Market Reaction to Earnings Management: The Incremental Contribution of Analysts

Transcription

1 International Research Journal of Finance and Economics ISSN Issue 8 (2007) EuroJournals Publishing, Inc Market Reaction to Earnings Management: The Incremental Contribution of Analysts Ilanit Gavious School of Management, Department of Business Administration P.O.Box 653, Ben-Gurion University, Beer-Sheva 84105, Israel address: madaril@bgu.ac.il Tel.: ; Fax: Abstract We investigate whether analysts have an incremental contribution, beyond financial statement information, to investors' ability to detect and react to earnings management. Using both return and price analyses in a short-window design, we show that investors fixate on reported earnings and seem unable to disentangle earnings management information, at least during the ten days following the publication of financial statements. Markedly, thirty days after the date of disclosure, with additional timely information released by analysts (as captured by analyst recommendations and target prices), investors appear to reassess the quality of managed earnings. We show that analysts react negatively to firms that artificially inflate earnings and that this negative reaction is followed by an even stronger negative reaction by the market. The results are supported by a set of sensitivity tests that consider the robustness of our findings to the measure and extent of earnings management. Keywords: Earnings management; Analysts; Analyst recommendations; Analyst target price; Accruals JEL Classification: M41, G Introduction This study investigates how investors in the market react to earnings management (hereafter, EM) and whether analysts provide added value to the market in interpreting EM information. We consider whether evidence of EM in financial statements influences market prices and returns, and whether analysts provide investors with incremental information that may help them assess the integrity of reported earnings. Following Balsam et al. (2002), we focus on market reaction to the release of the full set of financial statements, and not to earnings announcements, as the latter may not include sufficient information (e.g., balance sheet and/or cash flow information) that can be used to disentangle the consequences of EM (see also, e.g., Baber et al., 2006; Balsam et al., 2002). Balsam et al. (2002) show that sophisticated investors as well as unsophisticated investors (proxied by the level of institutional ownership) are unable to recognize earnings management around earnings announcement date. Literature on market reaction to EM includes two strands of research; studies that examine the associations between security returns and accruals or accruals management in the context of longwindow designs (Sloan, 1996; Subramanyam, 1996; Rangan, 1998; Teoh et al., 1998a,b; Xie, 2001) and studies that examine these associations in the context of short-window designs (DeFond and Park,

2 197 International Research Journal of Finance and Economics - Issue 8 (2007) 2001; Balsam et al., 2002; Baber et al., 2006). Sloan (1996) provides evidence that investors naively fixate on earnings, failing to distinguish fully between the different properties of the accrual and cash flow components of earnings. Hence, stocks of firms with relatively high (low) levels of accruals are overpriced (underpriced), and given the differential persistence in cash flows and accruals, these stocks experience negative (positive) future abnormal returns around future earnings announcements. Xie (2001) also finds that the market overestimates the persistence (one-year-ahead earnings implications) of discretionary accruals, consequently overpricing these accruals. Rangan (1998) and Teoh et al. (1998a,b) investigate the long-run price effects of discretionary accruals in the context of equity issues and show that investors are deceived by EM. Subramanyam (1996) documents a positive correlation between annual discretionary accruals and twelve-month period stock returns (ending three months after fiscal year-end). In contrast to long-window studies, the short-window studies attempt to isolate reactions of investors to earnings disclosure to facilitate examination of their ability to detect and react to EM. DeFond and Park (2001) find higher (lower) earnings response coefficients (ERCs) when abnormal accruals unexpected working capital suppress (exaggerate) the magnitude of earnings surprises, consistent with market participants anticipating the reversing implications of abnormal accruals. However, based on analysis of subsequent stock returns, DeFond and Park conclude that market participants do not fully adjust for suspected EM at the earnings announcement date. Balsam et al. (2002) focus on investors' reaction to the release of the full set of financial statements in Form 10-Q (quarterly report), rather than on the reaction to the earnings announcement, because the Form 10-Q filing with the SEC provides information that can be used to asses the integrity of reported earnings. Balsam et al. find that unsophisticated investors, as indicated by low institutional ownership, react to EM revealed in the disclosure of financial statements; specifically, they document a negative association between evidence of EM and cumulative abnormal returns over a 17-day window around the 10-Q filing date (CAR(-1,+15)). In contrast, sophisticated investors react prior to this date (over a period ending two days prior to the filing date of Form 10-Q with the SEC). According to Balsam et al., the reaction of sophisticated investors precedes the release of Form 10-Q, probably due to their access to other, more timely sources of information (e.g., conference calls and private conversations with management). Baber et al. (2006) also investigate investor reaction to EM. Like DeFond and Park (2001), they focus on the influence of EM on interpretations of earnings at the time of earnings announcement. They find that security price reactions to quarterly earnings announcements depend on whether balance sheet and/or cash flow information is released concurrently with earnings press releases. Notably, in cases where balance sheet and/or cash flow information is voluntarily released, the negative price reaction to upward managed earnings is more substantial and more significant statistically. Baber et al. conclude that investors attempt to price-protect themselves against EM, and that their ability to do so is enhanced when firms disclose information that can be used to disentangle the consequences of EM. Thus, existing studies of EM and security pricing either provide evidence that investors are, in fact, misled by the accruals component of earnings or that they are, at least partially, able to detect and react to EM. Neither of these studies considers the role that analysts play in the process of market detection and interpretation of EM. Analysts (as well as external auditors of financial statements) play the role of independent intermediaries between companies and market participants (including investors, creditors and employees), and as such, are needed to minimize the risk of accounting frauds. Massive accounting frauds such as those revealed over the past several years (e.g., the cases of Enron, WorldCom, Global Crossing) catalyzed public criticism of the failure of analysts and auditors to warn of accounting irregularities, manipulated financial statements and imminent bankruptcies. In this study, we focus on the incremental contribution of analyst information over and above the information in

3 International Research Journal of Finance and Economics - Issue 8 (2007) 198 audited financial statements in the context of EM, in the era after revelations of high-profile accounting scandals. 1 In the first stage of our study, we examine the stock price reaction to EM in a short-window design (of up to thirty days following financial statements disclosure) using both price and return analyses. Moreover, we investigate if and how this reaction is affected by analyst information issued during this period. We capture analyst information using analyst target prices calculated based on updated valuations in response to the release of financial statements. 2 To evaluate EM, we utilize the contemporary measure of discretionary accruals estimated using the widely applied modified Jones (1991) model (see, e.g., Erickson and Wang, 1999; Klein, 2002; Geiger et al., 2005). Each firm's estimated discretionary accruals are adjusted for a performance-matched firm's discretionary accruals based on Kothari et al.'s (2005) study which advocates performance-matching to control for the impact of firm performance on accruals. 3 We observe average performance-matched discretionary accruals of 18% of lagged total assets for firms with upward managed earnings as well as for firms with downward managed earnings. Additionally, the performance-matched discretionary accruals are at least one percent of total assets (current and lagged). In their study, Balsam et al. (2002) also require that managed accruals 4 be at least one percent of total assets; they explain that Given the measurement error inherent in discretionary accrual estimation, this requirement helps to eliminate firms that are less likely to have managed earnings. Furthermore, a relatively large discretionary component of earnings helps to assure that investors are able to detect its presence. (Balsam et al., 2002). We present evidence that investors in the market fixate on earnings and are unable to interpret EM information during a time-window of at least ten days following the release of financial statements. Markedly, thirty days after disclosure of financial statements, with the additional timely information released by analysts, 5 discretionary accruals become value relevant to stock pricing. At this point in time, investors in the market seem to utilize the incremental information provided by analysts to reassess the quality of reported earnings. We show that analysts react negatively to firms that artificially inflate earnings (expressed by reduction of analyst target prices for these firms), and that this negative reaction is followed by an even stronger negative reaction by the market, according to both price and return analyses. In the second stage of the study we perform sensitivity analyses that consider the robustness of our findings to our measure of EM and to the extent of EM (magnitude of abnormal accruals). First, we use estimated discretionary accruals, without the adjustment for a performance-matched firm's discretionary accruals. Second, we employ an alternative measure of EM, the non-estimated measure of nonoperating accruals introduced by Givoly and Hayn (2000). Third, we sort our sample based on the magnitudes of abnormal accruals and divide it into three thirds; we repeat the analyses on a subsample that includes only the top and bottom thirds. This procedure, which limits the sample to a subset of extreme abnormal accruals, further increases the likelihood that earnings have indeed been managed and that this will be detected by investors. The results of the sensitivity tests show that our findings are robust to the measure and to the extent of EM. Altogether, the three sensitivity tests yield results that support our confidence in the validity of our findings. 1 Prior literature has focused on analysts' ability to anticipate and forecast EM; some studies show that analysts anticipate and incorporate EM in their forecasts (e.g., Burgstahler and Eames, 2003), whereas others find analysts are unable or unmotivated to anticipate fully firms' earnings management in forecasts (Abarbanell and Lehavy, 2003a) (see also, e.g., Degeorge et al., 1999; Abarbanell and Lehavy, 2003b). This study, in contrast, focuses on analysts' reaction ex-post to EM to investigate their contribution to investor ability to detect and interpret EM in disclosed financial statements. 2 Analyst information can be alternatively captured by the short recommendations issued to firms. Analyst recommendations are based on a six point scale: strong buy, buy, accumulate, hold, underperform and sell, and are usually accompanied by target prices. Target prices are preferable to our analyses over recommendations, because they are a quantitative financial measure, like the rest of variables used in our analyses. Additionally, given that recommendations can take one out of only six categories makes it a rough measure compared to the continuous variable of target price that can take any value and is calculated for each firm specifically. Indeed, when we apply the categorical variable of recommendations in our tests instead of target prices, we generally obtain same qualitative results, however statistically weaker. 3 Performance matching is done on the basis of industry, year and current ROA, consistent with Kothari et al.'s (2005) performance-matching approach. The same approach is also employed in the study of Gavious (2006) which explores analysts' reaction to EM. 4 Balsam et al.'s (2002) proxy for managed accruals is unexpected discretionary accruals calculated as the seasoned difference in discretionary accruals. 5 The majority of analyst recommendations are issued between ten and thirty days following the release of annual financial statements.

4 199 International Research Journal of Finance and Economics - Issue 8 (2007) The remainder of the paper is organized as follows. Section 2 outlines our research design, and Section 3 describes the data. Section 4 presents the results, and Section 5 provides a summary and concluding remarks. 2. Research Design We analyze market reaction to EM by exploring whether the presence of EM affects price reactions to the disclosure of full set of financial statements. We employ price regressions analysis and event study analyses over four time-windows around the date of disclosure. The equity price regressions are estimated using stock prices one day, three days, ten days and thirty days following this date. Concomitantly, cumulative abnormal returns (CARs) are estimated over a 3-day, 5-day, 12-day and 32- day window (where day 0 is the date of financial statements disclosure); i.e., windows of (-1,+1), (- 1,+3), (-1,+10), (-1,+30), respectively. The wider time windows are expected to allow the marginal investor to make sense of the reported accounting figures, interpret EM information (e.g., disaggregate accruals into their discretionary and nondiscretionary components) and assess earnings integrity (see also Balsam et al., 2002). We focus on the period around the disclosure of annual financial statements and not on earnings announcements, because for our analysis, we require that investors have all the information necessary to assess the integrity and quality of the reported earnings (Balsam et al., 2002; Baber et al., 2006). Our database indicates that the lion s share of analyst target prices is issued after the release of financial statements and not during the period between earnings announcement and the release of financial statements. Moreover, the vast majority of analyst target prices are issued in the period between ten and thirty days following financial statements disclosure. Following prior value-relevance studies, we run a price regression model on a combination of reported accounting fundamentals, under the assumption that the latter provide investors with valuerelevant information (Amir and Lev, 1996): MVit = β 0 + β1bvit + β 2 Eit + β 3NEG _ Eit + ε it (1) where MV is market value of equity measured one day, three days, ten days or thirty days after financial statements disclosure; BV is book value of equity; E is earnings before extraordinary items; NEG_E is earnings before extraordinary items if earnings before extraordinary items <0, 0 otherwise. We differentiate between value implications of positive and negative earnings based on prior literature that documents differences in the valuation of profits and losses (e.g., Hayn, 1995; Collins et al., 1997; Basu, 1997). We also run the same specification with cash flows from operations (CFO) instead of earnings. Because cash flows are considered a more substantive component of earnings than the accrual components, we seek to explore whether investors react differently to the former in the presence of EM. To test whether and how the presence of EM affects the pricing of earnings news, we add to regression model (1) an explanatory variable to proxy for EM. Equation (1) thus becomes: MVit = β 0 + β1bvit + β 2Eit + β 3NEG _ Eit + β 4DISC _ ACCit + ε it (2) where DISC_ACC is unexpected discretionary accruals, adjusted for firm performance (the estimation of DISC_ACC is explained henceforth). We hypothesize a negative association between unexpected discretionary accruals and security price reactions; that is, β 4 <0. As positive (negative) unexpected discretionary accruals indicate that the recently reported earnings have been artificially inflated (deflated), security prices of these firms are expected to decrease (increase) if investors are able to detect the presence of EM. Estimating Eq. (2) over the four alternative time-windows allows us to examine not only if, but also how long it takes investors to detect and interpret EM. As long as EM is undetected, we expect the estimated coefficient on DISC_ACC to be insignificantly different from zero.

5 International Research Journal of Finance and Economics - Issue 8 (2007) 200 We deflate Eq. (2) by the book value of equity. 6 Our deflated regression model is: MVit / BVit = β 01/ BVit + β1 + β 2Eit / BVit + β 3NEG _ Eit / BVit (3) + β 4DISC _ ACCit / BVit + ε it The intercept in the deflated model can be interpreted as the coefficient on book value of equity in the undeflated model. Consistent with prior studies, we retain the inverse of book value of equity as an explanatory variable in the deflated regression model because we include the intercept in the unscaled model to explain economic variation in market values that is not captured by our other explanatory variables (Core et al., 2003). In the regression of market value measured thirty days after financial statements disclosure we add another explanatory variable, median analyst target price (TP) deflated by book value of equity. Using mean analyst target price rather than median to proxy for a composite target price for a company does not change our qualitative results. The inclusion of information published by analysts in response to the disclosure of financial statements allows us to explore whether analysts provide incremental information content to investors in the market, over and above that provided by the audited financial statements. This is particularly interesting in the case of manipulated financial statements. EM is proxied for by the discretionary accruals component of reported earnings. Discretionary accruals are unexpected accounting accruals identified by the modified Jones (1991) model. For each firm in our sample, we estimate the following model: TACC = α + β *( REV _ DEL AR _ DEL ) + β * GPPE + v (4) i, t i 1i i, t i, t 2i i, t i where TACC is total accruals, REV_DEL is the change in revenues from previous year, AR_DEL is the change in accounts receivable, and GPPE is gross fixed assets. Each variable, including the intercept, is deflated by beginning-of-year total assets. Consistent with prior research, total accruals are net income minus cash flows from operations. The residual in the regression model ( v ) is the measure of discretionary accruals. We find that discretionary accruals differ significantly (p-value 0.000) between industries; thus to control for industry effects, we allow in regression (4) for different industries to have different slope coefficients, as well as different intercepts. 7 Because EM is related to firm performance (i.e., firms with extreme financial performance are likely to engage in earnings management; see Kothari et al. 2005), discretionary accruals models such as the above might be misspecified, as they ignore the impact of performance on accruals. Kothari et al. (2005) show that discretionary accruals estimated using the Jones or the modified-jones model, and adjusted for a performance-matched firm's discretionary accrual, tend to be the best specified measures of discretionary accruals Thus, consistent with Kothari et al.'s performancematching approach, we adjust each firm's estimated discretionary accruals by subtracting the corresponding estimated discretionary accruals of a matched firm (henceforth, adjusted discretionary accruals ). Our matching procedure is as follows: from the Yif at Capital Disk Co. database (see Section 3), we identify firms with sufficient data to compute discretionary accruals and return on assets (current ROA computed as net income divided by total assets). 8 Then, in keeping with Kothari et al., we match each firm-year observation in our sample with that observation in the control sample from the same industry and year with the closest performance (ROA) 9. The performance-matching approach distinguishes between normal and abnormal EM, where abnormal EM firms are those that manage more than would be expected given their level of performance (Kothari et al., 2005). We complement the price-levels regressions analysis with a return analysis. Using an event study methodology, we explore changes in analyst target prices and excess returns (CARs) around the release of financial statements by income-increasing/decreasing management firms. Change in analyst target price is measured as the median percentage change in target price relative to last target price 6 Our sample is restricted to firms with positive book value (see also, e.g., Collins et al., 1997; Brown et al., 1999; Core et al., 2003). Lo (2004) advocates deflating financial data in accounting research by a proxy for scale, rather than including a scale proxy as an independent variable. The advantages of 2 deflation by a scale proxy include, inter alia, mitigation of hetroscedasticity, R bias and coefficient bias. 7 Industries are as defined in the databases described in Section 3 8 Kothari et al. (2005) show that matching based on current ROA performs better than matching based on prior year's ROA. 9 The same performance-matching procedure is also applied in the study of Gavious (2006).

6 201 International Research Journal of Finance and Economics - Issue 8 (2007) issued to the company by the same investment house prior to disclosure of the financial statements. Taking mean percentage change in target price rather than median does not change inferences from the analysis. CARs are computed using the market model estimated from 210 trading days up to 30 days prior to the date of financial statements disclosure. In the second stage of the study, we perform sensitivity analyses that consider the robustness of our findings to our measure of EM and to the extent of EM. First, we use estimated total discretionary accruals, without the adjustment for a performance-matched firm's discretionary accruals. Second, we use a non-estimated measure of EM, nonoperating accruals (Givoly and Hayn, 2000). Following Geiger et al. (2005), we chose nonoperating accruals as a second measure of EM to address empirical concerns regarding the Jones model (Dechow et al., 1995; Erickson and Wang, 1999) (Geiger et al., 2005). 10 Based on Givoly and Hayn (2000), nonoperating accruals are calculated as net income plus depreciation and amortization minus cash flows from operations minus operating accruals; where operating accruals are defined as change in accounts receivables plus change in inventories plus change in prepaid expenses minus change in accounts payable minus change in taxes payable. To control for size effects, we scale nonoperating accruals by beginning-of-year total assets (consistent with the scaling of the modified Jones model). According to Givoly and Hayn (2000), nonoperating accruals consist primarily of such items as loss and bad debt provisions (or their reversal), restructuring charges, the effect of changes in estimates, gains or losses on the sale of assets, asset write-downs, the accrual and capitalization of expenses, and the deferral of revenues and their subsequent recognition. Given that nonoperating accruals include items which are under the discretion of management (in terms of timing and/or estimation of recorded amounts), they are used to indicate whether firms actively engage in EM. Third, we repeat our analysis on a subsample that includes only extreme managed accruals - the top and bottom thirds of adjusted discretionary accruals in our sample. This procedure is used to further ensure that firms that are less likely to have managed earnings are eliminated from the analysis. This requirement also helps increase the likelihood that investors detect EM, if it exists. To deal with outliers, we winsorize extreme observations for all variables (top and bottom 1%). We winsorize rather than delete outliers in order to conserve data. The results do not change qualitatively when outliers are deleted. The regressions are estimated using panel data (same companies in successive years) with industry and year fixed effects. Namely, we include intercept dummies for each industry and year to capture constant industry-specific and year-specific factors. The regressions include White's (1980) correction. 3. Data We obtain analyst target prices issued for Israeli firms listed on the New York Stock Exchange (NYSE) from 2001 to 2004, as taken from TheMarker database. 11 The following information was extracted from the database: date of analyst recommendation, analyst name, investment house, analyst's target price for the firm's share, 12 prior target price, market price at time of recommendation, and firm's industry. Firms' accounting information, the identity of the auditor/s for current and previous year and daily share prices were obtained from the database in the Yif at Capital Disk Co. Following prior literature, we exclude from the analysis financial and utility firms that are subject to more complex earnings-management incentives due to regulation or other factors (Burgstahler and Eames, 2003; see also Burgstahler and Dichev, 1997 and Rosner, 2003). 10 See also Gavious (2006). 11 Our sample includes the target prices issued by US-based analysts for Israeli firms listed on the NYSE. Israeli firms that are listed on the NYSE report their financial statements in accordance with US Generally Accepted Accounting Principles. 12 Target price relates to the period of one year from time of recommendation.

7 International Research Journal of Finance and Economics - Issue 8 (2007) 202 Our sample includes 367 target prices issued for 60 firms during the one-month period after disclosure of financial statements for fiscal years (resulting in 128 firm-years). The sample is categorized into four industries: software systems and devices; electronics, electrics and optics; pharmaceuticals and chemicals; and other. Other primarily consists of commerce and industrial firms operating in traditional industries (e.g., textile, real estate, food). Table 1 presents descriptive statistics for our sample. We divide the sample into two subsamples by the sign of adjusted discretionary accruals. Positive (negative) adjusted discretionary accruals are considered to signal income-increasing (decreasing) management. Table 1 shows that analyst target price scaled by book value (TP/BV) is lower for firms engaging in income-increasing management. According to Gavious (2006), analysts' negative reaction to income-increasing management, expressed in reduced target prices, is consistent with evidence of analyst reluctance to follow or start covering firms that extensively manage their earnings upwards. It is shown in Table 1 that the average number of analysts following an incomeincreasing firm is lower than that for an income-decreasing firm (3.333 and 5.759, respectively). Table 1: Descriptive statistics Income-increasing (decreasing) management firms are those with positive (negative) adjusted discretionary accruals. TP/BV is median target price per share, published by the analyst within a month after financial statements disclosure, divided by book value of equity per share. MVt/BV is share market price t days after financial statements disclosure divided by book value of equity per share. DISC_ACC/TA is adjusted discretionary accruals divided by total assets (beginning of year). Discretionary accruals are from the modified Jones (1991) model, and are adjusted for a performancematched firm's discretionary accruals. Performance matching is on the basis of industry, year and current ROA. Size is total assets (in millions). MV is end-of-year market value of equity (in millions). ROA is net income before extraordinary items divided by total assets (beginning of year). Leverage is long-term debt divided by total assets. Sales_gr is average percentage sales growth over the last three years. CurrRatio is current assets divided by current liabilities. Average number of analysts following a firm is calculated as the total number of analyst recommendations available for income-increasing (decreasing) management firms divided by the number of these firms. Big4 Auditor represents the percentage of firms that were audited by one or more of the Big4 accounting firms. Auditor change represents the number of cases in which an auditor has been replaced by another. Variable TP/BV MV1/BV MV3/BV MV10/BV MV30/BV DISC_ACC/TA Size (in $ millions) MV (in $ millions) ROA Leverage Sales_gr CurrRatio Income-increasing management (N=63) Income-decreasing management (N=65) Mean Median Mean Median Mean Median Mean Median Mean Median Mean Median Mean Median Mean , Median Mean Median Mean Median Mean Median Mean Median

8 203 International Research Journal of Finance and Economics - Issue 8 (2007) Average number of analysts following a firm Big4 Auditor (% of cases) 82.5% 81.5% Auditor change (#cases) 6 5 Concomitantly to the findings of lower analyst target price-to-book for income increasing firms, we find that market-to-book ratios (MV/BV) measured one to thirty days after disclosure of financial statements are also consistently lower for these firms. Adjusted discretionary accruals are, on average (median), 18% (8% and 7%) of total assets for income-increasing and income-decreasing firms, respectively. Minimum adjusted discretionary accruals for both groups of firms reach higher than one percent of total assets, the same scale of managed accruals required in the study of Balsam et al. (2002) in order to exclude firms that are less likely to have managed earnings. Firm size, as measured by total assets, is lower for income-increasing firms (mean (median) of $447 ($248) million compared with $659 ($339) million for income-decreasing firms). Concomitantly, market value is lower for income-increasing firms (mean (median) of $684 ($232) million compared with $1,287 ($295) million for income-decreasing firms). 13 Firm size is generally correlated with the sophistication of firm's accounting and internal control systems, which increase the difficulty of strategically falsifying accounting numbers and managing accounting choices. It is thus possible that the likelihood of a fraudulent activity would be higher in smaller firms. Manipulation of accounting figures in a deliberate attempt to deceive investors usually characterizes income-increasing management, whereas income-decreasing management is many times a result of management necessity. For example, in cases of severe financial distress, management may engage in liquidity-enhancing transactions (e.g., delaying payables or factoring receivables) that result in large negative accruals (see Butler et al., 2004). Hence, abnormal negative accruals in such firms would be a consequence of liquidity-related transactions due to financial distress rather than a consequence of EM (in the following section we review different scenarios in which income-decreasing management generally occurs). Measures of firm performance indicate that sales growth in income-increasing firms is higher, whereas profitability as measured by return on assets is close to that of income-decreasing firms. The differences in sales growth and profitability between income-increasing and income-decreasing firms are statistically insignificant. The artificial inflation/deflation of earnings seems to obscure performance differences between the two groups of firms. Income-decreasing firms in our sample have an insignificantly higher degree of leverage and a significantly lower degree of liquidity (both mean and median at the 10% level). The lower (higher) liquidity as measured by the ratio of current assets to current liabilities may be the result of abnormal negative (positive) accruals in income-decreasing (increasing) firms, and not necessarily the result of lower (higher) economic performance. Finally, Table 1 shows that income-increasing and income-decreasing firms do not differ in the size or reputation of their auditors (82% of both subsamples were audited by Big4 accounting firms). Also, the frequency of auditor replacements is quiet similar in both subsamples (6 (5) cases within incomeincreasing (decreasing) firms). The low frequency of change of auditors is not surprising given that audit engagements are typically long term, even in the presence of EM. Moreover, Butler et al. (2004) show that audit opinions are not affected by EM. According to Butler at al., The auditor and the company likely resolve earnings-management-related issues before the audit report is issued because the costs of not doing so are so high. As a result an audit opinion modification due to earnings management should be rare. 13 The requirement that sample firms be followed by analysts may have biased our sample towards larger firms of interest to analysts.

9 International Research Journal of Finance and Economics - Issue 8 (2007) Empirical Findings 4.1. Regression analysis Table 2 reports the parameter estimates together with their significance levels for regression model (3). The regressions results presented in panel a suggest that market value one, three and ten days after financial statements disclosure (MV1/BV, MV3/BV and MV10/BV, respectively) is affected only by earnings information. Positive earnings are, as expected, significantly positively related to market values (p-value=0.000). The coefficient on negative earnings (sum of coefficients on E/BV and NEG_E/BV) is insignificantly negative, indicating that negative earnings are value irrelevant. A possible explanation for the value irrelevance of negative earnings may be related to the fact that it is harder to base forecasts on losses. For example, management of a firm with losses may choose to change its business strategy or take other actions to turn to positive (profitable) results. Such actions are generally difficult to predict; even if management releases information regarding its plans for the future, the actual realization of these plans and their outcome are still difficult to predict. The intercept in the regressions of MV1/BV, MV3/BV and MV10/BV, which can be interpreted as the coefficient on book value of equity in the undeflated model, is insignificantly positive. The coefficient on adjusted discretionary accruals is, as hypothesized, negative however insignificant. 14 These results imply that investors in the market, at least during the ten days following the disclosure of financial statements, fixate on reported earnings and are unable to interpret EM information. In other words, the market seems to be misled by the accruals components of earnings during this short-time window. However, in the regression of market value thirty days after financial statements disclosure (MV30/BV), where additional timely information from analysts is available to investors, both book value of equity and discretionary accruals become value relevant. Investors seem to make use of this incremental information to reassess the quality of reported earnings. As shown in Table 2, the coefficient on analyst target price is, as expected, significantly positive. The coefficient on positive (negative) earnings remains (in)significantly positive (different from zero), however smaller relative to the regressions where only accounting information is included. The intercept the coefficient on book value of equity in the undeflated model becomes highly significant and, as expected for a price level regression, close to 1. The coefficient on adjusted discretionary accruals becomes markedly more negative and highly significant (p-value=0.000). The negative market price reaction to upward managed earnings is consistent with findings of prior studies, such as Baber et al. (2006) and Balsam et al. (2002). Overall, the regressions results demonstrate that information from analysts has a significant incremental contribution to the ability of investors to detect and react to managed earnings. In panel b of Table 2, earnings are replaced with cash flows from operations. When we include cash flows in the regressions instead of earnings, inferences are generally the same. Up until ten days post full financial disclosure, positive (negative) cash flows from operations are (in)significantly positively (negatively) related to market values. The intercept and the coefficient on adjusted discretionary accruals are insignificant. However, in the regression of MV30/BV, with analyst target prices included in the regression (and significantly positively related to market values), the intercept and coefficient on adjusted discretionary accruals become significantly positive and negative, respectively. We point to one qualitative difference found when regressing MV30/BV on cash flows instead of earnings. Whereas the coefficient on NEG_E/BV is significantly negative (panel a), the coefficient on NEG_CF/BV is insignificantly different from zero. This means that negative and positive cash flows do not differ in their valuation implications, i.e., both are significantly positively related to market value; the more negative (positive) are the cash flows from operations, the lower (higher) is the market valuation of the firm. This difference in the valuation of negative earnings and negative cash flows one month after financial statements disclosure implies that negative cash flows are more value relevant than negative earnings. Investors in the market seem to regard negative cash flows from operations as a value-destructing feature. It may be related to investor ability to appreciate 14 We also run regressions of market values on the accruals variable alone; the coefficient on managed accruals whether adjusted discretionary accruals or nonoperating accruals used in the sensitivity analysis (Section 4.3) remains insignificant.

10 205 International Research Journal of Finance and Economics - Issue 8 (2007) that negative cash flows from operations are a more substantive component of negative earnings (relative to accruals).

11 International Research Journal of Finance and Economics - Issue 8 (2007) 206 Table 2: Price regressions on accounting and analyst information The table reports the parameter estimates together with their significance levels for regression model: MV / BV = β 1/ BV + β + β E / BV + β NEG _ E / BV it it 0 it 1 2 it it 3 it it + β DISC _ ACC / BV + ε 4 it it it where MV is market value of equity measured one day, three days, ten days or thirty days after financial statements disclosure; BV is book value of equity; E is earnings before extraordinary items; NEG_E is earnings before extraordinary items if earnings before extraordinary items <0, 0 otherwise; DISC_ACC is adjusted discretionary accruals. In panel b earnings are replaced with cash flows from operations (CFO). The intercept ( β ) can be interpreted as the coefficient on book value of equity in 1 the undeflated model. In the regression of market value measured thirty days after financial statements disclosure we add another explanatory variable, median target price (TP) published by analysts during this period. Using mean analyst target price rather than median to proxy for a composite target price for a company does not change our qualitative results. We include intercept dummies in the regressions for each industry and year to control for industry and time effects. To deal with outliers, the independent and dependent variables are winsorized (top and bottom 1%). P-values of the coefficients are presented in parentheses. Panel a: accounting fundamentals include earnings and book value of equity MV1/BV MV3/BV MV10/BV MV30/BV Independent variables: Intercept (0.371) (0.494) (0.526) (0.002) 1/BV (0.920) (0.977) (0.991) (0.023) E/BV (0.000) (0.000) (0.000) (0.016) NEG_E/BV (0.000) (0.000) (0.000) (0.057) DISC_ACC/BV (0.450) (0.381) (0.360) (0.000) TP/BV (0.000) Adj_R-squared F-value (0.000) (0.000) (0.000) (0.000) Panel b: accounting fundamentals include cash flows from operations and book value of equity MV1/BV MV3/BV MV10/BV MV30/BV Independent variables: Intercept (0.845) (0.954) (0.928) (0.018) 1/BV (0.815) (0.778) (0.757) (0.074) CFO/BV (0.000) (0.000) (0.000) (0.009) NEG_CFO/BV (0.000) (0.000) (0.000) (0.977) DISC_ACC/BV (0.650) (0.702) (0.733) (0.017) TP/BV (0.000) Adj_R-squared F-value (0.000) (0.000) (0.000) (0.000)

12 207 International Research Journal of Finance and Economics - Issue 8 (2007) 4.2. Return analysis Table 3 presents the percentage change in analyst target price ( TP(%) ) and CARs around the disclosure of financial statements by income-increasing and income-decreasing firms. The results show that analysts significantly reduced their target prices to income-increasing firms by 3.5%, on average. Market response to income-increasing management during the ten days following financial statements disclosure, as measured by CAR(-1,1) and CAR(-1,10), is insignificantly negative (-1% and -2%, respectively). 15 However, thirty days after financial statements disclosure, investor response becomes significantly negative with CAR (-1,30) of -5.2%. Table 3: Return analysis The table presents the average change in analyst target price ( TP(%) ) and excess returns (CAR) in response to the release of full set of financial statements by income-increasing and income-decreasing management firms. Income-increasing (decreasing) management firms are those with positive (negative) adjusted discretionary accruals. TP(%) is median percentage change in analyst target price relative to prior target price (last target price issued to the company by the same investment house before financial statements disclosure). Taking mean percentage change in target price rather than median does not change inferences from the analysis. CARs are cumulative abnormal returns around the date of financial statements disclosure. We compute the abnormal returns using the market model estimated from 210 trading days up to 30 days prior to the date of financial statements disclosure; the parameter estimates are used to compute riskadjusted abnormal returns. * and ** indicate 10% and 5% significance level in a two-tailed t-test/wilcoxon signed ranks test. Income-increasing management Income-decreasing management TP(%) * CAR(-1,1) CAR(-1,10) CAR(-1,30) ** The results of the return analysis support our inferences from the regression analysis. As expected, analysts, on average, react negatively to income-increasing management by reducing target prices issued for these firms. This is consistent with the documented lower TP/BV for incomeincreasing firms (Table 1). During the short-time windows around the publication of financial statements, while analysts have not yet issued their updated recommendations and target prices, investors in the market seem to be hesitant in their response to the reported figures. Thirty days after firms' financial reporting, with analyst responses to these reports being released, abnormal returns become significantly negative for income-increasing firms. Hence, like in the regression analysis, analyst information appears to be incrementally value-relevant for investors in the market. Whereas on their own investors appear to be unable to detect and respond to EM, with the additional information released by analysts, they seem to reassess earnings integrity and coincide with analyst response. Moreover, negative investor response to income-increasing management is even stronger than that of analysts; with an average CAR of -5.2% on shares of firms that artificially inflated earnings compared with 3.5% reduction in analyst target prices for these firms. For income-decreasing firms, analyst target prices increased by 2.5%, on average, followed by an excess return (CAR(-1,30)) of 3.4%; however, these changes are statistically insignificant. To try to explain this insignificant positive response to income-decreasing management, we explore the reasons that may drive firms to end up with abnormal negative accruals (see also Gavious, 2006). One possible scenario for downward EM is when management chooses to take a big bath as it realizes the company 15 Balsam et al. (2002) document an insignificant response to EM by unsophisticated investors during the short time-windows of three and seven days around the filing date of financial statements.

13 International Research Journal of Finance and Economics - Issue 8 (2007) 208 cannot meet the benchmark set by analysts (Hayn, 1995; Brown, 1997; Burgstahler and Dichev, 1997). Another possible scenario is that of management change; where poor earnings are attributed to the former management in order to necessitate the change (see, e.g., DeAngelo, 1988). DeAngelo et al. (1994) and Butler et al. (2004) relate negative abnormal accruals to financial distress. DeAngelo et al. (1994) document incentives to reduce earnings by managers of financially distressed firms engaged in contractual renegotiations. According to Butler et al. (2004), firms in severe financial distress may engage in liquidity-enhancing transactions resulting in large negative accruals. They claim that traditional abnormal accruals models do not adequately control for distress-induced variation in accruals, and thus may misclassify these accruals as discretionary, leading to the belief that a given distressed firm manages its earnings when, in fact, it does not. If abnormal negative accruals represent management necessity rather than an attempt to deceive investors and other stakeholders, than analyst reactions may be non-negative. The non-significant reaction of analysts, followed by a non-significant reaction of the market, implies an uncertainty with which these negative accruals firms are judged. This is as opposed to the significant negative reaction to abnormal positive accruals (indicating income-increasing management) Sensitivity analysis This section reports the results of three sensitivity analyses. First, we replicate our regression and return analyses using estimated discretionary accruals, without the adjustment for a performancematched firm's discretionary accruals, and obtain same qualitative results (untabulated). Second, we replicate the analyses using nonoperating accruals as an alternative measure of EM. The results of the regressions and returns analyses are displayed in Tables 4 and 5, respectively. Again, the results indicate that our findings are robust to the measure used to proxy for EM. The coefficient on nonoperating accruals (NONOP_ACC/BV) in all three regressions of MV1/BV, MV3/BV and MV10/BV is more negative than the coefficient on adjusted discretionary accruals (panel a of Tables 4 and 2, respectively), however still statistically insignificant. Thirty days after financial statements disclosure, analysts provide incremental information content to the market that helps investors acknowledge the value implications of nonoperating accruals. At this point in time, the intercept (0.920) and the coefficient on nonoperating accruals (-0.831) become statistically significant at the 1% and 5% level, respectively. The regressions presented in panel b of Table 4, where earnings are replaced with cash flows, also yield the same inferences as those obtained from the regressions based on discretionary accruals measure of EM (Table 2 panel b). In all, the power of explanation of the price 2 regressions (measured as the adjusted R ) with EM proxied for by nonoperating accruals is close to that obtained using adjusted discretionary accruals to proxy for EM.

14 209 International Research Journal of Finance and Economics - Issue 8 (2007) Table 4: Price regressions on accounting and analyst information using an alternative measure to proxy for EM The regression model and variables are as defined in Table 2, except for the measure of earnings management, nonoperating accruals (NONOP_ACC). Based on Givoly and Hayn (2000), nonoperating accruals are calculated as net income plus depreciation and amortization minus cash flows from operations minus operating accruals. Operating accruals are defined as: Accounts Receivables + Inventories + Prepaid Expenses - Accounts Payable - Taxes Payable. We include intercept dummies in the regressions for each industry and year to control for industry and time effects. To deal with outliers, the independent and dependent variables are winsorized (top and bottom 1%). P-values of the coefficients are presented in parentheses. Panel a: accounting fundamentals include earnings and book value of equity MV1/BV MV3/BV MV10/BV MV30/BV Independent variables: Intercept (0.571) (0.714) (0.752) (0.005) 1/BV (0.804) (0.876) (0.890) (0.001) E/BV (0.000) (0.000) (0.000) (0.013) NEG_E/BV (0.000) (0.000) (0.000) (0.100) NONOP_ACC/BV (0.108) (0.113) (0.106) (0.045) TP/BV (0.000) Adj_R-squared F-value (0.000) (0.000) (0.000) (0.000) Panel b: accounting fundamentals include cash flows from operations and book value of equity MV1/BV MV3/BV MV10/BV MV30/BV Independent variables: Intercept (0.950) (0.842) (0.816) (0.047) 1/BV (0.939) (0.904) (0.878) (0.019) CF/BV (0.000) (0.000) (0.000) (0.009) NEG_CF/BV (0.001) (0.001) (0.000) (0.941) NONOP_ACC/BV (0.608) (0.569) (0.582) (0.095) TP/BV (0.000) Adj_R-squared F-value (0.000) (0.000) (0.000) (0.000) The return analysis with nonoperating accruals used as measure of EM, displayed in Table 5, also provides results consistent with those obtained using adjusted discretionary accruals. A significantly negative abnormal return of 5.8%, on average, on shares of income-increasing firms follows a significant reduction in analyst target prices by 3.1%. No immediate response by the market is apparent (insignificant CAR(-1,1) and CAR(-1,10)). As for income-decreasing firms, neither analysts nor the market responded significantly, regardless of the method used to proxy for EM.