Asset Write-downs and Information Asymmetry: Do Big Baths Muddy the Waters or Clear the Air?

Transcription

1 Asset Write-downs and Information Asymmetry: Do Big Baths Muddy the Waters or Clear the Air? K. Stephen Haggard Department of Finance and General Business College of Business Administration Missouri State University John S. Howe Department of Finance Robert J. Trulaske, Sr. College of Business University of Missouri Andrew A. Lynch Department of Finance Robert J. Trulaske, Sr. College of Business University of Missouri July 2011

2 Asset Write-downs and Information Asymmetry: Do Big Baths Muddy the Waters or Clear the Air? We examine the intertemporal relation between large asset write-downs (big baths) and information asymmetry. Theory suggests that baths facilitate future earnings smoothing (Kirschenheiter and Melumad (2002)), but does not predict whether this smoothing improves or degrades earnings quality. We use baths as a natural experiment to test changes in earnings quality, measuring changes in firm information asymmetry to determine whether a bath improves or degrades earnings quality. Analyzing 12 generally accepted proxies of information asymmetry, we find strong evidence that a firm s information environment degrades substantially following baths. The increase in information asymmetry is strongest in the year following a bath and weakens over the two following years. We conclude that baths facilitate future earnings smoothing, and that the increase in smoothing degrades earnings quality. JEL Classification code: M41 Key words: Liquidity, information asymmetry, cost of capital, earnings smoothing

3 I. Introduction Research has shed considerable light on the causes of large asset write-downs ( baths ) and short-term market reactions to them (e.g., Francis, Hanna and Vincent (1996)). Yet there remains uncertainty as to the consequences of a bath for a firm s information environment. Kirschenheither and Melumad (2002) propose a model in which managers use a bath to facilitate future earnings smoothing, suggesting that baths alter the quality of future earnings disclosures. 1 However, they do not address whether a bath positively or negatively impacts future earnings quality. If a bath clears the firm s balance sheet of materially impaired assets, future smoothed earnings become more economically representative of future cash flows. In this case, a bath improves earnings quality and reduces information asymmetry. If managers write down assets to facilitate manipulation of future reported earnings, such earnings become less economically representative of future cash flows. In this case, a bath harms earnings quality and increases information asymmetry. The purpose of this paper is two-fold. First, we empirically investigate the theoretical relation proposed by Kirschenheiter and Melumad (2002) between baths and future earnings smoothing. To our knowledge we are the first to do so, and we confirm their central prediction that the post-bath period will be characterized by earnings smoothing. Second, we use the change in information asymmetry around baths to test whether the increased earnings smoothing following baths results in increased or decreased earnings quality. In addition to being our metric of earnings quality, information asymmetry also influences the cost of capital (Easley and O'Hara (2004)), equity trading costs (Huang and Stoll (1997)), and uninformed investor returns 1 A similar argument is made by LaFond and Watts (2008), though they additionally assume increased smoothing post bath will increase information asymmetry. 1

4 (Easley, Hvidkjaer and O'Hara (2002)). Thus, changes in information asymmetry are important to a number of market participants. The ambiguity of the relation between earnings smoothing and earnings quality is discussed by Dechow and Skinner (2000). 2 Accruals-based accounting, designed to create more economically meaningful earnings by intertemporally matching revenues and expenses, mechanically smoothes earnings. By construction, these smoothed earnings can convey more information to outsiders than non-smoothed cash-based earnings by stripping away volatile transient cash flows, matching revenues to expenses and making earnings more persistent. 3 However, the flexibility granted to managers by accruals-based accounting can also result in excess earnings smoothing. Motivated by performance compensation and the desire to retain their jobs, managers might over-smooth earnings and, consequently, reduce their information content. The difficulty in disentangling economically meaningful earnings smoothing from excess smoothing has resulted in contradictory theoretical and empirical evidence about the relation between earning smoothing and earnings quality. There is substantial evidence suggesting earnings smoothing improves earnings quality. Trueman and Titman (1989) propose a model in which high quality firms smooth earnings as a signal of low risk by demonstrating less volatile revenue. Chaney and Lewis (1995) report evidence supportive of this model. Verrecchia (1986) argues that the balance between agency conflict and compensation can benefit shareholders by encouraging managers to produce earnings with less noise on a consistent basis, under appropriate compensation contracts. 2 See Dechow, Ge and Schrand (2010) for a similar discussion. Demski (1998) develops a model that finds that, dependent on the conditions, earnings smoothing by managers can produce either less informative or more informative earnings. Empirically, Jayaraman (2008) finds that spreads and PIN increase when earnings are smoothed either more or less than cash flows, implying both over smoothing and under smoothing harm the informativeness of earnings. 3 Kirschenheiter and Melumad (2002, 2004) contend that earnings smoothed by informed managers have higher inferred precision, allowing investors to extract more information from them. 2

5 Empirically, Subramanyam (1996) finds that managers use discretionary accruals to smooth earnings, and that those smoothed earnings are more economically representative of future earnings. Affleck-Graves, Callahan and Chipalkatti (2002) find that firms with more predictable earnings have lower spreads. Tucker and Zarowin (2006) find that firms with smoother earnings have more earnings information incorporated into returns than those with less smooth earnings. Conversely, there is also considerable theoretical and empirical evidence suggesting that earnings smoothing degrades earnings quality. Several theories have been proposed based on agency conflicts that motivate managers to obscure true earnings in order either to ensure bonus compensation or to retain their jobs ((Lambert (1984), Dye (1988), Stein (1989)). These theories are empirically supported by papers that find that managers with performance compensation incentives and/or weak governance actively smooth earnings (see, e.g., Healy (1985), Skinner (1993), DeFond and Park (1997), Leuz, Nanda and Wysocki (2003)). The result of agencymotivated smoothing (often referred to as earnings manipulation) is usually interpreted as degrading the quality of a firm s earnings and a firm s information environment (see, e.g., Bhattacharya, Daouk and Welker (2003), Leuz, Nanda and Wysocki (2003), Huang, Zhang, Deis and Moffitt (2009)). Thus, the literature on the relation between earnings smoothing and earnings quality is decidedly mixed (Dechow, Ge and Schrand (2010)). We provide additional evidence on this issue in the context of baths, using proxies for information asymmetry as our metrics. Because theory suggests a link between baths and earnings smoothing, baths are a natural setting for investigating the relation between earnings smoothing and quality. If information asymmetry decreases after a bath, the post-bath change in earnings smoothing clears the air, i.e., makes 3

6 earnings more informative. If information asymmetry increases, the change in earnings smoothing muddies the water, i.e., makes earnings less informative. We begin by testing the Kirschenheiter and Melumad (2002) model prediction that a bath facilitates future earnings smoothing. Analyzing 6,666 asset write-downs of amounts greater than 1 percent of total assets from 5,590 firms between 1965 and 2006, we find that firms that take baths experience substantial negative earnings prior to the bath and small positive profits following the bath. For all four generally accepted earnings smoothing measures utilized, firms demonstrate significantly smoother earnings following baths. The volatility of earnings decreases post-bath in relation to the volatility of cash flows. The number of firms experiencing small positive profits increases by percent following baths. Similarly, the number of firms experiencing a small earnings increase rises by percent following baths. Thus, our results confirm the central prediction of Kirschenheiter and Melumad. Directly testing the channels through which baths might facilitate future earnings smoothing, we find that firms write up assets in the years following large baths. Overall, it appears that baths do facilitate future earnings smoothing. We next test changes in the firm-level information environment in the period surrounding baths. We find strong evidence of a positive intertemporal relation between large asset writedowns and information asymmetry. Ten of 12 measures of information asymmetry three years before and three years after baths show statistically and economically significant increased information asymmetry. For instance, average daily share turnover declines by 2 basis points, effective spread increases by 12 basis points, analyst following increases by 0.23, and the 4

7 Amihud price impact measure increases by The increase in information asymmetry following baths is consistent with increased earnings smoothing that degrades earnings quality. 5 Our study contributes to the literature in at least three ways. First, we add to the big bath literature by assessing the long-term ramifications of baths. The big bath literature is predominantly concerned with the causes of baths. This paper examines the consequences of baths. Second, we contribute to the earnings quality literature by empirically linking baths and earnings smoothing. While Kirschenheiter and Melumad (2002) theoretically connect the two actions, to our knowledge no study has empirically identified an intertemporal relation between baths and earnings smoothing. Third, our study contributes to both the earnings quality and information asymmetry literatures by enhancing the understanding of how discretionary managerial behavior impacts a firm s information environment. The remainder of the paper is organized as follows. Section II discusses literature relevant to our study and Section III outlines our hypotheses. Section IV discusses sample selection and empirical methods. Sections V and VI report our results concerning earnings smoothing and information asymmetry around baths. Section VII concludes. II. Relevant Literature Bath has become the general term used to describe a large loss or asset write-down, usually discretionary. Moore (1973), one of the first to study baths, finds that new managers use 4 For economic comparison, the time series mean of the monthly cross sectional average for all observations in our sample is 0.36 percent for daily turnover, 1.91 percent for effective spread, 7.37 for analyst following and 4.56 for the Amihud measure. 5 Several papers have identified a strong relation between earnings quality and information asymmetry. Bhattacharya, Desai and Venkataraman (2007) find poor earnings quality leads to increased information asymmetry. Other work, such as Aboody, Hughes and Liu (2005) and Hughes, Liu and Liu (2007) treat the two as similar, using earnings quality as a proxy of information asymmetry. We posit any change in information asymmetry to be evidence of changes in earnings quality. 5

8 their discretion to minimize current income. 6 He argues that taking a bath alters the benchmark for future comparisons and increases future earnings by removing future losses from future income statements. Analyzing accounting decisions in light of bonus compensation contracts, Healy (1985) proposes a theory of bath accounting. He argues that managers select accounting practices to ensure bonuses are earned (i.e., income before taxes exceeds a set amount). In this setting, if no accounting maneuvers can ensure a bonus is paid in a particular period, managers have an incentive to accelerate write-offs, maximize realized losses, and defer income. This action has two direct effects. First, current net income is deflated. Second, future net income is inflated relative to the income that would have been reported without the writeoffs. Although the study of the causes of baths is important, there are also reasons to study the consequences of baths. Completely writing off an asset (a write-off) whenever the abandonment value exceeds the present value of all subsequent future cash flows is a simple decision for a manager which does not require discretion (Gaumitz and Emery (1980)). However, as discussed by Strong and Meyer (1987), it is more difficult to choose an accurate residual value or correctly identify an incremental decline in an asset s value (a write-down). If the asset is actually disposed of for more than its residual value or future events show that an incremental write-down was too large, the excess is credited back to income. Strong and Meyer (1987), basing their analysis on the incentive for managers to maximize write-downs in order to maximize future 6 The idea that managers use unique events to justify baths is also discussed by Bernstein (1970) in the context of mergers and reserves for future costs and losses. 6

9 income, 7 find evidence of excess write-downs following managerial turnover or poor firm performance. 8 A bath does not appear to be a positive event for shareholders in the short run. Firms experience poor earnings and financial health in the years preceding discretionary write-downs and suffer severe negative market reactions to the write-down announcement (Elliot and Shaw (1988), Zucca and Campbell (1992), Francis, Hanna and Vincent (1996)). Francis, Hanna and Vincent (1996) find that write-downs are inversely related to both large negative and large positive earnings, suggesting they are not used to manage earnings. Rees, Gill and Gore (1996) analyze abnormal accruals around baths and find significantly negative abnormal accruals in the write-down year that do not reverse in subsequent years, evidence that discretionary write-downs are not tied to earnings management. Following Healy (1985), big bath accounting practices have been considered distinct from earnings smoothing accounting (see Dyer (1988), Trueman and Titman (1988), Fudenberg and Tirole (1995)). Kirschenheiter and Melumad (2002) join the two concepts, proposing a model based on the inferred precision of long-term earnings. As investors cannot know either a firm s true long-term earnings or the precision of announced earnings, they must infer the accuracy of earnings over multiple periods. Managers, therefore, have an incentive to underreport positive earnings surprises (smoothing) to save income for future periods and to increase the inferred precision of earnings announcements. However, when faced with large negative earnings surprises, managers have an incentive to introduce additional noise into the 7 We note that while Strong and Meyer (1987) make a case that managers use write-downs to maximize future earnings, that conclusion does not imply that future earnings are either more or less economically meaningful. 8 Several bath studies, such as Wahlen (1994), Collins, Shackelford and Whalen (1995), and Cornett, McNutt and Tehranian (2009), limit their samples to bank holding companies. Although this research is important, because bank balance sheets differ significantly from industrial firms and banks have much more flexibility in writing down financial assets, their findings are not directly relevant to this paper. 7

10 announcement by maximizing losses, that is, by taking a bath. The noise both reduces the inferred precision of the announcement and preserves discretionary income for future smoothing. An important consequence of imperfect disclosure is information asymmetry. Information asymmetry imposes costs on firms and investors. For example, information asymmetry might be a priced risk factor (Easley, Hvidkjaer and O'Hara (2002), Duarte and Young (2009), Hwang and Qian (2010)), can impact a firm s cost of equity (Botosan, Plumlee and Xi (2004), Easley and O'Hara (2004)), and influence investor trading costs (Krinsky and Lee (1996), Huang and Stoll (1997)). These costs motivate the study of accounting decisions that impact a firm s information environment. III. Hypotheses Asset write-downs can facilitate future earnings smoothing in one of two ways (Kirschenheither and Melumad (2002)). First, future losses can be recognized currently. Consequently, reported earnings in future periods are higher than they would have been otherwise. Second, excess losses can be realized currently. In future periods these excess losses can be reversed and income can be artificially inflated. Since smoothing is the preferred steady state of management (Kirschenheither and Melumad (2002)), and managers have discretion over how early losses are recognized or when excess losses are reversed, they can used both these bath tactics to smooth future earnings. These possibilities motivate Hypotheses 1 and 2: Hypothesis 1: Earnings smoothing increases following a bath. Hypothesis 2: Large asset write-downs are reversed in post-bath periods. Given evidence supporting Hypothesis 1, we then examine whether the increase in earnings smoothing following baths increases or decreases the inferred precision of earnings. If future smoothed earnings are more informative (Trueman and Titman (1989), Subramanyam 8

11 (1996)) they make private information public, thereby reducing firm information asymmetry (Easley, Keifer and O Hara (1997)), prompting Hypothesis 3a: Hypothesis 3a: Information asymmetry decreases following a bath. Conversely, if future smoothed earnings are less informative (Dye (1988), Luez, Nanda and Wysocki (2003)), the ratio of private to public information increases and information asymmetry increases, leading to Hypothesis 3b: Hypothesis 3b: Information asymmetry increases following a bath. Asset write-downs can impact future earnings in two ways. First, they create a reserve of losses that, with some discretion, can be reversed in the future (Moore (1973), Healy (1985)). Second, if baths are used merely to accelerate losses, a larger portion of those bath losses are transferred from the near future than from the far future. If baths are used to manage future earnings using either method, then baths should have the strongest impact on a firm s information environment the year following the bath, with the impact weakening over time. If baths are used to facilitate earnings smoothing which degrades earnings quality, we should expect Hypothesis 4a: Hypothesis 4a: Information asymmetry is lowest immediately following a bath, and increases as time passes. If, on the other hand, baths facilitate informative earnings smoothing, then we should expect Hypothesis 4b: Hypothesis 4b: Information asymmetry is highest immediately following a bath, and decreases as time passes. IV. Data and Methods 9

12 IV.A. Sample Selection There is disagreement in the literature about what constitutes a bath. Several studies limit their samples to announced asset write-downs (see, e.g., Strong and Meyer (1987), Zucca and Campbell (1992), Francis, Hanna and Vincent (1996)). This approach tends to limit sample periods to five years or less and can miss multiple write-downs that, in aggregate, constitute a bath. We therefore adopt the measure of Elliott and Shaw (1988) and classify a bath as any fiscal year end 9 observation in Compustat for which Special Items is negative and exceeds 1 percent of lagged firm total assets. 10 For robustness and comparison purposes, we create three other samples, defining baths as 2, 5 and 10 percent of lagged total assets. By construction, there is overlap in these sample sets. For example, all 10 percent baths are also included in the 5 percent bath sample. We run our analyses on all four samples. Using multiple bath definitions provides insight into whether different sized baths have different ramifications for a firm s information environment. It is common for firms to take multiple baths over the course of several decades. In fact, there are two firms which, between 1964 and 2009, had 18 years with baths at the 1 percent level. We limit our sample to clean baths of firms which do not have other baths within surrounding years. We require at least two years pre-bath and two years post-bath to adequately test long-term changes in information asymmetry. Our analysis also requires sufficient pre-bath 9 We choose the annual approach of Elliott and Shaw (1988) for three reasons. First, quarterly identification would result in multiple quarterly asset write-downs during one year being removed from our sample (as we use only baths which occur outside of three years of another firm bath). Second, comprehensive analysis dictates we analyze annual earnings management measures around baths and annual identification makes such analysis cleaner. Third, annual identification is less precise and, therefore, more conservative it biases against finding any change in information asymmetry. In untabulated results we conduct our regression analyses on a sample of quarterly identified baths and find qualitatively similar results. 10 Elliott and Shaw (1988) remove asset write-downs they determine to be non-discretionary. For our sample, we include all special item observations. While only discretionary write-downs are theoretically tied to future earnings management, our goal is to be as conservative as possible in our sample selection. Manual deletion of observations induces subjectivity into sample selection, whereas leaving potentially non-discretionary write-downs in our sample biases against our findings (that is, there is no a priori reason to believe non-discretionary write-downs facilitate future earnings management). 10

13 observations to establish a baseline for information asymmetry. However, too many years between baths might eliminate valid observations and weaken the power of our tests. Given these constraints, we limit our sample to baths that occur at least three years before and three years after another bath at the same firm. 11 After identifying clean baths, we trim possible outliers. Several hundred baths exceed 67 percent of lagged total assets. These observations are either data errors or firms that are most likely in severe distress, and are therefore discarded. We also remove baths of firms with total assets of less than 5 million dollars. 12 Financial firms possess different asset write-down incentives and avenues. We therefore remove all financial firms by excluding four-digit SIC code between 6000 and Finally, to allow for three years of pre- and post-bath data, we include only baths that occurred between 1965 and Table 1 shows the summary statistics for each of our four final samples. The 1 percent sample has 6,666 baths from 5,590 unique firms; the mean (median) bath is 6.92 (3.40) percent of lagged total assets. With our stricter 10 percent sample we find 3,675 baths. Examining the differences between the four samples, we find that firm total assets decrease substantially on average as the bath definition increases in size. However, that difference is driven by the right tail of the distribution, as median total assets are fairly similar. IV.B. Information Asymmetry Measures 11 We have conducted all of our empirical analyses using 2 years and 4 years (pre and post) and have found qualitatively similar results. 12 There are two reasons to remove such firms. First, the accounting variables used to calculate earnings management measures are often missing. Second, these small firms often have extremely high information asymmetry and their inclusion might skew our analysis. 11

14 There is a wealth of information asymmetry measures from which to choose. We use 12 measures which cover a broad spectrum, from market-based to analyst-based and from monthly to daily to intra-daily measurement frequency. We calculate seven market-based measures from the CRSP daily file, measured monthly for 1962 through We define turnover as the monthly mean of daily volume divided by shares outstanding. Daily Q spread is the monthly mean of the daily high/closing ask price minus the low/closing bid price reported in CRSP. HL spread is the spread component of the decomposition of daily spread into volatility and spread in accordance with Corwin and Schultz (2010). Roll is the Roll covariance measure defined as, where is the change in log price from end of day t-1 until the end of day t, set to zero if the covariance is positive (Roll (1984)). Amihud is the monthly mean daily price impact measure proposed by Amihud (2002) and defined as,where is in millions of dollars. Realized volatility is the monthly sum of daily squared returns (French, Schwert and Stambaugh (1987), Andersen, Bollerslev, Diebold and Ebens (2001)). Return standard deviation is the monthly standard deviation of daily returns. We use two analyst-based measures obtained from Institutional Brokers' Estimate System (I/B/E/S) for 1976 through Analyst standard deviation is the standard deviation of all current 1-year forward earnings forecasts. Number of analysts is the number of analysts in any given month for which I/B/E/S records a one-year forward earnings forecast. Finally, we use three transaction-level measures calculated for years 1993 through 2009 from the Trade and Quote database (TAQ). Effective spread is the monthly trade dollar volume weighted mean of the transaction level calculation of ( ) (Goyenko, Holden and Trzcinka (2009)). Quoted spread is the monthly trade dollar volume weighted mean 12

15 of the transaction level calculation of. 13 We calculate PIN (probability of informed trading) quarterly from daily buy and sell orders (classified as such by the Lee and Ready (1991) algorithm) in accordance with Easley, Kiefer and O'Hara (1997). To mitigate the effect of extreme tails, we Winsorize all measures at the 1st and 99th percentiles. IV.C. Earnings Smoothing Measures Much like information asymmetry, there is no single, universally accepted metric for earnings smoothing. Therefore, we employ four widely accepted measures of earnings smoothing. Smooth(std) is defined as the standard deviation of earnings divided by the standard deviation of cash flows 14 (Dechow, Ge and Schrand (2010)), each over three years before and three years after a bath. Smooth(var) is defined as the variance of earnings minus the variance of cash flows (Jayaraman (2008)), each over three years before and three years after a bath. Small positive is defined as an indicator variable equal to 1 if a firm has income before extraordinary items between 0 and 5 percent of lagged total assets and 0 otherwise. Small increase is defined as an indicator variable equal to 1 if a firm has an increase in income before extraordinary items between 0 and 1.3 percent of lagged total assets and 0 otherwise (Burgstahler and Dichev (1997), DeGeorge, Patel and Zeckhauser (1999)). 15 V. Earnings Smoothing around Baths 13 In accordance with Chordia, Roll and Subrahmanyam (2000), we filter transactions by throwing out trades with a bid-ask spread greater than 5 dollars, and effective-to-quoted spread ratio greater than 4, or a quoted proportional spread greater than 40 percent. 14 Cash flow from operations is measured directly in COMPUSTAT starting in For observations prior to 1987, we calculated cash flows in accordance with Bowen, Burgsthler and Daley (1987). We also conducted our analyses using only post 1986 data and find similar results. 15 Each of the smoothing measures are Winsorized at the 1 st and 99 th percentiles. 13

16 The purpose of this section is to test two earnings smoothing implications of the Kirschenheiter and Melumad model, earlier identified as Hypotheses 1 and 2.If baths are used by managers to facilitate future earnings smoothing, then we expect traditional earnings smoothing measures to increase following baths. Table 2 reports mean accounting variables and earnings smoothing measures around baths. Our analysis begins with net income in Panel A. As seen in all four bath samples, firms experience considerable negative earnings in the years leading up to baths, consistent with Elliott and Shaw (1988) and Zucca and Campbell (1992). Following baths, firms experience small positive (or negative) earnings. Using a Wilcoxon test, there is a significant increase in earnings following a bath for all but the 10 percent bath sample. It is possible for such a change to be driven by something other than earnings smoothing. If firms are disposing of underperforming assets and restructuring, then such a recovery would not be unexpected. It is also possible that firms are simply using poor performing years to cover house cleaning with no intention to alter future earnings. So we turn to our earnings smoothing measures to examine whether this post-bath recovery is real or possibly manufactured. Panel B of Table 2 shows that all four of our earnings smoothing measures show increased smoothing in the three years following a bath. In all four of our bath samples, Smooth(std) and Smooth(var) decrease by more than 50%, a statistically and economically significant increase in earnings smoothing. For instance, considering Smooth(std) in our 1 percent sample, the standard deviation of earnings goes from being 7.63 times the standard deviation of cash flows pre-bath to only 3.06 times post-bath. Examining our small profit measure, the proportion of firms experiencing small earnings grows in all four samples. For instance, small profit in our 2 percent bath sample rises from percent 3 years prior to the 14

17 bath to percent 3 years after the bath. Examining small earnings increases, and throwing out the year following a bath, we observe a similar pattern. 16 Combining these results with our results for net income in Panel A, it appears that firms are not just reporting increased earnings, but also smoother earnings. These results are consistent with the theoretical model proposed by Kirschenheiter and Melumad (2002), in which baths facilitate future earnings smoothing, and provide evidence in support of Hypothesis 1. They also give us insight into the channel through which baths might impact the information environment of a firm. While it is possible that the bath itself has some impact on earnings quality at the time of the bath, we are concerned with the intertemporal relation between baths and information. That relation must, to some extent, flow through earnings smoothing. The second variable in Panel A of Table 2 sheds light on Hypothesis 2, that baths facilitate earnings smoothing by reversing write-downs in future years. We find mixed evidence for this hypothesis. For the 1 and 2 percent samples, special items scaled by lagged total assets actually decreases following baths. In our 5 and 10 percent bath samples the opposite is true. Firms appear to reverse, to some extent, the write-downs taken in the bath. This increase in special items inflates earnings, facilitating earnings management. These findings provide evidence for both channels of baths influencing earnings quality. Given that there is a relation between baths and future earnings smoothing, it appears that baths facilitate earnings management in different ways based on the size of the bath. The decrease in special items for the smaller bath samples suggests that smaller baths aid in earnings smoothing by accelerating loss recognition. However, the increase in special items for the larger bath 16 Because earnings are exceptionally low the year of the bath, the number of firms with small increases in earnings the following year is mechanically very low. 15

18 samples suggests that larger baths aid in earnings smoothing by creating a reserve of losses that are subsequently reversed. VI. Information Asymmetry around Baths Earnings become more smooth after baths, but that finding in isolation does not indicate whether baths increase or decrease earnings quality (Dechow and Skinner (2000), Dechow, Ge and Schrand (2010)). As noted earlier, smoothed earnings can produce either increased precision (Kirschenheiter and Melamud (2004)) or, if excessively smoothed, decreased information precision (Dye (1988)). We therefore test for changes in firms information environments around baths to determine whether the increased smoothing following baths improves or degrades earnings quality. All of our information asymmetry proxies are highly time dependent in two ways. First, there is a strong time trend in information asymmetry. Spreads, price impact and volatility all decline over our sample period. Therefore, a simple pre- and post-bath univariate comparison will, by construction, reveal information asymmetry declines on average following baths. Second, commonality in many of these measures (e.g., spreads and volatility) leads to systematic shocks that might mask the true impact of baths on firms information environments. For instance, all baths that occur in the year preceding October 1987 will appear to be related to future information asymmetry increases. This effect would wash out across our entire sample if baths are independent of systematic information shocks, but that independence is not an assumption we think is justified. The time dependence suggests the use of a fixed effects model. Our first analysis, reported in Table 3, runs a panel regression of each information asymmetry measure on a bath 16

19 indicator (0 for months prior to the bath, 1 for months following the bath) and indicator variables for each month in our sample. 17 Because we calculate our information measures monthly (except PIN 18 ), we have a maximum of 71 observations per bath, 35 months prior to the bath and 36 months following the bath. 19 The month indicator variables absorb time trends and systematic information shocks, allowing for a cleaner analysis of the impact of baths on information asymmetry. Panel A of Table 3 reports the coefficients of bath indicators using pooled ordinary least squares (OLS) regressions with fixed effects. The far left column reports the expected sign of the bath indicator coefficients if they are consistent with Hypothesis 3b, that information asymmetry increases following baths. We find that 10 of the 12 information asymmetry proxies show support for Hypothesis 3b, many with economic significance. In the 1 percent bath sample, daily turnover decreases, all four of our spread measures increase, both the Roll measure and the Amihud price impact measure increase, as does realized volatility, return standard deviation and analyst following. Although some measures show small changes, such as the 0.02 percent decrease in turnover (compared to the 0.36 percent sample mean), others demonstrate an economically meaningful deterioration of the firms information environments. The Amihud measure increases by 1.90, or a percent increase when compared to the sample mean value 17 We have also run the regressions with a variety of control variables such as market capitalization, book-to-market ratio, share price, leverage, asset tangibility, and research and development expenditures. We find qualitatively similar results. Although these controls are often included in adverse selection and information asymmetry studies, they also present a considerable endogeneity problem, as baths mechanically impact all of them. We therefore exclude them from this paper. 18 Calculated quarterly, there are 11 observations pre-bath, and 12 post-bath, with the quarter the bath occurs excluded. 19 The month of the bath is not included in these regressions. For robustness we have also conducted our analysis on a sample which excludes the month of the bath and the three following months (to account for the uncertainty surrounding release of the firm's annual report) and find qualitatively similar results. We do not require firms to have complete data coverage for the full six years surrounding the bath, so actual observations per bath may be less than

20 of Realized volatility increases by 0.51 percent, or percent of the sample mean value of 3.26 percent. Turning our attention to the three other bath samples, we make two observations. First, the previous results are robust to differences in the bath size threshold. Each of the 10 measures displaying significant change around baths retains its sign and significance as bath size increases. Second, the magnitude of the change around baths increases as our bath size threshold increases for most of our information asymmetry measures. In the 10 percent sample, turnover declines by 0.06 percent (13.04 percent of the sample mean), Amihud increases by 3.16 (54.86 percent), effective spread by 0.23 percent (10.50 percent), and realized volatility by 0.87 percent (19.21 percent). Taken as a whole, we interpret these results as strong evidence in support of Hypothesis 3b, that information asymmetry increases following baths. If baths influence a firm s information environment, we should observe mean reversion in information asymmetry in the periods following baths. This reasoning is behind Hypotheses 4a and 4b. The change in information asymmetry should be strongest in the year following the bath and revert in subsequent years. To test these hypotheses, we conduct pooled OLS regressions identical to those reported in Table 3, except that we include indicator variables for year t+1 and year t In conjunction with the bath indicator, these three indicator variables provide a comprehensive picture of both the level effect of the bath and the trend in information asymmetry following the bath. 21 Panel A of Table 4 reports the coefficients for all three indicators for regressions using the 1 percent bath sample. To illustrate our interpretation of these coefficients, we examine one 20 An indicator variable for year t+2 is left out to avoid perfect collinearity, so its effect is folded into the level effect of the bath indicator variable. 21 The level effect, measured as the coefficient from the post-bath indicator variable, shows the impact of the bath on information asymmetry common to all three post-bath years. The indicator variables interacted with the post-bath indicator (year t+1 and t+3) show the deviation of the effect from this common level during years t+1 and t+3. 18

21 of the stronger effects from Table 3, the Amihud price impact measure. The bath indicator shows a significant level effect of The t+1 indicator has a coefficient of 1.05, showing that the Amihud measure is 1.05 higher than the level effect the year following a bath (2.70 higher than pre-bath). The t+3 indicator has a coefficient of -0.61, showing that the Amihud measure is 0.61 lower than the level effect (though still 1.04 higher than pre-bath). This result is as predicted by Hypothesis 4b. In total, 10 of the 12 information asymmetry measures provide support for Hypothesis 4b. All four spread measures, Roll, realized volatility and return standard deviation display the same pattern as the Amihud measure: positive level effect, positive t+1 coefficient, negative t+3 coefficient. Analyst standard deviation and PIN, both insignificant in Table 3, also present some noteworthy results. Although analyst standard deviation shows a significant negative level effect of percent, its t+1 coefficient is 1.47 percent. Similarly, while PIN has an insignificant level effect, it does have a t+1 coefficient of The level effects of both measures do not support Hypothesis 3b; there appears to be no persistent impact in the post-bath period. However, the significant increase in information asymmetry in the first year following a bath lends credence to Hypothesis 4b. Turning our attention to the other three bath samples in Panels B, C and D, we find similar results. A handful of coefficients do not line up with Hypothesis 4b, but the large majority of our measures support Hypothesis 4b across all four bath samples, strongly suggesting that baths have a large immediate positive effect on information asymmetry which decays over time. Overall, we interpret the results as supportive of both Hypotheses 3b and 4b. Baths appear to facilitate future earnings smoothing, and this smoothing degrades the quality of earnings. 19

22 VII. Conclusions A firm s information environment is of vital importance to managers, shareholders, regulators, and prospective investors. However, although managers have some incentive to publicly disclose material information, they might also have personal incentives to obscure such information releases. We study baths (asset write-downs), an important discretionary component of disclosures, and how they impact a firm s information environment. Consistent with the accounting theory of Kirschenheiter and Melumad (2002), we find increased earnings smoothing following baths. The smoothing is both directly facilitated by baths (i.e., write-downs are reversed in future periods) and the result of broader corporate change, as abnormal accruals also increase following large baths. To our knowledge, ours is the first paper to empirically link baths and future earnings smoothing. Analyzing whether this increased smoothing improves or degrades earnings quality, we find a statistically and economically significant increase in information asymmetry following large asset write-downs. The decrease in transparency is apparent in analyst coverage, marketbased measures such as spread, share turnover and volatility, and transaction-based spread measures. Also consistent with accounting theory, we find that the increase in information asymmetry is strongest in the year following a bath and decays over the following two years. This finding suggests the changes in firm information asymmetry are directly caused by baths, and not structural change in managerial or accounting behavior. In summary, we find a strong negative relation between asset write-downs and a firm s information environment. This finding means that baths are an accounting event with real effects. By increasing information asymmetry, baths have a material impact on firm cost of capital (Easley and O'Hara (2004)), equity trading costs (Huang and Stoll (1997)), and 20

23 uninformed investor returns (Easley, Hvidkjaer and O'Hara (2002)). Our evidence is strongly supportive of the notion that big baths muddy the waters instead of clearing the air. 21

24 References Aboody, David, John Huges and Jing Liu, 2005, Earnings quality, insider trading and cost of capital, Journal of Accounting Review 43, Amihud, Yakov, 2002, Illiquidity and stock returns: Cross-section and time-series effects, Journal of Financial Markets 5, Affleck-Graves, John, Carolyn M. Callahan and Niranjan Chipalkatti, 2002, Earnings predictability, information asymmetry, and market liquidity, Journal of Accounting Research 40, Andersen, Torben, Tim Bollerslev, Francis X. Diebold and Heiko Ebens, 2001, The distribution of realized volatility, Journal of Financial Economics 61, Beidleman, Carl R.,1973, Income smoothing: The role of management, The Accounting Review 48, Bernstein, Leopold A., 1970, Reserves for future costs and losses: Threat to the integrity of the income statement, Financial Analyst Journal 26, Bhattacharya, Neil, Hemang Desai and Kumar Venkataraman, 2007, Earnings quality and information asymmetry: Evidence from trading costs, Working, Southern Methodist University. Botosan, Christine A., Marlene A. Plumlee and Yuan Xi, 2004, The role of information precision in determining the cost of equity capital, Review of Accounting Studies 9, Burgstahler, David, and Ilia Dichev, 1997, Earnings management to avoid earnings decreases and losses, Journal of Accounting and Economics 24, Chaney, Paul K., and Craig M. Lewis, 1995, Earnings management and firm valuation under asymmetric information, Journal of Corporate Finance 1, Chordia, Tarun, Richard Roll and Avanidhar Subrahmanyam, 2000, Commonality in liquidity, Journal of Financial Economics 56, Collins, Julie H., Douglas A. Shackelford and James M. Whalen, 1995, Bank differences in the coordination of regulatory capital, earnings, and taxes, Journal of Accounting Research 33, Cornett, Marcia Millon, Jamie John McNutt and Hassan Tehranian, 2009, Corporate governance and earnings management at large U.S. bank holding companies, Journal of Corporate Finance 15,

25 Corwin, Shane A., and Paul Schultz, 2010, A simple way to estimate bid-ask spreads from daily high and low prices, working, Mendoza College of Business, University of Notre Dame. Dechow, Patricia, Weili Ge and Catherine Schrand, 2010, Understanding earnings quality: A review of the proxies, their determinants and their consequences, Journal of Accounting and Economics 50, Dechow, Patricia and Douglas J. Skinner, 2000, Earnings management: Reconciling the views of accounting academics, practitioners, and regulators, Accounting Horizons 14, Degeorge, Francois, Jayendu Patel and Richard Zeckhouser, 1999, Earnings management to exceed thresholds, Journal of Business72, Demski, Joel S., 1998, Performance measure manipulations, Contemporary Accounting Research 15, Duarte, Jefferson, and Lance Young, 2009, Why is PIN priced?, Journal of Financial Economics 91, Dye, Ronald A., 1988, Earnings management in an overlapping generations model, Journal of Accounting Research 26, Easley, David, Soeren Hvidkjaer and Maureen O'Hara, 2002, Is information risk a determinant of asset returns, Journal of Finance 57, Easley, David, Nicholas M. Kiefer and Maureen O'Hara, 1997, One day in the life of a very common stock, Review of Financial Studies Easley, David, and Maureen O'Hara, 2004, Information and the cost of capital, Journal of Finance 59, Elliott, John A., and Wayne H. Shaw, 1988, Write-offs as accounting procedures to manage perceptions, Journal of Accounting Research 26, Francis, Jennifer, J. Douglas Hanna and Linda Vincent, 1996, Causes and effects of discretionary asset write-offs, Journal of Accounting Research 34, French, Kenneth R., G. William Schwert and Robert F. Stambaugh, 1987, Expected stock returns and volatility, Journal of Financial Economics 19, Fudenberg, Drew, and Jean Tirole, 1995, A theory of income and dividend smoothing based on incumbency rents, Journal of Political Economy 103, Gaumitz, Jack E., and Douglas R. Emery, 1981, Asset growth, abandonment value and the replacement decision of like-for-like capital assets, Journal of Financial and Quantitative Analysis 15,

26 Goyenko, Ruslan Y., Craig W. Holden and Charles A. Trzcinka, 2009, Do liquidity measures measure liquidity?, Journal of Financial Economics 92, Guidry, Flora, Andrew J. Leone and Steve Rock, 1999, Earnings-based bonus plans and earnings management by business-unit managers, Journal of Accounting and Economics 26, Healy, Paul M., 1985, The effect of bonus schemes on accounting decisions, Journal of Accounting and Economics 7, Healy, Paul M., and Krishna G. Palepu, 2001, Information asymmetry, corporate disclosure, and the capital markets: A review of the empirical disclosure literature, Journal of Accounting and Economics 31, Huang, Roger D., and Hans R. Stoll, 1997, The components of the bid-ask spread: A general approach, Review of Financial Studies 10, Huang, Pinghsun, Yan Zhang, Donald R. Deis and Jacquelyn S. Moffitt, 2009, Do artificial income smoothing and real income smoothing contribute to firm value equivalently?, Journal of Banking and Finance 33, Hughes, John S., Jing Liu and Jun Liu, 2007, Information asymmetry, diversification, and cost of capital, The Accounting Review 82, Hwang, Chuan-Yang, and Xiaolin Qian, 2010, Is information risk priced? Evidence from the price discovery of large trades, working, Nanyang Business School. Jayaraman, Sudarshan, 2008, Earnings volatility, cash flow volatility and informed trading, Journal of Accounting Research 46, Jones, Jennifer J., 1991, Earnings management during import relief investigations, Journal of Accounting Research 29, Kirschenheiter, Michael, and Hahum D. Melumad, 2002, Can big bath and earnings smoothing co-exist as equilibrium financial reporting strategies?, Journal of Accounting Research 40, Kirschenheiter, Michael, and Hahum D. Melumad, 2004, Earnings quality and smoothing, Working, Krannert School of Management, Purdue University. Kothari, S.P., Andrew J. Leone and Charles E. Wasley, 1995, Performance matched discretionary accrual measures, Journal of Accounting and Economics 39, Krinsky, Itzhak, and Jason Lee, 1996, Earnings announcements and the components fo the bidask spread, Journal of Finance 51,

27 Krishnaswami, Sudha, and Venkat Subramaniam, 1999, Information asymmetry, valuation, and the corporate spin-off decision, Journal of Financial Economics 53, LaFond, Ryan and Ross L. Watts, 2008, The information role of conservatism, The Accoutning Review 83, Lambert, Richard A., 1984, Income smoothing as rational equilibrium behavior, The Accounting Review 59, Lee, Charles M.C., and Mark J. Ready, 1991, Inferring trade direction from intraday data, Journal of Finance 46, Moore, Michael L., 1973, Management changes and discretionary accounting decisions, Journal of Accounting Research 11, Rees, Lynn, Susan Gill and Richard Gore, 1996, An investigation of asset write-downs and concurrent abnormal accruals, Journal of Accounting Research 34, Richardson, Vernon J., 2000, Information asymmetry and earnings management: Some evidence, Review of Quantitative Finance and Accounting 15, Roll, Richard, 1984, A simple implicit measure of the effective bid-ask spread in an efficient market, Journal of Finance 39, Skinner, D., 1993, The investment opportunity set and accounting procedure choice: Preliminary evidence, Journal of Accounting and Economics 16, Stein, Jeremy C., 1989, Efficient capital markets, inefficient firms: A model of myopic corporate behavior, Quarterly Journal of Economics 104, Strong, John S., and John R. Meyer, 1987, Asset writedowns: Managerial incentives and security returns, Journal of Finance 42, Subramanyam, K. R., 1996, The pricing of discretionary accruals, Journal of Accounting and Economics 22, Trueman, Brett, and Sheridan Titman, 1988, An explanation for accounting income smoothing, Journal of Accounting Research 26, Tucker, Jennifer W., and Paul A. Zarowin, 2006, Does income smoothing improve earnings informativeness?, The Accounting Review 81, Verrechia, Robert E., 1986, Managerial discretion in the choice among financial reporting alternatives, Journal of Accounting and Economics 8,