Errors in Estimating Accruals: Implications for Empirical Research Daniel W. Collins a *, Paul Hribar b a Henry B. Tippie Research Chair in Accounting Tippie College of Business, University of Iowa, Iowa City, IA 52242 b Johnson Graduate School of Management, Cornell University, Ithaca, NY 14853 Current Version: September 10, 2000 *Corresponding Author. Phone (319) 335-0910; Fax: (319) 335-1956; e-mail: danielcollins@uiowa.edu. We gratefully acknowledge the helpful comments and suggestions made by Kevin Den Adel, Mary Barth, Bill Beaver, Mike Cipriano, S.P. Kothari, Nicole Jenkins, Maureen McNichols, Mort Pincus, Scott Vandervelde, Charles Wasley, Hong Xie and workshop participants at the University of Iowa, Tulane University, and the Stanford Summer Research Camp.
Abstract This paper examines the impact of measuring accruals as the change in successive balance sheet accounts, as opposed to measuring accruals directly from the statement of cash flows. Our primary finding is that studies using a balance sheet approach to test for earnings management are potentially contaminated by measurement error in accruals estimates. In particular, if the partitioning variable used to indicate the presence of earnings management is correlated with the occurrence of mergers and acquisitions or discontinued operations, tests are biased and researchers are likely to erroneously conclude that earnings management exists when there is none. Additional results show that the errors in balance sheet accruals estimation can confound returns regressions where discretionary and non-discretionary accruals are used as explanatory variables. Moreover, we demonstrate that tests of market mispricing of accruals will be understated due to erroneous classification of extreme accruals firms. 1
Errors in Estimating Accruals: Implications for Empirical Research I. Introduction The measurement of accruals plays a central role in a wide body of literature in accounting. This literature includes studies on the relative informativeness or value relevance of cash flows versus accruals [Rayburn (1986), Wilson (1987), Dechow (1994)], tests of earnings management and income smoothing [e.g. Healy (1985), DeAngelo (1986, 1988), Jones (1991), Dechow, Sloan and Sweeney (1995), Rees, Gill and Gore (1996), DeFond and Subramanyam (1998), Teoh, Welch, Wong (1998), Rangan (1998)], the pricing of discretionary versus nondiscretionary accruals [Subramanyan (1996), Guay, Kothari and Watts (1996), Xie (1999)], and the market's mispricing of accruals [Sloan (1996), Xie (1999), Collins and Hribar (2000)]. Despite the availability of accurate accruals data in the statement of cash flows since 1988, the majority of these studies use an indirect balance sheet approach to calculate accruals. The balance sheet approach relies on the presumed articulation between changes in working capital balance sheet accounts and accrual components of revenues and expenses on the income statement. This presumed articulation breaks down when nonoperating events such as reclassifications, acquisitions, divestitures, accounting changes and foreign currency translations occur [Drtina and Largay (1985), Huefner, Ketz and Largay (1989), Bahnson, Miller and Budge (1996) and Revsine, Collins and Johnson (1999)]. Using actual accruals taken from cash flow statements subsequent to SFAS No. 95, Bahnson et al. (1996) estimate that approximately 75% of Compustat firms present nonarticulated changes in current accounts. The purposes of this paper are to assess the error introduced by the indirect balance sheet approach to accruals estimation and demonstrate the implications for empirical studies that have 2
relied on this approach. The severity of the mismeasurement of accruals and/or operating cash flows is evaluated for samples comprised of a broad cross section of NYSE and AMEX firms and for samples of firms involved in mergers and acquisitions, divestitures and foreign operations through subsidiaries. We demonstrate the effect of using balance sheet accruals estimates relative to accruals taken directly from the statement of cash flows in three popular applied settings: (1) estimating the discretionary and nondiscretionary component of accruals and tests of earnings management; (2) the relation between security returns and accruals, discretionary accruals, nondiscretionary accruals, and cash flow from operations; and (3) testing for market mispricing of accruals. Our major finding is that the error induced by using a balance sheet estimation approach contaminates computations of so-called discretionary or abnormal accruals, and can lead to erroneously concluding that earnings management exists when no such opportunistic activity is present. We demonstrate that balance sheet accruals estimates are predictably biased in studies where the partitioning event is correlated with either mergers and acquisitions or discontinued operations. To illustrate how this bias can lead to unwarranted inferences, we replicate sections of two prior studies that test for earnings management using a balance sheet approach. Additionally, we document the probability of making a type I error as the percent of the sample that is contaminated by these non-articulation events increases. Other findings show that when discretionary and non-discretionary balance sheet accruals estimates are used as regressors in a second stage regression of returns on earnings components, the resulting coefficient bias can yield incorrect statistical inferences about the equality of parameter estimates. Finally, we demonstrate that tests of market mispricing of accruals calculated under the balance sheet 3
approach tend to understate the true extent of the mispricing by roughly 35% due to misclassification of extreme accruals firms. The remainder of the paper proceeds as follows. Section two explains the balance sheet approach to estimating accruals and how the presumed articulation between changes in balance sheet working capital accounts and income statement accruals breaks down when nonoperating events like mergers and acquisitions, divestitures and foreign currency translations are present. This section also documents the frequency and magnitude of accrual estimation errors by comparing total accruals taken directly from the cash flow statement to estimates derived using the indirect balance sheet method. In section three we use the modified Jones model to evaluate the effect of non-articulation on estimates of discretionary and nondiscretionary accruals and tests of earnings management. We compare results for two samples of firms: (a) a broad crosssection of NYSE/AMEX firms; and (b) samples of firms affected by one or more of the nonoperating events that contribute to non-articulation. We also investigate the potential bias in selected studies on earnings management that are confounded by non-articulation events. Section four investigates the relation between returns, discretionary and non-discretionary accruals, and operating cash flows determined using the balance sheet approach versus the cash flow statement approach. Section five demonstrates the effect of measurement error on tests of market mispricing of accruals using the approach outlined in Sloan (1996). Section six summarizes our findings and discusses the implications for prior and future research. 4
II. The Non-articulation Problem: Its Frequency, Magnitude and Contributing Factors 2.1 An example of the non-articulation problem Much of the literature to date that focuses on accruals uses a balance sheet approach to determine the accruals component of earnings. Specifically, accruals (ACC bs ) are typically calculated as follows (firm and time subscripts omitted for convenience): ACC bs = ( CA - CL - Cash + STDEBT DEPTN) (1) where CA = the change in current assets during period t (Compustat #4); CL = the change in current liabilities during period t (Compustat #5); Cash = the change in cash and cash equivalents during period t (Compustat #1); STDEBT = the current maturities of long-term debt and other short-term debt included in current liabilities during period t (Compustat #34); and DEPTN = depreciation and amortization expense during period t (Compustat #14). All variables are deflated by lagged total assets (TA t-1 ) to control for scale differences. The H.J. Heinz 1997 annual report demonstrates the magnitude of the accrual estimation problem caused by non-articulation between the balance sheet and the cash flow statement and the kinds of events that contribute to this problem. 1 The first column of numbers in Exhibit 1 shows the operating section of Heinz s cash flow statement for the year ended, April 30, 1997 and the implied directional changes in the working capital accounts after excluding the effects of acquisitions and divestitures. Note that the net change in working capital and its corresponding effect on earnings is + $253.3 million (i.e., income increasing accruals). The directional changes in the individual working capital accounts taken from Heinz s comparative balance sheets for 1996-1997 are presented as the second column of numbers. 2 1 This illustration is adapted from Revsine, Collins and Johnson (1999, Chapter 16). 2 These amounts exclude (in $ millions) a reduction of short-term debt of $404.7, an increase in the current portion of long-term debt of $486.0 and an increase in accrued restructuring costs of $210.8 which are captured elsewhere in Heinz s cash flow statement. Also omitted is $324,418 of depreciation and amortization expense. 5
These numbers are presented analogous to the statement of cash flows in that they represent the adjustment to net income to arrive at cash from operations. Notice that the net change in these working capital accounts is only + $13.1 million. Thus, the indirect approach to calculating accruals based on using comparative balance sheets understates Heinz s working capital accruals by approximately $240.2 million, or 79.6% of Heinz s 1997 reported net income of $301.9 million. The main reason for the understatement of Heinz s accruals (and associated overstatement of operating cash flows) stems form two major divestitures that Heinz made in fiscal 1997 related to its Project Millenia reorganization and restructuring program. As shown on the cash flow statement and detailed in footnotes to their annual report, Heinz divested of its New Zealand ice cream business and its U.K. real estate business in 1997. Accordingly, a portion of the decrease in the working capital balance sheet accounts relates to these divestitures that would erroneously be shown as income decreasing accruals under the balance sheet approach to estimating accruals. In contrast to divestitures of on-going businesses that introduce a negative bias to estimated accruals, mergers and acquisitions introduce a positive bias to estimated accruals using the approach in eqn. (1). Heinz reported cash outflows (net of cash acquired) of $208.4 million in 1997 for acquisitions of several businesses accounted for under the purchase method. A portion of this Investing Activity cash outflow would most likely contribute to a positive increase in working capital accounts that did not have a corresponding positive effect on accrual earnings. Finally, Heinz has numerous foreign subsidiaries whose statements are translated using the current rate approach specified in SFAS No. 52. If the dollar falls (rises) in relation to the 6
functional currencies of these subsidiaries, this change would result in increases (decreases) in balance sheet accounts that would not have a corresponding effect on accruals [Huefner et al.(1989)]. Unlike divestitures and acquisitions where one can reliably sign the direction of the bias in accruals estimated via eqn. (1), the direction of the bias due to foreign currency adjustments is indeterminate and, hence, is more akin to error or noise. The bias in estimated accruals under the balance sheet approach would depend on whether the dollar strengthened or weakened relative to the local currency of the countries in which a company operates. While there are no doubt other reasons for non-articulation (e.g., accounting changes and reclassifications that affect current accounts), we believe that mergers and acquisitions, divestitures and foreign currency translation of foreign subsidiary account balances are likely to be the most important and most pervasive factors contributing to the non-articulation problem. Accordingly, the remainder of the paper focuses on understanding how these transactions affect the estimation of accruals and cash flows using the balance sheet approach reflected in eqn. (1) and (2) above, which has been the dominant approach used in the literature to date. 2.2 Estimating the severity of the non-articulation problem Our sample is comprised of all NYSE/AMEX firms on the Compustat Primary, Supplementary and Tertiary Annual Industrial File and Research File with requisite balance sheet and earnings data to compute accruals according to eqn. (1). 3 Accruals are determined for each year over the ten-year period from 1988 1997 resulting in a total sample of 14,558 firmyears. 3 The primary tests are also replicated including all NASDAQ firms in the sample as well, with no qualitative change in results. The NYSE/AMEX sample is used for comparability with prior accruals research which focuses primarily on this subsample of firms. 7
To provide some evidence on the severity of the non-articulation problem, we first calculate accruals using the balance sheet approach (ACC bs ) as shown in eqn. (1). We next compute two alternative definitions of accruals directly from the cash flow statement. The first measure captures total accruals and is computed as follows (firm and time subscripts are suppressed for convenience): TACC cf = EBXI CFO cf (2) where TACC cf = the total accrual adjustments provided on the cash flow statement under the indirect method; EBXI = earnings before extraordinary items and discontinued operations (Compustat #123); and CFO cf = operating cash flows (from continuing operations) taken directly from the statement of cash flows (Compustat #308 - Compustat #124). 4 The second measure of accruals includes only the changes in the working capital accounts plus depreciation expense and, hence, is more directly comparable to the balance sheet definition of accruals commonly used in prior research. 5 This definition of accruals is computed as follows: ACC cf = - (CHGAR cf + CHGINV cf + CHGAP cf + CHGTAX cf + CHGOTH cf + DEP cf ) (3) where CHGAR cf is the decrease (increase) in accounts receivable (Compustat #302); CHGINV cf is the decrease (increase) in inventory (Compustat #303); CHGAP cf is the increase (decrease) in accounts payable (Compustat #304); CHGTAX cf is the increase (decrease) in taxes payable (Compustat #305); CHGOTH cf is the net change in other current assets (Compustat #307); and 4 The cash portion of discontinued operations and extraordinary items (Compustat #124) is subtracted from total cash from operations to provide a cash flow from continuing operations. This is consistent with our definition of net income. Alternatively, one could include bottom line net income (Compustat #172) instead of net income before extraordinary items, which is consistent with including cash from discontinued operations in the measure of operating cash flow. 5 An alternative approach is to include a funds from operations variable in the balance sheet definition of accruals to arrive at a measure of total accruals using the balance sheet. This metric, however, can only be computed prior to 1988. 8
DEP cf is depreciation expense (Compustat #125). All of these variables are taken from the operating section of the statement of cash flows and, hence, are not affected by non-operating changes in these accounts. Because this second measure approximates the balance sheet definition more closely, we calculate the difference in accruals estimated under the balance sheet approach given in eqn. (1) and the accrual amount from the statement of cash flows given in eqn. (3) for each firm and scale by alternative deflators as follows: DIFF = ACC bs - ACC cf / Deflator (4) where ACC bs the alternative deflators are the absolute value of EBXI, ACC bs, or total assets at the beginning of the year (TA t-1 ). 6 Table 1 presents descriptive statistics for the error or bias in balance sheet accruals induced by the three non-articulation events noted earlier. Panel A presents results for firmyears where the firm is involved in a merger or acquisition, Panel B presents results for firmyears where the firm is involved in divestitures (discontinued operations), Panel C presents results for firms with foreign operations, and Panel D presents results for firms without any of the three non-articulation events. To determine observations that are affected by one of the three non-articulation events, we use proxies that are available on Compustat. Mergers and acquisitions are determined by examining annual footnote code #1, which reflects whether a merger or acquisition impacts current sales. A firm is considered to have discontinued operations if the absolute value of discontinued operations (Compustat #66) exceeds $10,000. 6 It should be emphasized that all predictions regarding error and bias in balance sheet estimation will carry over regardless of which measure of accruals is used, as it is the non-operating changes in working capital accounts from the balance sheet that lead to the non-articulation problem. 9
Similarly, a firm is considered to have foreign operations if the absolute value of the foreign currency adjustment account (Compustat #150) exceeds $10,000. Consistent with our priors, the mean, median and inter-quartile range of ACC bs values for the merger and acquisition sample are less negative than the ACC cf amounts (i.e., positively biased). For example, the median DIFF in estimated accruals under the balance sheet approach is 16.95% of EBXI, indicating a positive and significant bias in balance sheet accruals estimates. For one-quarter of the firms in this sample, the accruals error exceeds 64.3% of EBXI. Similarly, the results for the discontinued operations sample (Panel B) reveal the expected negative bias in estimated accruals under the balance sheet approach. The mean, median and inter-quartile range values for the scaled ACC bs values are consistently more negative than for the ACC cf amounts. For example, the median DIFF is -29.41% of EBXI, demonstrating a negative and significant bias in balance sheet accruals estimates in this subset of firms. For the lower quartile of the firms in this sample, the negative bias is substantial, measuring -96.3% of EBXI. The results for the foreign operations subsample in Panel C are less dramatic. The median difference is 0.12% of TA t-1 or 1.65% of EBXI. This is expected, however, as foreign currency translations can cause positive or negative differences in balance sheet accruals estimates depending on the fluctuations of the local (foreign) currency relative to the U.S. dollar. Hence, unlike the case of mergers and acquisitions or discontinued operations, foreign currency translations do not induce an ex-ante predictable bias in balance sheet accruals estimates. Thus, the resulting difference between accruals measured using the balance sheet and cash flow approaches is more akin to error or noise. Finally, Panel D reports the difference for firms that do not have any of these three events. Similar to the case of foreign currency translations, the 10
median difference is quite small, equaling 0.11% of total assets, or 1.85% of EBXI. Additionally, the interquartile range is smaller, indicating that there is less noise in these firms accruals estimates than in the foreign operations subsample. The univariate results presented in Table 1 do not control for the fact that in any given year, firms might have more than one of the three events affecting the articulation between the balance sheet, income statement and the statement of cash flows. To account for this potential overlap, Table 2 provides multivariate estimates of the average bias in estimated accruals induced by each of the three events noted above. The first model regresses DIFF/TA t-1 on three indicator variables, D MERGER, D DISC-OP, and D FC, each of which equals one in firm-years in which mergers and acquisitions, discontinued operations, or foreign currency translations occur, respectively, and zero otherwise. The benefit to this approach is that the average bias can be examined after controlling for the other non-articulation events. Model 2 is similar to the first model, but also adds an indicator variable for significant mergers and acquisitions. 7 Results in Table 2 show first that there is a significantly positive intercept. Thus the balance sheet estimation method tends to induce a positive bias across all firms, before considering non-articulation events. This is likely due to the omission of non-current accruals other than deprecation and amortization when computing total accruals using the balance sheet approach. 8 Turning to the non-articulation events, mergers and acquisitions induce an upward bias in balance sheet accruals estimates of approximately 1.48% of total assets. In contrast, discontinued operations induce a downward bias of approximately -1.54% of total assets, and 7 A merger or acquisition is classified by Compustat as significant when the impact on current sales is greater than 50% of the prior year s sales. 8 Bahnson et al. (1996) use a more precise measure of accruals from the balance sheet that might resolve this overall positive bias. Instead of using Bahnson et al.'s approach, we opted to use the balance sheet estimation approach most commonly used in prior empirical accrual studies. 11
foreign currency translations have a smaller negative bias of approximately -0.49% of total assets. Moreover, Model 2 demonstrates that large mergers and acquisitions induce an incremental bias (relative to small mergers and acquisitions) of 0.91% of total assets, for a total bias of approximately 2.39% of total assets. Because of the small number of large merger firms relative to the total merger subsample (56 larger merger firms compared to 2991 merger firms in total), the coefficient on the small mergers decreases only slightly in the expanded model (from 1.48% to 1.44%). Finally, to provide some sense of the pervasiveness of the non-articulation problem caused by the three events noted above, Figure 1 plots the percentage of Compustat firms affected by each of these events over the ten-year period from 1988 to 1997. The percentage of Compustat firms involved in a merger or acquisition has increased steadily from a low of roughly 14% in 1991 to slightly over 27% in 1997. The percentage of firms with material foreign subsidiaries has remained fairly steady throughout the sample period ranging from 18% to 21% of Compustat firms. Firms reporting divestitures (discontinued operations) ranges from just over 15% in 1988 to around 9% in 1997. The percentage of Compustat firms having one or more of the three events that introduce error in ACC bs estimates is shown in the top line of Figure 1a. Clearly, a significant portion of Compustat firms are affected, with the proportion ranging from a low of 38.6% in 1991 to 46.5% in 1997. Figure 1b depicts a time series plot of the absolute value of the difference (deflated by total accruals) between accruals measured using the balance sheet and the cash flow statement approach. A cursory examination of the plot reveals that the average error resulting from a balance sheet approach appears to follow the same pattern as the relative frequency of nonarticulation events presented in Figure 1a. Specifically, both the frequency of non-articulation 12
events in Figure 1a and the degree of measurement error in balance sheet accruals estimates plotted in Figure 1b decrease from 1988-1991, and then follow more of a gradual upward trend until 1997. Collectively, the results reported in this section suggest that non-operating events or transactions (i.e., mergers and acquisitions, divestitures and foreign operations) can cause significant error and bias in accrual estimates using the balance sheet approach. Moreover, the number of firms affected by one or more of these activities represents a nontrivial proportion of the firms that comprise the Compustat files. Thus, a considerable body of literature to date that utilizes balance sheet-based accruals estimates suffers from a potentially large measurement error problem. The following sections evaluate how the errors embedded in ACC bs estimates affect inferences in studies where accruals are the major object of interest. III. Estimating Discretionary and Nondiscretionary Accruals and Detecting Earnings Management 3.1 Bias in discretionary accrual calculations McNichols and Wilson (1988) outline a general discretionary accruals framework that is the foundation for most earnings management studies in accounting. In the model, accruals are partitioned into a discretionary (DACC*) and non-discretionary (NDACC*) component, such that: ACC* = NDACC* + DACC* (5) Since DACC* is unobservable, it is typically estimated using one of several alternative empirical models. 9 The estimate of discretionary accruals (DACC) that emerges from the empirical model inevitably measures the true and unknowable discretionary accruals (DACC*) with error: 9 Examples of discretionary accruals models include the Healy model, the DeAngelo model, the Jones model, the modified Jones model, and the Kang and Sivaramakrishnan model. Because of their prevalence in accounting research, most of the analysis in this section focuses on the Jones and modified Jones models. 13
DACC = DACC* + η (6) where η is the measurement error associated with the estimate. Ultimately, the researcher uses this estimate of discretionary accruals to test for earnings management surrounding a particular event. Examples of these events are numerous, but include such diverse events as seasoned equity offerings, proxy contests, stock-for-stock mergers, management buyouts, asset writedowns, and management forecasts. Typically, the test is conducted by regressing discretionary accruals on a partitioning variable (PART), which is a dummy variable equaling one in the period(s) in which earnings management is hypothesized to occur. Thus, a theoretical model to test for earnings management can be represented as follows: DACC* = α + β PART + ε (7) Where α estimates average discretionary accruals across all firms, and α + β estimates the average discretionary accruals for the experimental group. The significance of β is used to draw inferences about the presence of earnings management (or lack thereof). However, because a discretionary accrual proxy, DACC, is used instead of DACC*, McNichols and Wilson (1988) demonstrate that equation (7) can be rewritten as follows: DACC = α + γ PART + ν, (8) where γ = β + ρ PART, η * σ η /σ PART = β + bias in γ Hence, as noted in McNichols and Wilson (1988), given measurement error in DACC, the coefficient used to test for the presence of earnings management will be biased. Moreover, this bias will be (1) increasing in the correlation between η and PART, (2) increasing in the variance of η, and (3) decreasing in the variance of PART. More importantly, with a significant bias in the discretionary accrual estimate, one could erroneously conclude there is earnings management 14
(i.e. observe non-zero values of γ) when, in fact, earnings may not be managed at all (i.e. β = 0). Alternatively, accruals estimates biased in the opposite direction of the hypothesized earnings management could yield insignificant results when, in fact, accruals were used to manage earnings. 3.2 Empirically estimating bias in tests of earnings management As shown in equation (8), empirically assessing the degree of bias in a test of earnings management requires an estimate of η, the measurement error in the estimate of discretionary accruals. As argued earlier, using a balance sheet approach to estimating accruals will induce error if the balance sheet working capital changes do not articulate with accruals from the income statement that are reflected in the statement of cash flows. The fact that in the post- SFAS 95 period we are able to measure both ACC bs and ACC cf allows us to estimate the resulting error in estimated discretionary accruals. Because we only have ten years of post-sfas 95 data available, however, a cross-sectional version of the modified Jones model is used to estimate discretionary accruals. 10 Thus, the following models are estimated by year and industry: 11 ACC bs = α + β 1 ( Rev) + β 2 PPE + ε bs ACC cf = α + β 1 ( Rev) + β 2 PPE + ε cf (10a) (10b) Where Rev is the change in revenue, and PPE is the level of gross property, plant, and equipment. The parameters in equations (10a) and (10b) are estimated over sub-samples of firmyears without a specific non-articulation event (for example, all firm-years without a merger or acquisition or divestiture). The proxies for non-discretionary accruals under the balance sheet and cash flow approach are then computed as follows: 10 This estimation procedure is a cross-sectional adaptation of the modified-jones model introduced by Dechow, Sloan, Sweeney (1995). It is similar to the cross-sectional Jones model used in Teoh, Welch, Wong (1998). 11 The industry classifications are those used in Barth, Beaver, Hand, and Landsman (1999). 15
NDACC NDACC bs cf = α + β = α + β 1 1 ( Re v AR) + β ( Re v AR) + β 2 2 PPE PPE (11a) (11b) Where Rev - AR represents the change in revenue less the change in accounts receivable, and PPE is the level of gross property, plant, and equipment. 12 Discretionary accruals are computed as total accruals less non-discretionary accruals under both the balance sheet (DACC bs ) and cash flow (DACC cf ) approaches. By comparing the estimates of discretionary accruals under the balance sheet and cash flow approaches, we can estimate the error in discretionary accruals that results from using a balance sheet approach. 13 Specifically, η = DACC bs - DACC cf (12) If the error in ACC bs (i.e. DIFF) is uncorrelated with the regressors, the majority of this error will be captured by the residual, DACC bs (Greene, 1991). Empirically, this turns out to be the case as there is a 0.94 correlation between η and DIFF. Thus, much of the error from a balance sheet approach to estimating accruals, which is induced by non-articulation events, transfers to the discretionary accrual proxy in tests of earnings management. Equation (8) demonstrates that the magnitude and direction of the bias in a test of earnings management depends on the correlation between PART and η. However, we have already identified at least three non-articulation events that contribute to η : mergers and acquisitions, divestitures, and foreign currency translations. Table 3 presents evidence of the bias in tests of earnings management that arises when PART (the dummy variable used to 12 Using the cash flow approach, AR is taken from the statement of cash flows, while under the balance sheet approach it is measured as the difference in subsequent balance sheets. 13 Clearly this is not the only source of measurement error. The discussion in McNichols and Wilson is geared towards the measurement error induced by the choice of an empirical model used to measure discretionary accruals. Our analysis, however, examines a source of measurement error that is knowable and hence allows us to empirically evaluate the impact that the resulting coefficient bias can have on inferences made about earnings management. 16
identify periods of expected earnings management) coincides with each of these three nonarticulation events. Panel A displays the standard deviations of the measurement error proxy, η, and the three partitioning variables, PART merger, PART disc-op, and PART fc, as well as the correlations between η and each of these partitions. Using equation (8), the magnitude of the bias associated with the three non-articulation events can be estimated. This is shown in the right-most column of Panel A. The bias associated with PART merger is 1.65% of total assets, which is both statistically and economically significant, given accruals are, on average, approximately 4.5% percent of total assets. The bias associated with discontinued operations is similarly large but in the opposite direction, averaging -1.64% of total assets. The bias associated with foreign currency translations also tends to be negative, but is not as large averaging only -0.25% of total assets. While Panel A shows evidence of a significant bias entering the calculation of discretionary accruals when using DACC bs, it does not provide direct evidence of the potential impact on statistical inferences in tests of earnings management. To address this issue, Panel B of Table 3 provides results of tests of earnings management (see eqn.(8)) using the three previously identified partitions and discretionary accruals computed under both the balance sheet approach and from the statement of cash flows. By testing for significant discretionary accruals under both approaches, we can examine if any statistical inferences change due to measurement error in discretionary accruals under the balance sheet approach. A priori, we have no reason to believe that there will be significant earnings management associated with any of the three nonarticulation events used as partitioning variables. Therefore, when accruals are measured correctly from the statement of cash flows, we expect the coefficient on PART to be insignificantly different from zero. 17
Discretionary accruals are regressed on each of the three partitioning variables on an annual basis. Additionally, the significance across all years is reported by taking the average of the sampling distribution of the ten individual-year parameter estimates. From Panel A, we expect that the balance sheet estimation method biases the coefficient on PART, especially for merger/acquisition firms and discontinued operation firms. More importantly, however, Panel B demonstrates that the researcher s inference of whether or not earnings management exists can change depending on how the accruals are measured. Specifically, the first two columns of Panel B demonstrate that a partition on mergers and acquisitions shows evidence of significant income increasing earnings management (in 10 of 10 individual years and across all years). However, the same partition using accruals from the statement of cash flows shows no evidence of earnings management. Similarly, the next two columns in Panel B demonstrate that a partition on discontinued operations shows evidence of significant income decreasing earnings management (in 8 of 10 individual years and across all years) when using accruals under the balance sheet approach, while a similar partition using accruals from the statement of cash flows shows evidence of earnings management in only one year. Finally, the last two columns of panel B demonstrate that a partition on foreign currency translation does not appear to be directionally biased, with insignificant results across all years under both the balance sheet and cash flow approaches. The results in Table 3 assume a 100% overlap between the partitioning variable used to test for the existence of earnings management and the three non-articulation events considered in this study. In general, however, the partitioning variable chosen by the researcher will rarely coincide perfectly with these non-articulation events. Rather, the sample will often be only partially contaminated by the non-articulation events, and the degree of contamination will vary 18
depending on the event chosen. For example, studies examining financially distressed firms might contain a higher proportion of firms with discontinued operations than the population in general, while studies examining rapidly growing firms might contain a higher proportion of merger and acquisition firms than the population in general. Therefore, this section attempts to estimate the potential bias in tests of earnings management where the partitioning variable is imperfectly correlated with the non-articulation events. The procedure we use is described only for the mergers and acquisitions subsample of firms, but the identical procedure is used for the discontinued operations subsample. We start by taking a random sample of 250 firms from the subsample of firms that do not have a non-articulation event in a given year. This is our 0% contamination sample. 14 Using 1000 iterations of this procedure, we then measure the amount of the bias in discretionary accruals as a percent of total assets (η in equation 12), and the probability of committing a type I error if there is no earnings management present. Based on 1000 trials, the rejection frequencies (i.e. type I error rates) are reported at the 1% and 5% significance levels using a one tailed t-test. Next, we increase the percent contaminated to 10% by taking a random sample of 25 firms from the merger and acquisition subsample and a random sample of 225 from the subsample of firms without a non-articulation event, repeat this procedure 1000 times, and measure the magnitude of the bias and the rejection frequency at this level of contamination. We continue this procedure until the percent of the sample contaminated by mergers and acquisitions is 100%. Results of this procedure for mergers and acquisitions are depicted in Figure 2 and the results for discontinued operations are depicted in Figure 3. Examining the merger and 14 A sample size of N=250 is arbitrarily chosen to approximate a typical sample size in a study of earnings management. As N is increased (decreased), the rejection rate frequencies will increase (decrease), although the magnitude of the bias will not be affected. 19
acquisition subsample in Figure 2, the first point to note is that at a 0% contamination level, the bias in panel A is minimal (0.03% of total assets) and the test statistics in panel B are relatively well specified (3.4% and 1.8% relative to the expected 5% and 1% levels). As the percent of the sample contaminated increases, however, the bias increases and the probability of committing a type I error increases significantly. For example, even when the sample is only 30% contaminated, the probabilities of committing type I errors in the absence of earnings management are approximately 45% and 33%, relative to the expected levels of 5% and 1%. Moreover, as the percent of the sample contaminated by mergers and acquisitions rises above 50%, the type I error rates increase to 90% and greater. Thus, at these levels of contamination, even if no earnings management were present, the researcher would almost certainly conclude that earnings had indeed been managed. Results for the discontinued operations in Figure 3 are similar. At the 0% contamination level, the bias in panel A is 0.05% of total assets and the probabilities of committing a type I error in panel B are 6.5% and 3.6% relative to the expected 5% and 1% levels. As the percent of the sample contaminated increases, however, the bias for discontinued operation firms decreases from approximately 0.19% of total assets at the 10% contamination level, to 1.33% of total assets at the 100% contamination level. Additionally, although the rejection frequencies do not rise as quickly for the discontinued operations subsample as they do for the mergers and acquisitions subsample, at 30% contamination the rejection frequencies are still 36% and 25%, relative to the expected 5% and 1% levels. Thus, it is obvious that even a relatively modest level of contamination by non-articulation events can have a potentially large impact on both the magnitude of the estimated accruals that are managed, as well as the inferences that the researcher draws from these results. 20
3.3 Implications for extant earnings management literature. The results of the previous section have significant implications for extant earnings management studies using a balance sheet approach to estimating accruals. Specifically, the measurement error in total accruals and the resulting coefficient bias for various partitions could lead the researcher to conclude that significant earnings management exists, when in fact there is none. Given the results of the previous section, this is most problematic for partitions that are correlated with either mergers and acquisitions or discontinued operations. For example, Rees, Gill and Gore (1996) investigate the potential for earnings management associated with asset write-downs using a balance sheet approach to measuring accruals (p. 159, equation 1). The authors hypothesize that if the primary motive for an asset write-down is opportunistic, firms are likely to concurrently manage operating accruals downwards. Accordingly, they test for income decreasing discretionary accruals. Yet, as noted by the authors, the majority of write-downs in their sample are part of an overall firm restructuring. Thus, the partition chosen by Rees, Gill, and Gore (i.e. asset write-downs) is likely to be positively correlated with the presence of discontinued operations. Furthermore, as shown above, discretionary accruals determined under the balance sheet approach are negatively biased for firms with discontinued operations. Therefore, it is predicted that this combination will induce a negative bias to their test of earnings management, which is exactly the predicted direction of the alleged earnings management. To empirically assess the impact of this potential bias, we collect a sample of asset writedowns from 1988-1993 using the same data source (NAARS), search string, and selection criteria as Rees et al. (1996, pp.158-159). 15 This procedure results in a sample of 307 firm-year 15 Rees et al. (1996) examine data from 1987-1992. However, to ensure that all cash flow statement data are available to correctly determine accruals, we use data from 1988-1993. 21
observations. Operating accruals are as defined as in Rees et al. (1996) whereby the definition only encompasses working capital accruals and depreciation expense. Additionally, accruals are adjusted for the impact of the restructuring charge on current liabilities. Finally, the control sample consists of all firms within the same two-digit SIC code in the year of the write-down. The test for abnormal accruals is then conducted using a separate regression for each sub-sample of observations, where the sub-sample consists of a sample and a control group, as follows: 16 ACC t = α 0 + α 1 ( Rev t ) + α 2 (PPE t ) +α 3 (CFO t ) +β 1 (PART t ) + ε t (13) where PART equals one for firms with an asset write-down, and zero otherwise. The original results from the Rees et al. (1996) study are shown in Table 4, Panel A. Abnormal operating accruals in the year of the write-down (i.e. the average coefficient on PART) are -3.05% of total assets. Additionally, in their 120 separate cross-sectional regressions estimated using equation (13), the coefficient on PART is negative 67.5% of the time. Using the balance sheet approach and sample we collected from 1988-1993, we first replicate the Rees et al. finding using a balance sheet estimation approach (Panel B) and then repeat the tests using a cash flow statement approach (Panel C). Using the balance sheet approach in Panel B, we find results that are qualitatively similar to those reported in Rees et al. (1996). There appears to be significant income-decreasing abnormal accruals of approximately 2.24% of total assets, and negative abnormal accruals in 62.3% of our 125 cross-sectional regressions. In Panel C, however, when operating accruals are taken directly from the statement of cash flows, the results are dramatically different. In the year of the asset write-down, abnormal operating accruals are only 0.40% of total assets, which is not significantly different from zero at conventional significance levels. Moreover, abnormal accruals are negative only 52.8% of the time, which is 16 This estimation procedure differs slightly from the procedure outlined in equations (10) and (11), but is used to maintain consistency with the original Rees et al. (1996) study. 22
not significantly different from 50% using a binomial test. Thus, the presence of discontinued operations has a significant impact on the magnitude of income-decreasing discretionary accruals estimated via the balance sheet approach. As shown, this can readily lead the researcher to conclude that accruals are being managed downward when, in fact, properly measured accruals indicate otherwise. Other studies of earnings management subject to similar concerns about negative bias would include studies that test for income decreasing earnings management in a sample that contains a disproportionate number of financially distressed firms (e.g. Perry and Williams 1994). In this case, the firms hypothesized to be engaged in earnings management are also likely to be divesting of unprofitable segments or divisions. As we have shown, this divestiture activity induces a negative bias to ACC bs estimates that is in the same direction as the hypothesized earnings management, thus confounding their hypothesis tests. Moreover, as shown earlier, the probability of committing a type I error increases substantially, even if the sample is only 30-40% contaminated by discontinued operations. Similar concerns can be raised about studies that test for income-increasing earnings management around events that are positively correlated with mergers and acquisitions. For example, Teoh, Welch, and Wong (1998) and Rangan (1998) examine earnings management in anticipation of seasoned equity offerings (SEOs). These studies investigate whether earnings management might contribute to the long-run underperformance of SEOs previously identified in the finance literature (e.g. Loughran and Ritter 1995). While on the surface, there is no obvious relation between mergers and acquisitions and SEOs, we find that firms listed by Securities Data Corporation as participating in SEOs are 55.1% (64.7%) more likely to be involved in a merger or acquisition in the year prior to (in the year of) an SEO than the population in general. Thus, 23
an estimate of discretionary accruals in this sub-sample of firms is more likely to be upward biased relative to the population in general, due to the fact that there is a higher proportion of merger and acquisition firms in this sub-sample. To investigate whether this bias impacts any inferences, Table 5 examines the impact of the balance sheet estimation approach for 775 firms conducting seasoned equity offerings from 1988-1997. Teoh, Welch, and Wong (1998) find that, on average, accruals are managed upwards in the year prior to and in the year of a seasoned equity offering. Panel A re-examines this issue by comparing the discretionary accruals of the SEO sample to discretionary accruals of a control sample in the year prior to a seasoned equity offering. 17 Because of our requirements for cash flow statement data availability, our sample periods are largely non-overlapping, with our period covering 1988-1997, and their period covering 1970-1989. 18 The first line of Panel A reveals that under the balance sheet approach, discretionary accruals are significantly larger in period t-1 for the firms that issue a seasoned equity offering in period t as compared to a set of control firms (DACC bs,seo DACC bs,control = 0.73, t=2.90). Using the cash flow approach to measuring accruals, however, we find no significant difference in discretionary accruals between the SEO firms and the control firms (DACC cf,seo DACC cf,control = 0.22%, t=1.13). Thus, the inference as to whether firms manage earnings upward in the year prior to a seasoned equity offering changes depending on whether the balance sheet or cash flow approach is used to estimate accruals. 19 17 The control sample is matched on the basis of net income in the year prior to the seasoned equity offering. Discretionary accruals are computed using the cross-sectional version of the modified Jones model, as outlined in equations (10) and (11). 18 Another issue is that their balance sheet accrual measure cannot be replicated during our sample period because total funds from operations (Compustat #110) is only reported prior to the issuance of the statement of cash flows. Thus, we use the ACC bs and ACC cf measures discussed earlier when replicating the study. 19 Additional findings not reported in Table 5 show that in the year of the SEO, discretionary accruals as a percent of total assets are 138% larger when estimated using the balance sheet approach (0.0238 vs. 0.0100, p<.01). 24
Panel B demonstrates that within the sample of SEO firms, it is those with mergers and acquisitions that cause the measure of discretionary accruals to be significantly positive under the balance sheet approach. For the 251 firms involved in a merger or acquisition, mean abnormal accruals are 1.69% of total assets. Conversely, for the 524 firms not involved in a merger or acquisition, discretionary accruals are 0.03% of total assets. Note that the magnitude of estimated discretionary accruals for the non-merger firms in Panel B is comparable to the discretionary accruals of 0.04% reported in Panel A for the SEO sample as a whole when the cash flow approach is used. Thus, one important implication of this result is that when cash flow statement data are not available to determine accruals (e.g. prior to 1988), a viable alternative is to separately analyze those firms involved in non-articulation events. Comparison of results for these firms with those not impacted by non-articulation events can be used to determine the potential contamination cause by balance sheet accrual estimation. In summary, our results suggest that inferences regarding the existence and magnitude of earnings management in anticipation of SEOs can change when the balance sheet approach is used to estimate accruals and the sample contains a significant number of firms involved in mergers and acquisitions. Moreover, the underperformance of SEO stocks which Teoh et al. (1998) and Rangan (1998) attribute to the market corrections related to earnings management is likely to be confounded by the well-documented long-run underperformance of firms involved in mergers and acquisitions (see, for example, Agrawal and Jaffe 1999). IV. Tests Using Discretionary and Non-Discretionary Accruals as Explanatory Variables Another stream of literature that is potentially affected by measurement error in accruals estimates includes studies using discretionary and non-discretionary accruals as explanatory 25
variables in a second-stage regression with returns as the dependent variable. The basic question addressed by such studies is whether discretionary accruals are priced differently from nondiscretionary accruals or cash flows. For example, Guay, Kothari, Watts (1996) and Subramanyam (1996) regress returns on non-discretionary earnings (operating cash flows plus non-discretionary accruals), discretionary accruals, non-discretionary accruals, and operating cash flows all determined using a balance sheet approach. As shown in the previous section, however, each of these variables will embed some (or all) of the error resulting from a balance sheet estimation approach. When these variables are used as regressors and the measurement errors embedded in these measures are correlated across variables, the resultant coefficient estimates are biased, but the direction of the bias is difficult to predict (Greene 1991, p.284). To provide some evidence on the impact of this measurement error, we estimate the same series of regressions used by Subramanyam (1996). As in the previous section, discretionary accruals are measured using the modified-jones model. Table 6 presents the results of estimating a series of equations using both the balance sheet and cash flow estimation approaches. In should be emphasized that while our approach is the same approach used by Subramanyam, our sample periods are largely non-overlapping (1973-1993 vs. 1988-1997) and hence the results may not be directly comparable. Panel A of Table 6 shows the basic regression of returns on accruals and cash from operations. Under a balance sheet approach, the coefficient on accruals is upward biased by about 10.4% (1.06 versus 0.96), and the coefficient on cash flows is downward biased by about 7.1% (1.06 vs. 1.12). 20 More importantly, using a balance sheet approach a test of equality of the coefficients would lead the researcher to conclude there is no difference between the 26
coefficients on accruals and cash flows (p-value=0.987), while using a cash flow approach shows a significant difference (p-value = 0.022). 21 Panel B examines the relation between size-adjusted returns and non-discretionary earnings and discretionary accruals. Here, the coefficient on non-discretionary earnings is downward biased by only 6.9% (1.08 vs.1.16), while the coefficient on discretionary accruals (DACC) is upward biased by approximately 23.5%(1.00 vs. 0.81). Again, a t-test on the equality of coefficients using the balance sheet estimation of accruals yields an insignificant difference between the coefficient on NDE and DACC (p-value = 0.44), while using accruals from the cash flow statement yields a statistically significant difference between these coefficients (p-value < 0.01). Panel C breaks down earnings into cash flow from operations (CFO), non-discretionary accruals (NDACC), and discretionary accruals (DACC). In this case, the coefficients on NDACC and CFO are not biased by large amounts (8.9% and 6.3% respectively), but the coefficient on DACC is upward biased by approximately 24% (0.98 vs. 0.79). Tests of equality of parameter estimates are largely similar, although a test of the difference between the coefficient on DACC and the coefficient on CFO is insignificant under a balance sheet estimation approach (p-value = 0.34) and statistically significant under a cash flow approach (pvalue = 0.002). In summary, when discretionary and non-discretionary accruals based on balance sheet estimates of accruals are used as regressors in a second stage regression, the coefficients are biased to varying degrees and the bias can contaminate statistical inferences made on the basis of these coefficient estimates. 20 To assess the significance of the bias, proxies for the measurement error (i.e. the DIFF variable) are added to the balance sheet model. In all cases, unreported coefficients on the measurement error proxy are statistically significant. 21 All tests in this section are based on White heteroskedasticity-consistent standard errors. 27
V. Tests of Market Pricing of Accruals Tests of market pricing of accruals [Sloan (1996), Collins and Hribar (2000) and Xie (1999)] are typically implemented by taking portfolio positions based on ranked values of accruals (scaled by lagged total assets). Ten equal sized portfolios are formed by ranking firms from the most negative accruals (Portfolio 1 = PORT1) to the most positive accruals (Portfolio 10 = PORT10). Zero investment hedge portfolios are formed by taking a long position in PORT1 firms and a short position in PORT10 firms. Evidence of market mispricing is indicated if the hedge portfolio yields positive abnormal returns net of transactions costs over the subsequent holding period. In the present context, we are concerned with how measurement error in ACC bs affects the ranking of firms conditional on accruals estimated under the balance sheet approach, and in particular, firms assigned to the extreme accruals decile portfolios. Panel A of Table 7 provides evidence on the percentage of firm-years that fall into a different decile ranking using ACC bs estimates of accruals compared to the correct ACC cf accruals measurements. Because events causing non-articulation errors in ACC bs can cause rankings to differ in either direction, we report the percentage of firm-years that change decile rankings irrespective of direction. Results are presented for all firm-years and for firm-years that fall into the extreme deciles only based on ACC bs. Across all firm-years, we find that only 38.2% of the time do the ACC bs accrual decile rankings agree with the ACC cf rankings. Thus, 61.8% (100% - 38.2%) of the observations fall into a different decile ranking when measured correctly (ACC cf ) as compared to using the balance sheet approach (ACC bs ) to estimating accruals. Interestingly, 27.9% of the observations shift by at least two deciles when measured by ACC cf compared to ACC bs. More relevant to tests 28
of market mispricing of accruals, we find that 38.7% (100% - 61.3%) of those observations assigned to the extreme accruals deciles (i.e., decile 1 or decile 10) using ACC bs, are misclassified. Thus, we conjecture that studies which rely on ACC bs measures to form hedge portfolios [e.g., Sloan (1996) and Xie (1999)] may understate the extent of market mispricing of accruals because a non-trivial proportion of the firms are misclassified as extreme accruals firms. Panel B of Table 7 examines the accrual portfolio misclassification as a result of the three non-articulation events described earlier. Given the bias documented for these non-articulation events, the misclassification is as expected. Specifically, 513 firms involved in a merger or acquisition end up in the highest accrual portfolio, but 284 (55.4%) of those firms are misclassified and fall into a lower accrual decile under the cash flow approach. The misclassification of high accrual firms is larger, both in absolute number and in percentage terms, than is the number of merger firms that are misclassified in the lowest accrual decile (83 firms, 49.7%). The opposite effect is observed for firms with discontinued operations. Here, 314 firms with discontinued operations fall into lowest accrual portfolio under the balance sheet approach, but 190 (60.5%) of these firms are misclassified and are classified into a higher accrual decile under the cash flow approach. Again, the misclassification of large negative accrual firms is greater, both in absolute number and in percentage terms, than the number of discontinued operation firms that are misclassified in the highest accrual decile (61 firms, 49.2%). Firms with foreign operations have a slightly larger proportion of firms in the lowest accrual portfolio, with 253 firms falling into the lowest accrual portfolio. Again, 125 (49.4%) of these firms are misclassified, whereas in the highest accrual portfolio only 82 firms (39.0%) are misclassified. 29
5.1 Assessing the impact of portfolio misclassification on the accrual anomaly. To assess the effect of measurement error in ACC bs on tests of accruals mispricing, we replicate the hedge portfolio analysis in Sloan (1996) using annual data and the hedge portfolio analysis in Collins and Hribar (2000) using quarterly data. We construct zero-investment hedge portfolios by taking a short position in PORT10 firms (highest decile of firms with most positive accruals) and a long position in PORT1 firms (lowest decile of firms with most negative accruals). The analysis is repeated twice in both the annual and quarterly settings: once using decile rankings based on ACC bs measures (as in Sloan (1996)), which are contaminated by the non-articulation problems noted above, and again using the correct ACC cf measures (as in Collins and Hribar (2000)). Our sample is comprised of all NYSE/AMEX firms on the Compustat Primary, Supplementary and Tertiary Annual Industrial File and Research File with required earnings and accruals data over the ten-year period from 1988 to 1997. Figure 4 replicates Sloan s (1996) results and displays the average size-adjusted returns accruing to both the ACC bs and the ACC cf hedge portfolios cumulated over the twelve months following the release of the annual report. As conjectured, the average annual size-adjusted return accruing to the ACC cf hedge portfolios is 11.88% and is roughly 36% higher than the 8.72% average return earned by the ACC bs -based strategy. Thus, when measured correctly, the accruals-based hedge portfolio yields much stronger evidence of market mispricing than documented in the previous literature. Figure 5a replicates the accruals-based hedge portfolio results in a quarterly setting for the 36 quarters from 1988 to 1996 in Collins and Hribar (2000). The average size-adjusted returns are measured over the two quarters subsequent to portfolio formation and are based on 30
using the correct quarterly accruals measures from the statement of cash flows. As shown, the average two-quarter return is 5.56% or an annualized return of 11.12%. In contrast, the average two-quarter abnormal return accruing to hedge-portfolios based on ACC bs presented in Figure 5b are much smaller and display considerably more variability. As shown, the average two-quarter return is 2.99%, or an annualized return of 5.98%. Note that the ACC bs -based strategy yields negative returns in 12 of the 36 quarters while the ACC cf -based strategy yields negative returns in only 5 quarters and they are much smaller in magnitude. Again, these results indicate that the measurement error introduced when using balance sheetbased accruals estimates can substantially bias tests of market mispricing towards zero. VI. Summary and Research Implications Accruals measurement plays a central role in a considerable body of research in accounting. Much of this research relies on estimates of accruals based on the presumed articulation between changes in balance sheet working capital accounts and accrued revenues and expenses on the income statement. However, this presumed articulation breaks down when non-operating events/activities like mergers and acquisitions, divestitures and translation of foreign subsidiary accounts are present. This paper demonstrates that the frequency and magnitude of errors introduced when using balance sheet-based accruals estimates can be substantial. Our findings have implications for studies designed to detect earnings management, the estimation of discretionary and nondiscretionary accruals, and the market s pricing (and mispricing) of these accruals components. The most significant finding relates to studies examining earnings management. The prevalence of the balance sheet approach to estimating accruals in these studies suggests that their results should be reevaluated in light of the potential 31
impact of mismeasured accruals. This is especially pertinent in cases where the partitioning variable used to identify instances of earnings management is correlated with mergers and acquisitions or discontinued operations. Our results suggest that many studies concerned with possible differential pricing implications of discretionary and nondiscretionary accruals are adversely affected by measurement errors introduced by the balance sheet approach to accruals measurement. In particular, tests that discretionary and nondiscretionary accruals are priced the same are less likely to be rejected when accruals are measured using the balance sheet approach as compared to when they are measured correctly from the cash flow statement. Finally, our findings suggest that tests of market mispricing of accruals suffer from significant classification errors of extreme accruals firms when accruals are estimated using the balance sheet approach. Using correctly measured accruals, we find that the degree of mispricing is significantly greater than that documented using balance sheet accruals. Going forward, we expect that accruals will continue to be the main object of interest in a broad cross-section of literature in accounting. Our results suggest that it would be prudent for researchers to rely on accruals measures taken directly from the cash flow statement. If the research context requires use of pre-sfas 95 data, additional specification tests should be conducted to control for possible errors in accruals measurements introduced by non-articulation events. Failing to do so may lead to unreliable tests and unwarranted inferences. 32
Bibliography Agrawal, A, and J. Jaffe, 1999, The Post-merger Performance Puzzle, University of Pennsylvania Working Paper. Bahnson, P., P. Miller and B. Budge, 1996, Nonarticulation in Cash Flow Statements and Implications for Education, Research, and Practice, Accounting Horizons 10(4): 1-15. Barth, M., B. Beaver, J. Hand, and W. Landsman, 1999, Accruals, Cash Flows, and Equity Values, Review of Accounting Studies, 4(3/4):205-229. Collins, D.W. and P. Hribar, 2000, "Earnings-based and Accrual-based Market Anomalies: One Effect or Two?" Journal of Accounting and Economics 29(1), forthcoming. DeAngelo, L., 1986, Accounting numbers as market valuation substitutes: A study of management buyouts of public stockholders, The Accounting Review 61: 400-420. DeAngelo, L., 1988, Managerial Competition, information costs, and corporate governance: The use of accounting performance measures in proxy contests, Journal of Accounting and Economics 10: 3-36. Dechow, P., Accounting Earnings and Cash Flows as Measures of Firm Performance: The Role of Accounting Accruals, Journal of Accounting and Economics (July 1994), pp. 3-42. Dechow, P., R. Sloan and A. Sweeny. 1995. Detecting Earnings Management. The Accounting Review April:193-226. Defond, M. and K. Subramanyam, 1998, Auditor Changes and Discretionary Accruals, Journal of Accounting and Economics 25(1): 35-68. Dritina, R. and J. Largay, 1985, Pitfalls in calculating cash flows from operations. The Accounting Review 60(2): 314-326. Erickson, M. and S. Wang, 1999, Earnings Management by Acquiring Firms in Stock for Stock Mergers, Journal of Accounting and Economics 27(2): 149-176. Francis, J., Hanna, D., and Vincent L., 1996, Causes and Effects of Discretionary Asset Write- Offs, Journal of Accounting Research Supplement, pp. 117-134. Greene, W., 1991, Econometric Analysis, Prentice-Hall, Englewood Cliffs, NJ. Guay, W., S.P. Kothari, and R. Watts, 1996, A Market-Based Evaluation of Discretionary Accrual Models, Journal of Accounting Research Supplement 34: 83-115. 33
Healy, P., 1985, The Effect of Bonus Schemes on Accounting Decisions, Journal of Accounting and Economics 7: 85-107. Huefner, R., J. Ketz, and J. Largay, 1989, Foreign Currency Translation and the Cash Flow Statement, Accounting Horizons 3(2): 66-75. Jones, J., 1991, Earnings Management During Import Relief Investigations, Journal of Accounting Research Supplement 26: 1-31. Kang, S.H. and K. Sivaramakrishnan, 1995, Issues in Testing Earnings Management and an Instrumental Variable Approach, Journal of Accounting Research Autumn: pp. 353-368. Loughran, T, and J. Ritter, 1995, The New Issues Puzzle, Journal of Finance (50): 22-31. McNichols, M. and P. Wilson, 1988, Evidence of Earnings Management from the Provision for Bad Debts, Journal of Accounting Research 26: 1-31. Perry, S. and T. Williams, 1994, Earnings Management Preceding Management Buyout Offers, Journal of Accounting and Economics 18(2): 157-180. Rangan, S., 1999, Earnings Management and the Performance of Seasoned Equity Offerings, Journal of Financial Economics 50: 101-122. Rayburn, J., 1986, The Association of Operating Cash Flow and Accruals with Security Returns, Journal of Accounting Research: 112-137. Rees, L., Gill, S. and Gore, R., 1996, An Investigation of Asset Write-Downs and Concurrent Abnormal Accruals, Journal of Accounting Research (Supplement): 157-169. Revsine, L., D. Collins, and B. Johnson, 1999, Financial Reporting and Analysis, Upper Saddle River, NJ: Prentice-Hall. Sloan, R., 1996, Do Stock Prices Fully Reflect Information in Accruals and Cash Flows about Future Earnings? The Accounting Review 71(3): 289-316. Subramanyan, K.R., 1996, The Pricing of Discretionary Accruals, Journal of Accounting & Economics 22(1-3): 249-281. Teoh, S., I. Welch, and T. Wong, 1998, Earnings Management and the Underperformance of Seasoned Equity Offerings, Journal of Financial Economics (50): 63-99. Wilson, P., 1987, The Incremental Information Content of the Accrual and Funds Components of Earnings after Controlling for Earnings, The Accounting Review 62(2): 293-322. Xie, H., 1999, Are Discretionary Accruals Mispriced? A reexamination, University of Arizona Working Paper. 34
Exhibit 1. H.J. Heinz Company and Subsidiaries Comparison of Accruals Determination from Change in Balance Sheet Accounts versus Cash Flow Statement Fiscal Year Ending April 30, 1997 Fiscal Year Ended April 30,1997 (Dollars in thousands) Statement of Cash Flows Income Effect Adjustments to arrive at CFO based on changes in the Balance Sheet Income Effect OPERATING ACTIVITIES: Net income $ 301,871 $301,871 Adjustments to reconcile net income to cash provided by operating activities: Depreciation 244,388 fl Amortization 96,102 fl Deferred tax (benefit) provision (33,450) Gain on sale of New Zealand ice cream (85,282) business and U.K. real estate Provision for restructuring 647,200 fl Other items, net (42,527) Changes in current assets and liabilities, excluding effects of acquisitions and divestitures: Receivables 74,445 fl 89,000 fl Inventories (5,329) 61,452 fl Prepaid expenses and other current 5,094 fl (36,809) assets Accounts payable 18,003 fl (5,183) Accrued liabilities (182,555) (42,063) Income taxes (162,962) (79,538) Working Capital Accruals +253,304 +13,141 Cash from Operations 874,998 35
Table 1. Summary statistics for accruals computed under the balance sheet and cash flow statement approaches by type of non-articulation event. Based on 14,558 firm-years over the years 1988-1997. Fractiles of Distribution Panel A: Merger & Acquisition Sample (2,991 firm-years) Mean Std. Dev. 25% Median 75% TACC bs / TA t-1-1.44 6.67-5.34-1.82 2.29 TACC cf / TA t-1-2.79 5.67-5.69-2.74 0.84 DIFF / TA t-1 1.35 4.62-1.01 1.21 3.38 DIFF / EBXI 28.32 214.60-19.25 16.95 64.34 Panel B: Discontinued Operations Sample (1,277 firm-years) TACC bs / TA t-1-6.29 7.49-10.31-5.86-2.03 TACC cf / TA t-1-4.22 6.17-7.50-4.11-0.97 DIFF / TA t-1-2.07 5.67-4.96-1.45 1.06 DIFF / EBXI -52.91 269.36-96.30-29.41 19.85 Panel C: Foreign Operations Sample (2,812 firm-years) TACC bs / TA t-1-4.01 6.09-7.31-4.00-0.77 TACC cf / TA t-1-3.62 5.58-6.63-3.60-0.54 DIFF / TA t-1-0.39 3.82-1.98-0.12 1.30 DIFF / EBXI -11.25 183.92-39.53-1.65 24.66 Panel D: Firms without a non-articulation event (7,233 firm-years) TACC bs / TA t-1-4.11 6.62-7.79-4.28-0.67 TACC cf / TA t-1-3.92 6.25-7.50-4.15-0.41 DIFF / TA t-1-0.16 3.58-1.21 0.11 1.31 DIFF / EBXI -8.06 144.22-23.14 1.85 25.19 Note: All ratios are reported after deleting the first and ninety-ninth percentiles. Variable Definitions: TACC bs equals the change in non-cash current assets less the change in current liabilities excluding short-term debt, less depreciation expense. TACC cf equals net income before extraordinary items less cash flow from operations, taken from the statement of cash flows. DIFF equals the difference between total accruals measured under the balance sheet and cash flow approaches. TA t-1 equals total assets at the end of the previous fiscal year. EBXI equals net income before extraordinary items and discontinued operations. * is the absolute value operator. 36
Table 2. Multivariate tests of the effect of mergers and acquisitions, discontinued operations, and foreign currency translations on the difference between accruals computed under the balance sheet approach and the cash flow statement approach. Based on 14,558 firm years over the years 1988-1997. Model 1: DIFF/TA t-1 = α + β 1 D MERGER + β 2 D DISC-OP + β 3 D FC + ε Model 2: DIFF/TA t-1 = α + β 1 D MERGER + β 2 D DISC-OP + β 3 D FC + β 4 D LARGE_MERGER + ε Coefficient Estimate Model 1 Model 2 (t-statistic) Intercept 0.0012 (16.15) 0.0012 (16.16) D MERGER 0.0148** (15.95) D DISC-OP -0.0156** (-10.62) D FC -0.0049** (-4.38) 0.0144 ** (15.49) -0.0157** (-10.67) -0.0049** (-4.64) D LARGE_MERGER 0.0091** (4.21) Adj. R 2 3.60% 3.62% * Significant at the α=0.05 level; ** significant at the α=0.01 level. Variable Definitions: DIFF equals the difference between total accruals measured under the balance sheet and cash flow approaches. TA t-1 equals total assets at the end of the previous fiscal year. D MERGER equals 1 if the firm-year contains a merger or acquisition (as measured by the Compustat annual footnote code #1), 0 otherwise. D DISC-OP equals 1 if the firm year contains discontinued operations greater than $10,000 (as measured by Compustat data item #66), 0 otherwise D FC equals 1 if the firm-year contains foreign currency gains or losses (as measured by Compustat annual data item #150), 0 otherwise D LARGE_MERGER equals 1 if the firm-year contains a significant merger or acquisition as classified by Compustat, 0 otherwise. 37
Table 3. Annual significance levels of different event partitions under the balance sheet and cash flow approaches to computing accruals. Based on 14,558 firm years over the years 1988-1997. Panel A: Potential bias in earnings management studies using the three non-articulation events as partitions. Standard Deviations (σ) σ η / σ PART Correlations (ρ) Bias as % TA t-1 (r *s h /s PART ) σ(η) 0.043 σ(part merger ) 0.409 0.105 ρ(η, PART merger ) 0.157 1.651% σ(part disc-op ) 0.272 0.158 ρ(η, PART disc-op ) -0.104-1.644% σ(part fc ) 0.392 0.110 ρ(η, PART fc ) -0.023-0.252% Panel B: Tests of earnings management using mergers and acquisitions, discontinued operations, and foreign currency translations as partitioning variables. BS Model: DACC bs = β 0 + β 1 PART + ε CF Model: DACC cf = γ 0 + γ 1 PART + ε PART merger PART disc-op PART fc Year BS Model (β 1 ) CF Model (γ 1 ) BS Model (β 1 ) CF Model (γ 1 ) BS Model (β 1 ) CF Model (γ 1 ) 1988 2.01%* -0.39% -0.21% 0.75% -1.08%* -0.00% 1989 1.85%** -0.27% -2.70%** -0.20% -0.10% 0.68%* 1990 1.72%** 0.06% -2.32%** -1.24%** 0.43% 0.48% 1991 1.94%** 0.31% -1.29%** 0.21% -0.14% 0.00% 1992 1.71%** 0.06% -1.49%** 0.41% -0.87%** -0.40% 1993 1.44%** -0.07% -2.16%** -0.14% -0.92%** -0.33% 1994 1.69%** 0.13% -0.74% 0.46% 0.14% 0.10% 1995 1.34%** -0.12% -1.08%** 0.22% 0.13% 0.39% 1996 2.50%** 0.87% -0.99%* 0.46% -0.51% 0.08% 1997 1.36%** -0.12% -1.38%** 0.23% 0.47% 1.01%** All Years 1.76%** 0.09% -1.43%** 0.11% -0.24% 0.20% * Significant at the α=0.05 level, two tailed test; ** significant at the α=0.01 level, two tailed test. Variable Definitions: DACC bs (DACC cf ) = discretionary accruals computed using a cross-sectional adaptation of the modified Jones model under the balance sheet (cash flow statement) approach; η = DACC bs - DACC cf. PART merger is an indicator variable equal to 1 if a firm reports a merger or acquisition in a given year, 0 otherwise. PART disc-op is an indicator variable equal to 1 if a firm reports discontinued operations in a given year, 0 otherwise. PART fc is an indicator variable equal to 1 if a firm has foreign currency gains or losses in a given year, 0 otherwise. 38
Table 4. Testing for abnormal operating accruals concurrent with asset write-downs. This table examines discretionary operating accruals in the year of an asset write-down under both the balance sheet and cash flow statement approaches. Total accruals are adjusted for the impact of current liabilities associated with the restructuring charge. Model: TACC = α 0 + α 1 ( Rev) + α 2 (PPE) +α 3 (CFO) +β 1 (PART) + ε i Panel A. Results as reported in Rees, Gill, Gore (1996, Table 3) Number of Separate Regressions = 120 DRev PPE CFO PART Avg. 0.141-0.023-0.457-0.030 Coefficient % positive 90.8 34.2 3.3 32.5 Z-statistic -5.54** Panel B. Write-down sample in the current paper using a balance sheet approach Number of Separate Regressions = 125 DRev PPE CFO PART Avg. 0.126-0.021-0.416-0.022 Coefficient % positive 86.7 27.5 6.7 37.7 Z-statistic -3.94** Panel C. Write-down sample in the current paper using a cash flow approach Number of Separate Regressions = 125 DRev PPE CFO PART Avg. 0.095-0.025-0.387-0.004 Coefficient % positive 95.2 17.6 2.4 47.2 Z-statistic -0.87 * Significant at the α=0.05 level; ** significant at the α=0.01 level. Variable Definitions: Rev equals the change in revenue from year t-1 to year t. PPE equals gross property plant and equipment in year t. CFO equals cash from operations in year t. PART is a partitioning variable equaling 1 if the firm has an asset write-down. Z-Statistic is computed as follows: Z = 1 N N t j= 1 k j / k j j, 2 where t j is the t-statistic for portfolio, N is the number of portfolios, and k j is the degrees of freedom for the corresponding t-statistic. 39
Table 5. Testing for Income-Increasing Discretionary Accruals in the year prior to a Seasoned Equity Offering (SEO) using the Balance Sheet and Cash Flow Approaches. The sample includes 775 firms involved in SEOs during 1988-1997, and a sample of control firms matched on the prior year s net income. All values are reported as a percent of total assets. Panel A: Differences in discretionary accruals between firms involved in an SEO and a control sample of firms that are matched on net income in the pre-offering year. SEO sample Control Difference t-statistic Balance Sheet Approach 0.63% -0.10% 0.73% 2.90** Cash Flow Approach 0.04% -0.19% 0.22% 1.13 Panel B: Differences in discretionary accrual calculations using the Balance Sheet Approach when SEO firms are involved in mergers and acquisitions N DACC bs t-statistic Firms Involved in a Merger/Acquisition 251 1.75% 3.95** Firms not Involved in a Merger/Acquisition 524 0.08% 0.09 Difference -1.72% 3.59** ** Significant at the α=0.01 level. Variable Definitions: DACC bs equals discretionary accruals calculated using a cross-sectional adaptation of the modified Jones model, with total accruals measured as the change in non-cash current assets less the change in current liabilities excluding short term debt, less depreciation expense. 40
Table 6. Cross-sectional regressions of size-adjusted returns on accruals, discretionary accruals, non-discretionary accruals, and cash from operations. Discretionary accruals are computed as the residual from a modified Jones model. Panel A: Returns on Accruals and Cash Flows Model 1: SRET = α + β 1 TACC bs + β 1 CFO bs + ε Model 2: SRET = α + β 1 TACC cf + β 1 CFO cf + ε a b 1 b 2 Model 1 (bs) -0.02 1.06 1.06 (-3.21) Model 2 (cf) -0.03 (-4.42) (14.44) 0.96 (14.79) (12.50) 1.12 (11.63) Model 1 Test β 1 = β 2 : p-value = 0.987 Model 2 Test β 1 = β 2 : p-value = 0.022 Panel B: Returns on non-discretionary earnings and discretionary accruals Model 1: SRET = α + β 1 NDE bs + β 2 DACC bs + ε Model 2: SRET = α + β 1 NDE cf + β 2 DACC cf + ε a b 1 b 2 Model 1 (bs) -0.02 1.08 1.00 (-3.97) Model 2 (cf) -0.02 (-4.72) (14.94) 1.16 (15.71) (10.77) 0.81 (8.04) Model 1 Test β 1 = β 2 : p-value = 0.44; Model 2 Test β 1 = β 2 : p-value = 0.001 Panel C: Returns on non-discretionary earnings and discretionary accruals Model 1: SRET = α + β 1 NDACC bs + β 2 DACC bs + β 3 CFO bs + ε Model 2: SRET = α + β 1 NDACC cf + β 2 DACC cf + β 3 CFO cf + ε a b 1 b 2 b 3 Model 1 (bs) -0.01 1.33 0.98 1.05 (-1.15) Model 2 (cf) -0.01 (-0.95) (9.23) 1.46 (10.93) (10.47) 0.79 (8.73) (14.19) 1.12 (14.64) Model 1 Test β 1 = β 2 : p-value = 0.022; Model 2 Test β 1 = β 2 : p-value = 0.001 Model 1 Test β 2 = β 3 : p-value = 0.340; Model 2 Test β 2 = β 3 : p-value = 0.002 Model 1 Test β 1 = β 3 : p-value = 0.076; Model 2 Test β 1 = β 3 : p-value = 0.032 Note: The bs (cf) subscript refers to variables estimated using a balance sheet (cash flow) approach. Variable Definitions: TACC equals total accruals computed using a balance sheet or cash flow approach DACC equals discretionary accruals computed as the residual from a modified Jones model NDACCequals non-discretionary accruals computed as the predicted value from a modified Jones model NDE equals earnings before extraordinary items discretionary accruals CFO equals cash from operations computed using a balance sheet or cash flow approach SRET equals 12-month size-adjusted abnormal returns 41
Table 7. Statistics on the difference in rankings between balance sheet and cash flow operational definitions of accruals. Based on 14,558 firm year observations over the years 1988-1997 for NYSE/AMEX firms Panel A. Differences in decile membership for accruals computed using the balance sheet and accruals from the statement of cash flows Absolute value of the difference in decile rankings between ACCDCL BS and ACCDCL CF. 0 1 2 3 4 5 6 7 8 9 All Firm Years: Percent (%) 38.2% 33.9 12.9 6.1 3.1 2.0 1.4 1.0 0.7 0.4 Extreme Firm-years: Percent (%) 61.3% 15.9 5.2 3.6 2.9 2.0 2.0 2.0 1.8 3.2 Panel B. Misclassification in the extreme (i.e. first and tenth) deciles. All Firms Merger firms Disc-op firms Foreign Curr. firms ACCDCL bs = 1: Number of firm-years 1451 169 314 253 Number Misclassified 594 83 190 125 Percent Misclassified 40.9% 49.7% 60.5% 49.4% ACCDCL bs = 10: Number of firm-years 1451 513 124 210 Number Misclassified 529 284 61 82 Percent Misclassified 36.5% 55.4% 49.2% 39.0% Variable Definitions: ACCDCL BS equals the decile ranking of total accruals, where total accruals are measured the change in non-cash current assets less the change in current liabilities excluding short term debt, less depreciation expense. ACCDCL CF equals the decile ranking of total accruals, where total accruals are measured as operating accruals, taken from the statement of cash flows. 42
Figure 1 Frequency of Non-articulation Events magnitude of the difference between accruals measured using the balance sheet and cash flow approaches. This table depicts the proportion of firms with one of three non-articulation events over the years 1988-1997. Panel A examines the proportion of firms in the sample that experience each of the non-articulation events individually, as well as the proportion of firms that experience at least one non-articulation event. Panel B depicts the absolute value of the difference between total accruals measured using the balance sheet approach and total accruals measured using the cash flow statement approach, as a percent of total accruals under the balance sheet approach. 50 45 40 Panel A. Frequency of firms with one of three non-articulation events Firms with at least one nonarticulation event Percent (%) 35 30 25 20 15 10 5 Merger firms Forr. Curr. firms Disc-op firms 0 88 89 90 91 92 93 94 95 96 97 Year Difference as a % of accruals 50 45 40 35 30 25 20 15 10 Panel B. Absolute value of the difference between accruals calculated using the balance sheet and cash flow approaches. 5 0 88 89 90 91 92 93 94 95 96 97 Year 43
Figure 2 Magnitude of bias and imputed rejection frequencies for samples contaminated by mergers and acquisitions. This figure documents the bias in discretionary accruals in a sample of 250 firms as the percent of the sample contaminated by mergers and acquisitions varies. The sample is contaminated at the X% level by taking a random sample of 250*X% from the subsample of firm years with a merger or acquisition and a random sample of 250*(100%-X%) from the subsample of firm years without a non-articulation event. The bias in Panel A is calculated as the average of the difference between discretionary accruals measured using the balance sheet and cash flow approaches, deflated by total assets. The magnitude reported is based on 1000 trials of the aforementioned procedure. The rejection frequencies in Panel B represent the proportion of times the difference between discretionary accruals measured using the balance sheet and cash flow approaches is significantly different from zero at the 1% and 5% levels and, hence, represents the probability of committing a type I error when using the balance sheet approach if no earnings management is present. Panel A: Bias to discretionary accruals induced by mergers and acquistions 1.6 1.4 Percent of total Assets 1.2 1 0.8 0.6 0.4 0.2 0 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% Percent of Sample Contaminated by Mergers Panel B: Rejection frequencies as the percent of the sample contaminated by merger and acquisition firms varies. 100.0% 90.0% 80.0% Rejection Frequency (Type I Error) 70.0% 60.0% 50.0% 40.0% 30.0% 20.0% 10.0% 0.0% 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% Percent of sample contaminated by mergers 44
Figure 3 Magnitude of bias and imputed rejection frequencies for samples contaminated by discontinued operations. This figure documents the bias in discretionary accruals in a sample of 250 firms as the percent of the sample contaminated by discontinued operations varies. The sample is contaminated at the X% level by taking a random sample of 250*X% from the subsample of firm years with discontinued operations and a random sample of 250*(100%-X%) from the subsample of firm years without a non-articulation event. The bias in Panel A is calculated as the average of the difference between discretionary accruals measured using the balance sheet and cash flow approaches, deflated by total assets. The magnitude reported is based on 1000 trials of the aforementioned procedure. The rejection frequencies in Panel B represent the proportion of times the difference between discretionary accruals measured using the balance sheet and cash flow approaches is significantly different from zero at the 1% and 5% levels and, hence, represents the probability of committing a type I error when using the balance sheet approach if no earnings management is present. Panel A: Bias to discretionary accruals induced by discontinued operations Percent of total Assets 0-0.2-0.4-0.6-0.8-1 -1.2-1.4 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% Percent contaminated by discontined operations Panel B: Rejection frequencies as the percent of the sample contaminated with discontinued operations firms varies. 100.0% 90.0% 80.0% Rejection Frequency (Type I Error) 70.0% 60.0% 50.0% 40.0% 30.0% 20.0% 10.0% 0.0% 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% Percent contaminated by discontinued operations 45
Figure 4. Abnormal returns to a hedge portfolio formed on the basis of total accruals, where accruals are measured using the balance sheet and cash flow statement approaches. This figure depicts the abnormal returns accruing to a hedge portfolio strategy based on total accruals. The top half of the graph represents abnormal returns to firms in the lowest accrual decile under both the balance sheet (denoted by circles) and cash flow statement (denoted by squares) approaches. The lower half of the graph represent abnormal returns to firms in the highest accrual decile under the both the balance sheet and cash flow statement approaches. Returns are size-adjusted and compounded over the 12 months following the releases of the annual report. Based on 12,813 firm years from 1988-1997. 0.08 0.06 0.04 Size-adjusted return 0.02 0-0.02 BS CF 8.72% 11.88% -0.04-0.06-0.08 t t+1 t+2 t+3 t+4 t+5 t+6 t+7 t+8 t+9 t+10 t+11 t+12 Months past annual report 46
Figure 5 Abnormal returns to a quarterly accrual strategy under both the balance sheet and cash flow statement approaches This figure depicts the magnitude and regularity of hedge portfolio abnormal returns earned by taking a long position in the lowest total accrual decile and an offsetting short position in firms in the highest total accrual decile. Portfolios are formed every quarter from 1988 through 1997 and positions are held for 120 trading days. Panel A depicts returns that are earned when the statement of cash flows is used to rank total accruals, while panel B depicts returns that are earned when the balance sheet is used to estimate and rank total accruals. Panel A. Quarterly hedge portfolio returns when using the statement of cash flows to estimate accruals 0.3 0.2 Abnormal Return 0.1 0-0.1 88 89 89 90 90 91 91 92 92 93 93 94 94 95 95 96 96 97 Mean Return 5.56% Sum Pos 211% Sum Neg -10.7% -0.2-0.3 Year 0.3 Panel B. Quarterly hedge portfolio returns when using the balance sheet approach to estimate accruals 0.2 Abnormal Return 0.1 0-0.1 88 89 89 90 90 91 91 92 92 93 93 94 94 95 95 96 96 97 Mean Return 2.99% Sum Pos 162% Sum Neg -51.1% -0.2-0.3 Year 47