-tax
OCI is incrementally associated with both one- and two-year-ahead pre-tax earnings (the
coefficients on PtOCI are positive and statistically significant), suggesting that fair value-
oriented OCI predicts future bank performance.15
15 In untabulated separate regressions of public (private) banks only, the coefficient on PtOCI is 0.1323 (0.1084),
when Pre-tax ROAt+1 is the dependent variable, and 0.0858 (0.1084), when Pre-tax ROAt+2 is the dependent
- 18 -
In Panel B, the results show that net unrealized gains and losses related to available-for-
sale securities are positively related to both one- and two-year-ahead pre-tax earnings (the
coefficients on PtOCI-AFS are positive and statistically significant), whereas net unrealized
gains and losses for derivative contracts classified as cash flow hedges are negatively related to
future earnings (the coefficients on PtOCI-Derivatives are negative and statistically significant).
These coefficients are statistically different from each other in both one- and two-year-ahead
regressions. Finally, while we do not have a prediction about the sign of the coefficient for
PtOCI-Other, the coefficient on net unrealized gains and losses from other OCI components is
negatively and significantly associated with one-year-ahead earnings but not significantly related
to two-year-ahead earnings.16 In conclusion, the evidence in Panel B suggests that unrealized
gains and losses on individual components of OCI have different implications for future
performance of banks.
4.3. Fair value accounting and predictability of pre-tax earnings before provision
In this section, we report the results of tests that examine the association between OCI
and future Pre-tax EBP. The results of equation (2), which estimates the predictive value of
aggregate OCI for future Pre-tax EBP, are shown in Panel A, whereas those of equation (4),
which estimates the predictive value of OCI components, are shown in Panel B of Table 3. As
before, the primary coefficient of interest for hypothesis 1 is on PtOCIt. The adjusted R-squared
variable. However, when we estimate our regressions in a sample comprising both public and private banks and
include an interaction between PtOCI and an indicator variable for public banks, the coefficient on the interaction is
insignificant for both dependent variables. Thus we do not separate public and private banks in our subsequent tests
and tabulate results only for the combined sample. 16 One year ahead, the coefficient on PtOCI-Other is statistically different from PtOCI-AFS (PtOCI-Derivatives) at
the 1% (10%) level. Two years ahead, the coefficient on PtOCI-Other is statistically different from PtOCI-AFS
(PtOCI-Derivatives) at the 10% (1%) level. If we replace PtOCI-Other with the other two OCI components for
which data is available (i.e., foreign currency translation adjustments and additional minimum pension adjustments),
results for PtOCI-AFS and PtOCI-Derivatives are qualitatively similar to those reported in Table 2. The coefficients
on both of the other OCI components are negative and significant for one-year-ahead Pre-tax ROA and statistically
indistinguishable from zero for two-year-ahead Pre-tax ROA.
- 19 -
of our model ranges from 34% (for models using future pre-tax EBP in year t+2) to 46% (for
models using future pre-tax EBP in year t+1), which is comparable to prior research (e.g.,
Altamuro and Beatty 2010). Similar to the evidence in Table 2, we find that current year earnings
predict future Pre-tax EBP (the coefficients on Pre-tax ROA are positive and statistically
significant) and larger banks have higher Pre-tax EBP (the coefficients on Log(Assetst) are
positive and statistically significant). More importantly, consistent with hypothesis 1, in Panel A,
we find that pre-tax OCI is incrementally associated with one- and two-year-ahead Pre-tax EBP
after controlling for contemporaneous pre-tax earnings (the coefficients on PtOCI are positive
and statistically significant). In Panel B, we find that net unrealized gains and losses related to
available-for-sale securities are positively related to both one- and two-year-ahead Pre-tax EBP
(the coefficients on PtOCI-AFS are positive and statistically significant), whereas the coefficient
on the net unrealized gains and losses for derivative contracts classified as cash flow hedges
(PtOCI-Derivatives) is negatively associated with one- and two-year-ahead Pre-tax EBP. While
the coefficient on PtOCI-Derivatives is statistically significant in the two-year-ahead regressions,
its significance is only marginal in the one-year-ahead regression. As before, the coefficients on
PtOCI-AFS and PtOCI-Derivatives are statistically different from each other in both the one- and
two-year-ahead regressions. The net unrealized gains and losses related to other components are
not significantly related to Pre-tax EBP (the coefficients on PtOCI-Other are negative but not
statistically significant).17
17 If we replace PtOCI-Other with the other two OCI components for which data is available (i.e., foreign currency
translation adjustments and additional minimum pension adjustments), results for PtOCI-AFS and PtOCI-
Derivatives are qualitatively similar to those reported in Table 3. The coefficient on the foreign currency translation
adjustment component is negative and significant (insignificant) for one-year-ahead (two-year-ahead) Pre-tax EBP,
and the coefficients on the pension adjustment are insignificant for both one- and two-year-ahead Pre-tax EBP.
- 20 -
In summary, we find robust evidence in support of H1 in Tables 2 and 3. Our evidence
suggests OCI predicts future bank earnings after controlling for contemporaneous earnings. In
addition, OCI components have different predictive implications for future earnings.
4. 4. Reliability of fair value estimates and their predictive ability
In this section, we report the results of the tests of our second hypothesis, which
examines the impact of reliability of fair value estimates on the relation between fair value
adjustments and future earnings.
4.4.1. Market-wide liquidity and the predictive ability of fair value estimates
To examine the impact of market-wide liquidity on the predictive ability of fair value
adjustments with respect to future performance, we estimate equation (1) separately for high
liquidity years (i.e., Liquid_Annual is below the sample period median) and low liquidity years
(i.e., Liquid_Annual is above the sample period median). The results from regressions where
Pre-tax ROA and Pre-tax EBP are measured one year (two years) ahead are presented in Panel A
(B) of Table 4. We continue to find that current year pre-tax earnings are associated with future
performance (for both Pre-tax ROA and Pre-tax EBP) during high as well as low liquidity
periods. With respect to H2, we find that PtOCI has a stronger effect on both future Pre-tax ROA
and Pre-tax EBP measured one and two years ahead during periods of high market-wide
liquidity. While the coefficient on PtOCI is positive and significant for both high and low
liquidity sample periods, the coefficient on PtOCI is almost twice as large during periods of high
market-wide liquidity relative to periods of low market-wide liquidity.18 This evidence is
consistent with H2.
18 In untabulated tests, we combine the high and low liquidity subsamples and interact PtOCI with an indicator for
high liquidity. We predict that the coefficient on this interaction will be positive and significant. Consistent with our
prediction, we find that the coefficient on the interaction of PtOCI with the indicator variable for high liquidity is
positive and statistically significant for each dependent variable.
- 21 -
4.4.2. Proportion of U.S. agency securities and the predictive ability of fair value estimates
In this section, to test the influence of reliability of fair value estimates on the predictive
ability of fair value adjustments, we partition the subsample of publicly traded banks in our main
sample based on their holdings of agency securities. As mentioned above, only banks with assets
in excess of $1 billion must report the details necessary to determine the extent to which their
investment securities are agency securities. Since a large majority of privately held banks have
less than $1 billion of assets, we cannot reliably estimate the proportion of agency securities in
their investment securities portfolio. Hence we run this analysis only for publicly traded banks.
Banks whose proportion of agency securities in their investment securities portfolio is above
(below) the sample median are classified as “High (Low) Percent of U.S. Guaranteed
Investments” banks. The results from regressions where Pre-tax ROA and Pre-tax EBP are
measured one year (two years) ahead are presented in Panel A (B) of Table 5. We find that the
coefficient on PtOCI is positive and significant in both one- and two-year-ahead regressions
when future performance is measured as Pre-tax ROA only for the subsample of banks with a
higher proportion of U.S. guaranteed securities, suggesting that fair values embedded in OCI
predict next-year bank performance only for banks whose fair value estimates can be measured
more reliably. In regressions where future performance is measured as Pre-tax EBP, the
coefficient on PtOCI is positive and marginally significant only for banks with higher holdings
of agency securities. This provides additional evidence in support of H2. Taken together, the
results reported in Tables 4 and 5 suggest that the predictive ability of fair value estimates for
future performance is enhanced when the fair values can be measured more reliably.
4.5. Additional Tests
- 22 -
During the 2007–2009 financial crisis many criticized fair value accounting because it
forced banks to record downward adjustments to asset values that were not justified by the
deterioration in the underlying fundamentals of the assets. Critics argued that, while many bank
assets were still performing during the financial crisis, fair value accounting required that the
assets be reported at value below their “true value” as some asset markets had become illiquid
(e.g., ABA 2008; Wesbury 2008). Banks argued that these impairments on their investment
securities, many of which were classified as available-for-sale, were temporary, and,
accordingly, they recorded the fair value adjustments in OCI. If the fair value adjustments
recorded during the financial crisis were indeed excessive and temporary, we expect that they
will not predictfuture bank performance. Accordingly, we examine the ability of fair value
adjustments included in OCI to predict future performance during the 2007–2009 financial crisis.
We re-estimate models (1) and (2) for a subsample spanning the years 2007 to 2009.
Laux and Leuz (2010) argue that fair value adjustments recorded during the crisis
reflected the actual underlying economic values of the assets. Our test seeks to provide the first
empirical evidence that complements their conclusions. The results for this analysis are
presented in Table 6. In Panel A (B), future performance is measured by one-year-ahead (two-
year-ahead) Pre-tax ROA and Pre-tax EBP. In all regressions the coefficient on PtOCI is positive
and statistically significant.
In untabulated analysis, we replace PtOCI with its components and re-estimate models
(3) and (4) for the years of the financial crisis. We find that during the crisis, unrealized gains
and losses on available-for-sale securities are positively associated with one- and two-year-ahead
bank earnings, while the unrealized gains and losses from derivative contracts classified as cash
flow hedges are negatively associated with future bank performance. Thus our results suggest
- 23 -
that fair value adjustments recorded in the OCI during the recent crisis were informative about
future bank profitability and consistent with the general conclusions of Laux and Leuz (2010).
These results contradict the claim that, on average, fair value accounting resulted in the recording
of unjustified downward adjustments to asset values.
Additionally, we examine whether the results are similar depending on whether total OCI
is positive or negative (i.e., whether it reflects net increases or net decreases in fair value). To do
so, we divide our sample into two subsamples where OCI is positive versus negative in year t
and then we estimate, in each subsample, equations (1) and (2) examining the association
between PtOCI and each of our four earnings proxies (one- and two-year-ahead pre-tax earnings
and one and two-year-ahead pre-tax earnings before the provision). In total, we estimate eight
separate regressions—four regressions for the positive OCI observations and four for the
negative OCI observations. In these untabulated analyses, we find that the coefficient on PtOCI
is positive and significant in six of the eight regressions. The two exceptions are when Pre-tax
ROAt+2 is the dependent variable in the negative OCI sample and when Pre-tax EBPt+1 is the
dependent variable in the positive OCI sample. In these two cases, the coefficient on PtOCI is
statistically insignificant. To summarize, we find that in general fair value adjustments included
in OCI predict future bank earnings irrespective of whether the net fair value adjustments
included in OCI for the period are positive or negative.
4.6. Robustness tests
We perform several robustness tests. First, we use an alternative measure of performance,
pre-tax residual earnings measured as income before taxes minus 12% times lagged book value
of common equity (BHCK3210), divided by lagged total assets. We use residual earnings as an
additional measure of future performance in our analysis as bank managers have increasingly
- 24 -
incorporated performance metrics that include an adjustment for the return on capital in their
decision-making (Kimball 1998). Also, residual earnings are frequently used in bank valuation
models (e.g., Begley et al. 2006; Kohlbeck and Warfeld 2007). We present the results using this
alternative performance measure in Table 7. Panel A shows the results of the predictive value of
summary OCI for one- and two-year-ahead pre-tax residual earnings, while Panel B shows the
results of individual OCI components. Using this measure, we obtain results that are consistent
with our main results. Specifically, in Panel A, the coefficient on PtOCI is positive and
significant for both one- and two-year-ahead residual earnings. In Panel B, similar to our main
analysis, net unrealized gains and losses on available-for-sale securities are positively associated
with both one and two year-ahead residual earnings, whereas net unrealized gains and losses
from derivative contracts classified as cash flow hedges are negatively associated with future
residual earnings. Also, the coefficients on PtOCI-AFS and PtOCI-Derivatives are statistically
different from each other in both one- and two-year-ahead regressions. Finally, the net unrealized
gains and losses from other OCI components are negatively related to one-year-ahead residual
earnings but not significantly related to two-year-ahead residual earnings.
Second, to alleviate the concern that our documented results could be due to the
differences in the asset holdings and business models of banks, we include several additional
controls and re-run our tests. Specifically, we include bank-year-specific controls for asset
composition (i.e., loans-to-asset ratio), risk buffers (i.e., equity-to-asset ratio), funding structure
(i.e., deposits-to-asset ratio), and asset risk (i.e., nonperforming assets-to-total assets ratio) in
Equations 1 to 4. In an alternate specification, we replace these additional control variables with
bank-fixed effects to control for the business model of the banks. The results of these robustness
tests are qualitatively similar to those reported earlier, and our inferences remain unchanged. The
- 25 -
only exception is that the coefficient on PtOCI-Derivatives is negative but not statistically
significant in the regression including firm fixed effects where one-year-ahead Pre-tax EBP is
the dependent variable.
Finally, in our primary analyses, we follow prior banking and predictive value research
and use assets to scale PtOCI and its components. We re-examine whether our results are
sensitive to the choice of the deflator by re-estimating Equations 1–4 after scaling PtOCI and its
components by the book value of equity (BHCK3210). The results are qualitatively similar to
those reported in Tables 2 and 3. If we instead scale by total interest income (BHCK4107), the
results are also qualitatively similar, with the exception that PtOCI-Derivatives is negative but
not statistically significant when one-year-ahead Pre-tax EBP is the dependent variable.
5. Conclusion
The increasing use of fair values in financial reporting has sparked an ongoing debate
about the merits of a fair value accounting-based reporting system. We contribute to this debate
by examining whether fair value adjustments embedded in OCI predict future performance in a
sample of banks. We find that fair value-based OCI and its individual components predict future
bank earnings both one and two years ahead. Importantly, different components of OCI have
different implications for future bank profitability, suggesting that not all unrealized gains and
losses included in OCI are similar. Furthermore, the predictive ability of fair value estimates for
future performance is enhanced when the fair values can be measured more reliably. Finally, we
present some of the first empirical evidence contradicting the claims that, during the financial
crisis, fair value accounting forced banks to record excessive downward adjustments to assets’
fair values unrelated to the deterioration in the underlying fundamentals. We find that fair value
- 26 -
adjustments recorded during the crisis predict future bank performance both one and two years
ahead.
Our research informs claims by the FASB and the IASB that fair value accounting meets
the objectives of financial reporting by providing decision-useful information helpful for the
prediction of future performance. Our results are also relevant for U.S. bank regulators, who
have faced criticism for the inclusion of unrealized gains and losses reported in accumulated
other comprehensive income in the calculation of regulatory capital, in line with the
recommendations of Basel III Capital Framework.
As a caveat, care should be exercised in generalizing our findings to firms beyond the
banking industry. Our findings are based on a sample comprised only of banks because banks
have experienced the most immediate and direct impact of the move toward more fair value-
based reporting. Relative to other firms, balance sheets of banks are comprised almost entirely of
financial assets. The fair values of nonfinancial assets may not be estimable with the same
degree of reliability (Barth and Landsman 1995), and this might impact the predictive value of
fair value estimates. Finally, our study speaks to only the mixed-attribute financial reporting
system in use today. It may be the case that our findings do not hold in a “full” fair value
reporting system.
- 27 -
References
Aboody, D., M. E. Barth, and R. Kasznik (1999). Revaluations of fixed assets and future firm
performance: Evidence from the UK. Journal of Accounting and Economics 26 (1–3): 149–178.
Altamuro, J., and A. Beatty (2010). How does internal control regulation affect financial
reporting? Journal of Accounting and Economics 49 (1–2): 58–74.
American Bankers Association (2008). Letter to the SEC. September 23.
American Bankers Association (2012). Treatment of unrealized gains and losses under the Basel
III capital framework. Available at
http://www.aba.com/Advocacy/commentletters/Documents/453b885719b74b46b59d56ca98bf04
d9cl_Basel_GainsLosses2012March.pdf.
Amihud, Y. (2002). Illiquidity and stock returns: cross-section and time series effects. Journal of
Financial Markets 5: 31–56.
Barth, M. E. (1994). Fair value accounting: Evidence from investment securities and the market
valuation of banks. The Accounting Review 69 (1): 1–25.
Barth, M. E. (2000). Valuation-based accounting research: Implications for financial reporting
and opportunities for future research. Accounting and Finance 41: 7–31.
Barth, M. E. (2006). Including estimates of the future in today’s financial statements. Accounting
Horizons 20 (3): 271–285.
Barth, M. E. (2007). Standard-setting measurement issues and the relevance of research.
Accounting and Business Research. Special Issue: International Accounting Policy Forum: 7–15.
Barth, M. E. (2014). Measurement in financial reporting: The need for concepts. Accounting
Horizons 28 (2): 331–352.
Barth, M. E., W. H. Beaver, and W. R. Landsman (1996). Value-relevance of banks’ fair value
disclosures under SFAS No. 107. The Accounting Review 71 (4): 513–537.
Barth, M. E., and W. R. Landsman (1995). Fundamental issues related to using fair value
accounting for financial reporting. Accounting Horizons 9: 97–107
Barth, M. E., W. R. Landsman, and J. M. Wahlen (1995). Fair value accounting: Effects on
banks’ earnings volatility, regulatory capital, and value of contractual cash flows. Journal of
Banking and Finance 19: 577–605.
Beatty, A.L, B. Ke, and K. R. Petroni (2002). Earnings management to avoid earnings declines
across publicly and privately held banks. The Accounting Review 77(3): 547–570.
- 28 -
Beaver, W. H., and M. Venkatachalam (2003). Differential pricing of components of bank loan
fair values. Journal of Accounting, Auditing and Finance 18 (1): 41–67.
Begley, J., S. L. Chamberlain, and Y. Li (2006). Modeling goodwill for banks: A residual
income approach with empirical tests. Contemporary Accounting Research 23 (1): 31–68.
Bhat, G., R. Frankel, and X. Martin (2011). Panacea, Pandora’s Box or placebo: Feedback in
mortgage-backed security holdings and fair value accounting. Journal of Accounting and
Economics, 52 (2–3): 153–173.
Bischof, J., U. Bruggeman, and H. Daske (2012). Fair value reclassifications of financial assets
during the financial crisis. Working paper. University of Manheim.
Bowen, R., and U. Khan (2014). Market reactions to policy deliberations on fair value
accounting and impairment rules during the financial crisis of 2008–2009. Journal of Accounting
and Public Policy 33: 233–259.
Campbell, J. (2015). The fair value of cash flow hedges, future profitability, and stock returns.
Contemporary Accounting Research 32 (1): 243–279.
CFA Institute (2005). A comprehensive business reporting model: Financial reporting for
investors. CFA Center for Financial Market Integrity.
Chambers, D., T. Linsmeier, C. Shakespeare, and T. Sougiannis (2007). An evaluation of SFAS
no. 130 comprehensive income disclosures. Review of Accounting Studies 12: 557–593.
Chisnall. P. (2001). Fair value accounting: An industry view. Balance Sheet 9, 1, 2001, 27–33.
Chordia, T., A. Sarkar, and A. Subramanyam (2005). An empirical analysis of stock and bond
market liquidity. Review of Financial Studies 18: 85–129.
Dechow, P. M. (1994). Accounting earnings and cash flows as measures of firm performance:
The role of accounting accruals. Journal of Accounting and Economics 18 (1): 3–42.
Dechow, P. M., S. P. Kothari, and R. L. Watts (1998). The relation between earnings and cash
flows. Journal of Accounting and Economics 25 (2): 133–168.
Dhaliwal, D., K. R. Subramanyam, and R. Trezevant (1999). Is comprehensive income superior
to net income as a measure of firm performance? Journal of Accounting and Economics 26: 43–
67.
Eccher, E. A., K. Ramesh, and S. R. Thiagarajan (1996). Fair value disclosures by bank holding
companies. Journal of Accounting and Economics 22: 79–117.
- 29 -
Evans, M. E., L. D. Hodder, and P. E. Hopkins (2014). The predictive ability of fair values for
future financial performance of commercial banks and the relation of predictive ability to banks’
share prices. Contemporary Accounting Research 31 (1): 1–12.
Federal Deposit Insurance Corporation (2013). Regulatory Capital Rules: Regulatory Capital,
Implementation of Basel III, Capital Adequacy, Transition Provisions, Prompt Corrective
Action, Standardized Approach for Risk-Weighted Assets, Market Discipline and Disclosure
Requirements, Advanced Approaches Risk-Based Capital Rule, and Market Risk Capital Rule;
Interim Final Rule. Available at https://www.fdic.gov/news/board/2013/2013-07-
09_notice_dis_a_res.pdf.
Financial Accounting Standards Board (FASB) (2006). Fair Value Measurements. Statement of
Financial Accounting Standard No. 157. FASB, Norwalk, CT.
FASB (2010). Conceptual Framework for Financial Reporting. Statement of Financial
Accounting Concept No. 8. FASB, Norwalk, CT.
Goh, B.W, J. Ng, and K.O. Yong (2009). Market pricing of banks’ fair value assets reported
under SFAS 157 during the 2008 economic crisis. Working paper. Massachusetts Institute of
Technology and Singapore Management University.
Hodder, L. D., P. E. Hopkins, and J. M. Wahlen (2006). Risk-relevance of fair value income
measures for commercial banks. The Accounting Review 81 (2): 337–375.
Jones, D. A., and K. J. Smith (2011). Comparing the value relevance, predictive value, and
persistence of other comprehensive income and special items. The Accounting Review 86 (6):
2047–2073.
Kanagaretnam, K., R. Mathieu, and M. Shehata (2009). Usefulness of comprehensive income
reporting in Canada. Journal of Accounting and Public Policy 28: 349–365.
Kanagaretnam, K., C. Y. Lim, and G. J. Lobo (2014). Effects of international institutional factors
on earnings quality of banks. Journal of Banking and Finance 39: 87–106.
Khan, U. (2015). Does fair value accounting contribute to systemic risk in the banking industry?
Working paper. Columbia University.
Kimball, R. C. (1998). Economic profit and performance measurement in banking. New England
Economic Review (July-August): 35–53.
Kohlbeck, M., and T. D. Warfield (2007). Unrecorded intangible assets: Abnormal earnings and
valuation. Accounting Horizons 21 (1): 23–41.
- 30 -
Kolev, K.S. (2008). Do investors perceive marking-to-model as marking-to-myth? Early
evidence from SFAS 157 disclosures. Yale School of Management Working Paper. Available at
SSRN: http://ssrn.com/abstract=1336368.
Landsman, W. R. (2007). Is fair value accounting information relevant and reliable? Evidence
from capital market research. Accounting and Business Research. Special Issue: International
Accounting Policy Forum: 19–30.
Liu, C., S. G., Ryan, and J. M. Wahlen (1997). Differential valuation implications of loan loss
provisions across banks and fiscal quarters. The Accounting Review 72 (1): 133–146.
Laux, C., and C. Leuz (2010). Did fair value accounting contribute to the financial crisis?
Journal of Economic Perspectives 24 (1): 93–118.
Makar, S., L. Wang, and P. Alam (2013). The mixed attribute model in SFAS 133 cash flow
hedge accounting: Implications for market pricing. Review of Accounting Studies 18 (1): 66–94.
Nelson, K. K. (1996). Fair value accounting for commercial banks: An empirical analysis of
SFAS No. 107. The Accounting Review 71 (2): 161–182.
Nissim, D. (2003). Reliability of bank’s fair value disclosure of loans. Review of Quantitative
Finance and Accounting 20 (4): 355–384.
Nissim, D., and S. Penman (2008). Principles for the application of fair value accounting. Center
for Excellence in Accounting and Security Analysis: Occasional Paper Series, Columbia
Business School.
Ohlson, J. A. (1999). On transitory earnings. Review of Accounting Studies 4: 145–162.
Park, M. S., T. Park, and B. T. Ro (1999). Fair value disclosures for investment securities and
bank equity: Evidence from SFAS No. 115. Journal of Accounting, Auditing and Finance 14 (3):
347–370.
Petroni, K. R., and J. M. Wahlen (1995). Fair values of equity and debt securities and share
prices of property-liability insurers. The Journal of Risk and Insurance 62 (4): 719–737.
Petroni, K. R., and J. M. Wahlen (1997). Fair value disclosure of debt securities portfolio:
Reliable information about future interest income? Working Paper, Indiana University.
Schipper, K., and L. Vincent (2013). Earnings quality. Accounting Horizons 17: 97–110.
Song, C.J., W.B. Thomas, and H. Yi (2010). Value relevance of FAS No. 157 fair value
hierarchy information and the impact of corporate governance mechanisms. The Accounting
Review 85 (4): 1375–1410.
Storey, R. K., and S. Storey (1998). The Framework of Financial Accounting Concepts and
- 31 -
Standards. Financial Accounting Series No. 181-C Special Report, FASB, Norwalk CT.
Venkatachalam, M. (1996). Value-relevance of banks’ derivative disclosures. Journal of
Accounting and Economics 22, 327–355.
Wahlen, J. M. (1994). The nature of information in commercial bank loan loss disclosures. The
Accounting Review 69 (3): 455–478.
Wesbury, B. (2008). How to start the healing now. The Wall Street Journal. October 1.
- 32 -
Appendix – Variable Definitions
Variable Definition Data
Source
Assets ($ MM) Total assets (BHCK2170). FR Y-9C
Pre-tax ROA Income before taxes (BHCK4301) divided by lagged total assets
(BHCK2170).
FR Y-9C
Pre-tax EBP Pre-tax net income (BHCK4301) plus provision for loan losses
(BHCK4230), divided by lagged total assets (BHCK2170).
FR Y-9C
Pre-tax RE Residual earnings divided by lagged total assets. Residual earnings are
calculated as income before taxes (BHCK4301) minus 12% times the
lagged book value of equity (BHCK3210).
FR Y-9C
PtOCI Pre-tax other comprehensive income divided by lagged total assets
(BHCK2170). Pre-tax other comprehensive income is calculated by
dividing other comprehensive income (BHCKB511) by one minus the
statutory tax rate (i.e., 35% during our sample period).
FR Y-9C
PtOCI-
Derivatives
Pre-tax unrealized gains and losses related to changes in the fair value of
derivative contracts classified as cash flow hedges
(CHG_FV_EFF_HEDGE divided by one minus the statutory tax rate)
divided by lagged total assets (BHCK2170).
FR Y-9C,
SNL
Financial
PtOCI-AFS Pre-tax unrealized gains and losses related to changes in fair values of
investment securities classified as available for sale (CHG_UNRL_GAIN
divided by one minus the statutory tax rate) divided by lagged total assets
(BHCK2170).
FR Y-9C,
SNL
Financial
PtOCI-Other All other adjustments included in PtOCI other than PtOCI-AFS and
PtOCI-Derivatives, calculated as [(TOT_OCI minus
CHG_FV_EFF_HEDGE minus CHG_UNRL_GAIN), divided by one
minus the statutory tax rate] divided by lagged total assets (BHCK2170).
FR Y-9C,
SNL
Financial
Liquid_Annual Measure of market-wide liquidity based on Amihud (2002). A ratio of
daily absolute returns to trading volume is calculated for all NYSE stocks
with positive trading volume. Next, a market-wide annual liquidity
measure is calculated as the daily market-cap weighted average of all
daily measures after excluding outliers in the top and bottom 1% of the
sample.
CRSP
Agency_Secs Proportion of securities issued or guaranteed by the U.S. government or
its agencies. Before 2009 [in 2009 and thereafter], this ratio is calculated
as (BHCK0211 + BHCK1289 + BHCK1294 + BHCK1698 + BHCK1703
+ BHCK1714 + BHCK1718 + BHCK1286 + BHCK1291 + BHCK1297
+ BHCK1701 + BHCK1706 + BHCK1716 + BHCK1731) / (BHCK1754
+ BHCK1772)
[(BHCK0211 + BHCK1289 + BHCK1294 + BHCKG300 + BHCKG304
+ BHCKG312 + BHCKG316 + BHCK1286 + BHCK1291 + BHCK1297
+ BHCKG302 + BHCKG306 + BHCKG314 + BHCKG318 +
BHCKK142 + BHCKK150 + BHCKK144 + BHCKK152) / (BHCK1754
+ BHCK1772)].
FR Y-9C
- 33 -
Table 1
Descriptive Statistics
Mean Std. Dev. 1st Pctl. 25th Pctl. Median 75th Pctl. 99th Pctl.
Assetst ($millions) 5,656 26,384 164 383 688 1,461 131,476
Log(Assetst) 13.68 1.40 11.98 12.76 13.38 14.12 18.98
Pre-tax ROAt+1 0.0111 0.0135 -0.0436 0.0073 0.0132 0.0180 0.0361
Pre-tax EBPt+1 0.0157 0.0098 -0.0152 0.0111 0.0160 0.0207 0.0430
Pre-tax REt+1 0.0003 0.0133 -0.0537 -0.0036 0.0023 0.0073 0.0259
PtOCIt 0.0002 0.0046 -0.0120 -0.0017 0.0000 0.0020 0.0133
PtOCI-Derivativest 0.0000 0.0004 -0.0015 0.0000 0.0000 0.0000 0.0015
PtOCI-AFSt 0.0004 0.0037 -0.0087 -0.0012 0.0002 0.0019 0.0106
PtOCI-Othert -0.0001 0.0008 -0.0038 0.0000 0.0000 0.0000 0.0015
Table 1 presents descriptive statistics. For all variables except PtOCI-Derivatives, PtOCI-AFS, PtOCI-Other,
the sample includes 14,781 bank-year observations (2,420 unique banks) from 2001 to 2012, taken from
bank FR Y-9C reports. Assets is equal to total assets. Pre-tax ROA is calculated as income before taxes
divided by lagged total assets. Pre-tax EBP is calculated as pre-tax net income plus the provision for loan
losses, divided by total assets. Pre-tax RE is defined as residual earnings, which is calculated as pre-tax
income minus 12% times the lagged book value of equity, divided by lagged assets. PtOCI is defined as pre-
tax other comprehensive income, which is calculated as other comprehensive income, divided by one minus
the maximum statutory corporate tax rate (35% for all years in our sample), divided by lagged total assets.
PtOCI-Derivatives, PtOCI-AFS, and PtOCI-Other are available for 6,485 bank-year observations (965
unique banks) and are defined similarly to PtOCI but using only the individual components of other
comprehensive income, taken from SNL Financial. All variables have been winsorized at the 1% and 99%
level.
- 34 -
Table 2
Predictability of pre-tax earnings based on fair value-oriented other comprehensive income
Panel A: Predictability of pre-tax earnings using total OCI
Pre-tax ROAt+1 (t+2) = α +β1Pre-tax ROAt+ β2Log(Assetst) +
β3[Pre-tax ROAt* Log(Assetst)] + β4ptOCIt + year effects + εt+1 (t+2) (1)
Prediction Dep. Var: Pre-tax ROAt+1 Dep. Var: Pre-tax ROAt+2
coef. t-stat coef. t-stat
Intercept
-0.0018 -1.27
-0.0022 -1.16
Pre-tax ROAt (+) 0.8912 10.34 *** 0.8413 7.44 ***
Log(Assetst) (+) 0.0004 4.07 *** 0.0006 4.82 ***
Pre-tax ROAt* Log(Assetst) -0.0161 -2.59 *** -0.0274 -3.38 ***
PtOCIt (+) 0.1136 3.77 *** 0.0928 4.27 ***
Year Fixed Effects
Included
Included
N
14,781
12,145
Adj. R2
0.536
0.385
Panel B: Predictability of pre-tax earnings using OCI components
Pre-tax ROAt+1 (t+2) = α +β1Pre-tax ROAt+ β2Log(Assetst) + β3[Pre-tax ROAt* Log(Assetst)] +
β4ptOCI-AFSt + β5ptOCI-Derivativest +β6ptOCI-Othert + year effects +ε (3)
Prediction Dep. Var: Pre-tax ROAt+1 Dep. Var: Pre-tax ROAt+2
coef. t-stat coef. t-stat
Intercept
-0.0045 -2.15 ** -0.0089 -3.67 ***
Pre-tax ROAt (+) 0.8918 7.09 *** 0.9297 6.38 ***
Log(Assetst) (+) 0.0006 4.03 *** 0.0011 6.33 ***
Pre-tax ROAt* Log(Assetst) -0.0191 -2.13 ** -0.0355 -3.55 ***
PtOCI-AFSt (+) 0.2663 3.69 *** 0.1847 4.35 ***
PtOCI-Derivativest (-) -1.3503 -2.88 *** -1.3429 -2.92 ***
PtOCI-Othert
-0.7252 -3.55 *** -0.2224 -1.29
Year Fixed Effects
Included
Included
N
6,485
5,566
Adj. R2 0.531 0.386
F-test: β4 = β5 F = 29.72*** F = 16.71***
Table 2 presents the results for regressions (1) and (3), testing the ability of fair value adjustments in other
comprehensive income to predict future earnings when the dependent variable is measured at t+1 and t+2.
Pre-tax ROA is calculated as pre-tax net income, divided by lagged total assets. Assets is total assets. In
Panel A, PtOCI is equal to pre-tax other comprehensive income, calculated as the reported after-tax other
comprehensive income divided by one minus the maximum statutory corporate tax rate of 35%, divided by
lagged total assets. In Panel B, PtOCI-Derivatives, PtOCI-AFS, and PtOCI-Other, available for a subsample,
are defined similarly to PtOCI but using only the individual components of other comprehensive income. t-
statistics are based on standard errors clustered by firm. One-tailed tests of significance are used where
predictions have been made. ***, **, * indicate statistical significance at the 1, 5, and 10 % levels,
respectively.
- 35 -
Table 3
Predictability of pre-tax earnings before the provision for loan losses (EBP) based on OCI
Panel A: Predictability of pre-tax EBP using total OCI
Pre-tax EBPt+1 (t+2) = α +β1Pre-tax ROAt+ β2Log(Assetst) +
β3[Pre-tax ROAt* Log(Assetst)] + β4ptOCIt + year effects + εt+1 (t+2) (2)
Prediction Dep. Var: Pre-tax EBPt+1 Dep. Var: Pre-tax EBPt+2
coef. t-stat coef. t-stat
Intercept
-0.0062 -3.32 *** -0.0052 -2.54 **
Pre-tax ROAt (+) 0.9304 9.05 *** 0.8010 6.86 ***
Log(Assetst) (+) 0.0009 7.12 *** 0.0010 6.87 ***
Pre-tax ROAt* Log(Assetst)
-0.0288 -3.85 *** -0.0268 -3.17 ***
PtOCIt (+) 0.1059 3.94 *** 0.0902 4.60 ***
Year Fixed Effects
Included
Included
N
14,781
12,145
Adj. R2
0.461
0.440
Panel B: Predictability of pre-tax EBP using OCI components
Pre-tax EBPt+1 (t+2) = α +β1Pre-tax ROAt+ β2Log(Assetst) + β3[Pre-tax ROAt* Log(Assetst)] +
β4ptOCI-AFSt + β5ptOCI-Derivativest +β6ptOCI-Othert + year effects +ε (4)
Prediction Dep. Var: Pre-tax EBPt+1 Dep. Var: Pre-tax EBPt+2
coef. t-stat coef. t-stat
Intercept
-0.0124 -5.84 *** -0.0130 -6.07 ***
Pre-tax ROAt (+) 1.0640 8.42 *** 0.9706 7.21 ***
Log(Assetst) (+) 0.0014 9.27 *** 0.0015 10.04 ***
Pre-tax ROAt* Log(Assetst)
-0.0424 -4.77 *** -0.0417 -4.44 ***
PtOCI-AFSt (+) 0.2427 4.11 *** 0.1762 5.15 ***
PtOCI-Derivativest (-) -0.4032 -1.62 * -0.6634 -2.25 **
PtOCI-Othert
-0.1848 -1.08
-0.0460 -0.41
Year Fixed Effects
Included
Included
N
6,485
5,566
Adj. R2 0.349 0.335
F-test: β4 = β5 F = 8.05*** F = 10.23***
Table 3 presents the results for regressions (2) and (4) testing the ability of fair value adjustments in other
comprehensive income to predict future earnings before the provision when the dependent variable is
measured at t+1 and t+2. Pre-tax EBP is calculated as pre-tax net income plus the provision for loan
losses, divided by lagged total assets. Pre-tax ROA is calculated as pre-tax net income, divided by lagged
total assets. Assets is total assets. In Panel A, PtOCI is equal to pre-tax other comprehensive income,
calculated as the reported after-tax other comprehensive income divided by 1 minus the maximum
statutory corporate tax rate of 35%, divided by lagged total assets; in Panel B, PtOCIt-Derivatives, PtOCIt-
AFS, and PtOCIt-Other, available for a subsample, are defined similarly to PtOCI but using only the
individual components of other comprehensive income. t-statistics are based on standard errors clustered
by firm. One-tailed tests of significance are used where predictions have been made. ***, **, * indicate
statistical significance at the 1, 5, and 10 % levels, respectively.
- 36 -
Table 4
Predictability of pre-tax earnings and EBP in periods of high versus low liquidity
Panel A: One-year-ahead predictability of pre-tax earnings and EBP by periods of high vs. low liquidity
Pre-tax ROAt+1 or Pre-tax EBPt+1 = α +β1Pre-tax ROAt+ β2Log(Assetst) + β3[Pre-tax ROAt* Log(Assetst)] + β4PtOCIt + year effects + εt+1 (t+2)
Dependent Variable = Pre-tax ROAt+1
Dependent Variable = Pre-tax EBPt +1
Prediction High Liquidity Low Liquidity High Liquidity Low Liquidity
coef. t-stat coef. t-stat coef. t-stat coef. t-stat
Intercept
0.0056 2.16 ** -0.0077 -3.99 *** 0.0055 2.10 ** -0.0078 -3.73 ***
Pre-tax ROAt (+) 0.6874 3.96 *** 1.0071 9.64 *** 0.5381 2.83 *** 1.1263 10.88 ***
Log(Assetst) (+) -0.0001 -0.61
0.0007 5.59 *** 0.0001 0.41
0.0015 9.92 ***
Pre-tax ROAt* Log(Assetst)
-0.0016 -0.13
-0.0243 -3.23 *** 0.0028 0.20
-0.0450 -6.00 ***
PtOCIt (+) 0.1740 4.47 *** 0.0866 2.23 ** 0.1312 3.80 *** 0.0963 2.86 ***
Year Fixed Effects
Yes
Yes
Yes
Yes
N
5,696
9,085
5,696
9,085
Adj. R2
0.482
0.564
0.511
0.443
- 37 -
Table 4 (continued)
Predictability of pre-tax earnings and EBP in periods of high versus low liquidity
Panel B: Two-year-ahead predictability of pre-tax earnings and EBP by periods of high vs. low liquidity
Pre-tax ROAt+2 or Pre-tax EBPt+2 = α +β1Pre-tax ROAt+ β2Log(Assetst) + β3[Pre-tax ROAt* Log(Assetst)] + β4PtOCIt + year effects εt+1 (t+2)
Dependent Variable = Pre-tax ROAt+2
Dependent Variable = Pre-tax EBPt +2
Prediction High Liquidity Low Liquidity High Liquidity Low Liquidity
coef. t-stat coef. t-stat coef. t-stat coef. t-stat
Intercept
0.0082 2.31 ** -0.0094 -4.07 *** 0.0047 1.59
-0.0070 -2.97 ***
Pre-tax ROAt (+) 0.6257 2.40 *** 0.9734 7.98 *** 0.3668 1.62 * 0.9823 8.10 ***
Log(Assetst) (+) -0.0001 -0.28
0.0011 6.69 *** 0.0003 1.37 * 0.0014 8.10 ***
Pre-tax ROAt* Log(Assetst)
-0.0155 -0.83
-0.0347 -3.97 *** 0.0043 0.26
-0.0398 -4.56 ***
PtOCIt (+) 0.1334 3.01 *** 0.0724 3.52 *** 0.1220 3.08 *** 0.0746 3.94 ***
Year Fixed Effects
Yes
Yes
Yes
Yes
N
4,478
7,667
4,478
7,667
Adj. R2
0.340
0.460
0.329
0.395
Table 4 presents the results for regressions (1) and (3) testing the ability of fair value adjustments in other comprehensive income to predict one-year-
ahead earnings (in Panel A) and two-year-ahead earnings (in Panel B) across periods of high vs. low market-wide liquidity. Pre-tax ROA is calculated as
pre-tax net income, divided by lagged total assets. Pre-tax EBP is calculated as pre-tax net income plus the provision for loan losses, divided by lagged
total assets. Assets is total assets. PtOCI is equal to pre-tax other comprehensive income, calculated as the reported after-tax other comprehensive income
divided by 1 minus the maximum statutory corporate tax rate of 35%, divided by lagged total assets. The sample is partitioned based on years of high
liquidity (t = 2005, 2006, 2007, 2010, 2011, 2012, and 2013) and low liquidity (t = 2001, 2002, 2003, 2004, 2008, and 2009). Sample years are classified
as high and low liquidity based on Liquid_Annual. See the appendix for details on the estimation of Liquid_Annual. t-statistics are based on standard
errors clustered by firm. One-tailed tests of significance are used where predictions have been made. ***, **, * indicate statistical significance at the 1, 5,
and 10 % levels, respectively.
- 38 -
Table 5
Predictability of pre-tax earnings and EBP in publicly traded banks based on characteristics of investment securities
Panel A: One-year-ahead predictability of pre-tax earnings and EBP by the extent of U.S. government guaranteed investment securities
Pre-tax ROAt+1 or Pre-tax EBPt+1 = α +β1Pre-tax ROAt+ β2Log(Assetst) + β3[Pre-tax ROAt* Log(Assetst)] + β4PtOCIt + year effects + εt+1 (t+2)
Dependent Variable = Pre-tax ROAt+1
Dependent Variable = Pre-tax EBPt +1
Percent of U.S. Guaranteed Investments Percent of U.S. Guaranteed Investments
Prediction High Low High Low
coef. t-stat coef. t-stat coef. t-stat coef. t-stat
Intercept
-0.0058 -1.30
-0.0041 -0.71
-0.0106 -2.32 ** -0.0084 -1.48
Pre-tax ROAt (+) 0.7354 2.59 *** 0.7917 3.45 *** 0.8524 2.87 *** 0.8350 3.21 ***
Log(Assetst) (+) 0.0005 1.61 * 0.0001 0.25
0.0010 2.93 *** 0.0006 1.73 **
Pre-tax ROAt* Log(Assetst) -0.0089 -0.47
0.0016 0.10
-0.0217 -1.08
-0.0084 -0.47
PtOCIt (+) 0.3088 2.66 *** 0.0954 1.22
0.1624 1.56 * 0.0937 1.24
Year Fixed Effects
Yes
Yes
Yes
Yes
N
1,644
1,650
1,644
1,650
Adj. R2
0.456
0.557
0.438
0.532
- 39 -
Table 5 (continued)
Predictability of pre-tax earnings and EBP in publicly traded banks based on characteristics of investment securities
Panel B: Two-year-ahead predictability of pre-tax earnings and EBP by the extent of U.S. government guaranteed investment securities
Pre-tax ROAt+2 or Pre-tax EBPt+2 = α +β1Pre-tax ROAt+ β2Log(Assetst) + β3[Pre-tax ROAt* Log(Assetst)] + β4PtOCIt + year effects εt+1 (t+2)
Dependent Variable = Pre-tax ROAt+2
Dependent Variable = Pre-tax EBPt +2
Percent of U.S. Guaranteed Investments
Percent of U.S. Guaranteed Investments
Prediction High Low High Low
coef. t-stat coef. t-stat coef. t-stat coef. t-stat
Intercept
-0.0085 -1.32
0.0090 1.31
-0.0140 -2.61 *** 0.0007 0.12
Pre-tax ROAt (+) 0.7539 2.02 ** 0.1196 0.38
1.0305 3.18 *** 0.4582 1.47 *
Log(Assetst) (+) 0.0011 2.57 *** -0.0002 -0.50
0.0016 4.33 *** 0.0004 1.12
Pre-tax ROAt* Log(Assetst)
-0.0276 -1.08
0.0352 1.59
-0.0433 -1.94 * 0.0111 0.52
PtOCIt (+) 0.1933 1.73 ** 0.0389 0.38
0.1019 1.30 * 0.0457 0.49
Year Fixed Effects
Yes
Yes
Yes
Yes
N
1,530
1,558
1,530
1,558
Adj. R2
0.355
0.454
0.299
0.396
Table 5 presents the results for regressions (1) and (3) testing the ability of fair value adjustments in other comprehensive income to predict one-year-
ahead earnings (in Panel A) and two-year-ahead earnings (in Panel B) as a function of the extent of U.S. guaranteed investments among publicly traded
banks. Pre-tax ROA is calculated as pre-tax net income, divided by lagged total assets. Pre-tax EBP is calculated as pre-tax net income plus the provision
for loan losses, divided by lagged total assets. Assets is total assets. PtOCI is equal to pre-tax other comprehensive income, calculated as the reported
after-tax other comprehensive income divided by 1 minus the maximum statutory corporate tax rate of 35%, divided by lagged total assets. Banks whose
proportion of agency securities in their investment securities portfolio is above (below) the median are classified as High (Low) percent of U.S.
Guaranteed Investment Banks. t-statistics are based on standard errors clustered by firm. One-tailed tests of significance are used where predictions have
been made. ***, **, * indicate statistical significance at the 1, 5, and 10 % levels, respectively.
- 40 -
Table 6
Predictability of other comprehensive income during the financial crisis
Panel A: One-year-ahead predictability of earnings using total OCI
Pre-tax ROAt+1 or Pre-tax EBPt+1 = α +β1Pre-tax ROAt+ β2Log(Assetst) +
β3[Pre-tax ROAt* Log(Assetst)] + β4PtOCIt + year effects + εt+1
Prediction Dep. Var: Pre-tax ROAt+1 Dep. Var: Pre-tax EBPt+1
coef. t-stat coef. t-stat
Intercept
-0.0012 -0.39
-0.0017 -0.62
Pre-tax ROAt (+) 1.1123 4.70 *** 1.0878 6.42 ***
Log(Assetst) (+) 0.0003 1.28
0.0010 5.10 ***
Pre-tax ROAt* Log(Assetst)
-0.0326 -1.95 * -0.0482 -3.99 ***
PtOCIt (+) 0.2128 3.20 *** 0.2270 4.15 ***
Year Fixed Effects
Yes
Yes
N
2,831
2,831
Adj. R2
0.508
0.437
Panel B: Two-year-ahead predictability of earnings using total OCI
Pre-tax ROAt+2 or Pre-tax EBPt+2 = α +β1Pre-tax ROAt+ β2Log(Assetst) +
β3[Pre-tax ROAt* Log(Assetst)] + β4PtOCIt + year effects + εt+2
Prediction Dep. Var: Pre-tax ROAt+2 Dep. Var: Pre-tax EBPt+2
coef. t-stat coef. t-stat
Intercept
-0.0056 -1.73 * -0.0070 -2.71 ***
Pre-tax ROAt (+) 0.8519 3.70 *** 0.6051 3.77 ***
Log(Assetst) (+) 0.0008 3.57 *** 0.0013 7.34 ***
Pre-tax ROAt* Log(Assetst)
-0.0301 -1.87 * -0.0199 -1.77 *
PtOCIt (+) 0.1716 4.16 *** 0.1637 4.73 ***
Year Fixed Effects
Yes
Yes
N
2,664
2,664
Adj. R2
0.247
0.277
Table 6 presents the results for regressions (1) and (2) testing the ability of fair value adjustments in
other comprehensive income to predict future earnings where the dependent variable is measured at t+1
(Panel A) and t+2 (Panel B), using only observations where t=2007, 2008, and 2009. Pre-tax ROA is
calculated as pre-tax net income, divided by lagged total assets. Pre-tax EBP is calculated as pre-tax net
income plus the provision for loan losses, divided by lagged total assets. Assets is total assets. PtOCI is
equal to pre-tax other comprehensive income, calculated as the reported after-tax other comprehensive
income divided by 1 minus the maximum statutory corporate tax rate of 35%, divided by lagged total
assets. t-statistics are based on standard errors clustered by firm. One-tailed tests of significance are used
where predictions have been made. ***, **, * indicate statistical significance at the 1, 5, and 10 % levels,
respectively.
- 41 -
Table 7
Alternative measure of performance: Predictability of pre-tax residual earnings (RE) based on OCI
Panel A: Predictability of pre-tax residual earnings using total OCI
Pre-tax REt+1 (t+2) = α +β1Pre-tax ROAt+ β2Log(Assetst)
β3[Pre-tax ROAt* Log(Assetst)] + β4ptOCIt + εt
Prediction Dep. Var: Pre-tax REt+1 Dep. Var: REt+2
coef. t-stat coef. t-stat
Intercept
-0.0110 -5.75 *** -0.0107 -4.68 ***
Pre-tax ROAt (+) 0.4123 4.22 *** 0.3451 2.84 ***
Log(Assetst) (+) 0.0003 2.25 ** 0.0005 2.95 ***
Pre-tax ROAt* Log(Assetst)
0.0113 1.61
0.0014 0.16
PtOCIt (+) 0.0828 2.80 *** 0.0558 2.67 ***
Year Fixed Effects
Included
Included
N
14,781
12,145
Adj. R2
0.467
0.340
Panel B: Predictability of pre-tax residual earnings using OCI components
Pre-tax REt+1 (t+2) = α +β1Pre-tax ROAt+ β2Log(Assetst) + β3[Pre-tax ROAt* Log(Assetst)] +
β4ptOCI-AFSt + β5ptOCI-Derivativest +β6ptOCI-Othert +εt
Prediction Dep. Var: Pre-tax REt+1 Dep. Var: REt+2
coef. t-stat coef. t-stat
Intercept
-0.0143 -5.65 *** -0.0182 -6.88 ***
Pre-tax ROAt (+) 0.4964 3.65 *** 0.5198 3.42 ***
Log(Assetst) (+) 0.0005 2.78 *** 0.0009 5.09 ***
Pre-tax ROAt* Log(Assetst)
0.0034 0.35
-0.0122 -1.17
PtOCI-AFSt (+) 0.2103 2.95 *** 0.1478 3.54 ***
PtOCI-Derivativest (-) -1.5023 -3.11 *** -1.4852 -3.23 ***
PtOCI-Othert
-0.6547 -3.12 *** -0.1395 -0.83
Year Fixed Effects
Included
Included
N
6,485
5,566
Adj. R2 0.484 0.361
F-test: β4 = β5 F = 30.29*** F = 18.88***
Table 7 presents the results for the above regressions testing the ability of fair value adjustments in other
comprehensive income to predict future earnings where the dependent variable is measured at t+1 and t+2.
Pre-tax RE is calculated as pre-tax net income minus 12% times the lagged book value of common equity,
divided by lagged total assets. Pre-tax ROA is calculated as pre-tax net income, divided by lagged total assets.
Assets is total assets. In Panel A, PtOCI is equal to pre-tax other comprehensive income, calculated as the
reported after-tax other comprehensive income divided by 1 minus the maximum statutory corporate tax rate
of 35%, divided by lagged total assets. In Panel B, PtOCIt-Derivatives, PtOCIt-AFS, and PtOCIt-Other,
available for a subsample, are defined similarly to PtOCI but using only the individual components of other
comprehensive income. t statistics are based on standard errors clustered by firm. One-tailed tests of
significance are used where predictions have been made. ***, **, * indicate statistical significance at the 1, 5,
and 10 % levels, respectively.