What Drives the Increased Informativeness of Earnings Announcements Over Time? Daniel W. Collins Department of Accounting University of Iowa Iowa City, IA 52242 Email: daniel-collins@uiowa.edu Oliver Z. Li Department of Accountancy University of Notre Dame Notre Dame, IN 46556 Email: oli@nd.edu Hong Xie Department of Accountancy University of Illinois at Urbana-Champaign Champaign, IL 61820 Email: hongxie@uiuc.edu March 2005 Deleted: November 2004 Formatted: Centered We appreciate the helpful comments of Paul Beck, Brian Cloyd, Rajib Doogar, Neil Fargher, Susan Krische, Jeff Miller, James Myers, Theodore Sougiannis, Martin Wu, David Ziebart and workshop participants at University of Illinois at Urbana-Champaign. We gratefully acknowledge the contribution of Thomson Financial for providing analyst forecast data through the Institutional Brokers Estimate System. We also gratefully acknowledge the financial support of the Department of Accountancy at University of Illinois at Urbana-Champaign and the Department of Accounting at University of Iowa.
What Drives the Increased Informativeness of Earnings Announcements Over Time? ABSTRACT Landsman and Maydew (2002) find that the informativeness of earnings announcements, measured by abnormal trading volume and abnormal return volatility around earnings announcement dates, has increased in the past three decades. Francis, Schipper and Vincent (2002) investigate three potential explanations for the Landsman and Maydew finding and conclude that expanded concurrent disclosures in firms earnings announcement press releases explain the over-time increase in the informativeness of earnings announcements. This paper investigates whether the market s intensified reaction to Street earnings, initially documented in Bradshaw and Sloan (2002), offers a competing explanation for the Landsman and Maydew finding. We find that the market reaction to Street earnings, in terms of return, trading volume and return volatility, is larger than that to GAAP earnings on average. More importantly, the market reaction to Street earnings increases over time whereas the market reaction to GAAP earnings decreases over time. For a randomly selected sample of 2,421 quarterly earnings announcements, we find that the market s intensified reaction to Street earnings explains the over-time increase in the informativeness of earnings announcements. 1
What Drives the Increased Informativeness of Earnings Announcements Over Time? I. INTRODUCTION Landsman and Maydew (2002) document that the information content of quarterly earnings announcements, as measured by abnormal trading volume and abnormal return volatility around earnings announcement dates, has increased over the past three decades (see also Kross and Kim 2000 and Lo and Lys 2001). Landsman and Maydew, however, provide no explanation for what drives their results. In a follow-up study, Francis, Schipper and Vincent (2002) investigate three competing explanations for the Landsman and Maydew (2002) finding: (1) increases in the absolute amount of unexpected earnings at earnings announcements; (2) increases in the intensity of investors average reaction to unexpected earnings; and (3) an over-time expansion in the amount of concurrent information disclosed in firms earnings announcement press releases. They reject the first two explanations and conclude that expanded concurrent disclosures in firms earnings announcement press releases, especially the inclusion of detailed income statements, explain the increased informativeness of quarterly earnings announcements over time. 1 Thus, the implication of the Francis et al. findings is that it is the disclosure of more detailed GAAP-based numbers that has contributed to the increased informativeness of quarterly earnings announcements. In this paper, we investigate a competing explanation for the Landsman and Maydew (2002) finding that the information content of quarterly earnings announcements has increased over time. Specifically, we posit that the increased frequency and 1 We use the information content of earnings announcements and the informativeness of earnings announcements interchangeably throughout this paper. 2
magnitude of cases where Street and GAAP earnings differ and the increased reliance of the market on Street earnings for equity valuation provide a competing explanation to those investigated by Francis et al. (2002) for the Landsman and Maydew (2002) finding. In contrast to GAAP earnings, which are prepared under Generally Accepted Accounting Principles, street earnings typically exclude certain expenses deemed to be nonrecurring or non-cash such as restructuring charges, write-downs and asset impairments, one-time charges related to mergers and acquisitions, goodwill amortization and research and development expenditures. Importantly, Street earnings are generally announced concurrently with firms earnings announcements, but are disseminated through analyst estimate clearing houses like I/B/E/S, Zacks and First Call. 2 Thus, Street earnings announcements represent a potentially important correlated omitted variable in the Francis et al. (2002) study. Bradshaw and Sloan (2002) document that over the past 20 years there has been a dramatic increase in the frequency and magnitude of cases where Street earnings differ from GAAP earnings. 3 Moreover, they show that the market response to the Street earnings number has displaced GAAP earnings as a primary determinant of stock prices (p. 41) in that stock returns are more closely related to Street earnings than GAAP earnings. Similarly, Brown and Sivakumar (2003) find that Street earnings are more value relevant than operating income derived from 10-Q and 10-K filings to the SEC, a 2 The I/B/E/S Glossary (2000) states that I/B/E/S strives to report actual earnings as soon as they are released into the market place. For the US & Canada, earnings reports are culled directly from the newswires, adjusted for comparability with estimates and reported to subscribers via the Intra Day Surprise Report, which is delivered five times each trading day (p. 7). 3 As explained below, we use I/B/E/S reported actual earnings as a measure of Street earnings, following Bradshaw and Sloan (2002), Brown and Sivakumar (2003) and Doyle, Lundholm and Soliman (2003). Street earnings are disseminated to the market place concurrently with firms earnings announcements. It should be noted that Street earnings are not the same as pro-forma or other non-gaap earnings that firms often report in their earnings announcement press releases (see Bhattacharya et al. 2003). 3
GAAP-based measure, in terms of (1) ability to predict future earnings, (2) association of earnings levels with stock price levels, and (3) correlation of earnings surprises with abnormal stock returns. One interpretation of these findings is that the increased emphasis on Street earnings represents attempts by managers and analysts to remove transitory components from GAAP earnings in order to make Street earnings an improved measure for determining future cash flows and hence firm value (Bradshaw and Sloan 2002, p. 42). If it is Street earnings rather than GAAP earnings or GAAP-based operating income that are driving the trading activity and share price movements around earnings announcement dates, then it is important to include Street earnings when seeking to explain the increased informativeness of earnings announcements over time as documented in Landsman and Maydew (2002) and other studies. The primary objective of this paper is to investigate whether the market s increased reliance on Street earnings is what drives the increased informativeness of earnings announcements over time. Following Landsman and Maydew (2002), we measure the information content of quarterly earnings announcements using abnormal trading volume and abnormal return volatility around quarterly earnings announcement dates. Similar to Bradshaw and Sloan (2002) and other recent studies, we use I/B/E/S reported actual earnings as our measure of Street earnings and Compustat reported earnings as a measure of GAAP earnings. We initially examine a large sample of 115,450 firm-quarter observations, spanning 16 years during 1985-2000. We begin by successfully replicating the Landsman and Maydew (2002) finding using our full sample that the information content of quarterly earnings announcements, as measured by abnormal trading volume and abnormal return volatility, has increased in the past two decades. 4
We then turn to investigating potential explanations for the over-time increase in the information content of earnings announcements. We start with three potential explanations investigated by Francis et al. (2002). First, we examine changes in the absolute amount of unexpected earnings over time. Consistent with Francis et al., we find no evidence that the increased informativeness of earnings announcements is due to increases in the absolute amount of unexpected earnings, regardless of whether unexpected earnings are measured based on GAAP earnings (the difference between Compustat reported earnings and the most recent median analyst forecast) or based on Street earnings (the difference between Street earnings as reported by I/B/E/S and the most recent median analyst forecast). Second, we investigate changes in earnings response coefficients (ERCs) over time. Consistent with the findings of Francis et al., we find that ERCs with respect to GAAP earnings surprises do not increase over time in our sample. However, we find that ERCs with respect to Street earnings surprises steadily increase in our sample period. 4 Thus, while we concur with Francis et al. s conclusion that the increased informativeness of earnings announcements cannot be attributable to the market s intensified reaction to GAAP earnings surprises, we cannot reject the possibility that the over-time increase in the information content of earnings announcements is attributable to investors intensified reaction to Street earnings. We then link excess trading volume and abnormal return volatility to GAAP as well as Street earnings surprises to examine (1) how the market reaction to GAAP and Street earnings surprises, respectively, change over time and (2) whether investors 4 As discussed in more detail below, both of Francis et al. s unexpected earnings, SRW-UE and Analyst- UE, are equivalent to our GAAP earnings surprises. They do not examine Street earnings in their study. 5
intensified reaction to Street earnings surprises explains the over-time increase in the informativeness of earnings announcements. We find that volume response coefficients (VRCs) and return volatility response coefficients (RVRCs) with respect to Street earnings surprises are, on average, larger than their counterparts with respect to GAAP earnings surprises. In addition, VRCs and RVRCs with respect to Street earnings surprises increase over time whereas those with respect to GAAP earnings decline over time. Thus, we have three distinct, albeit related, pieces of evidence ERCs, VRCs and RVRCs all suggesting that investors have come to rely more on Street earnings than on GAAP earnings over time. However, the market s intensified reaction to Street earnings surprises does not fully explain the over-time increase in the information content of earnings announcement for our full sample. Our results, therefore, suggest that Francis et al. s expanded disclosure hypothesis at least partially explains the over-time increase in Deleted: explain the informativeness of earnings announcements. Finally, we investigate whether investors intensified reaction to Street earnings or Francis et al. s expanded disclosure or both explain the increased informativeness of earnings announcements over time using a randomly selected sample. It is important to investigate our intensified reaction to Street earnings hypothesis together with Francis et al. s expanded disclosure explanation because expanded disclosure, such as the provision of detailed income statements, could potentially proxy for Street earnings when investors back out non-recurring items to derive more permanent core earnings. We randomly select 58 firms from our full sample and form a representative sample of 2,421 firmquarter observations. We hand collect earnings announcement press releases for each observation. For this random sample, we demonstrate that investors intensified reaction 6
to Street earnings surprises explains the over-time increase in the information content of earnings announcements. Consistent with Francis et al., we find that expanded concurrent disclosure in firms earnings announcement press releases also explains the over-time increase in the information content of earnings announcements after controlling for Street earnings. Contrary to Francis et al., however, we do not find that the over-time increase in the market reaction to earnings announcements is attributable to the disclosure of detailed income statements. Rather, we find that the disclosure of detailed statement of cash flow data is the primary concurrent disclosure that contributes to the increased informativeness of earnings announcements over time. We conjecture that the market s increased reaction to statement of cash flow data is due to an over-time decrease in firm profitability (e.g., Fama and French 2001) or an over-time increase in losses or frequency of delisting from the stock exchange (e.g., Hayn 1995 and Collins, Pincus and Xie 1999) that makes cash flow statement information more relevant to investors assessment of firms performance and future prospects. This paper contributes to the literature by identifying the market s intensified reaction to Street earnings as a competing explanation for the Landsman and Maydew (2002) finding that the information content of earnings announcements has increased over time. Our findings are important because the implications are different from those of Francis et al. (2002). While they suggest that the expanded disclosure of GAAP-based accounting information explains the over-time increase in the informativeness of earnings announcements, we find that it is the market s intensified reaction to non-gaap based Street earnings that explains the increased informativeness of earnings announcements over time. If the purpose of expanded disclosure is to help investors better assess a firm s Deleted: is Deleted: of our finding Deleted: s Deleted: at Deleted: more detailed Deleted:, not necessarily the expanded disclosure per se, Deleted: After all, 7
future prospects, then our findings suggest that Street earnings are a better measure of a firm s future prospects than GAAP earnings (e.g., Bradshaw and Sloan 2002, Brown and Sivakumar 2003). Thus, this finding suggests that traditional GAAP-based earnings Deleted: current performance and Deleted:. Deleted: O Deleted: s Deleted: current performance and measures may, in fact, include components that the market discounts or ignores in setting current prices and that Street earnings does a better job of identifying those components that the market deems more sustainable and, therefore, more relevant for valuation purposes. In addition, we contribute to the literature by providing volume-based evidence that investors rely more on Street earnings and react increasingly more intensely to Street earnings over time. Prior studies on GAAP versus Street earnings typically use returnbased measures of value relevance or information content (e.g., Bradshaw and Sloan Deleted: and that the market s intensified reaction to Street earnings is an important factor that explains the overtime increase in the informativeness of earnings announcements. [Dan, please critically revise this paragraph. We may not be able to make the above assertion (it is Street earnings not expanded disclosure per se) because we did not find that Street earnings subsume GAAP earnings. In any event, we need to explain why our research question and finding are important.] 2002, Brown and Sivakumar 2003, Doyle, Lundhom and Soliman 2003 and Bhattacharya et al. 2003). Volume-based evidence is not redundant to return-based evidence. Rather, it is an important complement because volume-based metrics and return-based metrics capture different aspects of the market reaction. Return-based metrics capture the average reaction of all investors whereas volume-based metrics capture the sum of all investors reaction (e.g., Beaver 1968, Bamber 1986 and 1987, Kim and Verrecchia 1991). Combining prior return-based studies with our volume-based findings, we conclude that investors, not only on average but also individually, have come to rely more on Street earnings than on GAAP earnings. The remainder of the paper is organized as follows. Section II describes data, variable measurement and model specification. Section III describes our sample and presents major findings. We conclude in Section IV. 8
II. DATA AND METHODOLOGY Sample Selection Data used in this study are obtained from the 2000 Compustat (quarterly), 2000 I/B/E/S, and 2000 CRSP (daily) tapes. We identify 155,559 firm-quarter observations between 1985 and 2000 where (1) quarterly earnings announcement dates and GAAP earnings (defined below) are available on the Compustat tape, (2) there are at least 50 but no more than 182 days between two adjacent quarterly announcements, (3) Street earnings (defined below) are available on the I/B/E/S tape. We further require that each firm-quarter observation has non-zero daily trading volume data on the 2000 CRSP tape for 91 days between day 45 to day 45 relative to the quarterly earnings announcement date (day = 0). This reduces the sample to 122,071 firm-quarter observations. To minimize the undue impact of extreme observations, we delete the top and bottom 0.5% of all variables in our regression analyses (described in more detail below). Our final sample contains 115,450 firm-quarter observations, covering 16 years between 1985 and 2000. Variable Measurement We measure GAAP earnings per share (E_GP it ) as income before extraordinary items available for common (Compustat quarterly item #25) divided by total number of shares outstanding (Compustat quarterly item #61) adjusted for stock splits and dividends (Compustat quarterly item #17). 5 Following Bradshaw and Sloan (2002) and other studies in the prior literature, we use I/B/E/S reported actual earnings as our measure of Street earnings (E_ST it ). Forecast errors are defined as the differences between actual earnings 5 The subscript t in E_GP it should be interpreted as indicating year-quarter since quarter subscript is omitted for ease of exposition. 9
and the most recent median consensus forecasts before earnings announcement dates. We, thus, have two measures of forecast errors, one based on GAAP earnings (FE_GP it ) and the other based on Street earnings (FE_ST it ). Both earnings measures and forecast error measures are scaled by split-adjusted stock price at the beginning of the quarter. We adopt two measures of information content of quarterly earnings announcements, excess trading volume and abnormal return volatility. We measure excess trading volume similar to Utama and Creedy (1997). Specifically, we define the eleven days from day 5 to day +5 relative to earnings announcement date (day = 0), as the announcement period, and the 80 days from day 45 to day 6 and from day +6 to day +45 as the non-announcement period. Following Utama and Creedy (1997), we measure excess trading volume over a 3-day window around the earnings announcement date, EXVOL it, as follows: VOL d [ 1, + 1] it, d EXVOL it = 1, 3NVOL it where VOL it,d is daily trading volume in millions of share on day d and NVOL it = d [ 45, 6] [ + 6, + 45] 80 VOL it, d is the normal daily trading volume during the non-announcement period. Thus, our excess trading volume measure gauges the percentage of changes in daily trading volume during the announcement period relative to the non-announcement period. 6 Similar to Landsman and Maydew (2002), we define abnormal return volatility, AVAR it, as follows: 6 We also examine a 7-day window, [-1, +5], around earnings announcement date. The results from the 7- day window are qualitatively identical to those reported in the paper for the 3-day window. 10
AVAR it = 2 u d [ 1, + 1] it, d, 2 3σ it where u it,d = 2 ˆ βˆ is abnormal return on day d, and σ is the variance RET it, d α it it MKTt, d it of abnormal returns over the 80-day non-announcement period. We estimate αˆ it and βˆ it using the market model over the 80-day non-announcement period: RET it,d = α it + β it MKT t,d + u it,d, (1) where RET it,d and MKT t,d are daily stock returns and value-weighted market returns, respectively. Empirical Models We first replicate the main finding of Landsman and Maydew (2002) that the information content of quarterly earnings announcement has increased over time using the following regression equations: EXVOL it = a 0 + a 1 T t + a 2 D it + a 3 T t D it + e it, (2) AVAR it = α 0 + α 1 T t + α 2 D it + α 3 T t D it + ε it, (3) where EXVOL it and AVAR it are excess trading volume and abnormal return volatility over the 3-day window, T t = log(year 1984) is a time trend where Year is a firmquarter observation s fiscal year (from 1985 to 2000), and D it is a dummy variable for December fiscal year-end firms. A significantly positive coefficient on T t (i.e., a 1 > 0 and α 1 > 0) replicates Landsman and Maydew (2002). Next, we relate our two measures of information content, EXVOL it and AVAR it, to both GAAP and Street earnings to investigate (1) how the market s volume response and return volatility response to GAAP as well as Street earnings change over time and (2) whether the inclusion of Street earnings as an explanatory variable in our models 11
eliminates a positive time trend on the information content of earnings announcements. We employ the following regression equations for our investigation: where: EXVOL it = b 0 + b 1 T t + b 2 ABSFE_GP it + b 3 ABSFE_GPT it + b 4 ABSFE_ST it + b 5 ABSFE_STT it + b 6 ABSARET it + b 7 LSIZE it + b 8 LPRICE it + b 9 LADVOL it + b 10 RISK it + e it, (4) AVAR it = β 0 + β 1 T t + β 2 ABSFE_GP it + β 3 ABSFE_GPT it + β 4 ABSFE_ST it + β 5 ABSFE_STT it + β 6 LSIZE it + β 7 LPRICE it + β 8 LADVOL it + β 9 RISK it + ε it, (5) EXVOL it = excess trading volume during [ 1, +1] as defined before, AVAR it = abnormal return volatility during [ 1, +1] as defined before, T t = log(year 1984) as defined before, 7 ABSFE_GP it = absolute value of forecast error based on GAAP earnings, FE_GP it, as defined before, ABSFE_GPT it = ABSFE_GP it T t, ABSFE_ST it = absolute value of forecast error based on Street earnings, FE_ST it, as defined before, ABSFE_STT it = ABSFE_ST it T t, ABSARET it = absolute value of the average daily abnormal returns during [ 1, +1]. The daily abnormal returns are estimated as the market model residuals, 31 ( RET3it αˆ it βˆ itmkt3 ), where RET3 t it and MKT3 t are the 3-day cumulative stock returns and value-weighted market returns during [ 1, +1]. αˆ and it βˆ are estimated using the market it model, RET it,d = α it + β it MKT t,d + u it,d, over the 80-day nonannouncement period from day 45 to day 6 and from day +6 to day +45, where RET it,d and MKT t,d are daily stock returns and valueweighted market returns, respectively. LSIZE it = natural logarithm of average market capitalization, in millions of dollar, in the 80-day non-announcement period, LPRICE it = natural logarithm of average daily stock price in the 80-day nonannouncement period, LDVOL it = natural logarithm of average daily dollar values of shares traded, in millions of dollar, in the 80-day non-announcement period, RISK it = firm-specific total risk estimated as σ it, where σ and it σ are Mt σ Mt standard deviations of RET it,d and MKT t,d over the 80-day nonannouncement period. 7 We take a natural logarithm transformation to allow for the possibility that the time trend is non-linear (see Bradshaw and Sloan 2002, Fig. 2). 12
Our first variable of interest is ABSFE_STT it, which is the interaction between ABSFE_ST it and the time trend. This interaction term captures the over-time change in the coefficient on the absolute value of Street-based forecast errors (ABSFE_ST it ). Bradshaw and Sloan (2002) find that investors return response to Street earnings (i.e., ERC) increases over time whereas that to GAAP earnings remains stagnate over time (p. 54). Based on the Bradshaw and Sloan finding, we expect that the volume response coefficients (VRCs) and return volatility response coefficients (RVRCs) with respect to Street earnings surprises will increase over time as well, i.e., we predict b 5 > 0 and β 5 > 0. Our second variable of interest is the time trend, T t, which captures the over-time increase in the information content of earnings announcements. If the market s intensified reaction to Street earnings fully explains the over-time increase in the informativeness of earnings announcements, we expect that the incorporation of Street earnings surprises in equations (4) and (5) eliminates the significant coefficient on T t, i.e., b 1 = 0 and β 1 = 0. We include several control variables in equations (4) and (5) based on the extant literature. Kim and Verrecchia (1991) suggest that excess trading volume associated with a public announcement is positively related to a measure of absolute price change over the event window. Similar to Atiase and Bamber (1994) and Utama and Creedy (1997), we use the absolute value of the average daily abnormal returns over the 3-day event window, ABSARET it, as our measure of absolute price change. We expect a positive coefficient on ABSARET it, i.e., b 6 > 0. Note, ABSARET it is not included as a control variable in equation (5) because AVAR it (abnormal return volatility during the 3-day event window) and ABSARET it (absolute value of abnormal returns during the 3-day event window) are potentially highly correlated. 13
Prior research finds that firm size is another variable that affects trading volume around earnings announcements. Bamber (1986 and 1987) finds that excess trading volume around earnings announcements is negatively related to firm size because larger firms typically release more information before earnings announcements, which reduces volume reactions to the earnings announcement itself. Similar to Utama and Creedy (1997), we use the natural logarithm of average market capitalization (LSIZE it ) during the 80-day non-announcement period as our control for size. We expect a negative coefficient on LSIZE it (i.e., b 7 < 0 and β 6 < 0). Prior studies show that transaction costs are negatively related to trading volume. Following Utama and Cready (1997), we use the natural logarithm of average daily stock price (LPRICE it ) and the natural logarithm of average daily dollar values of shares traded (LADVOL it ) in the non-event window to control for transaction costs. Since larger LPRICE it and LADOVL it indicate lower transaction costs, we expect positive coefficients on LPRICE it and LADVOL it (i.e., b 8 > 0, b 9 > 0, β 7 > 0 and β 8 > 0). Finally, Michaely and Vila (1995, 1996) find that trading volume around exdividend days is negatively related to risk. They argue that the higher a stock s risk, the smaller the position traders are willing to assume in the ex-dividend day trading. We use a firm s return volatility scaled by the market return volatility over the 80-day nonannouncement period as our measure of risk (RISK it ). Consistent with Mickaely and Vila (1995, 1996), we expect the coefficient on RISK it to be negative (i.e., b 10 < 0 and β 9 <0). 14
III. RESULTS Descriptive Statistics Table 1 presents descriptive statistics for our final sample of 115,450 firm-quarter observations. The mean and median EXVOL it are 0.4180 and 0.1589, respectively, suggesting that daily trading volume during the 3-day announcement window exceed the average daily trading volume during the non-announcement period by 41.80% (15.89%). The mean (median) of abnormal return volatility, AVAR it, is 2.0085 (1.0858). Both Deleted:, Deleted: our first measure of information content of earnings announcements, Deleted: the mean (median) Deleted:, our second measure of information content of earnings announcements, values exceed one, suggesting that the return volatility during the 3-day announcement window is larger than normal return volatility during the non-announcement period. 8 In short, both excess trading volume and abnormal return volatility measures of information content indicate more active trading during the earnings announcement window than the non-announcement period, which is consistent with findings in prior studies (e.g., Beaver Deleted: many 1968; Bamber 1986 and 1987; Utama and Creedy 1997). [Insert Table 1 here] The mean and median of GAAP-based forecast errors (FE_GP it ) are -0.0022 and 0.0000, respectively, whereas the mean and median Street-based forecast errors (FE_ST) are -0.0010 and 0.0000. Note that forecast errors based on GAAP or Street earnings are both negative on average, consistent with prior findings that analysts forecasts, on average, are optimistic. In addition, the mean forecast error based on Street earnings is less negative than its GAAP earnings counterpart, suggesting that Street earnings are larger than GAAP earnings. This finding is consistent with Street earnings excluding expense items included in GAAP earnings (Doyle, Lundholm and Soliman 2003). 8 AVAR it, by construction, is a non-negative number. When AVAR it is less (greater) than one, it indicates smaller (larger) than normal return volatility. 15
Untabulated results suggest that the differences between Street and GAAP earnings increase substantially after 1990, consistent with Bradshaw and Sloan (2002). Confirmation of an Over-Time Increase in the Informativeness of Earnings Announcements We replicate Landsman and Maydew (2002) on our sample using excess trading volume and abnormal return volatility measures and report the results in Panel A, Table Deleted: Deleted: our sample and Deleted:. We estimate equation (2) 2. Since we estimate equation (2) and other equations in the cross-section and over time, we report the t-statistics using the Huber-White maximum likelihood estimation procedure to correct for heteroscedasticity and serial correlation among earnings surprises (see for example, Garvey and Milbourn 2000 and Gleason and Lee 2003). The coefficient Deleted: potentially present in the data on the time trend, T t, is significantly positive (0.0863, t = 10.06) when equation (2) is estimated using the full sample, suggesting that excess trading volume around earnings announcement dates increases over time. The coefficient on T t is also significantly positive when equation (2) is estimated separately for each quarter. Similar to Landsman and Maydew (2002), we also find that the positive time trend for December fiscal yearend firms is somewhat smaller since the incremental coefficient on T t for December fiscal year-end firms are often significantly negative. The F-statistic, however, suggests that the total coefficient on T t (a 1 + a 3 ) for December fiscal year-end firms remains significantly positive for all four quarters combined and for each individual quarter. [Insert Table 2 here] Panel B, Table 2, presents our findings from estimating equation (3) that tests for a time trend on abnormal return volatility. Consistent with Landsman and Maydew, the coefficient on T t is again significantly positive for all quarters and for each individual quarter. This suggests that, like excess trading volume, abnormal return volatility around 16
earnings announcements has increased over time as well. Taken together, the results in Table 2 confirm the main findings in Landsman and Maydew (2002) that the information content of quarterly earnings announcements, measured by excess trading volume and abnormal return volatility, has increased over time in the past two decades. Tests of Competing Explanations for an Over-Time Increase in the Informativeness of Earnings Announcements Francis, Schipper and Vincent (2002) explore three potential explanations for the increased informativeness of earnings announcements over time. These explanations are (1) increases in the absolute amount of unexpected earnings at earnings announcements; (2) increases in the intensity of investors average reaction to unexpected earnings; and (3) an over-time expansion in the amount of concurrent information disclosed in firms earnings announcement press releases. They measure unexpected earnings as the difference between quarterly earnings per share before extraordinary items (Compustat item #9) and the most recent analyst forecast from Zacks Investment Research database. 9 Their unexpected earnings, therefore, are equivalent to our GAAP-based forecast errors, FE_GP it. Francis et al. find that the absolute value of their GAAP-based unexpected earnings actually decreases, rather than increases, over time (Table 3). Moreover, earnings response coefficients to their GAAP-based unexpected earnings also decrease over time (Table 4). Based on these findings, Francis et al. reject their first two potential explanations for the increased information content of earnings announcements over time. They, however, do not examine Street-based unexpected earnings and ERCs with respect to Street-based unexpected earnings. 9 They also use the seasonal random-walk difference in quarterly earnings per share before extraordinary items (Compustat item #9) as a second measure of unexpected earnings, with qualitatively unchanged results. 17
Changes in the Absolute Amount of Unexpected Earnings and in the Intensity of Investors Return Response to GAAP versus Street Earnings Surprises As a first step in our investigation of an alternative explanation for the increased Deleted: search for informativeness of earnings announcements over time, we examine the absolute value of Street-based unexpected earnings and earnings response coefficients to Street-based unexpected earnings. Column 3 of Table 3 shows the annual mean (median) absolute value of our GAAP-based forecast errors, ABSFE_GP it, over the sample period 1985-2000. The rows labeled Trend report slope coefficient estimates and p-values from regressing the annual mean or median ABSFE_GP it on the time trend, T t. The time trend coefficient of -0.0002 (p-value = 0.3109) indicates that the mean ABSFE_GP remains relatively unchanged over the sample period. The median ABSFE_GP, on the other hand, decreases slightly over the sample period (slope coefficient = -0.0003, p-value = 0.0001). In contrast, both the mean and median absolute Street-based earnings forecast errors, ABSFE_ST it, decline steadily over time (Column 4). Columns 3 and 4, thus, provide no evidence that the absolute amount of unexpected earnings, either based on GAAP or Street earnings, has increased over time. Thus, our finding corroborates the finding in Francis et al. (2002) that the over-time increase in information content of earnings announcements cannot be attributed to an over-time increase in the absolute value of unexpected earnings, either GAAP_based of Steet-based. [Insert Table 3 here] Columns 5 and 6, Table 3, present test results for whether the intensity of the market reaction to a given amount of unexpected GAAP earnings surprises and Street earnings surprises, respectively, has intensified over time. Specifically, we regress signed average daily abnormal returns during the 3-day announcement window, ARET it, on 18
signed forecast errors, FE_GP it or FE_ST it, each quarter in a year to estimate earnings response coefficients by year. We report the mean of four quarterly ERCs each year. The mean ERCs based on GAAP-earnings remain relatively unchanged throughout the 1985-2000 sample period. The slope coefficient on the time trend is 0.0057, which is not significantly different from zero (p-value = 0.4677). In sharp contrast, the mean ERCs based on Street-earnings increase steadily over our sample period. The slope coefficient on the time trend is 0.1786 and is highly significant (p-value = 0.0001). Our finding that ERCs with respect to Street earnings increase over time is consistent with Bradshaw and Sloan (2002, p.54) who also find that ERCs with respect to Street earnings increase over time. 10 The results reported in Columns 5 and 6 suggest that although the intensity of investors reaction to GAAP-based earnings surprises remains relatively unchanged over our sample period, the intensity of investors reaction to Street-based earnings surprises increases steadily over time. Therefore, the increased intensity of investors reaction to Street-based unexpected earnings provides a potential competing explanation for the Deleted: is one increased informativeness of earnings announcements over time. Changes in the Intensity of Investors Volume and Return Volatility Responses to GAAP versus Street Earnings Surprises In this section, we examine (1) how the volume response coefficient (VRC) and return volatility response coefficient (RVRC) with respect to GAAP and Street earnings surprises have change over time; and (2) whether the market s intensified reaction to Deleted:, respectively, Street earnings surprises explains the over-time increase in the information content of 10 Our ERCs with respect to Street earnings are much smaller in magnitude compared to those reported in Bradshaw and Sloan (2002). This is because our signed abnormal return, ARET it, is the average daily abnormal returns during the 3-day announcement window whereas their return metric is quarterly raw returns cumulated from two days after the previous quarterly earnings announcement until one day after the current quarterly earnings announcement. Our ERCs are comparable to those reported in Francis et al. (2002, Table 4) in magnitude since Francis et al. also use daily abnormal returns. 19
earnings announcements. We do so by linking our two measures of information content of earnings announcements, EXVOL it and AVAR it, to GAAP as well as Street earnings surprises in equations (4) and (5). Panel A, Table 4, reports our findings from estimating equation (4) and several variations of equation (4). Model 1 shows the coefficient estimates when excess trading volume is regressed on GAAP-based earnings surprises and five control variables. The coefficient on ABSFE_GP is positive and significant (0.8683, t = 2.90). This is consistent with findings in prior studies that excess trading volume around earnings announcement dates is positively related to the magnitude of earnings surprises (e.g., Bamber 1986, 1987). The coefficients on control variables are generally consistent with our predictions based on the prior literature. 11 [Insert Table 4 here] The results from regressing excess trading volume on Street-based earnings surprises and five control variables are presented in Model 2, Panel A of Table 4. The coefficient on ABSFE_ST is 2.0058 (t = 4.05), which is considerably larger than the coefficient on ABSFE_GP in Model 1. When both GAAP-based and Street-based earnings surprises are included in Model 3, the coefficient on Street-based earnings surprises remain significantly positive (1.8012, t = 3.14), whereas the coefficient on GAAP-based earnings surprises becomes insignificant. This suggests that the volume Deleted: ly different from zero. response at earnings announcement dates, similar to the return response documented in 11 For example, the coefficients on ABSARET are highly significantly positive, consistent with the theoretical prediction of Kim and Verrecchia (1991) and empirical findings of Atiase and Bamber (1994) and Utama and Creedy (1997) that abnormal trading volume is positively related to the magnitude of price changes. Consistent with Bamber (1986, 1987), the coefficient on LSIZE is significantly negative. The coefficient on LPRICE, an inverse proxy for transaction costs, is significantly positive as expected. However, the coefficient on LDVOL, also an inverse proxy for transaction costs, is significantly negative. Utama and Creedy (1997) predict a positive coefficient on LDVOL but find the coefficient is insignificantly different from zero. Finally, consistent with Michaely and Vila (1995), the coefficient on RISK is significantly negative. 20
Bradshaw and Sloan (2002), is more closely associated with Street earnings surprises than with GAAP earnings surprises. We present evidence on over-time changes in trading volume response to GAAP earnings surprises versus Street earnings surprises in Model 4. The coefficient on ABSFE_GPT is significantly negative (-1.9351, t = -3.23), suggesting that the association between excess trading volume and GAAP earnings surprises decreases over time. On the other hand, the coefficient on ABSFE_STT is significantly positive (2.3730, t = 3.42), Formatted: Underline Formatted: Underline indicating that the volume response to Street earnings surprises increases over time. Our finding that the market s volume response at earnings announcement dates to Street (GAAP) earnings surprises increases (decreases) over time is consistent with the returnbased findings in Bradshaw and Sloan (2002, Table 1 and Figure 2). However, the coefficient on the time trend, T, remains significantly positive (0.0117, t = 2.07). This result suggests that the market s intensified reaction to Street earnings surprises does not fully explain the over-time increase in the informativeness of earnings announcements for our full sample. We present our findings from estimating equation (5) for abnormal return volatility in Panel B, Table 4. Consistent with the excess trading volume results, we find that abnormal return volatility at earnings announcement dates, on average, is also more closely related to Street earnings surprises than to GAAP earnings surprises (Model 3). In addition, the association of abnormal return volatility with Street earnings surprises increases over time whereas its association with GAAP earnings surprises becomes weaker over time. Once again, the coefficient on T remains significantly positive (0.1610, 21
t = 9.95). Thus, our abnormal return volatility results are consistent with our excess trading volume results. To summarize, findings in Table 4 suggest that the market reaction to Street (GAAP) earnings surprises, both in terms of volume and return volatility, increases (decreases) over time. The market s intensified reaction to Street earnings surprises, however, does not fully explain the over-time increase in the information content of earnings announcements for our full sample. This suggests that the expanded concurrent disclosure investigated in Francis et al. (2002) could at least partially explain the increased informativeness of earnings announcements over time. We investigate our Deleted: s Deleted: further intensified reaction to Street earnings hypothesis and Francis et al. s expanded disclosure explanation in the next section. Expanded Concurrent Disclosures versus Increased Intensity of Investors Reaction to Street Earnings Surprises The third potential explanation for increased informativeness of earnings announcements investigated by Francis et al. (2002) is the expanded concurrent disclosures made in firms earnings announcement press releases. Francis et al. randomly select 30 firms from their sample and hand collect and code 2,190 earnings announcement press releases during 1980-1999. They find that there is a significant increase in the amount of concurrent information disclosed along with bottom line earnings during their sample period. They conclude that these concurrent disclosures, especially the inclusion of detailed income statements, explain the increased informativeness of earnings announcements over time. In this section, we investigate whether it is the expanded disclosures or intensified reaction to Street earnings, or both, that explain the increased informativeness of earnings 22
announcements over time. We compare expanded disclosures with Street earnings because detailed income statements may simply proxy for Street earnings. That is, with detailed line-items disclosed concurrently with bottom-line GAAP earnings, investors can back out non-recurring items from bottom line earnings to derive a measure of core earnings, which is likely correlated with Street earnings. We randomly select 58 firms from our full sample. 12 These 58 firms generate a sample of 2,421 firm-quarter observations (hereafter, referred to as the Disclosure Deleted: d sample ). We then hand collect earnings announcement press releases for each of these 2,421 observations from the Lexis-Nexis newswire archives. We read and code each press release for the following four variables: ABSUS it = absolute value of firm i s unexpected sales, measured as reported sales for year-quarter t less sales for same quarter of last year, scaled by the market value of equity at the end of previous quarter, IS it = 1 if firm i s press release for year-quarter t contains a detailed quarterly income statement, 0 otherwise, BS it = 1 if firm i s press release for year-quarter t contains a detailed quarterly balance sheet, 0 otherwise, SCF it = 1 if firm i s press release for year-quarter t contains a detailed quarterly statement of cash flow, 0 otherwise. 13 The percentages of press releases including a detailed income statement, balance sheet or statement of cash flow for our disclosure sample are 48.5%, 27.9% and 4.5% (results untabulated). These percentages are somewhat higher than their counterparts reported in Francis et al., due in part to our sample period being more recent than theirs. Consistent with Francis et al., we also find that the percentage of press releases Deleted: s 12 To increase the chance of finding their earnings announcement press releases, our 58 firms are randomly selected from the top 50% of our full sample in terms of market capitalization. An average firm in our random sample, therefore, is larger in size than an average firm in our full sample. 13 Francis, Schipper and Vincent (2002, p. 539) code five additional concurrent disclosure variables. [Hong we should probably state what these additional five disclosure variables are in this footnote]. These five variables are all insignificant in explaining increased informativeness of earnings announcements (p. 540). Moreover, coding these five variables involves a high degree of subjectivity. Consequently, we do not include these five variables in our investigation. Deleted: additional 23
containing a detailed income statement, balance sheet or statement of cash flow increase over time. We first confirm that the information content of earnings announcements for the disclosure sample also increases over time as was the case for our full sample. We regress EXVOL and AVAR, respectively, on the time trend, T. The slope coefficients are 0.09132 (p-value = 0.0001) for EXVOL and 0.15810 (p-value = 0.0652) for AVAR (results untabulated). 14 We then expand equations (4) and (5) by including the four concurrent disclosure variables and their respective interactions with the time trend. Table 5 reports our findings. Panel A, Table 5, presents our excess trading volume results. As in Francis et al., we are interested in whether the coefficient on the time trend, T, becomes insignificant when we include concurrent disclosure variables and/or Street earnings as independent variables. Our Model 1 is equivalent to Model 3, Table 7, in Francis et al. (2002) and tests whether the incorporation of concurrent disclosure variables eliminates the significance on T. The coefficient on T is indeed insignificant (0.0436, t = 1.12). Untabulated results reveal that the coefficient on T remains significantly positive (0.0569, t = 2.18) when the four concurrent disclosure variables (ABSUS, IS, BS and SCF) and their interaction terms with time trend (ABSUST, IST, BST and SCFT) are excluded from Model 1. Therefore, incorporating these concurrent disclosure variables and their interaction terms eliminates the significantly positive time trend, consistent with Francis et al. [Insert Table 5 here] 14 Thus, consistent with the results reported previously for our full sample, both excess trading volume and abnormal return volatility appear to increase over time for the disclosure sample. 24
However, it is important to note that the coefficients on our concurrent disclosure variables and the interaction terms are quite different from those reported in Francis et al. For example, Francis et al. find a significantly negative coefficient on the dummy variable, IS, which indicates the inclusion of a detailed income statement in the earnings announcement press release (-0.015, p-value = 0.001) and a significantly positive coefficient on the interaction of IS with the time trend, IST, (0.002, p-value = 0.001). We find that the coefficients on these two variables are both insignificant in Model 1 in our sample. A closer examination reveals a relatively high correlation between IS or BS and SCF in our sample. Specifically, firms that disclose a detailed statement of cash flow in their earnings announcement press releases (i.e., SCF = 1) almost invariably also disclose a detailed income statement (IS =1) and a detailed balance sheet (BS =1). To mitigate the potential multicolinearity problem between IS, BS and SCF, we re-estimate Model 1, Panel A of Table 5, excluding BS, SCF, BST and SCFT. Untabulated results suggest that the coefficient on IS becomes significantly positive (0.1989, p-value = 0.075) although the coefficient on IST remains insignificant. Thus, in contrast to Francis et al. who find the coefficient on IS increases over time, we do not find a significantly positive time trend on detailed income statement data. Thus, the concurrent disclosure of detailed income statements does not explain the increased informativeness of earnings announcements over time for our disclosure sample. Model 1, Panel A of Table 5, reveals another important difference between our findings and those of Francis et al. We find a significantly negative coefficient on SCF (- 1.1627, t = -2.65) and a significantly positive coefficient on SCFT (0.4791, t = 2.71), 25