Mixing fair-value and historical-cost accounting: predictable other-comprehensive-income and mispricing of bank stocks

Abstract

Other comprehensive income (OCI) items are often considered to be transitory (Chambers et al. 2007; IASB 2013; CFA2014). In this paper, we show that a significant portion of OCI, namely unrealized gains and losses (UGL) from available-for-sale (AFS) debt securities, is non-transitory: a negative correlation between accumulated unrealized gains and losses in the current period and next period UGL is predicted, and we show that this correlation is economically and statistically significant. This correlation is due to a mix of accounting methods of measurement of income from fixed-income securities: UGL are recognized based on fair values, whereas interest income is measured based on historical cost. We document that (1) this negative correlation helps explain a previously unexplained negative correlation in other comprehensive income (OCI), and (2) investors seem to price total UGL disregarding (or not understanding) the predictable, accounting-driven component of UGL.

1 Introduction

The current GAAP approach to incorporating fair value accounting information for available-for-sale (AFS) securities into a primarily historical-cost-based accounting system involves the separation of two income concepts, net income and comprehensive income, as well as “recycling” from comprehensive income to regular income.¹ Thus two different income measures are reported in the same set of financial statements.

We show that this mix of accounting methods has two side-effects: (1) unrealized accounting holding gains and losses (UGL) differ from true economic holding gains and losses, and (2) there is an induced, sizable negative correlation between UGL and accumulated unrealized holding gains and losses (AUGL) at the end of the previous fiscal year. This correlation varies systematically with the percentage of AFS securities that are invested in fixed-income securities as well as with the relative amount of accumulated unrealized gains (AUG) versus unrealized losses (AUL). We document that (1) the negative correlation between UGL and lagged AUGL helps explain the previously observed negative serial correlation in other comprehensive income, and (2) investors seem to price the total amount of UGL as real economic gains and losses, disregarding (or not understanding) the fact that reported UGL includes a predictable, accounting-driven component.

Statement of Financial Accounting Standards 115: Accounting for Certain Investments in Debt and Equity Securities (SFAS 115) requires differential treatment of unrealized holding gains and losses based on management’s intended strategy for the security (FASB 1993).² For securities that management intends to resell in the near term (“trading” securities), unrealized holding gains and losses are recognized in earnings. For securities that management intends to hold to maturity (“held to maturity” securities), unrealized holding gains and losses typically are not recognized in net income or other comprehensive income (OCI). For all other securities, (“available for sale” or AFS securities) unrealized holding gains and losses are typically recognized in OCI.

Unrealized holding gains and losses from investment in AFS debt securities are measured based on fair value. In contrast, interest income from these securities, which is reported on the income statement, is measured based on their historical cost. This combination of accounting methods results in two components of unrealized holding gains and losses: (1) change in the fair value of debt securities due to changes in expected future cash flows, the discount rate, or both, and (2) change in the difference between fair value and the corresponding amortized cost due to the difference between the fair-value-based and historical-cost-based amortization. In an efficient market, the first component is not predictable. The second component, however, is predictable.³ This is at odds with the pervasive notion in the literature that other comprehensive income items, which include UGL, are transitory (e.g., Linsmeier et al. 1997; Chambers et al. 2007; Bamber et al. 2010; Black 2015).⁴ Such a feature may lead investors to misinterpret bank financial statements, which, in turn, could lead to mispricing of bank stocks.

We analyze reported UGL of all U.S. commercial banks traded on the NYSE, AMEX, and NASDAQ. Our sample period starts in 1998 when banks were first required to disclose, in detail, UGL under Statement of Financial Accounting Standards No. 130 Reporting Comprehensive Income (SFAS 130) (FASB 1997). We predict a negative correlation between reported UGL on AFS debt securities and AUGL on AFS debt securities at the beginning of the fiscal year, and we show that this correlation is economically and statistically significant.

Since the predictable component of UGL results from the application of the effective yield method in the calculation of amortized cost, the effects of the mixed accounting methods apply to debt securities only. We show that, as the percentage of fixed-income AFS debt securities increases, both the magnitude and the significance of the negative correlation between UGL and lagged AUGL increases.

The predicted negative correlation of UGL and lagged AUGL will be observed only if the bank continues to hold the AFS debt security; a sale will lead to a realized gain or loss and no further UGL. Banks may have an incentive to hold AFS debt instruments with a market value below amortized cost because realizing a loss lowers regulatory capital and earnings while holding the instruments to maturity (or until their prices recover) avoids these effects (Moyer 1990). On the other hand, selling securities with unrealized losses brings tax benefit to the banks (Scholes et al. 1990; Warfield and Linsmeier 1992). Since these incentives for banks to hold securities with unrealized gains and unrealized losses may be asymmetric, the correlation between UGL and lagged accumulated unrealized gains (AUG) may be different from the correlation between UGL and lagged accumulated realized losses (AUL). This possible asymmetry is one of the reasons why we allow for differences in these correlations in our empirical analyses.

Another reason for an asymmetric relation between UGL and lagged AUG versus between UGL and lagged AUL is that GAAP imposes a conservative bias on the treatment of unrealized gains versus unrealized losses associated with holding AFS securities. Unlike unrealized holding gains, unrealized holding losses are sometimes required to be recognized in the income statement, even though the underlying securities are not sold; equivalent recognition is not required for unrealized gains. This recognition of unrealized holding losses happens when there is significant doubt whether the bank can hold the security until the fair value recovers to amortized cost. The resulting income statement charge is called an “other-than-temporary impairment” (OTTI).⁵ As a result of this asymmetric accounting, unrealized holding losses at the end of the fiscal year are more likely than unrealized holding gains to be associated with securities that banks plan to hold rather than to sell.⁶ Since securities with unrealized losses are more likely to be held, the conservative accounting bias implies that the correlation with UGL in the following year is expected to be greater than the correlation between unrealized gains and next year’s UGL. Our analysis shows that the correlation between AUL and UGL in the following year is greater than the correlation between AUG and UGL in the following year.

It is commonly assumed that items in OCI are transitory as they relate to volatile changes in market value (Linsmeier et al. 1997; Chambers et al. 2007; Bamber et al. 2010; Black 2015). This perceived transitory and volatile nature of OCI is the primary reason given by standard setters for permitting items of OCI to bypass the income statement. Both the FASB and the International Accounting Standards Board (IASB) view the lack of persistence as a key characteristic of OCI items (IASB 2013). In its 2014 comment letter to the IASB, the Chartered Financial Analyst Institute (CFA) argues for using lack of persistence as the main criteria for the distinction between OCI and net income (CFA 2014). Nonetheless, Jones and Smith (2011) documented a puzzling negative serial correlation in OCI. Jones and Smith posited that this serial correlation may be due to price reversion in securities investments (which seems to be at odds with the assumption of market efficiency) or to “recycling.” They were, however, unable to provide evidence consistent with either explanation. In this paper, we provide evidence that, for our sample of bank stocks, the negative correlation in OCI is due to the negative correlation between UGL and lagged AUGL, created by the accounting for UGL. When we control for this negative correlation, the serial correlation in UGL (and in OCI) disappears. This suggests that the serial correlation in OCI of banks is not an indication of price reversion; instead, it is driven by the mix of fair value and historic cost accounting.

Research has provided evidence suggesting that investors often fail to fully understand the properties of accounting measures (e.g., Sloan 1996; Penman and Zhang 2002; Campbell et al. 2015). This evidence casts doubt on investors’ ability to appreciate the subtlety/nuances of the accounting, which we describe. Investors may, for example, price the total amount of UGL as real economic gains and losses, disregarding the fact that reported UGL includes a predictable, accounting-driven component.

To examine the market pricing of UGL, we isolate a predictable component of UGL based on a linear regression of UGL on variables that reflect information on AFS securities available at the end of the previous year. We conduct three sets of analyses based on this predictable component.

First, we regress next-period stock return on predicted next year UGL and known risk factors. The estimated coefficient on predicted UGL is highly significant, consistent with investor misinterpretation of the information in UGL. Second, we form portfolios each year based on the magnitude of predicted UGL. We show that, for these portfolios, the magnitude of the mispricing is economically significant—a hedged portfolio strategy yields significant annual excess return during the sample period. Third, for the same set of portfolios, we show significant excess returns (alpha) in a Fama and French (2015) and Carhart (1997) factor model, in which we regress monthly portfolio returns on the returns on six factors—market risk premium, book-to-market factor (HML), firm size (SMB), profitability (RMW), investment (CMA), and momentum (UMD)— as well as debt-specific risk factors (i.e., proxies for shocks to the yield curve (Viale et al. 2009)).⁷

Our paper makes the following contributions to the literature.⁸ First, we explain the observed economically and statistically significant negative correlation between UGL and lagged AUGL, and we show how this correlation varies with the composition of AFS securities and with the composition of unrealized gains and losses. Second, we show that the negative autocorrelation in banks’ OCI is not an indication of price reversion or a result of the “recycling” of UGL due to sales of AFS securities. Instead, it is due to the combination of a negative correlation between UGL and lagged AUGL, and a positive correlation between current UGL and current AUGL. Third, we document evidence that the mix of historic cost and fair value accounting leads to market mispricing of bank stocks. This suggests that requiring separate disclosure of the amortization-driven component of unrealized gains and losses and the market-price-change-driven component may have informational benefits to investors.

Our paper proceeds as follows. Our main research questions and predictions are developed in Section 2. Section 3 describes the sample selection and the data gathering procedure as well as providing selective descriptive statistics. Section 4 reports the test results. Section 5 concludes the paper with a discussion of sensitivity tests and a brief summary.

2 Research questions and research design

We begin by describing the accounting mechanism by which a negative correlation arises between current UGL and lagged AUGL. We present simple algebraic and numerical illustrations. These illustrations show that unrealized losses (gains) are ceteris paribus followed by unrealized gains (losses) because the book and fair values of a fixed-income security converge at maturity.

2.1 Predictive power of AUG and AUL for UGL

Let ${\mathrm{BV}}_{\mathrm{t}}^{\mathrm{C}}$ and ${\mathrm{BV}}_{\mathrm{t}}^{\mathrm{F}}$ denote the amortized cost and the fair value of an AFS debt security held by the bank. Let UGL_t+1 denote unrealized gains and losses incurred in year t + 1. Let E_t[UGL_t+1] denote expected UGL_t+1, and ε_t+1 denote the unexpected UGL_t+1 due to change in market conditions.

Suppose that the market discount rate (${\mathrm{r}}_{\mathrm{t}}^{\mathrm{F}}$) exceeds the historical-cost-based discount rate (r^C) such that ${\mathrm{BV}}_{\mathrm{t}}^{\mathrm{C}}$ > ${\mathrm{BV}}_{\mathrm{t}}^{\mathrm{F}}$. That is,

$$ {\displaystyle \begin{array}{l}{\mathrm{r}}_{\mathrm{t}}^{\mathrm{F}}>{\mathrm{r}}^{\mathrm{C}}\hfill \\ {}{\mathrm{BV}}_{\mathrm{t}}^{\mathrm{C}}>{\mathrm{BV}}_{\mathrm{t}}^{\mathrm{F}}.\hfill \end{array}} $$

Note that, by definition, accumulated unrealized gains and losses equal the difference between the fair value and the amortized cost:

$$ {\mathrm{AUGL}}_{\mathrm{t}}={\mathrm{BV}}_{\mathrm{t}}^{\mathrm{F}}-{\mathrm{BV}}_{\mathrm{t}}^{\mathrm{C}}. $$

In this scenario, AUGL_t is less than 0.

Interest revenue reported on the income statement is calculated based on historical-cost-based interest rate (r^C). This interest revenue is less than that based on the fair-value discount rate (${\mathrm{r}}_{\mathrm{t}}^{\mathrm{F}}$) and ${\mathrm{BV}}_{\mathrm{t}}^{\mathrm{F}}$:

$$ {{\mathrm{r}}_{\mathrm{t}}^{\mathrm{F}}}^{\ast }{\mathrm{BV}}_{\mathrm{t}}^{\mathrm{F}}>{{\mathrm{r}}_{\mathrm{t}}^{\mathrm{C}}}^{\ast }{\mathrm{BV}}_{\mathrm{t}}^{\mathrm{C}}. $$

This implies:

$$ {\mathrm{E}}_{\mathrm{t}}\left[{\mathrm{BV}}_{\mathrm{t}+1}^{\mathrm{F}}\right]-{\mathrm{BV}}_{\mathrm{t}}^{\mathrm{F}}>{\mathrm{E}}_{\mathrm{t}}\left[{\mathrm{BV}}_{\mathrm{t}+1}^{\mathrm{C}}\right]-{\mathrm{BV}}_{\mathrm{t}}^{\mathrm{C}}, $$

and thus:

$$ {\mathrm{BV}}_{\mathrm{t}}^{\mathrm{C}}-{\mathrm{BV}}_{\mathrm{t}}^{\mathrm{F}}>{\mathrm{E}}_{\mathrm{t}}\left[{\mathrm{BV}}_{\mathrm{t}+1}^{\mathrm{C}}\right]-{\mathrm{E}}_{\mathrm{t}}\left[{\mathrm{BV}}_{\mathrm{t}+1}^{\mathrm{F}}\right]. $$

That is, over time, the expected difference between amortized cost of the fixed-income security and the market value of the security will gradually reduce to zero as the bond approaches its maturity. Since the difference between the amortized cost and the fair value is expected to decrease in the following year, that is, E_t[${\mathrm{BV}}_{\mathrm{t}+1}^{\mathrm{C}}$ − ${\mathrm{BV}}_{\mathrm{t}+1}^{\mathrm{F}}$] < [${\mathrm{BV}}_{\mathrm{t}}^{\mathrm{C}}$ − ${\mathrm{BV}}_{\mathrm{t}}^{\mathrm{F}}$]_, this expected decrease will be captured in UGL_t+1. As a result,

$$ {\mathrm{E}}_{\mathrm{t}}\left[{\mathrm{UGL}}_{\mathrm{t}+1}\right]=\mathrm{E}\left[{\mathrm{BV}}_{\mathrm{t}+1}^{\mathrm{F}}-{\mathrm{BV}}_{\mathrm{t}+1}^{\mathrm{C}}\right]-\left[{\mathrm{BV}}_{\mathrm{t}}^{\mathrm{F}}-{\mathrm{BV}}_{\mathrm{t}}^{\mathrm{C}}\right]>0. $$

That is, conditioning on AUGL_t being less than 0, E_t[UGL_t+1] will be greater than 0. Similar logic can be applied to show that AUGL_t greater than 0 implies E_t[UGL_t+1] will be less than 0 when ${\mathrm{BV}}_{\mathrm{t}}^{\mathrm{C}}$ < ${\mathrm{BV}}_{\mathrm{t}}^{\mathrm{F}}$.

We also illustrate the above effects via a numerical example, summarized in Appendix 2.⁹ Suppose a bank purchased a $100 corporate bond and classified the investment as AFS. The bond, which was a three-year, 10% annual coupon bond, was issued at par on 12/31/×0. Assume that, due to favorable news, the market discount rate decreased to 8% on 12/31/×1 (see Scenario 1a, Appendix 2). The fair value of the bond would increase to $103.57, and, under SFAS 115, an accumulated unrealized holding gain of $3.57 would be recorded in year 1. In year 2, the fair value of the bond would be $101.85, while the amortized cost of the bond would remain at $100. The reduction of AUGL to $1.85 (i.e., $101.85 - $100) causes the recognition of an unrealized holding loss of $1.72 (i.e., $103.57 - $101.85). Similarly, a $1.85 unrealized holding loss would be recognized in year 3.

In contrast, if true mark-to-market accounting were applied in the sense that interest expenses are also recognized based on the market interest rate as opposed to the historical rate, then in year 2 interest income would be $8.28 (i.e., $103.57*8%) as opposed to $10 under SFAS 115. As a result, the amortized cost of the bond would be $101.85, equal to its fair market value. Therefore, no unrealized holding gains or losses would be recognized in year 2. Similarly, no unrealized holding gains and losses would be recognized in year 3.

The key insight from the above example is that under true fair value accounting, UGL in both year 2 and year 3 are zero, which represents the economic reality that the market discount rate did not change in years 2 and 3. That is, there was no economic holding gain or loss in these years. Under SFAS 115, however, the company shows unrealized holding losses in both years, with UGL of ($1.72) and ($1.85) in years 2 and 3. Such losses, however, are simply due to the fact that interest income, which is measured based on historical cost, was $1.72 and $1.85 more than the economic interest income in these years. Note that the example illustrates the negative correlation, which is due to the mix of historic cost and fair value accounting; an accumulated unrealized gain of $3.57 at the end of year 1 is associated with an unrealized loss of $1.72 in year 2 and an accumulated unrealized gain of $1.85 at the end of year 2 is associated with an unrealized loss of $1.85 in year 3.

In Scenario 1b, Appendix 2, the bank chooses to sell the security at the beginning of year 2, in which case the UGL in year 2 will be zero, and hence there will be no correlation between UGL in year 2 and AUGL at the beginning of year 2.

An alternative scenario (see Scenario 2, Appendix 2) is that the discount rate increases to 12%. There will be an unrealized loss of $3.38. This will be followed by a predictable unrealized gain in the next year of $1.59 because the fair value will increase to $98.21 (and we would, again, predict a negative correlation between UGL and lagged AUGL).

Our algebraic and numerical examples are based on a single debt security. When a bank has a portfolio with thousands of securities, with some having unrealized holding losses and some having unrealized holdings gains, it is unclear whether the accounting illustrated via these examples will lead to a detectable pattern in the time-series properties of the aggregated unrealized holding gains and losses from both debt and equity securities. Nevertheless, our empirical results show that the negative correlation is statistically and economically significant.¹⁰

A key assumption underlying the predictability of UGL is that the AFS security will continue to be held by the bank. As we discussed in the introduction, a bank’s incentive to hold securities with unrealized gains versus unrealized losses may be asymmetric. In addition, due to the conservative bias in accounting, unrealized holding losses are sometimes required to be recognized in the income statement even though the underlying securities are not sold. As a result of the possible asymmetry in banks’ security holdings as well as the conservatism in accounting, accumulated unrealized holding losses at the end of the fiscal year may have different predictive power with respect to UGL, compared to accumulated unrealized gains. Therefore we estimate the following regression to assess the statistical significance of the relation between UGL and lagged AUG and AUL:

$$ {\mathrm{UGL}}_{\mathrm{jt}}={\alpha}_0+{\upalpha}_1{\mathrm{AUG}}_{\mathrm{jt}-1}+{\upalpha}_2{\mathrm{AUL}}_{\mathrm{jt}-1}+{\mathrm{e}}_{\mathrm{jt}}. $$

(1)

Another feature of the predicted negative correlation between UGL and lagged AUG and UGL and lagged AUL is that the argument only applies to fixed-income security investments. Therefore, we expect the correlation between UGL and lagged AUG and UGL and lagged AUL to increase with the percentage of AFS securities invested in fixed-income securities. We estimate the percentage of fixed-income AFS securities as follows:

$$ {\mathrm{FI}}_{\mathrm{jt}}=\frac{{\mathrm{TB}}_{\mathrm{jt}}+{\mathrm{MBS}}_{\mathrm{jt}}+{\mathrm{BOND}}_{\mathrm{jt}}+{\mathrm{MUNI}}_{\mathrm{jt}}}{{\mathrm{COSTAFS}}_{\mathrm{jt}}}, $$

where TB_jt, MBS_jt, BOND_jt, and MUNI_jt are the amortized cost of AFS securities invested in Treasury bills, mortgage-backed securities, corporate bonds, and municipal obligations for bank j and the end of year t. The denominator, COSTAFS_jt is the amortized cost of all AFS securities for bank j at the end of year t. Each year we sort firms based on the relative amount of FI into portfolios; as expected, the magnitude and significance of the negative correlation between UGL and lagged AUGL increases with the portion of debt securities in the AFS portfolio.

where TB_jt, MBS_jt, BOND_jt, and MUNI_jt are the amortized cost of AFS securities invested in Treasury bills, mortgage-backed securities, corporate bonds, and municipal obligations for bank j and the end of year t. The denominator, COSTAFS_jt is the amortized cost of all AFS securities for bank j at the end of year t. Each year we sort firms based on the relative amount of FI into portfolios; as expected, the magnitude and significance of the negative correlation between UGL and lagged AUGL increases with the portion of debt securities in the AFS portfolio.

2.2 Negative serial correlation in OCI

Jones and Smith (2011) study the total amount of OCI, including additional minimum pension liability adjustments, foreign currency translation gains and losses, changes in the fair value of derivative instruments classified as cash flow hedges as well as unrealized gains and losses from AFS securities of 236 companies from 1986 to 2005. They document a negative serial correlation in OCI, but they are unable to determine whether this correlation reflects reversion in market values or recycling.¹¹

We document a negative serial correlation in UGL and show that, once we control for the negative correlation between current UGL and lagged AUG and lagged AUL, the negative relation between current UGL and lagged UGL becomes insignificant. We also find that current UGL and current AUGL are significantly positively correlated, which is not surprising given that UGL of the current period makes up a significant portion of AUGL at the end of the current period. These findings suggest that a negative serial correlation in UGL is not necessarily indicative of reversion of market values; instead, it is due to the combination of a negative correlation between current UGL and lagged AUGL and the positive correlation between current UGL and current AUGL.

We further conduct the following regression analyses:

$$ {OCI}_{jt}={\beta}_0+{\beta}_1{OCI}_{jt-1}+e{\prime}_{jt} $$

(2a)

$$ {OCI}_{jt}={\gamma}_0+{\gamma}_1{AUG}_{jt-1}+{\gamma}_2{AUL}_{jt-1}+{e}^{\prime }{\prime}_{jt} $$

(2b)

$$ {OCI}_{jt}={\delta}_0+{\delta}_1{OCI}_{jt-1}+{\delta}_2{AUG}_{jt-1}+{\delta}_3{AUL}_{jt-1}+{e}^{\prime \prime }{\prime}_{jt} $$

(2c)

A negative estimate of β ₁ confirms the negative correlation documented by Jones and Smith (2011). Negative estimates of the coefficients γ ₁ and γ ₂ suggest that the combination of historic cost and fair value accounting discussed in section 2.1 affects the predictability of OCI. And an estimate of the coefficient δ ₁ that is not significantly different from zero suggests that this accounting effect helps explain the observed negative serial correlation in OCI.

Unlike Jones and Smith (2011), who include both the recycled component of OCI as well as new UGL arising during the year in their measure of UGL, our hand-collected data separates the recycling component of OCI (i.e., the income that is recycled out of accumulated OCI upon sale and recognized as a gain or loss in net income) from the UGL that arise during the period. This enables us to explicitly test Jones and Smith’s conjecture that the negative correlation they document might be due to reclassification. We show that, although reclassification (RECL_t) is negatively correlated with UGL_t-1, it is positively correlated with RECL_t-1. The overall correlation between RECL_t and OCI_t-1is insignificantly different from zero. These results suggest that “recycling” is unlikely to be a major contributor to the observed negative serial correlation in banks’ OCI.

2.3 Market mispricing

Studies have provided ample evidence on the value relevance of items of OCI. Chambers et al. (2007), for example, show that investors value the components of OCI approximately dollar-for-dollar.¹² Our analysis in Section 2.1 reveals, however, that UGL has two components, one of which will not be priced in an efficient market. These components are (1) change in the fair value of debt securities due to changes in expected future cash flows, the discount rate, or both, and (2) change in the difference between fair value and the corresponding amortized cost, due to the difference between the fair-value-based and historical-cost-based amortization. Unlike the first component that reflects real economic changes, the second component is purely accounting-driven. In an efficient market, the first component should be priced approximately dollar-for-dollar, but the second components should not.¹³

We conduct three sets of analyses to seek evidence regarding investor consideration of the negative correlation between UGL and lagged AUGL when pricing bank equity. (Investors may, for example, fail to understand the complexity of the accounting and price the entire UGL as an economic gain or loss.) First, we regress next period abnormal stock return on known risk factors as well as predicted next year UGL, based on banks’ currently reported AUG and AUL. Second, each year, we form portfolios based on the magnitude of predicted UGL and show that the magnitude of the mispricing is economically significant. Third, we show that there are significant excess returns over and above that explained by the Fama and French (2015) and Carhart (1997) factor model as well as by proxies for shocks to the yield curve (Viale et al. 2009), which have been posited as indicators of bank risk. We elaborate on these three sets of analyses later in the paper.

3 Sample and descriptive statistics

Our initial sample includes all U.S. commercial banks traded on the NYSE, NASDAQ, and AMEX during 1998–2012. We choose commercial banks because they often hold significant portfolios of AFS securities with unrealized gains and losses representing an important portion of the reported comprehensive income. We hand-collect data on AFS securities, including UGL, the amount of reclassified gains and losses (RECL), AUG, and AUL, from bank 10-K filings. The sample period begins in 1999, which is one year after SFAS 130 became effective. (This requirement is necessary because we require a beginning balance of AUGL.) The sample consists of 4066 observations covering 546 banks.

Table 1 reports statistics describing our sample. Because our sample period includes the years of the global financial crisis, we split our sample into two sub-periods: years 2007 to 2009, which are the years when the crisis likely affected variables key to our analyses (i.e., UGL for 2008 and 2009 and lagged UGL, AUG, AUL, and AUGL for 2007 and 2008), and 1999 to 2006 and 2010 to 2012, which are noncrisis years.¹⁴ Differences between the crisis- and the noncrisis years can be seen in the distribution of AUGL; the average AUGL, as a percentage of total assets, is 0.06 in the noncrisis years and −0.06 in the crisis years.¹⁵

Table 1 Descriptive statistics

For the whole sample and for each of the subsamples, it is evident that the variables at the core of our study may be large relative to the net income and the total assets of the banks. For example, for the full sample, the median absolute value of UGL is 14% of net income. Similarly, the fifth percentile of AUGL is −0.42% of total assets, and the 95th percentile is 0.50%. The average cost of the available-for-sale securities equals 19% of total assets.

Table 2 reports the cross-sectional correlations among the variables at the core of our analyses: UGL_t, UGL_t-1, AUGL_t-1, AUG_t-1, and AUL_t-1. The Pearson correlation between UGL_t and UGL_t-1 is negative and significant (−0.12 in the whole sample, for example).¹⁶ The correlations between UGL_t-1 and AUGL_t-1 are high and significant; for example, for the full sample the Pearson (Spearman) correlation is 0.58 (0.49), which is not surprising in light of the fact that UGL of the current period makes up a considerable portion of AUGL at the end of the current period. Also, consistent with the prediction from our analyses in section 2.1, we observe a significant negative correlation between UGL_t and AUGL_t-1 in every subsample.

Table 2 Correlations among key variables

The asymmetry in the relation between UGL_t and AUG_t-1 versus UGL_t and AUL_t-1 is also evident in Table 2; for example, the Pearson and Spearman correlation between UGL_t and AUL_t-1 in the noncrises years are −0.39 (see Panel B), whereas these correlations between UGL_t and AUG_t-1 are 0.05 and 0.02.¹⁷

4 Results

4.1 Predictive power of AUG_t and AUL_t with respect to UGL_t+1

The results of running regression (1) are summarized in Table 3.¹⁸ The first column of Panel A reports the results for the entire sample of observations. The estimate of the coefficient on AUL is significantly negative (−0.17 with a t-statistic of −3.30). In comparison, the estimated coefficient on AUG is less negative (−0.07 with a t-statistic of −2.65). The difference between the two coefficient estimates is highly significant. That is, the predicted asymmetry due to either the tendency to hold securities with unrealized losses to a greater extent than those with unrealized gains, and/or the conservative bias in accounting recognition of unrealized gains and losses, is evident in the data.

Table 3 Relation between UGL and lagged AUG and lagged AUL

In the second and third columns of Panel A, we repeat the analysis for two subsamples: the financial crisis years (2007–2009) and the noncrisis years (1999–2006, 2010–2012). This set of tests also addresses the issue of how the length of holding AFS securities affects the relation between AUG(AUL)_t-1 and UGL_t, but from a different angle. In the financial crisis years, many banks were required to recognize unrealized holdings losses, due to the recognition of other-than-temporary impairment, even though they were still holding the underlying securities. This, essentially, shortens the holding time for these securities, potentially weakening any association between AUG(AUL)_t-1 and UGL_t. As the results show, both the magnitude and the statistical significance of the correlations between AUL_t-1 and UGL_t decreased significantly during the financial crisis years. The correlation between AUG_t-1 and UGL_t became more negative, although no longer significant at the 10% level, suggesting that more unrealized gains turn into unrealized losses during the crisis years.

To further assess the possibility that the negative correlation reported in Table 3 might be due to factors other than the accounting reason described in section 2, we conduct a placebo test in which we replace UGL with reported unrealized gains and losses from comprehensive income items other than AFS securities (including additional minimum pension liability adjustments, foreign currency translation gains and losses, and changes in the fair value of derivative instruments classified as cash flow hedges) as well as with the accumulated amount of these items of OCI. In contrast to UGL and lagged AUG and lagged AUL, the sum of these other items of OCI in the current year is not correlated with the accumulated total of the items in the prior year. Specifically, the coefficient on lagged non-AOCI is not significantly different from zero (0.02, with a t-statistic of 0.77). This is consistent with the common belief that these other comprehensive income items, which result from changes in market value of the underlying instruments, are transitory and unpredictable in nature.

4.2 Investment composition

As shown in Table 1, the distribution of the portion of AFS securities that are fixed income debt securities (i.e., FI) is clustered between 70 and 100%. Hence, to obtain a meaningful separation, we group banks based on their quintile ranks of FI. Each year, banks in the lowest quintile are designated as the Low-FI group. Banks in the highest quintile are included in the High-FI group. All remaining banks are included in the Medium-FI group. The results from regression (1) for each of these groups are included in Panel B of Table 3. As expected, as FI increases, the coefficients relating UGL to lagged AUG become more negative (−0.05, −0.07, and −0.09), the coefficients relating UGL to lagged AUL become more negative (−0.04, −0.22 and −0.39) and the differences between these coefficients (i.e., the asymmetry) increases (0.01, −0.15, and −0.30).

4.3 Explaining the negative autocorrelation in UGL and OCI

To examine the extent to which the negative correlation between UGL and lagged AUG and UGL and lagged AUL contributes to the previously documented negative correlation in UGL and OCI, we begin by adding lagged UGL to regression (1), and then we conduct the series of regression analyses (regressions 2a, b, c). The results are reported in Table 4.

Table 4 Explaining the negative autocorrelation in OCI

Panel A shows that, in the simple regression of UGL on lagged UGL, the estimate of the coefficient on lagged UGL is significantly negative (−0.11 with a t-statistic of −2.84). But, in the multiple regression, when we add lagged AUG and lagged AUL, the relation between UGL and lagged UGL is no longer statistically significant (−0.01 with a t-statistic of −0.23), but the estimates of the coefficients on lagged AUG and AUL remain statistically significantly negative.

We then confirm that, for our sample of banks, we observe negative serial correlation in OCI as in Jones and Smith (2011). As shown in the first column of Panel B, Table 4, OCI exhibits strong negative serial correlation (coefficient estimate of −0.15 with a t-statistic of −3.52), consistent with the finding of Jones and Smith (2011). Second, we regress OCI_t on AUG_t-1 and AUL_t-1, and show that OCI is significantly negatively correlated with AUG_t-1 and AUL_t-1, consistent with the finding for UGL. Third, we include lagged OCI and total accumulated other comprehensive income (AOCI, consisting of the accumulated gains and losses from foreign currency translation, pensions, and derivatives but excluding AUGL) in the regression. The result shows that the estimated coefficients on OCI_t-1 and AOCI_t-1 are not significantly different from zero, suggesting that the negative correlation with lagged AUG and lagged AUL explains the negative serial correlation in OCI for our bank sample.¹⁹

We also explicitly test the extent to which the negative serial correlation in OCI is due to the reclassification of available-for-sale gains and losses (i.e., “recycling”). Jones and Smith (2011) study the autocorrelation of the combined amount of UGL and reclassification (RECL). They conjecture that the negative autocorrelation they document might be due to reclassification. However, because of a lack of data on UGL and RECL, the issue was left unresolved; we examine UGL and RECL separately. The results are shown in Panel C. The estimate of the correlation between the RECL and lagged UGL is negative (−0.04 with t-statistic of −1.79). However, RECL has a significant positive, not negative, autocorrelation (correlation of 0.22 with a t-statistic of 2.96). The overall correlation between REGL and lagged OCI is not significantly different from zero (correlation of −0.06, with a t-statistic of −1.09). These findings suggest that reclassification is not the cause of the observed negative serial correlation in OCI.

4.4 Market pricing of the predictable component of UGL²⁰

To examine the market pricing of the predictable component of UGL, we first isolate a predictable component based on current accounting information. We do this via a regression of reported UGL on lagged AUG and lagged AUL within each of the FI groups.²¹ Then we apply the average estimated coefficients from each of the available past years to AUG and AUL in the current year to form a predicted value for next year UGL.²² We denote this predicted amount as PUGL_t+1. To avoid potential look-ahead bias in estimating model coefficients using data from banks with different fiscal year-ends, we restrict our analysis to December fiscal year-end banks in the remainder of the paper. This restriction affects less than 2% of the observations since the majority of banks have a December fiscal year-end.

We conduct three sets of analyses of the relation between PUGL and future returns: (1) we regress next period abnormal stock return on known risk factors as well as predicted next year UGL, based on banks’ currently reported AUG and AUL; (2) we examine the future returns from hedge-portfolios formed on the basis of PUGL; and (3) we examine the monthly excess returns on portfolios based on PUGL over and above the Fama and French (2015) and Carhart (1997) factor returns as well as estimated shocks to the yield curve (Viale et al. 2009).

4.4.1 Prediction of future stock returns

We regress one-year-ahead, bank-specific buy-and-hold stock returns (R_jt+1), minus the risk-free rate of return for the year, on PUGL for next year, together with common risk factors including size, book-to-market, CAPM-beta, profitability, growth and momentum as well as interest rate sensitivity and provision for loan losses.²³ That is,

$$ {R}_{jt+1}-{RF}_{t+1}={\alpha}_t+{\beta}_1{PUGL}_{jt+1}+{\beta}_2{BETA}_{jt}+{\beta}_3{BM}_{jt}+{\beta}_4{SIZE}_{jt}+{\beta}_5{ROE}_{jt}+{\beta}_6 GTA+{\beta}_7{MOMENTUM}_{jt}+{\beta}_8{GAP}_{jt}+{\beta}_9{PCL}_{jt}+{e}_t, $$

(3)

where BETA_jt is estimated via a regression of the difference between the monthly return on the stock of bank j and the corresponding return on three-month T-bills on the difference between the CRSP value-weighted market return and the corresponding risk-free rate for the 60 months prior to the end of the third month of fiscal year t, SIZE_jt is the logarithm of the market capitalization of bank j at the end of the third month after the fiscal year end t-1, the book-to-market ratio (BM_jt) is calculated as the book value of equity of bank j divided by the market value of equity of bank j at the end of the third month after fiscal year-end t-1, ROE_t (return on equity) and GTA_t (growth in total asset) are used to measures the profitability and investment growth of the firm (Fama and French 2015), and MOMENTUM_jt is the return on the equity of bank j for the year ending at the third month of fiscal year t. Following Viale et al. (2009), we use GAP_jt to estimate the interest rate sensitivity of net short-term assets, which is measured as the difference between short-term investments and short-term liabilities for bank j at the end of year t-1, deflated by total assets at the end of year t-1. PCL_jt is the provision for credit and loan losses for bank j as a percentage of the bank’s total interest income. A positive estimate of the coefficient on PUGL, that is, β_1, indicates that investors do not fully understand the predictability of UGL.

The results from the estimation of regression (3) are reported in Table 5. The first column of Panel A shows that, among the Fama and French (2015) and Carhart (1997) factors, BM and ROE show significant predictive power for our sample of banks. The second column shows that PCL, provision for credit and loan loss, also predicts future stock returns, but with a negative sign. The third column shows that the estimated coefficient on PUGL is 0.007, significant at the 1 % level. This suggests market mispricing. However, the results may also suggest that PUGL captures certain aspects of risk in banks operations. We address this suggestion next.

Table 5 Predictable component of UGL and future stock returns

Note that accumulated unrealized gains and losses reflect changes in interest rates; that is, AUGL captures changes in expected returns on the stocks and bonds held by the bank. If such changes in expected returns on AFS securities somehow affect the overall expected return on bank equity and debt, perhaps due to the fact that such holdings represent a significant portion of bank assets, we may expect AUGL to be correlated with the value of bank total assets and total equity as well as expected return on bank equity. Such an argument was first advanced to explain the negative correlation between SIZE and future stock return (Berk 1995).

In Panel B of Table 5, we include AUG and AUL in the regression where future stock returns is the dependent variable. If any relation between AUG(AUL) and future stock return is due to changes in expected return of AFS securities, we would expect SIZE, BM, ROE, GTA, and MOMENTUM to subsume the correlation between AUG(AUL) and future stock return, since change in equity value captures the overall effect on the expected return on bank equity due to changes in the expected return of the bank AFS security holdings. As shown in model 1, this is not the case. Consistent with market mispricing, both AUG and AUL have negative correlation with future stock returns (with t-statistics −1.02 and −3.31 respectively) when all common proxies for risk factors are included in the regression. In addition, since both AUG and AUL capture changes in the interest rates, we would expect them to have a similar relation with future stock return based on the change-in-expected-return argument. In contrast, our analysis predicts an asymmetric relation between AUG and AUL and future UGL, and it follows that the market mispricing notion would suggest an asymmetric relation with future stock returns after controlling for known risk factors; this asymmetric relation is seen in the results for Model 1 in Panel B.

In Model 2, we include PUGL, AUG, and AUL in the same return regression. With PUGL included in the regression, the estimated coefficients on AUG and AUL become insignificantly different from zero. This indicates that PUGL captures the predictive power of AUG and AUL with respect of future stock return.

4.4.2 The economic significance of the market mispricing

In Table 6, we assess the economic significance of market mispricing by documenting the average return on portfolios based on PUGL. Specifically, each year, banks are sorted into three equal-sized portfolios based on PUGL for the next year. We then calculate the portfolio average one-year-ahead, buy-and-hold return for each portfolio. Table 6 reports the average portfolio return for all three PUGL portfolios, over all sample years. As shown in the first column, the average annual return for the high-PUGL portfolio is 13.23%. The average return for the low-PUGL group is 7.98%. The difference, 5.25%, represents a significant return to a hedged portfolio that shorts the low-PUGL banks and goes long in the high-PUGL banks. The difference is even greater (5.77%) if we remove the crisis years from the analysis.

Table 6 Portfolio return analysis

To further tie the mispricing to the predictability of UGL, we repeat the portfolio return analysis for the three sub-sample groups based on the FI index. As discussed in Section 2, the level of predictability of UGL_t, based on AUGL_t-1, increases with the percentage of fixed-income investments. It follows that, if the predictability of UGL is the impetus behind the documented mispricing, we would expect to see more significant hedge portfolio returns as we move from the low-FI subsample to the high-FI subsample.

As shown in Panel A of Table 6, both the statistical significance and the magnitude of the hedge-portfolio return, increases as the percentage of fixed-income investment increases. For the low-FI group, the hedge portfolio return is 1.71%, which is not significant at the 10% level. In contrast, for the medium- and the high-FI groups of banks, the hedged-portfolio returns increase to a significant 5.60% and 7.97%, respectively.

The results are similar, but the hedge-portfolio returns are greater, when the crisis years are removed from the analyses. This is expected in light of the fact that, during the financial crisis, banks were forced to write down a significant portion of their investment holdings. Badertscher et al. (2014) document that the top U.S. commercial banks recorded roughly 5.5% of their 2008 total market capitalization as other-than-temporary impairments during the financial-crisis years, 2008 and 2009. Such recognition of OTTI, as discussed in Section 2, reduces the negative correlation between AUGL_t and UGL_t+1. As shown in Table 2, Panel C, during the crisis years, the average annual correlation between UGL_t and AUGL_t-1 is −0.03, which is not significantly different from zero, whereas, for the entire sample this correlation is −0.25 and significant at the 1 % level.

We also repeat the portfolio return analysis for large and small banks, defined as banks with market capitalization above versus below the sample median market capitalization. Untabulated results show that the average hedge portfolio return is significantly larger for large banks. This suggests that the excess hedge-portfolio return is more likely due to mispricing as opposed to trading/transactions costs, as these costs are, on average, lower for larger banks.²⁴

4.4.3 Controlling for the Fama, French, Carhart factors and shocks to the yield curve

As a final test of market mispricing, we examine the returns on portfolios based on PUGL after controlling for the Fama and French (2015) and Carhart (1997) factor returns. In addition, we also include estimates of shocks to the yield curve. Viale et al. (2009) show that these estimates of shocks to the yield curve, based on the ICAPM, perform better in capturing the time-series return variation in bank stocks. Specifically, each year, we divide banks into three, equal-size groups based on the magnitude of PUGL. We run the following regression of monthly portfolio returns on the risk factor returns:

$$ {\displaystyle \begin{array}{l}{R}_{pt}={\alpha}_0+{\beta}_1\left({R}_{Mt}-{RF}_t\right)+{\beta}_2{HML}_t+{\beta}_3{SMB}_t+{\beta}_4{RMW}_t\hfill \\ {}\kern1.44em +{\beta}_5{CMA}_t+{\beta}_6{UMD}_t+{\beta}_7 TB3{M}_t+{\beta}_8 GS10{Y}_t+{\beta}_9 CB10{Y}_t+{e}_5,\hfill \end{array}} $$

(4)

where R_pt is the return on a portfolio formed by going long stocks with high PUGL and short stocks with low PUGL. R_Mt is the monthly return on the CRSP value-weighted index. RF_t is the monthly return on a three-month T-bill. HML_t is the monthly return on a portfolio that is long in stocks with high book value-to-market value and short in stocks with low book-to-market. SMB_t is the monthly return on a portfolio that is long in small stocks and short in large stocks. And UMD_t is the monthly return on a portfolio that is long in stocks with high past returns and short in stocks with low past returns. RMW_t is the difference between the returns on portfolios of stocks with robust and weak profitability, and CMA_t is the difference between the returns on portfolios of the stocks of low and high investment (Fama and French 2015). The factor returns are obtained from Ken French’s data library.²⁵ TB3 M, GS10Y, and CB10Y are the residuals from a vector-autoregression using the three-month Treasury constant maturity, 10-year Treasury constant maturity, and 10-year Moody’s AAA corporate bond yields over the 10-year Treasury yield, respectively. These interest rate data are obtained from the Federal Reserve’s online FRED database.²⁶

Table 7 is a summary of the results from estimation of regression (4). Again, we see evidence of market mispricing. For example, for the subsample of observations that excludes the crisis years, the intercept is significantly positive (0.005), indicating a risk-adjusted return of 0.5% per month. The loadings on various risk factor returns change as the percentage of fixed-income securities change. The intercept is highly significant for the two-thirds of the observations that contain the most fixed-income debt securities.

Table 7 Predictable component of UGL and future stock returns

Collectively, our results presented in this section indicate that investors misinterpret the financial information regarding UGL, leading to mispricing of bank stocks.²⁷

4.5 Robustness tests

We performed an array of sensitivity tests to assess the robustness of our results, including (1) replacing the deflator (total assets) with the number of shares; (2) increasing the horizon from one year to three years in conducting the portfolio return analysis; (3) using the level of net interest income and non-interest income, as opposed to the change, in the return regressions; (4) replacing the cost of investment securities with the fair value of the securities in calculating the FI index. Our results are robust to these changes in test design.

5 Conclusions

The introduction of fair value accounting has significantly transformed our accounting system, which is primarily based on historical cost. Changes in the fair value of assets and liabilities, as reported in the Shareholders’ Equity Statement, or in the newly introduced Statement of Comprehensive Income, provide potentially highly value-relevant information to investors. Studies have provided ample evidence on investor pricing of such information. Chambers et al. (2007), for instance, document that investors price other comprehensive income items almost dollar-for-dollar, consistent with the transitory nature of fair value changes.

We show that a significant portion of OCI, namely the unrealized gains and losses (UGL) from available-for-sale securities (AFS), is non-transitory: a negative correlation between accumulated UGL in the current period and next period UGL is predictable and economically and statistically significant. This is caused by a mixture of fair value and historical cost accounting with respect to the measure of income from fixed-income securities: UGL are recognized based on fair values, whereas interest income is measured based on historical cost accounting. We show that the predictable component varies systematically with the amount of fixed-income investment and with the relative amount of unrealized gains versus unrealized losses. This predictable component seems to be overlooked by investors, leading to mispricing of bank stocks.

A potential policy implication of our findings is that a change to disclosure rules to mandate separation of the amortization-driven component of UGL from the remainder of UGL, which would reflect true holding gains and losses due to changes in market prices, may be useful to investors. These two components capture different economic forces and hence have different implications regarding banks’ future comprehensive income.

Appendices

Appendix 1: Variable defintions

AOCI::

Accumulated other comprehensive income items. This is the sum of derivative gains and losses (Compustat item AOCIDERGL), pension gains and losses (Compustat item AOCIPEN), foreign currency gains and losses (Compustat item RECTA), other (Compustat item AOCIOTHER), plus accumulated unrealized gains and losses from AFS securities.

AUGL::

Accumulated unrealized holding gains and losses from available-for-sale securities. Data are hand-collected from sample bank annual reports.

AUG (AUL)::

Accumulated unrealized holding gains (losses) from available-for-sale securities. Data are hand-collected from sample bank annual reports.

BETA::

CAPM beta estimated using 60 monthly return data up to the third month after fiscal year-end.

BM::

Book-to-market ratio, calculated as the book value of the bank (Compustat item CEQ) divided by the CRSP market capitalization of the bank at the end of the third month after fiscal year-end.

BOND::

Amortized cost of AFS securities invested in corporate bonds.

CB10Y: :

Residual from a vector-autoregression of the 10-year Moody’s AAA corporate bonds yield over the yield on 10-year Treasury securities, together with three-month Treasury bill and 10-year Treasury securities. Data are obtained from the Federal Reserve’s online FRED database.

CMA::

Return on a portfolio of stocks that is long in banks with high growth in total assets and short in stocks with low growth in total assets.

COSTAFS: :

Amortized cost of all AFS securities.

FI::

Estimate of the percentage of AFS securities invested in fixed-income securities.

GAP::

Difference between short-term investment (Compustat item IST) and short-term liabilities (Compustat item DLC), deflated by total assets (Compustat item AT).

GTA::

Annual growth in total assets.

GS10Y: :

Residual from a vector-autoregression of 10-year Treasury securities, together with three-month Treasury bill and 10-year Moody’s AAA corporate bonds. Data obtained from the Federal Reserve’s online FRED database.

HML::

Monthly return on a portfolio of stock, which is long in stock with a high book-to market ratio and short in stocks with a low book-to-market ratio.

MBS::

Amortized cost of AFS securities invested in mortgage-backed securities.

MOMENTUM: :

Return on the equity of the bank for the 12 months ending in the third month of the trailing fiscal year.

MUNI::

Amortized cost of AFS securities invested in Municipal obligations.

NI::

Net income (Compustat item IBCOM).

OCI::

Other comprehensive income items, which is the sum of foreign currency gains and losses (Compustat item CICURR), derivative gains and losses (Compustat item CIDERGL), pension gains and losses (Compustat item CIPEN), other (Compustat item CIOTHER), plus accumulated unrealized gains and losses from AFS securities.

OTTI::

Other than temporary impairments.

PCL::

Provision for credit and loan loss (Compustat item PCL), as a percentage of total interest income (Compustat item IDIT).

PUGL::

Predicted unexpected gains and losses.

RECL: :

Reclassified AFS security gains and losses.

RGL::

Realized gains and losses on AFS securities.

ROE::

Return on equity, used as a measure of profitability.

RF::

Rate of return on three-month treasury bills, measured over 12 months in regression (3) and over one month in regression (4).

R _j ::

One-year-ahead buy-and-hold returns on the stock of bank j.

R _M : :

Rate of return on the CRSP Value weighted index, measured over 12 months in regression (3) and over one month in regression (4).

R _p ::

Monthly return on a portfolio of stocks formed by going long stocks with high PUGL and short stocks with low PUGL.

RMW::

Monthly return on a portfolio of stocks that is long in stocks with high profitability and short in stocks with low profitability.

SIZE::

The logarithm of the equity market value on the last trading date in the third month after the fiscal year-end. Price and number of shares outstanding are obtained from the CRSP.

SMB::

Monthly return on a portfolio of stock, which is long in small stocks and short in stocks with a high market capitalization.

TA::

Total assets (Compustat item AT).

TB::

Amortized cost of AFS securities invested in Treasury bills.

TB3 M::

Residual from a vector-autoregression of three-month Treasury bill, together with 10-year treasury securities and 10-year Moody’s AAA corporate bonds. Data obtained from the Federal Reserve’s on-line FRED database.

UGL::

Un realized gains and losses on AFS debt securities.

UMD::

Return on a portfolio of stocks that is long in stocks with high returns over the past 12 months and short in stocks short in stocks with low past returns.

Appendix 2: Numerical example of the accounting mechanism leading to negative correlation between UGL_t and AUGL_t-1

$100 invested in AFS, three-year, 10% annual coupon debt security issued at par.

Scenario 1a: Discount rate decreases to 8% at end of year 1; bank holds the AFS debt security.

		Post SFAS 115				Pre SFAS 115
Date	Fair Value	AUGL	UGL	RGL	Interest	AUGL	RGL	Interest
12/31/×0	$100.00
12/31/×1	$103.57	$3.57	$3.57	$0.00	$10.00	$0.00	$0.00	$10.00
12/31/×2	$101.85	$1.85	($1.72)	$0.00	$10.00	$0.00	$0.00	$10.00
12/31/×3	$0.00	$0.00	($1.85)	$0.00	$10.00	$0.00	$0.00	$10.00

Scenario 1b: Discount rate decreases to 8% at end of year 1; bank sells the AFS debt security at the beginning of year 2.

			Post SFAS 115				Pre SFAS 115
Date	Fair Value	AUGL	UGL	RECL	RGL	Interest	AUGL	RGL	Interest
12/31/×0	$100.00
12/31/×1	$103.57	$3.57	$3.57	$0.00	$0.00	$10.00	$0.00	$0.00	$10.00
12/31/×2	$0.00	$0.00	0.00	-$3.57	$3.57	$0.00	$0.00	$3.57	$0.00
12/31/×3	$0.00	$0.00	$0.00	$0.00	$0.00	$0.00	$0.00	$0.00	$0.00

Scenario 2: Discount rate increases to 12% at end of year 1; bank holds the AFS debt security.

		Post SFAS 115				Pre SFAS 11528
Date	Fair Value	AUGL	UGL	RGL	Interest	AUGL	RGL	Interest
12/31/×0	$100.00
12/31/×1	$96.62	($3.38)	($3.38)	$0.00	$10.00	($3.38)	$0.00	$10.00
12/31/×2	$98.21	($1.79)	$1.59	$0.00	$10.00	($1.79)	$0.00	$10.00
12/31/×3	$0.00	$0.00	$1.79	$0.00	$10.00	$0.00	$0.00	$10.00