Expected Reporting Speeds: Information in Firms’ Relative and Aggregate Behavior

Ed deHaan⇤ Stanford University Graduate School of Business Eric C. So Massachusetts Institute of Technology Sloan School of Management Prepared for Berkeley-Stanford Workshop November 2015 Extremely early draft. Please do not cite or distribute. Abstract Existing studies show that intra-firm variation in earnings announcement speeds correlates with earnings news, such that firms are more likely to decrease (increase) the speed of announcing bad (good) earnings news relative to their fiscal period end. We examine firms’ “expected abnormal reporting speeds” not just in relation to their own past behaviors, but also in relation to other firms announcing in the same month. Relative expected abnormal reporting speeds predict future earnings realizations and stock returns not only for firms that deviate from their own historic reporting patterns, but also for the supermajority of firms that elect not to alter their reporting speeds. These results suggest that a firm’s choice not to change its reporting speeds is informative when viewed relative to the behavior of other firms. Further, aggregating firms’ month-end expected abnormal reporting speeds indicates that relative changes do not offset one another, but rather give rise to predictable changes in aggregate earnings news, treasury spreads, and market uncertainty. JEL Classifications: G10, G11, G12, G14, M40, M41 ⇤ We thank Wall Street Horizon for generously providing data on expected earnings announcement dates. Contact emails: Ed deHaan, [email protected] and Eric So, [email protected]

Expected Reporting Speeds

1

1. Introduction At least as far back as Kross (1981) and Givoly and Palmon (1982), prior research documents a positive link between firms’ reporting speeds and the nature of their earnings news. A persistent finding is that firms with bad (good) news tend to increase their reporting speeds relative to past behavior, where reporting speeds are measured as the time between firms’ fiscal period ends and their earnings announcements. These findings suggest that managers both consider, and convey, earnings news when scheduling their earnings announcements. However, recent studies find that stock prices fail to fully respond to the predictable earnings information contained in firms’ announcements of expected reporting dates (deHaan, Shevlin, and Thornock (2015), So and Weber (2015)). Prior research also shows that firms’ earnings realizations are correlated (Foster (1981); Freeman and Tse (1992)), and that managers consider peer firms’ releases (Dye and Sridhar (1995)) and other information events (Acharya, DeMarzo, and Kremer (2011)) in timing their disclosures, which suggests that managers strategically time news relative to other sources of information. Building upon this prior research, we posit that there is information not only in firms’ choices to change their reporting speeds relative to their past behavior, which has been the focus of prior studies, but also in their choices to maintain their reporting speeds in times when other firms are delaying or accelerating news, which is a key innovation and insight of this study. Further, if firms’ collective expected delays or accelerations are due to a shift in the distribution of forthcoming earnings as opposed to a reordering within a fixed distribution, then we expect aggregate reporting speeds to convey information about aggregate earnings news and market trends. We provide empirical evidence consistent with the aforementioned predictions and, in doing so, offer new evidence on the information content of firms’ expected earnings announcement dates. Specifically, we show that trends in firms accelerating or delaying their reporting speeds are not only informative about those firms’ forthcoming earnings, but also

Expected Reporting Speeds

2

informative about earnings for the two-thirds of firms that do not alter their expected reporting reporting speeds from the prior year. We also show that firms’ reporting speeds in aggregate signal predictable changes in aggregate earnings news, treasury spreads, and market uncertainty. Finally, in conducting these tests, we design, implement, and validate a novel methodology for studying firms’ expected reporting speeds using monthly earnings calendar data. Our empirics are based on a new measure of “expected abnormal reporting speed,” or EARS, that can be calculated for a broad cross-section of firms on a synchronized, monthend basis. The central input to calculate EARS is earnings calendar data, which provides expected earnings announcement dates for the majority of U.S. public companies. We use the earnings calendar to identify all firms at the end of each month that are expected to announce earnings in the subsequent month. We then use the month-end expected announcement dates to calculate within-firm changes in each firm’s reporting speed relative to the same fiscal quarter in the prior year, where higher (lower) values correspond to cases when a firm is expected to report abnormally fast (slow). Importantly, EARS captures changes in reporting speed both for the minority of firms that materially delay or accelerate the earnings announcement relative to past behaviors (and have been the subject of previous studies), as well as for the majority of firms that do not significantly deviate from their past reporting speeds. In our firm-level tests we divide expected announcers into three EARS portfolios at each month-end, whereby the largest (smallest) EARS values are designated as being in the “fast” (“slow”) reporting group. Importantly, these portfolio assignments measure a firm’s expected abnormal reporting speed relative to other firms, so the firm can be assigned to the fast or slow reporting group even if its nominal speed is unchanged from the prior year. The minimal data requirements for calculating EARS allows us to achieve a coverage ratio of approximately 80% of the earnings announcements at the intersection of the CRSP, Compustat, and IBES databases from 2006 - 2013, which supports the likelihood that aggregate EARS predicts

Expected Reporting Speeds

3

market-level conditions. Our firm-level tests show that EARS has strong predictive power for firms’ earnings news as well as future returns. Specifically, the average analyst-based earnings surprise and profitability growth in the fast EARS portfolio are significantly more positive than in the slow EARS portfolio, and a long-short trading strategy formed at the end of month m produces an average return of 140 basis points in month m + 1. Further, since the combined long-short portfolios are formed synchronously and average 430 firms per month, these results suggest the capacity for economically significant trading strategy returns. Our findings continue to hold even in two-thirds of firms with no change in expected reporting speed from the previous year, and the pricing results are robust to value-weighting and standard risk adjustments. Combined, these results indicate that not only are within-firm changes in expected reporting speeds informative about a firm’s forthcoming earnings, but that across-firm relative changes are also indicative of earnings realizations. We next turn our attention to whether aggregate EARS is predictive of aggregate news as measured by one-month-ahead aggregate earnings surprises, aggregate returns, treasury spread changes, and market uncertainty as proxied by changes in the VIX. Our motivation for examining aggregate earnings surprise and returns is straight-forward: given our evidence that EARS predicts firm-level earnings and returns, it is plausible that the aggregation of earnings calendar information provides information about the broader economy. Our motivation for examining treasury spreads and VIX is based on prior findings that earnings surprises are associated with uncertainty (Barth and So (2014)) and discount rates (Hirshleifer et al 2009; Kothari et al 2006). Since EARS predicts firm-level earnings news not yet reflected in stock prices, if EARS predicts aggregate earnings then it is plausible that EARS also predicts changes in market-level uncertainty and risk. There are also several reasons to expect no relations between aggregate EARS and market-level outcomes. First, it is plausible that observed changes in EARS are a result of firms shuffling their earnings announcement dates according to idiosyncratic positive or

Expected Reporting Speeds

4

negative earnings realizations, such that the overall earnings distribution is unchanged from one month to the next. If so, EARS will aggregate to roughly zero and have no association with market-level conditions. Second, because prior literature finds that earnings news is informative about both future cash flows and risk, it is plausible that these effects counteract each other such that we find no relation between aggregate EARS and returns (Kothari et al 2006). Finally, prior papers find evidence that markets are better able to anticipate earnings in aggregate rather than at the firm level (Sadka and Sadka 2009), so it is plausible that EARS is impounded at the aggregate level even if not at the firm-level. We first show that both average analyst-based earnings surprises and abnormal expected reporting speeds monotonically decrease across consecutive months of a calendar quarter. More specifically, aggregate earnings surprises and abnormal reporting speeds are reliably more positive in “earnings season” months (i.e., January, April, July, and October) and become increasingly negative toward the end of the calendar quarter (i.e., March, June, September, and December). This predictable seasonality in aggregated earnings news and abnormal reporting speeds follows a striking jigsaw pattern over time, where good news tends to arrive early and bad news tends to arrive late. For our tests of earnings calendars and market outcomes, we develop two aggregate EARS measures that reflect changes in aggregate reporting behavior relative to the trailing 12-month average: (1) equal-weighted EARS, referred to as “speed average” (or SA); and (2) the month-end imbalance between positive and negative EARS firms, referred to as “speed imbalance” (or SI). Univariate and regression tests using a sample of 95 calendar-month observations find consistent evidence that both aggregate earnings innovations and analystbased earnings surprises are significantly greater in high relative to low SA and SI months, which is consistent with aggregate EARS predicting aggregate earnings news. We also find that both SA and SI are predictive of changes in the risk-free rate and VIX, consistent with aggregate EARS predicting changes in discount rates and uncertainty. However, we find evidence that SA and SI does not predict market-index returns, potentially due to the

Expected Reporting Speeds

5

offsetting effects of cash flow and risk changes. Our final set of tests link together our firm- and market-level tests by examining whether the predictive power of aggregate EARS for market-level earnings news gives rise to added predictability in firm-level returns. First, we examine whether aggregate SA and SI are predictive of firm-level returns, and in particular whether the predictive powers of SA and SI are strongest among firms with greater aggregate earnings sensitivity. Second, we examine whether the documented association between firm-level EARS and firm-level returns is stronger for firms that are more sensitive to aggregate earnings news.1 We estimate firms’ aggregate earnings sensitivities as the sensitivity of a firm’s monthly return in month t to the average analyst-based earnings surprise of all firms announcing in month t, measured over the 60 calendar months ending in month m

1. Univariate and regression tests find

consistent evidence that both SA and SI are positively associated with future firm-level returns among firms with greater aggregate earnings sensitivity, and that the EARS-returns relation is stronger in the cross-section among high-sensitivity firms. This study makes several contributions. First, we contribute to the literature examining managers’ strategic choices in disclosing private information, and in particular studies examining firms’ choices to accelerate (delay) announcements of positive (negative) earnings news. We find that the absence of abnormal reporting speeds is indicative of forthcoming earnings when viewed relative to other firms’ behavior. Specifically, in periods where the majority of firms are expected to announce abnormally slow, observing that a firm has not delayed its expected reporting speed is a positive predictor of earnings and returns. While prior studies have found evidence that managers consider their own earnings news in timing earnings releases, our findings indicate that timing decisions involve dynamic, across-firm considerations. Examining managers’ incentives and benefits of timing earnings announcements in 1

For example, consider firms with zero change in expected speed but that are assigned to a low EARS portfolio due to their position in relation to other firms at month-end (i.e., because most other firms accelerate their timing). Our firm-level tests indicate that these firms experience relatively negative returns in month m + 1, which is in part likely driven by news contained in peer firms’ earnings announcements. Thus, we expect that these firms’ experience incrementally smaller (larger) negative returns in m + 1 when the firms are less (more) sensitive to peer firms’ earnings news.

Expected Reporting Speeds

6

relation to peer firms is a potentially interesting avenue for future research. Second, we contribute to the academic and practitioner literatures on earnings-based trading strategies. Our approach of using end-of-month earnings calendar data is computationally simple and facilitates pricing tests using balanced long-short positions based on synchronized signals. By contrast, prior studies tend to focus on the returns to positions formed in event-time (e.g., selling after a firm misses its expected date), which increases the difficulty of hedging risks and scaling position sizes in response to fluctuations in the existence and nature of events within a given time period.2 Additionally, our methodology permitting a larger sample also allows for layering multiple signals and screening firms along multiple dimensions when forming positions, which more closely mimics how investment strategies are implemented in practice. Finally, we contribute to the literatures studying macroeconomic forecasting and aggregate associations between earnings news and market trends (e.g., Konchitchki and Patatoukas (2013); Konchitchki and Patatoukas (2014)). Our study shows that aggregate reporting speeds have strong predictive power for aggregate earnings surprises, treasury spreads, and changes in the VIX. Together, these findings show that information from earnings calendars may be useful in both generating and improving macroeconomic forecasts. The rest of the paper is organized as follows. Section 2 details the data and methodology. Section 3 discusses our firm-level tests and Section 4 discusses our market-level tests. Section 5 concludes. 2

For example, our sample used in long-short positions increases ten-fold compared to So and Weber (2015) despite using the same underlying calendar data over the same time period. Specifically, the revisionbased pricing tests in So and Weber (2015) involve approximately 125 observations per quarter compared to over 1,250 per quarter in this study. Further, our methodology permits forming synchronized long and short positions at the end of each month. By contrast, because calendar revisions are non-synchronized across time, the main tests in So and Weber (2015) involve forming unbalanced long and short positions in event-time and at uneven intervals. However, So and Weber (2015)’s sample documents greater return predictability, suggesting that firm-initiated calendar revisions may provide a more precise signal of firms’ earnings news than expected reporting speeds. This tradeoff suggests that the appropriate methodology may vary across contexts and depend on the goals of the researcher. The data limitations imposed by prior methodologies are even more pronounced in studies such as Penman (1984) and Bagnoli, Kross, and Watts (2002) that measure reporting speeds only ex post, after a firm misses an expected date or unexpectedly announces early.

Expected Reporting Speeds

7

2. Data and Methodology The main analyses of this paper examine the information content of firms’ expected reporting speeds, using variation measured both within-firm and across-firm. Empirically, we test whether the month-end landscape of an earnings announcement calendar predicts future firm-specific news and macroeconomic conditions. Forward-looking earnings announcement calendar data are sourced from Wall Street Horizon, which has been used in related studies including So and Weber (2015) and deHaan, Shevlin, and Thornock (2015).3 We begin by calculating firm-specific EARS at the end of each month for firms that are expected, per the earnings calendar, to report earnings in the subsequent month. Notationally, EARS is calculated as follows:

EARSi,m,q,y =

1 ⇤ (ExpectedSpeedi,m,q,y RealizedSpeedi,q,y 1 ) RealizedSpeedi,q,y 1

(1)

where ExpectedSpeedi,m,q,y is the expected number of days between firm i’s announcement date and its fiscal quarter end as of the end of month m for quarter q of year y, and RealizedSpeedi,q,y

1

is defined as the difference between the realized announcement date and

fiscal quarter-end in the previous year.4 The difference between ExpectedSpeedi,m,q,y and RealizedSpeedi,q,y

1

is multiplied by -1 so that higher (lower) values of EARS corresponds

to firms reporting abnormally fast (slow) relative to the prior year. To mitigate the influence of data errors and outliers, we require firms to have an expected reporting speed between 15 and 75 trading days relative to their fiscal period end, although the main results do not appear sensitive to this requirement. To study the predictive power 3

As discussed in So and Weber (2015) and deHaan, Shevlin, and Thornock (2015), Wall Street Horizon began disseminating daily snapshots of the earnings calendar in 2006, where each snapshot lists expected announcement dates for a broad cross-section of firms. The earnings calendar is updated daily in response to public information including, but not limited to, firms’ investor relations webpages, press releases, and direct correspondence with firms. The daily snapshots reflect information available to investors by 4am ET of each trading day. Wall Street Horizon data is commercially distributed via the Toronto Stock Exchange’s TMX Datalinx service, as well as other sources. 4 RealizedSpeed is not indexed to m as it is a known date and, therefore, do not vary by month.

Expected Reporting Speeds

8

of EARS for firms earnings news and returns, we also require firms to have at least six months of historical return data in CRSP, fundamental information and prior year earnings announcement dates in Compustat, and at least one analyst-based earnings forecast in IBES. The resulting sample consists of 83,411 observations spanning 2006-2013. Our firm-level tests examine relative comparisons of firms’ EARS in calendar time based on synchronized, month-end signals. For our firm-level tests, we assign each firm into one of three portfolios on the last day of each month, where the 30% of firms with the highest (lowest) EARS are assigned to the abnormally fast (slow) portfolio relative to the remaining 40% of firms at the same month-end. The timeline below illustrates the calculation of EARS at the end of January for three hypothetical firms, referred to as ‘A’, ‘B’, and ‘C’, which all have fourth quarter ending date of December 31st and are expected to announce earnings in the month of February.

2/1/Y-1

Historical Speed: Calculate reporting speeds from prior calendar year using firms’ realized dates | {z } A’s RA 2/5

B’s RA 2/10

C’s RA 2/15

Calendar observed: EARS calculated for firms A, B, and C on 1/31/Y 2/28/Y-1

+

2/1/Y A’s EA 2/5

2/28/Y B’s EA 2/8

C’s EA 2/10

The timeline above provides an example of how EARS would be calculated on Jan. 31 of year y, where RA refers to a firm’s realized announcement date in (y

1) and EA refers to

a firm’s expected announcement date in year y. Assuming there are 31 days in January and (for the sake of this example) ignoring weekends and holidays, RealizedSpeedi,q,y

1

for firms

A, B, an C would be 36, 41, and 46, respectively. Similarly, based on the earnings calendar data observed on Jan. 31, ExpectedSpeedi,m,q,y for firms A, B, an C would be 36, 39, and 41, respectively. Thus, EARS would be 0 (=-1*(36-36)/36) for firm A, 0.05 (=-1*(39-41)/41) for firm B, and 0.11 (=-1*(41-46)/46) for firm C. In portfolio assignments, firm C would be characterized as being abnormally fast, firm B as being in the middle group, and firm A as being abnormally slow.

Expected Reporting Speeds

9

The above example demonstrates two important characteristics of our EARS measure. First, our use of within-firm changes helps screen out variation in earnings announcement speed that may be driven by firms’ innate reporting processes. Second, our use of monthend EARS portfolio assignments allows us to characterize firms as being relatively early or late even for firms that do not change their expected timing from the previous year. For example, firm A is characterized as being in the slow portfolio group despite being the first to announce earnings as well as having no change in timing from the previous year. The above timeline also helps emphasize two important differences between our EARSbased analyses as compared to previous studies of early versus late earnings announcements. First, our EARS measure can be constructed without look-ahead bias and even before the firm surpasses its realized earnings speed from the prior year, while prior studies could often only identify early and late announcements on an ex post basis (e.g., Penman (1984); Bagnoli, Kross, and Watts (2002)). Second, the sample examined in So and Weber (2015) relies on explicit, firm-initiated announcements of changes in earnings timing, while our approach includes firms that do not confirm or revise their expected earnings announcement date. These two EARS characteristics allow us to construct a much broader sample of firms representing over 80 percent of the merged Compustat, CRSP, and IBES universe.5 3. Main Findings We begin this section by providing descriptive statistics. Section 3.2 examines the association between EARS and future earnings news, both for firms that do and do not change their reporting speed from the prior year. Section 3.3 discusses the predictive ability of EARS for firms’ returns. 5

Of course, these EARS characteristics are not without tradeoffs. For example, firm-initiated announcements of changes in previously-scheduled earnings times, as examined in So and Weber (2015), likely provide a much cleaner signal about the content of future earnings news than do our tests of month-end earnings calendar snapshots. The usefulness of the approach advocated in this paper relative to those in prior research involves tradeoffs and thus depends on the goals of the researcher.

Expected Reporting Speeds

10

3.1. Descriptive Statistics Panel A of Table 1 presents annual sample averages of the data contributing to our EARS measure. OBS equals the number of firm-month observations and F IRM S indicates the number of unique firms, which show that the sample consists of approximately ten thousand observations per year and roughly three thousand unique firms. %SAM P measures the number of observations in the sample relative to the full CRSP/Compustat/IBES universe, which shows that the sample in this study achieves a coverage ratio of approximately 80% of the earnings announcements.

RS is the level change in a firm’s expected reporting speed

relative to the realized reporting speed in the prior year, which shows that most firms are expected to report within one trading day of their past reporting speed. %SAM EDAT E equals the proportion of firms that are expected to report earnings on roughly the same date as the prior year, defined as firms with an absolute

RS of one or

zero. Roughly two-thirds (66%) of firms are scheduled to announce earnings within one trading day of their past reporting speed, which highlights the magnitude of the sample that are omitted when only studying firms that materially change their expected announcement date. DEV equals the number of days between a firm’s expected announcement date and the future realized announcement date, showing that most firms eventually announce within one day of their expected date. LRS is firms’ lagged reporting speed, where the average firm reported earnings 25 trading days after their fiscal period end. Finally, the pooled average EARS indicates that the average firm is expected to announce one-half of one percent faster than in the prior year, although there is variation across years. Figure 1 shows that nearly 50% of EARS observations fall within ±2.5 of zero, and over 20% of observations fall within the range of +2.5% to 7.5%. Panel B of Table 1 presents descriptive statistics across tercile portfolios of EARS, where the top (bottom) 30 percent are assigned to the ‘High’ (‘Low’) portfolio and the remaining are assigned to the ‘Mid’ portfolio at the end of each calendar month. Panel B also reports average differences across the high and low portfolios, where the reported t-statistics corre-

Expected Reporting Speeds

11

spond to the time-series average monthly difference. There are approximately 228 firms per month in the high portfolio and 201 firms per month in the low portfolio, indicating that the long-short strategy proposed in this paper consists of approximately 428 firms per month.6 RS row shows that abnormally fast firms decrease their reporting speed by an average of 3.9 days and abnormally slow firms increase by their reporting speed by an average of 3.9 days. DEV indicates that abnormally slow firms are more likely to report late relative to their expected date than abnormally fast firms, but the average difference is less than one trading day. Panel B of Table 1 also reports firms’ market capitalization, M CAP , reported in millions, and return momentum, M OM EN , defined as the cumulative market-adjusted return over the prior 12-months. Abnormally fast firms tend to be roughly the same size as abnormally slow firms, although both groups are larger than the middle-tercile firms. Abnormally fast firms have higher return momentum compared to abnormally slow firms. 3.2. EARS and Future Earnings News Table 2 examines the association between EARS and firms’ subsequently reported earnings news. Panel A presents sample averages of earnings news proxies across EARS portfolios. SU RP equals the actual EPS number reported in IBES minus the last consensus forecast available immediately prior to the announcement, and scaled by beginning-of-quarter assets.7 SU E is the standardized unexplained earnings, defined as the realized EPS minus EPS from four quarters prior, divided by the standard deviation of this difference over the prior eight quarters. The first two rows of Panel A of Table 2 present results for the pooled sample. Analystsbased earnings surprises are significantly higher among abnormally fast firms compared to 6

By comparison, the average number of positions in So and Weber (2015) is less than one-tenth the size at approximately 42 per month. 7 We scale EPS forecast errors by assets as opposed to price to avoid a the mechanical relation between prices and earnings expectations noted in Cheong and Thomas (2011) and Ball (2011). However, results are qualitatively unchanged if EPS surprise is scaled by market value. Throughout this paper, we use the term "qualitatively unchanged" to mean that the tests of interest are of the same sign as the reported results and are significant at 10% or higher.

Expected Reporting Speeds

12

abnormally slow firms, where the average high-low spread accounting for 0.11% of firms’ assets (t-statistic = 4.91). Average SUE’s monotonically increase across EARS portfolios, consistent with expected abnormal reporting speeds predicting subsequently reported earnings news. Panel B (C) presents similar analysis for only those firms without (with) a change in their expected reporting speed relative to the prior year (i.e., with SAM EDAT E = 1 and 0, respectively). We again find significantly more positive SU RP and SU E for firms in the highest EARS tercile relative to the lowest tercile, which indicates that EARS is predictive of earnings news even for firms that do not alter their earnings timing from the previous year. Panel D contains results from monthly Fama-MacBeth regressions of earnings news proxies on EARS (in its continuous form) and firm controls. LBM and SIZE are the log of one plus the book-to-market ratio and log of market capitalization, and M OM EN is the cumulative market-adjusted return over the prior 12-months ending in month m. The parentheses contain t-statistics from the Fama-MacBeth regressions after Newey-West adjustments for autocorrelation up to three lags. The results show that the positive link between EARS and firms’ earnings news is incremental to standard controls. 3.3. EARS and Future Returns Table 3 provides evidence on the predictive power of EARS for firms’ returns in month m + 1. Specifically, Table 3 contains both equal- and value-weighted average returns across EARS portfolios, using five measures of firm-level returns. RET(1) is the the firm’s raw return in the month of its expected earnings announcement month. The Five-Factor Alpha is the intercept from a regression of raw returns minus the risk-free rate, regressed on the excess market return (MKTRF); two Fama-French factors (SMB and HML); the PastorStambaugh liquidity factor (LIQ), and the momentum factor (UMD). Similarly, the OneFactor Alpha results from returns regressed MKTRF; the Three-Factor Alpha results from returns regressed on MKTRF, SMB, and HML; and the Four-Factor Alpha results from returns regressed on MKTRF, SMB, HML, and UMD. The reported t-statistics correspond

Expected Reporting Speeds

13

to time-series average monthly difference across High and Low EARS portfolios. Univariate tests in Panel A of Table 3 shows that EARS has strong predictive power for future returns. The average spread in equal-weighted returns across high and low EARS portfolios is relatively constant across the five return metrics at approximately 1.4 percent per month with a corresponding t-statistic of approximately six. Panel B of Table 3 shows that the results weaken somewhat using value-weighted returns, but that the return spreads remain economically and statistically significant. Figure 2 shows that a long-short strategy based on EARS portfolios generates positive equal-weighted returns in 76 of 95 months in our sample period. Figure 3 plots daily longshort strategy returns within month m+1 relative to firms’ expected earnings announcement dates t. The majority of the strategy return appears to occur within days [0, 2] relative the earnings announcement, indicating that the predictive power of EARS is indeed a result of EARS forecasting earnings news that is not yet reflected in market prices. Panel C of Table 3 contains results from monthly Fama-MacBeth regressions of raw returns on EARS, firm-level risk proxies, and alternative measures of firms’ expected announcement speeds. Column (1) shows that EARS predicts returns incremental to standard risk proxies including firm size, book-to-market, return momentum, and volatility. Columns (2) and (3) of Panel C are intended to further distinguish this paper from prior research on the informativeness of changes in earnings announcement timing. REV is the cumulative change in firms’ expected earnings announcement date over the prior two months. Consistent with So and Weber (2015), column (2) shows that REV is predictive of future returns. Finally, column (3) shows that EARS and REV both retain statistically significant predictive power for future returns when both variables are included in the regression, indicating that EARS and calendar revisions capture distinct aspects of firms’ earnings news. To mitigate concerns that the announcement-month returns are driven by return reversals, column (4) of Panel C of Table 3 repeats the model from column (3) but includes a control for the prior month’s return. Results are qualitatively unchanged. Column (5)

Expected Reporting Speeds

14

shows that the level change in expected reporting speed,

RS, the numerator of Equation

(1), also has predictive power for announcement-month returns. However, column (6) shows that when both EARS and

RS are included together, only EARS retains incremental

explanatory power, indicating that the informativeness of abnormal announcement speeds for firms returns depends on the magnitude of the change relative to the level of the firms’ reporting speed. Panel A (B) of Table 4 presents average portfolio returns across EARS portfolios for the subsample of firms that are expected to report earnings at the same (different) reporting speed as in the prior year. Panel A shows that EARS portfolio assignments predict equalweighted returns even among firms that do not alter their reporting speed from the prior year. Among non-change firms, the high EARS portfolio outperforms the low EARS portfolio by 115 basis points per month (t-statistic = 2.44) using firms’ raw returns. These results speak to the value of identifying cross-sectional variation in reporting speeds using all announcing firms rather than only those that change their behavior from the prior year. Panel B of Table 4 confirms that there is strong predictive power for EARS portfolios among firms that alter their reporting speeds. Among these firms, the high EARS portfolio outperforms the low EARS portfolio by 164 basis points per month (t-statistic = 4.70) using firms’ raw returns, which is slightly larger than the excess performance among firms in Panel A that did not change their reporting speed. By showing that EARS predicts firms’ returns even for firms that did not change their expected reporting speeds, the results in Panels A and B of of Table 4 demonstrate that their is information contained in the relative speed of announcing earnings and not simply in the within-firm variation in reporting speeds. Panels C and D of Table 4 extend these results by examining the predictive content of SAM EDAT E in states where most firms are reporting faster versus slower in a given reporting period. We predict that a firm choosing to keep its reporting speed unchanged is a strong signal of positive news when the average reporting firm is reducing their reporting speeds.

Expected Reporting Speeds

15

To test this prediction, we calculate two aggregate EARS measures for each month m. Our first measure is referred to as “speed average,” or SA, and is intended to capture the overall average EARS observed at the end of the month. Specifically, SA is defined as the equal-weighted firm-level EARS at the end of month m, minus the twelve-month historical average. We subtract the twelve-month historical average to adjust for the overall trend towards faster reporting speeds observed in Panel A of Table 1. Our second measure is intended to capture the imbalance between abnormally fast and slow announcers. Specifically, our measure of “speed imbalance” (SI) is calculated the difference in number of firms that are expected to report faster relative to slower than in the prior year, scaled by the total of firms expected to report faster and firms expected to report slower, minus its twelve-month historical average. The "High" ("Low") SA and SI subsamples include those months where SA and SI is greater than or equal to (less than) zero. The results in Panels C and D show that the positive relation between SAM EDAT E and announcement returns is driven by firms that do not change their reporting speed in periods when the majority of announcing firms have opted to reduce their reporting speed. These findings underscore the importance of considering the behavior of firms relative other firms in the economy, suggesting that a firm’s choice to report at the same speed as in prior years is a positive signal of firm performance if other firms are opting to report slower. To summarize the results so far, we develop a simple measure of expected abnormal reporting speed, EARS, that can be calculated for a broad cross-section of firms and used to measure relative changes in speed at month-end. We find strong associations between relative EARS and future earnings, both for firms that do and do not change their speed from the prior year, suggesting that there is information in both within-firm and across firmvariation in expected reporting speeds. We also find that the market does not appear to price this information about future earnings, as indicated by predictable future returns. In the next section, we extend the firm-level results to show that earnings calendars also have strong predictive power for market-level information.

Expected Reporting Speeds

16

4. Market-Level Tests Section 4.1 discusses our analysis of aggregate EARS for predicting macro-economic conditions. Section 4.2 discusses cross-sectional tests of the whether the predictive ability of firm-level EARS for firms’ earnings and returns varies depending on the firms’ sensitivity to macroeconomic conditions. 4.1. Market-Level Outcomes The objective of the tests in this section is to investigate whether there is information about the macroeconomy in aggregate variation in firms’ reporting speeds. We conduct these tests by examining whether the overall landscape of the earnings calendar in month m is predictive of market-level results and macroeconomic conditions in month m + 1. To begin, Figures 4 and 5 provide descriptive information on average earnings announcement frequency, SA, SI, and analyst-based earnings surprise (SU RP ) by month. Panel A of Figure 4 shows that the first and second months of each calendar quarter contain the most earnings announcements, which is expected given that most firms’ fiscal quarters align with calendar quarters. Panels B and C of Figure 4 plot the average SA and SI in each month m, as calculated at the end of m

1. Both SA and SI tend to decrease by

month within each calendar quarter, which is consistent with firms with positive (negative) news accelerating (delaying) their earnings announcements. Figure 5 provides striking evidence regarding the arrival of earnings news across months of an earnings season. The downward jigsaw pattern within each reporting cycle (i.e., calendar quarter) is consistent with this intuition in showing that average SU RP decreases by month within each calendar quarter. Since SU RP is based on analyst-based earnings surprises, the pattern in Figure 5 is also consistent with analysts (and plausibly other market participants) not incorporating EARS information into their expectations of forthcoming earnings. We also supplement the findings in Figure 4 by presenting descriptive statistics regarding how aggregate measures of news relate to aggregated calendar information. The first row of

Expected Reporting Speeds

17

Panel A of Table 5 tabulates average equal- and value-weighted SU RP for the first month of each calendar quarter (i.e., the “base” period) relative to the second and third months (i.e., the “comparison” period). Consistent with the visual evidence in Figure 5, we find that equal- and value-weighted SU RP are significantly higher in the base period relative to the comparison period. Tests also show that average SU RP is higher in the high SA months relative to low SA months, as well as high SI versus low SI months. The regressions in Panels B and C of Table 5 investigate whether the aggregate signals are predictive of future earnings news after including year-quarter fixed effects to control for seasonal patterns. Column (1) of Panel B regresses equal-weighted SU RP on an indicator for the first month of each reporting period (variable F M D for "first month dummy"). Consistent with the results in Panel A, SU RP is significantly more positive in the first month of each quarter. Columns (2) and (3) find similar results when regressing SU RP on indicators for high SA and SI, respectively. Importantly, columns (4) and (5) include F M D as well as SA and SI, respectively, and find that SA and SE both have incrementally stronger associations with SU RP than does F M D. Panel C of Table 5 finds similar results for valueweighted SU RP , except that SA is not statistically significant in column (4) when controlling for the first month indicator. In sum, the results in Table 5 find that SA and SI are strongly associated with future analyst-based earnings surprises and that the aggregation of earnings calendar data provides information incremental to the sequence of months within a given reporting period. In the tests below, we extend these results by examining the usefulness of aggregate calendar information in forecasting alternative market-level outcomes. In Panel A of Table 6, we examine how SA and SI at the end of month m predict the change in SP500 earnings in month m + 1, SP500 returns, and change in the riskfree rate.8 Column (1) finds that high SA months are associated with positive changes in SP500 earnings, which is consistent with the results found in Table 5. Column (2) finds an insignificant result for SI, and column (3) finds a significantly positive result for months 8

Our thanks to Robert Shiller for providing this data on his website

Expected Reporting Speeds

18

that have both SA and SI equal to one. In columns (4) through (6) we find insignificant associations with SP500 returns. Thus, while EARS predicts future firm-level returns, no such association is observed at the market level. Columns (7) through (9) find positive associations with changes in the future risk-free rate, consistent with prior evidence that market-level earnings news results in increases in discount rate that investors apply to firms’ earnings (e.g., Cready and Gurun (2010)). 4.2. Cross-sectional analysis: firm-level tests based on macroeconomic sensitivities In this section, we provide evidence that the predictive power of aggregate expected reporting speeds for market-level information yields added return predictability among stocks sensitive to aggregate news. Specifically, we predict aggregated market-level calendar data is informative for firm-level returns among firms that are more sensitive to aggregate earnings news. Akin to an ‘earnings-surprise-beta’, we calculate firms’ market earnings sensitivities as the sensitivity of a firm’s monthly return in month t to the average analyst-based earnings surprise of all firms announcing in month t, measured over the 60 calendar months ending in month m

1. A firm is classified as “High” (“Low”) aggregate earnings sensitivity if it is

above the median of all firms expected to announce in month m + 1, and is identified as such using the binary variable M ACRO. Panel A of Table 7 tabulates average equal-weighted returns in the high and low MACRO firms, by high versus low SI. High MACRO firms earn an average of 2.261% in the high SA months, while low MACRO firms earn 1.550%. The difference of 0.711% is statistically significant, consistent with EARS having a stronger association when firms are more sensitive to market earnings news. No significant returns are observed in the low SA months. The difference-in-differences is 0.831 and significantly positive. Results in Panels B for SI are qualitatively similar. These results show that combining the market-level information in aggregate expected reporting speeds with firm-level tests further improves the predictive power of expected reporting speeds for firm-level earnings and returns realizations.

Expected Reporting Speeds

19

Table 8 examines the interaction effect between SA and M ACRO in predicting future returns. All regressions include untabulated controls for firm size, book-to-market, momentum, and return volatility. Results in column (1) of Panel A finds that EARS is positively associated with one-month-ahead returns, and the interaction between SA ⇤ M ACRO indicates that this association is stronger in high-SA months. Column (2) includes an additional control for stock market beta (BET A) and results are qualitatively unchanged. Columns (3) and (4) find that EARS is insignificantly associated with two-months-ahead returns, and that the SA ⇤ M ACRO interaction is significantly negative. Together the results in columns (1) through (4) show that firms with high sensitivities to aggregate news initially earning higher returns, consistent with their prices predictably rising in response to initially positive aggregate news, but also subsequently earn lower returns, consistent with their returns predictably falling in response subsequently announced negative aggregate news. Columns (5) and (6) find that EARS is positively associated with three-months-ahead returns, but that the interaction between SA ⇤ M ACRO is insignificant. Panel B provide similar results for SI. Collectively, the results in Table 8 show that predictable variation in aggregate news leads to predictable variation in firm-level returns. Recall that high SA months corresponds to periods when the average announcing firm is more likely to report positive news. As a result, firms with sensitivities to aggregate news tend to experience an uptick in prices. However, in months following the realization of high SA, the average firm is less likely to report positive news. In fact, empirically we show that non-high (i.e., low) SA months tend to correspond to more negative earnings news. As a result, firms with sensitivities to aggregate news tend to also experience a subsequent downtick in prices as negative aggregate earnings are reported. The symmetry and significance of these findings show that earnings calendar information is useful in understanding not only aggregate patterns in earnings information but also predictable return patterns among firms sensitive to aggregate news.

Expected Reporting Speeds

20

4.3. Market-Level Uncertainty A central insight from this paper is that there is information in inter-temporal variation in aggregate reporting behavior, rather than simply in within-firm variation that has been the focus on prior research. The analyses so far have primarily focused on predicting the sign and magnitude of earnings news and returns. In our final analyses, we examine whether earnings calendar information also provides information about the second moment of earnings news and returns as captured by innovations in the CBOE volatility index, commonly referred to as the VIX. To the extent that pervasive changes in reporting speeds corresponds to greater dispersion in firms’ subsequently announced earnings performance, we expect earnings calendar information to help predict changes in market-level uncertainty. Table 9 contains time-series regressions of changes in the CBOE volatility index, VIX, on the average absolute value of EARS denoted last ABS(AGGEARS). The dependent variable in these regressions is Log(V IX m+1 /V IX m ) which equals the log change in the VIX in month M +1, where M is the month of the earnings calendar. All independent variables are measured in month M and V IX m denotes the level of the VIX in month M . The results in Table 9 show that ABS(AGGEARS) has strong univariate and incremental explanatory power for innovations in the VIX, even after controlling for past innovations and levels. These findings add strong support to the idea that aggregated information regarding firms’ reporting speeds, and earnings calendars more broadly, contain information useful in forecasting market-level outcomes. 5. Conclusion Prior research shows that firms’ earnings realizations are correlated (Foster (1981); Freeman and Tse (1992)), and that managers consider peer firms’ releases (Dye and Sridhar (1995)) and other information events (Acharya, DeMarzo, and Kremer (2011)) in timing their disclosures, which suggests that managers strategically time news relative to other sources of information. Building upon this prior research, we provide evidence that there is

Expected Reporting Speeds

21

information not only in firms’ choices to change their reporting speeds relative to their past behavior, which has been the focus of prior studies, but also in their choices to maintain their reporting speeds in times when other firms are delaying or accelerating news, which is a key innovation and insight of this study. We show that trends in firms accelerating or delaying their reporting speeds are not only informative about those firms’ forthcoming earnings, but also informative about earnings for the two-thirds of firms that do not alter their expected reporting reporting speeds from the prior year. We also show that firms’ reporting speeds in aggregate signal predictable changes in aggregate earnings news, treasury spreads, and market uncertainty. Collectively, this paper makes three contributions. First, conceptually, while prior studies have found evidence that managers consider their own earnings news in timing earnings releases, our findings indicate that timing decisions involve dynamic, across-firm considerations. Examining managers’ incentives and benefits of timing earnings announcements in relation to peer firms is a potentially interesting avenue for future research. Second, methodologically, our approach of using end-of-month earnings calendar data is computationally simple and facilitates pricing tests using balanced long-short positions based on synchronized signals. Our methodology permitting a larger sample also allows for layering multiple signals and screening firms along multiple dimensions when forming positions, which more closely mimics how investment strategies are implemented in practice. Finally, we contribute to the literatures studying macroeconomic forecasting by showing that aggregate reporting speeds have strong predictive power for aggregate earnings surprises, treasury spreads, and changes in the VIX, suggesting that information from earnings calendars may be useful in both generating and improving macroeconomic forecasts.

Expected Reporting Speeds

22

References Acharya, V.V., DeMarzo, P.M., Kremer, I., 2011. Endogenous information flows and the clustering of announcements. American Economic Review 2955–2979. Bagnoli, M., Kross, W., Watts, S.G., 2002. The information in management’s expected earnings report date: A day late, a penny short. Journal of Accounting Research 40, 1275–1296. Ball, R., 2011. Discussion of why do eps forecast error and dispersion not vary with scale? implications for analyst and managerial behavior. Journal of Accounting Research 49, 403–412. Barth, M.E., So, E.C., 2014. Non-diversifiable volatility risk and risk premiums at earnings announcements. The Accounting Review 89, 1579–1607. Cheong, F., Thomas, J., 2011. Why do eps forecast error and dispersion not vary with scale? implications for analyst and managerial behavior. Journal of Accounting Research 49, 359–401. Cready, W.M., Gurun, U.G., 2010. Aggregate market reaction to earnings announcements. Journal of Accounting Research 48, 289–334. deHaan, E., Shevlin, T., Thornock, J., 2015. Market inattention and the strategic scheduling and timing of earnings announcements. Journal of Accounting and Economics 60, 36–55. Dye, R.A., Sridhar, S.S., 1995. Industry-wide disclosure dynamics. Journal of accounting research 157–174. Foster, G., 1981. Intra-industry information transfers associated with earnings releases. Journal of accounting and economics 3, 201–232. Freeman, R., Tse, S., 1992. An earnings prediction approach to examining intercompany information transfers. Journal of Accounting and Economics 15, 509–523. Givoly, D., Palmon, D., 1982. Timeliness of annual earnings announcements: Some empirical evidence. Accounting Review 486–508. Konchitchki, Y., Patatoukas, P.N., 2013. Taking the pulse of the real economy using financial statement analysis: Implications for macro forecasting and stock valuation. The Accounting Review 89, 669–694. Konchitchki, Y., Patatoukas, P.N., 2014. Accounting earnings and gross domestic product. Journal of Accounting and Economics 57, 76–88. Kross, W., 1981. Earnings and announcement time lags. Journal of Business Research 9, 267–281. Penman, S.H., 1984. Abnormal returns to investment strategies based on the timing of earnings reports. Journal of Accounting and Economics 6, 165–183. So, E.C., Weber, J., 2015. Time will tell: Information in the timing of scheduled earnings news. Working paper, Massachusetts Institute of Technology.

Expected Reporting Speeds

23

Figure 1. Histogram of Expected Abnormal Reporting Speed The figure contains a histogram of values of expected abnormal reporting speed, EARS. EARS is the percentage change in a firm’s expected reporting speed relative to the same fiscal quarter in the prior year. A firm’s expected reporting speed is defined as the number of days between its expected earnings announcement date and fiscal period end date and the lagged reporting speed is defined as the difference between a firm’s realized earnings announcement date from the prior year and its corresponding fiscal period end date. Expected earnings announcement dates of firms expected to announce in month M + 1 are measured in the earnings calendar data on the final trading date of calendar month M . The histogram plots the percentage of sample correspond to each bucket falling within ±2.5 of the X-axis value. The sample for this analysis consists of 83,411 firm-month observations spanning 2006-2013.

Expected Reporting Speeds

24

Figure 2. Monthly Strategy Returns The figure contains the monthly spread in returns across high and low EARS portfolios. EARS is the percentage change in a firm’s expected reporting speed relative to the same fiscal quarter in the prior year. A firm’s expected reporting speed is defined as the number of days between its expected earnings announcement date and fiscal period end date and the lagged reporting speed is defined as the difference between a firm’s realized earnings announcement date from the prior year and its corresponding fiscal period end date. Expected earnings announcement dates of firms expected to announce in month M +1 are measured in the earnings calendar data on the final trading date of calendar month M . Returns are measured in month M +1. Observations are assigned to portfolios at the end of each calendar month, where the top (bottom) 30 percent are assigned to the ‘High’ (‘Low’) portfolio and the remaining are assigned to the ‘Mid’ portfolio. The return spread corresponds to the return in month M +1 from a long (short) position in firms within the high (low) EARS portfolios. The sample for this analysis consists of 83,411 firm-month observations spanning 2006-2013.

Expected Reporting Speeds

25

Figure 3. Cumulative Returns in Event-Time The figure contains cumulative difference in returns across high and low EARS portfolios in the 21 trading days relative to firms’ expected announcement date, t. EARS is the percentage change in a firm’s expected reporting speed relative to the same fiscal quarter in the prior year. A firm’s expected reporting speed is defined as the number of days between its expected earnings announcement date and fiscal period end date and the lagged reporting speed is defined as the difference between a firm’s realized earnings announcement date from the prior year and its corresponding fiscal period end date. Expected earnings announcement dates of firms expected to announce in month M +1 are measured in the earnings calendar data on the final trading date of calendar month M . Returns are measured relative to the expected announcement date in month M +1. Observations are assigned to portfolios at the end of each calendar month, where the top (bottom) 30 percent are assigned to the ‘High’ (‘Low’) portfolio. The sample for this analysis consists of 83,411 firm-month observations spanning 2006-2013.

Expected Reporting Speeds

26

Figure 4. Aggregate Earnings Announcements Statistics The top panel presents the average number of expected earnings announcements for each month within a given year of our 2006-2013 sample. Expected earnings announcement dates of firms expected to announce in month M + 1 are measured in the earnings calendar data on the final trading date of calendar month M . The second panel plots the average expected abnormal reporting speed, EARS, of all observations where the firm is expected to announce in a given calendar month. EARS is the percentage change in a firm’s expected reporting speed relative to the same fiscal quarter in the prior year. A firm’s expected reporting speed is defined as the number of days between its expected earnings announcement date and fiscal period end date and the lagged reporting speed is defined as the difference between a firm’s realized earnings announcement date from the prior year and its corresponding fiscal period end date. Expected earnings announcement dates of firms expected to announce in month M +1 are measured in the earnings calendar data on the final trading date of calendar month M . The bottom panel plots the aggregate speed imbalance, SI, in a given calendar month. SI equals the difference in number of firms that are expected to report faster relative to slower than in the prior year, scaled by the total of firms expected to report faster and firms expected to report slower in a given calendar month. Each earnings season is divided into the first, second, and third months, where the first month denotes January, April, July, and October (shown in black bars); the second month denotes February, May, August, and November (shown in grey bars); and the third month denotes, March, June, September, and December (shown in white bars). The sample for this analysis consists of 83,411 firm-month observations spanning 2006-2013.

Expected Reporting Speeds

Figure 4: [Continued] Aggregate Earnings Announcements Statistics

27

Expected Reporting Speeds

28

Figure 5. Average Earnings Surprises by Month The figure presents the average analyst-based earnings surprise, SU RP , for each month within a given year of our 2006-2013 sample. Expected earnings announcement dates of firms expected to announce in month M + 1 are measured in the earnings calendar data on the final trading date of calendar month M . Each earnings season is divided into the first, second, and third months, where the first month denotes January, April, July, and October (shown in black bars); the second month denotes February, May, August, and November (shown in grey bars); and the third month denotes, March, June, September, and December (shown in white bars). The sample for this analysis consists of 83,411 firm-month observations spanning 2006-2013.

Expected Reporting Speeds

29

Table 1. Descriptive statistics Panel A presents annual sample averages of the main variables used throughout the paper. OBS equals the number of firmmonth observations and F IRM S indicates the number of unique firms. EARS is the percentage change in a firm’s expected reporting speed relative to the same fiscal quarter in the prior year. A firm’s expected reporting speed is defined as the number of days between its expected earnings announcement date and fiscal period end date and the lagged reporting speed is defined as the difference between a firm’s realized earnings announcement date from the prior year and its corresponding fiscal period end date. Expected earnings announcement dates of firms expected to announce in month M + 1 are measured in the earnings calendar data on the final trading date of calendar month M . RS is the level change in a firm’s expected versus lagged reporting speeds and LRS indicates the firm’s lagged reporting speed. DEV equals the number of days between a firm’s expected announcement date and the actual announcement date. %SAMP measures the number of observations in the sample relative to the full CRSP/Compustat/IBES universe. SAM EDAT E is a binary variable that equals one if the firm is expected to announce earnings at the same speed as in the prior calendar year. Panel B presents descriptive statistics across portfolios of EARS, where observations are assigned to portfolios at the end of each calendar month, where the top (bottom) 30 percent are assigned to the ‘High’ (‘Low’) portfolio and the remaining are assigned to the ‘Mid’ portfolio. MCAP equals firms’ market capitalization reported in millions and MOMEN is the cumulative market-adjusted return over the prior 12-months ending in month M . The reported t-statistics correspond to time-series average monthly difference across High and Low EARS portfolios. The sample for this analysis consists of 83,411 firm-month observations spanning 2006-2013. Panel A: Sample Averages by Year OBS

Firms

%SAMP

%SAMEDATE

DEV

LRS

EARS

2006 2007 2008 2009 2010 2011 2012 2013

8,399 10,064 10,399 10,812 11,028 10,794 11,253 10,662

2,678 3,056 3,028 3,090 3,120 3,100 3,157 3,097

0.654 0.762 0.816 0.881 0.848 0.832 0.850 0.798

0.216 0.326 0.728 0.388 0.677 0.756 0.858 0.520

RS

0.690 0.698 0.569 0.687 0.742 0.738 0.471 0.694

0.399 0.275 0.468 0.351 0.049 0.145 0.429 0.301

24.501 25.772 25.606 25.336 25.434 25.155 25.721 26.635

-1.176 -1.049 0.924 0.084 1.153 1.285 2.124 0.679

Avg

10,426

3,041

0.805

0.559

0.661

0.302

25.520

0.503

Panel B: Monthly Sample Averages by EARS Portfolios EARS Portfolios OBS EARS RS LRS DEV MCAP MOMEN

High-Low

High (Faster)

Mid

Low (Slower)

Mean

t-statistic

227.6 10.90 3.91 24.06 0.70 41.09 0.81

449.3 2.33 0.62 28.53 0.23 27.97 -0.75

201.1 -17.44 -3.90 28.27 -0.08 44.25 -2.47

26.56 28.33 7.81 -4.21 0.78 -3.16 3.28

(3.52) (45.77) (26.52) -(13.11) (8.48) -(1.26) (5.90)

Expected Reporting Speeds

30

Table 2. Predicting Earnings News Panel A presents sample averages of earnings news proxies across EARS portfolios. EARS is the percentage change in a firm’s expected reporting speed relative to the same fiscal quarter in the prior year. A firm’s expected reporting speed is defined as the number of days between its expected earnings announcement date and fiscal period end date and the lagged reporting speed is defined as the difference between a firm’s realized earnings announcement date from the prior year and its corresponding fiscal period end date. Expected earnings announcement dates of firms expected to announce in month M + 1 are measured in the earnings calendar data on the final trading date of calendar month M . Observations are assigned to portfolios at the end of each calendar month, where the top (bottom) 30 percent are assigned to the ‘High’ (‘Low’) portfolio and the remaining are assigned to the ‘Mid’ portfolio. SU RP equals the actual EPS number reported in IBES minus the last consensus forecast available immediately prior to the announcement, and scaled by beginning-of-quarter assets and SU E is the standardized unexplained earnings, defined as the realized EPS minus EPS from four quarters prior, divided by the standard deviation of this difference over the prior eight quarters. The reported t-statistics correspond to time-series average monthly difference across High and Low EARS portfolios. Panel B (C) presents sample averages across EARS portfolios for the subsample of firms that are expected to report earnings at the same (different) reporting speed as in the prior year. A firm is categorized as announcing at the same speed (the Panel B sample) if it the difference between its expected earnings announcement date and fiscal period end date is within one day of the difference between a firm’s realized earnings announcement date from the prior year and its corresponding fiscal period end date. Panel D contains results from monthly Fama-MacBeth regressions of earnings news proxies on EARS and additional firm controls. LBM and SIZE are the log of one plus the book-to-market ratio and log of market capitalization, respectively. M OM EN is the cumulative market-adjusted return over the prior 12-months ending in month M . The parentheses contain t-statistics from the Fama-MacBeth regressions after Newey-West adjustments for autocorrelation up to 3 lags. The notations ***, **, and * indicate the coefficient is significant at the 1%, 5%, and 10% level, respectively. The sample for this analysis consists of 83,411 firm-month observations spanning 2006-2013. Panel A: Earnings Metrics Across EARS Portfolios EARS Portfolios High (Faster)

Mid

Low (Slower)

Low-High

SURP

0.071 (4.33)

-0.056 -(2.02)

-0.034 -(1.64)

0.105 (4.91)

SUE

0.043 (0.87)

-0.069 -(1.28)

-0.263 -(3.52)

0.305 (6.67)

Panel B: Same Reporting Speed Subsample EARS Portfolios High (Faster)

Mid

Low (Slower)

High-Low

SURP

0.117 (9.28)

-0.060 -(1.95)

-0.069 -(1.11)

0.186 (3.16)

SUE

0.064 (1.22)

-0.092 -(1.46)

-0.115 -(1.38)

0.179 (2.24)

Panel C: Changed Reporting Speed Subsample EARS Portfolios High (Faster)

Mid

Low (Slower)

High-Low

SURP

0.056 (2.61)

-0.247 -(1.70)

-0.029 -(1.34)

0.085 (3.39)

SUE

0.049 (0.87)

-0.323 -(1.21)

-0.293 -(3.63)

0.342 (5.94)

Expected Reporting Speeds

Table 2: [Continued] Predicting Earnings News Panel D: Fama-MacBeth Regressions of Earnings Metrics SURP EARS SIZE LBM MOMEN INT R2 (%)

SUE

(1)

(2)

(3)

(4)

0.002*** (4.31) – – – – – – -0.013 (-0.61)

0.002*** (3.62) 0.050*** (4.70) -0.093*** (-2.81) 0.002*** (3.66) -0.616*** (-3.77)

0.009*** (5.59) – – – – – – -0.081 (-0.84)

0.007*** (5.28) 0.063*** (4.15) -0.275*** (-4.70) 0.011*** (6.54) -0.747*** (-3.05)

0.431

3.692

0.762

5.842

31

Expected Reporting Speeds

32

Table 3. Predicting Future Returns Panel A presents results from monthly Fama-MacBeth regressions of raw returns on EARS and additional firm controls, where returns are measured in month M + 1 and all explanatory variables are measured in M . EARS is the percentage change in a firm’s expected reporting speed relative to the same fiscal quarter in the prior year. A firm’s expected reporting speed is defined as the number of days between its expected earnings announcement date and fiscal period end date and the lagged reporting speed is defined as the difference between a firm’s realized earnings announcement date from the prior year and its corresponding fiscal period end date. Expected earnings announcement dates of firms expected to announce in month M + 1 are measured in the earnings calendar data on the final trading date of calendar month M . Observations are assigned to portfolios at the end of each calendar month, where the top (bottom) 30 percent are assigned to the ‘High’ (‘Low’) portfolio and the remaining are assigned to the ‘Mid’ portfolio. LBM and SIZE are the log of one plus the book-to-market ratio and log of market capitalization, respectively. M OM EN is the cumulative market-adjusted return and V LT Y is the standard deviation of monthly returns over the prior 12-months ending in month M . RET (0) is the firm’s raw return in month M . The parentheses contain t-statistics from the Fama-MacBeth regressions after Newey-West adjustments for autocorrelation up to 3 lags. The notations ***, **, and * indicate the coefficient is significant at the 1%, 5%, and 10% level, respectively. Panel B (C) presents equal-weighted (value-weighted) sample averages of return metrics across EARS portfolios. RET(1) is the the firm’s raw return in the month of its expected earnings announcement month, M + 1. The Five-Factor Alpha is the intercept from a regression of raw returns minus the risk-free rate, regressed on the excess market return (MKTRF); two Fama-French factors (SMB, and HML); the Pastor-Stambaugh liquidity factor (LIQ), and the momentum factor (UMD). The One-Factor Alpha results from returns regressed MKTRF; the Three-Factor Alpha results from returns regressed on MKTRF, SMB, and HML; and the Four-Factor Alpha results from returns regressed on MKTRF, SMB, HML, and UMD. The reported t-statistics correspond to time-series average monthly difference across High and Low EARS portfolios. The sample for this analysis consists of 83,411 firm-month observations spanning 2006-2013. The sample for this analysis consists of 83,411 firm-month observations spanning 2006-2013. Panel A: Equal-Weighted Returns Across EARS Portfolios EARS Portfolios High (Faster)

Mid

Low (Slower)

High-Low

RET(1)

1.636 (2.40)

1.127 (1.64)

0.196 (0.30)

1.440 (6.23)

One-Factor Alpha

0.847 (3.29)

0.333 (1.27)

-0.551 -(2.10)

1.399 (6.04)

Three-Factor Alpha

0.832 (4.15)

0.314 (1.49)

-0.561 -(2.75)

1.392 (5.94)

Four-Factor Alpha

0.836 (4.35)

0.320 (1.70)

-0.555 -(2.91)

1.391 (5.91)

Five-Factor Alpha

0.843 (4.38)

0.321 (1.70)

-0.554 -(2.89)

1.397 (5.91)

Panel B: Value-Weighted Returns Across EARS Portfolios EARS Portfolios High (Faster)

Mid

Low (Slower)

High-Low

RET(1)

1.174 (2.23)

1.034 (1.89)

0.287 (0.52)

0.887 (2.62)

One-Factor Alpha

0.559 (3.03)

0.397 (1.98)

-0.313 -(1.16)

0.871 (2.54)

Three-Factor Alpha

0.569 (3.06)

0.379 (1.90)

-0.295 -(1.11)

0.864 (2.51)

Four-Factor Alpha

0.569 (3.05)

0.378 (1.89)

-0.295 -(1.10)

0.864 (2.50)

Five-Factor Alpha

0.554 (3.01)

0.370 (1.85)

-0.302 -(1.13)

0.857 (2.47)

Expected Reporting Speeds

33

Table 3: [Continued] Predicting Future Returns Panel C: Fama-MacBeth Regressions EARS REV RS SIZE LBM MOMEN VLTY RET(0) INT R2 (%)

(1)

(2)

(3)

(4)

(5)

(6)

0.033*** (5.04) – – – – -0.038 (-0.48) 0.663 (1.46) -0.004 (-0.44) -0.363* (-1.69) – – 1.995 (1.64)

– – 0.131*** (4.46) – – -0.066 (-0.78) 0.638 (1.47) -0.003 (-0.37) -0.361* (-1.67) – – 2.419* (1.91)

0.024** (2.41) 0.073* (1.70) – – -0.042 (-0.53) 0.618 (1.40) -0.004 (-0.49) -0.364* (-1.70) – – 2.106* (1.71)

0.022** (2.35) 0.078* (1.84) – – -0.025 (-0.32) 0.504 (1.30) -0.004 (-0.47) -0.418* (-1.75) -0.048*** (-3.07) 1.991 (1.57)

– – – – 0.077*** (3.71) -0.029 (-0.38) 0.538 (1.43) -0.004 (-0.41) -0.416* (-1.73) -0.048*** (-2.97) 1.975 (1.59)

0.030*** (2.72) – – 0.029 (0.73) -0.020 (-0.26) 0.569 (1.44) -0.004 (-0.43) -0.400* (-1.66) -0.049*** (-3.10) 1.810 (1.45)

4.435

4.345

4.722

5.807

5.575

5.908

Expected Reporting Speeds

34

Table 4. Sample Partitions Based on Changes in Reporting Speed Panel A (B) presents sample averages across EARS portfolios for the subsample of firms that are expected to report earnings at the same (different) reporting speed as in the prior year. A firm is categorized as announcing at the same speed (the Panel A sample) if it the difference between its expected earnings announcement date and fiscal period end date is within one day of the difference between a firm’s realized earnings announcement date from the prior year and its corresponding fiscal period end date. EARS is the percentage change in a firm’s expected reporting speed relative to the same fiscal quarter in the prior year. A firm’s expected reporting speed is defined as the number of days between its expected earnings announcement date and fiscal period end date and the lagged reporting speed is defined as the difference between a firm’s realized earnings announcement date from the prior year and its corresponding fiscal period end date. Expected earnings announcement dates of firms expected to announce in month M + 1 are measured in the earnings calendar data on the final trading date of calendar month M . Observations are assigned to portfolios at the end of each calendar month, where the top (bottom) 30 percent are assigned to the ‘High’ (‘Low’) portfolio and the remaining are assigned to the ‘Mid’ portfolio. RET(1) is the the firm’s raw return in the month of its expected earnings announcement month, M + 1. SU RP equals the actual EPS number reported in IBES minus the last consensus forecast available immediately prior to the announcement, and scaled by beginning-of-quarter assets and SU E is the standardized unexplained earnings, defined as the realized EPS minus EPS from four quarters prior, divided by the standard deviation of this difference over the prior eight quarters. OBS equals the number of firm-month observations. The reported t-statistics correspond to time-series average monthly difference across High and Low EARS portfolios. Panels C and D present results from monthly Fama-MacBeth regressions of raw returns on SAM EDAT E and additional firm controls when the sample is partitioned based on aggregate changes in firms’ reporting speeds. SA is the monthly average value of EARS minus its twelve-month historical average. SI equals the difference in number of firms that are expected to report faster relative to slower than in the prior year, scaled by the total of firms expected to report faster and firms expected to report slower within in a given calendar month, minus its twelve-month historical average. The ‘High’ (‘Low’) subsamples for SA and SI corresponds to months where the aggregate measures are greater than or equal to (less than) zero. SAM EDAT E is a binary variable that equals one if the firm is expected to announce earnings at the same speed as in the prior calendar year. LBM and SIZE are the log of one plus the book-to-market ratio and log of market capitalization, respectively. M OM EN is the cumulative market-adjusted return and V LT Y is the standard deviation of monthly returns over the prior 12-months ending in month M . RET (0) is the firm’s raw return in month M . The parentheses contain t-statistics from the Fama-MacBeth regressions after Newey-West adjustments for autocorrelation up to 3 lags. The sample for this analysis consists of 83,411 firm-month observations spanning 2006-2013. Panel A: Returns for Same Reporting Speed Subsample EARS Portfolios High (Faster)

Mid

Low (Slower)

High-Low

RET(1)

1.706 (2.32)

1.084 (1.59)

0.739 (0.99)

1.154 (2.44)

One-Factor Alpha

0.776 (2.49)

0.296 (1.15)

-0.006 -(0.01)

1.090 (2.27)

Three-Factor Alpha

0.873 (3.48)

0.279 (1.35)

-0.055 -(0.14)

1.000 (2.08)

Four-Factor Alpha

0.827 (3.52)

0.285 (1.52)

-0.054 -(0.14)

0.996 (2.05)

Five-Factor Alpha

0.837 (3.63)

0.285 (1.51)

-0.097 -(0.24)

1.053 (2.21)

Panel B: Returns for Changed Reporting Speed Subsample EARS Portfolios High (Faster)

Mid

Low (Slower)

High-Low

RET(1)

1.751 (2.40)

2.502 (1.14)

0.107 (0.16)

1.644 (4.70)

One-Factor Alpha

0.946 (2.77)

1.912 (1.09)

-0.636 -(2.32)

1.582 (4.50)

Three-Factor Alpha

0.920 (2.96)

1.019 (0.56)

-0.649 -(2.82)

1.568 (4.42)

Four-Factor Alpha

0.921 (2.95)

1.352 (0.75)

-0.643 -(2.98)

1.564 (4.46)

Five-Factor Alpha

0.923 (2.94)

1.252 (0.69)

-0.636 -(2.94)

1.559 (4.42)

Expected Reporting Speeds

Table 4: [Continued] Sample Partitions Based on Changes in Reporting Speed

Panel C: Regression of Returns on SAMEDATE Full Sample

Low SA

High SA

(1)

(2)

(3)

0.315*** (3.30) -0.069 (-0.97) 0.604 (1.43) -0.003 (-0.32) -0.378** (-2.01) 2.217* (1.91)

0.357* (1.82) -0.236*** (-2.68) 0.757 (0.89) 0.012*** (2.92) -0.729*** (-3.26) 4.850*** (3.28)

0.265 (1.18) 0.125** (2.24) 0.427* (1.86) -0.020 (-1.44) 0.028 (0.09) -0.836 (-0.72)

R2 (%)

4.130

4.215

4.031

OBS

83,411

47,682

35,729

SAMEDATE SIZE LBM MOMEN VLTY Intercept

Panel D: Regression of Returns on SAMEDATE Full Sample

Low SI

High SI

(1)

(2)

(3)

0.315*** (3.30) -0.069 (-0.97) 0.604 (1.43) -0.003 (-0.32) -0.378** (-2.01) 2.217* (1.91)

0.317* (1.73) -0.188 (-1.60) 0.702 (0.77) 0.009*** (2.73) -0.689*** (-3.45) 4.226** (2.03)

0.312 (1.35) 0.075 (0.70) 0.485 (1.56) -0.018 (-1.22) -0.002 (-0.01) -0.213 (-0.14)

R2 (%)

4.130

3.894

4.416

OBS

83,411

44,661

38,750

SAMEDATE SIZE LBM MOMEN VLTY Intercept

35

Expected Reporting Speeds

36

Table 5. Aggregate Earnings Surprises This table provides equal- and value-weighted average analyst-based earnings surprises, SU RP , across monthly subsamples. SU RP equals the actual EPS number reported in IBES minus the last consensus forecast available immediately prior to the announcement, and scaled by beginning-of-quarter assets. The first two rows compares SU RP across the first versus second and third months of an earnings season. The first month denotes January, April, July, and October; the second month denotes February, May, August, and November; and the third month denotes, March, June, September, and December. A firm is included in a given month if it is expected to announce in the month according to the earnings calendar measured at the end of prior calendar month. The second two rows compares SU RP across subsamples partitioned by SA, where SA is the monthly average value of EARS minus its twelve-month historical average. The ‘High’ (‘Low’) subsample corresponds to months where the value of SA is greater than or equal to (less than) zero. EARS is the percentage change in a firm’s expected reporting speed relative to the same fiscal quarter in the prior year. A firm’s expected reporting speed is defined as the number of days between its expected earnings announcement date and fiscal period end date and the lagged reporting speed is defined as the difference between a firm’s realized earnings announcement date from the prior year and its corresponding fiscal period end date. The bottom two rows compare SU RP across subsample partitioned by SI, where SI equals the difference in number of firms that are expected to report faster relative to slower than in the prior year, scaled by the total of firms expected to report faster and firms expected to report slower within in a given calendar month, minus its twelve-month historical average. The ‘High’ (‘Low’) subsample corresponds to months where the average value of SI is greater than or equal to (less than) zero. Expected earnings announcement dates of firms expected to announce in month M + 1 are measured in the earnings calendar data on the final trading date of calendar month M . Reported t-statistics are shown in parentheses underneath the sample averages. Panel B (C) present results from regressing monthly equal-weighted (value-weighted) average earnings surprises on indicator variables for aggregate earnings news, where calendar quarter fixed-effects are included throughout. The sample for this analysis consists of 95 calendar month observations spanning 2006-2013. Panel A: Average Earnings Surprises Equal-Weighted Averages

Value-Weighted Averages

Comparison

Base

Comparison

Difference

Base

Comparison

Difference

FMD vs. Non-FMD

0.071 (7.24)

-0.011 -(0.50)

0.084 (4.38)

0.109 (11.06)

0.060 (7.70)

0.049 (5.02)

High vs. Low SA

0.057 (4.11)

-0.055 -(1.48)

0.116 (4.06)

0.102 (11.09)

0.050 (5.99)

0.051 (4.64)

High vs. Low SI

0.061 (4.47)

-0.060 -(1.58)

0.124 (4.10)

0.104 (10.79)

0.048 (5.14)

0.056 (4.97)

Panel B: Regressions of Equal-Weighted Surprise FMD vs. Non-FMD High vs. Low SA High vs. Low SI R2 (%)

(1)

(2)

(3)

(4)

(5)

0.128*** (5.29) – – – –

– – 0.138*** (5.43) – –

– – – – 0.141*** (5.50)

0.050 (1.05) 0.097* (1.96) – –

0.034 (0.75) – – 0.114** (2.39)

20.139

22.738

25.227

23.719

25.711

Panel C: Regressions of Value-Weighted Surprise FMD vs. Non-FMD High vs. Low SA High vs. Low SI R2 (%)

(1)

(2)

(3)

(4)

(5)

0.040*** (3.49) – – – –

– – 0.044*** (3.86) – –

– – – – 0.047*** (4.35)

0.015 (0.63) 0.031 (1.31) – –

0.004 (0.19) – – 0.044** (1.99)

10.530

12.005

14.801

12.484

14.842

Expected Reporting Speeds

37

Table 6. Market-Level Regressions This table provides regression results of monthly S&P500 earnings growth, S&P500 index returns, and changes in the 10-year treasury yield. SA is the monthly average value of EARS minus its twelve-month historical average. The ‘High’ (‘Low’) subsample corresponds to months where the value of SA is greater than or equal to (less than) zero. EARS is the percentage change in a firm’s expected reporting speed relative to the same fiscal quarter in the prior year. A firm’s expected reporting speed is defined as the number of days between its expected earnings announcement date and fiscal period end date and the lagged reporting speed is defined as the difference between a firm’s realized earnings announcement date from the prior year and its corresponding fiscal period end date. SI equals the difference in number of firms that are expected to report faster relative to slower than in the prior year, scaled by the total of firms expected to report faster and firms expected to report slower within in a given calendar month, minus its twelve-month historical average. The ‘High’ (‘Low’) subsample corresponds to months where the average value of SI is greater than or equal to (less than) zero. Expected earnings announcement dates of firms expected to announce in month M + 1 are measured in the earnings calendar data on the final trading date of calendar month M . High SA & High SI is a dummy variable that equals one for months where SA and SI are greater than zero. Reported t-statistics are shown in parentheses underneath the sample averages, where calendar quarter fixed-effects are included throughout. The sample for this analysis consists of 95 calendar month observations spanning 2006-2013. S&P500 Earnings High vs. Low SA High vs. Low SI High SA & High SI R2 (%)

S&P500 Returns

Risk-Free Rate

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

2.435** (2.28) – – – –

– – 1.387 (1.45) – –

– – – – 2.247** (2.00)

0.008 (0.98) – – – –

– – -0.000 (-0.02) – –

– – – – 0.003 (0.41)

6.659* (1.72) – – – –

– – 8.739** (2.44) – –

– – – – 7.425** (2.04)

4.924

1.706

4.476

1.087

0.001

0.188

3.553

6.533

4.716

Expected Reporting Speeds

38

Table 7. Predicting Firm-Level Returns Using Aggregate Earnings News This table presents the difference-in-differences of firms’ raw return in their expected announcement month partitioned by their sensitivity to aggregate earnings news. Expected earnings announcement dates of firms expected to announce in month M + 1 are measured in the earnings calendar data on the final trading date of calendar month M . Aggregate earnings sensitivity is measured as the sensitivity of a firm’s monthly return in month t to the average analyst-based earnings surprise of all firms announcing in month t, measured over the 60 calendar months ending in month M 1. A firm is classified as ‘High’ (‘Low’) aggregate earnings sensitivity if it is above the median of all firms expected to announce in month M +1. Panel A compares returns across subsamples partitioned by SA, where SA is the monthly average value of EARS minus its twelve-month historical average. The ‘High’ (‘Low’) subsample corresponds to months where the value of SA is greater than or equal to (less than) zero. EARS is the percentage change in a firm’s expected reporting speed relative to the same fiscal quarter in the prior year. A firm’s expected reporting speed is defined as the number of days between its expected earnings announcement date and fiscal period end date and the lagged reporting speed is defined as the difference between a firm’s realized earnings announcement date from the prior year and its corresponding fiscal period end date. Panel B compares returns across subsample partitioned by SI, where SI equals the difference in number of firms that are expected to report faster relative to slower than in the prior year, scaled by the total of firms expected to report faster and firms expected to report slower within in a given calendar month, minus its twelve-month historical average. The ‘High’ (‘Low’) subsample corresponds to months where the average value of SI is greater than or equal to (less than) zero. Reported t-statistics based on the time-series of calendar quarters are shown in parentheses underneath the sample averages. The sample for this analysis consists of 83,411 firm-month observations spanning 2006-2013. Panel A: Returns Across High and Low SA Months Aggregate Earnings Sensitivity High

Low

High-Low

High SA

2.261 (1.57)

1.550 (1.22)

0.711 (2.47)

Low SA

0.715 (0.64)

0.836 (0.80)

-0.121 -(0.52)

Difference-in-Difference: [H0 : High vs. Low SA]

0.831 (2.33)

Panel B: Returns Across High and Low SI Months Aggregate Earnings Sensitivity High

Low

High-Low

High SI

1.998 (1.43)

1.357 (1.10)

0.641 (2.31)

Low SI

0.476 (0.44)

0.717 (0.69)

-0.241 -(1.10)

Difference-in-Difference: [H0 : High vs. Low SI]

0.882 (2.54)

Expected Reporting Speeds

39

Table 8. Regression of Returns on Aggregate Earnings News This table presents results from regressing firms’ raw returns on calendar-based measures of aggregate earnings news. RET(X) is the firm’s return X months after month M , where earnings calendar data is measured on the final trading date of calendar month M for all firms expected to announce earnings in month M + 1. Panel A measures expected aggregate news using SA measured in month M , where SA is the monthly average value of EARS minus its twelve-month historical average. EARS is the percentage change in a firm’s expected reporting speed relative to the same fiscal quarter in the prior year. A firm’s expected reporting speed is defined as the number of days between its expected earnings announcement date and fiscal period end date and the lagged reporting speed is defined as the difference between a firm’s realized earnings announcement date from the prior year and its corresponding fiscal period end date. Panel B measures expected aggregate news using SI measured in month M , where SI equals the difference in number of firms that are expected to report faster relative to slower than in the prior year, scaled by the total of firms expected to report faster and firms expected to report slower within in a given calendar month, minus its twelve-month historical average. Aggregate earnings sensitivity, M ACRO, is measured as the sensitivity of a firm’s monthly return in month t to the average analyst-based earnings surprise of all firms announcing in month t, measured over the 60 calendar months ending in month M 1. All regressions include controls for firm size, book-to-market, momentum, and return volatility. The reported t-statistics are based on two-way cluster robust standard errors, clustered by firm and quarter. ***, **, and * indicate significance at the 1, 5, and 10% level, respectively. The sample for this analysis consists of 83,411 firm-month observations spanning 2006-2013. Panel A: Regression Results on Earnings Season Month Indicators RET(1) EARS SA SA*MACRO MACRO SA*BETA BETA R2 (%)

RET(2)

RET(3)

(1)

(2)

(3)

(4)

(5)

(6)

0.021*** (4.35) 0.172 (0.97) 0.354** (2.43) 0.396 (1.45) – – – –

0.021*** (4.35) 0.174 (0.98) 0.355** (2.43) 0.396 (1.45) -0.002 (-0.15) 0.001 (0.02)

0.000 (0.09) -0.350 (-1.26) -0.364** (-2.52) 0.070 (0.30) – – – –

0.000 (0.04) -0.329 (-1.20) -0.357** (-2.47) 0.058 (0.25) -0.020 (-1.38) 0.044 (1.09)

0.005 (1.22) 0.281 (1.22) 0.031 (0.19) -0.041 (-0.17) – – – –

0.005 (1.19) 0.292 (1.28) 0.035 (0.21) -0.049 (-0.20) -0.011 (-0.56) 0.026 (0.49)

0.304

0.305

0.599

0.608

0.216

0.219

Panel B: Regression Results on Earnings Season Month Indicators RET(1) EARS SI SI*MACRO MACRO SI*BETA BETA R2 (%)

RET(2)

RET(3)

(1)

(2)

(3)

(4)

(5)

(6)

0.018*** (3.91) 2.776 (1.41) 4.975*** (2.73) 0.482* (1.82) – – – –

0.018*** (3.92) 2.702 (1.38) 4.951*** (2.71) 0.488* (1.84) 0.067 (0.69) -0.020 (-0.76)

0.002 (0.44) -4.800* (-1.73) -3.120* (-1.96) -0.026 (-0.11) – – – –

0.002 (0.41) -4.667* (-1.71) -3.071* (-1.93) -0.033 (-0.14) -0.127 (-1.17) 0.024 (0.79)

0.006 (1.35) 2.723 (1.13) -0.076 (-0.04) -0.026 (-0.10) – – – –

0.006 (1.35) 2.753 (1.15) -0.065 (-0.03) -0.028 (-0.11) -0.028 (-0.20) 0.006 (0.14)

0.473

0.474

0.800

0.806

0.193

0.194

Expected Reporting Speeds

40

Table 9. Predicting Changes in the VIX This table contains time-series regressions of changes in the CBOE volatility index, VIX, on the average absolute value of EARS and additional controls. The dependent variable is Log(V IX m+1 /V IX m ) which equals the log change in the VIX in month M +1, where M is the month of the earnings calendar. All independent variables are measured in month M. V IX m is the level of the VIX in month M . Expected earnings announcement dates of firms expected to announce in month M + 1 are measured in the earnings calendar data on the final trading date of calendar month M . F M D is a dummy variable that equals one for the first month of each calendar quarter. The first month denotes January, April, July, and October. EARS is the percentage change in a firm’s expected reporting speed relative to the same fiscal quarter in the prior year. A firm’s expected reporting speed is defined as the number of days between its expected earnings announcement date and fiscal period end date and the lagged reporting speed is defined as the difference between a firm’s realized earnings announcement date from the prior year and its corresponding fiscal period end date. The sample for this analysis consists of 95 calendar month observations spanning 2006-2013.

ABS(AGGEARS) Log(VIXm /VIXm VIXm FMD R2 (%)

1)

(1)

(2)

(3)

(4)

(5)

2.940*** (3.48) – – – – – –

2.733*** (3.40) -0.284** (-2.17) – – – –

2.772*** (3.82) – – -1.595*** (-4.23) – –

3.825*** (3.97) – – – – 4.059 (1.11)

2.985*** (3.17) -0.153 (-1.38) -1.481*** (-3.72) 1.460 (0.44)

8.818

16.241

37.963

9.769

40.239