Usefulness of OCI

Is Comprehensive Income Useful?

Gary C. B i d d l e a * and Jong-Hag C h o i a , b *

“Hong Kong University of Science & Technology

“Seoul National University

Received November 2004; Accepted March 2006

Abstract

With the lnternational Accounting Standards Board (IASB) now considering the question, we examine 16 different definitions of income across three applications: information content, predictive ability and executive compensation contracting. Our results reveal that comprehensive income defined by FASB Statement 130 (NIIJ dominates both traditional net income (NI) and fully comprehensive income (NIb,,,,d) in explaining equity returns, but that NI dominates NI,,, and Nlh,odd in explaining executive compensation. These findings are strikingly consistent with prior lobbying positions. In predictive ability, no definition clearly dominates. When income co,npunent.y are considered, NIbnrad dominates in all three applications, thus lending support to the disclosure of comprehensive income components.

JEL Classifications: M41, G12, G31

Keywords: comprehensive income, information content, executive compensation, predictive ability

*We gratefully acknowledge comments received from participants at the Columbia University Burton Workshop, the American Accounting Association Annual Meetings, and Summer Symposium on Accounting Research, Hong Kong University of Science & Technology, especiaily Chul Park and Michelle Yetman. Fenny Cheng provided expert research assistance. Both authors gratefully acknowledge financial support from the Research Grants Council of Hong Kong. The second author gratefully acknowledges the financial support from the Institute of Management Research at the College of Business Administration of Seoul National University.

2 Gary C. Biddle and Jong-Hag Choi Journal of Contemporary Accounting & Economics Vol2, No 1 (June 2006) 1-32

1. Introduction

The International Accounting Standards Board is presently considering a proposal for firms to report fully comprehensive income.’ This study provides evidence on the usefulness of comprehensive income in three applications: information content, predictive ability and executive compensation contracting. We use data from the US, where a comprehensive income requirement was introduced in 1997. Following lengthy deliberation and by a split vote, the Financial Accounting Standards Board (FASB) issued Statement of Financial Accounting Standards No. 130 (SFAS 130), Reporting Comprehensive Income, requiring the disclosure of both net income and a more ‘comprehensive’ measure of income (NI13,) for fiscal years beginning after December 15, 1997. SFAS 130 NI,,, includes four items recorded as owners’ equity under previous FASB pronouncements: adjustments to unrealized gains and losses on available-for-sale marketabIe securities (SFAS I 15), foreign currency translation adjustments (SFAS 5 2 ) , minimum required pension liability adjustments (SFAS 87) and changes in the market values of certain futures contracts qualifying as hedges (SFAS 80).2

In perspective, these are only the latest rounds in a longstanding debate going back to at least the 1930’s.’ A review of this debate reveals that proponents of ‘comprehensive income’ have argued that a ‘clean surplus’ that reflects only earnings and (nct) dividends is internally consistent, less subject to manipulation and more in accordance with valuation t h e ~ r y . ~ Opponents have argued that income should abstract from transitory effects to reflect only ‘current operating performance.’ Prominent among them are managers who also have lobbied effectively to exclude from net income items over which they have lesser control.

This contest between the ‘all-inclusive (comprehensive)’ and ‘current operating performance’ views of income has also spawned considerable research, including several recent studies that examine specifically the usefulness of SFAS 130 disclosures. The evideence is mixed. For example, Hirst and Hopkins (1998) find in a controlled behavioral laboratory setting that the ability of analysts to estimate share values is enhanced by the disclosure of comprehensive income component^.^ However, Dhaliwal et al. ( 1999) conclude that “comprehensive income is no more strongly associated with returndmarket value or better predicts future cash flowshncome than net income.” This, they argue, calls into question the FASB’s requirement for SFAS 130 comprehensive income disclosures. 0’ Hanlon and Pope ( 1999) similarly find little evidence of incremental information content for comprehensive income items excluded from net income for a sample of UK companies.

I See International Accounting Standards Committee (I 999). International Accounting Standards Board

NI,,,, however, is not fully comprehensive, as would be a measure that explains the change in owners’

See Brief and Peasnell (I 996) and O’Hanlon and Pope (I 999).

(2002,2003).

equity net of dividends (Nlh,,J -see Section 2 below.

.I See, for example, Robinson (199 I ) . Association for Investment Management and Research (AIMR) (1993). Sutton and Johnson (1993). American Institute of Certified Public Accountants (1994), Beresford et A. (1996), Johnson et al. (1995) and Smith and Reither (1996).

Maines and McDaniel (2000) find similarly that nonprofessional investors utilize comprehensive income measures acquired in an experimental lab setting. Hirst et al. (2001) report consistent results for bank analysts.

Gary C. Biddle and Jong-Hag Choi Jorirnal of Contemporary Accounting & Economics Vol2, No I (June 2006) 1-32

3

This divergence of views and evidence regarding an issue of such central importance to accounting begs for explanation. In this study we propose that conceivably, both sides of the ‘all-inclusive’ versus ‘current operating performance’ debate could be correct. How? By positing that different definitions of income may be more ‘decision useful’in different applications. Such a proposition has been alluded to in previous studies and frameworks,6 but it has not heretofore been tested systematically. Its confirmation would help explain the conflicting views and help inform ongoing deliberations regarding appropriate comprehensive income disclosures.

We proceed by comparing the decision usefulness of alternate income definitions for identical firm-years in three applications: information content, executive compensation and prediction. We examine the performance of both income definitions (aggregated earnings) and income component sets (disaggregated income components). Data are obtained from SFAS 130 disclosures, which are attractive for several reasons. First, SFAS 130 disclosures allow 16 different definitions of income to be compared with ample data availability. Second, SFAS 130 is described by the FASB as a “first step in implementing the concept of comprehensive income” (FASB 1997, SFAS 130, para. 5, p. 2), with explicit provision for future modifications that could benefit from research findings. Third, prior research has reached differing conclusions regarding their usefulness.

Our findings for income de$nitions reveal that for information content, the definition of comprehensive income adopted by the FASB in SFAS 130 dominates both net income and fully comprehensive income. In fact, it is not beaten by any of 15 other definitions of income examined. Thus, in terms of information content, the FASB appears to have chosen wisely. For predictive ability, no income definition dominates clearly. For executive compensation contracting, net income, rather than comprehensive income is most decision relevant, consistent with the view of managers and prior contracting research (e.g., Holthausen and Watts, 2001). Thus, these results for income definitions reveal a pattern strikingly consistent with the divergent views of investors and managers, in that comprehensive income dominates for information content and net income for compensation contracting.

Our findings for income component sets reveal that fully comprehensive income provides the most decision usefulness for all three applications: information content, predictive ability and executive compensation contracting (except for predicting future NI,,,, where NI,,, exhibits the largest R2, but is insignificantly different from Nlbroad). Also, component sets provide generally greater decision usefulness than corresponding income definitions across applications. These findings, when combined, support two primary conclusions: ( 1 ) different definitions of income provide different decision usefulness in different applications; and (2) it is decision useful to disclose separately comprehensive income components. These conclusions lend support to the IASB’s (2003) proposal fo require the disclosure of comprehensive income and the view that “different components of comprehensive income are useful in different ways to users of the accounts” (IASB 2002, lo).

For example, Holthausen and Watts (2001, 16) ask “is a ‘better summary measure of performance’ one that more accurately measures permanent income, one that is more highly associated with value, or something completely different (e.g., one that measures the effect of the manager’s actions on firm value for contracting purposes)?’

4 Gary C. Biddle and Jong-Hag Choi Journal of Contemporary Accounting & Economics Vol2, N o I (June 2006) 1-32

The paper is organized as follows. Section 2 provides background on SFAS 130 and previous research findings. Section 3 develops our approach and tests. Section 4 describes the sample selection, variable measures and data characteristics. Section 5 presents our empirical results. Section 6 provides a summary and discussion.

2. Comprehensive Income: Debate and Evidence

The longstanding debate between proponents and opponents of comprehensive income is well documented in several previous studies, notably Kiger and Williams (l977), Robinson (1 99 1) and Brief and Peasnell (1 996). Since at least the 1930’s, advocates of ‘current operating performance’ have argued that decision makers are better served by income insulated from external and transitory effects.’ These arguments have originated from both valuation and contracting perspectives. Analysts and investor interests have argued that nonrecurring items impair the forecasting of future cash flowsR Managers have argued that they should not be held responsible for earnings components over which they have little control, a reasoning fundamental in contracting theory (Holthausen and Watts, 2001; Lambert, 2001). These arguments and associated lobbying have been instrumental in the passage of several FASB and Accounting Standards Board (ASB) standards that explicitly introduce ‘dirty surplus’ items to balance sheet owners’ equity and bypassing income. Notable among these are adjustments to unrealized gains and losses on available-for-sale marketable securities (SFAS 12 and 115, effective 1975 and 1993, respectively), foreign currency translation adjustments (SFAS 8 and 52, effective 1976 and 1982, respectively) and minimum required pension liability adjustments (SFAS 87, effective 1986). In the UK, similarly, members of the ASB dissented from restrictions on ‘dirty surplus’ items introduced in Financial Reporting Standards (FRS) 7 and 10 (O’Hanlon and Pope, 1999).

Proponents of ‘comprehensive income’ have countered that ‘dirty surplus’ items violate the articulation of the income statement and balance sheet and allow managers to manipulate earnings. Their arguments have also met with success as the FASB and its predecessor in the US, the Accounting Principles Board (APB), have periodically swung toward more comprehensive income measures. Examples include APB Opinion 9, effective lL966; Statement of Financial Accounting Concepts 5, issued 1984; and most recently, SFAS 130, effective 1997. In the UK, examples of pronouncements restricting ‘dirty surplus’ items include FRS 3 (issued 1992), FRS 7 (issued 1994), FRS 10 (issued 1997) and FRS 15 (issued 1999). Tellingly, arguments for more comprehensive income measures have originated primarily from the valuation (versus contracting) perspective.‘ For example,

’ Paton (1934), May (1937) and Littleton (1940). “If we want to maximize the information about value in the earnings figure and minimize noise, we can

choose accounting rules that make earnings look more like value and less like change in value. In other words, we can choose rules that minimize transitory components of earnings, while leaving the permanent components” (Black, 1993.5).

See Robinson (1991). AIMR (1993), Sutton and Johnson (1993), American Institute of Certified Public Accountants (1994), Beresford et at. (1996), Johnson et al. (1995), Smith and Reither (1996) and American Accounting Association Financial Accounting Standards Committee (Linsmeier et al., 1997a).

Gary C. Biddle and Jong-Hag Choi Journal of Contemporary Accounting & Economics Vol2, No I (June 2006) 1-32

5

the American Accounting Association Financial Accounting Standards Committee (Linsmeier et al., 1997b) argued that “an analyst’s forecast can be used to value a stock only if it is a forecast of comprehensive income and a price/earnings ratio only has a precise interpretation if the earnings are comprehensive ... For reported income to be most useful in equity price valuation, it must be comprehensive.” As noted by the FASB, the report of the AIMR ( I 993) was particularly influential in motivating the SFAS 130 comprehensive income requirement (SFAS 130, para. 40, p. 16).’O The IASB is presently working in conjunction with the ASB, which has had an active project on this issue since 2000, when it issued Financial Reporting Exposure Draft 22 advocating a comprehensive measure of income.

From the US experience, two noteworthy features of SFAS 130 are the controversy surrounding its passage and its work-in-process nature. As documented in SFAS 130 and beyond the advance lobbying above, the FASB received 28 1 comment letters on the Exposure Draft (ED) for SFAS 130 (issued June 1996) and 22 individuals and organizations presented their views at public hearings in November 1996. The FASB further discussed the ED with “constituents, the Financial Instruments Task Force and the Financial Accounting Advisory Council” (SFAS 130, para. 50, p. 19). Despite this considerable input, views among Board members remained split, leading to a 5-to-2 vote for passage. The two dissenting Board members, Messrs. Cope and Foster, objected that SFAS 130 “permits an enterprise to display the items of other comprehensive income [...I with less prominence and to characterize them differently from other items of comprehensive income that are currently included in net income” (SFAS 130, p. 9).

The work-in-process nature of SFAS 130 is evidenced by its partial implementation of SFAC 6. In SFAC 6 (issued 1984), the FASB had previously defined comprehensive income as “the change in equity [net assets] of a business enterprise during a period from the transactions and other events and circumstances from non-owner sources. It includes all changes in equity during a period except those resulting from investments by owners and distributions to owners” (SFAC 6, para. 70). In other words, the FASB had previously defined comprehensive income as ‘clean surplus’ income, to include all changes in owners’ equity except (net) dividends. SFAS 130 goes only part way towards this definition by retaining several ‘dirty surplus’ items (detailed below) and provides specifically that “future accounting standards will provide guidance on items that are to be included in comprehensive income and its components” (SFAS 130, para. 7, p. 3).

SFAS 130 includes in its definition of comprehensive income four owners’ equity adjustments explicitly excluded from net income in previous pronouncements: adjustments to unrealized gains and losses on available-for-sale marketable securities (SFAS 1 15), foreign currency translation adjustments (SFAS 52), minimum required pension liability adjustments (SFAS 87) and changes in the market values of certain futures contracts

“’Also influential, as noted in SFAS 130 (para. 41-45), was a meeting of the FASB with the Accounting Policy Committee of Robert Morris Associates (a credit rating agency now called Risk Management Association), a similar comprehensive income requirement by the ASB (effective 1992) and a then pending comprehensive income requirement by the International Accounting Standards Committee (International Accounting Standards I , 1998, esp. para. 87-88, p. 34). See also Wahlen et al. (2000).

6 Gary C. Biddle and Jong-Hag C h i Journal of Contemporary Accounting & Economics Vol2, No 1 (June 2006) 1-32

qualifying as hedges (SFAS 80). SFAS 130 omits from comprehensive income several additional ‘dirty surplus’ items, including gains or losses on foreign currency transactions designated as and effective as, economic hedges of a net investment in a foreign entity (SFAS 52); gains or losses on intercompany foreign currency transactions that are of a long-term investment nature and accounted for using the equity method (SFAS 52); and subsequent increases or decreases in the fair value of available-for-sale investment securities previously written down as impaired (see Beresford et al., 1996, 70). Below we examine empirically both comprehensive income as defined by SFAS 130 (NI,30) and a fully comprehensive measure that includes all changes to owners’ equity except (net) dividends (NIbroad).”

Several recent studies have examined specifically the usefulness of SFAS 130 disclosures. Hirst and Hopkins (1998) find in a behavioral lab setting that disclosure of the SFAS 130 marketable securities component helps buy-side equity analysts estimate more accurately the stock price of a company that upwardly manages its net income using the available-for-sale investment portfolio (see also discussant’s comments by Lipe (1998)). A follow up study of bank analysts’ risk and value judgments reaches a similar conclusion (Hirst et al., 2001). As Hirst and Hopkins (1998) correctly admit (p. 48), “[a] strict interpretation of the efficient markets hypothesis suggests that SFAS 130 constitutes a simple reformatting of the financial statements (i.e., there are no new recognition or measurement rules included in SFAS 130), [that] will have no effect on financial statement users’ judgments” (emphasis in the original). Nevertheless, they argue from a behavioral perspective that “analysts judgments will be affected by the clarity of disclosure of value- relevant information.”’2 Consistent with this prediction, Ahn and Choi (2002) find using Korean data that comprehensive income is incrementally value-relevant over net income. Choi and Das (2003) find that analysts use information contained in comprehensive income in updating their forecasts.

In contrast, Dhaliwal et al. (1999,43) conclude that, “[wlith the exception of financial firms, we find no evidence that comprehensive income is more strongly associated with retumdmarket value or better predicts future cash flows/income than net income. Moreover, the only component of comprehensive income that improves the association between income and returns is the marketable securities adjustment. Our results do not support the claim that comprehensive income is a better measure of firm performance than net incorr~e.”’~ O’Hanlon and Pope (1999,459) similarly find “little evidence that other flows excluded

As explained below, our empirical measure of NI,,, (NIhruad) also omits (includes) changes in the market values of certain futures contracts qualifying as hedges (SFAS 80) due to coding limitations in the Cornpustat database.

I 2 This argument, while plausible for individual decision makers, is problematic in markets where incentives exist to detect and arbitrage reiated valuation effects, particularly where this component is observable at trivial cost (by turning a page to the balance sheet). A more compelling ‘behavioral’ rationale for decision usefulness derives from the contracting perspective discussed below.

I 3 Even so, selected results in Dhaliwal et al. (1999) provide evidence of decision usefulness for information content. Specifically, on page 52 it is reported that “the adjusted R2 using NI,,, is larger than the adjusted R2 using net income and the difference [...I is significant at the 0.01 two-tailed level.” On page 54 it is reported that “[tlhe adjusted R2 using [comprehensive income defined as net income plus the marketable securities component of SFAS 1301 is larger than the adjusted R2 using net income and the difference I...] is significant at the 0.01 two- tailed level.” We augment these findings below.

Gary C. Biddle and Jong-Hag Choi 7 Journal of Contemporary Accounting & Economics Vol2, No I (June 2006) 1-32

from ordinary profit are value-relevant” for UK firms. Cahan et al. (2000) “find no evidence that [comprehensive income] items provide information that is incrementally relevant above comprehensive income” for New Zealand firms (see also discussant’s comments by O’Hanlon (2000)).

Recent controversies regarding pro-forma earnings can likewise be viewed as part of the larger debate regarding the comprehensiveness of income. Whereas managers have claimed that the narrower ‘pro forma earnings’ provide a better measure of operating results, financial commentators and accounting regulators have countered that pro forma earnings can misrepresent operating results. Once again, research evidence is mixed. For example, Bradshaw and Sloan (2002) and Brown and Sivakumar (2003) find greater market reactions around earnings announcements for pro-forma earnings than for net income. Lougee and Marquardt (2002) and Johnson and Schwartz (2001) find no significant differences. Brown and Sivakumar (2003) find pro-forma earnings more predictive whereas Doyle et al. (2002) find with a different design that items excluded from pro-forma earnings are themselves predictive. l 4

This continuing debate and conflicting evidence are in a sense surprising, in that they relate to an issue as fundamental to accounting as the comprehensiveness of income. In another sense, this debate is revealing. Over decades, investor and management interests have consistently held to differing views. Investors have advocated income measures that include all value-relevant components, with further distinction between permanent and transitory effects. Managers have consistently advocated less inclusive measures that exclude items over which they have lesser control. In this study we allow that these divergent views may be justified. How? By allowing that different measures of income may be more decision useful in different applications. Given the parties involved, three obvious applications to examine are information content, predictive ability and executive compensation contracting.

This notion that different measures of income should exhibit different decision usefulness in different applications also has been alluded to in previous research. For example, Holthausen and Watts (2001,48) assert that “the nature of dirty surplus can be explained by the existence of multiple roles for financial reporting.” Gaver and Gaver (1998, 252) express hope that “research [...I could provide additional insights into the process by which compensation is determined and the aspects in which the types of accounting information useful for compensation contracting differ from the accounting numbers used for valuation purposes.” However, no study heretofore has directly addressed this proposition. This study therefore offers the opportunity to assess whether compensation committees consider adjustments between net income and comprehensive income to be transitory components or noise that should not factor into management compensation. In SFAS 130, the FASB created a natural experiment by identifying components that in combination create 16 different measures of income that are here compared in terms of decision usefulness.

To enhance comparability, we assess decision usefulness using similar firm-years. To provide broader inferences, we further distinguish between income definitions and income

lJSee also Smith (2001) and Liesrnan and Weil (2001).

8 Gary C. Biddle and Jong-Hag Choi Journal of Contemporary Accounting & Economics Vol2, No I (June 2006) 1-32

component sets. By income definition we refer to the number obtained by algebraically summing all components of a given income measure. For example, net income equals the sum of component revenues and expenses. Introducing a summary measure into a regression model has the effect of constraining all of its components to have the same coefficient value. Analyzing income component sets separately allows each component in a regression model to have its own coefficient value and provides inferences regarding the decision usefulness of their separate disclosures.

3. Approach and Tests

Following from the above, this study compares for a similar firm-year sample the decision usefulness of alternative income definitions and component sets in three applications: information content, executive compensation contracting and prediction of future values of income and cash flows. These hypotheses (in alternate form) are:

Different definitions of income and /income component sets provide different decision usefulness for:

H,: Information content (equity returns).

H,: Predicting future values of net income, operating income and operating cash flows.

H,: Compensation contracting.

Combined, these results address the question of whether different income measures are more decision useful in different applications:

H,: More (less) comprehensive definitions of income are more decision useful for explaining equity returns (compensation contracting).

Finally, there is the question of whether it is decision useful to disclose comprehensive income components:

H,: It is decision useful to disclose comprehensive income components separately.

Relative association comparisons are used to address hypotheses H,, H, and H, (Biddle et al., 1995). Hypothesis H, is assessed by comparing the findings for H,, H, and H,. Hypothesis H, is assessed by comparing results for income definitions and component sets.15

As our focus is on the relative decision usefulness of income definitionkomponent sets, we do not report results for tests of incremental information content of individual components (see also Biddle et al. (1995)), except for Table 3 below.

Cavy C. Biddle and Jong-Hag Choi Joiirnal of Contenzporary Accounting C? Economics Vol2, No I (June 2006) I-32

9

Testing HI: We test hypothesis H, by regressing equity returns (raw and excess above a value-weighted market index) on the levels and lagged levels of different income definition/ component sets and conducting relative association comparisons. As detailed in Biddle et al. ( 1 993, this approach is econometrically equivalent to a ‘levels and changes’ formulation, but holds the advantage of allowing response coefficients to be observed directly. It further endogenously incorporates income forecast errors based on autoregressive time-series predictions.

Testing H,: We test hypothesis H, using relative association tests of the ability of different income definitiodcomponent sets to explain future net income, operating income, operating cash flows, NI,,, and NIbIUad. These measures are selected based on their use in previous studies of predictive ability (e.g., Dhaliwal et al., 1999; Brown and Sivakumar, 2003; Doyal et al., 2002) and their appearance in valuation models. We examine both one-year- ahead and two-year-ahead forecasts.

Testing H,: We test hypothesis H, using similar relative association tests of the ability of different income definition/component sets to explain chief executive compensation. Following the approach of Gaver and Gaver (1998) and Bushman et al. (2000), we focus on salary and bonus compensation.Ih Excluding compensation components based explicitly on share grants or changes in the values of owned shares or options reduces noise introduced by other factors affecting share values and provides a sharper focus on alternative income definitions that may be used in executive compensation contracting. Executive compensation data are obtained from the ExecuComp database.I7

Testing H,: Hypothesis H, is assessed by comparing qualitatively the findings for HI , H, and H,. It cannot be tested quantitatively using conventional association tests because it involves comparisons of other tests that themselves employ different dependent variables. A qualitative assessment is obtained by comparing rankings of relative decision usefulness of the different income definition/component sets for information content, compensation contracting and predictive ability, as informed by levels of statistical significance obtained for each.

Testing H,: Hypothesis H, is also assessed by comparing qualitatively the findings for H, , H, and H,. Specifically, it is decision useful to disclose sets of income components rather

l6 As observed by Bushman et al. (2000, 1 I ) , “[tlhe association between changes in cash compensation and earnings is well documented in prior research (see Lambert and Larcker (1987) and Sloan (!993)), among others.” Gaver and Gaver (1998, 235) “find that cash compensation is significantly positively related to above the line earnings, as long as results are positive. Compensation is shielded from the effects of above the line losses. Similarly, nonrecurring transactions that increase income flow through to compensation, but nonrecurring losses do not.” We employ similar partitions based on the signs of income components.

Previous studies that have considered associations between earnings components and executive compensation include Clinch and Magliolo (1993) (earnings components of bank holding companies), Dechow ( 1994) (differential potential for manipulation), Natarajan ( 1996) (cash versus accrual components in optimal contracting) and Baber et al. (1998) (persistent versus transitory components).


than only definitions if either association is enhanced or patterns of association differ in any or all of the application settings examined. An assessment of statistical significance can be provided within each application, with economic significance remaining judgmental.

4. Research Methodology

4.1 Sample selection

Examined in this study are firm-year observations with required Compustat and CRSP (and later, ExecuComp) data from the period 1994-98. All 48,649 firm-year observations from Compustat are originally retrieved. After deleting firm-years with missing earnings components and share returns data, 23,427 firm-year observations remain: 4,232 observations from 1994,4,607 from 1995, 4,768 from 1996,4,931 from 1997 and 4,889 from 1998. Below we restrict our sample to about 18,700 observations that have available future valuesI8 and 5,349 observations that have chief executive compensation data for the analyses on compensation contracting.

We examine 16 definitions of income including net income (NI), fully comprehensive income (NIbioad) and a11 combinations of SFAS 130 components in between. These latter combinations are compared within three groupings in which one, two, or all three SFAS 130 components (NJ3,J are added to NI. The three SFAS 130 components are: adjustments for unrealized gains and losses on available-for-sale marketable securities (SEC), foreign currency translations (FCT) and minimum required pension liabilities (PEN). We do not examine separately the fourth SFAS 130 component, changes in the market values of certain futures contracts qualifying as hedges, as it is not separately identified in Compustat. Rather, we combine this SFAS 130 component with any other ‘dirty surplus’ items (OTH) that explain the difference between the SFAS 130 definition of comprehensive income and the fully comprehensive measure (NIbmad). Thus, NIbroad = NI,,, + OTH = NI + SEC + FCT + PEN + OTH.

4.2 Variable measurement

Our income and returns variables are defined as in Dhaliwal et al. (1999), with only slight differences in notation and the addition of abnormal returns. All variables except equity returns and the future changes are scaled by beginning-of-period market value of equity.

NI Net income Bottom line net income after extraordinary items and discontinued operations, measured as Compustat data item # I 72.

The exact sample sizes for future value predictions are 18.7 14 for the prediction of future NI; 16,321 for future operating cash flow; 18,679 for future operating income; 18,701 for future NI,30; and 18,322 for future NIhroad. We also performed analyses for the subset of data with all five future variables available. Because the results are qualitatively similar, we do not separately report them.

SEC

FCT

PEN

NI,,,

NLad

OTH

CRR

CAR

Gary C. Biddle and Jong-Hag Choi I 1 Journal of Contemporary Accounting & Economics Vol2, No I (June 2006) 1-32

Adjustment for unrealized holding gains/losses on marketable securities Measured as the change in Compustat data item #238. Adjustment for foreign currency translation. Measured as the change of Compustat data item #230. Adjustment for pension liability Measured as the change in additional minimum pension liability in excess of unrecognized prior service costs (Compustat data item #297 - #298, if less than zero). Comprehensive income defined as SFAS 130 Measured as NI,3, = N1+ SEC + FCT + PEN. Comprehensive income defined broadly Fully comprehensive income measured as the change in retained earnings plus common stock dividends (change in Compustat data item #36 + #21). Other comprehensive income Measured as OTH = NIbrond - NII3". Cumulative raw return Equity returns (cum dividend) cumulated for 12 months, from eight months prior through three months after the fiscal year end. Cumulative abnormal return Cumulative abnormal return measured as the difference raw equity return and expected return, where expected return is estimated using the conventional market model:

R,l= a + bR,I,l + E ,

where RII = raw return to shares of firm i during time t R,,,l = value-weighted market index return during time t

Parameters a and b are estimated by using 100 days' actual data ending one year before the event year. s l is a random, mean-zero error term. Returns are cumulated for 12 months, from eight months prior through three months after the fiscal year end month (month 0). The Pearson (Spearman) correlation between CAR and CRR for the extended sample is 0.9133 (0.8980).19

Iy Because the results using CAR and CRR are not qualitatively different, we henceforth report only results using CAR. We also measure the CAR and CRR variables by cumulating returns for 15 months (CAR2. CRR2), starting from the beginning of year t (month -1 1) to three months after the end of the fiscal year (month +3) and 12 months (CARS, CRR3), from the beginning of the fiscal year (month - I 1) to the end of the fiscal year (month 0). Because all the empirical results are qualitatively similar regardless of the measurement period, we do not report the results based on these alternate returns measures. The Pearson correlation for the extended sample between CRR and CRRZ is 0.8807, between CRR and CRR3 is 0.729 I and between CRRZ and CRR3 is 0.8588, all significantly different from zero (p<O.OOI). The Pearson correlation between CAR and CAR2 is 0.8961, between CAR and CAR3 is 0.75 10 and between CAR2 and CAR3 is 0.8617, all significantly different from zero (p<0.001). The Pearson correlation between CRR and CAR is 0.9133, between CRR and CAR2 is 0.8949 and between CRR3 and CAR3 is 0.9236, all significantly different from zero (p<O.OOI).

12 Gary C. Biddle and Jong-Hag Choi Journal of Conteniporary Accounting & Economics Vol2, No I (June 2006) 1-32

CF Operating cash flow Measured as the level of cash flow per share scaled by the year t's beginning stock price. (Compustat data item #123 + #125).

Measured as the level of operating income per share scaled by the year t's beginning stock price. (Compustat data item #178).

0 1 Operating iricome

Chief executive compensation is measured as salary plus bonus, excluding shares and share options granted or owned. This follows the approach of Gaver and Gaver (1998) and Bushman et al. (2000) and the reasoning that: (1) cash compensation constitutes a material portion of executive compensation (Lambert and Larcker, 1987); (2) bonus plans typically make reference to earnings measures (Gaver and Gaver, 1998); and (3) cash compensation is likely to be more closely related to accounting measures compared with other forms of compensation (Sloan, 1993). Thus, cash compensation is likely to be more revealing of the utilization of alternative earnings measures in compensation contracting.

COMP Executive cash compensation Total cash compensation (cash salary plus cash bonus) of the chief executive officer from the ExecuComp database and following Gaver and Gaver (1998). scaled by equity value at the beginning of the fiscal year and multiplied by 1,000.

4.3 Data characteristics

In Table 1, Panel A, we report Pearson and Spearman rank correlations among NI, NI,,, and NIbroad.2" The Pearson (Spearman) correlation between NI and NI,,, is 0.9979 (0.9542); between NI and NIbrond, 0.6884 (0.8750); and between NI,,, and NIbrond, 0.6898 (0.9 152). These correlations are all highly significant (p<O.OO 1). Variable distribulions are described in Panel B of Table 1. Noticeable is the preponderance of zero values for the SFAS 130 components. The resulting close correspondence between NI, NI, 1o and PJIbrOad is evident in their similar mean values and high correlations noted in Panel A. It might also be noted that observations for PEN are strictly negative, as it refers to the recognition of minimum pension liability. Generally, the means of future observations of NI,+,, NI13,,, ,+, and N L d , t+l are greater than the means of NI, NI,,, and NIhroad. This is because a few extreme observations having very small values in year t do not appear in year t+l, possibly due to the bankruptcy or acquisition of the company. There is little difference in median values among them.

2o We also calculated complete correlations and distributions for the 5,349 firin-year observations with compensation data. Because the results are qualitatively similar, we do not report them separately.

Gary C. Biddle and Jong-Hag Choi 13 Journal of Contenzporary Accounting & Economics Vol2, No I (June 2006) 1-32

Table 1

Descriptive Statistics

Panel A: Pearson (Spearman) correlations among the three primary income measures ' I b

N1 1

NI,,, 0.9979 (0.9542) I

NIhrood 0.6884 (0.8750) 0.6898 (0.9152) 1

Panel B: Variable distribictiorts

Variable Mean Variance 1% 25 70 50 % 75 % 99 %

-0.0533

-0.0539

-0.05 I5

0.000 I

-0.0004

-0.0004

0.0024

0.0085 0.0173

0.0995 0.1287

0.0278 0.0 I32

1.3693

0.6822

0.6819

0.9677

0.0024

0.0004

0.0001

0.53 13

0.4064

0. I930

0. I238

0. I067

0.0932

0.3693

8.5789

-1.6214

- 1.7124

- 1.6355

-0.0777

-0.0275

-0.0052

-0.2097

-0.8525

-0.9584

-0.6754 -0.5987

-0.88 I7

- I .0208

0.0279

-0.0052

-0.0090

-0.0090

0 0

0

-0.0013

-0.324 I -0.006 I 0.01 19

0.0320

-0.007 1

-0.02 13

0.3282

0.0467

'0.0454

0.04 10

0

0 0

0

-0.0569

010621 0.0939

O.Il220

0.0589

0.0532

0.7697

0.0745

0.0742

0.0724

0

0

0

0

0.2082

0.1081

0.1810

0.2149

0.1084

0.1063

1.6306

0.2493

0.2830

0.3232

0.0786

0.02 I5 0

0. I296

2.1214

0.4834

0.9934

0.9742 0.5255

0.5852

9.6790

a Correlation significance levels not reported since all differ significantly from zero (p<O.O01). Distributions for NI through CAR are calculated by using 23,427 observations; for NI,+,, CF,+,, 01,+1, NI130,,+,, NIhr,,ad,,+,, approximately 18,700 observations; and COMP 5,349 observations.

NI NI,,, NIhrc,d SEC FCT PEN OTH CAR

NI,+, CF,+, OIf+, NIl,o,,+l =future comprehensive income defined by SFAS 130 (NI13,J at year t+l NIbrtud,t+l = future fully comprehensive income (NIhr<,.J at year t+l COMP = executive cash compensation

= net income (where all variables are scaled by beginning of fiscal year equity value) = comprehensive income defined by SFAS 130 = NI + SEC + FCT + PEN = fully comprehensive income = NI + SEC + FCT + PEN + OTH = change in the balance of unrealized gains and losses on marketable securities = change in the cuinulatjve foreign currency translation adjustment = change in additional minimum pension liability in excess of unrecognized prior service cost = other components of Comprehensive income that belong to NIb,t,ad but not to NI,,,, =cumulative abnormal return measured for 12 months, from month -8 to month +3 = future net income (NI) at year t c l =future cash flow at year t+l = future operating income at year t + l


5. Empirical Results

5.1 Information content (H,)

We test hypothesis HI by examining the relative information content of 16 definitions of income that are available in SFAS 130 disclosures. We employ the following regression model:

CAR, = a +b, “I, + Added Component(s),] + b, “I,., + Added Component(s),., It- E , (2)

where Added Components are SEC, FCT, PEN and OTH; a, b, and b, are estimated regression coefficients, E , is a random, mean-zero error term, time subscripts are omitted for expositional clarity; and the quantities in square brackets create (levels and lagged levels of) different definitions of income. The R2 from this equation is used to determine whether an income definition is better (or worse) in explaining contemporaneous stock returns than another when compared using an association test.

To compare the decision usefulness of income component sets, we also estimate a separate regression that allows the coefficients on each component to differ (where terms are as defined above):21

CAR, = a +b, NI, + b, NI,., + bi Added Component(s), + bj Added Component(sj,-, + E , (3)

The empirical results using equations (2) and (3) are reported in Panels A and €3 of Table 2, respectively.22 Each cell in the second column of Table 2 shows adjusted R2 when the added components are those in the first column. We perform statistical tests to compare these R2 against four benchmarks. The first comparison is against NI, with the corresponding value of the Biddle et al. (1995) (hereafter BSS) Wald statistic reported in the third column of the table.23 The second comparison is against NI,,,, with the corresponding BSS statistic reported in the fourth column of the table. The third comparison is against NIbroad, with the corresponding BSS statistic reported in the fifth column of the table. The last comparison is against the income definition that provides the largest adjusted R2 among models having the same number of added components. The resulting BSS Wald statistic is reported in the rightmost column of the table.24 In the table, even though the BSS test is non-directional, we add a “+” or “-” sign to help the readers understand the results more easily. For example,

2 1 It is reasonable to conduct these tests using data prior to the effective date of SFAS 130, since these components were required disclosures under prior FASB pronouncements (listed above).

22 The empirical analyses are performed after removing outliers having an absolute value of the standardized error greater than 3. Thus, the sample size of each regression is slightly smaller than 23,427. Using different cutoff values (2, 2.5,3.5 and 4 standard errors) does not qualitatively affect the results. In addition, all the results are corrected for the heteroskedasticity by using White’s (1980) method.

23 The distribution of BSS statistic is approximately 9. 24 We also performed Vuong (1989) tests to examine the R* differences among models. Results are qualitatively

similar.

Gary C. Biddle and Jong-Hag Choi Journal of Conteniporary Accounting C? Economics Vol2, No 1 (June 2006) 1-32

15

Table 2

Relative Information Content of Alternative Earnings Measures (H,)

CAR,= a + b, “I, + Added Component(s)l] + b, “I,., + Added Cornponent(s),.,]+ E ,

CAR, = a + b, NI, + b,NI,., + b, Added Component(s), + bjAdded Component(s), , + E ,

Panel A: Analyses with income dejinition (equation 2)

Biddle et al. (1995) Wald statistics for the comnarison against:

NI NI,, NIbrwd Best model with

(Adj. R2 (Adj. R2 (Adj. R2 the same number

Added Component(s) Adj. R’ =0.0296) = 0.0304) = 0.0253) of added components

None (=NU 0.0296 NA -9.54*** 1.84 NA

SEC

FCT

PEN

OTH

0.0303

0.0296

0.0297

0.0246

SEC, FCT

SEC, PEN

SEC, OTH FCT, PEN

FCT, OTH

PEN, OTH

0.0303 0.0304

0.0252

0.0297

0.0247

0.0247

1 1.24***

0.05 15.29***

-5.94**

8.50*** 12.26***

-2.00

0.13

-6.25**

-5.50**

-0.10

-12.22***

-8.48*** -9.19***

- 15.00***

-0.07

-5.28**

- 1 1.21***

- 1 0.29***

-8.80***

5.82**

1.60

2.08

-13.66***

5.16** 6.26**

-2.85*

1.82

-12.06***

-12.58***

NA

-10.44***

-10.38***

-10.41 *** -0.67

NA

-6.35**

- 10.44***

- 1 1.33***

- 1 0.39***

0.0304 9.54*** NA 5.56** NA SEC, FCT, PEN

( = NI,30) SEC, FCT, OTH 0.0253 -2.10 -5.96** 5.59** -5.96** SEC, PEN, OTH 0.0253 - I .75 -4.92** 0.83 -4.92*** FCT, PEN, OTH 0.0247 -5.79** -9.86*** - 1,1.13*** -9.86***

-1.84 -5.56** NA SEC, FCT, PEN, OTH

0.0253 ( = NLad )

NA


Table 2 (Cont.)

Panel B: Analyses with income components (equation 3)

Biddle et al. 11995) Wald statistics for comDarison against:

NI N L " NIbcad Best model with (Adj. Rz (Adj. R' (Adj. R' the same number

Added Component(s) Adj. R' =0.0296) = 0.0316) = 0.0335) of added components

None (=NI) 0.0296 NA -5.62** -15.51*** NA

SEC

FCT

PEN

OTH

0.0298 0.24 -5.44** -15.32*** -4.15**

0.0305 0.05 -5.12** -1 0.55*** -0.12

0.0298 0.35 -5.25** - 1 4.48*** -4.08** 0.0305 4.04** -S.lO** -10.39*** NA

SEC, FCT 0.03 I5 5.49** -0.42 -7.9 I *** -1.84

SEC, PEN 0.0296 0.41 -5.12** -14,89*** -6.62*** SEC, OTH 0.03 17 5.91*** 0.28 -6.82** -0.84 FCT, PEN 0.0306 4.1 1 ** -4.98** -9.87*** -5.46** FCT, OTH 0.0322 7.24*** 2.51 -5.94** NA

PEN, OTH 0.0307 4.93** -4.52** -9.52*** -4.49**

0.03 I6 5.62** NA -7.67*** -5.99** SEC, FCT, PEN

( = NI,,, )

SEC, FCT, OTH 0.0333 10.92*** 8.79*** -5.n8** NA

SEC, PEN, OTH 0.03 I9 I1.25*** 0.92 -10.45*** -5.92**

FCT. PEN. OTH 0.0324 4.95** 1.86 -4.79 -1.81

15.51*** 7.67*** NA SEC, FCT, PEN, OTH

0.0335 ( = NI,,,,, )

NA

Adjusted R2 statistics for returns-earnings equations (2) and (3). The dependent variable is CAR. The predictors are the levels and the lagged levels of alternative income definitions defined by progressively adding one, two, three or all four comprehensive income components to net income (NI). The analyses are performed after removing outliers with absolute values of standardized errors greater than three. The adjusted R? of each model is compared with those of four benchmark models, with the values of Biddle et al. (1995) statistics reported in the corresponding rowskolumns. The second column reports the adjusted R2 of each regression model. The Biddle et al. (1995) statistics reported in the third column is a comparison against NI, the statistic in the fourth column is a comparison against NI,,,, the statistic in the fifth column is a comparison against NI,,, and the statistic in the rightmost column is a comparison against the model with the largest adjusted R2 among the grouped models with equal numbers of added components. The symbol '***' (**, *) represents that the result is significant at the I (5, 10) percent level. Variables are as defined in Table 1 .

Gary C. Biddle and Jong-Hag Choi 17

Joirnial of Coriteinporary Accounting & Economics Vol2, No I (June 2006) 1-32

when we compare the model where none of the other components are added to the NI, the R2 is 0.0296 and it is significantly smaller than that of the NI,,, model (0.0304) as indicated by the BSS statistic of -9.54 in the fourth column of Panel A of Table 2.

The results in Panel A of Table 2 reveal that: ( I ) contrary to the conclusion of Dhaliwal et al. (1999), the definition of comprehensive income chosen by the FASB in SFAS 130 (NI,,”) exhibits greater information content for equity returns than NI and NIbrna,; and (2) NI and NIblOd, are indistinguishable (at conventional levels of statistical significance).

Among income definitions that add one SFAS 130 component to NI, the results in Table 2 reveal that: (3) the addition of SEC provides the greatest information content (statistically significant at conventional levels against FCT, PEN and OTH), consistent with Dhaliwal et al. (1999); (4) the addition of SEC to NI also provides an income definition with information content greater than NI and NIbroad but not NI,,,, also consistent with Dhaliwal et al. ( 1 999); (5) the addition of FCT, PEN, or OTH to NI yields information content smaller than NI,,,, (all significant at conventional levels); and (6) the addition of OTH to NI yields information content smaller than NIblndd (significant at conventional levels).

Among income definitions that add two SFAS 130 components to NI, Table 2 reveals that: (7) the additions of SEC and PEN provides the greatest information content (statistically significant at conventional levels above all the other combinations, except for the additions of SEC and FCT, against which it is indistinguishable); (8) the additions of SEC and FCT and SEC and PEN, provide income definitions with information content greater than N1 and Nlhrmd (significant at conventional levels); (9) Except the addition of SEC and PEN, all the other combinations yield information content smaller than NI,,, (all significant at conventional levels); ( lo) the additions of SEC and OTH, FCT and OTH and PEN and OTH, yield information content smaller than NIbroJd (significant at conventional levels).

Among income definitions that add three SFAS 130 components to NI, Table 2 reveals that: (1 1) the addition of SEC, FCT and PEN = NI,,, provides the greatest information content than all other combinations (all significant at conventional levels); (1 2) the addition of SEC, FCT and OTH is better than NIbro, but indistinguishable from NI; (13) the addition of SEC, PEN and OTH is indistinguishable from both NI and NIbro,; (14) the addition of FCT, PEN and OTH yields an income definition with information content smaller than both NI and NIbroad (significant at conventional levels).

Considered together, the findings in Panel A of Table 2 reveal in one view the information contents of all definitions of income available from SFAS 130 disclosures. They suggest that based on information content, the FASB chose its definition wisely.Among 16 different definitions of income, NI,,, wins. Additional industry-by-industry analyses (not reported) show that NI,,, also wins in information content for three industry groupings of financial, manufacturing and other firms.25

2s While i t may be observed that adjusted R’s in Table 2 are small, they are comparable to those observed in other related studies (e.g., Brown and Sivakumar, 2003) for pooled samples. When analyses are performed industry-by-industry and year-by-year, adjusted R’s increase dramatically. For example, the average R? of NI,,, in industry-by-industry and year-by-year analyses (total 15 combinations = three industries x five years) is 0. 05 I3 (and 0.0599 for CRR). If we remove additional outliers, adjusted R’s also increase; for example, removing observations with absolute values of standardized errors greater than 2 (versus 3) increases the adjusted R’ for NI,,, for the pooled sample to 0.0465 (from 0.0304). Results are otherwise qualitatively similar.


Panel B of Table 2 reports adjusted R2s when we allow each component to have its own coefficient using equation (3). The results reveal that: (1) contrary to the findings based on income definitions above, fully comprehensive income Nlbn,ad now exhibits greater information content than’both NI and NII3,; and (2) NI,,, again dominates NI. Thus, the separate consideration of comprehensive income components yields inferences that differ from their combined definitions.

Among income component sets that add one SFAS I30 component to NI, the results in Table 2 reveal that: (3) the addition of OTH provides the greatest information content (statistically significant at conventional levels against SEC and PEN); (4) the addition of any one component to NI yields information content smaller than NI,,, and NIbrDBd; (5) the addition of FCT or OTH to NI yields information content greater than NI. Thus, these findings also differ from the definitional setting (equation (2)), where SEC added the most explanatory power.

Among income component sets that add two SFAS 130 components to NI, Table 2 reveals that: (6) the additions of FCT and OTH provides the greatest information content (statistically significant at conventional levels above the additions of SEC and PEN, FCT and PEN and PEN and OTH); (7) all the combinations of the components provide information content greater than NI except the addition of SEC and PEN; (8) the addition of SEC and PEN, FCT and PEN and PEN and OTH yields information content smaller than NI,,,; (9) all the combinations of the components provide information content smaller than NIhroad (significant at conventional levels).

Among income component sets that add three SFAS 130 components to NI, Table 2 reveals that: (10) the addition of SEC, FCT and OTH provides greater information content than other combinations (significant at conventional levels except against FCT, PEN and OTH); ( 1 1 ) all the combinations provide higher information content than NI; (12) the addition of SEC, PEN and OTH and FCT, PEN and OTH are indistinguishable from NI,,, but worse than NIbrosd; and (1 3) the adjusted RZs are generally larger for income component sets than for corresponding income definitions ( 13 of 15 cases).

Taken together, the results reported in Table 2 suggest that the FASB chose wisely its definition of comprehensive income in SFAS 130. Among 16 definitions of income, NI,,, exhibits the greatest information content. However, when income component sets are considered, the rankings change, with fully comprehensive income exhibiting the greatest information content and other rankings also changing. This divergence of findings for income definitions versus component sets and the generally larger adjusted R2s for income component sets, lend support to hypothesis H, and to the decision usefulness of disclosing income components.26

To provide additional insights, Table 3 presents full results for equation (3) applied to net income (NI) and to the components of fully comprehensive income, NIbroad, for the pooled sample and for industry subsamples. Table 3 also compares the adjusted R2s obtained

26 We also perform the same tests after restricting our samples to those used in the predictive ability tests of N1 ( 1 8,714 observations). When the income definitions are tested, the R’ of NI model is 0.0261, that of NI,,, is 0.0272 and that of NI,,, is 0.0241. When the income component sets are considered, the R* of the NI model is 0.0261, that of NI,,,, is 0.0298 and that of NIhrwd is 0.0315. In summary, all the results are consistent with those reported in Table 2.

Gary C. Biddle and Jong-Hag Choi Journal of Contenzporary Accounting & Economics Vol2, No I (June 2006) 1-32

19

from equation (3) with corresponding adjusted R2s obtained from equation (2) for NI only. These are shown in the rightmost column by the percentage decrease of the sum of squared error (SSE) for each regression and statistical significance of the change of the explanatory power (F value). The results reveal that: ( I ) all income components exhibit incremental information content significant at conventional levels across all samples;27 (2) adjusted R2s increase significantly for component sets when compared with the corresponding NI only model, especially for the financial industry, where the sum of squared error decreases by 2.42%;28 and (3) adjusted R2s are larger for all the three industry subsamples than for the pooled sample as evidenced by the significant F-statistics reported in the rightmost column.29 As this study focuses on comparing the decision usefulness of income definitions and components for information content, predictive ability and executive compensation, it does not consider in detail the valuation and contracting implications of individual comprehensive income components. For an excellent survey of this literature, see Wahlen et al. (2000).

5.2 Predictive ability (H, )

Hypothesis H, is tested by comparing the ability of different income definition/ component sets to predict future levels in five series: net income, operating cash flows, operating income, NI,,, and fully comprehensive income. In this regard, our tests parallel those of Brown and Sivakumar (2003), Dhaliwal et al. (1999) and Doyle et al. (2002). We employ equations (4) and (5) for income definitionkomponent sets, respectively:

27 A recent study by Louis (2003) finds foreign currency translation adjustments to be negatively associated with equity returns, inconsistent with the positive sign observed in Table 3. Additional year-by-year analyses reveal that the positive sign of FCT in Table 3 for the pooled sample and manufacturing and other subsamples is driven by 1997 data. For 1994 and 1995, results (not reported) show a negative and significant coefficient for FCT for all samples. For 1996 and 1998, it is slightly positive but insignificant. Correspondingly, the lagged FCT coefficient is only positive and significant in 1998, while all results prior to 1998 show negative signs (among them, only 1994’s is significant.) Thus, the lack of a negative significant coefficient for FCT for the pooled sample and manufacturing and other subsamples and significant positive coefficient for lagged FCT appear to be driven by 1997 data. I t happens that 1997 contained a major financial crisis (which started from late 1996 and ended in 1998) that produced a dramatic strengthening of the US dollar. Among 202 non-zero observations of FCT in 1997, 182 are negative, consistent with sample firms suffering foreign currency translation losses. Thus, a positive coefficient for FCT may reflect foreign currency losses impounded in the stock prices of US firms. Financial firms may have been less affected due to hedging activities or lower net foreign investments.

*‘Alternatively, the adjusted RZ increase by 54.3% from 0.029 I to 0.0449 for the financial industry subsample. 2y Dhaliwal et al. (1999) do not examine incremental information content directly, but their relative comparisons

for income definitions that add one component to NI suggest incremental information content for SEC for financial firms only. O’Hanlon and Pope (1999) do examine incremental information content, finding no evidence for the comprehensive income components of UK firms. The results in Table 3 reveal incremental information content for all components for all samples. One explanation for the different results is that Dhaliwal et al. (1999) and O’Hanlon and Pope (1999) examine different time periods and smaller samples ( I 1,425 and 3,160, respectively, versus the 23,427 firm years used in this study).

Tabl

e 3

Incr

emen

tal I

nfor

mat

ion

Con

tent

of C

ompr

ehen

sive

Inco

me

Com

pone

nts

CA

R, =

a+

b,N

l, +

b2N

Ir.,+

b,SE

C,+

b,S

EC,.,

+ b

,FC

T, +

b,FC

T~.,+

b,PE

N,+

b,PE

N,,+

b,O

TH,

+ b,

,OTH

,.,+

E

,

Sam

ple

Con

stan

t b,

b2

b,

b,

b5

b6

b,

b"

b,

(3)

NI

SEC

FC

T

PEN

O

TH

b,,,

N

Adj

.R*

% S

SE

decr

ease

(F

val

ue)

Tota

l -0

.027

9 0.

I 13

7 0.

01 80

23

,139

0.

0296

N

A

(-9.

67**

*) (8

.30*

**)

(1.1

7)

Tota

l -0

,026

8 0.

1235

0.

0078

0.

3396

0.

0909

0.

0356

1.

0419

0.

7702

-0

.167

8 0.

0337

0.

0068

23,

139

0.03

35

0.55

%

(-9.

20**

*) (8

.21*

**)

(0.4

4)

(4.2

9***

) (1

.23)

(0

.15)

(4

.12*

**)

(5.0

7***

) (-

0.53

) (3

.29*

**)

(1.1

5)

(16.

00**

*)

Fina

ncia

l 0.

0818

0.

1365

0.

0705

0.

4067

0.

0562

-2

.547

8 -1

.766

7 6.

9420

-0

.796

5 0.

0623

0.

0074

5,

198

0.04

49

2.42

%

(14.

79**

*) (5

.81*

**)

(1.8

1*)

(3.7

7***

) (.7

9)

(-2.

29**

) (-

1.30

) (2

.58*

**)

(-2.

28**

) (9

.89*

**)

(1.5

7)

(15.

77""

")

Man

ufac

- -0

.054

4 0.

3154

-0

.108

9 0.

9788

-0

.360

3 0.

8927

1.

5057

0.

8785

-0

.105

8 0.

0702

0.

0487

9,

295

0.05

08

0.40

%

turi

ng

(-10

.98*

**) (

9.47

***)

(-2

.89*

**) (

3.12

***)

(-0.

73)

(I .8

1*)

(4.2

3***

) (4

.96*

**)

(-0.

45)

(1.6

2)

(0.9

8)

(4.6

0***

)

Oth

ers

-0.0

618

0.07

87

0,00

10

0.23

79

0.19

08

-0.0

235

0.52

59

3.30

98

-0.9

263

0.01

20

-0.0

223

8,64

3 0.

0271

0.

31 %

(-12

.86*

**) (

5.28

***)

(0

.07)

(4

.30*

**)

(0.6

0)

(-0.

09)

(2.5

6**)

(1

.42)

(-

0.71

) (2

.77*

**)(

-0.8

7)

(3.3

1***

)

Incr

emen

tal

info

rmat

ion

cont

ent f

or c

ompr

ehen

sive

inco

me

com

pone

nts f

rom

equ

atio

n (3

). Th

e de

pend

ent v

aria

ble

is C

AR

. Th

e pr

edic

tors

are

the

leve

ls a

nd th

e la

gged

le

vels

of c

ompr

ehen

sive

inco

me

com

pone

nts.

The

anal

yses

are

perfo

rmed

afte

r rem

ovin

g ou

tlier

s with

abs

olut

e va

lues

of s

tand

ardi

zed

erro

rs g

reat

er th

an 3

. A

ll re

sults

are

he

tero

sked

astic

ity c

orre

cted

usi

ng W

hite

's (1

980)

met

hod.

Rep

orte

d ar

e pr

edic

tor e

stim

ates

and

asso

ciat

ed t-

stat

istic

s ag

ains

t a n

ull o

f zer

o. T

he re

lativ

e de

crea

se o

f the

sum

of

squa

red

erro

r (SS

E) fo

r the

regr

essi

on m

odel

ver

sus

the

corr

espo

ndin

g sa

mpl

e us

ing

equa

tion

(2) f

or N

I onl

y is

show

n in

the

righ

tmos

t co

lum

n, w

ith

F st

atis

tics

re

port

ed b

elow

eac

h in

par

enth

eaes

. Th

e sy

mbo

l '**

*' (*

*, *

) re

pres

ents

that

the

res

ult

is s

igni

fica

nt a

t the

1 (5

, 10)

perc

ent l

evel

. V

aria

bles

are

as d

efin

ed in

Ta

ble

I.

3

3 E

4


Journal of Contemporary Accounting & Economics Vol2, No I (June 2006) 1-32

Future value = a + b, [NIl +Added Component(s),] + b, [NZ,.,+

Future value = a + b, NI, + b, NI,., + bjAdded Component(s), + Added Component(s),-,]+ E, (4)

bjAdded Component(s),-, + s1 ( 5 )

where Future value = {NIt+l, CFl+,, Oil+,, N1130,1+1, NIbroad,t+l) and terms are as defined in Table 1. Selected results from the 16 different income measures are shown in Table 4. The adjusted R2 of each model in Table 4 is compared with those of the other models, with the values of Biddle et al. (1995) statistics reported in parentheses.The leftmost statistic in parenthesis is a comparison against NI, the second statistic is a comparison against NI,,, and the rightmost statistic is a comparison against NIbruad.

Table 4

Relative Decision Relevance of Alternative Earnings Measures for Future Prediction (H,)

Future value = a + b, IN/, +Added Component(s),l +b, "I,., +Added Cornponent(s),.,]+ s,

Future value = a + b,N/, + b,NI,., + bfAdded Component(s), +b, Added Component(s),., + B,

(4) (5)

Panel A: Dependent variable = Future net income (NI )

Added Component(s) Definition (equation 4) Components (equation 5)

NONE ( = NI) 0.0998 0.0998 (NA, 4.00**, -10.48***) (NA, -1.04, -2.94*)

SEC, FCT, PEN 0.0961 0.1006

( = NI,,, 1 (-4.00**,NA,-9.25***) (1.04,NA,-1.94)

SEC, FCT, PEN, OTH 0.1039 0.1261

( = NIhmdd ) (10.48***, 9.25***, NA) (2.94*, 1.94, NA)

Panel B: Dependent variable = Future operating cash j low (CF)


NONE ( = NI) 0.0354 0.0354

(NA, 0.93,6.80***) (NA, -6.62**, -6.93***)

SEC, FCT, PEN 0.0347 0.0389 ( = NI,,,) (-0.93, NA, 6.20**) (-6.62**, NA, -0.48)

SEC, FCT, PEN, OTH 0.0257 0.0398 (-6.80***, -6.20**, NA) (-6.93***, -0.48, NA) ( = "hmd


Table 4 (Cont.)

Panel C: Dependent variable = Future operating income (01)


NONE ( = NI) 0.0668 0.0668 (NA, -0.98, -0.20) (NA, -2.74*, -5.67**)

SEC, FCT, PEN 0.0679 0.0707

( = NI,," ) (0.098, NA, -0.08) (2.74". NA, -3.17*)

SEC, FCT, PEN, OTH 0.0845 0.0922

( = NIhmd ) (0.20, 0.08, NA) (5.67**, 3.17*, NA)

Panel D: Dependent variable = Future comprehensive income (NI , TJ

Added Component(s) Definition (equation 4) Components (equation 5 )

NONE ( = NI) 0.0997 0.0997

(NA, -16.15***, -6.05**) (NA, -28.68***, -14.74***)

SEC, FCT, PEN 0.1551 0.1564

( = NI,," ) (16.15***, NA, 11.09***) (28.68***, NA, 1.00)

SEC, FCT, PEN, OTH 0.1242 0. I320

( = N L & J ) (6.05**, - 1 1.09***, NA) (14.74***, -1.00, NA)

Panel E: Dependent variable = Future fully comprehensive income (NIb,,,J


NONE ( = NI) 0.0457 0.0457

(NA, -0.01, -5.07**) (NA, -2.95*, -37.54***)

SEC, FCT, PEN 0.0465 0.0485

( = NI,,, ) (0.01, NA, -5.31**) (2.95*, NA, -50.84***)

SEC, FCT, PEN, OTH 0.05 19 0.0557

( = "broad ) (5.07**, 5.31**, NA) (37.54***, 50.84***, NA)

Adjusted R2 statistics for future prediction equations (4) and (5). The dependent variable is either future net income (N1: Panel A), future cash flow (CF: Panel B), future operating income (01: Panel C), future Comprehensive income defined by SFAS 130 (Nl,,": Panel D) and future fully comprehensive income (NIhrodd: Panel E) The predictors are the levels and the lagged levels of alternative income definitions defined by progressively adding three (NIlso) or all four (NIhrodd) comprehensive income components to net income (NI). The analyses are performed after removing outliers with absolute values of standardized errors greater than 3. The adjusted R2 of each model is compared with those of the other models, with the values of Biddle et al. (1995) statistics reported in parentheses. The leftmost statistic is a comparison against NI, the second statistic is a comparison against Nl,,, and the rightmost statistic is a comparison against NIhruud. Variables are as defined in Table I .

Gary C. Biddle and Jong-Hag Choi Journal of Contenzporaty Accounting C? Economics Vol2, No 1 (June 2006) 1-32

23

Results for income dejnifion equation (4) are reported in the second column of Table 4. The results for predicting future N1 reported in Panel A reveal that: (1) NI model provides the greater predictive ability than NI,,,; ( 2 ) NIbroad provides the greatest predictive ability among the three. The results for predicting future operating cash flow are reported in Panel B. The results reveal that: (3) There is no statistically significant difference between NI and NI,,,; (4) Both NI and NI,,, provide greater predictive ability than NIbroad. This differs from the results in Dhaliwal et al. (1 999), where NI dominates both NI,,,, and NIbmJd. The results for predicting future operating income are reported in Panel C. The results reveal that: (5) There exist no statistically significant differences among the three definitions. The results for predicting future NI,,, are reported in Panel D. The results reveal that: (6) NI,,, provides the greatest predictive ability among the three definitions; (7) NIbroad provides a greater predictive ability than NI. The results for predicting future NIbroad are reported in Panel E. The results reveal that: (8) There exist no statistically significant differences between NI and NI,,"; (9) NIhroad provides the greatest predictive ability among the three.

Considered altogether, the results which compare income definitions reported in the second row of Table 4 reveal that: NI,,, provides the highest predictive ability for future N1 and NIbroad; NI,,, provides the highest predictive ability for future NI,,,; NI provides the highest predictive ability for future operating cash flow; and there exists no dominant measure for predicting future operating income. These findings indicate that no single definition of income clearly dominates in predictive ability and that different measures are superior for different predictions.

Results for income component sets using equation ( 5 ) are reported in the rightmost column of Table 4. The results show that: (1 ) for all five future values, NIhrodd exhibits larger explanatory power than NI,,,, which in turn exhibits larger explanatory power than NI; ( 2 ) the sole exception being the prediction of NI,,,, where NI,,, exhibits the largest explanatory power but is indistinguishable statistically from NIbroad (Panel E); (3) all differences are statistically significant except that the components of NI,,, do not differ significantly from those of NI and NIbroad for predicting NI (Panel A) and the components of NI,,, do not differ significantly from those of NIbroad for predicting operating cash flow (Panel B). Thus, the results in Table 4 for income component equation (5) showing that NIbroad usually dominates other measures in predictive ability and the differences in predictive ability results between income definition equation (4) and income component equation (3, both lend support to hypothesis H, and the decision usefulness of disclosing income components.30*

3n We also examined separately predictive ability for year t+2, with qualitatively similar results for the income components. However, for income definitions, the orderings are generally consistent with Table 4, but few significant differences in predictive abilities are observed, consistent with the inherent difficulty of predicting over longer forecast horizons.

3 ' Reported explanatory powers are relatively low because we report tests that use a pooled sample. When we perform year-by-year tests, explanatory powers increase significantly. For example, the average Rz of NI,,,od components in predicting NIbmad using year-by-year data is 0.1989 compared with 0.0557 for pooled data in Panel E of Table 4. In terms of RZ order, individual year results are qualitatively similar to those for the pooled sample.

24 Gary C. Biddle and Jong-Hag Choi Journal of Contemporary Accounting & Economics Vol2, N o I (June 2006) 1-32

5.3 Executive cash compensation (H,)

We test hypothesis H, by examining the relative decision usefulness of 16 definitions of income for explaining executive cash compensation. Following the approach of Gaver and Gaver (1998), we separate observations by the sign of an income component. We also continue to distinguish between income definitions (equation (6)) and component sets (equation (7)):

COMP, = a +b, "1, + Added Component(~)~]+b, (NEGI x [Nil + Added Component(s),]) +b, "I,-, + Added Cornponent(s),-/]+ b, (NEG,, X "I,., +Added Component(s)l-,]i E, (6)

COMP, = a +b, N l , + b, (NEG, x NI,) + b, NI,., + b, (NEG,., x NI,-/ ) + bi Added Component(s)l + bj (NEGr x Added Component(s),) + b, Added Component(s)l-, +b, (NEG,., x Added Component(s)r., )+ E~ (7)

where NEG = 1 if the associated variable is negative; and other terms are as defined in Table 1. Empirical results are reported in Table 5. Panel A of the Table 5 reports the results using income definitions (equation 6) and Panel B reports the results using income components (equation 7).

Similar to Table 2 above, we present adjusted R2s with decision usefulness compared with four benchmarks. The first comparison is against NI, with the corresponding value of the BSS Wald statistic reported in the third column of the table. The second comparison is against NI,,,, with the corresponding BSS statistic reported in the fourth column. The third comparison is against NIbroad, with the corresponding BSS statistic reported in the fifth column. The last comparison is against the income definition that provides the largest adjusted R2 among models having the same number of added components. The resulting BSS statistic is reported in the rightmost column of the table.

The results in Panel A of Table 5 for income definitions equation (6) reveal that: (1) in contrast to Table 2 for equity returns and Table 4 for predictive ability, NI now dominates both NI,,, and NIbrOad in explaining executive cash compensation (significant at conventional levels); and (2) NI,,, and NIbroad are not distinguishable in decision usefulness.

Among income definitions that add one SFAS 130 component to NI, the results in Table 5 reveal that: (3) the addition of PEN (rather than SEC as in Table 2 ) provides the greatest value-relevance (statistically significant at conventional levels beyond SEC, but not beyond FCT and OTH); (4) the addition of FCT and PEN yield an income definition with decision usefulness greater than NI,30; ( 5 ) the addition of FCT, PEN and OTH to NI also provides an income definition with decision usefulness greater than NIbroad; and (6) the addition of SEC to NI yields decision usefulness smaller than for NI. Thus, these findings for executive compensation differ from those for equity returns in terms of which income measure dominates (NI versus NI,,,), whether individual components subtract versus add decision usefulness and in terms of the contributions of individual components (notably PEN versus SEC). They also differ from the predictive ability results in Table 4 where the results are inconclusive.

Among income definitions that add two SFAS 130 components to NI, Table 5 reveals that: (7) the addition of FCT and PEN provides the greatest decision usefulness, larger

Gary C. Biddle and Jong-Hag CIzoi 25 Jvcrrnal of Coiiteriiporary Accounting & Ecoriomics Vol2, No 1 (June 2006) 1-32

Table 5

Relative Decision Relevance of Alternative Earnings Measures for Executive Cash Compensation (H,)

COMP,= a +b, [Nf, + Added Component(s),]+ b,(NEG, x [",Added Cornponent(s),])+ b,[N/,~, + Added Component(s),., ] + b, (NEG,., x "I,., + Added Component(s),., ]+ E,

COMP,=a +b,NI,+ b,(NEG, X Nl!) + b,NI,., + b,(NEG,.,x NI,.,) + b,Added Component(s),+ 6, (NEG,X Added Component(s),) + b, Added Component(s),., +

(6)

6, (NEG,., x Added Component(s),.,)+ E, (7)

Panel A: Analyses with income definition (equation 6 )

Biddle et al. (1995) Wald statistics for comparison against:

NI NI,,,, NI,,ro4d Best model with

(Adj. R' (Adj. R' (Adj. R' the same number

Added Component(s) Adj. R* = 0.1303) = 0.1264) = 0.1226) of added components

None (=NI) 0. I303 NA 6.08** 3.47* NA

SEC 0. I270 -8.40"** 0.33 1.42 -9.24*** FCT 0.1302 -0.02 7.04*** 3.39* -0.23

PEN 0.1303 -0.54 7.63*** 3.68* NA

OTH 0. I267 -0.87 0.0 I 7.12*** -0.97

SEC, FCT 0.1261 -7.98*** -0.96 0.97 - l0.37***

SEC, PEN 0.1267 -6.66*** 0. I5 I .28 -6.33**

SEC, OTH 0.1227 -3.48* - 1.05 0.02 -3.43*

FCT, PEN 0.1301 -0.08 9.22*** 3.63" NA

FCT, OTH 0. I267 -0.86 0.0 I 8.34*** -0.95

PEN, OTH 0. I268 -0.8 1 0.02 8.49*** -0.86

SEC, FCT, PEN

( = NI,,, ) 0.1264 -6.08** NA 1.15 -0.02

SEC, FCT, OTH 0.1224 -3.69* -9.71 *** -0.25 -10.84***

SEC, PEN, OTH 0.1228 -3.33* -6.52*** 0.12 -7.01 *** FCT, PEN, OTH 0. I268 -0.77 0.02 10.27*** NA

0. I226 -3.47* -1.15 NA NA SEC, FCT, PEN, OTH

( = NIbmad )

26 Gary C. Biddle and Jong-Hag Choi Journal of Contemporary Accounting & Econornics Vol2, No I (June 2006) 1-32

Table 5 (Cont.)

Panel B: Analyses with income components (equation 7)

Biddle et al. (1995) Wald statistics for comparison against:

NIbrmd Best model with NI NLO (Adj. R* (Adj. R2 (Adj. R2 the same number

Added Component(s) Adj. R2 =0.1303) = 0.1368) = 0.1433) of added components

None (=NI) 0.1303 NA -3.21* -12.80*** NA

SEC 0.1313 2.03 - I .66 - 1 1.12*** -5.31** FCT 0.1354 1.13 -2.53 -13.47*** -3.91 ** PEN 0.1304 0.60 -2.8 1 * -12.33*** -9.50***

OTH 0.1369 11.71*** 0.94 -3.16* NA

SEC, FCT 0.1370 2.79* 0.46 -1 1.62*** -4.24**

SEC, PEN 0.1314 2.58 -1.26 -10.60*** -5.18**

SEC, OTH 0. I376 13.00*** 2.48 - I O M * * * -0.05 FCT, PEN 0.1354 1.53 -2.13 -13.02*** -9.73***

FCT, OTH 0.1421 10.74*** 3.49* -2.39 NA

PEN, OTH 0.1405 12.68*** 0.49 -2.63 -0.87

0.1368 3.2 1 * NA - I 1.13*** -9.07*** SEC, FCT, PEN

( = NI,,, )

SEC, FCT, OTH 0.1431 I1.99*** 9.07*** -0.7 1 NA

SEC, PEN, OTH 0.1413 13.84*** 1.43 -1.55 - I .02

FCT, PEN, OTH 0.1423 1 I .64*** 4.47** - 1.74 -0.72

NA NA SEC, FCT, PEN, OTH o, 1433 12.80*** 11.13*** ( = N I b d ) Adjusted R2 statistics for equations (6) and (7) for 5,349 firm-year observations. The dependent variable is COMP. The predictors are the levels and the lagged levels of alternative income definitions defined by progressively adding one, two, three or all four comprehensive income components to net income (NI). NEG equals I if the corresponding variable is negative. The analyses are performed after removing outliers with absolute values of standardized errors greater than 3. The adjusted R2 of each model is compared with those of four benchmark models, with the values of Biddle et a]. (1995) statistics reported in the corresponding rows/columns. The second column reports the adjusted RZ of each regression model. The Biddle et al. (1995) statistics reported in the third column is a comparison against NI, the statistic in the fourth column is a comparison against NI,,,, the statistic in the fifth column is a comparison against NIbn,& and the statistic in the rightmost column is a comparison against the model with the largest adjusted RZ among the grouped models with equal numbers of added components. The symbol '***' (**, *) represents that the result is significant at the 1 (5, 10) percent level. Variables are as defined in Table I .


Journal of Contemporary Accounting & Economics Vol2, No I (June 2006) 1-32

than for the additions of SEC and FCT, SEC and PEN and SEC and OTH, but indistinguishable from the additions of other components; (8) the additions of FCT and PEN, FCT and OTH and PEN and OTH, provide income definitions with decision usefulness indistinguishable from NI, but greater than NIbro, (significant at conventional levels); (9) the additions of FCT and PEN provides income definitions with decision usefulness greater than NI,,, (significant at conventional levels); and (10) the additions of SEC and FCT, SEC and PEN and SEC and OTH yield decision usefulness smaller than for NI , but not for

Among income definitions that add three SFAS 130 components to NI, Table 5 reveals that: (1 1 ) the addition of FCT, PEN and OTH provides decision usefulness indistinguishable from the addition of SEC, FCT and PEN = NI,,o, but greater than the other combinations (statistically significant at conventional levels); (12) the addition of FCT, PEN and OTH yields an income definition indistinguishable from NI but greater than NIbroad (significant at conventional levels); (13) the addition of SEC, FCT and PEN = SFAS,,,, SEC, FCT and OTH and SEC, PEN and OTH yields decision usefulness smaller than for NI but indistinguishable from NI,,, and NIbraad; and (14) the addition of SEC, FCT and PEN = NI,,{, yields decision usefulness greater than the addition of SEC, FCT and OTH and SEC, PEN and OTH.

The results in Panel B of Table 5 for income component sets equation (7) reveal that, consistent with the results for Table 2 equation (3) and Table 4 equation (5) , the addition of more components generally enhances decision usefulness. For executive compensation, OTH adds the most explanatory power among individual components, The adjusted R2 for NIbroad is larger than for NI,,, and NI and NI,,,’s is larger than for NI (all significant at conventional levels). Analogously, the patterns of relative decision usefulness differ between income definition equation (6) and components equation (7) and adjusted R2s are uniformly larger for equation (7), lending further support to hypothesis H, and the disclosure of comprehensive income components.

Considered together, the findings in Table 5 reveal in one view the decision usefulness of 16 definitions of income available from SFAS 130 components for executive cash compensation. For income definitions, NI dominates both NI,,, and NIb,t,ad, in contrast to the results for equity returns and predictive ability. For income components, once again the more comprehensive set provides greater decision usefulness.

NI,,, and NIbroad.

5.4 Income measures and disclosures ( H , and H,)

When combined across all three applications of information content, predictive ability and executive compensation contracting, the results in Tables 2-5 for income definitions lend support to hypothesis H,, namely that more (less) comprehensive definitions of income are more decision useful for explaining equity returns (compensation contracting). The results in Tables 2-5 for income component sets and comparisons between these results and those for income definitions, lend support to hypothesis H, that it is decision useful to disclose separately comprehensive income components.

28 Gary C. Biddle and Jong-Hag Choi Journal of Contemporary Accounting & Econonzics Vol2, No 1 (June 2006) 1-32

5.5 Robustness checks

Several alternative formulations were examined with results qualitatively similar to those described above. First, observations were partitioned into financial, manufacturing and other subsamples. Second, formulations in levels only, changes only and with NEG dummies for returns specification predictors (similar to those used for executive compensation) were examined, with similar findings. The results for executive compensation became insignificant in the absence of the NEG dummies, consistent with the findings of Gaver and Gaver (1998). Third, the information content results were found robust to the introduction of control variables for year effects, firm size (natural logarithm of the previous year’s ending market equity), growth potential (previous fiscal year’s ending book-to- market ratio) and earnings persistence (dummy variable equal to 0 if the rank of firm’s prior period’s earnings-to-price ratio is in the highest or lowest quartile, otherwise I ) , following Ali et al. (1992), Ali and Zarowin (1992) and Cheng et al. (1996). Fourth, when we use the change of future values rather than future values for predictive ability tests, the results are consistent for income components (equation 5); for income definitions (equation 4), order is similar though results are generally less significant statistically. Fifth, the compensation analysis was conducted using the undeflated change in executive cash compensation, rather than the level of executive cash compensation scaled by market equity, with qualitatively similar findings. We also performed the analyses year-by-year, 1994- 1998. For both income definitions and income components, results are qualitatively similar with those in Tables 2-5. Overall, our findings appear to be robust to alternative specifications.

6. Discussion and Summary

This study is motivated by a decades-long debate regarding a fundamental definition in accounting, the comprehensiveness of income and the IASB’s current consideration of this question. A review of this debate reveals that investor interests have consistently lobbied for more comprehensive measures of income whereas managers have lobbied to exclude items over which they have less control. In this study we allow that these divergent views may be justified. How? By allowing that different measures of income may be more decision useful in different applications. Given the parties involved, three obvious applications to examine are information content, predictive ability and executive compensation contracting. This is the first study to systematically examine this proposition. The adoption in the US of SFAS 130 in 1997 creates a natural experiment for comparing 16 different measures of income created from comprehensive income components. We examine both income definitions and component sets. These comparisons provide insights for the IASB’s consideration.

Our findings for income definitions reveal that for information content, the definition of comprehensive income adopted by the FASB in SFAS 130 (NI,30) dominates both net income (NI) and fully comprehensive income (Nlbroad). In fact, it is not beaten by any of 15 other definitions of income examined. Thus, in terms of information content, the FASB appears to have chosen wisely. For predictive ability, no income definition clearly dominates in decision usefulness for the prediction of future operating income. But NIbroad is better


than all other measures in predicting future NI and NIbroAd, while NI,,, is better than NI in predicting itself, consistent with dominance in information content. NI better predicts future operating cash flow. For executive compensation contracting, NI dominates both comprehensive income measures in decision usefulness, consistent with the view of managers and contracting arguments (e.g., Holthausen and Watts, 200 1). Thus, these results for income definitions reveal a pattern strikingly consistent with the divergent views of investors and managers, in that comprehensive income dominates for information content and net income for compensation contracting.

Our findings for income component sets reveal that in all applications, decision usefulness in enhanced with the introduction of more components. Here, NIbroad provides the greatest decision usefulness across all three applications: information content, predictive ability and executive compensation contracting. The only exception is in the prediction of NJ3,; although indistinguishable from NIhnwd statistically, NI,,, exhibits the largest predictive ability. Also, the component sets provide generally greater decision usefulness than corresponding income definitions.These findings, when combined, support two primary conclusions: ( 1) different definitions of income provide different decision usefulness in different applications; and (2) it is decision useful to disclose separately comprehensive income components. These conclusions lend .support to the IASB’s (2003) proposal to require the disclosure of comprehensive income and the view that “different components of comprehensive income are useful in different ways to users of the accounts” (IASB, 2002, 10).

References

Ahn, T.S. and W.S. Choi, 2002, “The value-relevance of comprehensive income: A Korean evidence”, Working Paper, Seoul National University.

Ali, A., A. Klein, and J. Rosenfeld, 1992, “Analysts’ use of information about permanent and transitory earnings components in forecasting annual EPS”, The Accounting Review

Ali, A. and P. Zarowin, 1992, “The role of earnings levels in annual earnings-returns studies”, Journal of Accounting Research 30, 286-296.

American Institute of Certified Public Accountants, 1994, Improving Business Reporting - A Customer Focus (AICPA, NY).

Association for Investment Management and Research, 1993, Financial Reporting in the 1990’s and Beyond (AIMR, Charlottesville, VA).

Baber, W., S. Kang, and K. Kumar, 1998, “Accounting earnings and executive compensation: The role of earnings persistence”, Journal of Accounting and Economics 25, 169- 193.

Beresford, D., L. Johnson, and C. Reither, 1996, “Is a second income statement needed?’, Journal of Accountancy 176,69-72.

Biddle, G., G. Seow, and A. Siege], 1995, “Relative versus incremental information content”, Contemporary Accounting Research 12, 1-23.

Black, F., 1993, “Choosing accounting rules”, Accounting Horizons 7, 1- 17. Bradshaw, M. and R. Sloan, 2002, “GAAP versus the street: An empirical assessment of

two alternative definitions of earnings”, Journal of Accounting Research 40,41-65.

67, 183-198.

Mohammad

Highlight

Mohammad

Highlight

Mohammad

Highlight

30 Gary C. Biddle and Jong-Hag Choi Journal of Contentpnrary Accounting & Economics VoL 2, N o I (June 2006) 1-32

Brief, R. and K. Peasnell (Eds.), 1996, Clean Surplus: A Link Between Accounting and Finance (Garland Publishing, NY).

Brown, L. and K. Sivakumar, 2003, “Comparing the value relevance of two operating income measures”, Review of Accounting Studies 8, 561-572.

Bushman, R., E. Engel, J. Milliron, and A. Smith, 2000, “An analysis of the relation between the stewardship and valuation roles of earnings”, Working paper, University of Chicago.

Cahan, S., S. Courtenay, P. Gronewoller, and D. Upton, 2000, “Value relevance of mandated comprehensive income disclosures”, Journal of Business, Finance and Accounting 27,

Cheng, C., C. Liu, and T. Schaefer, 1996, “Earnings permanence and the incremental information content of cash flows from operations”, Journal of Accounting Research

Choi, J.H. and S. Das, 2003, “Analysts’ use of comprehensive income in forecasting EPS”, Working Paper, Hong Kong University of Science and Technology.

Clinch, G. and J. Magliolo, 1993, “CEO compensation and components of earnings in bank holding companies”, Journal of Accounting and Economics 16, 24 1-272.

Dechow, P., 1994, “Accounting earnings and cash flow as measures of firm performance: The role of accounting accruals”, Jourqal of Accounting and Economics 18, 3-42.

Dhaliwal, D., K. Subramanyam, and R. Trezevant, 1999, “Is comprehensive income superior to net income as a measure of firm performance?’, Journal ofAccounting and Economics

Doyle, J., R. Lundholm, and M. Soliman, 2002, “The predictive value of expenses excluded from ‘pro forma’ earnings”, Working Paper, University of Michigan.

Financial Accounting Standards Board, 1997, Statement of Financial Accounting Standards No. 130: Reporting Comprehensive Income (FASB, Stamford, CT).

Gaver, J. and K. Gaver, 1998, “The relation between nonrecurring accounting transactions and CEO compensation”, The Accounting Review 73,235-253.

Hirst, D. and P. Hopkins, 1998, “Comprehensive income reporting and analysts’ valuation judgments”, Journal of Accounting Research 36,47-74.

Hirst, D., P. Hopkins, and J. Wahlen, 2001, “Fair values, comprehensive income reporting and bank analysts’ risk and valuation judgments”, Working paper, University of Texas.

Holthausen, R. and R. Watts, 2001, “The relevance of the value relevance literature for financial accounting setting”, Journal of Accounting and Economics 31, 3-75.

International Accounting Standards Committee, 1999, G4+1 Position Paper: Reporting Financial Performance, London.

International Accounting Standards Board, 2002, The Income Statement (Reporting Performance) - Project Summary (October), London.

International Accounting Standards Board, 2003, Reporting Comprehensive Income (September), London.

Johnson, L., C. Reither, and R. Swieringa, 1995, “Toward reporting comprehensive income”, Accounting Horizons 9, 128-137.

Johnson, W. and W. Schwartz, Jr., 2001, “Are investors misled by “Pro Forma” earnings”, Working Paper, University of Iowa and University of Arizona.

Kiger, J. and J. Williams, 1977, “An emerging concept of income presentation”, The Accounting Historians Journal 4, 63-77.

1273-130 1.

34, 173-181.

26,43-67.


Lambert, R., 200 I , “Contracting theory and accounting”, Journal of Accounting and Economics 32,347.

Lambert, R. and D. Larcker, 1987, “An analysis of the use of accounting and market measures of performance in executive compensation contracts”, Journal of Accounting Research 25,85125.

Liesman, S. and J. Weil, 2001, “Different views of ‘special items’ lead to various figures for S&P profit”, Wall Street Journal (August), 24.

Linsmeier, T., J. Gribble, R. Jennings, M. Lang, S. Penman, K. Petroni, D.Shores, J. Smith, and T. Warfield, 1997a, “Response to FASB exposure draft, Proposed statement of financial accounting standards - Reporting comprehensive income”, Accounting Horizons 11, 117-119.

Linsmeier, T., J. Gribble, R. Jennings, M. Lang, S. Penman, K. Petroni, D. Shores, J. Smith, and T. Warfield, 1997b, “An issues paper on comprehensive income”, Accounting Horizons 1 1, 120- 126.

Lipe, M., 1998, “Discussion of comprehensive income reporting and analysts’ valuation judgments”, Journal of Accounting Research 36,7743.

Littleton, A., 1940, “The integration of income and surplus statement”, Journal of Accountancy 69,30-40.

Lougee, B. and C. Marquardt, 2002, “Earnings quality and strategic disclosure: An empirical examination of ‘pro-forma’ earnings”, Working paper, New York University.

Louis, H., 2003, “The value relevance of the foreign translation adjustment”, The Accounting Review 78 (4), 1027- 1047.

Maines, L. and L. McDaniel, 2000, “Effects of comprehensive-income characteristics on nonprofessional investors’ judgments: The role of financial statement presentation format”, The Accounting Review 75, 177-204.

May, G., 1937, “Eating peas with your knife”, Journal ofAccountancy 63, 15-22. Natarajan, R., 1996, “Stewardship value of earnings components: Additional evidence on

O’Hanlon, J., 2000, “Discussion of value relevance of mandated comprehensive income

O’Hanlon, J. and P. Pope, 1999, “The value-relevance of UK dirty surplus accounting

Paton, W., 1934, “Shortcomings of present-day financial statements”, Journal of

Robinson, L., 1991, “The time has come to report comprehensive income”, Accounting

Sloan, R., 1993, “Accounting earnings and top executive compensation”, Journal of

Smith, P. and C. Reither, 1996, “Comprehensive income and the effect of reporting it”,

Smith, R., 2001, “Debate over earnings puts Wall Street analysts at odds”, Wall Street

Sutton, M. and J. Johnson, 1993, “Current values: Finding a way forward”, Financial

the determinants of executive compensation”, The Accounring Review 7 1, 1-22.

disclosures’’, Journal of Business, Finance and Accounting 27, 1303- 1309.

flows”, British Accounting Review 3 1, 459-482.

Accountancy 57, 15-22.

Horizons 5 , 107- 1 12.

Accounting and Economics 16,55-100.

Financial Analysts Journal 52, 14-19.

Journal (August), 29.

Executive 9,39-43.

32 Gary C. Biddle and Jong-Hug Choi Journal of Contemporary Accounting & Economics Vol2, No I (June 2006) 1-32

Vuong, Q., 1989, “Likelihood ratio tests for model selection and non-nested hypotheses”, Econornetrica 57, 307-333.

Wahlen, J., J. Boatsman, R. Herz, G. Jonas, K. Palepu, S. Ryan, K. Schipper, C. Schrand, and D. Skinner, 2000, “Response to the special report of G4+1, Reporting financial performance: A proposed approach’, Accounting Horizons 14, 365-379.

White, W., 1980, “A heteroskedasticity-consistent covariance matrix estimation and a direct test for heteroskedasticity”, Econornetrica 48, 8 17-838.

Usefulness of OCI

Documents

comprehensive income

comprehensive income

accounting research

traditional net income

predictive ability

executive compensation

financial support

hedges sfas