Bundled Earnings Guidance and Analysts' Forecast Revisions*

Bundling managerial earnings guidance with quarterly earnings announcements (EAs) has become an increasingly common practice. This study investigates the impact of bundled guidance on analysts’ forecast revisions. Our findings indicate that analysts respond more to bundled guidance than non-bundled guidance. This effect increases with analysts’ time pressure and cognitive constraints around the EA. Analysts’ revisions also incorporate more of the bundled management guidance when accompanied by additional information, such as conference calls. We further find that analysts revise their forecasts more quickly following bundled guidance than non-bundled guidance. Together, these findings are consistent with the notion that analysts place more weight on bundled guidance than non-bundled guidance in their forecast revisions as bundled guidance facilities analysts’ timely forecast revisions following EAs. Finally, we find that analysts’ forecast revisions following bundled guidance generate significant market reactions. Our findings enhance our understanding of analysts’ information processing and shed light on why bundling can be an effective guidance strategy.


Introduction
Financial analysts are important information intermediaries in capital markets. The market makes wide use of their consensus earnings estimates to assess a firm's performance.
Given the market influence of financial analysts, managers have an incentive to issue earnings guidance that helps a firm's analysts update their earnings expectations (e.g., Ajinkya and Gift 1984;McNichols 1989). Firms have commonly issued earnings guidance at the time of an earnings announcement (EA), in a practice that has been termed "bundled guidance" (e.g., Anilowski, Feng, and Skinner 2007). Although this disclosure practice has generated considerable interest in the accounting literature (see Rogers and Van Buskirk 2013), its economic consequences for financial analysts are still not well understood. To shed light on this issue, we investigate whether bundling (i.e., issuing a firm's guidance at the time of its EA) affects the weight financial analysts place on managerial guidance in revising their own earnings forecasts. 1 We also explore the mechanisms of how bundling influences analysts' forecast revisions, and examine how investors value the analysts' forecast revisions following bundled guidance as opposed to non-bundled guidance.
We posit that analysts respond more to the news in bundled guidance than in nonbundled guidance when revising their earnings forecasts for three reasons: (i) time pressure, (ii) cognitive constraints, and (iii) disclosure credibility. First, prior studies suggest that it is important for analysts to quickly issue forecast revisions around EAs (e.g., Guttman 2010; Keskek, Tse, and Tucker 2014). Given the importance of issuing a timely forecast after an EA, analysts may face greater time pressure to process managerial information around EAs than at other times. Bundled guidance, which contains the information most directly related to future earnings, is of great use, and is released during a time when their clients demand timely the concern that bundled and non-bundled guidance are inherently different in their guidance and firm characteristics.
It is possible that an omitted correlated variable that potentially relates to both analysts' forecast revisions and managers' decisions to issue bundled guidance is driving our results. For example, managers may choose to issue bundled guidance when it is more informative because of the increased attention it will receive at an EA. To address this issue, we follow the Frank (2000) method and find that the possibility of an unobserved confounding variable to significantly affect our results is likely very small. Nevertheless, we find that our inferences do not change if we (i) remove non-bundled guidance around concurrent corporate events, (ii) control for time trend effects that coincide with the increase in popularity of bundled guidance, and (iii) control for prior disclosure policy or firm fixed effects to mitigate potential concerns that our observed effects of bundled guidance could be driven by certain omitted time-invariant firm characteristics. Overall, these results reinforce our inference that analysts place greater weight on bundled guidance than on non-bundled guidance.
To better understand our baseline analyst revision results, we also investigate each of the potential explanations (mechanisms) discussed earlier. First, we expect analysts to rely more on bundled guidance to issue timely forecasts for more salient firms (e.g., larger firms, firms with greater institutional ownership, or firms that attract greater analyst following) since analysts might have more pressure to issue a quick response for these firms. Second, analysts may rely more on bundled guidance when it is costly for them to produce information on their own due to their cognitive constraints (e.g., more EAs clustered on the same day). Third, analysts may place a greater weight on bundled guidance while producing their own forecasts when analysts receive additional concurrent information at an EA (e.g., conference calls and 10-K/10-Q filings) that might verify the information in the earnings guidance. Our empirical results are consistent with the above expectations. Note that the three mechanisms mentioned

Accepted Article
This article is protected by copyright. All rights reserved.
above are not mutually exclusive. Collectively, they help explain why analysts place a greater weight on bundled guidance than on non-bundled guidance. However, we also note that the economic significance of each mechanism is small.
The limited economic magnitude observed for each mechanism could be due to the difficulty in measuring the underlying constructs (e.g., difficulty in finding exogenous variation in those constructs, or potentially noisy proxies), or the difficulty in fully disentangling the effect of managerial guidance from that of the EA. To mitigate this concern, we examine the bundling effect before versus after the implementation of Regulation Fair Disclosure (Reg FD).
Analysts likely face greater time pressure and cognitive constraints during EAs after Reg FD because Reg FD prohibits selective disclosure and analysts might rely more on EAs to obtain and process public information. In addition, analysts might find bundled guidance more credible after Reg FD with more concurrent disclosures (e.g., conference call, 10K/10Q) available to help verify the guidance news. Our results show that the bundling effect significantly increases after Reg FD, which is consistent with the idea that the bundling effect is stronger when its underlying mechanisms are more likely to be present. In addition, the economic significance of Reg FD is much larger than that documented using proxies of each mechanism. 3 Furthermore, we examine the effects of bundling on the speed of individual analyst's forecast revision. Our results show that analysts revise their forecasts more quickly after bundled guidance than non-bundled guidance, which is consistent with the notion that bundling helps analysts issue forecast revisions quickly when they are facing greater time pressure, 3 It is worth noting that the setting of Reg FD also has its own limitations, such as the disadvantages of not being able to isolate one mechanism from another, the low number of bundled observations before Reg FD (i.e., October 2000), and the potential confounding events such as the Global Analyst Research Settlement (Global Settlement). Global Settlement was an enforcement agreement reached in April 2003 between the Securities and Exchange Commission (SEC) and the 10 largest US investment firms to address conflicts of interest between investment banking and analyst research within these firms.

Accepted Article
This article is protected by copyright. All rights reserved. cognitive constraints, and when they find managerial guidance more credible. Taken together, these additional results enhance our understanding of how bundling facilitates analysts' forecast revisions.
Finally, we address whether investors value analysts' forecast revisions following bundled guidance. Investors may value these timely forecast revisions because they help resolve information uncertainty shortly after an EA. However, investors may discount these forecast revisions if they perceive them as more likely to be piggybacking on managerial guidance. We find that the intraday market return is more sensitive to each unit of analysts' forecast revisions immediately following bundled guidance than to non-bundled guidance. Our results thus suggest that investors do not simply dismiss analysts' revisions as piggybacking.
That is, the market benefits of issuing timely revisions following managerial guidance for analysts appear to outweigh the potential costs of possible piggybacking, which is consistent with prior studies that indicate the importance of analysts issuing timely forecasts immediately after EAs (e.g., Guttman 2010; Keskek et al. 2014).
Our study contributes to the literature in several ways. First, we extend the literature on managerial earnings guidance and analysts' forecast revisions. Prior studies examine analysts' revisions as a function of various guidance attributes (e.g., accuracy, precision, horizon, and bias). 4 We add to this literature by investigating the impact of issuing guidance at the time of an EA versus at a different time on analysts' forecast revisions. Our findings indicate that bundling is associated with greater analysts' responses to guidance news and quicker analysts' forecast revisions. We further explore the mechanisms (time pressure, cognitive constraints, and disclosure credibility) that influence the effect of bundling on analysts' revisions. The results of the mechanism analysis enhance our understanding of the role of bundled guidance in facilitating analysts' timely forecast revisions and shed light on factors that influence analysts' information processing around EAs (e.g., Blankespoor, deHaan, and Marinovic 2020;Driskill, Kirk, and Tucker 2020).
Second, we add to the literature that examines the usefulness of analyst output subsequent to managerial disclosure (e.g., Altınkılıç, Balashov, and Hansen 2013;Li et al. 2015). Altınkılıç et al. (2013) show that investors, on average, do not react to analysts' forecast revisions following corporate disclosure. Li et al. (2015), on the other hand, find that analysts' recommendation revisions following corporate disclosure continue to generate significant market reaction. 5 Our study extends these prior studies by showing that investors value analysts' timely forecast revisions following bundled guidance.
Importantly, although we document the benefits of bundling to financial analysts, it should be noted that because bundling is determined by various factors (e.g., Skinner 1994;Cheng and Lo 2006;Kothari, Shu, and Wysocki 2009;Billings and Cedergren 2015), not all firms may find it optimal to issue bundled guidance. The main purpose of this study is to document the incremental positive effects of bundling for firms that adopt this disclosure practice. We further note that although we adopt various methods to empirically control for firm characteristics and managerial incentives related to the bundling decision, we cannot fully rule out the possibility that there remains a correlated omitted variable.

Empirical design
We begin our analysis by investigating the impact of bundled guidance on analysts' forecast revisions. We define our key variable of interest, Bundled, as being equal to one if the managerial earnings guidance is in the form of bundled guidance and zero otherwise. We define bundled guidance as the earnings guidance issued on the EA date. Conversely, we define nonbundled guidance as managerial earnings guidance that is not issued on the EA date. 6 Note that we exclude the managerial earnings guidance issued after the corresponding fiscal quarter-end but before the EA (so-called "pre-announcements").
To test the effect of bundling on analysts' forecast revisions following managerial guidance, we estimate the following regression: where AFRev represents analysts' forecast revisions following managerial guidance, both of which are for the same future period of earnings. MFNews represents managerial guidance news. We define AFRev as the difference between the consensus (i.e., mean) analyst forecast issued within five days after the managerial guidance date and the last consensus analyst forecast issued before the managerial guidance date, scaled by the firm's stock price one trading day before the guidance date (Kross, Ro, and Suk 2011;Rogers and Van Buskirk 2013). 7 We use a period of five days to measure AFRev and other analyst revision variables because previous studies (e.g., Zhang 2008) find that most analysts revise their forecasts within five days. We define MFNews as the difference between managerial guidance and the most recent consensus analyst forecast prior to the guidance date, scaled by the firm's stock price one trading day before the guidance date. Following Rogers and Van Buskirk (2013), we adjust our bundled guidance news (MFNews) to mitigate potential measurement error that may be positively correlated with earnings surprise (note that we do not make a similar adjustment for 6 Our main results are similar if we use the measurement window of one day or two days around the EA date to define bundled guidance (we provide the details in section 6). 7 We manually calculate the consensus of revised forecasts based on the analysts' forecasts available. In our main analysis, we set AFRev to zero if no analysts issue a revision for a specific firm within five days following the guidance. As a robustness check (reported in section 6), we drop observations where no analysts revise their forecasts within five days after the guidance date and find similar estimation results.

Accepted Article
This article is protected by copyright. All rights reserved.
non-bundled guidance news). Essentially, this adjustment helps us obtain a refined consensus analyst forecast that represents analysts' expectations of future earnings immediately after the EA in the absence of bundled guidance. Therefore, by construction, this adjustment filters out the information contained in the actual earnings surprise as well as other supplementary information that is associated with the actual earnings surprise (but not bundled guidance), and hence enables us to obtain a more accurate measure of the guidance news contained in the bundled guidance.
Similarly, since our hypothesis focuses on the analyst reaction to guidance, rather than the reaction to the EA, we also adjust AFRev for any measurement error associated with the actual earnings news on the guidance date. Empirically, we follow Rogers and Van Buskirk (2013) and calculate AFRev as the difference between the unadjusted and fitted value of AFRev, estimated from an overall sample of EAs with no bundled guidance. 8 By doing so, we are able to mitigate the concern that the effect of bundled guidance using unadjusted analyst activities is mechanically driven by analysts' reactions to EAs rather than managerial guidance (Rogers and Van Buskirk 2013). In a similar vein, one of the benefits of testing AFRev is that it allows us to more precisely measure analysts' reactions to each unit of managerial guidance news (instead of measuring analysts' responses to non-guidance financial information contained in the EA). 9 In equation (1), if bundled guidance is associated with a stronger analyst response than non-bundled guidance, we expect the coefficient a2 to be positive.
To help isolate the bundling effect from the potential effects of other guidance characteristics, we also control for the importance of guidance news (AbsMFNews) (Lipe, Bryant, and Widener 1998;Rogers and Stocken 2005), the extent of guidance bias ( In all of our analyses, we include both year-quarter (γ) and industry (w) fixed effects based on the 48 industry classifications of Fama and French (1997) to account for any intertemporal or cross-industry differences in managerial guidance and firm characteristics beyond the controls we discussed above. 10 Finally, we correct our standard errors for heteroskedasticity as well as the clustering of observations by firm and year-quarter. 10 Our results are robust to further controlling for the interaction terms between industry fixed effects and guidance news and the interaction terms between year-quarter fixed effects and guidance news (p-value < 0.01, untabulated). Although this approach essentially controls for potential differences in analysts' responses to guidance news across time and industry, it significantly reduces our sample size to 5,183.

Accepted Article
This article is protected by copyright. All rights reserved.

Sample selection and summary statistics
To obtain our sample, we use managerial quarterly EPS estimates reported in the that the guidance value would be at the low (high) end of the range, we use the low (high) end of the range as the guidance value (Hutton, Lee, and Shu 2012). We next match our guidance data with the corresponding earnings and analysts' forecasts reported in I/B/E/S. We also remove firm-quarters with missing actual earnings or actual earnings releases that are issued later than 90 days after the corresponding quarter-end. In addition, we eliminate any guidance observations issued after the corresponding fiscal quarter-end (so-called "pre-announcements").
Finally, we match our guidance data to the other data used in our analyses. We obtain our financial accounting data from Compustat and stock returns data from CRSP.

Accepted Article
This article is protected by copyright. All rights reserved.
The lower percentage in 1998-2000 could be attributed to the fact that firms preferred issuing private guidance before Reg FD (Wang 2007).
Next, we calculate the summary statistics for the variables used in our main analyses and provide the univariate comparative statistics for various guidance characteristics across our bundled and non-bundled subsamples in panel B of Table 2. These statistics show that managers who bundle their guidance were, on average, more accurate in the past (HMFAccu), but are less precise (MFWidth) than their non-bundling counterparts. 11 We also find that the bundled guidance in our sample has a greater percentage of loss forecasts (MFLoss) and has a longer horizon (MFHorizon). These findings suggest that it is necessary to control for these guidance characteristics in our regression analysis in order to estimate the incremental effect of bundling. Panel C of Table 2 reports the summary statistics of conditioning variables used for the cross-sectional analyses in section 3. We discuss these variables in more detail in section 3. Table 3 reports the results from equation (1). Specifically, in column (1), we only control for industry and year-quarter fixed effects. In column (2), we include guidance characteristics in the model. We include firm characteristics in column (3). We find that the coefficient on MFNews×Bundled is significant at the less-than-one-percent level and the magnitude remains stable across the first three columns. Economically, for example, column

Main results
(3) shows that the effect of bundled guidance is approximately twice as large as that of non- 11 The term HMFAccu represents the historical guidance accuracy, defined as −1 multiplied by the average absolute management forecast error (scaled by stock price of one trading day before the guidance date) over the last two years (prior to the current quarter). Untabulated results further show that, ex post, bundled guidance is more accurate than non-bundled guidance (p-value < 0.01).

Accepted Article
This article is protected by copyright. All rights reserved.
bundled guidance. 12 Note that we use the column (3) specification as the baseline model in all of our subsequent analyses of analysts' reactions to managerial guidance news.
These results are robust when we include concurrent actual earnings news in column (4) to mitigate concerns that actual earnings surprise may not be fully adjusted. Together, these results support our prediction that bundling has a significant influence on analysts' reactions to managerial guidance. In other words, on average, analysts create more forecast revisions in response to each unit of bundled guidance news compared to their revision activity prompted by non-bundled guidance news.

Entropy balancing results
Next, we adopt an entropy balancing approach to address concerns that bundled and non-bundled guidance are inherently different in their guidance and firm characteristics. This approach allows us to use more homogenous subsamples to test the treatment (i.e., bundling) effect. Entropy balancing is a reweighting procedure that assigns a scalar weight to each sample unit such that the moments of the control variables are equal between the reweighted control group and the treatment group, creating a balanced sample for the subsequent estimation of the treatment effect (Hainmueller 2012). The entropy balancing approach reduces the effect of potential misspecification (e.g., omitted variables) in the estimation of the treatment effect (Ho et al. 2007; Abadie and Imbens 2011). We first use the entropy balancing method to balance the mean, variance, and skewness of our baseline control variables (i.e., firm and guidance characteristics). 13 The results reported in panels A and B of Table 4 show that the differences in mean, variance, and skewness values of control variables between bundled and non-bundled 12 Note that MFNews×Bundled only captures the incremental effect of bundling on MFNews controlling for other interaction terms with MFNews. We calculate the incremental effect of bundling using the ratio of the coefficient on MFNews×Bundled and the coefficient on MFNews (0.577/0.623 + 1 = 1.926). 13 In an additional analysis, we also include implied volatility from Billing Jennings, and Lev (2015) as an additional covariate in the entropy matching. Although our sample is reduced by 28% due to the smaller data coverage of implied volatility, our results based on the reduced sample are unaffected (p-values < 0.01, untabulated).

Accepted Article
This article is protected by copyright. All rights reserved.
guidance samples become negligible after the entropy balancing procedure is implemented.
This suggests that the level of homogeneity between bundled and non-bundled guidance samples is high. Next, we re-estimate equation (1). Panel C of Table 4 shows that MFNews×Bundled remains significant across different specifications.

Mechanism analyses
In this section, we explore three explanations for why analysts respond more to bundled guidance than to non-bundled guidance. We have previously labeled these explanations as "time pressure," "cognitive constraints," and "disclosure credibility."

Time pressure
Prior studies suggest that it is important for analysts to issue quick forecast revisions around

Accepted Article
This article is protected by copyright. All rights reserved.
Empirically, we replace Bundled in equation (1) with two indicator variables,

Bundled_HighPressure and
Bundled_LowPressure. Bundled_HighPressure (Bundled_LowPressure) is equal to one for the bundled guidance if the conditioning variable (i.e., firm size, institutional ownership, and analyst coverage) is higher than (lower than or equal to) the sample median of bundled guidance by each industry-year-quarter group, and zero otherwise. Panel A of Table 5 reports the results. We find statistically significant evidence consistent with our "time pressure" explanation. That is, the effect of bundling is larger (smaller) for firms with high (low) time pressure for analysts to issue timely forecasts. However, our results also suggest that the economic magnitude of the "time pressure" mechanism is small.

Cognitive constraints
In line with our argument that analysts rely more heavily on bundled guidance when they face greater cognitive constraints during EAs (i.e., the "cognitive constraint" explanation), we predict that the bundling effect increases with analysts' cognitive constraints. We use three proxies for cognitive constraints. We first consider cognitive constraints with respect to analyst ability. Following Clement (1999), we construct our first proxy for cognitive constraints using analysts' forecasting experience, i.e., the average number of quarters in which analysts issue forecasts for the firms they follow. As Clement (1999) points out, only capable analysts can survive competition with peers as reflected in longer tenure, and analysts' skills and knowledge about a firm improve over time as they follow the firm. We, therefore, expect the effect of bundling to be greater for bundling firms followed by less experienced analysts than for other bundling firms.
Next, we consider cognitive constraints with respect to analyst busyness. Our second

Accepted Article
This article is protected by copyright. All rights reserved.
likely to face constraints on time, energy, and resources. We thus expect the effect of bundling to be greater for bundling firms followed by analysts with larger portfolios than for other bundling firms. Our third measure of cognitive constraints incorporates the notion of analyst busyness due to the clustering of EAs. Driskill et al. (2020) show that analysts' constraints due to EA clustering limit their ability to process information from EAs and slow down their response. For example, Driskill et al. show that analysts with three or more concurrent announcements delay their forecasts by three times as much as those with one concurrent announcement. In our setting, when multiple firms from a given analyst's portfolio announce their respective earnings on the same day, the analyst might be busy processing multiple firms' information within a short period of time, and may thus find it difficult to produce her own information beyond public guidance. Thus, we expect the effect of bundling on analysts' revisions to be greater for bundling firms followed by a higher percentage of busy analysts (over the total number of analysts following the firm). We follow Driskill et al. and define busy analysts as those who have three or more concurrent EAs issued by firms in their portfolios on the same day. We define analyst experience, analyst portfolio size, and analyst busyness in more detail in panel C of Table 2.
To test our "cognitive constraints" explanation, we replace Bundled in equation (1) with two indicator variables, Bundled_HighConstraint and Bundled_LowConstraint. When we use analyst experience as the constraint proxy, Bundled_HighConstraint (Bundled_LowConstraint) is equal to one for the bundled guidance if analyst experience is lower than or equal to (higher than) the sample median of bundled guidance, and zero otherwise. When we use portfolio size as the constraint proxy, Bundled_HighConstraint (Bundled_LowConstraint) is equal to one for the bundled guidance if the size of the analyst coverage portfolio is higher than (lower than or equal to) the sample median of bundled guidance, and zero otherwise. Finally, when we use the percentage of busy analysts as the constraint proxy, Bundled_HighConstraint

Accepted Article
This article is protected by copyright. All rights reserved.
(Bundled_LowConstraint) is equal to one for the bundled guidance if the proportion of busy analysts of a firm described above is higher than (lower than or equal to) the sample median of bundled guidance, and zero otherwise. Panel B of Table 5 reports the results. Consistent with the notion that lower ability and busier analysts are more likely to rely on managerial guidance rather than on their own analysis, we find statistically significant evidence that the effect of bundling increases with analysts' constraints. However, we note that the economic significance of this mechanism is small.

Disclosure credibility
Next, we examine whether additional information concurrent with bundled guidance could increase disclosure credibility and explain why analysts react more to bundled guidance than to non-bundled guidance ("disclosure credibility" explanation). Although we adjust the bundling error to remove the potential impact of the actual earnings news on the EA date, other concurrent information may still exist at the EA. Additional concurrent disclosure may increase the credibility of bundled guidance, leading analysts to place greater weight on the guidance in revising their forecasts. We first consider conference calls as a source of additional disclosure.
Chapman and Green (2018) show that bundled guidance is often disclosed in the accompanying conference call. Analysts ask managers about and receive forward-looking information during conference calls. Conference calls therefore allow analysts to obtain additional information to verify bundled guidance. We next consider 10-K or 10-Q filings as another source of concurrent disclosures (Arif et al. 2019). Similarly, analysts may find the bundled guidance more credible because of the additional disclosure they can obtain from a 10-K or 10-Q filing on the same day. Furthermore, analysts may obtain more additional information if a firm issues a longer EA than other firms (e.g., Bird and Karolyi 2016). Panel C of Table 2 provides the summary statistics of concurrent conference calls, concurrent 10-K/Q filings, and the length of concurrent 8-K filings.

Accepted Article
This article is protected by copyright. All rights reserved.
To examine the "disclosure credibility" explanation, we replace Bundled in equation (1) with two indicator variables, Bundled_MoreDisc and Bundled_LessDisc, capturing whether more concurrent information is along with the bundled guidance. More specifically, Bundled_MoreDisc (Bundled_LessDisc) is equal to one for the bundled guidance with (without) a concurrent conference call, with (without) a concurrent 10-K/10-Q filing, or with a longer (shorter) concurrent 8-K filing, and zero otherwise. We report the results of our comparative statics in Table 6. 14 We find that the effect of "disclosure credibility" is statistically significant.
We, however, note that the economic significance of this mechanism is also small.

Reg FD
We note that the three explanations (mechanisms) discussed above are not mutually exclusive.
Collectively, they help explain why analysts place a greater weight on bundled guidance than on non-bundled guidance. We also note that the economic significance of each mechanism is small. The limited economic magnitude of each individual mechanism could be due to the difficulty in measuring the underlying constructs (e.g., it is difficult to find exogenous variation in these constructs and their empirical proxies can be noisy), or the difficulty in fully disentangling the effect of managerial guidance from that of the EA. To mitigate this concern, we examine whether Reg FD affects how analysts react to managerial bundled guidance. After Reg FD, the above mechanisms underlying the bundling effect are likely to be more present.

Accepted Article
This article is protected by copyright. All rights reserved.
credible because EAs are accompanied by more concurrent disclosures (e.g., conference calls) after Reg FD (e.g., Rogers and Van Buskirk 2013). The above reasoning leads us to posit that analysts' responses to the news in bundled guidance will significantly increase after Reg FD if the underlying mechanisms of "time pressure," "cognitive constraints," and "disclosure credibility" explain the bundling effect.
To examine whether the bundling effect increases after Reg FD, we conduct two tests.
In the first test, we replace Bundled in equation (1)  to obtain the same length of window before and after Reg FD or (ii) implementing an entropy balancing procedure to reweight the bundled guidance issued after Reg FD as opposed to the remaining guidance. In contrast to the findings in Tables 5 and 6 that show the small economic significance of each individual mechanism (e.g., the differences between the coefficients on MFNews×Bundled_HighPressure and MFNews×Bundled_LowPressure in Table 5 are small), the results in panel A of Table 7 show that the coefficients on MFNews×Bundled_PostRegFD are much larger than the coefficients on MFNews×Bundled_PreRegFD across the three columns. Taken together, these results indicate that the magnitude of the bundling effect related to Reg FD is greater than the ones related to the three mechanisms when tested individually.
This is consistent with the notion that the joint effect of the three mechanisms is likely more present after Reg FD.

Accepted Article
This article is protected by copyright. All rights reserved.
To mitigate the concern that omitted variables could correlate with both firm decisions to switch to issuing bundled guidance after Reg FD and firm fundamentals (e.g., Wang 2007), we conduct a second test, in which we adopt a difference-in-differences (DID) research design with an entropy balanced sample centered on the implementation date of Reg FD (October 23, 2000). We focus on the analysis covering January 1998-August 2003, which comprises the same number of months before and after the implementation of Reg FD (October 23, 2000) with available guidance data. We define the bundled (non-bundled) firms as those that consistently issued bundled (non-bundled) guidance and did not issue any non-bundled (bundled) guidance in the above-mentioned sample period. Next, we follow a similar entropy balancing approach as in the previous section to ensure that our non-bundled firms are comparable to our bundled firms, although their decisions on the timing of guidance differ. 15 Appendix 2 illustrates the effectiveness of the entropy balancing process. It shows that the differences between the two groups are significant for all variables before reweighting, but no difference is significant afterward, suggesting that our balancing process is effective. In this design, bundled (non-bundled) firms' decisions to issue bundled (non-bundled) guidance do not change around Reg FD, therefore, it is less likely that these firms have significant changes in their disclosure practices around Reg FD compared to other firms. As such, our DID design helps mitigate the omitted variable issues regarding both a firm's decision to switch to issuing bundled guidance after Reg FD, and firm fundamentals, especially after entropy balancing. We expect that the impact of Reg FD on analysts' reactions to guidance news is greater for bundled firms than for non-bundled firms because the regulation likely increases analysts' time pressure, cognitive constraints, and disclosure credibility around EAs, which coincide with the guidance issued by bundled firms but not the guidance issued by non-bundled firms.
We then estimate the following DID model to examine whether bundled firms generate larger analysts' forecast revisions in response to guidance news after Reg FD than before, as compared with non-bundled firms: where PostRegFD is equal to one for observations after the implementation of Reg FD, and zero otherwise. BundledFirms is an indicator variable equal to one for bundled firms, and zero otherwise. Control variables (CONTROLS) are the same as in column (3) of Table 3. We include both year-quarter (γ) and firm (f) fixed effects to account for any inter-temporal or cross-firm differences in managerial guidance and firm characteristics beyond the control variables. We present the results of estimating equation (2) in panel B of Table 7. We first find that the coefficient on BundledFirms×MFNews is insignificant, suggesting that analysts react to the guidance of bundled firms and non-bundled firms similarly prior to Reg FD. More importantly, we find that the coefficient on BundledFirms×PostRegFD×MFNews is significantly positive, which is consistent with analysts reacting more strongly to bundled guidance after Reg FD. We also conduct, in untabulated analyses, a trend analysis by creating an indicator variable, PostRegFD -1 , equal to one for guidance issued within the six-month window immediately before Reg FD (i.e., from May 1, 2000 to October 22, 2000), and zero otherwise. We find that the coefficient continues to be significantly positive on BundledFirms×PostRegFD×MFNews, but not on BundledFirms×PostRegFD -1 ×MFNews, which is consistent with the assertion that Reg FD has an impact on bundled firms only after Reg FD, but not before. These results corroborate the parallel trend assumption of our DID design and mitigate concerns that an endogenous relation is responsible for the results.

Accepted Article
This article is protected by copyright. All rights reserved.
In sum, the evidence from our mechanism analyses lends support to the notion that bundled guidance facilitates analysts' forecast revisions more when analysts have greater time pressure and cognitive constraints to produce information on their own at an EA, and when they receive additional information that increases the credibility of bundled guidance. Our results in this section should be interpreted with caution because it is inherently difficult to fully address the underlying causes of our results, although we have provided a series of tests to investigate the underlying mechanisms.

Speed of analysts' forecast revisions
Up to this point, our empirical results are consistent with the idea that analysts place more weight on bundled guidance than on non-bundled guidance when issuing their own forecasts. Our primary analysis in section 2 examines the sensitivity of analysts' responses to news in managerial earnings guidance. An alternative way to test whether analysts respond more to bundled guidance than to non-bundled guidance is to examine whether analysts revise their forecasts sooner after bundled guidance than after non-bundled guidance.
To explore this, we create AFSpeed, which accounts for the speed with which analysts issue their forecasts subsequent to guidance. To calculate AFSpeed, we first determine the average number of days between managerial guidance and the subsequent individual analysts' forecast revisions within the five-day post-guidance period, scaled by five. We code the number of days to be five for cases where no analysts issue forecasts within five days after the guidance but as discussed in footnote 16, our results are robust to removing these cases. Next, for analysts' revisions following a bundled guidance, we adjust the value of revision speed for the bundling error similar to the adjustment for AFRev following Rogers and Van Buskirk (2013). As mentioned earlier, we make this bundling adjustment to mitigate the concern that the effect of bundled guidance using unadjusted values could be driven by analysts' reactions to EAs rather than managerial guidance. More specifically, we use the Rogers and Van Burskirk (2013)

Accepted Article
This article is protected by copyright. All rights reserved.
approach to obtain a fitted value for the speed of analysts' revisions following EAs with no guidance. We then subtract the fitted value from the raw speed of analysts' revisions following bundled guidance to obtain the adjusted speed of revisions. We further winsorize this adjusted value of speed to be between zero and one to ensure that each value of speed in our analysis has practical meaning such that our results are not sensitive to this winsorization. Here, the value of zero (one) represents the fastest (slowest) speed, i.e., it takes zero (five) days, on average, scaled by five, for analysts to issue their forecasts following guidance within our fiveday measurement window. Finally, we multiply the speed of analysts' revisions by -1 to indicate the higher value of AFSpeed, the quicker their responses.
To estimate the effect of bundling on analyst revision speed, we estimate the following OLS regression: where the dependent variable (AFSpeed) has been defined above. The indicator variable (Bundled) and control variables (CONTROLS) are the same as in column (3) of Table 3. If analysts revise their forecasts more quickly after bundled guidance than after non-bundled guidance, we expect c1 to be positively significant.
Column (1) of Table 8 reports the results from the OLS estimation of equation (3). Our results show that analysts revise their forecasts more rapidly after bundled guidance than after non-bundled guidance (with a coefficient of 0.496 and p-value < 0.01). In terms of the economic magnitude, it suggests that analysts are approximately 2.5 days (i.e., 0.496 × 5 days) faster in issuing their forecasts after bundled guidance than after non-bundled guidance. 16 Since AFSpeed is censored between negative one and zero, the OLS estimators in equation (3) could be biased and inconsistent. To address this concern, we also use a Tobit regression to estimate the effect of bundling on analyst revision speed. We find similar results from the estimation of this Tobit regression in column (2) of Table 8.
In summary, our results regarding the speed of analysts' forecast revisions are consistent with our framework discussed in the previous section. Because bundled guidance is issued at a time when analysts likely have greater time pressure and cognitive constraints and when firms often provide additional disclosures, analysts not only place a greater weight on the guidance news in revising their own forecasts, but also issue forecasts more quickly following the guidance. 17

Bundled guidance and the informativeness of analysts' revisions
In this section, we investigate whether bundled guidance affects the usefulness of analysts' revisions to the market. Our results in section 2 suggest that analysts respond more to managerial earnings guidance issued at the time of an EA (i.e., bundled guidance) than to non- forecasts following bundled guidance than to forecasts following non-bundled guidance. On 17 According to our framework, analysts likely have less time pressure and fewer cognitive constraints outside the EA window, and they may find non-bundled guidance less credible. As a result, analysts may spend more time and effort gathering and producing information on their own and put less weight on managerial guidance in the non-bundling case. Thus, our framework does not predict whether analysts' forecasts are more or less accurate after a bundled guidance than after a non-bundled guidance. We find no evidence that analyst forecast accuracy is higher for the forecasts issued after bundled guidance (not tabulated). We also explore whether "time pressure" and "cognitive constraints" would lead analysts to "over-rely" on bundled guidance. To address this issue, we follow the research design in Feng and McVay (2010) and test whether the estimated analyst forecast revision in response to bundled guidance is associated with absolute analyst forecast errors. Our untabulated results suggest that relying on bundled guidance does not lower the accuracy of analysts' forecasts. We interpret this result as suggesting that analysts do not over-weight bundled guidance.
Accepted Article the other hand, bundled guidance might subsume the usefulness of subsequent revised forecasts if analysts tend to piggyback on managerial guidance (e.g., Altınkılıç et al. 2013;Kim and Song 2015). In this section, we examine whether investors react more or less strongly to individual analysts' forecasts following bundled guidance than non-bundled guidance. 18 More specifically, we investigate whether the market incorporates the news contained in analysts' forecast revisions into stock prices. If the market finds analysts' revised forecasts to be useful, then we expect a significant market reaction to analysts' forecast revisions (Ivković and Jegadeesh 2004). We follow Ivković and Jegadeesh (2004) and examine the stock price response to individual analysts' forecast revisions (per unit of revision news) following bundled guidance versus their revisions following the issuance of non-bundled guidance. More specifically, we regress the intraday market reaction around the analyst forecast announcement (MktReaction) on the AFRev, the interaction between AFRev and Bundled, and a set of control variables related to firm and analyst forecast characteristics as follows: We follow Altınkılıç et al. (2013) and measure the intraday market reaction (MktReaction) over a window of four 10-minute intervals around the analyst forecast announcement within five days after the managerial guidance, where non-trading nighttime (or weekend or holiday) periods are folded into one interval. This relatively narrow 40-minute window helps isolate the market reaction to individual analysts' forecast revisions from reactions to the preceding guidance news. We use a similar specification as that in equation (1)

Accepted Article
This article is protected by copyright. All rights reserved.
also follow Altınkılıç et al. (2013) and measure all analyst forecast related variables of each individual analyst. 19 Table 9 reports the results of this analysis. Column (1) shows that the interaction term, AFRev×Bundled, is positive and significant at the one-percent level. This result does not change if we further control for actual earnings surprise (column (2)). Thus, our results here suggest that market reaction sensitivity to analysts' forecast revisions is greater for revisions issued after bundled guidance than for those issued after non-bundled guidance. Economically, for example, column (1) shows that bundled guidance is associated with an increase of approximately 19% in market reaction to analysts' forecast revisions. 20 Our evidence in this section suggests that investors do not simply dismiss analysts' revisions after bundled guidance as piggybacking, but view analysts as still playing a role as an information intermediary in the presence of bundled guidance. This evidence is consistent with prior studies that indicate the importance of analysts in issuing timely forecasts immediately after EA (e.g., Guttman 2010; Keskek et al. 2014).

Unobserved correlated variables
It is possible that the effects of bundling that we observe are due to some unobservable firm or CEO characteristics, confounding year trend effects, or incentives omitted from our baseline regression models. For example, the decision to issue bundled versus non-bundled 19 We also consider the potential late time stamps of I/B/E/S data, but we do not think it is a major issue in our setting. First, late time stamps, if any, would lead us to misclassify some early guidance as late guidance, which in turn would weaken the association between delayed market reactions and guidance news. Second, such errors, if they exist, would affect both forecasts issued after bundled guidance and those issued after non-bundled guidance. This means that the potential errors would not systematically bias towards finding results for a particular group of analysts' forecasts. Third, Bradley et al. (2014) andHoechle, Schaub, andSchmid (2015) document that time stamp errors are relatively rare after 2002. As an additional robustness check, we re-run equation (3) using post-2002 data and find similar results (the p-value associated with AFRev×Bundled is less than 0.01, untabulated). 20 The incremental effect of bundling is equal to the ratio of the coefficient on AFRev×Bundled and the coefficient on AFRev (0.712/3.793 = 0.188).

Accepted Article
This article is protected by copyright. All rights reserved.
guidance has been shown to be affected by a number of factors, including the disclosure venue, a firm's pre-committed disclosure policy, the type and magnitude of news provided within the guidance, firm characteristics, and the firm's information environment (e.g., Rogers and Van Buskirk 2013). Although we control for a number of guidance properties and conduct an entropy balancing analysis, in this section we perform additional checks to address these concerns.
A traditional way to address unobserved correlated variables is to implement a Heckman (1979) two-stage procedure. We note that the most critical step using the Heckman (1979) two-stage procedure is to find an instrumental variable (IV) that is correlated with the independent variable of interest but is exogenous relative to the dependent variable. However, finding such an exogenous variable in the setting of disclosure policy is generally difficult (e.g.,

Lennox, Francis, and Wang 2012). Previous studies suggest that the Heckman (1979) approach
should not be used to bolster confidence in the OLS estimate if the IV is weak or semiendogenous (Larcker and Rusticus 2010). Therefore, we follow Larcker and Rusticus (2010) and assess the sensitivity of baseline results to unobserved correlated variables using the method developed by Frank (2000). This alternative approach is used to assess the likelihood of an unobserved confounding variable to significantly affect the results estimated from our baseline model. Our results from this analysis, reported in Appendix 3, suggest that the possibility of an unobserved confounding variable to significantly affect our results is very small.

Additional robustness checks
Next, we conduct a set of additional robustness checks based on alternative samples, estimation methods, and model specifications to further address the omitted variable issues, such as sticky firm disclosure policy and time trend effects.

Accepted Article
This article is protected by copyright. All rights reserved.
Prior studies show that non-bundled guidance can be "contaminated" by other concurrent events (e.g., conference calls, executive changes, product announcements) (Billings et al. 2015).
Indeed, based on multiple data sources (e.g., Capital IQ's Key Developments, Thomson Reuters' Street Event, and Ravepack's Press Releases), we verify that 66% of the non-bundled guidance in our sample has a contaminating event. To address this concern, we re-estimate our baseline model by removing non-bundled guidance with contaminating events and report the robust results in row (1) of Table 10.

No analyst forecast revisions
In our main analysis, we set AFRev to zero if the consensus analyst forecast after the managerial guidance is missing (i.e., there is no analyst revision). This implicitly assumes that the analyst issued a revision of zero when they did not change their expectations. As a robustness check, we drop any guidance with no analysts issuing their forecasts within five days after the guidance date. We find similar estimation results on MFNews×Bundled (row 2 of Table 10).

Managerial guidance issuance policy
We also consider that the decision to issue bundled versus non-bundled guidance could be a firm's pre-committed disclosure policy. For example, Billings et al. (2015) suggest that managers with a history of providing guidance are more likely to commit to providing bundled guidance to mitigate investor uncertainty. Our effect of bundling could be driven by a firm's committed disclosure policy. To address this alternative explanation, we create two indicator variables on whether a firm issued (i) managerial guidance in the past (MFIssued) and (ii) bundled guidance in the last quarter (LagBundled). Our results are robust to further controlling for the interactions of guidance news and these two indicator variables (row 3 of Table 10).

Time trend
As shown in Table 2, the frequency of bundling increases over time. To mitigate the concern that the bundling effect is due to a time trend, we examine whether our AFRev results are robust

Accepted Article
This article is protected by copyright. All rights reserved.
to controlling for a year-quarter trend variable (Year-Quarter) and its interaction with guidance news. This approach essentially controls for potential differences in the quality of guidance news and analysts' responses to guidance news over time. It also controls for other time-variant potentially confounding effects. The results in row (4) of Table 10 show that the effect of bundling remains unaffected.

Time-invariant effects
To control for potential omitted variables related to fixed firm characteristics, we re-estimate equation (1) while including the interaction terms between firm fixed effects and MFNews.
Note that the presence of firm fixed effects also allows us to interpret the coefficients as indicating the effects of changes in variables of interest (e.g., Balakrishnan, Core, and Verdi 2014). Thus, if we find a significantly positive coefficient on MFNews×Bundled, we can interpret it as a change in guidance bundling associated with a change in analysts' revision activities. The results in row (5) of Table 10 show that our main findings continue to hold.

Other robustness checks
Prior studies suggest that both analysts and managers have incentives to include a certain level As shown in row (6) of Table 10, our main findings are not affected.

Accepted Article
This article is protected by copyright. All rights reserved.
In our main analysis, we define bundled guidance as guidance issued on the EA date.
Some prior studies define bundled guidance as being issued within three days (e.g., Rogers and Van Buskirk 2013) or five days (e.g., Billings et al. 2015) centered on the EA. The percentage of bundled guidance slightly increases from 75.03% to 77.21% (77.38%) of our sample if we use the measurement window of three days (five days) centered on the EA. This confirms that the majority of bundled guidance is issued on the EA date. Our results in rows (7) and (8) of Table 10 show that our main findings are robust to using these alternative windows to define bundled guidance. on AFRev hold for both good and bad news guidance (relative to analyst consensus) or optimistic and pessimistic guidance (relative to actual earnings). These results suggest that it is unlikely that our findings are driven by the sign of the news or the sign of the bias in managerial guidance (e.g., the downward bias in managerial guidance). 21 In summary, the results of various analyses in this section continue to show that bundled guidance is positively associated with the magnitude of analysts' revisions in response to guidance news. Although we cannot fully rule out the possibility that a correlated omitted variable remains, these results provide greater confidence in the conclusion that it is unlikely that our main findings are due to endogeneity issues.

Summary and conclusions
In this study, we find that bundled guidance influences analysts' expectations more than nonbundled guidance. This result is robust to a series of robustness tests. We further examine three possible explanations ("time pressure," "cognitive constraints," and "disclosure credibility") for the bundling effect. Collectively, these mechanisms help explain the usefulness of bundled guidance for analysts, but the economic significance of each mechanism is small.
Next, we find that analysts revise their forecasts more quickly following bundled guidance than non-bundled guidance. This evidence corroborates our main hypothesis that bundled guidance helps address analysts' time pressure and cognitive constraints and is more credible due to additional disclosures around EAs, which in turn facilitates analysts' forecast revisions. Finally, we show that investors respond more strongly to analysts' forecasts that follow bundled guidance than to those forecasts that follow non-bundled guidance. Our results thus suggest that the market values analysts' timely forecast revisions following bundled guidance.
Overall, our study contributes to the literature by investigating the effects of the increasingly common practice of bundled guidance on analysts' forecast revisions. Importantly, our results show that by aligning managerial guidance with analysts' routine revision activity, bundling facilitates analysts' forecast revisions. This finding provides insight into the growing body of literature that examines the impact of management guidance attributes on capital markets. Our study on the consequences of bundling on financial analysts also enhances our understanding of how analysts process information at the time of an EA versus at other times.
Lastly, our study extends prior studies on the interpretation role of analysts around corporate public information events by showing that investors value analysts' forecast revisions issued after bundled guidance more than those revisions issued after non-bundled guidance. This finding confirms the importance of analysts in issuing timely forecasts immediately after EAs, which helps explain why bundling can be an effective guidance strategy.

Accepted Article
This article is protected by copyright. All rights reserved. Notes: This appendix presents the effectiveness of entropy balancing for our DID analysis in panel B of Table 7 based on mean, variance, and skewness of control variables for bundled firms (BundledFirms=1) versus nonbundled firms (BundledFirms=0) derived before and after the application of the entropy balancing approach, respectively. The sample of bundled firms consists of bundled guidance issued by firms consistently issuing bundled guidance during January 1998 and August 2003, while the sample of non-bundled firms consists of nonbundled guidance issued by firms consistently issuing non-bundled guidance during January 1998 and August 2003. All variables are defined in Appendix 1.

Accepted Article
This article is protected by copyright. All rights reserved.

Frank (2000) method
The Frank (2000) method is based on the notion that for an unobserved variable (e.g., the variable that is not controlled by the model) to affect the results, it needs to be correlated with both  Table 3, is 0.193. To determine whether this ITCV is high or low, we compute the impact scores of other observed covariates in our baseline specification for comparison.
We find that MFNews, MFNews×HMFAccu, and MFLoss are most highly correlated with AFRev and MFNews×Bundled and, thus, have the highest impact scores. As shown in column (2) of the following  (1), the impact statistic (ITCV) indicates the minimum impact of a confounding variable that would be needed to render the coefficient on MFNews×Bundled statistically insignificant. The ITCV is defined as the product of the correlation between MFNews×Bundled (independent variable of interest) and the confounding variable (a simulated omitted variable) and the correlation between AFRev (dependent variable) and the confounding variable.
To assess the likelihood that such a variable exists, column (2) shows the impact of the inclusion of each control variable in our baseline model on the coefficient on MFNews×Bundled. The impact score is calculated as the product of (i) the partial correlation between MFNews×Bundled and the control variable and (ii) the correlation between AFRev and the control variable (teasing out the effect of the other control variables). All variables are defined in Appendix 1.

Accepted Article
This article is protected by copyright. All rights reserved.

Accepted Article
This article is protected by copyright. All rights reserved. Notes: This table presents descriptive statistics for the bundled and non-bundled guidance used in our study. Panel A reports the yearly distribution of our sample. Panel B reports the summary statistics of variables used in our main analyses. We indicate results in bold if significant at the 5% level. Panel C reports the summary statistics of variables used for partitioning bundled guidance in the cross-sectional analyses. Firm size is measured as the natural logarithm of market capitalization at the end of the last quarter. Institutional ownership is measured by percentage of shares owned by institutional investors at the end of the last quarter. Analyst coverage is measured by the number of analysts following a firm. Analyst experience is measured by the average experience of analysts following a firm, where an analyst's experience is calculated as the number of quarters for which the analyst has issued forecasts up to the current quarter. Analyst portfolio size is the size of the analyst coverage portfolio, measured by the average number of firms covered by analysts who follow a firm in a given quarter. Analyst busyness during EA, measured as the number of busy analysts, i.e., analysts who have more than three EAs issued by firms (including the focal firm) in their portfolios on the same day, divided by the total number of analysts following the focal firm. Conference call is an indicator variable equal to one if a bundled guidance is issued on the same day of a conference call, and zero otherwise. 10-K/Q filing is an indicator variable equal to one if a bundled guidance is issued on the same day of a 10-K or 10-Q filing, and zero otherwise. Length of 8-K filing is measured as the number of words in the 8-K filing issued on the same day of bundled guidance. We exclude 15,206 bundled guidance that do not have a concurrent 8-K filing on the same day from our cross-sectional analysis conditional on the length of 8-K filings. All other variables are defined in Appendix 1.

Accepted Article
This article is protected by copyright. All rights reserved. AFRev = a0 + a1MFNews +a2MFNews×Bundled +a3Bundled +∑akCONTROLS + ∑bkMFNews×CONTROLS + γ + w +ε. (1) In column (1), we do not include control variables. In column (2), we control for managerial guidance characteristics and their interaction terms with MFNews. In column (3), we further control for firm characteristics and their interaction terms with MFNews. In column (4), we control for confounding actual earnings news (EANews) and its interaction terms with firm characteristics. All variables are defined in Appendix 1. Constant terms and industry and year-quarter fixed effects are included but not tabulated for brevity. Robust t-statistics are reported in parentheses. Standard errors are clustered at the firm and year-quarter level. Coefficients that are significant at the 10%, 5%, and 1% levels are marked with *, **, and ***, respectively.

Accepted Article
This article is protected by copyright. All rights reserved. Notes: This table presents the robustness analysis on entropy balancing of mean, variance, and skewness of control variables between bundled guidance and non-bundled guidance. Panels A and B show descriptive statistics for control variables for bundled versus non-bundled guidance derived before and after the application of the entropy balancing approach, respectively. Panel C reports the estimation results based on the specifications in Table 3 (columns (1)-(4)). Variables are defined in Appendix 1. Robust t-statistics are reported in parentheses. Standard errors are clustered at the firm and year-quarter level. Coefficients that are significant at the 1% level are marked with ***.

Accepted Article
This article is protected by copyright. All rights reserved.