Analysts and Hedge Fund Activism

Analysts and Hedge Fund Activism Ryan Flugum John S. Howe July 31, 2017 Abstract Using a sample of hedge fund activist events from the years 2001 to 2014, we assess analysts reaction to hedge fund activism by considering changes in their recommendations. We find a preponderance of recommendations that move to or are reinstated at the Hold level following the arrival of a hedge fund activist, suggesting analysts are uncertain in this environment. Supporting our uncertainty hypothesis, we find that, over the course of an activist campaign, analysts experience a decrease in their ability to forecast earnings and a reduction in their earnings forecast activity levels. Overall, our findings suggest that analysts are uncertain in the presence of a hedge fund activist, limiting their ability to add substantial value in this setting. JEL classification: G10, G14, G23, G24, G34 Keywords: Analysts, Analyst Forecasts, Analyst Recommendations, Hedge Fund Activism Department of Finance, Collins College of Business, University of Tulsa, 118H Helmrich Hall, Tulsa, OK 74104. Email: ryan-flugum@utulsa.edu. Phone: (918)-631-2956 Department of Finance, Trulaske College of Business, University of Missouri, 401 Cornell Hall, Columbia, MO 65211. Email: howej@missouri.edu. Phone:(573)-882-5357

1 Introduction Analysts have a prominent role in capital markets. Beginning with Jensen and Meckling (1976), the literature has evaluated the socially productive role of analyst activities. These agents interpret and disseminate new information primarily through recommendations and earnings forecasts, activities which have been shown empirically to produce investment value (Womack, 1996; Barber, Lehavy, McNichols, and Trueman, 2001; Mikhail, Walther, and Willis, 2004; Loh and Mian, 2006; Barber, Lehavy, and Trueman, 2010). More important, the material produced by analysts is widely used by a variety of market participants. Consequently, it is critical to identify situations or events that could impair or enhance the social good that analysts attempt to provide. One such event whose frequency has increased dramatically over the recent decade is the intervention of a hedge fund activist. 1 Hedge fund activists intervene in a firm to engage with or change management, to seek board representation, to reallocate resources, to demand changes to the capital structure, and in extreme cases, to force the firm into a takeover (Clifford, 2008; Brav, Jiang, Partnoy, and Thomas, 2008; Greenwood and Schor, 2009). Although the literature supports the view that hedge fund activists improve the firms they target, such interventions simultaneously make the job of an analyst more difficult. Not only must the analyst continue assessing the firm, but he or she must also assess the intentions and consequences of the actions taken by the hedge fund activist. How analysts handle this tradeoff is unclear and we fill this gap by assessing their reaction through changes in their recommendations. Hedge fund activists have been shown to improve their target firm s fundamentals, reallocate resources to mitigate agency costs and improve efficiency (Clifford, 2008; Brav et al., 2008; Klein and Zur, 2009; Boyson and Mooradian, 2011; Brav, Jiang, and Kim, 2015a; Krishnan, Partnoy, and Thomas, 2015). Additionally, the announcement of a hedge fund activist s intervention is frequently met with positive short and long-term cumulative abnormal returns (CARs) around the hedge fund activist s intervention date (Brav et al., 2008; Boyson and Mooradian, 2011; Krishnan et al., 2015; Bebchuk, Brav, and Jiang, 2015). Regarding hedge funds in general (not just activists), analysts have been shown to revise their recommendations upward for stocks recently purchased by hedge 1 JP Morgan s report The Activist Revolution: Understanding and navigating the new world of heightened investor scrutiny, January 2015, finds that assets under management by hedge fund activists has increased from $12 billion in 2003 to $112 billion in 2014. 1

funds (Teo and Chung, 2011). Taken together, this evidence makes it reasonable to conjecture that analysts react positively, increasing the probability that target firms receive an upgrade following the arrival of a hedge fund activist. On the other hand, critics of hedge fund activism argue that these activists engage in myopic activities, seeking out short-term gains to the detriment of a firm s long-term sustainability (Coffee Jr. and Palia, 2015). As anecdotal evidence, a prominent reason that hedge fund activists target a firm is to increase dividends and shareholder buybacks, actions that create short-term gains, but are regarded by many managers and industry practitioners as detrimental to long-term growth. 2 Greenwood and Schor (2009) find that the positive CAR associated with the entrance of a hedge fund activist is concentrated primarily in the subsample of target firms that are eventually sold. Consequently, hedge fund activists may be better salespeople than true stewards of better corporate governance and firm profitability. In this scenario, it is reasonable to conjecture that analysts might react negatively to the arrival of a hedge fund activist, increasing the probability that target firms receive a downgrade following the arrival of a hedge fund activist. Last, it may be the case that analysts are unable to make clear recommendation assessments in the presence of a hedge fund activist. Not only are hedge funds notorious for their secrecy and behind-the-scenes negotiations, activists materially change the information environment of their targets. For example, Chen and Jung (2016) find that targeted firms cease providing financial guidance following the arrival of a hedge fund activist. Furthermore, hedge fund activists introduce other uncertainties, such as increases in a firm s level of tax avoidance and increases in the level of conservatism in financial statements (Cheng, Huang, Li, and Stanfield, 2012; Cheng, Huang, and Li, 2015). The combined lack of transparency regarding the hedge fund activist and the targeted firm might increase the uncertainty levels of analysts to a point at which they are unable to issue decisive recommendations. In this case, analysts are most likely to issue or move their recommendations to the neutral level of Hold following the arrival of a hedge fund activist. Our results suggest that analysts initial reaction to the arrival of a hedge fund activist is consistent with uncertainty. Using a sample of 2,138 hedge fund activist campaigns occurring over the years 2001 to 2014, we consider the changes in analyst recommendations for each targeted firm (henceforth referred to as target ) in the 25-month window centered at the hedge fund activist s 2 See As Activism Rises, U.S. Firms Spend More on Buybacks Than Factories, Wall Street Journal, May 26, 2015. 2

initial 13D filing date. 3 In the cross section of firms, we find hedge fund activism has no effect on the probability that an analyst will upgrade or downgrade a targeted firm, on average. Most consistent with uncertainty, Hold is the most frequent recommendation level chosen by analysts following the initial 13D filing and analysts are 19.52% more likely to move their recommendation to Hold from a prior level other than Hold. We consider which analyst, activist, and activism characteristics influence the observed recommendation changes and find that analysts experience with hedge fund activism and the size of the hedge fund activist are the primary factors influencing subsequent recommendation changes. For example, a 10% increase in the number of recommendations an analyst has given to target firms leads to a 1.36% increase in the probability an analyst will upgrade a target firm in the future. Similarly, a 10% increase in the size of the hedge fund activist increases the probability of upgrade by 0.43%. When we consider analyst experience more broadly, we find the most experienced analysts in our sample to be more likely to issue or reiterate a recommendation at the Hold level. In contrast, activists experience has a limited impact on the type of recommendation change occurring for target firms and there is variation in the degree with which the type of activist campaign influences recommendation changes. A possibility regarding our empirical finding that the arrival of a hedge fund activist increases the likelihood of an analyst moving their recommendation to Hold is that this action is purely mechanical. For example, Brav et al. (2008), Klein and Zur (2009), Boyson and Mooradian (2011), and Bebchuk et al. (2015), among others, demonstrate that the filing of the 13D by a hedge fund activist is frequently accompanied by abnormal returns and a subsequent positive drift in stock price. Therefore, analysts may be moving their recommendations to Hold simply because their assessment of the value of a hedge fund activist is captured by the abnormal announcement returns. To test this conjecture, we run our main results controlling for each firm s stock price in proximity to the most recent target price estimate from each respective analyst. We continue to find a roughly 20% increase in the probability that a target firm s recommendation will be moved to Hold following the arrival of a hedge fund activist. To substantiate our uncertainty hypothesis, we turn to the accuracy of analysts earnings fore- 3 Here targeted firm refers to the firm listed on the hedge fund activist s initial 13D filing. The SEC requires that a hedge fund activist with a greater than 5% stake and an intent to invoke change within the firm must file a Form 13D within 10 days of exceeding the 5% threshold. 3

casts and analysts forecast activity levels in the presence of a hedge fund activist. If hedge fund activism increases analysts uncertainty level, we expect to find a decrease in forecast accuracy taking place during the campaign, as well as greater dispersion in the cross section of analyst forecasts. Additionally, more uncertainty would imply that analysts have less information to convey through their earnings estimates, and consequently, we should observe fewer earnings revisions in the quarters following the arrival of a hedge fund activist. Consistent with our expectations, we find decreases to both analysts forecast accuracy and revision activity. Specifically, we find that analysts absolute annual forecast error and the crosssectional dispersion of their forecasts to be nearly twice as large for target firms, on average, when compared to the average firm. Furthermore, analysts issue more optimistic forecasts for target firms in the sense that target firms are unable to meet the earnings expectations set by analysts. Regarding forecast activity, we find target firms receive 6.61 (3.10) revisions, on average, in annual (quarterly) earnings forecasts in quarter t 4, t denoting the 13D filing quarter. This number compares to 5.78 (2.50) revisions, on average, in annual (quarterly) earnings forecasts in quarter t + 4. In a multivariate setting, we find statistically and economically significant decreases in annual (quarterly) earnings forecast activity of roughly 8% (5%), on average, following the arrival of a hedge fund activist. Overall, these reductions in accuracy and activity levels support our conjecture that analysts are uncertain in the presence of a hedge fund activist. The primary source of investment value that analysts are able to provide is through the predictive content of their recommendations. Although we find hedge fund activism to increase the level of uncertainty experienced by analysts, this may not directly translate into poorer predictive content from their recommendations. It is reasonable to assume recommendations hold less predictive power, given our evidence of increased analyst uncertainty, but hedge fund campaigns are often opaque and therefore investors may rely more on analyst recommendations. Such an effect may actually cause the predictive content of analyst recommendations to increase. Our remaining empirical tests address this issue by considering the short and long-term buy-and-hold abnormal returns following analyst recommendations for target firms. We find no superior predictive content from Buy or Sell recommendations that are issued to firms targeted by a hedge fund activist. Whereas we find Buy and Sell recommendations to have predictive content through six months in general, when we restrict our test to only the target sample, Buy and Sell recommendation 4

have no predictive content for the longer term horizons we consider. Our findings suggest that the increase in analysts uncertainty indeed translates into diminished investment value from their recommendations. Our contributions to the financial analyst and hedge fund activism literature are threefold. First, financial analysts continue to be important gatherers of information and therefore, the opinions expressed by these individuals following the arrival of a hedge fund activist should provide an alternative, and in many cases, independent view of the consequences of this type of activism. However, our results suggest that turning to analysts might not be prudent in this setting, as they appear to struggle with a greater level of uncertainty in the presence of a hedge fund activist. Second, we demonstrate yet another unforeseen external consequence attributable to the actions of hedge fund activists - a reduction in analyst ability that is most apparent in their earnings forecast accuracy. Last, our results are consistent with the view that analysts piggyback on existing news and ultimately bring little clarity to the hedge fund activism environment (Altinkiliç, Balashov, and Hansen, 2013). 2 Data 2.1 Activist sample To construct our sample of hedge fund activist events, we collect all 13D and 13D/A filings over the years 2001 to 2015 from the Audit Analytics (AA) database. The 1934 Securities Exchange Act requires any investor to file a 13D within a 10-day window of the date on which the investor exceeds a 5% stake in a publicly traded firm and has the intention to pursue some form of activist agenda. We consider the date on this form to be the hedge fund activist s initial target date and the beginning of their campaign with the firm. Additionally, any change in the investor s initial intentions or stake in the respective target firm requires the activist to file an amended Form 13D/A. The AA database includes all 13D and 13D/A filings from the SEC over the years 2001 to 2015 along with data classifications that include the name of the activist filer, the target firm s name, the percentage stake, and the stated intentions of the activist, which are disclosed in Item 4. Purpose of Transaction of the filing. The AA database categorizes these intentions into 36 designated subclasses, and then aggregates the sub-classes into six overall classifications of Investment Purposes, 5

Agreements, Concerns, Dispute, Discussions, Control and Other. For our tests, we collect all reason codes listed in the Form 13D and 13D/A that are filed throughout a campaign. Hedge fund activists are the primary group of interest in our study and AA does not distinguish this class of investors among their comprehensive database of 13D filers. Therefore, we take the following steps, outlined in Table 1, to identify whether the listed activist filer is a hedge fund activist. First, we begin with the complete sample of 13D and 13D/A filings in the AA database and identify the names of hedge fund activists by matching filer names to the those of hedge fund activists in the sample used by Brav et al. (2008) and Brav, Jiang, and Kim (2015b). 4 We restrict the remaining filings to only those labeled as an initial 13D, to those filings whose target has a share code of 10 and 11, and to those filings whose target has an available PERMNO and GVKEY for the CRSP and Compustat databases. These additional restrictions leave us with 6,259 unique shareholder activist filer names that must be verified as that of a hedge fund activist. To identify these remaining names, we follow a procedure initially put forth by Brunnermeier and Nagel (2004). 5 First, we check each filer s SEC-ADV filing to confirm that they charge an incentive fee and that they have at least 50% of their assets in hedge funds or at least 50% of their assets owned by high net worth individuals. In this filing, hedge fund investors fall under the Pooled investment vehicles (other than investment companies) classification in Section D of Item 5: Information About Your Advisory Business. As additional confirmation and for those filers that do not file a Form ADV, we use the web to confirm that the respective filer is a hedge fund. Because managers and hedge funds can sometimes be assigned different Central Index Keys(CIK) by the SEC, we aggregate all filings at the hedge fund level when possible. 6 After identifying all possible hedge fund activists, we make the following final restrictions to our 13D filing sample that spans the year 2001 to 2014. We delete those filers having only one initial 13D and whose only stated intention classification throughout the campaign (i.e. 13D and 13D/A) consists of Investment purposes, Not applicable, no change or no intent stated, Bankruptcy settlement, Stock delisted, Intends to sell or reduce stake, or any combination of these five 4 We thank Wei Jiang and her coauthors for providing us with their updated hedge fund activist sample used in Brav et al. (2008) and Brav et al. (2015b). 5 See Griffin and Xu (2009) and Ben-David, Franzoni, and Moussawi (2012) for other published studies using this method. 6 For example, Phillip Goldstein (CIK 1067621) is the manager of Bulldog Investors (CIK 1504304). We would aggregate these under Bulldog Investors. 6

because these filers are not likely to be true hedge fund activist campaigns. Moreover, if an activist filing is labeled as a 13D in the AA database and the same activist has filed a 13D for the same firm in the previous three years, we treat this as an amendment to the initial filing labeled as 13D. Additionally, we require that our target firms have annual financial data in Compustat for the 12 months before the initial 13D filings date, and that the target firm has a stock price greater than $1 on the initial 13D filing date. Including this stock price restriction keeps our results from being biased by micro cap and illiquid firms. In the event that the initial 13D filing date is not a trading day, we use the first trading day after the filing date. Our activist sample consists of 2,138 initial 13D filings from 382 unique hedge fund activists for 1,577 unique target firms. 7 Panel A of Table 2 shows the distribution of events by year along with the aggregate dollar value of the initial stakes of hedge fund activists within the year. An increase in activist activity precedes the financial crisis, peaking at 296 events during 2007 with an aggregate dollar value in initial stakes of roughly $24 billion. The ensuing decrease in activity during the financial crisis is expected because the hedge fund asset class struggled during this time period. Moreover, the frequency and dollar value of hedge fund activist events in the post-crisis period have yet to regain their respective pre-crisis highs. Panel B of Table 2 provides a distribution of hedge fund activist activity across the 12 Fama- French industry classifications. Within our activist sample, Business equipment (487 events), Financials (365 events), Consumer discretionary (270 events) and Healthcare (242 events) account for nearly 64% of our observations. In panel C of Table 2, we show that activists in our sample target five firms, on average, with few targets being engaged by more than one activist (1.36, on average). Table 3 describes the variation among the stated intentions listed in the 13D filings. Classification of intentions in Table 3 are not mutually exclusive. The most frequent intentions listed by hedge fund activists can be categorized as the intent to engage in some form of discussion with management (1,406 events; 65.76% of the sample) or to express some type of concern regarding the firm (936 event; 43.78% of the sample). The most contentious activist campaigns fall under the 7 Consistent with Greenwood and Schor (2009), we exclude 439 target events from Gabelli Investment. Gabelli Investment includes GAMCO Investors, Gabelli Funds and Gabelli Asset Management entities that are not considered activists. Many of the filings pertaining to Gabelli frequently listed Investment Purposes as their only stated intention. 7

Disputes category and in aggregate account for 22.68% of our sample. Aside from the standard investment purposes terminology that is used in nearly every 13D filing, the intent to change or nominate the board of directors and reserve the right to hold discussions with management account for nearly 55% of our sample. 2.2 Financial / Analyst / Institutional Ownership Data We obtain financial data from the Compustat database and we winsorize all Compustat data each year at the 1% level. We obtain stock price and shares outstanding data from CRSP and require the firm to have a share code of 10 or 11. In many of our multivariate tests, we require data regarding the hedge fund activist s overall size and initial stake in the target firm. AA does provide an initial stake number, but in some cases the database does not aggregate the listed shares held by all entities given in the 13D filing. We obtain a more accurate proxy for the initial state by collecting equity holdings data from the Thomson Reuters 13F database for the hedge fund activists in our sample who file a 13F. Our analyst coverage, recommendation, target price estimates, annual earnings forecasts and actual annual earnings per share data are taken from Thomson Financials Institutional Brokers Estimate System (I/B/E/S). Recommendations are given on a five-point scale in the database and we reverse the orientation so that one represents Sell and five represents Strong Buy. Additionally, we use the unadjusted I/B/E/S files containing the forecasted and actual annual earnings per share. Therefore, to assure that the unadjusted forecast is on the same per share basis as the actual earnings, we use the split adjustment factors in CRSP. 8 Last, we exclude all analyst data that have an anonymous analyst code (ANALYS or AMASKCD of 000000 ) because many of our empirical tests are at the analyst-firm level. Table 4 shows summary statistics for our sample of hedge fund activist target firms over the years 2001 to 2014. The characteristics of our sample are consistent with those identified in previous studies. 9 Many hedge fund activists can be categorized as classic value investors. As shown in Table 4, target firms are smaller in size, they have low market-to-book ratios, and they have negative market-adjusted buy-and-hold stock returns leading up to the arrival of the hedge fund activist. 8 We use the iclink file to merge I/B/E/S data with CRSP data, requiring that the Permno-Ticker match have a score of 0, 1, or 2 (out of a 0 to 6 range). 9 See Brav et al. (2008), Klein and Zur (2009), and Boyson and Mooradian (2011), among others. 8

The actual 13D filing announcement is met with an average cumulative abnormal return of 2.32% in the three days around the announcement date. Higher levels of institutional ownership and analyst coverage are also present: the average level of institutional ownership stands at 55.63% in our sample and there are 3.44 and 6.18 analysts providing recommendations and annual earnings forecasts for our target firms, on average. A majority of the events in our sample (1,393 events or 65.15%) involve target firms that have recommendations as well as annual earnings forecasts and the average consensus recommendation level immediately before the 13D filings is 3.51. 10 3 Empirical Methods and Results Our empirical analysis uses analysts recommendation changes around the start of a campaign to assess their reaction to hedge fund activism. In particular, we identify how the level of a target firm s recommendation and the probabilities of the firm being upgraded or downgraded are changed by the arrival of a hedge fund activist. We also determine the important analyst, activist, and activism characteristics that influence changes in these probabilities. A priori, it is not clear how an analyst should react to a hedge fund activist s intervention. Research by Brav et al. (2008), Boyson and Mooradian (2011), Brav et al. (2015a), and Bebchuk et al. (2015) would suggest that analysts reaction will be positive, with an increase in the probability of the target firm receiving an upgrade in recommendation or annual earnings forecast. However, analysts may be inhibited by the view that hedge fund activists exercise myopic investment tactics that are detrimental to the long-term profitability of the target firm (Greenwood and Schor, 2009; Coffee Jr. and Palia, 2015) suggests a negative reaction. Alternatively, analysts may grapple with greater uncertainty, thereby making Hold the most likely recommendation level analysts will choose to issue or reiterate in the presence of a hedge fund activist. 3.1 Upgrade and Downgrade Probabilities - Univariate To get a general sense of the impact the arrival of a hedge fund activist has on analyst recommendations, we initially use a univariate approach to track the movement of recommendations around the initial 13D filing date. Using a 25-month window centered at the initial 13D filing date, 10 We calculate the consensus recommendation level as the average of the last recommendation issued by each analyst in the 12 months before the 13D filing. 9

we measure the change of the recommendation prior to the initial 13D filing date to the recommendation issued during 12 months following the initial 13D date. We focus on a 25-month window for two reasons. First, prior literature has found that the average holding period of a hedge fund activist ranges from 12 to 24 months (Brav et al., 2008; Greenwood and Schor, 2009; Boyson and Mooradian, 2011). Second, our study is focused on the initial reaction the analysts have to a hedge fund activist s entrance into a target firm, requiring the use of a relatively short post-13d filing period window. In Figure 1 we plot the aggregate proportions of recommendation levels in the event time quarters around the initial 13D filing quarter (event time 0). The recommendation levels we consider are Sell (i.e. Sell (1) and Underperform (2) grouped together), Hold (3), and Buy (i.e. Buy (4) and Strong Buy (5) grouped together) and in each event time quarter, we consider each analyst s last outstanding recommendation in the current quarter. 11 We then calculate the proportion of recommendations at each level. Figure 1 shows little variation in the proportion of Sell recommendations, suggesting the analysts do not view the arrival of a hedge fund activist as negative. However, there is a clear increase in the proportion of recommendations that are at the Hold level and a contemporaneous decrease in the proportion of recommendations at the Buy level. Whereas the proportion of recommendations at the levels of Buy and Hold are nearly equivalent in quarters -4 and -3 at 46%, by quarter 4 the proportion of recommendations at Hold increases to 53.14% and the proportion of recommendations at Buy decreases to 36.89%. These recommendation movements are not consistent with a positive reaction from analysts. Rather, the increase in recommendations moving to Hold suggests that analysts are uncertain about the consequences of the entrance of a hedge fund activist and, therefore, exercise caution by choosing a neutral recommendation level. In Table 5 we take a more nuanced approach in analyzing target firm recommendation change frequencies and probabilities in the 25-month window centered at the 13D filing month. In particular, we consider each analyst s last outstanding recommendation in the pre-13d time period and first issued recommendation in the post-13d period. To be included in the analysis of Table 5, the analyst must have a newly announced recommendation in both the pre- and post-13d filing 11 Recommendation outstanding for longer than 12 months are not considered, as well as those from anonymous analysts (AMASKCD= 000000 ). 10

periods. These requirements result in a sample of 2,642 recommendation changes for 985 unique target events and 796 unique firms. In Panel A of Table 5 we show the recommendation change probabilities. The most common recommendation prior to the initial 13D date is at the Hold level, with nearly 46% of the recommendations at this level. The Hold level increases its majority share in the post 12-month period to roughly 50%. Also apparent is the greater number of pre-recommendations at levels Buy and Strong Buy (42.39% of sample) as compared to those at Underperform and Sell (11.81% of sample). We also find an increase in the number of Underperform and Sell recommendations in the post period (14.00% of sample) as compared to a decrease in the number of Buy and Strong Buy recommendations (35.91% of sample). More important, the actual recommendation change probabilities show movement to Hold dominates every pre-13d recommendation category. For example, those analysts whose recommendation is Buy in the 12 months prior to the arrival of a hedge fund activist have a 66.13% chance they will change their recommendation to Hold in the 12 months following the arrival date. Panel B of Table 5 provides additional evidence of analyst uncertainty about hedge fund activism. In this table we consider the frequencies of each magnitude change in recommendation level. For example, a recommendation change of -4 denotes a change of a Strong Buy recommendation in the pre-12 month window to a Sell recommendation in the post-12 month window. As shown in the table, those recommendations moving downward outweigh those moving upward by 7.96% (42.18% versus 34.22%, respectively). The difference is partly attributable to the greater number of Buy and Strong Buy recommendations in the pre-period. Overall, the univariate analysis of recommendation changes presented in Table 5 is most consistent with analysts being uncertain in the presence of a hedge fund activist. 3.2 Upgrade and Downgrade Probabilities - Multivariate To obtain more robust estimates of the effect that hedge fund activists have on analyst recommendation changes, we use a logit model at the analyst-firm level on a panel of firms in the I/B/E/S database from the years 2001 to 2015. 12 In addition to being able to identify the effect 12 Our event sample spans the years 2001 to 2014. However, we consider the changes in recommendations taking place in the following 12 months, thereby requiring us to extend the sample to 2015. 11

of being targeted, the multivariate framework allows us to control for firm characteristics that have been shown to influence both the level and the change in an analyst s earnings forecast and recommendation (Jegadeesh, Kim, Krische, and Lee, 2004). Our model specification is as follows. P r( Dep i,j,t ) = Λ(γ 0 + γ 1 T arget j +γ 2 Size j,t 1 + γ 3 ME/BE j,t 1 + γ 4 Return j,t 1 + γ 5 V olume j,t 1 + γ 6 IO j,t 1 + ɛ i,j,t ) (1) With regression (1) we use four different dependent variable specifications pertaining to recommendation changes. Dep i,j,t is an indicator variable that is one if the recommendation change from analyst i for firm j in quarter t is an upgrade, a downgrade, a recommendation that moves to Hold from a previous recommendation other than Hold, or a recommendation that is issued at the Hold level. The independent variable of interest in regression (1) is T arget j, which is an indicator variable that is one if firm j has been targeted in any of the previous quarters, [t 4, t], and zero otherwise. We ensure that T arget j is one only if the recommendation change comes after the initial 13D filing date. 13 The remaining independent variables control for those firm characteristics identified in Jegadeesh et al. (2004) as influential for analyst recommendations. They find that financial analysts tend to prefer high growth firms, firms with strong stock performance in the preceding months, and firms with high trading volume. The log of the market-to-book ratio of firm j, ME/BE j,t 1, is as of the most recent year end before the recommendation quarter t. Size j,t 1 is the log of the market cap of firm j as of the end of the most recent quarter, t 1. Return j,t 1 and V olume j,t 1 are the stock return and average daily trading volume of firm j over the previous two quarters, [t 2, t 1], respectively. Institutional ownership, IO j,t 1, is a firm s average institutional ownership over the quarters t 4 to t 1 computed from 13F filings. All of the regressions include industry fixed effects using the Fama-French 12 industry classifications. To control for within-firm and within-year correlation in residuals, we cluster standard errors by firm and year (Petersen, 2009; Thompson, 2011). 14 13 For those firms that are targeted by more than one activist in the same quarter, we consider the earliest target date as the only target date for the quarter. 14 All of our multivariate analyses in the remainder of the paper include fixed effects for industry and standard errors that are clustered by firm and year. For brevity, we do not discuss these variables in the remainder of our empirical results. 12

Table 6 presents our results using the above logit model. Models (1) through (4) use a dependent variable that is an indicator that is one if the recommendation is an upgrade, a downgrade, moved to Hold, or issued at Hold. Evident in the results of models (1) and (2) is the absence of any effect that hedge fund activists have on analysts propensity to upgrade or downgrade targeted firms. In each of these models, the loadings on T arget j are not statistically significant. On the other hand, there is a statistically significant shift in the preference of analysts toward moving their recommendations to a more neutral level (i.e., Hold). In model (3) of Table 6 we use a dependent variable indicating whether a recommendation has been moved to Hold from a level other than Hold. 15 In this model, the loading on T arget j is 0.178 (significant at the 1% level) which corresponds to a 19.52% higher likelihood that an analyst will move their recommendation to the Hold level within the four quarters following the arrival of a hedge fund activist. Similarly, when we use a dependent variable indicating whether a recommendation is issued at the Hold level, as in model (4) of Table 6, we again find a statistically significant, positive T arget j loading of 0.174 (significant at the 1% level). This loading corresponds to an 18.98% higher likelihood that the analyst will issue or reiterate a recommendation at the Hold level. Taken together, our multivariate results confirm our univariate findings and suggest that analyst recommendation preferences regarding target firms shift toward a more neutral level after the hedge fund activist intervenes. 3.3 Conditional Upgrade and Downgrade Probabilities - Multivariate We have shown that hedge fund activism materially affects analysts recommendation change probabilities. Yet it is entirely possible that some analysts are immune to the subsequent effects of a hedge fund activist. For example, Mikhail, Walther, and Willis (1997) show that analyst experience influences performance. Therefore, not only could analysts overall experience effect the probability of upgrading or downgrading a target firm, analysts experience covering targeted firms in the past might also play a role. In a similar vein, the hedge fund activist s experience could also influence analyst recommendations; hedge funds activists with more experience might provoke a positive recommendation change because prior literature has shown these activists earn higher long-term stock returns, partake in more aggressive tactics, and improve their target firms 15 Because we are concerned with those recommendations that actually have a chance to move to the Hold level, we exclude from our sample those observations where the recommendation is reiterated at the Hold level. 13

(Krishnan et al., 2015; Boyson, Ma, and Mooradian, 2016). Conditional on a firm being targeted, we would like to determine which characteristics influence an analyst s probability of upgrading or downgrading target firms. Therefore, we reduce our analysis to the sample of target firm recommendation changes occurring in the four quarters following each initial 13D filing date. 16 We use the following logit regression model specification. P r( Rec i,j,t ) = Λ(γ 0 + γ 1 EXP 1,i,t 1 + γ 2 EXP 2,i,t 1 + γ 3 EXP 3,k,t 1 + + γ 8 CAR j + γ 9 AST AKE + γ 10 ASIZE + 14 m=11 7 γ n AT Y P E n n=4 γ m CONT ROL m,j,t 1 + ɛ i,j,t ) (2) The dependent variables we use in regression (2), Rec i,j,t, are the same recommendation indicator variables we use in Table 6 regarding the recommendation change from analyst i for firm j at time t. The first set of independent variables of interest include those measuring the analyst and activist experience, EXP 1,i,t 1, EXP 2,i,t 1, and EXP 3,k,t 1. We calculate an analyst s overall experience, EXP 1,i,t 1, as the log of one plus the number of years prior to the current year that the analyst has at least one annual earnings forecast in the I/B/E/S database. We calculate an analyst s experience with activism, EXP 2,i,t 1, as the log of one plus the number of recommendations that the analyst has made prior to the current quarter for a firm currently targeted by a hedge fund activist. Activist experience, EXP 3,k,t 1, is the log of one plus the number of target events the activist has had prior to the recommendation quarter. We measure both analyst experience with activism and activist experience over the time period beginning in the first quarter of 2001. We use the first four years as the accumulation period for these variables, reducing the sample period of our analysis to the years 2005 through 2015. In addition to analyst and hedge fund activist experience, we would like to determine the influence of various activism and hedge fund activist characteristics. For example, Gantchev (2013) shows that the most contested activist campaigns can generate substantial expenses which are often shared by the activist and target firm. Consequently, analysts may be more apt to downgrade firms involved in these type of campaigns. We include indicator variables, denoted as AT Y P E, that represent each of the four non-mutually exclusive categories that each campaign falls under: (1) 16 We do include recommendation changes occurring in the current target quarter so long as the announcement date comes after the initial 13D target date. 14

Corporate Governance / Control, (2) Discussions, (3) Dispute, and (4) Concerns. Additionally, we include the hedge fund activist s initial stake, AST AKE, because a larger initial stake may signal a stronger commitment toward the long-term performance of the target firm, leading to a more positive recommendation change from analysts. 17 The hedge fund activist s size, ASIZE, is included because it is frequently the case that experienced hedge fund activists have greater assets under management (Krishnan et al., 2015; Boyson et al., 2016). 18 Because the market s reaction to the arrival of a hedge fund activist might be a valuable assessment tool for the analyst, we include the target firm s cumulative abnormal return, CAR j, over the three day window centered at the initial 13D filing date. 19 Last, we include the same set of firm control characteristics used in Table 6. The results in Table 7 consist of eight different models that are variations of regression (2). Models (1) and (2) have a dependent variable that is one if the recommendation is an upgrade and zero otherwise, while models (3) and (4) have a dependent variable that is one if the recommendation is a downgrade and zero otherwise. Models (5) through (8) use dependent variables indicating recommendations that are moved to Hold from some other level or issued / reiterated at the Hold level. The results of our models suggest that overall analyst experience and analysts experience with activism have competing effects. In models (1) and (2), analyst experience has a loading of -0.087 and -0.098 (significant at the 10% and 5% levels) suggesting that the probability of a target firm receiving an upgrade is decreasing in the years of experience of the analyst. 20 Moreover, the most experienced analysts have a higher likelihood of issuing or reiterating a recommendation at the Hold level, as analyst experience has positive loadings of 0.094 and 0.120 in models (7) and (8), significant at the 10% and 5% levels. Interestingly, we find evidence supporting a more positive recommendation propensity from analysts having the most experience with activism. In models (1) and (2), the loadings on analyst experience with activism are 0.136 and 0.178, each significant at the 1% level. Put differently, these 17 We obtain the hedge fund activist s stake, AST AKE, from the most recent 13F filing following the initial 13D filing date. To do this, we match each activist by name to their respective manager number in the Thomson Reuters Financial 13F database. We are able to match 153 activist hedge funds with manager numbers in the 13F database. 18 The hedge fund activist s size, ASIZE, is computed as the aggregate dollar value of all holdings reported by the activist in their most recent 13F filing following the initial 13D filing date. 19 We compute the cumulative abnormal return over the three-day window around the initial 13D filing date using a market model estimated over the prior 12 months. 20 For example, an increase in overall experience from four to five years equates to a decrease in the probability of upgrade of exp( 0.087 (ln(6) ln(5))) 1 = 0.0157 or -1.57% 15

loadings suggest that for a 10% increase in the number of recommendations an analyst has provided to a targeted firm in the past, the probability of an upgrade is approximately 1.36% and 1.78% more likely. Moreover, models (5) and (6) suggest that analysts with more activism experience are more likely to upgrade to a level other than Hold; the experience with activism variable has significant negative loadings of -0.134 and -0.191 in these models, significant at the 5% and 1% levels. Analysts are unaffected by the experience of the activist, as this variable has limited statistical significance across any of the models. On the other hand, analysts are much more likely to upgrade and less likely to downgrade firms targeted by larger activists. In models (2) and (4), the loading on activist size is 0.043 and -0.032, significant at the 1% and 5% levels. Activist campaign characteristics have varying degrees of impact on analyst recommendation probabilities. The initial market reaction to the 13D filng (CAR[-1,1]) has positive and statistically significant loadings in models (6), (7), and (8), which suggests that some analysts view much of the value of a hedge fund activist to be encapsulated in the abnormal returns from the announcement of their arrival. When we consider the type of activist campaigns, the statistically significant negative loadings of -0.175 and -0.130 (-0.206 and -0.144) on the Corporate Governance / Control and Discussions indicator variables in model (3) (model (4)) suggest that analysts are 16.06% and 12.20% (18.64% and 13.41%) less likely to downgrade firms targeted by hedge fund activists with these campaign agendas. The remaining campaign indicator variables and activist initial stake have little or no influence on the recommendation change probabilities. 3.4 Analysts Move to Hold and Target Price Proximity We find an increased liklehood that analysts will move or issue a recommendation at the Hold level following the arrival of a hedge fund activist. We attribute this effect to analysts uncertainty regarding the intentions of the hedge fund activist and the effects of the ensueing campaign. However, the market s reaction to the filing of a 13D by a hedge fund activist is generally positive, with substantial abnormal returns on the announcement date and subsequent stock price drift. 21 Consequently, the movement of analysts recommendations to the Hold level may be purely mechanical 21 See Brav et al. (2008), Klein and Zur (2009), Boyson and Mooradian (2011), and Bebchuk et al. (2015), among others, for a thorough documentation of the positive abnormal returns following hedge fund activist 13D filings. 16

in the sense that analysts view the stock price as having reached fundamental value following the 13D announcement. To test this conjecture, we use the same regression model defined in equation (1), with additional controls for the firm s stock price proximity to the respective analyst s target price estimate. Although we control for each firm s prior stock performance in our prior regression model (1), doing so does not directly measure the analyst s expectation of what the stock price should be. This issue is alleviated by considering how far the stock price is from the analyst s target price estimate at the time when the recommendation is announced. The logit regression specification we use is as follows: P r(mhold i,j,t ) = Λ(γ 0 + γ 1 T arget j + γ 2 P roximity i,j + γ 3 Recency i,j + γ 4 Size j,t 1 +γ 5 ME/BE j,t 1 + γ 6 Return j,t 1 + γ 7 V olume j,t 1 + γ 8 IO j,t 1 + ɛ i,j,t ) (3) We implement regression (3) using the identical recommendation sample used in Table 6 and imposing an additional restriction that the respective analyst have an outstanding target price estimate in the 12 month prior to the recommendation announcement month. 22 The dependent variable in regression (3) is the move-to-hold indicator used in Table 6. T arget j continues to be the independent variable of interest. The additional controls regarding analyst i s target price estimate include the primary measure of stock price proximity to the target price estimate, P roximity i,j, and a measure of how recent the target price estimate is to the recommendation as stale target prices may be less relevant to subsequent recommendations. Our measure of recency, denotedrecency i,j, is log of the number of days between the target price announcement date and the recommendation announcement date. We measure P roximity i,j using three different specifications beginning with a percentage proximity to the target price estimate. Specifically, we compute this percentage as the difference between analyst i s target price estimate for firm j and firm j s stock price as of the recommendation announcement date, scaled by firm j s stock price as of the recommendation announcement date. Second, we use a proximity indicator variable that is one if the difference between analyst i s target 22 We obtain the unadjusted target price files from I/B/E/S and use the cumulative adjustment factors in CRSP to assure that the target price estimate is on the same basis as the stock price as of the recommendation announcement date. 17

price estimate for firm j and firm j s stock price as of the recommendation announcement date is positive, and zero otherwise. Last, we measure proximity as the absolute value of the previously specified difference. If analysts increased propensity to move target firm recommendations to hold is purely mechanical, we would expect our T arget j indicator variable to be insignificant throughout our tests using regression model (3). We present the results from regression model (3) in Table 8. The increase in the likelihood of analysts moving their recommendations to Hold continues to be present. The loadings on our T arget j indicator variable in models (1), (2), and (3) are 0.194, 0.189, 0.191, each significant at the 1% level. These loadings equate to an increase in the probability of moving a recommendation to Hold of 21.47%, 20.77%, and 20.99%. The loadings on our target proximity controls suggest that the liklehood of analysts moving their recommendation to Hold is decreasing in the time horizon between their target price estimate and recommendation announcement. Furthermore, despite our Absolute Proximity to Target Price variable being negatively related to the liklehood of movement to Hold, the Above Target Price Indicator loading of 0.139 in model (3) (significant at the 1% level) does suggest that analysts are 14.87% more likely to move their recommendations to Hold, on average. 3.5 Discussion of Analysts Reaction to Hedge Fund Activism The results given in sections 3.1 to 3.4 suggest a material level of uncertainty exists in the analyst community regarding the intentions and consequences of hedge fund activism. There is a higher probability of an analyst moving, issuing, or reiterating their recommendation of a targeted firm to the Hold level in the four quarters following the arrival of a hedge fund activist. This result is robust to the consideration of analysts expectation of a firm s stock price, as measured by their target price estimate, as well as other traditional firm characteristics that have been shown to influence analysts recommendation decisions. Moving to Hold is not consistent with either strand of academic literature that paints hedge fund activism in a positive or negative light. When we look within the sample of target firms, we find there to be few characteristics, other than experience, that influence recommendation changes for target firms. Analysts having experience with firms targeted by hedge fund activists in the past are more likely to upgrade a target firm, while the most senior analysts appear to have a lower likelihood of upgrading target firms. 18