Have Earnings Announcements Lost Information Content? Manuscript Steve Buchheit

Transcription

1 Have Earnings Announcements Lost Information Content? Manuscript Steve Buchheit University of Houston College of Business Administration Department of Accountancy and Taxation Houston TX, (713) Mark Kohlbeck * University of Wisconsin Madison Department of Accounting & Information Systems 975 University Avenue Madison, WI mkohlbeck@bus.wisc.edu (608) May 2001 * Corresponding Author The authors would like to thank the editor, Bala Balachandran, two anonymous reviewers, Roland Atiase, Robert Freeman, Scott Jackson, Adam Koch, Sandeep Nabar, Tom Noland, Per Olsson, Austin Reitenga, Terry Warfield, Simon Yang and seminar participants at the 2000 Southwest Regional Meeting of the American Accounting Association for comments on previous versions of this paper.

2 Have Earnings Announcements Lost Information Content? Abstract We investigate the potential historic decline in the information content of earnings announcements. Like Beaver (1968) we focus on the incremental information content of accounting earnings announcements - whether the announcements convey "new news" to the market as evidenced by unusual price movements. We investigate the price reaction to earnings announcements of a broad range of firms over a 23-year time period (1975 through 1997). We find that, on average, the price reaction to earnings announcements has increased over time. However, the robustness of this finding is called into question based on two factors. First, we find that the change in price reaction to earnings varies depending on the size of firms analyzed. Specifically, we find some evidence that small firms exhibit a decreasing price reaction to earnings announcements over time while larger firms consistently exhibit an increasing price reaction to earnings announcements. Second, we find that a minority of firms drives the observed increase in price reaction indicating that a cross-sectional mean reaction does not imply such a reaction is typical for individual earnings announcements.

3 Have Earnings Announcements Lost Information Content? 1.0 Introduction Whether or not financial reporting "matters" to investors and other market participants is an important issue to accountants. Ball and Brown (1968) and Beaver (1968) are seminal papers demonstrating that accounting matters. While Ball and Brown (1968) demonstrate that accounting information is consistent with information used to price securities (thus providing evidence that accounting information is value relevant), Beaver (1968) demonstrates significant stock price and trading volume reactions to accounting earnings information (thus providing evidence that accounting earnings have information content). During the past three decades, researchers have continued to document that accounting matters to market participants; however, there are recent concerns that accounting information has lost a significant portion of usefulness to investors (cf. American Institute of Certified Public Accountants, AICPA, 1994). Responding to such concerns, recent research has documented a historic decline in the value relevance of accounting earnings information (Brown, Lo, and Lys 2000; Lev and Zarowin 1999; Francis and Shipper 1999; Collins, Maydew, and Weiss 1997), 1 but an increase in the information content of earnings information (Landsman and Maydew 1 These studies provide somewhat contradictory results when balance sheet data is considered. Both Francis and Shipper (1999) and Collins, Maydew, and Wiess (1997) document an overall increase in financial statement value relevance because the increased value relevance of balance sheet information is greater than the decreased value relevance of earnings information. Brown, Lo, and Lys (2000) provide evidence that this overall increase in value relevance is an artifact of a scaling factor that, when corrected, reveals no increase in the value relevance of balance sheet information over time. In contrast, Lev and Zarowin (1999) speculate that the non-decreasing value relevance observed in the Collins et al. study is due to the examination period. By examining a more recent sample period, Lev and Zarowin document decreasing value relevance of accounting information.

4 2000; Kross and Kim 1999). Although the preceding evidence is consistent with analytic possibilities (Kim and Verrecchia 1991), decreasing value relevance coupled with increasing information content sends a mixed signal about the over-time usefulness of accounting earnings information. Given this mixed evidence, the implications of altering the current reporting model in an effort to make accounting earnings more useful are not clear. The purpose of this paper is to investigate the consistency and pervasiveness of the historical change in stock price reactions to accounting earnings announcements. 2 Like Landsman and Maydew (1999) and Kross and Kim (1999), we find that, on average, there has been a historical increase in price reactions to earnings announcements (PREA); however, we document that the change is not consistent for all firms. Specifically, we find that historical changes in PREA are positively associated with firm size (market value). For firms with relatively small market values, some measures of PREA actually decrease over time. We also investigate potential changes in the pervasiveness of "large" PREA over time. As documented by Bamber et al. (2000, 105), investigating only cross-sectional mean market reactions obscures the fact that most individual earnings announcements are not associated with unusual price reactions. Similar to Bamber et al., we employ a nonparametric rank-order test to define "large" PREA and examine the pervasiveness of historical changes in large PREA. Like our parametric tests of cross-sectional mean PREA, we find that the pervasiveness of large PREAs is positively associated with firm 2 Price reaction to earnings announcements is one form of information content that reflects changes in the expectations of the market as a whole whereas volume reactions reflect changes in the expectations of individual investors (Beaver 1968, 69). 2

5 size. For relatively small firms, the proportion of large PREAs decreases over time; however, the opposite is true for large firms. The remainder of our paper is organized as follows. The next section presents our motivation for investigating the consistency and pervasiveness of the historical changes in stock price reactions to earnings announcements. Empirical methodology and sample selection are described in Section 3. Section 4 discusses our results and concluding remarks follow. 2.0 Motivation Two aspects of our analysis distinguish this paper from current investigations of historical information content change. First, we investigate a broader sample of firms relative to Landsman and Maydew (1999) and Kross and Kim (1999). While Landsman and Maydew sample the 1,000 largest firms during each year of their study and Kross and Kim examine a constant sample of 385 firms, we examine the entire universe of Compustat firms that pass our selection criteria (described in the next section). 3 On average, we test over 2,100 different firms per year (approximately 1,000 firms meet our testing criteria in 1975, increasing to over 4,000 firms in 1997). Therefore, our sample includes relatively small firms that are excluded from other current studies. Beaver (1968) examines the information content of earnings announcements and finds that both trading volume and return volatility increase at the time of earnings announcements. However, his sample design limits his findings of information content 3 Kross and Kim limit their sample selection given their research objective of investigating earnings precision while Landsman and Maydew limit their sample to larger firms for practical data gathering reasons. 3

6 of earnings announcements to small firms. Atiase (1985) extends Beaver (1968) by providing evidence that during the 1971 and 1972 time period, cross-sectional differences exist between the PREA of relatively small versus relatively large firms. Specifically, Atiase posits that differences in the amount of predisclosure information lead to PREA being inversely related to firm size. He documents a significant price reaction to earnings announcements for small firms, but an insignificant PREA for large firms. As such, we believe that any conclusions about the historical change in PREA should explicitly consider firm size because empirical results are likely dependent upon the size of firms investigated. We also believe that examining the historical change in small firms' PREA is important because, all else being equal, small firms should have the most easily detectable decrease in PREA. To illustrate, investigating the historical change in large firms' PREA potentially fails to detect degradation in information content because large firms are historically close to the "floor" of no observable abnormal price volatility. If across-time variability in PREA exists (i.e. sometimes significant price reactions to earnings occur and sometimes they do not), investigations of large firms will easily detect upward movements in PREA, but will not easily detect across-time degradation in PREA. In contrast, small firms have a well-documented historical price reaction to earnings. As such, if PREA has declined (increased) over time, degradation (improvement) should be relatively easy to detect for small firms. 4 As such, we argue that examining small firms is intuitively more appropriate than examining large firms given a research objective of 4 For example, improved information technology such as Internet disclosure (Ashbaugh, Johnstone, and Warfield, 1999) may preempt earnings announcements. 4

7 investigating the potential for decreasing information content of earnings over time. We also distinguish our investigation from current research based on our historical investigation of the pervasiveness of "large" PREA. Bamber et al. (2000) documents that analyzing the cross-sectional mean information content effect obscures the fact that most individual earnings announcements are not associated with unusual price reactions. Although early information content studies examine non-parametric measures such as the percentage of firms exhibiting "large" PREA (Beaver 1968 and Patell 1976), current studies focus on the cross-sectional mean price reaction to earnings (Landsman and Maydew 1999, Kross and Kim 1999). When considering a change to the accounting model, knowing whether or not relatively few firms drive historical increases in PREA is important. Bamber et al. argue that accounting academics generally misinterpret Beaver's PREA result as being a universal result, but the actual proportion of firms exhibiting large PREA is relatively small. We therefore analyze the over-time change in large PREA where "large" is defined by a rank-order test of abnormal price volatility during the earnings announcement window relative to abnormal price volatility during surrounding event windows. If PREA is "large" then abnormal price volatility should be greater during the earnings announcement window relative to other event windows (i.e. when ranked by the size of abnormal price volatility, PREA should have a high rank if they are truly "large"). 5

8 3.0 Methodology and Sample Selection 3.1 Price Reaction Metrics Evidence of price reactions to earnings announcements has been investigated over the past thirty years using variants of Beaver's (1968) "U ratio." The U ratio compares a firm's unexpected returns during an event window (numerator) to the estimated variance of the firm's unexpected returns over comparable event windows (denominator). A U ratio during the earnings announcement window that is significantly greater than the U ratios in surrounding event windows suggests that earnings announcements contain information used to price securities. In our analysis of firm size and information content of earnings announcements, we use Atiase's (1985) variant of Beaver s U ratio to investigate historical changes in PREA. 5 The major innovation of Atiase's price reaction metric is to create an endogenous control for predisclosure information because the amount of price-relevant predisclosure information, such as news events, varies depending on firm size. For example, small firms likely receive less news and analyst coverage relative to large firms. As such, the earnings announcements of small firms are assumed to be relatively more informative because earnings information has not been preempted by other sources of information. Atiase's price reaction metric is identical to Beaver's price reaction metric in that they both (1) estimate market model parameters, (2) compute unexpected returns during 5 Minor modifications exist between our study and Atiase's (1985) methodology. For example, we estimate our market model using daily returns and analyze 3-day event windows surrounding accounting earnings announcements (as opposed to Atiase's weekly estimation and analysis intervals). Data availability improvements allow such modifications (i.e. we use the COMPUSTAT and CRSP databases whereas Atiase was forced to hand-collect data) that should improve the precision of market reaction tests. 6

9 an event window of interest, and (3) compare earnings announcement windows to comparable non-earnings windows. However, Atiase's "revaluation index" (the price reaction metric used in this study) extends Beaver's price reaction metric by (1) incorporating the prediction error term used by Patell (1976) and (2) explicitly measuring firm-specific predisclosure information as discussed above. 6 We calculate Atiase's price reaction metric for firms in our sample using the four steps presented in Figure 1 (see Atiase 1985 for further detail). The price reaction metric is presented below. RI where : AVG U it* RI = revaluation index U = it U * it = AVG U = U ratio based on Patell (1976) 1 P U Pi Pi 2 for t * report period P = number of days in the predisclosure period i i P i t=1 it it for t predisclosure period (1) Like Beaver s U ratio, a revaluation index between zero and one implies smaller than normal price volatility while an index greater than one implies larger than normal price volatility (i.e. a significant price reaction to earnings information). In order to assess the pervasiveness of price reactions to earnings announcements, we also consider the percentage of "large" PREA as an alternative dependent variable. Similar to Bamber et al. (2000), "large" is defined by a non-parametric rank-ordering test. Specifically, five revaluation indices are calculated for each firm for each quarterly 6 Atiase (1985) presents both a "revaluation index" and a "standardized revaluation index." The former is identical to the U statistic in Patell s (1976) extension of the Beaver price reaction metric and controls for variance due to prediction outside the estimation period. The latter adds a control for predisclosure information. Because (1) we are interested in PREA based on firm size and (2) predisclosure information has been shown to vary with firm size, the index used in this study is what Atiase labels "standardized revaluation index." Similar results are obtained when using Atiase's "revaluation index." 7

10 earnings announcement (i.e., five three-day windows centered on the earnings announcement). When the five revaluation indices are ranked from largest to smallest, the revaluation index during the earnings announcement window should be ranked largest more often than revaluation indices in surrounding windows if the PREA is truly "large." 3.2 Sample Our sample is selected from all U.S. companies with quarterly earnings announcements on the Compustat research file from 1975 to To be included in the study, each firm must have sufficient daily return data (up to three years preceding the earnings announcement and at least 150 return dates) available in the Center for Research in Security Prices (CRSP) database. Further sample reductions occur for observations with incomplete Compustat data; earnings release dates more the 90 days past the end of the applicable quarter; or earnings release dates within five days of the Securities and Exchange Commission s (SEC) due dates for annual and quarterly reports. The first two criteria were selected for practical, data gathering purposes and to ensure that sufficient CRSP and Compustat data are available to calculate market model parameters. The third criterion avoids including "bad news" delinquent earnings announcements. The fourth criterion avoids confounding the earnings announcement release with the release of other financial statement information in connection with the filing of annual and quarterly reports with the SEC. 7 In total, the sample consists of 196,811 firm-year observations over the 23-year period (see Table 1). In order to 7 In order to verify that the SEC due date is a reasonable approximation of the 10-K or 10-Q release date, we randomly selected 100 firms from our sample and compared the 10-K or 10-Q release dates to SEC due dates. We find that SEC reports are consistently filed on or about the due date. 8

11 investigate across-time differences in PREA attributable to firm size, for each quarter, we partition the full sample into five size-based sub-samples. 8 Panels A and B of Table 2 present descriptive statistics for the full sample and the size-based sub-samples of firms, respectively. 4.0 Empirical Analysis Figure 2 presents aggregate results for all firms in our sample irrespective of time (i.e., over-time changes are not considered). Panel A illustrates an aggregate PREA based on the revaluation index (RI). Overall, the RI metrics obtained from the earnings announcement period (window zero) are significantly greater than one at the 0.05 level and significantly greater than the corresponding measures in either of the two preceding or two following report period windows (true for the full sample and for each size-based portfolio). 9 Consistent with Atiase's (1985) size effect, the smallest quintile of firms exhibits larger RI metrics relative to larger firms. Panel B illustrates the percentage of times that the earnings announcement window contains the largest revaluation index relative to surrounding windows (RANK%). Similar to Bamber et al. (2000), we find relatively large PREA when using the rank order test. We find that the earnings announcement window contains the 8 Partitioning by size quintiles is consistent with Carter and Soo (1999). Note that our size-based grouping (each group representing 20% of the market) approximates the 27% size-based grouping that optimizes testing power (Lys and Sabino 1992). 9 The sample data are leptokurtic as market model disturbances for many firms deviate from normality with respect to skewness and kurtosis. Marais (1984) demonstrated that when these conditions exist, the asymptotic distributions of the Patell (1976) U statistic, the basis for the Atiase (1985) indices used in this paper are affected. Although the expected value of the U statistic is still one, the expected variance equals two plus the kurtosis of the disturbance distribution. Accordingly, we adjusted our variance in determining significance levels for the discussion of the data presented in Figure 2. 9

12 "largest" abnormal return approximately 30% of the time, which is significantly greater than the null hypothesis (i.e. no price-reaction to earnings) of 20%. The following subsections present our empirical tests and additional analysis that investigate over-time changes to these price reaction metrics. 4.1 Empirical Tests We first explore the change in price reaction to earnings announcements over time by separately considering the RI and the RANK% as dependent variable measurements of PREA. We investigate changes in PREA over time by regressing each dependent variable on TIME for the full sample and for each of the five size quintiles (see Panels A and B of Table 3, respectively). As shown in Table 3, Panel A, we find significant intercepts for RI (1.851; greater than 1.0 at p <.001), and RANK% (27.5%; greater than the expected 20% at p <.001) suggesting that, on average, investors receive price-relevant information from earnings announcements. Consistent with Landsman and Maydew (1999) and Kross and Kim (1999), we find that the over-time change in PREA is significantly increasing for both PREA metrics suggesting that accounting earnings are increasingly changing the expectations of the market as a whole over time (evidenced by the positive and significant coefficients on TIME). 10 As shown in Table 3, Panel B, our results are cross-sectionally dependent on firm size. We replicate Atiase's size effect (the significant inverse relationship between PREA and firm size is reflected via the monotonically decreasing intercept as firm size 10 In a separate analysis, we document that the results do not vary significantly by announcement quarter. 10

13 increases). We also find that the over-time change in PREA is dependent on firm size. Specifically, the largest quintile of firms in our sample have significantly increasing PREA measures over time whereas the smallest quintile of firms in our sample either exhibit insignificant over-time changes in PREA (the RI metric) or a significantly decreasing over-time change (the RANK% metric). Although statistically significant, our RANK% metric indicates that, consistent with Bamber et al. (2000), significant price reactions are not typical for most firms. Based on the estimated intercepts for RANK%, on average, only 24% to 32% of all firms exhibit a "large" PREA (recall that five revaluation index windows per firm are considered; therefore, by chance, the earnings announcement window will contain the largest revaluation index 20% of the time). We also analyze the mean price reaction metrics by considering only observation with PREAs in the top ten percent each quarter. Results (not presented) corroborate the finding that a minority of firms is driving the mean price reaction results. Collectively, the results in Table 3 support the consistency and pervasiveness of over-time increases in price reactions to earnings. While the mean PREA of all firms increases over time, this result is driven by (1) relatively large firms and (2) a minority of firms within each size-based partition (reflected by the RANK% metric). 4.2 Additional Analysis In this subsection, we further investigate the pervasiveness of historical changes in price reaction to earnings by (1) controlling for industry composition, (2) considering alternative measures of "large" PREA, and (3) analyzing a constant sample of firms. We 11

14 consider industry membership in order to control for the differing cross-sectional influence on PREA attributable to the relative decline in the number of traditional manufacturing firms and the relative rise in the number of high-tech and service firms during our sample period. 11 As shown in Table 4, firms vary substantially by industry therefore controlling for alternative factors that may affect historical changes in PREA in addition to size. Tables 5 and 6 present results for the RI and RANK% price-reaction metrics, respectively. 12 Table 5 shows that, even after controlling for industry membership, historical changes in PREA are positively associated with firm size (i.e., as size increases, RI decreases). However, the traditional fixed asset intensive firms (group 1), and to a lesser extent, the intangible-intensive firms (group 2), are driving the results for the mean revaluation index presented earlier. When we consider the RANK% metric (Table 6), we again find that a minority of firms drive the observed price reaction to earnings announcements across all industry groupings (evidenced by the intercept on the RANK% regressions). Further, all industry groupings except utilities exhibit a significant positive coefficient for the TIME SIZE interaction, indicating the positive correlation between firm size and RANK% increases over time. Overall, these results suggest that size (market value) is still important in understanding historical changes in PREA after controlling for industry membership. 11 We use industry classifications proposed by Landsman and Maydew (1999) that include traditional fixed asset intensive firms, intangible-intensive firms, utilities, financial services, and service firms. We also consider a second industry classification (not presented) that expands the fixed asset intensive firms into five of its components the results are generally unaffected. 12 In order to parsimoniously display results, a single regression for each of the five industry groups is presented in Tables 5 and 6. A regression for each size quintile (as presented in Table 3) would lead to 25 separate regressions for each price reaction metric. 12

15 As a second sensitivity test, we consider alternative measures of "large" PREA other than RANK% (not presented). We consider how the percentage of all firms exhibiting a "large" PREA changed as the threshold defining "large" increased. Approximately 46% of firms have a RI greater than 1.0, 33% of firms have a RI greater than 1.5, and only 25% of firms have a RI greater than 2.0. Consistent with the nonparametric RANK% results, (1) firms exhibiting an abnormally large price reaction are in the minority and (2) there is a significant positive relationship between firm size and PREA regardless of the "large threshold" PREA metric considered. Finally, we consider only firms that are present in both the beginning and the end of our sample period (this "constant sample" of firms is approximately 22 percent of the full sample). Firms are placed into size portfolios based on their quarterly ranking in the full sample. As such, the distribution of the constant sample firms includes a disportionate number of firms from the largest quintile. 13 Analyzing the constant sample addresses concerns that the observed change in PREA is due to new firms or mortality bias. As shown in Table 7, results are consistent with those previously reported (i.e., the intercept generally decreases as firm size increases and the over-time change in price reaction metrics is positively correlated with firm size). However, the differences among the size quintiles are less pronounced and only the differences among the intercepts of RANK % are significant. This is likely due to larger firms dominating the constant sample (the largest firms comprise 36 percent of the constant sample in 1975 increasing 13 The constant sample consists of over 2,000 firms, approximately 700 of which were in the largest size quintile throughout the sample period. Over 80% of the remaining 1,300 firms moved to larger size quintiles, including over 500 firms moving into the largest size quintile. Only 54 firms started in the smallest quintile and remained there throughout the sample period. 13

16 to 62 percent in 1997) and smaller firms that survive over twenty years are more similar to larger firms with respect to providing information to the market and investors. 5.0 Conclusion Like Landsman and Maydew (1999) and Kross and Kim (1999), we document that, on average, accounting earnings announcements provide an increasing amount of new information to market participants as evidenced by an increasing price reaction to earnings announcements (PREA) over time. We find that the over-time increase in PREA is positively associated with firm size (market value) and that relatively small firms exhibit either no historical change in PREA or a significant decrease in PREA over time. In addition, we document that a minority of firms drives the mean over-time increase in PREA. Finally, we show that the preceding results are robust to industry membership although two industry groups (traditional fixed asset intensive firms and intangibleintensive firms) exhibit the strongest associations. Although our results suggest that, on average, investors and analysts increasingly extract price relevant information from earnings announcements, we believe this result should be interpreted cautiously for two reasons. First, our results do not imply that earnings announcements have become increasingly useful over time. Historical increases in PREA may be caused by changes in voluntary disclosure, analyst following, etc. and these changes may have no bearing on the usefulness of accounting disclosures. Second, when debating changes to the reporting model in an effort to provide more price-relevant disclosures, our evidence suggests that focusing only on mean results may lead to a 14

17 reporting model that is an improvement for only a relatively small range of publicly held firms. 15

18 References American Institute of Certified Public Accountants Improving business reporting - a customer focus: Meeting the needs of investors and creditors (Comprehensive Report of the Special Committee on Financial Reporting). New York, NY: AICPA. Ashbaugh, H., K. Johnstone, and T. Warfield Corporate reporting on the internet. Accounting Horizons 13: Atiase, R. K Predisclosure information, firm capitalization, and security price behavior around earnings announcements. Journal of Accounting Research 23: Ball, R. and Brown, P An empirical evaluation of accounting income numbers. Journal of Accounting Research 6: Bamber, L. S., T. E. Christianson, and K. M. Gaver Do we really "know" what we think we know? A Case Study of Seminal Research and its Subsequent Overgeneralization. Accounting, Organizations, and Society 25: Beaver, W The information content of annual earnings announcements. Journal of Accounting Research 6: Brown, S., K. Lo, and T. Lys Use of R 2 in accounting research: measuring changes in value relevance over the last four decades. Journal of Accounting and Economics 28: Carter, M. E. and B. S. Soo The relevance of Form S-K reports. Journal of Accounting Research 37: Collins, D., E. Maydew, and I. Weiss Changes in the value relevance of earnings and book values over the past forty years. Journal of Accounting and Economics 24: Francis, J. and K. Schipper Have financial statements lost their relevance? Journal of Accounting Research 37: Kim, O. and R. Verrecchia Trading volume and price reaction to public announcements. Journal of Accounting Research 29: Kross, W. and M. Kim Differences between market responses to earnings announcements in the 1990s v. 1960s. Working paper: Purdue University and the University of Missouri - Columbia. 16

19 Landsman, W. R. and E. L. Maydew Beaver (1968) revisited: Has the information content of annual earnings announcements declined in the past three decades? Working paper: University of North Carolina - Chapel Hill. Lev, B. and P. Zarowin The boundaries of financial reporting and how to extend them. Journal of Accounting Research Supplement 37: Lys, T. and J. S. Sabino Research design issues in grouping-based tests. Journal of Accounting and Economics 32: Marais, L An application of the bootstrap method to the analysis of squared standardized market model prediction errors. Journal of Accounting Research 22: Patell, J. M Corporate forecasts of earnings per share and stock price behavior: Empirical Tests. Journal of Accounting Research 14:

20 Figure 1 Event Study Windows and Price-Reaction Metric Procedure Step 1 Step 2 Step 3 Step 4 q-9 q-1 q t-2 t+2 Earnings Release Date Quarter q, t=0 Step 1: Estimate firm-specific market model parameters. Step 2: Estimate firm-specific unexpected returns using model parameters obtained in step 1. Step 3: Determine firm-specific estimate of average predisclosure information (e.g., news reports, press release information, etc.) available to investors prior to the quarterly earnings announcement. Step 4: Examine the information content of the earnings announcement three-day window and the surrounding four three-day windows (t-2 to t+2). t is a three-day trading window. q is quarter. 18

21 Figure 2 Aggregate Analysis of Price Reaction Metrics (Irrespective of Time) Panel A: Revaluation Index Report Period Interval (earnings announcement is window 0) All firms Smallest Quintile Middle 60% Largest Quintile Panel B: Rank % Report Period Interval (earnings announcement is window 0) All firms Smallest Quintile Middle 60% Largest Quintile 19

22 Table 1 Sample Determination Compustat data from with quarterly earnings announcements 338,962 Less: Incomplete financial information 57,750 CRSP market data not available 42,280 Earnings release dates more the 90 days past the end of the applicable quarter 6,696 Earnings release dates within 5 days of SEC filing deadline 35,425 Available firm-year observations 196,811 Available firm-year observations distribution by quarter: 1 st quarter 46,468 2 nd quarter 47,482 3 rd quarter 47,488 4 th quarter 55, ,811 20

23 Table 2 Description of Sample Panel A: Full Sample (196,811 firm quarter observations) ($ in millions except per share data) Mean Median Standard Deviation 25 th Percentile 75 th Percentile Assets 2, , ,111 Common equity , Market Value 1, , Sales , Earnings per share Days past quarter end for earnings release Revaluation index Panel B: Size Quintiles Means ( 39,000 quarter-year observations per quintile) Size 1 (smallest) Size 2 Size 3 Size 4 Size 5 (largest) Assets ,465 10,701 Common equity ,157 Market value ,603 Sales ,342 Earnings per share Days past quarter end for earnings release Revaluation index Rank % Dependent Variable Definitions Revaluation index = The earnings announcement window revaluation index. Rank % = The percentage of observations where the earnings announcement revaluation index was the largest among the revaluation indices for five, three-day windows centered on the earnings announcement window. 21

24 Table 3 Price Reaction to Earnings Announcements Regressed on Time Variables (t-statistics) Dependent t = α 0 + α 1 TIME + ε t Panel A: Overall Analysis (n=92 quarters) Dependent Variable RI RANK% All firms * (14.40) Intercept TIME Adj. R 2 Intercept TIME Adj. R * (1.99) 3.1% * (72.13) Panel B: Quintile Analysis (n=92 quarters for each size quintile) * (8.38) Dependent Variable RI RANK% Size 1 (smallest firms) * (11.87) Size * (12.88) Size * (12.97) Size * (8.51) Size 5 (largest firms) * (7.36) F-statistic over size quintiles 43.2% Intercept TIME Adj. R 2 Intercept TIME Adj. R (-1.03) 0.0% * (54.41) * (-2.48) % * * (0.29) (48.59) (3.27) 0.019** 2.2% * * (1.74) (55.09) (5.76) * 8.8% * * (3.13) (37.46) (7.97) * 11.8% * * (3.63) (35.24) (10.72) 7.67 * 4.19 * * * 5.8% 9.6% 26.1% 40.8% 55.6% * / ** Significant at the 0.05 / 0.10 level. Variable Definitions RI = The mean quarterly revaluation index during the earnings announcement window. RANK% = The percentage of observations where the earnings announcement revaluation index was the largest among the revaluation indices for five, three-day windows centered on the earnings announcement window. TIME = Calendar year number where 1975=0 and 1997=22. 22

25 Table 4 Industry Analysis Mean Descriptive Statistics Industry Grouping ($ in millions) n 111,650 16,274 12,740 34,784 20,552 Assets 1, ,798 7, Common equity Market value 1,100 1,439 1, Sales Earnings per share Days past quarter end for earnings release Revaluation index Rank % Dependent Variable Definitions Revaluation index = The earnings announcement window revaluation index. Rank % = The percentage of observations where the earnings announcement revaluation index was the largest among the revaluation indices for five, three-day windows centered on the earnings announcement window. Industry Groupings Group 1 = traditional fixed asset intensive firms; SIC codes , , , , and Group 2= intangible-intensive firms; SIC codes , , and Group 3 = utilities; SIC codes Group 4 = financial services; SIC codes Group 5 = services; SIC codes

26 Table 5 Industry Analysis of Revaluation Index RI t = α 0 + α 1 TIME + α 2 SIZE2 + α 3 SIZE3 + α 4 SIZE4 + α 5 SIZE5 + α 6 (TIME*SIZE2) + α 7 (TIME*SIZE3) + α 8 (TIME*SIZE4) + α 9 (TIME*SIZE5) + ε t Industry Grouping n= Intercept * * * * * TIME SIZE ** SIZE * SIZE * ** SIZE5 (largest) * TIME * SIZE TIME * SIZE * TIME * SIZE * * TIME * SIZE * * Adjusted R 2 7.4% 14.3% 0.5% 0.1% 0.0% * / ** Significant at the 0.05 / 0.10 level Variable Definitions RI = The mean quarterly revaluation index during the earnings announcement window. TIME = Calendar year number where 1975=0 and 1997=22. SIZE (x) = Indicator variables indexed to the size quintile indicated. For example, SIZE 5 represents the largest quintile of sample firms. Industry Groupings: Group 1 = traditional fixed asset intensive firms; SIC codes , , , , and Group 2= intangible-intensive firms; SIC codes , , and Group 3 = utilities; SIC codes Group 4 = financial services; SIC codes Group 5 = services; SIC codes

27 Table 6 Industry Analysis of Rank % RANK% t = α 0 + α 1 TIME + α 2 SIZE2 + α 3 SIZE3 + α 4 SIZE4 + α 5 SIZE5 + α 6 (TIME*SIZE2) + α 7 (TIME*SIZE3) + α 8 (TIME*SIZE4) + α 9 (TIME*SIZE5) + ε t Industry Grouping n= Intercept * * * * * TIME * SIZE * * SIZE * * * SIZE * ** * * SIZE * * * TIME * SIZE * * TIME * SIZE * ** * TIME * SIZE * * * * TIME * SIZE * * * * Adjusted R % 19.4% 1.4% 10.6% 9.7% * / ** Significant at the 0.05 / 0.10 level Variable Definitions RANK% = The percentage of observations where the earnings announcement revaluation index was the largest among the revaluation indices for five, three-day windows centered on the earnings announcement window. TIME = Calendar year number where 1975=0 and 1997=22. SIZE (x) = Indicator variables indexed to the size quintile indicated. For example, SIZE 5 represents the largest quintile of sample firms. Industry Groupings: Group 1 = traditional fixed asset intensive firms; SIC codes , , , , and Group 2= intangible-intensive firms; SIC codes , , and Group 3 = utilities; SIC codes Group 4 = financial services; SIC codes Group 5 = services; SIC codes

28 Panel A: Mean Descriptive Statistics ($ in millions) n = 43,680 Size 1 (smallest) Table 7 Constant Sample Analysis Market Value Quintiles Size 2 Size 3 Size 4 Size 5 (largest) % of constant sample 7.3% 10.5% 13.5% 22.7% 46.0% Assets ,687 13,125 Common equity ,584 Market value ,727 Sales ,794 Earnings per share Revaluation index Rank % Dependent Variable Definitions Revaluation index = The earnings announcement window revaluation index. Rank % = The percentage of observations where the earnings announcement revaluation index was the largest among the revaluation indices for five, three-day windows centered on the earnings announcement window. 26

29 Table 7 (continued) Constant Sample Analysis Panel B: Price Reaction to Earnings Announcements Regressed on Time (t-statistics) Quintile Analysis (n=92 for each size quintile) Dependent t = α 0 + α 1 TIME + ε t Dependent Variable RI RANK% Size 1 (smallest firms) * (4.94) Size * (8.12) Size * (8.75) Size * (10.35) Size 5 (largest firms) * (7.56) F-statistic over size quintiles Intercept TIME Adj. R 2 Intercept TIME Adj. R * (3.25) * (2.73) * (2.65) * (3.60) * (3.88) 9.5% * (16.83) 6.6% * (19.80) 6.2% * (24.00) 11.6% * (25.71) 13.4% * (28.68) * (3.04) * (5.04) * (4.22) * (5.85) * (8.84) * % 21.1% 15.6% 26.7% 45.9% * / ** Significant at the 0.05 / 0.10 level. Variable Definitions RI = The mean quarterly revaluation index during the earnings announcement window. RANK% = The percentage of observations where the earnings announcement revaluation index was the largest among the revaluation indices for five, three-day windows centered on the earnings announcement window. TIME = Calendar year number where 1975=0 and 1997=22. 27