Accruals Management to Achieve Earnings Benchmarks: A Comparison of Pre-managed Profit and Loss Firms

Transcription

1 Journal of Business Finance & Accounting, 33(5) & (6), , June/July 2006, X doi: /j x Accruals Management to Achieve Earnings Benchmarks: A Comparison of Pre-managed Profit and Loss Firms Abhijit Barua, Joseph Legoria and Jacquelyn Sue Moffitt Abstract: This study examines whether firms with profits before accruals management are more likely than firms with losses before accruals management to meet or exceed earnings benchmarks when pre-managed earnings are below those benchmarks. We extend Brown (2001) by documenting that the differential propensity to achieve earnings benchmarks by profitable and nonprofitable firms results from differential accruals management behavior. We find that firms with profits before accruals management are more likely than firms with losses before accruals management to have pre-managed earnings below both analysts forecasts and prior period earnings and reported earnings above these benchmarks. Keywords: earnings benchmarks, abnormal accruals, accruals management, analysts forecasts, pre-managed earnings 1. INTRODUCTION We examine whether firm profitability is associated with earnings management to achieve two benchmarks: analysts forecasts and prior period earnings. The use of earnings benchmarks to investigate earnings management is well established in the literature. Burgstahler and Dichev (1997a) document the existence of two earnings management benchmarks (zero earnings and prior period earnings). Degeorge et al. (1999) identify one additional benchmark and show the emergence of an earnings management hierarchy: avoiding losses, avoiding earnings decreases and avoiding negative surprises relative to analysts forecasts. Brown and Caylor (2005) find that this hierarchy has reversed in recent years ( ); that is, avoiding negative surprises The first author is from the School of Accounting, Florida International University. The second and third authors are from the Department of Accounting, Louisiana State University. They would like to thank Bruce Billings, Jeff Boone, Andrew Christie, William Cready, Richard Morton, J.K. Reynolds, John Rigsby, the editor, an anonymous referee, and workshop participants at the 2004 AAA Southeast Regional Conference (Best Paper Award Recipient), Louisiana State University and Mississippi State University for their helpful comments. (Paper received June 2004, revised version accepted September 2005) Address for correspondence: Jacquelyn Sue Moffitt, Department of Accounting, CEBA 3101, E.J. Ourso College of Business, Louisiana State University, Baton Rouge, LA 70803, USA. jsmoff22@lsu.edu Journal compilation C 2006 Blackwell Publishing Ltd, 9600 Garsington Road, Oxford OX4 2DQ, UK and 350 Main Street, Malden, MA 02148, USA. 653

2 654 BARUA, LEGORIA AND MOFFITT has become predominant. Brown (2001) finds a nontrivial positive temporal trend in avoiding negative surprises and reports that the avoidance of negative surprises is higher for profit-reporting firms than for loss-reporting firms. Brown (2001) does not investigate the mechanisms firms apply to avoid negative earnings surprises. Research (e.g., Burgstahler and Eames, 2005; Bartov et al., 2002; and Matsumoto, 2002) suggests that firms both manipulate their earnings (earnings management) and analysts expectations (expectations management) to achieve earnings benchmarks. In this study, we investigate whether differential achievement of earnings benchmarks by firms with pre-managed profits versus firms with pre-managed losses results from differential accruals management. More specifically, we test the differences in accruals management behavior to achieve two benchmarks, analysts forecasts and prior period earnings, conditional on firms profitability. Since earnings management to achieve these two benchmarks can also affect the reported profit, we use pre-managed earnings, rather than reported earnings, as the measure of our conditional variable, profitability. 1 Prior studies (Hayn, 1995) suggest that the price-earnings relation differs for firms reporting profits and losses. Hayn (1995) reports that the earnings of profit-reporting firms are more strongly linked with current stock prices than the earnings of lossreporting firms and that the magnitude of current losses is not related to stock price. Burgstahler and Dichev (1997b) suggest that earnings response coefficients (ERCs) are positively associated with the level of earnings to book value. These findings imply that the earnings-return response differs for more profitable and less profitable firms and are consistent with more profitable firms having greater market-based incentives than less profitable firms for managing earnings to meet the earnings benchmarks of analysts forecasts and prior period earnings. To compare accruals management behavior between firms exhibiting profits and losses before accruals management activity, we conduct two tests based on analysts forecasts and prior period earnings benchmarks. We proxy abnormal accruals using a cross-sectional version of the Modified Jones Model (Dechow et al., 1995) after adjusting for performance (Kasznik, 1999). We estimate pre-managed earnings by subtracting this measure from reported earnings. Logistic regression is used to model the probability that pre-managed earnings are below a specified benchmark and reported earnings meet or exceed the benchmark, and test whether this probability varies with the sign of pre-managed earnings. Our results suggest that firms with profits before accruals management are more likely than firms with losses before accruals management to use accruals to meet or exceed analysts forecast and prior-period earnings benchmarks when pre-managed earnings are below the benchmarks. These findings imply that relatively more profitable firms having greater incentives than less profitable firms to manage earnings to meet these earnings benchmarks. Our findings are consistent with prior studies (e.g., Hayn, 1995; Matsumoto, 2002; and Abarbanell and Lehavy, 2003) that suggest the returns of profit-reporting firms are more sensitive to reported earnings. While our findings do not rule out the possibility of expectations management to avoid negative earnings surprises, the results suggest that managers use accruals management to achieve earnings benchmarks. 1 We define pre-managed earnings as reported earnings less abnormal accruals. See Section 4 (iii) for additional detail.

3 ACCRUALS MANAGEMENT TO ACHIEVE EARNINGS BENCHMARKS 655 We contribute to the literature by showing that firms exhibiting pre-managed profits are more likely than firms exhibiting pre-managed losses to use accruals management to achieve the benchmarks of analysts forecasts and prior period earnings. This study extends Brown (2001) by providing evidence that the differential propensity to achieve earnings benchmarks by relatively more profitable and less profitable firms results in differential accruals management behavior. The remainder of the paper is organized as follows. Section 2 discusses the hypotheses; Section 3 describes the data and sample selection process; Section 4 describes the research design; Section 5 presents the empirical results; and Section 6 concludes the paper. 2. HYPOTHESIS DEVELOPMENT Accounting research documents that firms manage earnings to meet or exceed earnings benchmarks (Burgstahler and Dichev, 1997a; and Degeorge, et al., 1999). A number of recent studies (Bartov et al., 2002; Kasznik and McNichols, 2002; and Lopez and Rees, 2002) report that the market rewards (penalizes) firms for meeting (missing) the earnings forecast benchmark. Studies also show that the relative frequency of meeting or exceeding analysts forecasts is increasing over time (Bartov et al., 2002; and Matsumoto, 2002). Another important benchmark is prior period earnings (Burgstahler and Dichev, 1997a; and Degeorge et al., 1999). Barth et al. (1999) document that firms with a consistent pattern of meeting prior period earnings have higher price-earnings multiples and that the earnings multiples decrease when this pattern breaks. These studies suggest that firms have incentives to achieve analysts forecasts and prior period earnings benchmarks. We argue that the incentives to achieve these earnings benchmarks are greater for firms experiencing profits or losses on a premanaged basis. Brown (2001) observes that the tendency to meet or beat analysts forecasts is consistently higher for profit-reporting firms than for loss-reporting firms. He also reports that the frequency of profit-reporting firms meeting or exceeding forecasts is more than double that for loss-reporting firms; however, his study does not examine how firms meet or beat analysts forecasts. When pre-managed earnings fall short of a benchmark, managers can exercise discretion in accounting accruals to achieve the benchmark. For example, Payne and Robb (2000) show that when pre-managed earnings are below analysts forecasts, firms use income-increasing abnormal accruals to meet or beat the forecasts. If the incentives differ between firms with profits before accruals management and firms with losses before accruals management, we expect differential accruals management behavior between these two sub-samples. Greater incentives for more profitable firms can be explained by the differential returns-earnings association documented by Hayn (1995), who reports earnings are significantly less strongly associated with returns for firms reporting losses than for firms reporting profits. Thus, profit-reporting firms have greater incentives to manage earnings to achieve benchmarks than loss-reporting firms since profit-reporting firms have higher perceived growth opportunities imbedded in stock prices (Hayn 1995). Matsumoto (2002) provides evidence suggesting that loss firms (i.e., firms with low value-relevance of earnings) have less incentive to meet or beat analysts forecasts. Burgstahler and Dichev (1997b) theorize that equity

4 656 BARUA, LEGORIA AND MOFFITT value is a convex function of both expected earnings and book value. They find that the coefficient on earnings increases with the ratio of earnings to book value while the coefficient on book value decreases with the ratio of earnings to book value. These results imply that the coefficient on earnings is positively associated with market value and that the earnings-return response differs for more profitable and less profitable firms. Additionally, these results are consistent with more profitable firms having greater market-based incentives than less profitable firms for managing earnings to meet earnings benchmarks. Abarbanell and Lehavy (2003) report that occurrences of meeting or beating forecasts increase with the sensitivity of stock price to earnings news. Their results provide evidence of equity market incentives for firms to manage earnings. These studies suggest that profitable firms will have more incentive than loss firms to manage earnings to achieve earnings targets. Differential incentives between profit- and loss-reporting firms are also evident in Lopez and Rees (2002), who find differential ERCs for firms meeting versus missing forecasts in the profit sample but not in the loss sample. Their findings suggest that the market is indifferent to whether a firm meets or misses an earnings forecast when it reports a loss. Differing incentives for profitable and unprofitable firms suggest profitable firms are more likely to manipulate earnings to achieve earnings benchmarks. We formulate our hypotheses as follows: H 1A : When pre-managed earnings are below analysts forecasts, profitable firms are more likely than loss firms to have reported earnings that meet or beat analysts forecasts. H 1B : When pre-managed earnings are below prior period earnings, profitable firms are more likely than loss firms to have reported earnings that meet or beat prior period earnings. 3. DATA AND SAMPLE SELECTION We use quarterly data to test our predictions and perform a sensitivity analysis using annual data. Investors view each interim period as a separate, self-contained reporting period as well as an integral component of a year, and from both points of view, managers have incentives to manage quarterly earnings to achieve earnings benchmarks (Burgstahler, 1997). Firms that report profits on an annual basis may not necessarily report profits for every quarter throughout a given fiscal year. To the extent that a firm reports both quarterly profits and losses within a given fiscal year, a manager will have differing incentives to manage earnings on a quarter-to-quarter basis. Degeorge et al. (1999) provide evidence of the existence of three earnings management benchmarks using quarterly data. A number of prior studies on accruals (e.g., Collins and Hribar, 2000; Balsam et al., 2002; and DeFond and Park, 2001) also use quarterly data. Furthermore, the financial press (e.g., Wall Street Journal, Barron s) extensively covers quarterly earnings announcements. We obtain financial data from the Compustat Full Coverage and Research database (COMPUSTAT) and actual and forecast earnings data from the Institutional Brokers Estimate System (I/B/E/S) detailed database from 1992 through We choose 1992 as a starting point for our sample period because Brown (2001) provides evidence of a temporal shift in analysts forecast bias starting in 1992

5 ACCRUALS MANAGEMENT TO ACHIEVE EARNINGS BENCHMARKS 657 Table 1 Sample Selection Firm Quarter Observations Total observations from COMPUSTAT from 1992 through 2002, inclusive 432,502 Less: Utilities Financial, and Service observations, (171,828) Observations with no I/B/E/S detailed forecasts (203,347) in last 30 days of fiscal quarter and observations not in I/B/E/S database, Observations with missing information, and (32,256) Extreme observations 1 (1,723) Total observations retained 23,348 Note: 1 To eliminate potential errors in the data, we eliminated observations in both the top and bottom 0.5% of observations for the following variables: AA (abnormal accruals), FERR (forecast errors), MBV (ratio of the market value of equity to the book value of equity), and OPP (opportunity to use abnormal accruals, measured as the sum of current assets (less cash) and current liabilities at the end of quarter t-1 over total assets at the end of quarter t-1). and Levitt (1998) argues that the pressure to meet consensus forecasts became more pronounced in the 1990s. Table 1 presents the sample selection process. 2 The consensus forecasts provided by the I/B/E/S summary database are splitadjusted and rounded. Payne and Thomas (2003) find that these adjusted and rounded forecasts may cause misleading results. As an alternative, they suggest researchers either (1) obtain unrounded and unadjusted data from I/B/E/S and create their own splitadjusted forecasts without rounding to the nearest penny or (2) use the I/B/E/S adjusted detailed database and compute their own consensus forecasts. We reviewed the detailed I/B/E/S database and found that the individual forecast numbers are rounded to four decimal places. We also reviewed the summary database and found that the consensus forecasts (mean and median) are rounded to two decimal places. Thus, using the detailed I/B/E/S database provides a more accurate estimate of the consensus forecast used in this study. Consistent with Payne and Thomas (2003) and Skinner and Sloan (2002) we compute consensus forecasts from the detailed I/B/E/S database using median earnings forecasts based on the latest forecast of each analyst in the final 30 days of the fiscal quarter. We exclude utilities, financial institutions, and service organizations (SIC codes and ) from the sample. We further narrow our sample size by eliminating observations with insufficient data to estimate abnormal accruals. To avoid the influence of extreme values, we truncate the sample by eliminating observations in both the top and bottom 1/2 percentile for abnormal accruals as well as for forecast errors, the market-to-book ratio, and the proxy for a firm s opportunity to use abnormal 2 Prior studies (e.g. Degeorge et al., 1999; and Kasznik and McNichols, 2002) use the analysts consensus earnings forecast (F cons ) from I/B/E/S as a proxy for ex ante market expectations of earnings.

6 658 BARUA, LEGORIA AND MOFFITT accruals. 3 Our final sample contains 23,348 firm-quarter observations, of which 16,368 are firms with profits before accruals management (pre-managed profits) and 6,980 are firms with losses before accruals management (pre-managed losses). We define firm-quarter observations with zero or positive earnings prior to accruals management as firms with pre-managed profits. (i) Forecast Errors 4. RESEARCH DESIGN Forecast errors (FERR) are calculated by subtracting the median forecast (F cons ) for a quarter from the corresponding earnings (FERR = EPS F cons ), where EPS is the actual earnings per share as reported by I/B/E/S. (ii) Abnormal Accrual Model We estimate abnormal accruals, a proxy for accruals management, by using the following variation of the Modified-Jones Model developed by Dechow et al. (1995). 4 where: TA it TA it = β 1 (1/A it 1 ) + β 2 ( REV it REC it ) + β 3 PPE it β 4 OCF it + ε it (1) = Total accruals for firm i for quarter t scaled by total assets (COMPUSTAT quarterly data item 44) for quarter t-1; total accruals are calculated by subtracting quarterly cash flow from operations (COMPUSTAT quarterly data item 108) from net income per I/B/E/S, 5 A it 1 = Total assets (COMPUSTAT quarterly data item 44) for quarter t-1, REV it = Revenues (COMPUSTAT quarterly data item 2) for firm i for quarter t less revenues for firm i for quarter t-1 scaled by total assets for quarter t-1, REC it = Receivables (COMPUSTAT quarterly data item 37) for firm i for quarter t less receivables for firm i for quarter t-1 scaled by total assets for quarter t-1. PPE it = Gross property plant and equipment (COMPUSTAT quarterly data item 118) for firm i for quarter t scaled by total assets for quarter t-1, and OCF it = Operating cash flow (COMPUSTAT quarterly data item 108) for firm i for quarter t less operating cash flow for firm i for quarter t-1 scaled by total assets for quarter t-1. Prior studies (Subramanyam, 1996; and Bartov et al., 2000) suggest that crosssectional versions of the Jones (1991) model and the Modified-Jones Model are better 3 To test the robustness of our method of eliminating extreme observations, we also used two alternative methods to eliminate extreme observations: (1) eliminate observations for which abnormal accruals are greater than 10% of firm market value, and (2) eliminate observations for which forecast errors are greater than 10% of firm market value. In both cases, the results (not tabulated) are consistent with reported results. 4 Existing literature suggests that the Modified-Jones Model is the most robust among all models (Dechow et al., 1995; and Bartov et al., 2001). 5 Our measure of total accruals follows the suggestion of Hribar and Collins (2002). We computed net income as earnings per share per I/B/E/S multiplied by common shares used to compute EPS (COMPUSTAT quarterly data item 15).

7 ACCRUALS MANAGEMENT TO ACHIEVE EARNINGS BENCHMARKS 659 specified than their time-series counterparts. We use the cross-sectional version of the Modified-Jones Model, in which total accruals are regressed on the change in cash basis sales and the level of property, plant and equipment. Prior studies (Bartov et al., 2000; Dechow et al., 1995; and Dechow et al., 2003) note that Jones (1991) type abnormal accrual estimation models suffer from correlated omitted variable issues and therefore are potentially misspecified. Two sources of this misspecification identified by prior research include operating cash flows (Dechow et al., 1995; and Subramanyam, 1996) and earnings performance (Bartov et al., 2000; Dechow et al., 1995; and Dechow et al., 2003). To control for the effect of operating cash flows, we include an additional variable in the abnormal accrual model computed as the change in operating cash flows from quarter t-1 to quarter t, scaled by assets at the end of quarter t, consistent with Kasznik (1999). To control for potential measurement error that is correlated with earnings performance, we follow the procedure described in Kasznik (1999). 6 We estimate equation (1) by year and quarter and by 2-digit SIC code using ordinary least squares regression and obtain the residual for each firm-quarter observation. 7 We rank the observations by return on assets (ROA) and assign them to percentiles. We measure ROA as earnings from continuing operations scaled by lagged assets, consistent with Kasznik (1999). Next we compute the median unadjusted abnormal accruals (residual from equation (1)) for each percentile and subtract the percentile median from each observation s unadjusted abnormal accruals in that percentile. We use this median adjusted abnormal accruals as our proxy for abnormal accruals, thus removing the effect of earnings performance on accruals. (iii) Pre-managed Earnings We test H1 A and H1 B using logistic regression analysis. Abnormal accruals estimated in Section 4 (ii) are scaled by lagged assets. These estimates are multiplied by lagged assets and divided by the common shares used to compute basic EPS (COMPUSTAT quarterly data item 15) to obtain a proxy for abnormal accruals per share. We estimate pre-managed earnings by subtracting abnormal accruals per share from actual earnings per share. (iv) Logistic Regression Model Our dependent variable, MOVE, is an indicator variable set equal to one if the firm has pre-managed earnings below a specified earnings benchmark and actual earnings that meet or beat the benchmark up to one percent of firm market value. MOVE is set equal to zero only when pre-managed earnings and reported earnings are both below the specified earnings benchmark. When a firm moves from a 6 We also controlled for earnings performance by including ROA directly in equation (1). The test results are qualitatively similar to the main results reported in Tables 3 and 4. 7 Subramanyam (1996) argues that the cross-sectional model has advantages over the time-series model ( Jones, 1991) and that parameter estimates are better specified. Fewer observations are likely to be lost using the cross-sectional model. At least 10 years of data are needed for each firm to estimate the time-series model; many firms would have to be dropped from the sample due to insufficient data, making estimates less precise. Nonstationarity may also be an issue since the time-series is estimated over a minimum of 10 years.

8 660 BARUA, LEGORIA AND MOFFITT situation where it has pre-managed earnings below a benchmark to reported earnings at or slightly above the benchmark, the movement is consistent with abnormal accruals management. The logistic regression model to test this movement is: PROB(MOVE) it = β 0 + β 1 PROFIT it + β 2 MBV it + β 3 (PROFIT it MBV it) + β 4 SIZE it + β 5 (PROFIT it SIZE it) + β 6 OPP it + β 7 (PROFIT it OPP it) (2) where: MOVE it = 1 if the firm-quarter observation has pre-managed earnings below the specified benchmark and reported earnings equal to or greater than the specified benchmark (up to 1% of firm market value) and 0 if the firm-quarter observation has both pre-managed earnings and reported earnings below the specified earnings benchmark, PROFIT it = 1 if firm i has profits before accruals management for quarter t and 0 otherwise, MBV it = Market value of equity for firm i for quarter t divided by book-value of equity, SIZE it = Log of total sales for firm i for quarter t, and OPP it = Current assets plus current liabilities less cash at the end of quarter t-1, scaled by total assets at the end of quarter t-1. PROFIT measures whether a firm has a quarterly profit prior to accruals management (pre-managed earnings). 8 To compute pre-managed earnings, we subtract the abnormal accruals per share estimate obtained in Section 4 (ii) from reported EPS. Based on H 1A and H 1B, we predict that the coefficient for PROFIT will be significantly greater than zero. Prior research (e.g., Watts and Zimmerman, 1986) suggests that, in addition to meeting earnings benchmarks such as analysts forecasts, additional variables can influence management s decision to use abnormal accruals. To allow for the possibility that our control variables behave differently for firms with pre-managed profits and firms with pre-managed losses, we include interaction terms between the control variables and PROFIT. Skinner and Sloan (2002) show that growth stocks suffer significantly greater negative market responses to earnings disappointments than value stocks, which suggests that growth firms have greater incentives to avoid negative surprises. Brown (2001) documents a greater frequency of positive surprises for growth firms than for value firms. Both studies use the market-to-book ratio (MBV) as a proxy for firm growth. Consistent with these studies, we predict MBV will be positively associated with MOVE. Brown (2001) also provides evidence that, conditional on reporting a profit, growth firms are more likely than value firms to report a small positive surprise ( a little bit good news ); therefore, we also predict a positive coefficient for the interaction term of PROFIT and MBV. Prior studies indicate that larger firms are followed by more analysts (Bhushan, 1989), which suggests that these firms face greater pressure to achieve earnings benchmarks (Das et al., 1998). On the other hand, Watts and Zimmerman (1986) 8 While profitability is a continuous variable, the use of an indicator variable avoids the need to scale profitability by an arbitrary number such as the number of shares outstanding.

9 ACCRUALS MANAGEMENT TO ACHIEVE EARNINGS BENCHMARKS 661 suggest that larger and more profitable firms have greater political costs due to external monitoring. Cahan (1992) finds that large firms facing antitrust investigations use abnormal accruals to reduce earnings, and Manzon (1992) finds that large firms subject to the alternative minimum tax also use abnormal accruals to reduce earnings. We measure firm size (SIZE) as the log of total sales. Other common proxies for size are market capitalization and total assets. We choose log of total sales as our measure of size as we use MBV to proxy for growth. We also use an alternative measure of SIZE (log of assets) and report the results in the Sensitivity Analysis in Section 5 (iv). To separate the differing incentives of larger firms to use abnormal accruals, we include two control variables. The first is SIZE, which controls for the pressure on larger firms to achieve earnings benchmarks. Consistent with Matsumoto (2002), we predict a positive coefficient for SIZE. The second is the interaction term PROFIT SIZE and is included to control for political costs. Managers of profitable large firms are more likely to avoid attracting the attention of regulators (e.g., SEC) by not engaging in earnings management through consistently meeting benchmarks. 9 We predict the sign of the coefficient on the interaction term will be negative. Burgstahler and Dichev (1997a) argue that firms with high levels of current assets and current liabilities will likely find it less costly to manage earnings through changes in working capital than firms with low levels of current assets and current liabilities. We call this variable opportunity (OPP) and define it as current assets plus current liabilities less cash at the end of quarter t-1, scaled by total assets at the end of quarter t-1. Firms with greater current assets and liabilities have a higher opportunity and are therefore more likely to manage earnings to move from below a benchmark to above the benchmark. We predict the sign of the coefficient of OPP will be positive and make no prediction for the sign of the coefficient on the interaction term, PROFIT OPP. (i) Summary Statistics 5. EMPIRICAL FINDINGS Table 2 presents summary statistics. Panel A includes descriptive statistics for the entire sample. Mean net income is approximately $48.85 million, while median net income is $7.0 million. Mean (median) quarterly abnormal accruals (scaled by lagged assets) (AA) is (0). The mean (median) forecast error per share (FERR) is (0.003). Table 2 includes descriptive statistics for control variables used in the multivariate analyses. 9 Levitt expressed concerns in a 1998 speech about earnings management, and especially about the tendency of firms to meet or beat earnings benchmarks. Anecdotal evidence since his speech suggests that the SEC began looking at firms earnings more closely and in particular whether a firm consistently met its consensus earnings estimates. For example, consider the SEC s Accounting and Auditing Enforcement release program where many large firms have been cited since 2001 (e.g., Enron, WorldCom, Xerox, TimeWarner). As further evidence that the SEC has begun looking at consistently meeting earnings estimates, Accounting and Auditing Enforcement Release No issued on June 2, 2005 (see indicates that Huntington Bancshares was sued by the SEC. The complaint cited that in 2001 and 2002 Huntington reported inflated earnings in its financial statements, enabling Huntington to meet or exceed Wall Street analysts earnings per share (EPS) expectations....

10 662 BARUA, LEGORIA AND MOFFITT Table 2 Descriptive Statistics Variables 1 Mean Median Standard Deviation Minimum 25% 75% Maximum Panel A: Entire Sample (23,348 observations) NI , AA FERR MBV SIZE OPP ROA Panel B: Pre-managed Profit Sub-sample 2 (16,368 observations) NI , AA FERR MBV SIZE OPP ROA Panel C: Pre-managed Loss Sub-sample 2 (6,980 observations) NI , AA FERR MBV SIZE OPP ROA Notes: 1 Variable definitions: NI = Net income (in millions), computed as actual EPS per I/B/E/S common shares used to calculate EPS (Compustat quarterly item No. 15), AA = Estimated abnormal accruals scaled by lagged assets, FERR = Forecast error (Actual EPS per I/B/E/S median forecast EPS, where the median forecast is computed using the latest forecast provided by each analyst in the last 30 days of each quarter), MBV = Market value of equity divided by book value of equity, SIZE = Log of total sales (in millions), OPP = Current assets plus current liabilities less cash at the end of quarter t-1 scaled by lagged assets, and ROA = Net income from continuing operations scaled by lagged total assets. 2 The pre-managed profit sub-sample contains those observations for which earnings prior to accruals management (i.e., EPS per I/B/E/S AA) for quarter t is zero or positive. The pre-managed loss sub-sample contains those observations for which earnings prior to accruals management for quarter t is negative. Panels B and C present descriptive statistics for the profit and loss sub-samples, respectively. Mean quarterly abnormal accruals for the profit sub-sample is (median of 0.004) and for the loss sub-sample is (median of 0.015). The means are significantly different from zero (p < ), suggesting that firms exhibiting profits before accruals management generally have income-decreasing abnormal accruals and firms exhibiting losses before accruals management generally have

11 ACCRUALS MANAGEMENT TO ACHIEVE EARNINGS BENCHMARKS 663 income-increasing abnormal accruals. The mean (median) forecast error (FERR) of the profit sub-sample is (0.098) and of the loss sub-sample is ( 0.322). The findings with respect to the mean and median FERR are consistent with Eames and Glover (2003). (ii) Tests of Accruals Management Around Analysts Forecasts We conduct two tests of accruals management around analysts forecasts. The first test (H 1A1 ) considers management to meet or exceed analysts forecasts. This test can be confounded by the management to meet or exceed prior period earnings. To be certain that earnings management to meet or beat last period s earnings does not confound our results, in a second test (H 1A2 ) we exclude observations where: (1) pre-managed earnings is below prior period earnings and (2) reported earnings meets or beats prior period earnings. Table 3 reports the parameter estimates and corresponding p-values from the logistic regression estimations of Model 2 to test H 1A. (a) Observations with Pre-managed Earnings Below Forecasts In the first test, the indicator variable MOVE equals one if the firm has pre-managed earnings below analysts expectations and reported earnings equal to or greater than expectations (up to 1% of firm market value). MOVE equals zero if both pre-managed earnings and reported earnings are below expectations. The model Chi-square statistic is and is highly significant (p < ) while the pseudo R 2 is 4.50%. The coefficient on PROFIT is positive and significant (p = ). This finding is consistent with H 1A, suggesting that firms having pre-managed profits are more likely than firms with pre-managed losses to use abnormal accruals to meet or beat analysts forecasts. The coefficients for MBV and PROFIT MBV are significantly positive (p < and p = 0.01, respectively), consistent with both our prediction and Brown (2001) that growth firms and profitable growth firms are more frequently associated with premanaged earnings below analysts forecasts and reported earnings at or above analysts forecasts than non-growth firms. The coefficient on SIZE is significantly positive ( p < ), consistent with our prediction that larger firms are relatively more likely to use income-increasing abnormal accruals to move from below to at or above earnings benchmarks. The coefficient on PROFIT SIZE is negative and significant (p < 0.04) consistent with those firms facing higher political costs than other firms. The coefficient on OPP is significantly positive (p = 0.004), suggesting the firms that have more opportunity to manage earnings move more frequently from below analysts forecasts to at or above analysts forecasts. The coefficient on PROFIT OPP is negative and significant (p = 0.004), which suggests that profitable firms with more opportunity have enough accruals from their normal operations to help them achieve earnings benchmarks. (b) Observations That Exclude Meeting or Exceeding the Prior Period Earnings Benchmark In the second test, MOVE is measured in the same manner as above; however, firmquarter observations with pre-managed earnings that are below prior period earnings

12 664 BARUA, LEGORIA AND MOFFITT Table 3 Results of Regression Testing Benchmark Analysts Forecasts PROB(MOVE) it = β 0 + β 1 PROFIT it + β 2 MBV it + β 3 (PROFIT it MBV it) + β 4 SIZE it + β 5 (PROFIT it SIZE it) + β 6 OPP it + β 7 (PROFIT it OPP it) (2) Observations with Observations Used to Test H 1A1 Pred Pre-managed Earnings Less Those That Met Prior Variables 1 Sign Below Analysts Forecasts 2 Period Earnings Benchmark 2 H 1A1 H 1A2 Intercept it? (<0.0001) (<0.0001) PROFIT it (0.0002) (0.005) MBV it (<0.0001) (<0.0001) PROFIT it MBV it (0.01) (0.002) SIZE it (<0.0001) (<0.0001) PROFIT it SIZE it (<0.04) (0.15) OPP it (0.004) (0.46) PROFIT it OPP it 1? (<0.004) (0.14) Model Chi-Square P-value (<0.0001) (<0.0001) Pseudo-R % 4.16% Number of Observations 11,278 7,017 Obs. where MOVE = 1 6,736 3,537 Obs. where MOVE = 0 4,542 3,480 Notes: 1 MOVE is an indicator variable set equal to one if the firm-quarter observation has pre-managed earnings that are below analysts forecasts and actual earnings that either meet or beat analysts forecasts (up to 1% of firm market value) and is set equal to zero when both pre-managed earnings and reported earnings are below analysts forecasts. PROFIT is an indicator variable set equal to one if the firm-quarter observation reports profits on a pre-managed basis and is zero otherwise. All other variables are described in Table 2. 2 The top row presents the estimated coefficient for each variable. P-values are given in parentheses below the coefficient. and actual earnings that meet or beat prior period earnings are eliminated from the sample. We define the prior period as quarter t-4 to control for seasonality in reported quarterly earnings. This test allows for a more direct test of whether firms exhibiting profits before accruals management activities use abnormal accruals to meet or beat analysts forecasts more frequently than firms exhibiting losses before accruals management activities since there are no firms in this sub-sample that managed earnings to meet the prior period earnings benchmark. The chi-square test for fit is (p < ) and the pseudo R 2 is 4.16%. Consistent with the first estimation,

13 ACCRUALS MANAGEMENT TO ACHIEVE EARNINGS BENCHMARKS 665 the coefficient on PROFIT is positive and significant (p = 0.005). The coefficients for the control variables PROFIT SIZE, OPP, and PROFIT OPP are no longer significant. The results for the remaining control variables are consistent with the results of H 1A1. (iii) Tests of Accruals Management Around Prior Period Earnings In this section, we conduct three tests of accruals management in relation to prior period earnings. The first test (H 1B1 ) includes all firm-quarter observations for which pre-managed earnings is below prior period earnings. The second and third tests (H 1B2 and H 1B3, respectively) are on negative and positive forecast error sub-samples, respectively. Table 4 reports our results. (a) All Observations In the first test, MOVE is coded one if the firm has pre-managed earnings below quarter t-4 (the prior period) earnings and reported earnings that exceed prior period earnings (up to 1% of firm market value). MOVE is coded zero if both pre-managed earnings and reported earnings are below prior period earnings. The model Chi-square is (p < ) and the pseudo R 2 is 6.89%. The coefficient for PROFIT is positive and significantly different from zero (p = 0.002). This is consistent with H 1B and suggests that firms with profits before accruals management are more likely to use income-increasing abnormal accruals to move from below to at or above prior period earnings than firms that have losses before accruals management. The results for the control variables MBV, PROFIT MBV, SIZE, OPP, and PROFIT OPP are consistent with those reported in Table 3, test (1). The coefficient for PROFIT SIZE is not significant, suggesting that there is no incremental size effect among profitable firms. (b) Earnings Observations That Do Not Meet or Exceed Analysts Forecasts We test, for firms that do not meet analysts forecasts, whether firms with profits before accruals management are more likely than firms with losses before accruals management to use accruals management to meet prior period earnings. The model chi-square and pseudo R 2 are (p < ) and 8.64%, respectively. The coefficient for PROFIT is positive and significantly different from zero ( p < ) consistent with H 1B. These results suggest that firms with profits before accruals management that are unable to achieve analysts forecasts still move more frequently from below prior period earnings to at or above prior period earnings compared to firms with losses before accruals management. The results for the control variables are consistent with the first test for H 1B, with two exceptions. PROFIT MBV is now insignificant, and PROFIT SIZE is now significantly negative (p < 0.06). (c) Earnings Observations That Meet or Exceed Analysts Forecasts We next limit our test to firms that meet or exceed analysts forecasts. This estimation allows us to test whether firms with profits before accruals management that have also

14 666 BARUA, LEGORIA AND MOFFITT Table 4 Results of Testing Benchmark Prior Period Earnings PROB(MOVE) it = β 0 + β 1 PROFIT it + β 2 MBV it + β 3 (PROFIT it MBV it) + β 4 SIZE it + β 5 (PROFIT it SIZE it) + β 6 OPP it + β 7 (PROFIT it OPP it) (2) H 1B1 H 1B2 H 1B3 Observations With Observations With Observations With Pre-managed Earnings Negative Forecast Positive Forecast Pred. Below Prior Period Negative Forecast Positive Forecast Variables Sign Earnings 2 Errors 2 Errors 2 Intercept it? (<0.0001) (<0.0001) (<0.0001) PROFIT it (0.002) (<0.0001) (0.40) MBV it (<0.0001) (<0.0001) (0.97) PROFIT it MBV it (<0.07) (0.31) (0.45) SIZE it (<0.0001) (<0.0001) (0.25) PROFIT it SIZE it (0.45) (<0.06) (0.52) OPP it (<0.0001) (<0.0001) (0.70) PROFIT it OPP it 1? (<0.0001) (<0.0001) (0.92) Model Chi-Square P-value (<0.0001) (<0.0001) (0.74) Pseudo-R % 8.64% 1.20% Number of Observations 9,363 7,811 1,548 Obs. where MOVE = 1 3,808 3, Obs. where MOVE = 0 5,555 4,049 1,502 Notes: 1 MOVE is an indicator variable set equal to one if the firm-quarter observation has pre-managed earnings that are below prior period earnings and actual earnings that exceed prior period earnings (up to 1% of firm market value) and is zero if both pre-managed earnings and reported earnings are below prior period earnings. PROFIT is an indicator variable set equal to one if the firm-quarter observation reports profits on a pre-managed basis and is zero otherwise. All other variables are described in Table 2. 2 The top row presents the estimated coefficient for each variable. P-values are given in parentheses below the coefficient. met analysts forecasts on a pre-managed basis are more likely to move from having premanaged earnings below prior period earnings to having reported earnings at or above prior period earnings than firms with losses before accruals management that have also met analysts forecasts. The model chi-square is 4.37 (p = 0.74), and all variables are insignificant. We offer two possible explanations for not finding significant results in this test. First, firm-quarter observations in this sub-sample have already met two earnings management thresholds: avoiding negative earnings and avoiding negative surprises. Through an analysis of incremental valuation consequences, Brown and Caylor (2005)

15 ACCRUALS MANAGEMENT TO ACHIEVE EARNINGS BENCHMARKS 667 show that avoiding negative surprises is the most attractive of the three earnings management benchmarks and achieving prior period earnings is the least attractive of the benchmarks. Having met the two most important thresholds, management of these firms therefore chooses not to manage accruals to meet or beat prior period earnings. Second, in the test of H 1B3, the dichotomous dependent variable (MOVE) has a value of one for only 3% of the observations in this sub-sample. 10 A logistic regression model with an unbalanced sample (many more ones than zeroes, or vice versa) has estimation problems and thus may not predict a 1 or 0 accurately (Greene 1993, p.652). We performed a sensitivity analysis using OLS regression; the results are consistent with the reported results. (iv) Sensitivity Analyses In this section we address the robustness of our primary findings by examining five potential validity threats: (1) use of median versus mean analysts forecasts to compute FERR, (2) fourth quarter effect, (3) measurement of size, (4) use of I/B/E/S earnings versus COMPUSTAT earnings, and (5) use of quarterly data versus annual data. When we use mean analysts forecast to compute FERR our results are qualitatively similar to the previously reported results. The coefficients for PROFIT using mean analysts forecasts are significantly positive in all tests. 11 Fourth quarter results could be driving the overall results if firms engage in more accruals management in the fourth quarter to impact annual reports. To test for this we estimate model (2) using all observations for the first three quarters only (results not tabulated). Results for PROFIT are qualitatively similar to the results reported in Tables 3 and 4. Overall these findings suggest that the fourth quarter is not driving our results. Our results may be sensitive to our measure of size. We substitute the log of total assets at the end of quarter t for the log of total sales for quarter t of the firm in the logistic regressions. The findings (not tabulated) are consistent with the reported results in Tables 3 and 4. We used EPS reported by I/B/E/S multiplied by common shares used to calculate EPS to calculate total accruals. When we use net income from COMPUSTAT to compute total accruals, our results are qualitatively similar to the reported results in Tables 3 and 4. We performed our main tests using annual data. The results are qualitatively similar to the reported results in Tables 3 and The lack of observations with a value of one for the dependent variable (MOVE) reinforces our first explanation. Management, having achieved the benchmarks of avoiding negative earnings and avoiding negative surprises, are less interested in achieving the benchmark of prior period earnings and therefore do not attempt to do so. Thus, there are few observations in this sub-sample where MOVE = We also estimated equation (2) using a two-way fixed effects model to resolve the dependence problem of panel data. The results from this estimation were consistent with the reported results in Tables 3 and 4.

16 668 BARUA, LEGORIA AND MOFFITT 6. SUMMARY AND CONCLUSIONS This study investigates whether firms with profits before accruals management activities are more likely than firms with losses before accruals management activities to meet or exceed earnings benchmarks when pre-managed earnings are below those benchmarks. Prior studies (e.g., Degeorge et al., 1999; and Brown and Caylor, 2005) document the existence of an earnings management hierarchy that includes three benchmarks: avoiding losses, meeting analysts forecasts, and meeting prior period earnings. Brown (2001) shows that the propensity of achieving benchmarks is systematically higher in profit-reporting firms compared to loss-reporting firms. We extend Brown (2001) by examining whether this differential propensity leads to differential accruals management behavior. The results suggest that firms with profits before accruals management are more likely than firms with losses before accruals management to have premanaged earnings below a specific earnings benchmark and actual earnings above that benchmark, regardless of whether the benchmark is prior period earnings or analysts forecasts. Our findings are consistent with firms with profits before accruals management having more incentives to achieve earnings benchmarks, since their earnings are more responsive to market valuation than the earnings of loss-reporting firms (e.g., Hayn, 1995; and Abarbanell and Lehavy, 2003). Brown (2001) finds that profit-reporting firms are more likely to meet or just beat analysts forecasts than loss-reporting firms. Our results suggest that Brown s (2001) findings are consistent with firms with profits before accruals management being more likely than firms with losses before accruals management to use income-increasing abnormal accruals to achieve the benchmark analysts forecasts. We extend Payne and Robb (2000), who find that when premanaged earnings are below analysts expectations, firms use income-increasing abnormal accruals. Our findings have implications for future research in the area of accruals management, suggesting that the differential incentives as well as the propensity to manage earnings by profitable and non-profitable firms should be considered. The findings in this study do not exclude expectations management as a mechanism used by managers to achieve analysts forecasts. The results overall suggest that managers use accrual management as one mechanism to achieve earnings benchmarks. Our findings also do not exclude the possibility that firms achieve earnings benchmarks by taking real action (e.g., working hard or reducing expenditures such as advertising and research and development). Further research could examine the trade-offs managers make among these choices to manage earnings or expectations. REFERENCES Abarbanell, J. and R. Lehavy (2003), Can Stock Recommendations Predict Earnings Management and Analysts Earnings Forecast Errors?, Journal of Accounting Research, Vol. 41, No. 1 (March), pp Balsam, S., E. Bartov and C. Marquardt (2002), Accruals Management, Investor Sophistication, and Equity Valuation: Evidence from 10-Q Filings, Journal of Accounting Research, Vol. 40, No. 4 (September), pp

17 ACCRUALS MANAGEMENT TO ACHIEVE EARNINGS BENCHMARKS 669 Barth, M.E., J.A. Elliott and M.W. Finn (1999), Market Rewards Associated With Patterns of Increasing Earnings, Journal of Accounting Research, Vol. 37, No. 2 (Autumn), pp Bartov, E., D. Givoly and C. Hayn (2002), The Rewards to Meeting or Beating Earnings Expectations, The Journal of Accounting and Economics, Vol. 33, No. 2 (June), pp , F.A. Gul and J.L. Tsui (2000), Discretionary Accrual Models and Audit Qualifications, The Journal of Accounting and Economics, Vol. 30, No. 3 (December), pp Brown, L.D. (2001), A Temporal Analysis of Earnings Surprises: Profits Versus Losses, Journal of Accounting Research, Vol. 39, No. 2 (September), pp and M.L. Caylor (2005), A Temporal Analysis of Quarterly Earnings Thresholds: Propensities and Valuation Consequences, The Accounting Review, Vol. 80, No. 2 (April), pp Burgstahler, D. (1997), Incentives to Manage Earnings to Avoid Earnings Decreases and Losses: Evidence from Quarterly Earnings, Working Paper (University of Washington). and I. Dichev (1997a), Earnings Management to Avoid Earnings Decreases and Losses, Journal of Accounting and Economics, Vol. 24, No. 1 (December), pp (1997b), Earnings, Adaptation and Equity Value, The Accounting Review, Vol. 72, No. 2 (April), pp and M. Eames (2005), Management of Earnings and Analysts Forecasts to Achieve Zero and Small Positive Earnings Surprises, Working Paper (University of Washington). Bhushan, R. (1989), Firm Characteristics and Analyst Following, Journal of Accounting and Economics, Vol. 11, No. 2/3 (July), pp Cahan, S. (1992), The Effect of Antitrust Investigations on Discretionary Accruals: A Refined Test of the Political-Cost Hypothesis, The Accounting Review, Vol. 67, No.1 (January), pp Collins, D.W. and P. Hribar (2000), Earnings-Based and Accrual-Based Market Anomalies: One Effect or Two?, Journal of Accounting and Economics, Vol. 29, No. 1 (February), pp Das, S., C.B. Levine and K. Sivaramakrishnan (1998), Earnings Predictability and Bias in Analysts Earnings Forecasts, The Accounting Review, Vol. 73, No. 2 (April), pp Dechow, P.M., S.A. Richardson and A.I. Tuna (2003), Why are Earnings Kinky? An Examination of the Earnings Management Explanation, Review of Accounting Studies, Vol. 8, No. 2/3 (June-September), pp , R.G. Sloan and A.P. Sweeney (1995), Detecting Earnings Management, The Accounting Review, Vol. 70, No. 2 (April), pp DeFond, M.L. and C.W. Park (2001), The Reversal of Abnormal Accruals and the Market Valuation of Earnings Surprises, The Accounting Review, Vol. 76, No. 3 (July), pp Degeorge, F., J. Patel and R. Zeckhauser (1999), Earnings Management to Exceed Thresholds, Journal of Business, Vol. 72, No. 1 (January), pp Eames, M.J. and S.M. Glover (2003), Earnings Predictability and the Direction of Analysts Forecast Errors, The Accounting Review, Vol. 78, No. 3 (July), pp Greene, W.H. (1993), Econometric Analysis (2nd ed., Macmillan Publishing Company, New York). Hayn, C. (1995), The Information Content of Losses, The Journal of Accounting and Economics, Vol. 20, No. 2 (September), pp Hribar, P. and D.W. Collins (2002), Errors in Estimating Accruals: Implications for Empirical Research, Journal of Accounting Research, Vol. 40, No.1 (March), pp Jones, J.J. (1991), Earnings Management During Import Relief Investigations, Journal of Accounting Research, Vol. 29, No. 2 (Autumn), pp Kasznik, R. (1999), On the Association Between Voluntary Disclosure and Earnings Management, Journal of Accounting Research, Vol. 37, No. 1 (Spring), pp and M. McNichols (2002), Does Meeting Expectations Matter: Evidence From Analyst Revisions and Share Prices, Journal of Accounting Research, Vol. 40, No. 3 (June), pp Levitt, A. (1998), The Numbers Game, Remarks delivered at the NYU Center for Law and Business (New York, NY, September 28).