When do banks listen to their analysts? Evidence from mergers and acquisitions

Transcription

1 When do banks listen to their analysts? Evidence from mergers and acquisitions David Haushalter Penn State University Phone: (814) Michelle Lowry Penn State University Phone: (814) April 12, 2010 Forthcoming, Review of Financial Studies Corresponding author: Michelle Lowry, 313 Business Building, University Park, PA 16802, We thank Lubomir Petrasek for excellent research assistance. We thank Richard Bundro, Laura Field, Peter Iliev, Sandy Klasa, Urs Peyer and seminar participants at Case Western Reserve University, Hong Kong University of Science and Technology, INSEAD, McGill University, National University of Singapore, Singapore Management University, the University of Colorado, the University of Dauphine, and the University of Lausanne for helpful comments and suggestions. 1

2 When do banks listen to their analysts? Evidence from mergers and acquisitions April 12, 2010 Abstract: We examine the conflicts of interest and the flow of information between divisions of financial institutions. Using data on analyst recommendations and stockholdings of investment banks advising acquirers in mergers, we find evidence that information from investment banking flows to other divisions of the bank. Specifically, following a merger announcement, changes in a bank s stockholdings of the acquirer are positively associated with changes in recommendations by its analyst. This relation, however, does not exist before the merger announcement. Additional tests show that the relation between stockholdings and recommendations following a merger announcement is strongest when conflicts of interest for analysts are likely smallest. 2

3 Financial institutions engage in a broad range of activities. Investment banks, for example, commonly act as underwriters, lenders, asset managers, and providers of investment recommendations. By offering multiple services, information generated from one division can be transferred to another in ways not possible when divisions stand alone. Along these lines, Sufi (2004) and Suarte-Silva (2009) show how information gained from a bank s lending activities can benefit other divisions, for example by reducing costs of underwriting debt, enabling better certification of equity issues, and increasing the probability of becoming a lead underwriter. 1 However, combining divisions can also be detrimental to the divisions of a bank. Of primary concern are the conflicts of interest that can arise between divisions. For example, a bank s sell-side analysts can face pressures to provide overly optimistic recommendations to support the activities of the investment banking division. Likewise, a bank's asset managers can face pressures to make investment decisions to support investment banking, leading managers to purchase shares of client companies for affiliated mutual funds. What is unclear for banks is when and to what extent conflicts of interest and information sharing arise. Indeed, in an extensive review of the literature, Mehran and Stulz (2007) describe the empirical evidence on conflicts of interest for analysts of investment banks as inconclusive. As they discuss, although the potential for conflicts of interest often exists, there are also important factors that can mitigate these conflicts. For example, pressures for a bank s analyst to inflate recommendations can be offset by the reputational capital that the analyst has at stake. Similar arguments can be made for information sharing. Although combining divisions can create the potential for information to be shared across divisions, regulatory issues and conflicts of interest can mitigate or even eliminate such transfers. As a result, the extent of information sharing between divisions is also not obvious. 3

4 The objective of this paper is to shed light on these issues by examining the activities of the divisions of investment banks that are advising acquirer companies in mergers. We use this setting to examine questions regarding whether information generated from a bank s investment banking division helps its analysts and asset managers, whether investment banking leads to conflicts of interest for analysts and asset managers, and how conflicts of interest potentially affect information flow between divisions. Although conflicts of interest and information sharing within a bank can be ongoing, both can be particularly large and easier to measure empirically when the bank is advising an acquirer in a merger. For one, mergers are a large source of revenues for investment banks. 2 Consequently, mergers can result in increased conflicts of interest between a bank s investment banking and other divisions. In addition, mergers are important information events. As highlighted by Moeller, Schlingemann, and Stulz (2005), the value of companies can change dramatically around mergers. Therefore, around the time of a merger, the information flow between divisions of the bank advising the acquirer can be especially valuable. Our analysis focuses on the activities of an investment bank s analysts and asset management divisions around the time the bank is advising an acquirer in a merger. We begin by examining changes in investment recommendations by a bank s analysts for the acquirer. We examine these recommendation changes both before and after a merger announcement. Of primary interest is whether changes in the bank s holdings of the acquirer s shares are associated with changes in the bank s analyst recommendations: does the asset management division assess the analyst recommendations to contain valuable information, and does this assessment vary around the time that the acquiring firm announces and completes the merger? Because conflicts of interest can affect just one division (e.g., analysts or asset management) or multiple divisions (e.g., analysts and asset management), there are several 4

5 possibilities for the relation between the bank s recommendations and stockholdings. One possibility is that around the time of a merger, the asset managers of the advising bank do not change holdings of the acquirer when analysts change their recommendations for the acquirer. This would suggest that asset management views the analyst recommendations as being uninformative. Asset managers might perceive Chinese Walls as being strict and analysts as therefore having no incremental information about the acquirer, or they might view an analyst upgrade as reflecting pressures for the analyst to support a merger rather than containing new positive information. Alternatively, the combined effects of information sharing and conflicts of interest may produce a positive relation between changes in recommendations and stockholdings for the advising bank. There are two potential explanations for such a finding. First, analyst recommendations might be more informative at this time, potentially due to information gleaned from the investment banking division. Second, analyst recommendations might be less informative due to conflicts of interest, but asset managers might face similar conflicts of interest and therefore behave similarly. That is, investment banking may pressure analysts to upgrade the acquirer company stock and pressure asset management to buy the acquirer company stock, for example through affiliated mutual funds and/or client accounts. Using a sample of 1,197 mergers between 1995 and 2007, we find that following the announcement of a merger, when the advising bank analysts change their recommendations of the acquirer, these banks correspondingly change their stockholdings of that acquirer. In contrast, there is no association between changes in analyst recommendations and changes in stockholdings of the acquirer prior to the merger. We conduct additional analysis to disentangle whether the increase in the correlation between activities of analysts and asset managers following the merger indicates an increase in information flow across divisions or an increase in 5

6 conflicts of interest between divisions. Results from additional tests suggest that the increased relation between recommendations and shareholdings primarily reflects an increase in information flow between divisions of the bank. In particular, changes in advising bank analyst recommendations are more likely to lead those of non-advising banks following a merger announcement than prior to the announcement. In addition, the increase in the association between an advising bank s analyst recommendations and shareholdings is most pronounced for higher quality analysts who would be expected to provide a more reliable signal of new information. The results also show that the extent of the increase in the relation between recommendations and shareholdings varies with the conflicts of interest analysts face. There is little evidence, however, of similar conflicts within the asset management division. Specifically, advising banks do not increase shareholdings of the acquirer following upgrades by their analysts. Rather, the relation between the advising bank s investment decisions and analyst recommendation changes is concentrated around analyst downgrades. To the extent that upgrades are more likely driven by conflicts of interest, the finding suggests that asset managers follow recommendations of their analysts when the recommendations are least likely to reflect conflicts of interest. We reach similar conclusions when we follow Agrawal and Chen (2008) and classify banks according to percent of revenues from investment banking. The relation between recommendations and shareholdings is only for banks that are less dependent on investment banking and conflicts of interest are likely smallest. Our findings support the hypothesis that investment banking produces information that is shared across divisions of a bank, but that conflicts of interest among analysts affect the dissemination of this information. Thus, the results highlight the benefits and problems that can arise when divisions are combined. The findings indicate that to understand the importance of 6

7 information sharing and conflicts of interest, they need to be considered together. Our paper proceeds as follows. Section 1 outlines the data and sample characteristics. Section 2 empirically examines the relation between changes in the advising bank analyst recommendations of the acquirer and changes in the bank s holdings of the acquirer s shares. Section 3 investigates the extent to which this relation reflects information sharing or conflicts of interest. Section 4 provides an analysis of stock returns, which quantifies the potential gains from considering both the presence of an analyst recommendation change and the likely information set and incentives behind this change. Section 5 examines whether non-advising banks follow the recommendations of the advising bank s analysts. Finally, Section 6 discusses several robustness checks, and Section 7 concludes. 1. Data 1.1 Sample Construction Our data consists of mergers and acquisitions between 1995 and 2007, as obtained from the Securities Data Company (SDC) database. To ensure that the merger is a material event for the acquiring firm, we require the market value of the target to be at least 5% of the combined market capitalization of the bidder and the target. Both targets and acquirers are public firms traded in the U.S., and the acquirer must be publicly traded for at least three years prior to the merger announcement. We require each bidder firm to be followed by at least one analyst, as listed on the IBES recommendation database, and to be partially owned by at least one institution, as listed in the Spectrum 13(f) filings, one year prior to the announcement of the acquisition. Our analysis necessitates merging the SDC merger data, the IBES recommendation data, and the Spectrum institutional holdings data. For each merger, we identify the advisory 7

8 investment bank from SDC. We match by hand the identity of this bank with the IBES broker code and with the Spectrum institutional name. In matching the institutions between the SDC, IBES, and Spectrum databases, we are careful to account for both mergers between investment banks and for banks reporting under different names (e.g., Smith Barney Inc. and Smith Barney & Co). We attempt to match every investment bank that served as an advising bank in at least 10 deals over our sample period. Banks not matched include those, such as Houlihan, Lokey, Howard & Zukin and Greenhill & Co, LLC, that do not have either an asset management division or analysts. Mergers in which the advising bank either did not have an advisory arm (i.e., wasn t listed in IBES), didn t have an asset management division (i.e., wasn t listed in Spectrum), or served as an advising bank in less than ten deals are omitted from our sample. Institutional holdings data are reported in Spectrum quarterly, on March 31 st, June 30 th, September 30 th, and December 31 st of each year. 3 We calculate total shares held by each advising bank institution and each non-advising bank institution over the period beginning five quarters prior to the merger announcement and continuing through five quarters following the merger completion. For our analysis of analyst recommendations, we obtain from IBES all analyst recommendations on each acquirer firm. 4 We identify the advising bank recommendation outstanding three days prior to each institutional trading date, and we aggregate all non-advising bank recommendations outstanding as of this same date into a non-advising bank average consensus recommendation. To remove confounding interests, we do not include advising banks to the target firms in this non-advising bank consensus measure. We compute analyst upgrades as cases where an analyst revised its recommendation upwards, and analogously for downgrades. Kadan, Madureira, Wang, and Zach (2008) note that many analysts revised their recommendations downward in the wake of the Global Settlement, to comply with regulations 8

9 and present a more balanced set of recommendations, i.e., more equal portions of optimistic versus pessimistic ratings. The process of the banks reclassifying substantial numbers of recommendations resulted in large numbers of recommendation changes that were not information based. As Kadan et al show, banks generally reclassified their outstanding recommendations within a very short period of time, and these changes did not result in significant stock price reactions. Based on these findings, changes in recommendations related to the Global Settlement are not classified in our sample as upgrades and downgrades Sample Characteristics As shown in Table 1, these requirements result in a sample of 1,197 mergers. Among these 1,197, 154 were announced but never completed. Across the mergers, 555 are stock acquisitions, 196 are cash, and 446 are mixed. Many of the mergers have more than one advising bank. Due to our focus on conflicts of interest at the investment bank level, many of our analyses use advising bank-level recommendations and stock ownership. Our sample includes 1,413 advising bank-level observations. The sample is spread over time, with the largest number of transactions occurring in the late 1990s. This concentration is consistent with the finding in prior literature that M&A activity tends to be particularly high when the stock market is strong. Looking at the industry distribution, the largest number of mergers is in the business equipment and finance industries. Table 2 provides descriptive statistics for the full sample and sub-samples. The sample is divided by whether the advising bank has an analyst covering the acquirer and by whether the advising bank owns shares in the acquirer, both measured one quarter prior to the merger announcement. Several differences become apparent. The acquirers that are covered by the advising bank s analyst and owned by the advising bank are larger than other acquirers. This 9

10 finding reflects the more general result that both analyst coverage and institutional ownership are greater in larger firms, as shown by Gompers and Metrick (2001) and Barth, Kasznik, and McNichols (2001). The acquirers that are covered by the advising bank s analyst and owned by the advising bank also have higher market-to-book ratios, higher leverage ratios, higher profitability, and lower working capital as a fraction of total assets. Finally, relative merger size is significantly lower among companies in which advising banks provide analyst coverage and own shares. This difference in relative merger size is potentially driven by differences in firm size companies in which the advising bank owns shares are significantly larger, meaning a given target size will be relatively smaller. Table 3 examines the extent to which a bank s tendency to issue analyst recommendations or own shares in a firm is related to either expected or recent M&A advisory business by the investment bank. The analysis begins five quarters prior to the merger announcement and continues through five quarters following the merger completion (or through the withdrawal date for non-completed mergers). In conducting this analysis, we assume that an investment bank s expectations regarding the acquirer can change substantially during this period. A bank likely has a much better idea that there is an opportunity to advise a firm in a merger one quarter prior to the merger announcement than five quarters prior to the announcement. Table 3 shows an increase in both the advising bank s analyst coverage of the acquirer and in the advising bank s stockholdings of the acquirer in the period leading up to merger. These increases are, however, comparable to those of other non-advising banks. 6 For example, Panel A shows that the percent of advising banks with analyst coverage increases from 48% five quarters prior to the merger announcement to 57% one quarter before the merger announcement. Although this increase is monotonic across quarters, we observe no systematic pattern in 10

11 advising bank analysts as a percentage of all analysts covering the acquirer, which varies between 11.4% and 12.0%. This increase suggests that other analysts are also picking up coverage of the acquirer during this time. Columns (3) and (4) show that advising bank analyst recommendations are consistently more optimistic than non-advising bank recommendations, where analyst recommendations are measured on a scale from one to five, with one being the most optimistic. The average recommendation among advising banks changes from 2.07 five quarters prior to the merger announcement to 2.04 one quarter before the merger announcement. By comparison, the average recommendation level among non-advising banks during this time remains around In the quarters immediately following merger completion, average recommendations by the advising banks become increasingly positive (average = 1.96 one quarter post-completion). However, in subsequent quarters, advising banks recommendations drift back (average = 2.03 five quarters post-completion). This trend in recommendations by analysts from the advising banks might reflect the tendency of acquirers to select advisors that are most positive about the prospects of a merger. Alternatively, the increasingly optimistic recommendations by the advising banks around the merger potentially reflect pressures among analysts to win business for its investment banking division, pressures that are eased once the bank secures the business (i.e., a conflict of interest). 7 We further explore the importance of conflicts of interest below. As shown in panel B of Table 3, the percent of advising banks owning shares of the acquirer increases from 54% five quarters prior to the merger announcement to 58% one quarter before the merger announcement. The increase in the number of advising banks owning shares, however, is part of a larger trend of increasing institutional ownership in the acquirer. Shares held by advising banks as a fraction of shares held by all institutional investors decreases slightly, from 0.94% to 0.89%. In sum, the results provide little evidence of disproportionate 11

12 changes in ownership by the advising bank during this period. Much of the increases in both advising bank analyst coverage and advising bank share ownership appear to be driven by increases in the size of the acquirer firm over the quarters prior to the merger announcement. 2. Are changes in advising banks analyst recommendations and stockholdings correlated? Our main analysis begins by examining the relation between advising bank analyst recommendations for the acquirer and advising bank stockholdings in the acquirer. We examine this relation both before and after the merger announcement. As discussed above, a finding of no relation between changes in recommendations by a bank s analysts and changes in the bank s stockholdings would indicate that the analyst recommendations are not informative either before or after the merger. The bank s asset managers, for example, might view upgrades by its analysts as reflecting pressures for the analyst to support a merger rather than new positive information, or they may view the analysts as not gaining informational advantages from the bank s investment banking division. There are two potential interpretations of a finding of an increased relation between changes in recommendations by the bank s analysts and changes in the bank s stockholdings from before to after the merger. One possibility is that analyst recommendations might be more informative at this time, potentially due to information gleaned from the investment banking division. A second potential explanation is that analysts recommendations are not informative and reflect conflicts of interest. Asset managers behave similarly only because they face similar conflicts. Our tests attempt to sort through these possibilities. 2.1 Univariate Analysis of Changes in Recommendations and Stockholdings Table 4 provides descriptive evidence on the relation between analyst recommendations 12

13 of the acquirer and stockholdings in the acquirer, by the advising bank. The panels in this table show the average change in stockholdings conditional on an analyst upgrade, downgrade, or zero change in recommendation. We examine four measures of changes in the advising bank stockholdings of the acquirer: (1) Raw changes in shares held (2) Percentage changes in shares held, where the percent change is measured as number of shares held in quarter t minus number of shares held in quarter t-1, all deflated by the number of shares outstanding in quarter t-1 (3) Change in investment in acquirer as a percent of advising bank portfolio, Δ% of MCap Adv IB holdings in Acq' rt advisor portfolio = MCap Adv IB holdings all firms t MCap Adv IB holdings in Acq' r MCap Adv IB holdings all firms (4) Change in investment in acquirer as a percent of advising bank portfolio, net of average change in percent of non-advising bank portfolio, t 1 t 1 Δ % of advisor portfolio NET = Δ% of advisor portfolio Δ% of non advisor portfolio The data underlying the analyses represent a panel dataset, with one observation for each acquirer firm advising bank in each quarter. Panel A of Table 4 shows results for the entire event period: five quarters preannouncement through five quarters post-completion. Although it might not be surprising that advising banks often upgrade acquirers around a merger, downgrading is also common. During this period, there were 472 advising bank downgrades of acquirer firms, 773 upgrades, and 8,708 firm quarters with no change in advising bank recommendation. The frequency of downgrades is notable and contrasts strongly with recommendation patterns following IPOs, where affiliated analysts almost always initiate with very positive recommendations (see, e.g., Michaely and Womack, 1999). On average across the 472 downgrades, advising banks increased their 13

14 shareholdings by 44,279 shares, compared to an increase of 84,621 shares in firm quarters with no recommendation change and 122,323 shares in firm quarters with an analyst upgrade (by the advising bank). 8 The t-stat for the difference between downgrades and upgrade quarters equals 1.66, significant at the 10% level. Similarly, we observe a monotonic increase in the percentage change in shares held, as we move from downgrades, to no recommendation change, to upgrades, however the difference in percentage changes between downgrade quarters and upgrade quarters is not significant at conventional levels. Finally, advising bank holdings of the acquirer as a percent of the bank s total portfolio decrease by 0.016% conditional on a downgrade, compared to an increase of 0.11% of their total portfolio conditional on an upgrade (t-stat=2.33, p-value < 0.05). Looking at Panel B, in the five quarters leading up to the merger announcement there are almost twice as many upgrades of the acquirer by the advising bank analysts as downgrades (386 to 198). Results, however, indicate that there is no relation during this pre-announcement period between changes in these recommendations and changes in advising bank stockholdings. The relation between changes in analyst recommendations and stockholdings is strongest following the announcement of a merger, defined as the period between the merger announcement and 5 quarters following merger completion (or through the withdrawal date for non-completed mergers). Regardless of the measure for change in stockholdings used, the results indicate that advising banks invest significantly more shares in acquirers that their analyst upgraded versus those that they downgraded. Moreover, there is a monotonic increase in all three measures (from downgrade, to no change, to upgrade). For example, percentage change in shares held equals 0.13% across the 274 downgrades, 0.45% across the 4,976 firm quarters with no recommendation change, and 0.88% across the 387 upgrade quarters. There are obviously many factors that may cause an advising bank to change its 14

15 shareholdings in the acquirer firm, for example earnings releases, corporate events such as equity or debt offerings, voluntary disclosures related to expected future performance, etc. Moreover, Altinkilic and Hansen (2009) suggest that these factors tend to be correlated with analyst recommendation changes. For example, a very poor earnings release may prompt the analyst to downgrade the stock and the asset managers to sell the stock. If such events are more important in the post-merger announcement period, for example because of greater uncertainty regarding how the new merged firm will perform, then we might be capturing the effects of these other events. Notably, such events are likely to cause not only the advising bank to change their portfolio weightings, but also all other institutions as well. The final row of each panel in Table 4 examines the extent to which changes in advising bank analyst recommendations are associated with greater portfolio re-allocations by the advising bank, relative to all non-advising bank institutions. Results using this final measure, change in percent of advising bank portfolio net of the average change in percent of non-advising bank portfolio, show a pattern similar to that observed with other measures of advising bank investment. We find a significantly greater increase in upgrade quarters relative to downgrade quarters, but only in the post-merger announcement period. These results suggest that advising bank institutions are buying and selling in response to advising bank analyst recommendation changes rather than in response to other public information. They also suggest that non-advising bank institutions do not change their portfolio allocations in response to advising bank analyst recommendations. The reasons that only the advising banks (but not non-advising bank institutions) choose to respond to advising bank analyst recommendation changes is investigated further in section 5. In sum, univariate results provide preliminary evidence of a distinct shift around the time of a client firm merger announcement in the extent to which financial institutions listen to their analysts. Financial institutions only invest consistent with their analyst recommendations in the 15

16 period following the merger announcement. The following subsection analyzes this relation in a multivariate framework, and subsequent sections investigate whether this relation is driven by common conflicts of interest across both the analyst and asset management divisions or whether it reflects increased information sharing from the investment banking division that increases the accuracy of the analyst forecasts. 2.2 Regression Analysis of Changes in Recommendations and Stockholdings Table 5 examines this relation between changes in analyst recommendations and percent changes in stockholdings in a regression framework. The dependent variable in each regression equals the percentage change in shares held, as defined previously (number of shares held in quarter t minus number of shares held in quarter t-1, all deflated by the number of shares outstanding in quarter t-1). In columns 1 and 2, regression observations include the period beginning five quarters prior to the merger announcement and extending through five quarters following the merger completion (or through the withdrawal date for non-completed mergers). In column 3, the sample is restricted to those quarters preceding the merger announcement, and in column 4 the sample represents those quarters following the merger announcement. Regressions are estimated with maximum likelihood, firm fixed effects, and standard errors clustered by calendar year. The independent variable of greatest interest in these regressions is the change in advising bank analyst recommendation, defined as the advising bank recommendation outstanding immediately prior to the quarter t holdings date minus the advising bank recommendation outstanding immediately prior to the quarter t-1 holdings date. Analyst recommendations range from one to five, with lower numbers being more positive. The change in recommendation is multiplied by -1, so that a positive recommendation change can be 16

17 interpreted as an upgrade and a negative recommendation change as a downgrade. Control variables include dummies for the level of the advising bank recommendation at the end of quarter t-1. We only include dummies for strong buy, buy, and hold, because there are fewer observations with lower recommendations (sells and strong sells). We control for the change in the consensus recommendation across all non-advising bank analysts and for the change in market capitalization of the acquirer. We also include lagged percent of shares held by the advising bank in the acquirer firm, to account for the fact that a bank may be less likely to increase its holdings if it already holds a substantial number of shares. Following Parrino, Sias, and Starks (2003), we include the log of acquirer market capitalization, a dummy for whether the acquirer firm decreased dividends during the quarter, industry adjusted EBIT/TA for quarter t, and acquirer stock return net of market return over quarter t. Finally, we include an estimate of the number of acquirer shares that the advising bank would obtain automatically following completion of a stock merger, as a result of shares previously held in the target. For stock mergers in the first quarter following merger completion, we estimate this as the number of shares owned in the target prior to merger completion times the ratio of target to acquirer price one day prior to merger completion. This variable equals zero for all other firm quarters. 9 Results show that the relation between changes in recommendations and stockholdings varies around mergers. Column 1 shows a positive coefficient on change in advising bank recommendation, consistent with advising banks changing their stockholdings in the same direction as the change in analyst recommendations. However, the coefficient is not significant at conventional levels. Notably, the results in column 2 indicate that the lack of significance over the entire event period actually combines two very different effects: a highly significant relation over the post-announcement period and a lack of any significant relation in the preannouncement period. The interaction term, change in advising bank analyst recommendation * 17

18 post-merger dummy, is significantly positive (t-statistic = 3.12). In contrast, the interaction term advising bank analyst recommendation * pre-merger dummy is negative and not significant at conventional levels (t-statistic = -0.08). Columns 3 and 4 yield similar inferences. In column 3, where the sample only includes pre-announcement firm quarters, we observe no significant relation between analyst recommendation changes and changes in stock positions. However, the relation is positive and highly significant during the post-merger announcement period (column 4). Our evidence is somewhat inconsistent with the findings of Chan, Cheng, and Wang (2009) who find a significant relation between analyst recommendations and in-house trading throughout time. As a robustness check we examine the possibility that our finding of a lack of significance in the pre-merger announcement period is in some way related to the merger. For example, if asset management knew of the merger ahead of time, they might be trading on inside information during this period. In contrast, even if analysts knew of the information ahead of time, they would be unlikely to convey it in public releases. To examine this argument, we reestimate the pre-merger announcement regression (column 3 of Table 5), using quarters -10 through -6, relative to the merger announcement. No one within the bank is likely to foresee the merger this far ahead of time, thereby lessening the probability that merger-related information flows are affecting results. Results (untabulated) are qualitatively similar to those for quarters -5 through -1: there is no evidence of a relation between analyst recommendation changes and investments by asset management. Table 6 replicates the analysis in Table 5, defining the dependent variable in terms of the advising bank s portfolio allocation. Specifically, the dependent variable equals the market value of the advising bank s investment in the acquirer as a percent of the advising bank s total portfolio in quarter t, minus this percent in quarter t-1. This measure captures the extent to 18

19 which the investment bank re-allocates its investments either toward or away from the acquirer firm, in response to changes in analyst recommendations. That is, how is the bank changing its weights in the acquirer, relative to other firms in the bank s portfolio? Results using this measure are qualitatively similar. Specifically, in the period following the merger announcement, an advising bank significantly changes its portfolio weights in the acquirer firm in the same direction as changes in recommendations by its analysts. However, there is no evidence of a similar relation in the pre-announcement period. Results in Tables 5 and 6 show dramatically different patterns across the pre-merger announcement versus post-merger announcement periods. In the pre-announcement period, advising banks investment decisions are completely unrelated to the advice being provided to clients, suggesting the banks themselves do not view the changes in recommendations by their analysts to contain important new information. However, following the merger announcement, advising banks investment decisions are positively related to the advice being provided to clients. An interpretation of this relation is that analysts have higher quality information in this post-announcement period, possibly as a result of information sharing from the investment banking division. An alternative is that pressure from the investment banking division causes both analysts to upgrade the acquirer stocks and asset management to purchase acquirer shares (either on its own account or through its mutual funds / client accounts). The distinction between these explanations is an important one: the first scenario implies a greater value in analyst recommendations in the post-announcement period, while the second scenario actually implies the reverse. The following section focuses on this distinction. 19

20 3. Information Sharing versus Conflicts of Interest 3.1 Analyst upgrades versus downgrades As a first step toward understanding the source of the stronger relation between analyst recommendation changes and advising bank stockholdings in the post-merger announcement period, we compare the extent to which asset management divisions (of the advising bank) invest in response to affiliated analyst upgrades versus downgrades. If the investment banking division places pressure on both analysts and asset management divisions to support the acquirer, this pressure would likely take the form of analyst upgrades and stock purchases. Thus, if such pressure from investment banking explains the increased consistency during the postannouncement period, we would expect this consistency to be greatest around analyst upgrades. Alternatively, a finding of a stronger relation around downgrades would suggest that conflicts of interest are stronger for the analysts, less for the asset management side. The asset management side might be more likely to respond to downgrades because they are less likely to be biased by conflicts of interest. The regressions in Table 7 are similar to those shown in Table 5 except that the sample in column 1 is limited to firm quarters with an analyst upgrade or no recommendation change, and the sample in column 2 is limited to firm quarters with an analyst downgrade or no recommendation change. The dependent variable in each is the percentage change in advising bank shareholdings in the acquirer, as defined earlier. Independent variables include the change in advising bank recommendation in the pre-announcement period, the change in advising bank recommendation in the post-announcement period, plus the same control variables as in Table 5. Across both regressions, upgrades are denoted as a positive recommendation change and downgrades as a negative recommendation change. Regressions are maximum likelihood, with firm fixed effects and standard errors clustered by calendar year. 20

21 In column 1, we find no relation between analyst upgrades and changes in advising bank holdings. In contrast, column 2 shows a significant relation between downgrades and changes in advising bank holdings of the acquirer. When the advising bank analyst downgrades the acquirer, the advising bank is significantly likely to sell more shares. The coefficient of 0.51 indicates that a downgrade in analyst recommendations is associated with decrease in the advising banks holdings that is roughly a half percentage point greater than when there is no change in the analyst recommendations. Results (not tabulated) are qualitatively similar when the dependent variable is defined as the change in the advising bank s investment in the acquirer as a percent of the advising bank s total portfolio. The finding that changes in the advising bank s shareholdings are only related to analyst downgrades provides preliminary evidence against the idea that the relation between analyst recommendations and asset management investments following the merger announcement reflects pressure (from the investment banking division) on both divisions to support the acquirer. Rather, results suggest that the asset management divisions tend to place more weight on analyst downgrades, perhaps because they are less likely to be affected by conflicts of interest. 3.2 Abnormal returns to analyst recommendations If information sharing across divisions increases the accuracy of the advising bank s analyst recommendations in the post-announcement period, we would expect a greater market reaction to the recommendation changes in this period. Alternatively, if conflicts of interest decrease the quality of advising bank analyst recommendations in the post-announcement period, we would expect less of a market reaction in this period, particularly for analyst upgrades. Panel A of Table 8 shows the abnormal return around analyst recommendation changes 21

22 in the pre-announcement and post-announcement periods, where abnormal returns are defined as the cumulative firm return over days -1 through 0, net of the value-weighted market return over this same period. 10 Day 0 represents the day of the recommendation change. Row 1 shows the absolute value of the abnormal return across all advising bank analyst recommendation changes, column 2 shows the abnormal return across upgrades, and row 3 across downgrades. Results indicate that the magnitude of the abnormal return is greater in the postannouncement period, particularly with respect to downgrades. The abnormal return to upgrades is insignificantly different between the pre-announcement and post-announcement periods. However, the abnormal return to downgrades is -3.3% in the pre-announcement period versus -6.0% in the post-announcement period, a difference that is significant at the 1% level (t-statistic = -2.91). The market infers more information from the downgrades in the post-announcement period. If advising bank analysts have more value-relevant information in the post-announcement period, then they may be more likely to issue recommendations prior to other analysts in this period. Panel B examines the ordering of analyst recommendations in the quarters preceding versus following the merger announcement. With respect to upgrades, we observe small differences in the ordering of recommendations. In the pre-announcement period, 20% of advising bank upgrades are followed within 14 days by a non-advising bank upgrade. This increases to 24% in the post-announcement period. In contrast, the pattern of advising bank analyst downgrades differs much more substantially. Only 11% of advising bank downgrades are followed by non-advising bank downgrades in the pre-announcement period, compared to 26% in the post-announcement period. The increased tendency of advising bank analysts to lead other analysts in the post-announcement period is consistent with these analysts having more information during this period. 22

23 Panel C examines the effects of recommendation order and analyst affiliation (advising bank versus non-advising bank) jointly. We regress the abnormal return around recommendation changes on a dummy equal to one if another analyst issued a similar recommendation change in the past 14 days, a dummy indicating if another analyst issued a similar recommendation change on the same day, a dummy indicating if the recommendation change was by the advising bank analyst during the pre-announcement period, and a dummy indicating if the recommendation change was by the advising bank analyst during the post-announcement period. Column 1 focuses on abnormal returns around upgrades, and column 2 around downgrades. Consistent with predictions, upgrades are associated with average positive announcement returns, and downgrades with average negative returns. When more than one upgrade (downgrade) is issued on the same day, the return is significantly higher (lower). In contrast, when a similar recommendation change has been issued by another analyst within the past 14 days, the return is attenuated. Interestingly, the greater market response to advising bank analyst recommendation changes in the post-announcement period observed in panel A is completely explained by the effects of recommendation order. Coefficients on advising bank*preannouncement and advising bank*post-announcement are approximately equal. Results across the three panels are consistent with advising bank analysts having more information in the postannouncement period, and this information enabling them to lead other analysts in their recommendation changes. Because these analysts are more frequently the first to disseminate certain information in the post-announcement period, the market tends to react more strongly to their recommendation changes. 3.3 Analyst quality and sources of banks revenue Results to this point suggest that information sharing across divisions of financial 23

24 institutions explains the strong relation between analyst recommendation changes and investments in the post-merger announcement period, but that in certain cases this relation is mitigated by conflicts of interest among analysts. Notably, prior literature suggests crosssectional variation in both conflicts of interest and analyst quality. This section attempts to partition banks on whether analysts are more susceptible to conflicts of interest and to partition analysts on their ability to convert information into value-relevant recommendations. If the relation between the advising bank s analyst recommendations and investments is caused by information sharing, we expect this relation to be strongest among the highest quality analysts and among the banks where conflicts of interest are lowest. To examine the extent of conflicts of interest within financial institutions, we classify financial institutions based on their sources of revenue. Following Agrawal and Chen (2008), we posit that analysts working for institutions in which investment banking is a more important source of revenue will face greater conflicts of interest, for example stronger pressures to upgrade stocks of companies for which the bank has recently served as advising bank on an acquisition. If pressure from investment banking also extends to asset management divisions, we would expect the higher conflict of interest banks to be more likely to buy these same stocks. For each publicly traded advising bank financial institution, we obtain source of revenue data from the bank s 10K. Because not all banks in our sample are publicly traded, this limits us to 25 of the financial institutions. However, these 25 banks served as advising banks in the majority of our acquisitions. Financial institutions are required to disclose the source of their revenues, and the banks generally break down the revenues into those from investment banking, as well as those from various other activities on which we are not focusing. Thus, for each of these banks in each year during which they were publicly traded, we are able to determine the fraction of revenues from investment banking. 24

25 To examine analyst quality we follow Loh and Mian (2006) and use the average prior forecast error of each advising bank analyst in our sample, i.e., the analysts at the advising bank issuing recommendations on the acquirer firm. For each of these analysts, we collect data on quarterly earnings forecasts he or she has made (on all firms he or she follows) over the prior three years, where forecasts consist of the last forecast made prior to the end of the forecasted firm s fiscal quarter. We calculate the forecast error as the absolute value of the difference between the forecast and actual earnings, deflated by the absolute value of earnings. Although we expect percent of revenues from investment banking to be positively related to the level of analyst recommendations, we do not expect any relation between analyst quality and the level of analyst recommendations. Table 9 confirms both these predictions. The dependent variable is the level of analyst recommendation, re-ordered such that a strong buy receives the highest possible value (5), while strong sell receives the lowest possible value (1). Analyst recommendations for each firm are measured at the end of the first quarter following the merger announcement. Consistent with Agrawal and Chen (2008), we find that recommendations are significantly more positive for firms that receive a greater portion of revenues from investment banking (t-stat = 2.35). In contrast, there is no relation between the analyst forecast error and the level of the recommendation. Control variables indicate that financial institutions tend to issue more positive recommendations about larger firms, and they tend to issue more positive recommendations about firms that are making stock acquisitions. Table 10 uses these proxies to categorize banks according to the magnitude of conflicts of interest and analyst quality. Specifically, for each year, we classify banks with above-median (below-median) percent of revenues from investment banking as high (low) investment banking. Similarly, analysts with above-median (below-median) forecast errors are classified as low (high) quality analysts. For each firm, the sample consists of the first quarter following the 25

26 merger announcement through five quarters following the merger completion. Similar to Table 5, Table 10 shows maximum likelihood regressions of the percent change in advising bank holdings of the acquirer on changes in their analysts recommendations, with firm fixed effects and standard errors clustered by calendar year. The only difference between column 1 in this table and column 4 in Table 5 is that the sample is restricted to those mergers for which we have source of revenues data for the advising bank and prior forecasts for the analyst. Similar to prior findings, we find a significant positive relation between changes in advising bank analyst recommendations and changes in advising bank shareholdings of the acquirer firm. To the extent that information is shared between the investment banking division and the analysts, we would expect to observe the strongest relation between advising bank recommendations and investments among high-quality analysts, who are better able to interpret the extra information and translate it into meaningful recommendations. A comparison of results across columns 2 and 3 provides only weak support for this prediction. The relation between analyst recommendation changes and changes in advising bank shareholdings is significant at the 10% level among the high quality analysts (t-statistic = 1.64). 11 In comparison, the relation between analyst recommendation changes and changes in advising bank shareholdings is significant at the 12% level among the low quality analysts. In unreported results, we divide the sample into terciles based on analyst quality: analysts in the highest tercile are considered high quality, those in the lowest tercile as low quality, and those in the middle tercile are omitted. Based on this categorization, we observe much more substantial differences. The relation between analyst recommendation changes and changes in advising bank shareholdings is significant at the 1% level among the high quality analysts (coef=1.04, t-stat=3.24), but not significant at conventional levels among the low quality 26