The Timing of Quarterly ‘Pro Forma’ Earnings Announcements

(i) Studies on Pro Forma Earnings Disclosures

In response to the debate over pro forma reporting, several studies examine investors’ response to alternative earnings metrics and find that investors often pay more attention to adjusted earnings than to GAAP earnings (see Bradshaw and Sloan, 2002; Bhattacharya et al., 2003 and 2007; and Lougee and Marquardt, 2004). This evidence suggests that, on average, adjusted earnings measures provide more value-relevant information to investors and that the disclosure of such measures reflects managers’ attempt to better portray the firm's core earnings performance. However, recent research documents that some managers opportunistically use pro forma earnings disclosures to obfuscate the firm's economic performance and that some investors (especially less-sophisticated investors) are affected by this practice. Specifically, several US- and European-based studies find that exclusions of recurring expenses from adjusted earnings are motivated by managers’ incentives to meet or beat earnings benchmarks (Doyle and Soliman, 2005; Black and Christensen, 2009; Isidro and Marques, 2011; and Frankel et al., 2011), and that recurring exclusions have significant predictability for negative future cash flows which investors fail to fully understand (Doyle et al., 2003; Landsman et al., 2007; Chen, 2010; and Frankel et al., 2011).² Further, Brown et al. (2011) find that the disclosure of pro forma earnings as well as the magnitude of managers’ earnings adjustments both increase during periods of high investor sentiment (i.e., when investors hold overly optimistic beliefs) and that this behavior partly reflects managerial opportunism. Finally, recent evidence suggests that less-sophisticated investors are more likely to rely on adjusted earnings measures (Bhattacharya et al., 2007) and that such disclosures may bias the valuation judgments of not only this investor group (Frederickson and Miller, 2004; Elliott, 2006; and Allee et al., 2007), but also financial analysts (Andersson and Hellman, 2007).

We extend this stream of research by examining an important disclosure tool — the deliberate timing of the earnings announcement when it contains an adjusted earnings number; and further, whether managers’ timing decisions reflect opportunism or the attempt to better inform investors. While past research broadly examines the timing of earnings announcements, there is scant evidence on the timing of announcements containing manipulated earnings measures. Insights about the timing of manager-adjusted pro forma earnings measures are important given investors’ increased reliance on such measures and regulators’ continued scrutiny over discretionary pro forma reporting (Flemmons, 2009; and Leone, 2010). Further insight is also warranted given current emphasis on the timeliness and transparency of reported financial information (Flemmons, 2009).

(ii) Studies on the Timing of Earnings Announcements

Prior evidence on the timing of GAAP earnings is consistent with the notion that managers may deliberately time the release of earnings news. Although this evidence is mixed, several studies suggest that managers frequently accelerate the release of ‘good’ (favorable) earnings news and delay the release of ‘bad’ (unfavorable) earnings news (e.g., Givoly and Palmon, 1982; Kross and Schroeder, 1984; Chambers and Penman, 1984; Bagnoli et al., 2002; Begley and Fischer, 1998; and Cohen et al., 2007a). Moreover, prior empirical evidence indicates that investors react more aggressively to accelerated earnings announcements (relative to the expected announcement date), regardless of the nature of the news (e.g., Kross and Schroeder, 1984; and Bagnoli et al., 2002). In other words, investors appear to reward early good-news announcements and punish early bad-news announcements. Prior research also suggests that, in the case of an announcement delay, investors tend to adjust stock prices downward during the delay, presumably in anticipation of bad news. Taken together, this body of evidence suggests that the acceleration or delay of the earnings announcement has a strong effect on investors’ perceptions of the information content of reported earnings.

Prior theoretical and empirical research discuss several managerial motives for deliberately timing the release of earnings news. First, past research suggests that managers, in an attempt to be informative, may accelerate the release of both good and bad earnings news to either provide more timely information to investors or build the firm's reputation for transparent reporting (Graham et al., 2005). Second, prior studies argue that the delay of bad earnings news may reflect managers’ attempts to give (less-informed) investors more time to anticipate and digest complex information (Patell and Wolfson, 1982; and Chen and Mohan, 1994).

While these arguments suggest that the timing of earnings news may be motivated by management's desire to be informative, prior research indicates that the timing of earnings news could reflect managerial opportunism arising from various incentives. For instance, managers may accelerate the release of good news to preempt the leakage of information from other sources, thereby ensuring a more positive price impact (Kim and Verrecchia, 1991; and Bowen et al., 1992). Alternatively, managers may accelerate the release of bad news to enable the firm to position adverse news in the ‘best possible light’ (Graham et al., 2005) and thus, minimize litigation and/or reputational risks as well as avoid a large negative price shock (Skinner, 1994 and 1997; Francis et al., 1994; and Kasznik and Lev, 1995). Moreover, managers could deliberately time the release of earnings-related news to boost the value of their equity holdings in the firm (e.g., Kothari et al., 2009).³

Finally, and most closely related to our setting, existing theory predicts that managers will deliberately time their earnings releases when reported earnings are managed upward. Specifically, Trueman (1990) demonstrates that managers will accelerate the earnings announcement when they report unmanaged earnings, but that some managers may intentionally accelerate the announcement when they engage in earnings management as a pooling strategy. Further, Trueman (1990) predicts that managers are more likely to accelerate the release of upward-managed earnings given investors’ greater ex ante expectation that delayed earnings reports are intentionally manipulated or otherwise contain bad news. In other words, firms that engage in earnings management will accelerate their earnings releases in order to avoid a potential price discount resulting from investors’ conjecture that delayed earnings reports are either manipulated or contain bad news. Trueman (1990) also predicts that under certain circumstances managers could choose to delay the release of upward-managed earnings. For example, in some cases, the income management itself could take time, thereby resulting in a reporting delay. In other cases, managers could delay the earnings report to further assess whether the benefits of income management is likely to outweigh the expected costs. However, even in such cases, managers’ propensity to delay the earnings report is decreasing in investors’ expectation that delayed earnings reports are either managed upward or contain bad news.

Despite Trueman's (1990) theoretical predictions, there is limited empirical evidence on the relation between earnings announcement timing and the manipulation of reported earnings. Moreover, existing evidence on managers’ opportunistic timing behavior focuses primarily on the delay of unfavorable earnings news, whereas the timing of potentially manipulated earnings is largely ignored by prior research. We extend this line of research by examining the timing of announcements containing manager-adjusted pro forma earnings measures. Our focus on pro forma earnings announcements provides a unique setting since prior studies argue that the disclosure of adjusted earnings measures is an alternative and relatively low-cost earnings manipulation tool (Doyle and Soliman, 2005).

(iii) Hypothesis Development

In light of the competing explanations for firms’ disclosure timing decisions, it is unclear whether managers, on average, accelerate or delay their earnings announcements when they report a pro forma earnings figure in the earnings press release. It is also unclear whether the acceleration or delay of pro forma earnings information is most likely to reflect managers’ information-related or opportunistic motives. Therefore, in the formulation of our hypotheses, we discuss two competing views that could explain the deliberate timing of pro forma earnings information: the informative and opportunistic views.

Our first hypothesis posits that managers time the release of their earnings announcements in quarters in which they report an adjusted (pro forma) earnings metric differently from quarters in which they do not. On the one hand, managers could accelerate or delay the release of adjusted earnings news for information-related reasons (the ‘informative view’). For instance, managers could accelerate pro forma earnings news in an attempt to provide value-relevant information to investors in a timely manner. In addition, managers could delay the release of pro forma earnings news to provide more time to analyze and decide on the specific earnings exclusions that provide the most value-relevant adjusted earnings figure or to give investors more time to anticipate and digest complex earnings information.

On the other hand, managers could intentionally accelerate or delay their pro forma earnings announcements for opportunistic reasons (the ‘opportunistic view’). As discussed previously, prior evidence suggests that some managers opportunistically use pro forma exclusions to embellish the firms’ reported earnings performance, irrespective of the information content of these excluded items (Doyle et al., 2003; Doyle and Soliman, 2005; Black and Christensen, 2009; Isidro and Marques, 2010; Frankel et al., 2011; and Brown et al., 2012). Consistent with Trueman's (1990) theoretical predictions, opportunistic managers could choose to accelerate the release of inflated pro forma earnings news in an effort to mitigate the potential price discount arising from investors’ conjecture that the earnings report is manipulated using income-increasing pro forma exclusions. Also, opportunistic managers could accelerate the release of inflated pro forma earnings news to gain private benefits such as to boost their equity holdings or minimize reputational risks (Kim and Verrecchia, 1991; Bowen et al., 1992; and Kothari et al., 2009). However, managers who use pro forma adjustments to embellish reported earnings could still choose to delay the earnings release to either assess the expected benefits and costs of income management or to provide more time to coordinate and analyze the pro forma adjustments (Trueman, 1990). Given these conflicting arguments, our first hypothesis (presented in the null form) does not predict a particular direction of the earnings announcement timing:

Our first hypothesis (H₁) addresses whether managers, on average, accelerate or delay their earnings announcement when it contains a pro forma earnings figure versus when it does not. If we find evidence of differential timing of announcements containing pro forma earnings information, then our second hypothesis (H₂) examines the relation between announcement timing and the exclusion of recurring (operating) earnings items. Prior research suggests that managers may exclude recurring items to better portray core earnings information (Bhattacharya et al., 2003; and Lougee and Marquardt, 2004). However, some studies argue that exclusions of recurring items such as depreciation and amortization costs and R&D expenses are ‘low-quality’ adjustments that persist in future periods, and therefore, are more likely to reflect managerial opportunism (Doyle et al., 2003; Elliott, 2006; Christensen, 2007; Choi et al., 2007; Kolev et al., 2008; Black and Christensen, 2009; and Frankel et al., 2011). Given these arguments, we cannot, on an ex ante basis, explicitly align the use of recurring exclusions with managers’ attempt to inform or mislead investors.^4,5 We also state our second hypothesis in the null form:

As discussed previously, managers’ timing behavior could reflect opportunism (the opportunistic view) or the desire to better inform investors (the informative view). In the spirit of prior studies (Doyle et al., 2003; Frankel et al., 2011; Kolev et al., 2008; and Brown et al., 2012), we attempt to disentangle managers’ motives by examining the average quality of managers’ recurring exclusions, conditional on the timing of the pro forma earnings announcement. We use the term ‘quality’ to refer to the transitory nature or ability of pro forma earnings exclusions to predict future earnings performance. If the timing of pro forma earnings announcements is opportunistically motivated, then we expect that the recurring exclusions contained in deliberately-timed pro forma announcements will be of relatively lower quality (i.e., less transitory). Alternatively, if the timing of pro forma earnings news is motivated by management's desire to better inform investors, then we expect that the recurring exclusions contained in deliberately-timed pro forma announcements will be of higher quality (i.e., more transitory). Given these competing arguments, we do not make directional predictions regarding the association between announcement timing and the average quality of managers’ recurring exclusions. We likewise present our third hypothesis in the null form:

Our fourth and final hypothesis provides further evidence on managers’ motives by examining the relation between announcement timing and the transparency of the pro forma earnings adjustments disclosed within the earnings press release. In particular, we examine the relation between announcement timing and the reconciliation of the differences between the pro forma and GAAP earnings figures within the earnings press release.⁶ Accounting regulators and other critics of pro forma reporting argue that managers may use their discretion to influence investors’ perceptions by obscuring the differences between the pro forma and GAAP earnings metrics. Given this concern, prominent industry groups and the SEC have issued guidelines requiring firms to provide a reconciliation of the pro forma and GAAP measures within the earnings press release.⁷ Notwithstanding these guidelines, companies continue to exercise discretion over the characteristics of pro forma reconciliations in quarterly earnings press releases. In fact, prior research suggests that after the adoption of regulatory guidelines a significant proportion of firms that report pro forma earnings still fail to provide a quantitative reconciliation (see Marques, 2010). Further, for firms that do provide a quantitative reconciliation, there is substantial variation in the disclosure transparency of the reconciliation.⁸

Prior research also suggests that more transparent disclosures lead to greater detection of earnings management (e.g., Hirst and Hopkins, 1998). Accordingly, managers often lobby for and choose less transparent disclosure formats. In addition, firms with a history of earnings management tend to choose less transparent reporting formats (Lee et al., 2006). Prior evidence is consistent with at least some managers believing that earnings management in conjunction with less transparent disclosure will improve stock price and not harm the firm's disclosure reputation, while earnings management with more transparent disclosure will harm both the former and the latter (Hunton et al., 2006; and Bamber et al., 2010).

Based on the preceding discussion, if the timing of pro forma earnings announcements is opportunistically motivated, then we expect a significant association between earnings announcement timing and the use of less transparent reconciliation formats. This result would provide further evidence of managers’ intentional timing of earnings news that is potentially manipulated and presented using less transparent disclosures. Alternatively, if the timing of pro forma earnings news reflects managers’ attempts to be more informative, then we expect no association between earnings announcement timing and the use of less transparent reconciliation formats. Given these competing arguments, we do not make directional predictions regarding the association between announcement timing and the use of less transparent reconciliation formats. We therefore present our fourth and final hypothesis in the null form:

(i) Data and Sample Selection

We hand-collect a comprehensive sample of quarterly earnings press releases by US firms that contain a manager-adjusted pro forma earnings per share (EPS) figure by searching the PR Newswire and Business Wire on LexisNexis for the calendar period, 1998 Q1 to 2006 Q1.⁹ This sample extends Bhattacharya et al.'s (2007) sample and is similar to Brown et al.'s (2012) sample. We begin our hand collection in 1998 Q1 since manager-adjusted pro forma earnings were not widely reported in the US prior to 1998 (see Bhattacharya et al., 2004, for a detailed analysis of time trends in pro forma reporting). Table 1 summarizes our sample selection procedures.

A typical press release in our sample contains the GAAP EPS figure for the current quarter, a pro forma earnings number (an adjusted EPS measure voluntarily disclosed by managers), and various other details. Appendix A provides examples of earnings press releases that disclose a pro forma EPS number. From our LexisNexis search, we find 17,199 pro forma earnings press releases for firms with non-missing identifiers in CRSP and Compustat, from which we gather stock returns and financial statement information, respectively. Firms with missing identifiers are primarily foreign firms, public debt-private equity firms, and firms that undergo re-organizations, mergers and acquisitions, name changes, or private buyouts.

To conduct our empirical analyses, we require each firm-quarter to have an actual quarterly earnings report date (RDQE) from Compustat for quarter q of year t and year t–1 over our sample period. We eliminate firm-quarters in which the actual announcement date is less than 7 calendar days or more than 45 calendar days (90 calendar days) after the fiscal quarter-end date for interim (annual) announcements. We also exclude earnings announcements where the fiscal quarter-end date reported in the pro forma press release is different from the quarter-end date reported in Compustat. We further eliminate earnings announcements where the pro forma report date is more than ±1 day from the Compustat quarterly earnings report date (RDQE). These restrictions ensure that our sample is free of data entry errors.

These data criteria result in a reduced sample of 11,215 pro forma announcements for 2,381 US firms. To mitigate potential selection biases, we do not impose further data restrictions when assessing H₁ in a univariate setting. For our multivariate analyses, we require firm-quarters to have earnings forecast and analyst-adjusted street earnings data available in the I/B/E/S split-unadjusted database. We also require firm-quarters to have a non-negative book value of equity and available data for our regression variables in the CRSP, Compustat, and Thomson Reuters 13f databases. These requirements result in 8,127 quarterly pro forma announcements issued by 2,134 firms.

We use a ‘within-firm’ research design to test H₁. This research design compares all quarters that the firm elects to report an adjusted earnings measure in its quarterly earnings press release (‘pro forma quarters’) with those quarters when the same firm does not disclose an adjusted earnings measure (‘non-pro-forma quarters'), i.e., all quarters in which our LexisNexis search does not identify a pro forma earnings press release. This research design uses each firm as its own control and thus minimizes the effects of cross-sectional variation in firms’ pro forma reporting decisions. For the same set of firms, we collect the quarterly earnings report date (RDQE) for all non-pro-forma quarters in year t and the corresponding fiscal quarter in year t–1, resulting in a matched within-sample of 11,215 pro forma and 39,842 non-pro-forma quarters for 2,381 firms for the univariate analysis of H₁.¹⁰ For our multivariate analyses, the reduced matched within-sample consists of 8,127 pro forma and 30,587 non-pro-forma quarters for 2,134 firms.

(ii) Variable Measurement

(a) Timing of the Earnings Announcement

We define earnings announcement timing using the following extrapolative model that is widely used in the literature (e.g., Givoly and Palmon, 1982; Chambers and Penman, 1984; Kross and Schroeder, 1984; Begley and Fischer, 1998; and Bagnoli et al., 2002):

(1)

where EAD_LAG is the number of trading days (excluding weekends and holidays) between the fiscal end date of quarter q and the actual earnings announcement date (RDQE) for firm i in quarter q of fiscal year t as reported in Compustat.¹¹ This model estimates the firm's deviation from its expected announcement timing, where the expected announcement timing for the current quarter is assumed to be the same as the reporting lag in trading days for the corresponding quarter of year t–1. More specifically, this model uses the actual announcement date for quarter q of year t–1 to estimate the expected report date for the current quarter.¹² Hence, a negative (positive) DELAY indicates that a firm's actual earnings announcement is earlier (later) than the estimated expected date. In robustness tests, we find similar results using four alternative time-series estimation models.¹³ We however, report results based on equation (1) to facilitate comparison with prior studies. Appendix B summarizes the specification of DELAY and all of the measures outlined below.

(b) Characteristics of Pro Forma Earnings Aannouncements

We measure several attributes of firms’ adjusted earnings information: the forecast error, the magnitude of managers’ recurring and nonrecurring exclusions, the transparency of the pro forma reconciliation, and specific meet/beat earnings benchmarks. Note that the respective pro forma variables are set to zero for quarters in which the firm does not report a pro forma earnings measure.

Forecast error. We first calculate three quarterly EPS measures: (1) GAAP diluted EPS from continuing operations (EPS_GAAP-OP); (2) the analyst-adjusted street EPS from the I/B/E/S split-unadjusted actual file (EPS_IBES); and (3) the hand-collected pro forma diluted EPS (EPS_PROFORMA).¹⁴ We then calculate three measures of unexpected earnings news or forecast error based on each of the three actual earnings metrics. Forecast errors based on GAAP operating EPS (FE_GAAP-OP), I/B/E/S actual EPS (FE_IBES), and pro forma EPS (FE_PROFORMA) are calculated by subtracting the mean (consensus) analyst forecast from each of the three actual earnings metrics. The mean analyst forecast is calculated for each firm using all split-unadjusted forecasts issued within 90 days prior to the earnings announcement date.

Managers’ recurring and nonrecurring earnings exclusions. As illustrated in Figure 1, we calculate the total amount per share of recurring and nonrecurring items excluded by management (TOTAL_EXCL) as the difference between EPS_PROFORMA and GAAP diluted EPS after extraordinary items (EPS_GAAP-AXI).¹⁵ We then decompose TOTAL_EXCL into the following components: (1) below-the-line items per share (BELOW_EXCL), calculated as GAAP diluted EPS before extraordinary items (EPS_GAAP-BXI) minus EPS_GAAP-AXI; (2) special items per share (SPECIAL_EXCL), calculated as EPS_GAAP-OP minus EPS_GAAP-BXI; and (3) managers’ exclusions of recurring items (RECUR_EXCL), calculated as TOTAL_EXCL minus the sum of SPECIAL_EXCL and BELOW_EXCL. To prevent the artificial creation of recurring exclusions, we follow Kolev et al., (2008) and set SPECIAL_EXCL and BELOW_EXCL to zero if TOTAL_EXCL equals zero.

Note that positive values of the earnings exclusion variables indicate that income-decreasing expenses were excluded from the respective GAAP earnings figure; whereas negative values indicate that income-increasing gains were excluded. For the sake of simplicity, we assume that managers exclude all nonrecurring items — i.e., below-the-line and special items — when arriving at the pro forma figure. Thus, any additional exclusions made by management are generally recurring or operating items such as depreciation, amortization, R&D, and interest expenses (Bhattacharya et al., 2004; and Black and Christensen, 2009).

Reconciliation transparency. We use the following five-point index to measure the transparency of the pro forma reconciliation provided within the earnings press release:

Form of Reconciliation	Transparency Index
Side-by-side	5 (most transparent)
Top-to-bottom	4
End of income statement	3
Text or footnote only	2
No reconciliation provided	1 (least transparent)

Prior evidence suggests that a simultaneous reconciliation of the GAAP and pro forma earnings figures in adjacent tabular columns is more transparent and easier to evaluate compared to a sequential presentation of the reconciled earnings items (Elliott, 2006; and Zhang and Zheng, 2010; see Appendix A for an example). We therefore classify side-by-side presentations as the most transparent forms of reconciliation. We code top-to-bottom reconciliations as the next highest transparency level since these reconciliations are separated from and presented directly after the income statement. The presentation of the reconciliation at the end of the income statement is similar to a top-to-bottom reconciliation. However, end-of-statement reconciliations are not separated from the income statement and in many cases appear to be a continuous income statement, which could be misleading to investors. We therefore treat end-of-statement reconciliations as less transparent and of lower informational value than top-to-bottom reconciliations. Finally, we classify text or footnote reconciliations and the absence of a reconciliation as the next lowest transparency levels.¹⁶ For ease of interpretation in our empirical analyses, we transform the transparency scores into a binary variable, LOW_RECON, that equals one if the transparency score of the pro forma earnings reconciliation is less than or equal to the median score of 3, and zero otherwise.

Earnings benchmarks. Following prior studies (e.g., Lougee and Marquardt, 2004; and Bhattacharya et al., 2003 and 2007), we measure whether GAAP operating earnings (EPS_GAAP-OP) is below or misses the consensus analyst forecast (NEG_FE). Specifically, we code NEG_FE as one if FE_GAAP-OP is negative, and zero otherwise. We also measure whether firms miss or fail to report a GAAP operating profit (GAAPOP_LOSS). We code GAAPOP_LOSS as one if EPS_GAAP-OP is a loss, and zero otherwise.

(iii) Descriptive Evidence

Table 2 provides summary statistics for the restricted sample of firm-quarters with available data for all our regression variables. We present separate statistics for the subsamples of pro forma and non-pro-forma quarters (8,127 and 30,587 firm-quarters, respectively). We do not report the pro forma variables for non-pro-forma quarters since these variables are set to zero by default. For ease of interpretation, we do not scale the per share values of the various earnings, forecast errors, and pro forma exclusions variables. However, our inferences are unchanged when we scale by the closing price five days before the earnings announcement date (Christie, 1987). For the pro forma sub-sample, the mean (median) EPS_PROFORMA is 35 (26) cents per share, while the mean (median) EPS_IBES is about 30 cents (23) per share and the mean (median) EPS_GAAP-OP is about 23 (20) cents per share. Consistent with prior evidence (Bhattacharya et al., 2003 and 2004; Lougee and Marquardt, 2004; and Bowen et al., 2005), these results indicate that firms typically report a pro forma earnings figure that is more favorable than the GAAP or street earnings figure (these differences are statistically significant at the 1% level). In fact, unreported statistics indicate that about 70% of our pro forma subsample report an adjusted earnings measure that exceeds the GAAP operating EPS figure. The forecast errors yield a similar story; firms often convert a negative GAAP operating earnings surprise (mean FE_GAAP-OP of −6 cents per share) into a positive pro forma surprise (mean FE_PROFORMA of 7 cents per share).

Table 2 also reports summary statistics for the magnitude of managers’ recurring and nonrecurring exclusions. The mean (median) TOTAL_EXCL is 0.2274 (0.0700), indicating that the mean (median) difference between the pro forma earnings figure and the bottom-line GAAP figure is about 23 (7) cents per share. Managers frequently argue that they exclude primarily nonrecurring items. However, we find that a large fraction of the pro forma subsample have zero values for BELOW_EXCL (80%) and SPECIAL_EXCL (46%). In contrast, only 18% of the sample have zero values for RECUR_EXCL with the mean (median) of RECUR_EXCL being about 12 (3) cents per share. This evidence indicates that managers are more likely to exclude recurring expenses than nonrecurring expenses.

Interestingly, we find that 50% of the pro forma subsample uses less transparent reconciliation formats (LOW_RECON). Consistent with Marques (2010), we find (not tabulated) that 40% of sample firms provide no reconciliation in the pre-SOX period and about 10% continue this practice after SOX, despite the issuance of industry and regulatory guidelines requiring the quantitative reconciliation of pro forma information (NIRI, 2001; and SEC, 2001 and 2003). We also find that GAAP operating earnings falls short of the analyst consensus forecast (NEG_FE) in about 55% of the pro forma quarters versus 39% for non-pro-forma quarters. We find a similar, but smaller difference, when we compare GAAPOP_LOSS across the two sub-samples (23% versus 16%). Hence, it appears that missing the analyst consensus is a stronger motivation for reporting pro forma earnings than turning a GAAP operating loss into a pro forma profit. Table 2 presents comparative statistics for our controls, though we do not discuss them in detail for the sake of brevity.

(i) Hypothesis 1 (H₁)

Using the unrestricted sample, we first conduct univariate tests of the differential timing of earnings announcements released during quarters in which firms report a pro forma earnings measure relative to quarters in which they do not. Panel A of Table 3 presents statistical comparisons of DELAY for the full sample of 51,057 firm-quarters, comprised of 11,215 pro forma and 39,842 non-pro-forma quarters. The mean DELAY for pro forma quarters is –0.0263, which is significantly different from the mean DELAY of 0.1542 for non-pro-forma quarters (t-statistic =–3.46). This result suggests that managers, on average, announce earnings earlier than expected when they disclose an adjusted earnings measure. To ensure that the statistical significance of this result is not simply an artifact of a large sample size, we conduct a bootstrapping exercise based on 1,000 iterations of randomly selecting 1,000 observations from our sample and then repeating the univariate comparisons. The bootstrap z-statistics confirm the significance of our tests.

In Panel A, we also note that the difference between the two sets of quarters may appear economically small. We however, attribute this result to the substantial clustering of announcements around the expected date (i.e., around DELAY = 0). This clustering, though similar to that found in prior studies (e.g., Bagnoli et al., 2002), is much lower in our sample. Nonetheless, we conduct two empirical procedures to address this issue. First, we examine the empirical distributions of DELAY for both sub-samples using the Kolmogorov-Smirnov (K-S) test, which is not sensitive to sample clustering at the mean or median (Sheskin, 2004). The K-S test confirms that the distribution of DELAY is significantly different between the two sets of quarters. Second, we repeat our tests after eliminating observations with an absolute DELAY less than three days (i.e., earnings announcements that are less than three days late or less than three days early). The results from this reduced sample (see Panel B) portrays a much more striking picture of the differential timing of earnings announcements during pro forma quarters versus non-pro-forma quarters.

Panel C of Table 3 examines the difference in DELAY for the pre- and post-SOX periods for the full sample, while Panel D examines the pre- and post-SOX differences for the subsample of firm-quarters with an absolute DELAY ≥ 3 days. In Panel C, we find that in the pre-SOX period the mean DELAY is –0.2021 for pro forma quarters, which is significantly different from zero (t-statistic =–3.01). This result contrasts dramatically with prior studies that document a mean announcement delay that is several times smaller and not significantly different from zero during years prior to the passage of SOX. For instance, Begley and Fischer (1998) report a mean delay of 0.06 for the 1982–1992 period. Using data that partially overlaps our pre-SOX period, Bagnoli et al., (2002) find a mean delay of zero over the 1995–1998 period and a mean of −0.01 during 1998.¹⁹ In the pre-SOX period, the mean DELAY for pro forma quarters is significantly lower than that for non-pro-forma quarters (–0.2021 versus −0.0042, t-statistic =−2.57). From Panel D, the magnitude of this pre-SOX difference is even greater (−0.0805 versus 0.5452, t-statistic =−3.14) when we exclude announcements clustered around the expected earnings date. In both panels, we observe a significant post-SOX increase in the average DELAY for both pro forma and non-pro-forma quarters. However, we find that the post-SOX increase in the average DELAY for pro forma quarters is not statistically different from that for non-pro-forma quarters, consistent with Pawlewicz (2011). Together, these results indicate that, while heightened regulatory scrutiny may have led to an increase in reporting delay, firms continue to accelerate their earnings announcements in quarters in which they report an adjusted earnings figure.²⁰

Using the restricted sample, we next investigate H₁ based on the multivariate model below:

(2)

where PROFORMA equals one if the firm reports pro forma earnings information in quarter q, and zero otherwise. Controls denotes a vector of control variables as detailed in Appendix B. We include FE_GAAP-OP to control for the magnitude of the GAAP operating earnings surprise. We find similar results (untabulated) when we include the I/B/E/S forecast error (FE_IBES) as opposed to FE_GAAP-OP to mitigate potential ‘errors-in-variables' biases arising from the use of an actual EPS metric that is not adjusted for the same items as the analyst forecast (Bradshaw and Sloan, 2002; Abarbanell and Lehavy, 2007; and Cohen et al., 2007b).

Column 1 of Table 4 presents the regression results for equation (2) based on the full restricted sample. Column 2 presents the results for the subsample of firm-quarters with an absolute DELAY ≥ 3 days. We winsorize all continuous variables at the 1% and 99% levels to reduce the effect of extreme outliers. The t-statistics (in parentheses) for all coefficients are based on robust standard errors corrected for heteroskedasticity and clustering by firm (Petersen, 2009). We also include fixed fiscal-quarter and Fama-French industry effects to control for unobserved time and within-industry factors. Again, for ease of interpretation, we do not deflate the per share values of FE_GAAP-OP. However, our inferences are unchanged when we scale by stock price five days before the earnings announcement date.

The evidence reported in Table 4 is consistent with our univariate results for H₁. Specifically, the estimated coefficient on PROFORMA is significantly negative in both regressions, indicating that firms tend to accelerate their earnings announcements when they disclose an adjusted earnings metric. Consistent with the good-news-early hypothesis, the estimated coefficient on FE_GAAP-OP is significantly negative, suggesting that firms tend to accelerate the release of favorable earnings news. Among our controls, we find that firms announce earnings later than expected if they report a GAAP operating loss (GAAPOP_LOSS), have negative GAAP earnings news (NEG_FE), engage in one-time events (SPECIAL_CHRG), have a high book-to-market ratio (BOOKMKT), are larger (SIZE), face high litigation risks (LITIGATE), and provide earnings guidance (GUIDANCE). In addition, firms with high earnings volatility (STDROA) tend to accelerate their earnings releases. Lastly, in unreported tests, we again find no significant difference in the association between PROFORMA and DELAY in the pre- versus post-SOX period, consistent with the univariate results reported in Table 3.²¹

(ii) Hypothesis 2 (H₂)

The results for H₁ suggest that managers tend to accelerate their earnings announcements in quarters in which they disclose a pro forma earnings metric relative to quarters in which they do not. Given this evidence, H₂ addresses the relation between announcement timing (DELAY) and the level of managers’ total recurring exclusions (RECUR_EXCL). To ensure that our results are not affected by the decision to report or not report pro forma earnings in a given quarter, we follow Doyle et al., (2003) and restrict this and all subsequent analyses to the subsample of pro forma quarters.²² We examine H₂ using the following model:

(3)

The specification in equation (3) is similar to that in equation (2), but instead replaces the PROFORMA variable with RECUR_EXCL. We include LAG_PROFORMA, GAAPOP_LOSS, NEG_FE, and our set of control variables (Controls) to account for cross-sectional differences in firms’ announcement timing and pro forma adjustment decisions. Note that FE_GAAP-OP incorporates managers’ nonrecurring exclusions (see Figure 1) and thus, implicitly controls for the level of below-the-line and special items exclusions.

Column 1 of Table 5 presents the results for equation (3) for all pro forma firm-quarters, while column 2 presents the results for pro forma firm-quarters with an absolute DELAY ≥ 3 days. We report t-statistics based on robust standard errors corrected for heteroskedasticity and clustering by firm (Petersen, 2009). We present results based on the unscaled values of FE_GAAP-OP and RECUR_EXCL for ease of interpreting the estimated coefficients. In column 1, we find significantly negative coefficients for FE_GAAP-OP (β₁=−1.4383, t-statistic =−2.83) and RECUR_EXCL (β₂=−1.0972, t-statistic =−2.08). To gauge the economic significance of these results, we note that a one-standard-deviation increase in the level of managers’ recurring exclusions (which is about 97 cents for the pro forma subsample) results in about a 1 day increase in the acceleration of the earnings report. Since the total pro forma earnings surprise is the sum of FE_GAAP-OP and RECUR_EXCL, we note that the sum of β₁ and β₂ suggests an average announcement acceleration of close to 3 days for every $1 increase in the total pro forma earnings surprise (β₁+β₂=−2.5355, t-statistic =−2.77). The effect also appears economically significant when we consider that a one-standard-deviation increase in the total pro forma earnings surprise (FE_PROFORMA) is about 67 cents for the pro forma subsample.

In column 2, the coefficients on FE_GAAP-OP (β₁=−3.5263, t-statistic =−2.91) and RECUR_EXCL (β₂=−1.4396, t-statistic =−1.97) are significantly more negative when we exclude announcements clustered around the expected release date. The economic significance of these results are even more striking when we consider that a one-standard-deviation increase in the total pro forma earnings surprise (FE_PROFORMA) is about 81 cents for this restricted sample. That is, the sum of β₁ and β₂ for this restricted sample suggests an average announcement acceleration of close to 5 days for every $1 increase in the total pro forma earnings surprise (β₁+β₂=−4.9659, t-statistic =−2.83). The control variables in both columns 1 and 2 are consistent with our results in Table 4, though we do not discuss them for brevity.

In untabulated tests, we still find a significantly negative association between DELAY and RECUR_EXCL when we directly control for the level of managers’ below-the-line (BELOW_EXCL) and special items exclusions (SPECIAL_EXCL). However, we find insignificant associations between DELAY and each of these nonrecurring exclusions. This evidence suggests that the level of nonrecurring adjustments does not influence the timing of pro forma earnings announcements. Also, we re-estimate equation (3) after deconstructing managers’ recurring exclusions into two components: (1) analysts’ recurring exclusions from street earnings (ANAL_RECUR), i.e., those recurring exclusions on which both managers and analysts agree, calculated as EPS_IBES minus EPS_GAAP-OP; and (2) managers’ incremental recurring exclusions (INCR_RECUR), i.e., those recurring exclusions on which managers and analysts disagree, calculated as RECUR_EXCL minus ANAL_RECUR. Consistent with H₂, the re-estimated results (not tabulated) indicate a significantly negative association between DELAY and INCR_RECUR.

Prior studies argue that the exclusion of recurring gains is more likely to reflect information-related motives (Christensen, 2007; and Curtis et al., 2011). On the other hand, prior evidence suggests that opportunistic managers may delay the earnings release when they manipulate earnings downwards (Chai and Tung, 2002). To investigate these issues, we re-estimate equation (3) after including a separate slope coefficient for negative values of RECUR_EXCL. Our results (not tabulated) show no significant association between announcement timing and the exclusion of recurring gains, indicating that the acceleration of pro forma earnings announcements is more prevalent when managers exclude recurring expenses in arriving at the pro forma earnings figure.

(iii) Hypothesis 3 (H₃)

The results in Table 5 suggest that some managers accelerate their pro forma earnings announcements when they make larger income-increasing recurring adjustments. To disentangle managers’ motives, H₃ investigates the average quality of managers’ recurring exclusions, conditional on the timing of the earnings announcement, and controlling for managers’ nonrecurring exclusions and other correlated factors. Following prior studies (Doyle et al., 2003; Frankel et al., 2011; Kolev et al., 2008; and Brown et al., 2011), we define ‘high-quality' earnings exclusions as those that are transitory or have the least predictive power for future earnings performance, where future earnings performance is operationalized as GAAP operating earnings per share (EPS_GAAP-OP) summed over four quarters beginning in quarter q+ 1 (FUTURE_GAAP-OP). Conversely, we define ‘low-quality' exclusions as those that persist in future periods and therefore, are not fully transitory or unimportant as some managers claim.

We estimate the regression model below, which follows directly from the specifications used in Doyle et al. (2003), Kolev et al. (2008) and Brown et al. (2012):

(4)

Exclusions is a vector of the recurring and nonrecurring components of managers’ pro forma earnings exclusions (i.e., RECUR_EXCL, SPECIAL_EXCL and BELOW_EXCL), and EARLY is an indicator variable that equals one for unexpectedly early announcements (DELAY < 0), and zero for on-time or late announcements (DELAY ≥ 0). We base our analysis on the binary variable, EARLY, to better interpret the model's interaction terms; however, our inferences are unchanged when we use the continuous DELAY variable.²³Controls is a vector of control variables: GAAPOP_LOSS, BOOKMKT, SIZE, STDROA and POSTSOX. If the acceleration of pro forma earnings announcements is more consistent with managerial opportunism, then we expect statistically significant coefficients on δ_j, indicating that the quality of managers’ pro forma exclusions is lower (i.e., less transitory) for accelerated earnings announcements.

Panel A of Table 6 presents the regression results for equation (4). Panel B presents the results of F-tests to examine the null hypotheses that the predictability of the pro forma earnings figure and the various earnings exclusions for early announcements are equal to zero (i.e., δ_i+δ_j= 0, i= [1, 4], j= [6, 9]). We again present separate regression results for the full sample of pro forma quarters (column 1) and the restricted sample with an absolute DELAY ≥ 3 days (column 2). The sample size for each regression reduces slightly due to missing values of FUTURE_GAAP-OP. We scale FUTURE_GAAP-OP, EPS_PROFORMA, and all the earnings exclusion variables by stock price five days before the announcement date to facilitate direct comparison with prior studies. The coefficients on EPS_PROFORMA from both regressions are about 2.20, where the theoretical coefficient for perfectly permanent earnings is 4.00 since we sum future operating earnings over the next four quarters. The main effect coefficients on RECUR_EXCL are significantly negative, indicating that the exclusions contained in on-time or unexpectedly late announcements are, on average, recurring expenses. The coefficient on the interaction term, EARLY × RECUR_EXCL, is significantly negative in both regressions, suggesting that the recurring exclusions contained in early pro forma announcements are of even lower quality. Specifically, from column 1, the total estimated predictability of recurring exclusions (–1.0101 – 0.3520 =–1.3621, F-test: p-value = 0.00) indicates that for early announcements, $1 of recurring exclusions is associated with $1.36 of expenses over the next four quarters. As a gauge, we note that for early announcements the total estimated predictability of EPS_PROFORMA is $2.07. Hence, the negative predictability of the recurring exclusions contained in early announcements account for more than 65% of the predictive value of the pro forma earnings figure itself. We find stronger evidence when we exclude pro forma quarters clustered around the expected date. That is, in column 2, we find that the negative predictability of the recurring exclusions contained in early announcements (–0.9366 – 0.6445 =–1.5811, F-test: p-value = 0.00) account for 85% of the predictive value of the pro forma earnings figure ($1.85).

In Table 6, we also observe that the coefficients on the main effects of SPECIAL_EXCL and BELOW_EXCL are positive in both regressions. Consistent with Burgstahler et al., (2002), this result suggests that nonrecurring exclusions are not fully transitory and thus, have predictive value for future earnings. We however, find no difference in the predictability of the nonrecurring exclusions contained in accelerated announcements.²⁴ Taken together, our results for H₃ provide further evidence that in some cases the acceleration of pro forma earnings news may be due at least partly to opportunistic motives.

(iv) Hypothesis 4 (H₄)

Our final hypothesis addresses the relation between announcement timing and the transparency of the pro forma earnings reconciliation. We investigate H₄ using the following regression model:

(5)

where LOW_RECON equals one if the transparency score of the firm's pro forma earnings reconciliation is less than or equal to three, and zero otherwise. This model extends equation (3) by including LOW_RECON as an additional explanatory variable. Table 7 presents the estimated results for equation (5). The estimated coefficients for LOW_RECON are significantly negative for both the full sample of pro forma quarters and the restricted sample with absolute DELAY ≥ 3 days. This result indicates that managers tend to accelerate the earnings release when the pro forma earnings reconciliation is less transparent. Combined with our previous results, this result provides further evidence that managers tend to accelerate the release of pro forma earnings information that is potentially inflated using low-quality recurring exclusions and then presented using less transparent reconciliation formats.

(i) Contemporaneous and Future Stock Returns

The results in Table 6 suggest that managers who accelerate their pro forma earnings announcements tend to exclude low-quality recurring expenses from the reported pro forma figure. Given this evidence, our next set of analyses investigates whether investors fully anticipate the negative predictive power of the recurring exclusions contained in accelerated pro forma earnings announcements. As argued in Doyle et al., (2003), if investors fully understand this relation, then the stock price reaction around the announcement date should completely reflect the negative implications of managers’ earnings exclusions. Conversely, if the contemporaneous market reaction is incomplete, then the firm's future stock returns should respond accordingly as the negative future earnings implications are materialized. Also, the contemporaneous and future market reaction to the recurring exclusions contained in accelerated pro forma earnings announcements provides evidence on investors’ ability to unravel managers’ opportunistic motives.

We employ the following regression model to estimate the association between managers’ earnings exclusions and the firm's contemporaneous and future stock returns, conditional on the timing of the pro forma earnings announcement:

(6)

where BHAR_3DAY is the compounded buy-and-hold return over the three-day window centered on the earnings announcement date less the value-weighted market return over the three-day window. BHAR_FUTURE is a vector of future market-adjusted buy-and-hold returns beginning two trading days after the announcement date and extending over the next 30, 60 or 90 trading days; i.e., over the windows, [+2, +31], [+2, +61] and [+2, +91]. If a firm delists during the future return window, we include the delisting return in our calculation of the compounded buy-and-hold return. Further, we correct for missing delisting returns as recommended by Shumway (1997) and Shumway and Warther (1999). Controls is a vector of the following controls: GAAPOP_LOSS, BOOKMKT, SIZE, STDROA and POSTSOX. We estimate equation (6) separately for the three-day window and each of the future return windows.²⁵ The independent variables in equation (6) correspond to those in our earnings predictability model (equation (5)) with the exception of FE_PROFORMA, which controls for the magnitude of the pro forma earnings surprise.

Panel A of Table 8 presents the regression results for each return window. Panel B presents the results of F-tests for the null hypotheses that the market response to the pro forma earnings surprise and the earnings exclusions contained in early announcements are equal to zero (μ₁+μ₆= 0 and μ_i+μ_j= 0). Column 1 of Panel A indicates a significantly positive coefficient on FE_PROFORMA, suggesting that investors respond positively to the pro forma earnings surprise contained in on-time and late announcements. The main effect coefficients on RECUR_EXCL, SPECIAL_EXCL and BELOW_EXCL are all insignificant. This insignificant result could suggest that the information content of the excluded items was already reflected in the stock price of on-time and late announcers prior to the report date. On the other hand, it could suggest that investors perceive the earnings exclusions of on-time and late announcers as value-irrelevant, despite the predictive power of these exclusions as documented in Table 7.

The estimated coefficient on the interaction term, EARLY × FE_PROFORMA, is positive but insignificant. Therefore, we cannot conclude that investors respond more aggressively to early pro forma earnings news. Interestingly, the coefficient on EARLY × RECUR_EXCL is significantly negative indicating an incremental negative response to the recurring exclusions contained in early announcements. The F-tests in Panel B indicate that the sum of the coefficients on RECUR_EXCL and EARLY × RECUR_EXCL is significantly negative at the 1% level (–0.0571 – 0.2385 =–0.2956, F-test: p-value = 0.00). This result provides evidence that, although investors respond favorably to positive pro forma earnings news, this response decreases with the level of recurring exclusions contained in accelerated earnings announcements. Further, investors appear to anticipate the negative predictability of the recurring exclusions contained in accelerated pro forma earnings announcements. However, it is still an open question whether investors’ response is complete.

We address this question in columns 2 to 5 of Table 8 by examining the subsequent stock returns after the earnings announcement date. With respect to on-time and late announcements, we find a significantly positive association between future stock returns and SPECIAL_EXCL in each of the three future return windows. This result suggests that investors fail to fully anticipate the positive predictability of the special items exclusions contained in on-time and late announcements. We find some evidence of a positive post-earnings announcement drift for on-time and late announcements as indicated by the positive coefficients on FE_PROFORMA; however, this association is significant in only the [+2, +91] window. For early announcements, we also find weak evidence of a positive incremental association between FE_PROFORMA and future stock returns. More importantly, we find a significant negative association between future stock returns and the recurring exclusions contained in early pro forma announcements. Specifically, the coefficient on EARLY × RECUR_EXCL is significantly negative across all the future return windows. Combined with the results in column 1, this evidence suggests that, while investors appear to discount the recurring exclusions contained in early pro forma announcements at the report date, this discount is far from complete. We find similar evidence (not tabulated) when we exclude pro forma quarters clustered around the earnings announcement date. Together, the results in Table 8 suggest that investors fail to fully anticipate the predictive power of the recurring exclusions contained in accelerated pro forma announcements. The results also suggest that the acceleration of pro forma news reduces investors’ ability to unravel managers’ opportunistic motives, thereby dampening the negative price response around the announcement date.²⁶

(ii) Cross-Sectional Analyses of Disclosure Transparency

We examine whether the quality of and the market reaction to pro forma earnings information varies with both announcement timing and disclosure transparency. We use the following measures to capture disclosure transparency: (1) reconciliation transparency and (2) information asymmetry.

Reconciliation transparency. For firms that provide less transparent reconciliations (LOW_RECON = 1), our predictability tests do not indicate a significant difference in the quality of the earnings exclusions contained in accelerated pro forma announcements. However, we find some evidence of a greater post-earnings announcement drift for early announcements with low reconciliation transparency. We also find that in each future return window the association between managers’ recurring exclusions (RECUR_EXCL) and the firm's future stock returns is significantly more negative for accelerated announcements with low reconciliation transparency. Consistent with Zhang and Zheng (2010), this result indicates that when reconciliation transparency is low, investors are even more likely to misjudge the negative predictive value of the recurring exclusions contained in early announcements. This result also suggests that low reconciliation transparency in conjunction with the deliberate timing of the pro forma earnings announcement has an important effect on investors’ perceptions of future firm performance.

Information asymmetry. Kothari et al. (2009) argue that firms have greater opportunities to intentionally time their news releases when there is high information asymmetry between managers and investors. We follow Kothari et al. (2009) and use factor analysis to construct an information asymmetry factor based on five measures: BOOKMKT, LEVERAGE, VOL, HITECH and REGULATE. BOOKMKT and LEVERAGE are as defined earlier. VOL is stock return volatility, measured as the standard deviation of monthly returns over the 12 months prior to the earnings report date. HITECH equals one for firms operating in high-tech industries, while REGULATE equals one for firms operating in regulated industries (i.e., utilities and telecommunication firms); and zero otherwise. We find that the factor loads negatively on BOOKMKT, LEVERAGE and REGULATE, and loads positively on VOL and HITECH. We classify firms with high information asymmetry (HIGH_ASYMM) as those with factor scores greater than the sample median. Our earnings predictability tests indicate that the quality of RECUR_EXCL contained in early announcements is considerably lower for high information asymmetry firms, suggesting that the acceleration of pro forma earnings announcements is more consistent with managerial opportunism for firms with greater opportunities to deliberately time their earnings releases. From our future returns tests, we find that the total coefficient on RECUR_EXCL is significantly more negative for early pro forma announcements issued by high information asymmetry firms. This result suggests that investors are less able to unravel managers’ disclosure timing motives when information asymmetry is high.

(iii) Robustness Tests

We perform several sensitivity analyses and find that all of our results are robust. First, we control for potential differences in the timing of fourth-quarter announcements due perhaps to the time-consuming nature of the audit process. Specifically, we replicate our empirical tests after including a separate indicator variable for fourth-quarter earnings announcements. Second, the timing of firms’ earnings announcements could be triggered by the announcement timing decisions of other related firms. We control for this potential follow-the-leader effect by including in our regressions the average announcement delay for all same-industry firms that reported earnings before firm i for the current fiscal quarter q. Our results and inferences are qualitatively similar, suggesting that the opportunistic timing of pro forma announcements is not fully attributable to the industry ordering of announcements. Lastly, our sample period coincides with two relevant changes in accounting standards — SFAS 142 (effective December 15, 2001) and SFAS 123R (effective June 15, 2005).²⁷ To control for these changes, we re-estimate our regressions after including separate indicator variables for those firm-quarters with non-zero stock compensation expense as reported in Compustat and firm-quarters with non-zero goodwill amortization and/or impairment charges.