Increased Information Content of Earnings Announcements in the 21st Century: An Empirical Investigation

Transcription

1 Increased Information Content of Earnings Announcements in the 21st Century: An Empirical Investigation William H. Beaver Joan E. Horngren Professor (Emeritus) Graduate School of Business, Stanford University, Stanford, CA, 94305, USA Maureen F. McNichols Marriner S. Eccles Professor of Public and Private Management and Accounting Graduate School of Business, Stanford University, Stanford, CA, 94305, USA Zach Z. Wang Assistant Professor University of Illinois at Urbana-Champaign, Champaign, IL, 61821, USA March 8, 2017 The authors thank Paul Reist and Nora Richardson for their assistance in data collection. The authors also appreciate the comments received from workshop and doctoral seminar participants at Stanford University.

2 Increased Information Content of Earnings Announcements in the 21st Century: An Empirical Investigation Abstract This study examines the factors contributing to a striking increase in information content of quarterly earnings announcements, measured as the absolute magnitude of stock price revision at earnings announcements relative to price revision at other times from 1999 to We provide evidence that one-day announcement windows exhibit roughly twice the price response observed for three-day windows. Furthermore, we find that management guidance, analyst forecasts, and disaggregated line items are more frequently bundled with earnings announcements over our sample period, and these variables help to explain the increase in the information content of earnings announcements over time. However, after controlling for concurrent disclosures, a striking increase in information content of earnings announcements remains. In addition, we compare the price response to earnings announcements without management guidance or analyst forecasts to the response to management guidance or analyst forecasts issued outside earnings announcements. We find that the relative return volatility on stand-alone earnings announcement dates is on average comparable to that on management guidance dates. However, management guidance exhibits a declining pattern of information content, whereas the information content of earnings is increasing and surpasses that of guidance from 2007 to Abnormal return variability at stand-alone earnings announcements is more than twice that at analyst forecast dates. Our findings suggest that although earnings announcements are quarterly and mandatory, they are informative to the market and after controlling for the effects of concurrent disclosures, are increasingly more informative to investors in the 21 st century. Keywords: capital markets; earnings announcements; information content; return volatility

3 Increased Information Content of Earnings Announcements in the 21st Century: An Empirical Investigation I. INTRODUCTION Beginning with Ball and Brown (1968) and Beaver (1968), a large literature documents that earnings announcements have significant information content. More recent literature explores whether and how the information content of earnings announcements varies across time and across firms. 1 This literature includes a recent study by Beaver, McNichols and Wang (2017), who find that the information content of earnings announcements increases dramatically after 2001, as captured by the absolute magnitude of stock price revision at earnings announcements relative to price revisions at other times. 2 This paper examines two questions concerning this surprising pattern: First, is the increase in relative abnormal price revision due to more information being released to investors at earnings announcements, and if yes, what is the nature of the information? Second, how does the magnitude of the price response to earnings announcements compare to the magnitude of the price response to information events on nonearnings announcement days, such as management guidance and analyst forecasts? In other words, does relative price variability at the time of other disclosure events also exhibit a similar pattern of increase, or is the increase in information content in recent years unique to earnings announcements. In order to study the time series and cross-sectional factors contributing to abnormal return variability at earnings dates, we seek the most powerful measure of stock price response 1 These studies include Francis Schipper and Vincent (2002b), Collins Li and Xie (2009), and Landsman and Maydew (2002), and more recently Beaver McNichols Wang (2017). 2 We refer to the information content of earnings announcements rather than earnings as we study the reaction of prices on earnings announcement days, which will reflect the reaction to all information released on those days. This potentially includes information about the balance sheet and statement of cash flows, as well as management guidance or other disclosures. 1

4 possible. For this reason, we develop a one-day U-statistic measure, hereafter referred to as USTAT, to examine the information content of earnings announcements in recent years. This measure incorporates the timing of earnings announcements and whether they occur before or during market hours or after the market close, to pinpoint the single day reaction to the earnings news. Our empirical analyses establish several significant findings. First, we find that the mean USTAT observed at earnings announcement periods is substantially above the 99 th percentile of the distribution of USTAT drawn from non-earnings announcement periods for each of the 14 years in our 1999 to 2012 sample period and shows a substantial increase overall, but also a decline during We find that the mean USTAT is almost twice the mean TCU, the abnormal return volatility over a three-day interval surrounding the earnings announcements. Our statistic measuring relative price revision activity on earnings announcement dates from 1999 to 2012 has a mean of 6.80, compared to a mean of 1.08 in non-announcement periods and a mean of 3.44 using three-day stock returns around earnings announcements. These findings suggest that the additional two days in the three day window around earnings announcements largely added noise. We also conduct median tests, R 2 tests (Ball and Shivakumar 2008), and abnormal trading volume tests (Landsman and Maydew 2002) which all exhibit a similar pattern of overall increase, with declines in These results provide assurance that, together with our analysis of medians as well as means, our findings are not attributable to either extreme observations or the choice of empirical measure. Given the wide use of many of these metrics in accounting research, they also point to the importance of understanding the temporal and crosssectional patterns affecting them. 2

5 Second, we examine what factors could explain the relative variability of returns at earnings announcements from 1999 to We observe the dramatic increase of the information content from 2001 onwards with the mean of USTAT reaching a peak of 9.94 in our final sample year We begin by decomposing USTAT into the numerator, Announcement Return Squared, and the denominator, Non-Announcement Return Volatility. We find that the increase in the USTAT ratio is driven by a relative decline in Non-Announcement Return Volatility, suggesting a decline in the price response to information released outside earnings announcements. The numerator can be further related to Unexpected Earnings, which captures the difference between realized street earnings and median analyst expectations, and Earnings Response Coefficient ( ERC ), which is the relation between announcement stock returns and Unexpected Earnings. We find that a decline in the ERC helps to explain the decline of USTAT in the financial crisis. Third, we examine the relation between relative return variability at earnings announcements and disclosures included in the earnings press release or in the same day as the earnings press release. Specifically, we examine whether management guidance, concurrent analyst forecasts, and financial statement line item disclosures help to explain the increase in USTAT over time. Prior research by Anilowski, Feng and Skinner (2007) and Rogers and Van Buskirk (2013) finds that firms are more likely to issue management guidance concurrent with earnings post-reg FD than pre-reg FD. Before the implementation of Reg FD, if managers issued management guidance outside earnings announcement dates, they could answer analyst questions privately about these guidance. In a post-reg FD world, such selective disclosure is prohibited, and bundling management guidance with earnings announcements allows managers to communicate with analysts in public venues such as ensuing earnings announcement 3

6 conference calls (Rogers and Van Buskirk, 2013). We find that the percentage of earnings announcements with concurrent management forecasts increases from less than 5% in 1999 to 43.27% in 2012, and the variability of returns at earnings announcement dates of firms with concurrent management guidance is significantly greater than that for firms without concurrent management guidance. We next examine the cross-sectional association between information content and concurrent analyst forecasts. Prior research by Lang and Lundholm (1996), Francis, Schipper and Vincent (2002b), and Frankel, Kothari and Weber (2006) find that analyst forecasts complement the informativeness of accounting information. This could occur because analysts update forecasts when the earnings news is most significant, because the analysts forecasts facilitate more rapid dissemination or processing of earnings information, or because analysts augment the information provided by management. We find that the percentage of earnings announcements with concurrent analyst forecasts increases from 30% in 1999 to 70% in 2012, and the variability of returns at earnings announcements dates of firms with concurrent analyst forecasts is significantly greater than that for firms without concurrent analyst forecasts. The third set of concurrent disclosures we examine are financial statement line items. Prior research by Hoskins, Hughs, and Ricks (1986), Livnat and Zarowin (1990), Francis, Schipper and Vincent (2002b), and Chen, Defond and Park (2002) find that disaggregated line items from the income statement, balance sheet, and statement of cash flows complement the informativeness of earnings. We find that firms continue to increase their disclosure of financial statement items from 1999 to 2012, and we observe a positive association between relative return variability at earnings announcements and concurrently disclosed financial statement items. In our multivariate analysis, concurrent management guidance, concurrent analyst forecasts, and 4

7 disclosed financial statement items combined with other control variables in Beaver et al. (2017) help explain more than half of the time trend in the information content of earnings announcements. Thus a substantial portion of the increase in information content at earnings announcements is indeed due to additional information being disclosed by firms and by analysts at earnings announcement dates. Finally, we compare the information content of earnings announcements to that of other important disclosure events such as management guidance and analyst forecasts. We find that the average information content of stand-alone earnings announcements is similar to that of standalone management guidance. However, stand-alone management guidance exhibits a declining pattern of information content, whereas the information content of stand-alone earnings is increasing and surpasses that of guidance from 2007 to We also find that the information content of stand-alone earnings announcements is more than twice that of stand-alone analyst forecasts, but analyst forecast days with at least one forecast or revision are six times more frequent than earnings announcements. Our paper makes the following contributions. First, we provide evidence that the move from three-day announcement windows to one-day announcement windows increases the measurement of price response two-fold, indicating previously documented results using threeday or longer announcement windows understate the information content of earnings announcements. Second, we find that management guidance, analyst forecasts, and disaggregated line items are more frequently bundled with earnings announcements, and these variables partially explain the increase in the information content of earnings announcements over time. However, after controlling for these concurrent items, there is still a significant increasing trend in abnormal price variability at earnings announcements. Last but not least, we 5

8 adapt the one-day information content measure to other events and compare the information content of earnings announcements to that of management guidance and analyst forecasts issued outside of earnings announcements. We find the relative return volatility of stand-alone earnings announcements is more than twice that of stand-alone analyst forecasts and exhibits the increasing pattern of price response to all earnings announcements, whereas the price response to analyst forecasts is essentially flat over our sample period. In contrast, the market reaction to stand-alone management guidance is approximately 7 times that for stand-alone earnings announcements in 1999, but declines over the sample period as the market reaction to earnings announcements experiences a striking increase. This results in a market reaction to stand-alone earnings announcements that is 50% greater than for management guidance by Taken as a whole, these findings suggest earnings announcements are an increasingly relevant source of information to investors in recent years. The remainder of the paper proceeds as follows. Section II discusses the background. Section III discusses the research design. Section IV presents sample properties. Section V discusses the results, and Section VI provides concluding remarks. II. BACKGROUND By showing both trading volume and return volatility increasing at the time of annual earnings announcements, Beaver (1968) provides evidence that earnings reports provide information to investors, as measured by the U-statistic and abnormal volume. 3 Beaver (1968) defines the U-statistic to be the announcement week squared residual returns divided by the 3 The U-statistic and the abnormal volume measures are based upon the notion that a necessary condition for a signal to have information content is that beliefs are altered and that the change in beliefs must be sufficient to alter actions across signals. 6

9 mean squared residual returns in the non-announcement period. After daily return data become available, it has been common practice to measure the mean U-statistic over a three-day interval surrounding the earnings announcement, due to uncertainty over the exact timing of the earnings release (e.g., Landsman and Maydew, 2002). For instance, to examine the time trend of the information content of earnings announcements from 1971 to 2011, Beaver, McNichols and Wang (2017) develop a measure that compounds returns over the three-day interval surrounding the earnings announcement to calculate a U-statistic (hereafter, TCU), and find that the movement from a weekly measure to a three-day return window increases the measure of information content. The movement from a three-day window to a one-day window can potentially further increase the magnitude of the information content given the three-day measure includes two days with less new information arrival. Prior research (Patell and Wolfson, 1982) identified the importance of after-hours announcements, which may not be reflected in price until the next trading day. Most recently, using announcement date and time data from WSJ.com, Berkman and Truong (2009) document that the proportion of after-hours earnings announcements increased to 40% in 2004, and that the market reaction is observed not on the earnings announcement date but one trading day later. Hence, the identification of a shorter window must include an adjustment for after-hours announcements. Because the exact timing of earnings releases is now available from several databases (E.g. IBES and Bloomberg) after 1999, we are able to construct a one-day U-statistic measure (hererafter, USTAT) for years 1999 to 2012 that captures the trading day of the earnings release. If earnings are released after normal trading hours, we define the earnings announcement day to be the following trading day in which market price is able to reflect the earnings announcement. 7

10 To evaluate the information content of earnings announcements based on USTAT, we first assess whether the variability of stock prices on the earnings announcement day is greater than that on randomly-chosen non-announcement days. The non-report period is not a no information benchmark because other information could arrive in the period. Because prior research has shown that earnings announcements have incremental information content using the three-day U-statistic, TCU, we then compare the mean and median values of USTAT to those of TCU to assess how the use of USTAT changes the measurement of price response to earnings announcements. The use of one-day return incorporating the timing of the announcement raises the prospect that results reported using weekly or three-day announcement windows understate the information content of earnings announcements. To ensure the robustness of our results, we also evaluate the information content using the R 2 measure from Ball and Shivakumar (2008) and the abnormal trading volume measure from Landsman and Maydew (2002). Similar to the USTAT, the R 2 and volume measures do not require an expectations model for earnings to capture information content. Several studies have examined whether the information content of earnings announcements has increased or decreased and explore potential explanations for this trend. Using a sample of firms from 1972 to 1998, Landsman and Maydew (2002) find that their three day U-statistic increases over time after controlling for factors such as firm size, intangible intensity and presence of a loss. Francis, Schipper and Vincent. (2002a) limit the sample to firms that are present in Compustat throughout the entire period from 1980 to 1999 and find that an increasing trend in market reactions exists in spite of declining absolute unexpected earnings and ERC s. Using a hand-collected sample of earnings press releases, they find that over-time increases in the amount of concurrent disclosures and over-time increases in investor reactions to 8

11 concurrent disclosures explain the increase in market reactions. Using a sample of firms drawn from IBES from 1985 to 2000, Collins, Li and Xie (2009) find an increasing trend in the informativeness of earnings announcements, which they attribute to an increasing reaction to Street earnings. Beaver et al. (2017) examine the time trend of the information content of earnings announcements from 1971 to 2011, and find that the rate of increase in relative price variability at earnings announcements after 2000 is significantly higher than before Our sample period is from 1999 to 2012, and the use of a one-day return incorporating the announcement allows more powerful tests of factors contributing to increased relative price variability at earnings announcements in the 21 st Century. The numerator of USTAT captures the price response to information released on earnings announcement days, and the denominator of USTAT captures the price response to information released during non-report windows. Similar to Francis et al. (2002a), the numerator can be further related to the absolute value of unexpected earnings and earnings response coefficient. Analyzing the time trend of these USTAT components can help identify the factors driving the intertemporal change of USTAT. The first specific form of concurrent information we examine is management guidance. Before the implementation of Reg FD, if managers issued management guidance outside earnings announcement dates, they could answer analyst questions privately about their guidance. In a post-reg FD world, such selective disclosure is prohibited, and bundling management guidance with earnings announcements allows managers to communicate with analysts in public venues such as ensuing earnings announcement conference calls (Rogers and Van Buskirk 2013). Because management forecasts contain informative forward-looking information (Waymire 1984) and prior research has documented a significant increase in bundled 9

12 management guidance post Reg FD (Anilowski, Feng and Skinner 2007), we expect that concurrent management guidance can explain at least some of the increase in the price response at earnings announcements. The second form of concurrent information we examine is analyst forecasts. Analysts serve as information intermediaries and play an important role in acquiring and disclosing information, leading to its incorporation into price. Lang and Lundholm (1996) conclude that timely accounting disclosures and analyst following are complements. Francis, Schipper, and Vincent (2002b) find that the information content of earnings announcements is positively associated with the information content of analyst reports. Frankel, Kothari and Weber (2006) provide evidence that analyst informativeness is positively associated with the timeliness of earnings information. Since selective disclosure is prohibited post Reg FD, analysts may choose to ask management questions in earnings conference calls and make timely forecasts and revisions to facilitate price formation around earnings announcements. Given prior findings on the complementary role of analysts, we conjecture that concurrent analyst forecasts can help explain the increase in price response to earnings announcements. The third factor we examine is concurrently disclosed financial statement line items. Hoskins, Hughs, and Ricks (1986) find that other information in earnings press releases besides earnings explains a significant portion of market reactions to earnings announcements. Livnat and Zarowin (1990) find that disaggregation of financing and operating cash flows into their components significantly improves the degree of association between cash flows and security returns. Francis, Schipper, and Vincent (2002a) provide evidence that the disclosure of earnings components has increased the usefulness of the earnings announcements from 1980 to Chen, Defond, and Park (2002) find that managers disclosing balance sheets in response to 10

13 investor demand for value relevant information to supplement earnings. D Souza, Ramesh and Shen (2010) conclude that controlling for incentives to limit disclosures, line item disclosures in earnings press releases highlight value-relevant information. By the end of 1999, out of the 30 companies in the Francis et al. (2002a) sample, 77.8% provide detailed income statements in their earnings press releases, but only 25.9% and 7.4% provide detailed balance sheets and cash flow statements respectively. We conjecture that management has increased disclosure of financial statement line items in earnings press release from 1999 to 2012, which may contribute to the increase in price response to earnings announcements. Last but not least, prior research has emphasized the importance of two disclosure events: management guidance and analyst forecast (Ball and Shivakumar 2008, Beyer, Cohen, Lys and Walther 2010). Using the time stamps of management guidance and analyst forecast from IBES, we adapt our one-day information content measure to quantify the information content of management guidance and analyst forecast. Instead of comparing the information content of earnings announcements to that of non-announcements, we compare the information content of stand-alone earnings announcements to that of stand-alone management guidance and standalone analyst forecast to assess the relative importance of earnings announcements. In particular, we examine whether each of these types of information exhibit the same pattern of increase that is observed at earnings announcement dates. To the extent the pattern we observe is common to other information events, it may reflect broader changes in trading behavior by investors, rather than an increase in relative informativeness of earnings announcements. 11

14 III. EMPIRICAL METHODOLOGY Main Measure of the Information Content of Earnings Announcements Our main measure of the information content of earnings announcements, the one-day U- Statistic ( USTAT ) captures the magnitude of the squared residual return on the date of the earnings announcement, hereafter the testing period TP, to the variance of residual returns during the non-announcement period, hereafter the estimation period EP. For each quarterly earnings announcement (day 0), an estimation period, EP, is defined as the period from 130 to 10 days prior to the earnings announcements and days 10 to 130 days after the announcement. We obtain the time stamps of earnings announcements from Bloomberg and IBES. If earnings are released after normal trading hours, we define the earnings announcement day to be the next trading day because market price is not able to reflect new information in the earnings announcement until the next trading day. Refer to Appendix 2 for more details. In a procedure to be explained shortly, we also apply the same estimation approach for each as if non-report earnings announcement randomly selected from the non-report period. Following Landsman and Maydew (2002) and Beaver, McNichols and Wang (2017), we aggregate quarterly earnings announcements into calendar years based on the dates of earnings release, and then calculate the means and medians of USTAT for each calendar year. Refer to Appendix 3 for further details on USTAT. Development of the Nonparametric Distribution for USTAT To generate a sampling distribution under the null hypothesis that earnings announcements do not have incremental information content, for each quarterly earnings announcement, we randomly select a day during its non-report or estimation period. That day is treated as if it were an announcement day. We then calculate USTAT in the same manner as the 12

15 actual earnings announcement, and obtain the mean and median of USTAT for that sample of randomly-chosen announcement dates. Repeating the procedure 1,000 times produces a sampling distribution, which is then compared with the values obtained for the actual earnings announcements. 4 The procedure is described in greater detail in Appendix 4. Three-Day Cumulative U-Statistic ( TCU ) TCU captures the magnitude of the squared residual return during the three days around the announcement, hereafter the testing period TP, to the variance of residual returns during the non-announcement period, hereafter the estimation period EP. For each quarterly earnings announcement (day 0), TP is defined as the period from day-1 to day+1, and EP is defined as the period from 130 to 10 days prior to the earnings announcements and days 10 to 130 days after the announcement. We obtain the earnings announcement dates from Compustat. 5 We aggregate quarterly earnings announcements into calendar years based on the dates of earnings release, and then calculate the means and medians of TCU for each calendar year. Refer to Appendix 5 for further details on TCU. Other Measures of the Information Content of Earnings Announcements Ball and Shivakumar (2008) R 2 Measure For each calendar year, we run regressions of calendar-quarter stock returns on quarterly earnings announcement-window returns, and the R 2 measures the proportion of total information incorporated in share prices quarterly that is associated with quarterly earnings announcements, and does not require a specific earnings expectation model. We define the earnings 4 Drawing the randomly chosen announcement date from the estimation period for each earnings announcements ensures that the randomly chosen date is from approximately the same calendar period as the actual earnings announcement date. Given that our sample is drawn over 14 calendar years, this procedure seems preferable to allowing the randomly chosen announcement date to occur at any time in the firm s history. 5 For comparability with prior literature using three day windows, including Landsman and Maydew (2002), Francis, Schipper and Vincent (2002a), Collins Li and Xie (2009) and Beaver, McNichols and Wang (2017), we don t adjust TCU for earnings announcements issued after the market close. 13

16 announcement-window by either the one-day window based on Bloomberg and IBES earnings time stamps or the three-day window around Compustat earnings announcement dates. Following Ball and Shivakumar (2008), we remove firm-quarter observations with fewer than 60 trading days in a calendar quarter. Under the null hypothesis that earnings announcements do not have incremental information, the R 2 is expected to be 1.67% (100%/60) for the one-day announcement window and 5% (100%/20) for the three-day announcement window. Abnormal Trading Volume ( AVOL ) Following Beaver (1968), Cready and Hurtt (2002) and Landsman and Maydew (2002), we create a volume-based measure AVOL to measure the information content of earnings announcements. The numerator is the difference between shares outstanding-scaled trading volume on the earnings announcement day (day 0) and average shares-outstanding scaled trading volume in the non-announcement period (day-130 to day-10, day+10 to day+130). The denominator is the standard deviation of shares-outstanding scaled volume in the nonannouncement period. Under the null hypothesis that earnings announcements do not have incremental information, AVOL should be zero. As Beaver (1968) notes, price-based tests reflect revisions in expectations of the market as a whole whereas volume can reflect revisions in individual investors expectations. The AVOL measure can reflect information that affects the investment decisions of individual investors even in the absence of significant price revision, whereas the USTAT measure can reflect significant price revision, even if it does not generate substantial trade. Thus the price and volume measures capture different dimensions of the effect of earnings information on investors investment holdings. Similar to our price-based measures, our volume measures are adjusted for after-hours earnings announcements. 14

17 Components of USTAT To analyze the intertemporal change of USTAT, we first decompose USTAT into the numerator and the denominator. The numerator is the squared residual return on the date of the earnings announcement, Announcement Return Squared, and the denominator is the variance of residual returns during the non-announcement period, Non-Announcement Return Volatility. We further analyze the residual return on the date of earnings announcement by considering two subcomponents: Unexpected Earnings, which is the difference between IBES realized street earnings and IBES median consensus earnings forecast at fiscal quarter end scaled by price, and Earnings Response Coefficient ( ERC ), which is the relation between residual return on the date of earnings announcement and unexpected earnings. For each calendar year, we regress earnings announcement residual return on Unexpected Earnings to obtain estimates of ERC. Because USTAT is unsigned, we focus on the absolute value of Unexpected Earnings. The Relation between USTAT and Other Variables We examine the association of USTAT with several variables, including time, management guidance, analyst forecasts, and financial statement line items. T is a trend variable that takes values from 0 to 14 for calendar years from 1999 to Guidance is an indicator variable that is equal to one when concurrent management guidance is issued on the earnings announcement date and zero otherwise. AF is an indicator variable that is equal to one when any analyst forecast is issued on the earnings announcement date and zero otherwise. Following D Souza et al. (2010), we collect financial statement line items disclosed in earnings press release from Compustat Snapshot. FS is the percentage of financial statement items disclosed in the earnings announcement. IS [BS] {SCF} is the percentage of income statement [balance sheet] 15

18 {cash flow statement} items disclosed in the earnings announcement. Refer to Appendix 6 for details. Following prior literature, we also include several control variables. 6 LOSS is an indicator variable that is equal to one when PTEBS is negative and zero otherwise, where PTEBS is the sum of Net Income before Extraordinary Items, Tax Expenses and Minority Interest before Special Items. CV is the market capitalization value of the firm s common stock at fiscal quarter end from Compustat. Following Atiase (1985), we take the natural log of CV, LCV, as a proxy for size. The variable for analyst following, NUMESTQ, or the number of analysts making forecasts of the upcoming quarterly earnings at fiscal quarter end, is drawn from the IBES analysts forecast database. For firm-quarters that are not in the IBES database, we assume zero analyst coverage. FIN is an indicator variable that is equal to 1 when the four-digit SIC code is between 6000 and 6999 and 0 otherwise. LAG is the number of days after the end of the fiscal quarter that earnings are announced. NONDEC31 is an indicator variable that is equal to 1 for firm-quarters with Non-Dec31 fiscal year end and 0 otherwise. Following Collins et al. (2009), ABSFE_ST is the absolute value of the difference between IBES realized earnings per share and IBES median consensus earnings per share at fiscal quarter end scaled by price per share. Initially, we explore the relations in an unconditional, bivariate manner (i.e., one variable at a time). In the subsequent section, we also conduct multivariate regression analysis to determine if the bivariate results are preserved conditional on all of the variables included in the regression. 6 Relevant studies motivating our choice of control variables include Landsman and Maydew (2002); Francis, Schipper and Vincent (2002a); and Collins, Li and Xie (2009). 16

19 The Information Content of Earnings Announcements and Other Disclosure Events To compare the information content of earnings announcements to that of management guidance and analyst forecasts, we obtain the time stamps of management guidance and analyst forecasts from IBES, adjust for after-hours announcements, and create one-day information content measures similar to USTAT. Refer to Appendix 3 for details. For multiple analyst forecasts issued on the same day, we count them as a single analyst-forecast event. We then separate all the management guidance (analyst forecast) into three portfolios based on their time stamps: Guidance (Forecast) Bundled with Earnings, Stand-Alone Guidance (Forecast), and Stand-Alone Earnings. The Guidance (Forecast) Bundled with Earnings portfolio includes the management guidance (analyst forecast) issued on the same days as earnings announcements. The Stand-Alone Guidance (Forecast) portfolio includes the management guidance (analyst forecast) not issued on the days of earnings announcements. The Stand-Alone Earnings portfolio includes earnings announcements without concurrent management guidance when compared to Stand-Alone Guidance and earnings announcements without concurrent analyst forecasts when compared to Stand-Alone Forecast portfolio. We calculate the mean and median of the information content of the above portfolios to assess how earnings announcements compare to management guidance and analyst forecasts in price response. IV. SAMPLE PROPERTIES We start with the universe of firms listed on the NYSE, AMEX and NASDAQ markets for which quarterly reporting dates are available from Compustat. We first match Compustat with Bloomberg and then with IBES to obtain a list of firm-quarter observations with earnings announcement time stamps. For after-hours earnings announcements, we use day+1 as the 17

20 earnings announcement date. For each quarterly earnings announcement included, we require that return data are available for the earnings announcement date, and the number of trading days with nonzero return data in estimation period is at least 40. We also make the same requirements for each simulated non-report earnings announcement. We finally have a sample of 240,521 quarterly earnings announcements for which we can calculate USTAT. Refer to Appendix 2 for details on sample selection. For the additional hypotheses, we impose additional restrictions on the sample to ensure each firm-quarter observation has data for the required variables. For our analysis on the Ball and Shivakumar (2008) R 2 measure, we require firm-quarters included to have at least 60 daily return observations in the calendar quarter, and the analysis has a sample size of 234,433. For our analysis on financial statement items, Compustat Snapshot coverage is required and the analysis has a sample size of 195,112. For our multivariate regression models, we require included firmquarters to have all required variables. We have a sample size of 148,102 observations, and we winsorize all variables at the 1st and 99th percentiles. For the information content of management guidance and analyst forecasts, we require the observations included to have non-missing time stamps from IBES, non-missing return data from CRSP, and the number of trading days with nonzero return data in estimation period is at least 40. Guidance Bundled with Earnings, Stand-Alone Guidance, and Stand-Alone Earnings have 64,195, 50,625, and 176,326 observations respectively. Forecast Bundled with Earnings, Stand-Alone Analyst Forecast, and Stand-Alone Earnings have 135,082, 1,491,628, and 105,439 observations respectively. 18

21 V. EMPIRICAL RESULTS Initial Evidence Figure 1 provides evidence concerning the impact of adjusting after-hours earnings announcements. It compares the mean of abnormal return volatility before (Panel A) and after (Panel B) we make adjustment for after-hours earnings announcements. Before adjustment, the mean of abnormal return volatility on day+1 (4.64) is higher than that on day 0 (3.82); after adjustment, the mean of abnormal return volatility on day 0 (6.80) is four times that on day+1 (1.72). The identification and adjustment of after-hours earnings announcements increase the measured price response to earnings announcements on day 0 from 3.82 to 6.80, a 78% increase. Consistent with Berkman and Truong (2009), the findings document the importance of incorporating time stamp information to accurately identify when earnings information is released. Figure 2 provides two forms of evidence concerning relative price variability at earnings announcements. It compares the mean (Panel A) and median (Panel B) of USTAT over each of the years from 1999 through The figure also compares these values with the respective sampling distribution constructed from the non-announcement period. For each calendar year, we plot the value of USTAT mean (USTAT_median) for our event sample against the values of 1000 USTAT means (USTAT_medians) from the null distribution. The mean and median values are above the 99th percentile in each of the 14 calendar years. Consistent with Beaver et al. (2017), the figure displays striking evidence that earnings announcements convey significantly more information than in non-announcement periods. Table 1 Panels A and B report the mean and median values of USTAT by year for each event period, along with the distribution of the respective values for the non-announcement period. For instance, in 2012, the mean value of 19

22 USTAT in Panel A is 9.94 and the 99th percentile of the non-report distribution is Similarly, the median value of USTAT for 2012 in Panel B is 2.00 and the 99th percentile is Over the 14 years, the mean value of USTAT in event windows is 6.80, which is 6.29 times 1.08, the mean value of USTAT in simulated non-report windows. Figure 3 provides two forms of evidence concerning one-day and three-day relative price variability at earnings announcements. It compares the mean (Panel A) and median (Panel B) of USTAT with those of TCU, the Three-Day Cumulative U-statistic, over each of the years from 1999 through The figure displays evidence that one-day relative price variability at earnings announcements is higher than three-day relative price variability at earnings announcements. Table 1 Panel A and B also report the mean and median values of both USTAT and TCU, by year for each event period. For instance, in 2012, the mean value of USTAT in Panel A is 9.94 and the mean value of TCU is Similarly, the median value of USTAT for 2012 in Panel B is 2.00 and the median value of TCU is Over the 14 years, the mean value of USTAT is 6.80, which is 1.98 times 3.44, the mean value of TCU. These findings and the evidence in Figure 1 indicate the price response to earnings largely occurs on the trading day of the earnings announcement. With respect to the time trend in USTAT and TCU, an increase is evident from 1999 to 2012 and the increase is more pronounced for USTAT than for TCU. For instance, the mean value of USTAT is 3.33 in 1999 and it reaches 9.94 in 2012; the mean of TCU is 2.40 in 1999 and it reaches only 4.48 in It is also clear from Figure 2 and 3 that the time trend is nonlinear. For instance, the mean value of USTAT is 9.78 in 2006 but it declines to 6.37 in We discuss potential reasons for both the overall increase and the decline in when we analyze some of the sub-components of USTAT and present our multivariate analysis. 20

23 Other Measures of the Information Content of Earnings Announcements Figure 4 provides evidence concerning R 2 from quarterly regressions of calendar-quarter returns on one-day (Panel A) and three-day (Panel B) announcement returns. The figure compares these values with the respective R 2 under the null hypothesis that earnings announcements do not have incremental information content. Similar to our USTAT findings, the figures display evidence that earnings announcements convey more information than would be expected under the null in each of the years from 1999 through The intertemporal trend of R 2 is also very similar to that of USTAT in Figure 2 and 3. Table 2 reports the R 2 by calendar year. Over the 14 years, the R 2 associated with one-day (three-day) announcement returns are 10.09% (13.95%), which is 6.04 (2.79) times the R 2 expected under the null 1.67% (5%). The R 2 multiple based on three-day announcement returns is 2.17 (6.04/2.79) times that based on oneday announcement returns, which is comparable to the ratio of the mean USTAT to that of TCU, Figure 5 provides evidence concerning relative trading volume at earnings announcements. Panel A and B show a significant day 0 impact after the adjustment of afterhours earnings announcements. Similar to our USTAT findings, the figure in Panel C displays evidence that earnings announcements convey more information than would be expected under the null in each of the years from 1999 through The intertemporal trend of AVOL is similar to that of USTAT in Figure 2 and 3. Overall, we find that the one-day information content of earnings announcements with adjustment for after-hours announcements is more than six times that of non-announcement days and twice as much as the three-day version. An increase in the information content of earnings announcements is evident from 1999 to We also observe a small decline in the information 21

24 content during the financial crisis. Next, we explore the factors that could explain the intertemporal pattern of price response to earnings announcements. Components of USTAT To analyze the intertemporal change of USTAT, we first decompose USTAT into the numerator, Announcement Return Squared, and the denominator, Non-Announcement Return Volatility. Figure 6 Panel A presents the mean of the numerator and the denominator by calendar year. We observe a relative decline of the Non-Announcement Return Volatility from 1999 to 2012 except during the financial crisis. For instance, the mean Non-Announcement Return Volatility declines from 1.92 in 2002 to one-third of that, 0.63 in 2006, and the mean Announcement Return Squared declines from 7.75 in 2002 only to 5.02 in Hence, the relative decline of new information in the non-announcement windows drives the increase in USTAT. During the financial crisis, the mean Non-Announcement Return Volatility increases to 3.29 in 2009, but the mean Announcement Return Squared only increases to in To further understand why Announcement Return Squared does not increase at the same rate as Non-Announcement Return Volatility during the financial crisis, we analyze two of the components of announcement return volatility, Absolute Unexpected Earnings ( ABUE ) and Earnings Response Coefficient ( ERC ). For consistency with our focus on unsigned return response, the earnings surprise is measured by its absolute value. We use the unsquared value to be consistent with the ERC coefficient. Figure 6 Panel B presents the mean ABUE and ERC by calendar year. Although the mean ABUE increases from 0.55 to 3.56 from 2006 to 2009, ERC declines from 1.15 to Firms having large unexpected losses during the financial crisis are likely to have large transitory components in their earnings, which can help explain why we 22

25 observe a decline in ERC, and also the decline of the information content during the financial crisis. We leave the explanation of the variation in the numerator and its relation to information in the earnings release about future earnings for further research. Firm-Quarter Specific Variables In this section, we examine the ability of management guidance, analyst forecast and disclosed financial statement items to explain the differences in USTAT. We present a series of figures that illustrate the differences over time. This form of presentation permits us to examine cross sectional differences, holding time constant, and time series differences holding the cross sectional variable constant. The subsequent section reports the findings of multivariate analysis that combines all of the variables examined with additional control variables. Management Guidance Figure 7 Panel A presents a figure with the percentage of quarterly earnings announcements with concurrent management guidance over time. Consistent with Anilowski, Feng and Skinner (2007) and Rogers and Van Buskirk (2013), the percentage of earnings announcements with concurrent management forecasts increases overtime from less than 5% in 1999 to 43.27% in Prior research has attributed the increase of concurrent management guidance to the implementation of Reg FD because issuing bundled management guidance allows managers to communicate with analysts in public venues such as earnings announcement conference calls (Rogers and Van Buskirk 2013). Figure 7 Panel B presents a description of the mean values of USTAT for both Guidance and No-Guidance firms. We assign firm-quarters with concurrent management guidance to the Guidance group and firm-quarters with no concurrent management guidance to the No-Guidance group. The figures indicate that firm-quarters with concurrent management guidance are 23

26 associated with higher information content than firm-quarters with no management guidance. The mean and median values for the Guidance group are higher than those for the No-Guidance group in each of the 14 years. These findings indicate that guidance either adds to the informativeness of earnings announcements or is issued with more informative earnings announcements. However, the finding for stand-alone earnings announcements without guidance exhibits the same pattern of increases and decreases as those with guidance. The increase in guidance over the sample period is thus not the sole factor contributing to the increase in informativeness at earnings announcements. Analyst Forecasts Figure 8 Panel A presents a figure with the percentage of quarterly earnings announcements with concurrent analyst forecast over time. The percentage of concurrent analyst forecasts increases overtime from 30% in 1999 to 70% in Figure 8 Panel B presents a description of the mean values of USTAT for both Forecast and No-Forecast firms. We assign firm-quarters with at least one concurrent analyst forecast to the Forecast group and firmquarters with no concurrent analyst forecast to the No-Forecast group. The figures indicate that firm-quarters with concurrent analyst forecasts are associated with higher information content than firm-quarters with no concurrent analyst forecast. The mean and median values for the Forecast group are higher than those for the No-Forecast group in each of the 14 years. However, similar to our findings for guidance, the no forecast group also exhibits an increasing pattern suggesting the increased frequency of analyst forecasts is not the sole explanation for the pattern in USTAT we observe. 24

27 Financial Statement Line Items Figure 9 Panel A presents a figure with the percentage of financial statement items over time. The percentage of total financial statement items ( FS ) increases from 22.34% in 1999 to 54.15% in The percentages of income statement items ( IS ), balance sheet items ( BS ), and cash flow statement items ( SCF ) increase from 57.40%, 26.11%, and 0.41% in 1999 to 77.30%, 69.42%, and 28.50% in 2012 respectively. The findings are consistent with firms issuing more concurrent financial statement line items in earnings announcements over time. Figure 9 Panel B presents a description of the mean values of USTAT in each of the FS quartiles by calendar year. For each calendar year, we assign firm-quarters with the lowest FS to quartile 1 and firm-quarters with the highest FS to quartile 4. The figures indicate that firmquarters with higher FS have higher information content than firm-quarters with lower FS in each of the 14 years. The findings suggest greater disclosure of financial statement items is a factor in the increase in information content over time. However, given the lowest quartile exhibits a similar though less pronounced pattern of increase and decline, the findings suggest these disclosures are at best a partial explanation for the increase in USTAT. Furthermore, the percentage of financial statement items is only slightly increasing between 2006 and 2011, whereas the USTAT shows a significant decline and then increase in these years. In summary, Figures 7 through 9 are helpful in analyzing the bivariate relation between the individual explanatory variables and USTAT. We next assess the incremental association of these variables to each other and to control variables using multivariate analysis. Multivariate Regression Results Table 3 reports the descriptive statistics on variables included in the multivariate model, and Pearson and Spearman correlations for the variables in the regressions. The descriptive 25