The Effect of Mergers on Human Capital: Evidence from Sell-Side Analysts PARTH VENKAT * NOVEMBER 2016 Abstract I find that when brokerage houses merge, the analysts of acquiring houses tem- porarily produce less accurate estimates. This temporary impairment suggests that the merging process can distract high-skill employees. Further, among redundant an- alysts, high-quality target-house analysts are the most likely to leave upon merger announcement. This suggests that employees exercise outside options when redeploy- ment requires abandoning human capital. As a consequence of these effects, forecast error in merging houses remains elevated by 10% in relation to non-merging houses for two years. I conclude that mergers can temporarily, but significantly impair firms’ ability to acquire, develop, and retain human capital. * Department of Finance, McCombs School of Business, University of Texas at Austin, [email protected]. I would like to thank Laura Starks, Jonathan Cohn, Cesare Fracassi, Andres Almazan, and Michael Clement for their invaluable advice and support; Jacelly Cespedes, William Grieser, Shuo Liu, Zack Liu, Gonzalo Maturana, Carlos Parra, Nathan Swem, Adam Winegar, and Ben Zhang for their comments as colleagues; the McCombs Finance Department for their seminar comments; Jarrad Harford and the other participants at the FMA Student Symposium; and Paul Irvine for his discussion at the FMA. All errors are my own.
49
Embed
The E ect of Mergers on Human Capital: Evidence from Sell-Side … · PARTH VENKAT NOVEMBER 2016 Abstract I nd that when brokerage houses merge, the analysts of acquiring houses tem-porarily
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
The Effect of Mergers on Human Capital:Evidence from Sell-Side Analysts
PARTH VENKAT∗
NOVEMBER 2016
Abstract
I find that when brokerage houses merge, the analysts of acquiring houses tem-
porarily produce less accurate estimates. This temporary impairment suggests that
the merging process can distract high-skill employees. Further, among redundant an-
alysts, high-quality target-house analysts are the most likely to leave upon merger
announcement. This suggests that employees exercise outside options when redeploy-
ment requires abandoning human capital. As a consequence of these effects, forecast
error in merging houses remains elevated by 10% in relation to non-merging houses
for two years. I conclude that mergers can temporarily, but significantly impair firms’
ability to acquire, develop, and retain human capital.
∗Department of Finance, McCombs School of Business, University of Texas at Austin, [email protected] would like to thank Laura Starks, Jonathan Cohn, Cesare Fracassi, Andres Almazan, and Michael Clementfor their invaluable advice and support; Jacelly Cespedes, William Grieser, Shuo Liu, Zack Liu, GonzaloMaturana, Carlos Parra, Nathan Swem, Adam Winegar, and Ben Zhang for their comments as colleagues;the McCombs Finance Department for their seminar comments; Jarrad Harford and the other participantsat the FMA Student Symposium; and Paul Irvine for his discussion at the FMA. All errors are my own.
While existing research characterizes the overall value implications of mergers, delving
into how mergers affect the value of specific assets, through which mergers create or destroy
value, remains challenging.1 Perhaps least understood is how mergers impact the value of
human capital, arguably the most important asset class in the modern economy.2 While
merger synergies may increase the value of human capital, the process of integrating two
workforces may impose costs that limit those synergies. In a recent survey, companies that
reported their own mergers as “failing” assigned some blame to “people and integration
issues.”3 Certain issues can be short-term, such as employee or management distraction,
while others can be longer term, such as unresolved cultural mismatch or lost and unreplaced
key talent. The goal of this paper is to understand how mergers impact merging firms’ ability
to acquire, develop, and retain human capital.
To understand how mergers impact the value of human capital assets, I explore the
performance and retention of sell-side analysts employed by merging brokerage houses. This
setting conveys several advantages. First, analyst groups consist almost solely of human
capital (i.e., their expertise and connections), allowing me to isolate the effect of mergers on
human capital from effects on other classes of assets.4 Second, analysts publish a stream of
corporate earnings forecasts that can be compared to actual earnings to construct frequently
time-varying measures of employee performance. Third, because analysts can be tracked
across brokerages, I can study the retention and separation of analysts. Finally, industry
consolidation provides a diverse set of mergers, creating a statistically powerful setting.
The primary contribution of my paper is to document and interpret two facts. First,
analysts who work for acquiring houses before a merger become less accurate after the merger.
This effect dissipates within four months, suggesting that distractions due to the integration
1See Betton, Eckbo, and Thorburn (2008) on combined announcement returns and Healy, Palepu, andRuback (1992) and Andrade, Mitchell, and Stafford (2001) a on operating performance.
2I follow Goldin (2016) and explicitly define human capital as “the stock of skills that the labor forcepossess.” I cite to Acemoglu (2002), Autor, Levy, and Murnane (2003), and Abowd, Haltiwanger, Jarmin,Lane, Lengermann, McCue, McKinney, and Sandusky (2005) with respect to the increasing importance ofhuman capital.
3The survey of almost 90 M&A professionals from McKinsey & Company lists issues such as “culturalmismatch, loss of key talent, lack of management commitment [and] lack of employee motivation” (Deutschand West (2010) and referenced in Shermon (2011)).
4See Brown, Call, Clement, and Sharp (2015) for evidence about the value of connections to managementand Swem (2016) about the value of connections to institutional investors
1
of operations or cultures temporarily impair employee performance. Second, target analyst
monthly attrition increases substantially after a merger is announced, especially for high
quality target analysts. This loss of key talent may have longer-lasting implications for
the value of a merged houses’ collection of human capital.5 As a consequence of these two
effects, the average forecast error of all forecasts produced by merging houses increases by
10% relative to forecasts of non-merging houses for the same set of covered firms. To put
this effect in context, it is slightly larger than the difference between a perfect estimate and
the estimate at the 25th percentile.6 Consistent with a drop in output quality, the combined
houses reduce overall equity coverage by 5%.
I document the first fact that acquiring house analysts suffer temporary output im-
pairment using a difference-in-differences framework that includes individual analyst fixed
effects.7 This drop is larger than the difference between the average star analyst’s accuracy
and non-star analyst’s accuracy, where star is defined by Institutional Investor All-America
Research Team designations. This impairment is consistent with Schoar (2002), which finds
that mergers can result in the impairment of assets already in place. The effect is short
lived, dissipating within a year, which implies merger impairments to individual analysts’
human capital are temporary rather than permanent. Anecdotal evidence from discussions
with sell-side analysts suggests that employees can be distracted by junior-staff shuffling,
training, client-base expansion, or moving offices.
In order to study how mergers impact brokerages collection of human capital, I construct
a new measure of analyst quality. Previous literature typically uses star designations to proxy
for quality. However, this approach omits public information that would be indicative of an
analyst’s historical performance and cannot differentiate between non-star analysts. I create
a more comprehensive measure by fitting a logit regression for non-merger analysts to predict
negative career outcomes.8 By conditioning on actual decisions brokerage houses make, such
5There is no significant change in the accuracy of target house analysts who are retained or any changein acquiring house analyst attrition.
6Forecast error increases in 27 of 34 mergers. This effect is not driven by outliers and is robust tononparametric specifications
7Target analysts who keep their jobs improve their accuracy at times, but this effect is not significantacross all mergers.
8I consider instances when an analyst leaves the data entirely or moves to a less prestigious house as a
2
as analyst terminations, this regression captures brokerage houses’ revealed preference for
analyst observables. I define quality as 1 minus the predicted values from this regression.
Using quality, I study attrition to document fact two. I find that during the time between
the merger announcement and its completion, which I term the interim period, target analyst
monthly attrition increases to 13% (compared to 1% unconditionally) relative to similar
analysts not involved in a merger, while the acquiring house analyst attrition does not
change. This is true even though acquiring house analysts are not systematically higher
in quality. Focusing on target analysts’ attrition, I show that the highest quality target
analysts’ (top quintile) monthly attrition increases to 20%, while the lowest quality target
analysts’ (bottom quintile) monthly attrition only increases to 8%.9 While this might be
partially driven by cost savings (because wages are likely correlated with quality), for the set
of target analysts who separate during a merger, higher quality analysts are more likely to
find another analyst job, suggesting that the availability of outside options impacts target
analyst separation decisions.
To understand why high-skill analysts leave, I examine redundancy. Analysts may antic-
ipate being asked to cover new firms (i.e., be redeployed) if they cover firms already covered
by the acquiring house analysts.10 However, redeployment may be costly for analysts, be-
cause it may require abandoning their firm and industry expertise and connections.11 I define
Redundancy as the fraction of firms an analyst covers before the merger that are also cov-
ered by the other merging house. Redundant target analyst attrition increases significantly
more (22%) than attrition for unique analysts (Redundancy equal to 1 and 0) from the same
house (6%). This effect is over 50% larger for the highest quality target analysts, which is
consistent with redeployment being costly and the existence of outside options increasing
with their quality. Further, when target analysts get new jobs, either at new firms or the
merged entity, they almost always continue to cover the firms that they previously covered.
negative career outcome and use observables, such as star rating, accuracy, productivity, optimism, firmscovered, tenure, and experience.
9This effect is monotonically increasing across quintiles.10Brokerage houses require a singular view on firms they cover.11Brown et al. (2015) cites the ability to communicate with senior management as an analysts primary
driver of value. These relationships take time to build.
3
In contrast to Tate and Yang (2015), who show that diversifying mergers can improve out-
put by facilitating human capital redeployment, my evidence suggests that redeployment
synergies may be difficult to unlock in industries that involve employees who have valuable
but specific human capital.
To explore whether high-quality analysts are choosing to leave during mergers, I exam-
ine how analysts react to an increasing probability of termination. Because redundancy is
generally only known after the merger announcement, it is a plausibly exogenous shock to
the probability that an analyst separates from the target house in the interim period. I
find that the increased probability of separation reduces productivity, which is measured as
the number of reports an analyst produces. This result suggest that when target analysts
know that job retention likely involves switching roles, they shift their efforts elsewhere (e.g.,
towards leisure or finding a new job) rather than increase their efforts to compete within the
merged firm.
A related industry-wide increase in forecast error has been documented by Hong and
Kacperczyk (2010), who attribute the merger-related performance decline to industry con-
solidation and the accompanying decrease in analyst competition. Hong and Kacperczyk
(2010) argue that forecast error increases because analysts intentionally decide to bias es-
timates upwards in order to cater to corporate clients. The differential impact cannot be
explained by competition declines because competition has the same impact on estimates
for the same firms. Similarly, because competition does not recover, it cannot explain the
temporary forecast error increase. Using whether an underlying firm is covered by both the
target and acquirer versus just one or the other as a shock to the intensity of the compe-
tition decline, I divide estimates into large competition shock and small competition shock
estimates. I find no differential impact on forecast error or positive bias further supporting
the idea that competition declines are not the cause of the increased forecast error.
To further differentiate unintended errors from workforce integration issues from intended
bias from changing priorities, I exploit a structural shift in analyst incentives, specifically
the 2003 Global Analyst Research Settlements (GARS), which created “brick walls” between
4
the research and investment banking divisions of large investment firms.12 Even after this
plausibly exogenous shock to analysts’ incentives, the results persist, implying that results
are more likely due to workforce integration issues as opposed to changing priorities.
Finally, I run two cross-sectional tests in order to confirm that workforce integration
issues are related to human capital issues and overall forecast error increases. I analyze the
merger announcements’ text and mark 14 mergers in which human capital or expanding
services is not mentioned as a merger motivation. Independently, I mark the tercile of
mergers that have the greatest ex post increases in forecast error. I restrict my tests to the
two subsets above, and in both subsets I find that the retained analyst accuracy declines
by 70% more than it does in the full sample and that high-quality target analyst attrition
increases almost 100% more than in the full sample. Thus, workforce integration issues (e.g.,
employee distraction, loss of key talent) are largest for mergers for which it is publicly stated
that human capital is not the primary motivation for the merger and for mergers that suffer
the largest overall forecast error declines.
Sheen (2014) and Hoberg and Phillips (2010) use micro-data to show that mergers create
value by consolidating production to reduce costs and by facilitating product differentiation.
Alternatively, I use micro-data to argue that while mergers create value via synergies, those
synergies may have implementation costs. These implementation costs, such as employee
distraction or a loss of key talent, are consistent with the theory of the firm literature that
finds limits to integration due to human capital ownership rights (Grossman and Hart (1986),
Fulghieri and Hodrick (2006), and Fulghieri and Sevilir (2011)). Ouimet and Zarutskie (2016)
find that some mergers aim to acquire human capital. My results suggest that managers
should consider the integration consequences of such acquisitions.
My results further suggest that brokerage house mergers are unlikely to provide an exoge-
nous shock to industry competition, because workforce integration effects are not excluded.
Alternatively, these mergers do negatively impact stock market information, as in Kelly and
Ljungqvist (2012). One concern with their method is that several brokerage house mergers
happen during the tech-bubble burst, and it is possible that firms impacted by the merger
I discuss my data and the mergers in the next section. In Section II, I present results and
the accompanying empirical strategies. In Section III, I discuss some identification issues.
In Section IV, I conclude.
13Many of their mergers occurred in 1999 to 2001 before the DotCom Crash.
6
1 Data
Information on analysts and earnings estimates comes from the Thomson Reuters Insti-
tutional Brokers Estimate System (IBES) database spanning the period 1980 through 2013.
IBES provides individual analyst earnings forecasts, buy-sell-hold recommendations, and re-
ported earnings. Analyst earnings estimate level information comes from the detail history
U.S. earnings estimate file. Unique analyst identifiers allow the tracking of analyst careers
across brokerage houses.14 In addition to the detail file, the recommendation file within
IBES is used to help map brokerage house names to news stories.
I identify star analysts using Institutional Investor magazine rankings (e.g., Gleason and
Lee (2003), Clement and Tse (2005), Cohn and Juergens (2014)). Because the magazine
does not contain IBES identifiers, I hand-match stars to I/B/E/S using name, brokerage
house and time of employment. I label analysts as stars only if all three match.
I also use news releases regarding merger announcements from Factiva and corporate
websites for company history.
1.1 Mergers
My sample includes 34 brokerage house mergers, listed in Table 1.15,16 These mergers
impact 2,327 distinct analysts: 884 from target houses and 1,718 from acquiring houses.
Matching the target and acquirer to the IBES data is difficult because the brokerage house
14I drop all observations with analyst ID number (ANALYS) equal to 1 or 0, as these are placeholders.I drop observations covering several indices (DOWI, MID1, RUS2, S4, S5, SAP1, SAP6), as these analystsupdate their estimates at a very high rate, which makes their activity measures outliers. Previous studiesfocus on annual estimates. While the majority of estimates are annual, 40% are also quarterly, so I focus onall estimates, but also run analyses on the annual estimates alone for robustness. I control for fiscal periodwhere appropriate.
15Thirteen are taken directly from Hong and Kacperczyk (2010), which those authors isolate by mappingSDC mergers that belong to SIC code 6211 (Investment Commodity Firms, Dealers, and Exchanges) tothe IBES database. I supplement with four mergers from Kelly and Ljungqvist (2012), and I also collectan additional 17 mergers by finding brokerage house closures in the data and by using news articles andcompany histories to determine whether the cause of closure was in fact a merger.
16Two financial crisis mergers, Bear Stearns being acquired by JP Morgan and Merrill Lynch by Bankof America, were omitted because of the federal government’s involvement in encouraging and subsidizingthe mergers. Also, because of the financial crisis, attrition and forecast error are uniquely high. I am lesscomfortable with the external validity from these mergers. All results are robust to their inclusion andusually have larger partial effects.
7
names in the recommendation file are shortened nicknames that are often based on historical
names as opposed to current brokerage house names.17 Thus, matching requires a careful
reading of each brokerage house’s corporate history to determine whether the IBES nickname
corresponds to any preious historical names of the brokerage house. I require that at least
some target analysts who leave the target join the acquiring firm at the merger dates, and I
require that the target house no longer appears in the data after merger completion.
For merger announcement dates, I use the earliest date that a merger is mentioned
in the Factiva, news aggregation service. I use details from these press releases to classify
mergers into two categories: those that appear to highly value the target’s human capital and
those that do not. Mergers in which research expansion, increased services or the analysts
themselves are mentioned as a primary motivation for merging are labeled as Labor Valued
while mergers for which increasing assets under management or access to new clients is the
primary driver are labeled as Labor Not Valued.
My list of mergers are provided Table 1. They cover a relatively long period, with
the earliest merger occurring in 1988 and the most recent in 2012. There is considerable
clustering in the late 90s and early 2000s. Eight of the merger targets have fewer than seven
analysts, while four have over 50 analysts. Justifications for the mergers vary, including (but
not limited to) acquiring an underperforming house, deregulation, industry-wide conditions,
and strategic or geographic expansion. Within four months after the merger announcement,
most mergers are completed and no analysts remaining under the target house name.
For most tests, I treat each merger as an independent event, even if they overlap in
calendar time. I create periods in event time, extending 30-day periods in each direction
from the merger announcement.
17For instance, Wachovia is represented by the name WHEAT from one of its predecessors, J.C. Wheat& Co.
8
2 Results and Empirical Design
2.1 Overall Estimate Level Changes - Difference in Differences
To document the overall impact of mergers on human capital output, I run difference-
in-difference specifications on forecast error, the absolute deviation of the estimate and the
actual earnings scaled by the firms previous stock price. The first difference is between before
merger announcements and after merger completions, and the second for estimates produced
by merging houses versus non-merging houses. I present the evidence both graphically and
as regressions. The regression specification is below.
where e denotes event, h house, a analyst, t, event-time. β1 is the main variable of interest
20The analysts in the acquiring house have redundancy 1̄ by construction21When events overlap in calendar time, the same control group analysts will appear multiple times in
the data with different cov per and different period definitions all based on the specific acquiring firm andspecific announcement date.
22For robustness I run all the tests with the total number of companies covered as sum(COVERED), anddummies SOME COV = 1 if cov per > 0 and HALF COV = if cov per > 0.5 or 0 otherwise.
13
and ω and α denotes unobserved heterogeneity. β1 measures the within analyst changes
in pr(separation) comparing 90 days prior to the merger announcement to 90 days after
the announcement prior to merger completion. I define the dependent variable, Separation
Month as equal to 1 for all analyst-months prior to the month an analyst separates from
their current house and 0 otherwise.23 The variables Past and InMerger are subsumed by
the fixed effects.
Table 4 shows these results. Column 1 compares the change in attrition of analysts
who are subject to a merger announcement (either as a target or acquirer) versus analysts
who cover similar underlying firms but are not subject to the merger announcement. Ana-
lysts subject to a merger announcement experience increased attrition of almost 4% (4x the
unconditional average of 1%).
In Table 4 column 2, I split the analysts impacted by the merger into acquiring and target
house analysts using the dummy Post*InMerger*TargetMerger. Attrition increases 12% for
target analysts in relation to similar control analysts while there is no significant increase
in attrition for acquiring house analysts. I confirm this result in column 3 by running the
regression on only target house analysts and their comparable control group analysts.
In Table 4 column 4, I subdivide target house analysts to help determine where attrition
is the largest using the triple difference specification of:
where redundancy is defined in the previous subsection. Attrition is much higher for re-
dundant analysts in the target house v unique analysts in the target house. Attrition for
unique (0 firms covered by this analyst were covered by the acquiring house prior to the
merger announcement) analysts increases by over 6% as measured by β1. β2 captures the
difference for control group analysts’ attrition differences between unique versus redundant
23unlike before, I do not remove promotion-like separations. Previously my goals was to measure the stockof human capital whereas now I am interested in all forms of separation.
14
analysts and tells us redundant analysts unaffected by the merger are slightly more likely to
keep their job. This is likely due to the fact that redundant analysts are often high quality
analysts who cover popular stocks (stocks covered by more analysts). β3 is the variable of
interest which tells us attrition for a fully redundant analyst (all firms covered by this analyst
were covered by the acquiring house) increases by over 21% when compared to unique target
house analysts.24 The takeaway is that merging firms downsize by reducing head count and
that reduction comes predominately from analysts within the target house, not the acquiring
house, who are redundant rather than unique.
Next I incorporate the proxy for analyst quality with target house separations within
mergers. In Table 5 column 1, I show that when I look at only redundant analysts from
the target and acquiring firm, quality has no differential impact on attrition within mergers
ruling out the possibility that acquiring house analysts are systematically higher in quality
than target house analysts. In Column 2, I look only at target house analysts with no fixed
effects, which allows interpretation of each coefficient of the triple difference. Post captures
the non-merger trend in attrition which is positive by construction. InMerger captures
the pre-merger announcement differential in attrition between target house and non-merger
analysts. This coefficient is indistinguishable from 0 meaning mergers are not being initiated
based on underlying analyst quality. z(Quality) and Post*z(Quality) load negatively which
provides an out of sample test of the quality proxy, high quality analysts outside of mergers
are less likely to separate than low quality analysts. Post*InMerger captures the increase
in attrition after the merger announcement for low quality target analysts and is small but
significantly greater than zero. The triple difference coefficient, Post*InMerger*z(Quality),
is the variable of interest and captures the differential impact in attrition for high versus
low quality analysts within merger targets. This coefficient is economically and statistically
significant and is interpreted as a standard deviation increase in analyst quality makes a
target house analyst 4% more likely to separate from the firm in a given month post merger
announcement.
24Most analysts are not either fully redundant or completely unique. I show the results using a standardizedredundancy measure and find for a standard deviation change in redundancy attrition increases by almost7%.
15
In Table 5 Column 3, fixed effects subsume Post, InMerger, z(Quality), and InMerger*z(Quality).
The triple difference coefficient is additional evidence that after merger announcements high
quality analysts are 4% more likely to leave than low quality analysts. In columns 4 through
8 the data is split into quality quintiles with 1 being analysts of the lowest quality and 5 be-
ing the highest in order to confirm the result from column 2 and 3. Attrition differential for
high quality analysts is 19% while the differential is only 6% for the lowest quality quintile.
The result monotonically increases across quintiles. Finally in column 9, I show the result
from column 3 is 50% larger when the sample is restricted to only analysts with redundancy
of over 1/2.
Next, I show evidence that the separation is at least in part due to higher quality target
analysts choosing to leave their firms upon the merger announcement. First, in Table 6 I
show that contingent on separation, higher quality analysts are more likely to find another
analyst job than lower quality analysts implying that at least some of the target analysts
choose to leave because they have stronger outside options. In Table 7, I study how often
analysts drop coverage. For the set of target analysts that find a new analyst job, either
in an alternate brokerage, New Job, or in the merged entity, Kept Job what percentage of
firms an analyst covered before the merger do they continue to cover after it. This fraction
is very high. It is 91% for the median analyst who switches to a new house and 97% for
the median analyst who keeps his job. This suggests that the human capital of an analyst,
their expertise and connections, are not easily transferable and that dropping coverage is
consistent with abandoning human capital. So, not only are higher quality analysts more
likely to find a new job contingent on separation, they are likely to perform the same job as
before just at a new house.
Second, if redundant target analysts expect they are unlikely to retain their job in their
current role because of redundancy, rather than work harder to keep their job, upon the
merger announcement they may shift their effort elsewhere, for example, towards finding
a new job. I use InMerger redundancy, shown above to be strongly related attrition, as a
plausibly exogenous shock to the probability an analyst separates from the firm and test how
16
the probability of separation impacts productivity. I define productivity as the number of
reports an analyst releases in a month by counting unique ticker-date pairs for each analyst
month.25
Table 8 shows within analyst changes in productivity around merger announcements.
Column 1 contains a triple difference comparing unique to redundant analysts who are
targets of the same merger. I find that when comparing production changes within target
house analysts, redundant analysts reduce their productivity by 0.8 reports in comparison
to unique analysts. This is a drop of about 16% of the mean productivity for target house
analysts prior to the merger announcement (Mean productivity is 5.1 reports). Because
redundant analysts are the ones facing the highest pr(separation) this is consistent with
high quality target analysts who are likely to leave shifting their effort in anticipation of
finding a new job.
In column 2, instead of using the triple difference specification I use redundancy to
instrument for the probability an analyst separates. As shown in table 4, for target analysts,
redundancy is associated with a 20% in attrition (relevance) and is arguably not associated
with an analyst’s within merger change in productivity for any other reason other than the
increase in attrition (exclusion). This makes in-merger redundancy a plausible instrument
for identifying the impact on a change in the pr(separation) on an analyst’s productivity.
Confirming the finding in column 1, we observe a large drop off in productivity for analysts
facing a 100% increase in pr(separation).
In columns 3-5 we run the specification from column 1 but on smaller subsets. In column
3 I exclude analysts for which we cannot estimate quality due to missing data and in columns
4 and 5 I split that group by above median and below median quality. Consistent with the
flight of human capital results, high quality analysts drop productivity over 50% more than
low quality ones do. This result is consistent with analysts analysts looking for a new job
when they expect to be redeployed.
25Results are robust to alternate productivity specifications such as total firms covered or days with areport. All analyst-periods containing less than 30 days due to analyst separation are removed to not biasthe results with partial months.
17
2.4 Competition, Crashes or Merger Integration Issues
The findings in the previous sections can be driven by two alternate channels other than
the merger related channel that is the subject of this paper. First, Hong and Kacperczyk
(2010) argue that mergers reduce competition between analysts at different firms because
at least some analysts leave the industry. They argue that analysts face a trade-off between
winning an external tournament by being accurate and pleasing corporate clients by being
optimistic. Because analysts in the tournament are judged by relative accuracy and not
absolute accuracy they argue, when competition is reduced, all analysts covering the same
firm can afford to be more optimistic. This is an important alternate channel because I
am arguing that forecast error increases are due to unintended errors whereas this story is
arguing that the increases are doing to intended decisions analysts make. A second alternate
theory is because analysts are on average optimistic, unexpected market crashes can cause
temporary increases in earnings forecast error across all analysts.26
Going back to the original graphs, Figure 1 Panel A shows that even though all estimates
experience some increase in forecast error, the increase is significantly larger in estimates pro-
duced by brokerage houses involved in mergers and this difference is temporary, lasting only
two years. Because competition should impact analysts who cover the same firms equally, it
is hard to reconcile the differential impact seen by either explanation. The temporary nature
of the effect is not consistent the competition story because there is not off-setting new entry
of analysts. While there are possibilities that analysts are choosing short term-catering,
Clarke, Khorana, Pate, and Rau (2007) cast doubt on that channel by studying star analyst
transitions and finding that optimism has no impact on investment banking deal flow.
To be even more specific, I divide the estimates into redundant estimates and not-
redundant estimates, where redundant estimates are of firms which are covered by the target
and acquirer prior to the merger announcement and non-redundant are covered by just one
or the other are much more likely to suffer a competition shock.27 As shown in Figure 1
26See Brav and Lehavy (2003) and Bradshaw, Brown, and Huang (2013) for evidence of analyst optimism.27Recall that estimates are only included for firms that are covered by at least one of the two and that
attrition is highest amongst redundant target analysts.
18
Panel B and Panel D, estimates divided by the intensity of the competition shock have no
significant difference from each other in forecast error increase. This is true for forecast error
and positive bias.
In order to treat the recession channel, first I argue that the difference-in-differences
framework with period fixed effects should mitigate most concerns. But I also run my
results for mergers that happen during or right before a recession versus those that do not.
While the overall increase in forecast error does not increase for the non-recession mergers,
there is still a significant and differential impact of mergers even for mergers that are not
related to recessions.
Because this differential impact is not caused by a drop in competition, and not fully
explained by external market downturns, that leaves the mergers themselves as the primary
driver of the impairment.
2.5 Cross-Sectional Results based on Merger Subsets
I examine the two channels for impairment of human capital output during the merger
process in more depth by creating three independently created subsets outlined in Table 9.
First, in order to gain an ex-ante measure of human capital importance, I conduct a textual
analysis on the merger announcements, marking 14 mergers in which human capital, labor
or expanding services is not mentioned as a motivation for the merger and 20 mergers in
which these motivations are mentioned.
Second, I note the considerable variation in the cross section of forecast error increases of
mergers. In three mergers forecast error increases by over 1% (doubling), and in ten mergers
forecast error increases by 0.5% (increasing by 50%). On the other end of the distribution,
there is one merger in which forecast error is reduced by over 1% and seven which have at
least some improvement. In order to ex-post test that my channels actually correspond to
the same mergers that have an overall increase in forecast error, my second subset includes
only the top tercile of mergers with respect to overall forecast error increase.
Third, in 2003 the industry underwent a structural shift when the U.S. regulatory bodies
19
reached the Global Analyst Research Settlements (GARS), forcing walls to be constructed
between the research and investment banking divisions of the largest investment firms.28
This event creates a source of exogenous variation to a brokerage houses ability to cater to
corporate clients and I test my results for mergers before and after the structural shift.
In table A4, I show that the overall forecast error increases are double for the subset of
mergers in which labor is not valued. Meanwhile there is no significant difference between
pre and post global settlement forecast error increases.
In table 11, the temporary deterioration of quality, channel 1, is almost 70% larger for
mergers which ex-ante labor is not valued and ex-post suffer the largest drops in forecast
error. The partial effects before and after the global settlement remain for the most part
unchanged.
In table 12, high quality target analyst attrition, channel 2, increases from 6 basis points
to 10 basis points for mergers which ex-ante labor is not valued and doubles for mergers
which ex-post suffer the largest drops in forecast error. The one surprising result is that
the affect disappears post global settlement but this may warrant extra attention due to the
small sample size.
2.6 Mergers Uncontrolled Impact on Human Capital Output
To confirm the overall difference-in-differences results are driven by changes in merging
houses and not changes in the control group, I run uncontrolled regressions for robustness.
The sample for this analysis includes the set of earnings estimates published by the acquirer
and the target prior to the merger announcement (the merging houses) as compared to the
set of earnings estimates produced by the merged entity after the merger completion. I
restrict the sample to estimates for firms that are covered both before and after the merger
to mitigate any coverage decision selection concerns. Table A2 presents the merger level
results. Overall, across the merged firms, I find that average forecast error increases from
1.03% to 1.41%. This increase can be seen visually in Figure 1, Panel A, represented by the
where e denotes event, t ticker, f fiscal period, p is pre (target or acquirer) or post
(merged entity), and β1 is the variable of interest.
In Table A3, Post captures the average change in forecast error in moving from two
separate houses to one combined house. Column 1 shows that the forecast error for estimates
produced 90 days before the merger announcement are 24 basis points lower than estimates
produced 90 days after the merger completion. Column 2 extends the windows to a year on
both sides and the forecast error increase becomes 36 basis points. In Column 3, I compare
estimates for the base quarter, the one before the merger announcement, to estimates for
the three quarters before that quarter (pre-trend) and to the estimates a year after, two
years after, and three years after the merger completion. I see no significant difference
before the merger announcement (i.e., YN1 is not different from zero), while Y1 and Y2
are both significantly greater than 0 (.30 and .45, respectively). Confirming the temporary
nature of the result, the Y3 coefficient is not significantly different from zero. Some might
21
argue that the Y3 coefficient is .20, so not technically zero (even though it is statistically
indistinguishable from zero), so in Column 4, I run the same regression broken down into
quarters, which show that the effect is in fact temporary, as the large partial effect is driven
only by the first quarter of the third year. I add Event*Fiscal Period (FPI) fixed effect
transformations to control for unobserved permanent heterogeneity in the events, whether
the estimates are annual or quarterly, and I cluster standard errors at the event level.
3 Further Discussion of Identification Issues
The main identifying assumption is that the factor that drives the mergers (and their
announcements) are not correlated with the changes to analyst attrition, forecast error or
productivity. Reverse causality is unlikely to be an issue because analysts’ career concerns
or future output changes are unlikely to drive the mergers. Additionally, by using triple
difference specifications I compare before and after changes of analysts within the same
brokerage house who are affected by the merger.
In the appendix, I also run pre-event falsification tests using a false merger date 2 months
prior to the merger announcement and show that there are no differential trends in analyst
behaviors.
Omitted variables may influence both the outcome and the explanatory variables. Results
in this paper are run including analyst#event#house and event-time (or sometimes period,
defined as event#event-time) fixed effects. Using within analyst variation (especially over
the short time window around the merger announcement) controls for variables such as
analyst ability. It also mitigates concerns over selection bias with regard to who gets fired,
resigns or stays at the firm. 29
The period fixed effects mitigate time trend concerns, the largest being quarterly cycli-
cality in earnings season and reports as well as major market crashes. Note there are several
overlapping events, thus these are NOT month fixed effects but significantly more conserva-
29All the results hold for analyst#event fixed effects, but I also interact brokerage house to only capturevariation from analysts within the target house who have not switched houses yet.
22
tive 30 day period fixed effects that are independently defined for each merger event.
The results are clustered at the event level. Because explanatory variables are constant
across periods within an event for a given analyst, clustering time periods is essential. I
cluster my main results at the event level to be conservative. I cluster my falsification tests
at the event-analyst levels to work against falsification.
4 Conclusion: Beyond Analysts
This paper provides evidence on how mergers impact the acquisition, performance and
retention of human capital by analyzing sell-side analyst output quality and career outcomes
around brokerage house mergers. I find evidence suggesting that analyst output quality is
impaired. This impairment is driven by a failure to retain high-quality analysts from the
target house and by the output quality deterioration of retained analysts from the acquir-
ing house. These effects are especially large in merger subsets for which human capital
acquisition does not appear to be of first-order importance.
These effects are unlikely unique to brokerage houses. Because analyst output is observ-
able to the labor markets and managers, one might expect it would be easier to measure
quality resulting in more complete contracts and thus this is a lower bound for individual
employees of acquiring firms. I observe the opposite because of the mobility and the lack of
contract completeness common to high human capital employees.
Finally, note that I can say little about overall merger efficiency. Sufficient value may
be transferred from labor to shareholders through cost savings, or the brokerage division
may be a small portion of a larger firm and merger gains may be earned elsewhere. How-
ever, given that sell-side research is considered a public good due to its positive impact on
informational efficiency (Kelly and Ljungqvist (2012)), impairment of research quality can
negatively impact investors and firms. The FTC and DOJ should more carefully review the
consumer impact of mergers that occur between firms that operate in industries in which
human capital is crucial, but the merging firms do not appear to value human capital.
23
REFERENCES
Abowd, John M., John Haltiwanger, Ron Jarmin, Julia Lane, Paul Lengermann, Kristin Mc-
Cue, Kevin McKinney, and Kristin Sandusky, 2005, The Relation among Human Capital,
Productivity, and Market Value: Building Up from Micro Evidence, in Measuring Capital
in the New Economy , NBER Chapters, 153–204 (National Bureau of Economic Research,
Inc).
Acemoglu, Daron, 2002, Technical change, inequality, and the labor market, Journal of
Economic Literature 40, 7–72.
Allen, Arthur, Bill Francis, Qiang Wu, and Yijiang Zhao, 2013, Monitoring or Pressure: The
Impact of Analyst Coverage on Corporate Tax Avoidance .
Andrade, Gregor, Mark Mitchell, and Erik Stafford, 2001, New evidence and perspectives
on mergers, Journal of Economic Perspectives 15, 103–120.
Andreoni, James, William Harbaugh, and Lise Vesterlund, 2003, The Carrot or the Stick:
Rewards, Punishments, and Cooperation, The American Economic Review 93, 893–902.
Autor, David H., Frank Levy, and Richard J. Murnane, 2003, The skill content of recent
technological change: An empirical exploration, The Quarterly Journal of Economics 118,
1279–1333.
Betton, Sandra, B. Espen Eckbo, and Karin S. Thorburn, 2008, Chapter 15 - corporate
takeovers*, in B. Espen Eckbo, ed., Handbook of Empirical Corporate Finance, Handbooks
in Finance, 291 – 429 (Elsevier, San Diego).
Bradshaw, Mark T., Lawrence D. Brown, and Kelly Huang, 2013, Do sell-side analysts
Observations 563,363 3,212,344 6,570,582 6,570,582Adjusted R2 0.069 0.059 0.050 0.050Event Period FPI FE YES YES YES YES
pval in parentheses, StErr Clustered at Event Level*** p<0.01, ** p<0.05, * p<0.1
Difference-in-differences are reported using merger announcement and completions as treatment events from1988 to 2012. I compare the difference in annual and quarterly estimate forecast error before the mergerannouncement from the target and acquiring house and the merged entity after merger completion, todifferences in non-merger estimates of the same firms over the same periods. Results are presented for 90days, 1 year, 3 years, and 12 quarters. Forecast Error is defined as the absolute deviation from actualearnings scaled by current stock price. The binary independent variable InMerger is equal to 1 for estimatesof the merging houses and 0 otherwise. Post*InMerger is the interaction of InMerger and an indicator equalto 1 for all estimates after merger completion and 0 otherwise. z(Timeliness) is defined as the number ofdays before the earnings announcement the estimate is released. All specifications include event, Periodand FPI fixed effects to control for unobserved heterogeneity. Parentheses contain p-values computed fromstandard errors clustered at the event level. 29
Table 3: Analyst Changes in Forecast Error
DepVar: (1) (2) (3) (4) (5)Analyst Forecast Error Merger Acq Target Acq By Period Acq Post
Post * InMerger 0.06* 0.11** -0.04(0.08) (0.01) (0.42)
pval in parentheses, StErr Clustered at Event Level*** p<0.01, ** p<0.05, * p<0.1
Difference-in-difference estimates are reported using merger announcements as treatment events from 1988to 2012. The sample compares analyst quarterly and annual earnings estimates from four months before themerger announcement to up to 3 years after merger completion for analysts who retain their job post merger.The dependent variable Forecast Error, is measured as the monthly average absolute difference between ananalysts estimates and the actual earnings per share scaled by the current stock price. The control group isrestricted to include analysts with at least a 50% overlap with the target or the acquiring house before themerger. Column 1 compares all merger analysts to non-merger analysts. Columns (2), (4) and (5) excludetarget house analysts while Column (3) excludes acquiring house analysts. Column 4 compares the originalpost period, four months after merger completion, to the next four months. Column 5 compares every fourmonth period to the original pre-merger period, four months before merger announcement. All specificationsinclude Event-Time and Event×Analyst×House fixed effects transformations to control for unobserved het-erogeneity and to mitigate selection bias concerns. Parentheses contain p-values computed from standarderrors clustered at the event level. Specification 1: ForecastErrore,h,a,t = β1postt∗InMergere,h+ωt+αe,a,h
30
Table 4: Attrition around merger announcements driven by Target Redundancy
DepVar: (1) (2) (3) (4)Separation Month Targ & Acq Targ v Acq Just Targ Redundancy
Post * InMerger 0.04*** 0.01 0.12*** 0.06*(0.00) (0.40) (0.00) (0.05)
pval in parentheses, StErr Clustered at Event Level*** p<0.01, ** p<0.05, * p<0.1
Linear probability model estimates are reported for difference-in-difference and triple-difference specifica-tions using merger announcements as treatment events from 1980 to 2012. The binary dependent variableSeparation is equal to 1 in months in which analysts separate from their brokerage house and 0 otherwise.The variable Redundancy is the fraction of coverage overlap that an analyst has with the acquiring housebefore the announcement. The control group is restricted to analysts with at least a 50% coverage overlapwith the target house. Specifications in Columns 1 and 2 include both acquirer and target house analystsas treated observations, while specifications in Columns 3 and 4 include only target house analysts. Allspecifications include Event-Time and Event×Analyst×House fixed effects transformations to control forunobserved heterogeneity and to mitigate selection bias concerns. Parentheses contain p-values computedfrom standard errors clustered at the event level.Specification 4: Separatione,h,a,t = β1postt ∗ InMergere,h + β2postt ∗ Redundancye,a,h + β3postt ∗InMergere,h ∗ Redundancye,a,h + αe,a,h + ωt where e denotes the event, h the house, a the analyst, trepresents the event time. β3 is the main variable of interest, and ω and α denotes unobserved heterogeneity.
31
Tab
le5:
Att
riti
onar
ound
mer
ger
annou
nce
men
ts-
Qual
ity
Dep
Var
:C
omb
Tar
gT
arg
Qual
ity
Quin
tile
Tar
g
Sep
arat
ion
Mon
thR
edN
oF
EF
E1
(Low
)2
34
5(H
igh)
Red
Pos
t*
InM
erge
r0.
10**
*0.
09**
*0.
12**
*0.
06*
0.09
***
0.13
***
0.16
***
0.19
***
0.20
***
(0.0
0)
(0.0
0)
(0.0
0)
(0.0
7)
(0.0
1)
(0.0
0)
(0.0
0)
(0.0
0)
(0.0
0)
Pos
t*
z(Q
ual
)-0
.03*
**-0
.01*
-0.0
3***
-0.0
3***
(0.0
0)
(0.0
8)
(0.0
0)
(0.0
0)
Pos
t*
InM
erge
r*
z(Q
ual
)-0
.00
0.04
**0.
04**
*0.
06**
(0.8
3)
(0.0
2)
(0.0
0)
(0.0
4)
Pos
t0.
04**
*(0
.00)
InM
erge
r0.
00(0
.96)
z(Q
ual
)-0
.02*
**(0
.00)
InM
erge
r*
z(Q
ual
)-0
.00
(0.8
3)
Con
stan
t0.
04**
*(0
.00)
Obse
rvat
ions
36,6
7943
,894
43,3
959,
003
8,74
18,
405
8,37
98,
867
17,9
27A
dju
sted
R2
0.30
40.
022
0.16
10.
166
0.15
50.
152
0.14
70.
160
0.15
9pva
lin
par
enth
eses
,StE
rrC
lust
ered
atE
vent
Lev
el***
p<
0.0
1,
**
p<
0.0
5,
*p<
0.1
Lin
ear
pro
bab
ilit
ym
od
eles
tim
ates
are
rep
orte
dfo
rtr
iple
-diff
eren
cesp
ecifi
cati
on
su
sin
gm
erger
an
nou
nce
men
tsas
trea
tmen
tev
ents
from
1988
to2012.
Th
ista
ble
splits
the
resu
ltfr
omT
able
4byz(Quality),
aco
nti
nu
ou
sst
an
dard
ized
qu
ali
tym
easu
regen
erate
din
Ap
pen
dix
A1.
Th
eb
inary
dep
end
ent
vari
ab
leSeparation
iseq
ual
to1
inm
onth
sin
wh
ich
anal
yst
sse
para
tefr
om
thei
rb
roke
rage
hou
sean
d0
oth
erw
ise.
Th
eco
ntr
ol
gro
up
isre
stri
cted
toin
clu
de
anal
yst
sw
ith
atle
ast
a50
%ov
erla
pw
ith
the
targ
eth
ou
se.
Colu
mn
1co
mb
ines
Acq
uir
ers
an
dT
arg
ets
as
asi
ngle
hou
sean
din
clu
des
on
lyan
aly
sts
wit
hre
du
nd
ancy
mea
sure
sov
er1/
2.C
olu
mn
2co
nta
ins
no
fixed
effec
ttr
an
sform
ati
on
wh
ile
inC
olu
mn
s3-9
,Post
,InMerger,
z(Quality),
an
dInMerger*z(Quality)
are
sub
sum
edby
the
even
t-ti
me
and
anal
yst
fixed
effec
ttr
an
sform
ati
on
s.S
pec
ifica
tion
s4
thro
ugh
8are
quality
qu
inti
les
wit
h1
bei
ng
an
aly
sts
of
the
low
est
qu
alit
yan
d5
bei
ng
anal
yst
sof
the
hig
hes
tqu
alit
y.T
he
fin
al
colu
mn
issi
mil
ar
toC
olu
mn
3b
ut
itin
clu
des
on
lyan
aly
sts
that
hav
ea
red
un
dan
cyof
at
least
1/2.
Par
enth
eses
conta
inp
-val
ues
com
pu
ted
from
stan
dard
erro
rscl
ust
ered
at
the
even
tle
vel.
32
Table 6: Pr(Find Analyst Job)|Separation for Target Analysts
Observations 468 468 437Adjusted R-squared 0.020 0.160Event FE No YES YESNumber of event 23
Robust pval in parentheses*** p<0.01, ** p<0.05, * p<0.1
Linear probability model estimates and conditional logit odds ratios are reported using merger announce-ments from 1980 to 2012 as treatment events. Table 6 studies only target house analysts who separatearound the merger announcement. The binary dependent variable Find New Job is equal to 1 if the analystfinds another analyst job after separation and 0 otherwise. Specifications in Columns 2 and 3 include Event-Time and Event×Analyst×House fixed effects transformations to control for unobserved heterogeneity andto mitigate selection bias concerns. Parentheses contain p-values computed from standard errors clusteredat the event level.
33
Table 7: Redeployment Post Job Transfer
New Job Kept Job
N 266 N 304
Mean 0.83 Mean 0.88
Level Quantile Level Quantile
100% Max 1 100% Max 1
99% 1 99% 1
95% 1 95% 1
90% 1 90% 1
75% Q3 1 75% Q3 1
50% Median 0.91 50% Median 0.97
25% Q1 0.73 25% Q1 0.82
10% 0.50 10% 0.64
5% 0.33 5% 0.50
1% 0.20 1% 0.17
0% Min 0.07 0% Min 0.07
Table 7 shows summary statistics for human capital abandonment. For target analysts that remain in thedatabase, either at a New Job or within the new merged entity, kept job, I calculate the fraction of firms theanalyst still covers that they covered previously.
34
Table 8: Target Analyst Productivity around Merger Announcements
DepVar: (1) (2) (3) (4) (5)# Reports Triple Diff Inst Qual 6=. Low Qual High Qual
Observations 37,790 40,292 31,108 15,201 15,904Adjusted R2 0.508 0.443 0.507 0.478 0.514Job Period FE YES YES YES YES YES
pval in parentheses, StErr Clustered at Event Level*** p<0.01, ** p<0.05, * p<0.1
Triple difference and instrumental-variable estimates are reported using merger announcements as treatmentevents from 1980 to 2012. The dependent variable, Productivity, is measured as the number of reportsan analyst produces in a given 30-0day period. The sample compares three months before the mergerannouncement to the three months after the merger announcement but before the merger closure excludingall periods less than 30 days due to analyst separation. Redundancy is the fraction of coverage overlap thatan analyst has with the acquiring house before announcement. The control group is restricted to includeanalysts with at least a 50% coverage overlap with the target house. Column 1 reports triple-differenceestimates comparing unique to redundant analysts within the target house. In Column 2, redundancy isused as an instrument for pr(separation). The IV specification allows inclusion of time-varying controls,so I add a control for Popularity. Popularity is the average number of other analysts who also coverthe stocks the analyst covers. Column 3 includes only analysts for which I can estimate quality. InColumns 4 and 5, I divide this group into high and low-quality. All specifications include Event-Time andEvent×Analyst×House fixed effects transformations to control for unobserved heterogeneity and mitigateselection bias concerns. Parentheses contain p-values computed from standard errors clustered at the eventlevel.
Specification 1: Productivitye,h,a,t = β1postt ∗ InMergere,h + β2postt ∗ Redundancye,a,h + β3postt ∗InMergere,h ∗Redundancye,a,h + ωt + αe,a,h where e denotes event, h house, a analyst, t period. β3 is themain variable of interest, and ω and α denotes unobserved heterogeneity.
35
Tab
le9:
Mer
gers
Subse
ts
Mer
ger
An
nC
omp
Targ
etA
cqu
irer
Fore
cast
Err
or
Lab
or
Post
Glo
bal
#D
ate
Dat
eD
iffV
alu
edS
ettl
emen
t
275/
079/
07C
och
ran
,C
aron
iaS
ecu
riti
esF
ox-P
itt
Kel
ton
2.87%
Yes
Yes
302/
086/
08F
erri
sB
aker
Watt
sR
BC
Wea
lth
Man
agem
ent
1.86%
No
Yes
2911
/07
1/08
Op
pen
hei
mer
CIB
C1.30%
No
Yes
285/
079/
07A
.G.
Ed
war
ds
an
dS
on
sW
ach
ovia
0.94%
No
Yes
261/
074/
07R
yan
Bec
k&
Co
Sti
fel
Fin
an
cial
0.90%
Yes
Yes
18/
887/
89B
utc
her
&C
o.,
Inc
Wh
eat
Fir
stS
ecu
riti
es0.74%
Yes
No
137/
0011
/00
Pai
ne
Web
ber
UB
S0.66%
Yes
No
159/
001/
01C
has
eM
anh
atta
n/
Ham
bre
cht
JP
Morg
an
0.61%
No
No
2510
/06
12/0
6M
ille
rJoh
nso
nS
teic
hen
Kin
nard
Sti
fel
Fin
an
cial
Corp
0.58%
Yes
Yes
111/
006/
00S
chro
der
sS
olo
mon
Sm
ith
Barn
ey0.52%
Yes
No
148/
0010
/00
Don
ald
son
,L
ufk
in&
Jen
rett
eC
redit
Su
isse
0.50%
No
No
82/
984/
98W
esse
lsA
rnol
d&
Hen
der
son
Dain
Rau
sch
er0.48%
Yes
No
124/
006/
00JC
Bra
dfo
rd&
Co.
Pain
eWeb
ber
Gro
up
0.45%
Yes
No
226/
0511
/05
Leg
gM
ason
Cit
igro
up
0.45%
No
Yes
169/
0011
/01
Dai
nR
ausc
her
Rb
cC
ap
ital
Mark
ets
0.34%
No
No
49/
9711
/97
Sal
omon
Bro
ther
sS
mit
hB
arn
ey0.34%
No
No
59/
973/
98Jen
sen
Sec
uri
ties
Co.
DA
Dav
idso
n0.34%
Yes
No
174/
0110
/01
Wac
hov
iaS
ecu
riti
esF
irst
Un
ion
0.32%
Yes
No
103/
9910
/99
EV
ER
EN
Cap
ital
Corp
Fir
stU
nio
nC
orp
0.29%
No
No
2410
/06
12/0
6P
etri
eP
arkm
an
&C
o.
Mer
rill
Lyn
ch&
Co
0.27%
Yes
Yes
210
/94
12/9
4K
idder
Pea
bod
y&
Co
Pain
eWeb
ber
0.23%
Yes
No
712
/97
2/98
Pri
nci
pal
Fin
an
cial
Sec
uri
ties
EV
ER
EN
Cap
ital
0.23%
Yes
No
3312
/11
3/12
Mor
gan
Kee
gan
&C
om
pany
Ray
mon
d0.19%
Yes
Yes
910
/98
6/99
Ale
xB
row
n-
Ban
kers
Tru
stD
euts
che
Ban
k0.18%
Yes
No
612
/97
6/98
Un
ion
Ban
kO
fS
wit
zerl
an
dS
wis
sB
an
kC
orp
ora
tion
0.15%
No
No
208/
0410
/04
Sch
wab
Sou
nd
vie
wC
ap
ital
Ub
s0.04%
No
Yes
239/
051/
06A
dam
sH
arkn
ess
Can
acc
ord
Cap
ital
Corp
ora
tion
0.02%
Yes
Yes
32/
974/
97D
ean
Wit
ter
Morg
an
Sta
nel
y-0.04%
No
No
212/
056/
05P
arke
r/
Hu
nte
rIn
cJan
ney
Montg
om
ery
Sco
tt-0.07%
Yes
Yes
3411
/12
2/13
Kee
feB
run
net
teW
oods
Sti
fel
Fin
an
cial
Corp
-0.10%
Yes
Yes
188/
0110
/01
Tu
cker
Anth
ony
Su
tro
Cap
ital
Rb
cC
ap
ital
Mark
ets
-0.32%
No
No
199/
012/
02Jos
ephth
alL
yon
&R
oss
Fah
nes
tock
-0.41%
Yes
No
324/
107/
10T
hom
asW
eise
lP
art
ner
sS
tife
lF
inan
cial
Corp
-0.48%
Yes
Yes
318/
0911
/09
Fox
-Pit
tK
elto
nm
acq
uari
e-1.70%
Yes
Yes
Th
ista
ble
ran
ks
all
34m
erge
rsby
dec
reas
ein
fore
cast
erro
r.M
erger
sfo
rw
hic
hm
erger
an
nou
nce
men
tsd
on
ot
men
tion
hu
man
cap
ital
or
exp
an
din
gse
rvic
esar
em
arke
das
No
forLaborValued
and
Yes
oth
erw
ise.
Th
eto
pte
rcil
eof
mer
ger
sd
ivid
edby
fore
cast
erro
rin
crea
se,
are
mark
edw
ith
ab
lack
lin
e.T
he
fin
al
colu
mn
mar
ks
mer
gers
that
occ
urr
edaf
ter
the
Glo
bal
An
aly
stR
esea
rch
Set
tlem
ents
.
36
Table 10: Estimate Level Operation Changes - Difference-in-Differences - by Merger Type
(1) (2) (3) (4) (5) (6)DepVar: Forecast Error LNV PGS PGS no 2009 LNV PGS PGS no 2009
Post * InMerger 0.26*** 0.09 0.14**(0.00) (0.20) (0.02)
pval in parentheses, StErr Clustered at Event Level
*** p<0.01, ** p<0.05, * p<0.1
Difference-in-difference estimates are reported using merger announcement and completions as treatmentevents from 1988 to 2012. I compare the difference in annual and quarterly estimate forecast error beforemerger announcement from the target and acquiring house to estimates of the merged entity after mergercompletion, to differences in non-merger estimates of the same firms over the same periods. Results arepresented for 90 days and three years. Columns (1) and (4) include only mergers in which labor is not highlyvalued while the remaining columns include only mergers after the GARS. In Columns (3) and (6), I removeall observations from 2009. Forecast Error is defined as absolute deviation from actual earnings scaled bycurrent stock price. The binary independent variable InMerger is equal to 1 for all estimates of the mergedentity, target, or acquirer, and 0 otherwise. Post*InMerger is the interaction of InMerger and an indicatorequal to 1 for all estimates after merger completion and 0 otherwise. z(Timeliness) is defined as the numberof days before the earnings announcement the estimate is released. All specifications include event, Periodand FPI fixed effects to control for unobserved heterogeneity. Parentheses contain p-values computed fromstandard errors clustered at the event level.
37
Table 11: Analyst Changes in Forecast Error - By Merger Type
pval in parentheses, StErr Clustered at Event Level*** p<0.01, ** p<0.05, * p<0.1
Difference-in-differences estimates are reported using merger announcements as treatment events from 1988to 2012. The sample compares analyst quarterly earnings estimates from 4 months and 12 months prior tothe merger announcement to 4 months and 12 months after merger completion for analysts who retain theirjob post merger. The dependent variable Forecast Error, is measured as the monthly absolute differencebetween an analysts estimates and the actual earnings per share scaled by the current stock price. Thecontrol group is restricted to include analyst’s with at least a 50% overlap with the target or the acquiringhouse beore the merger. Columns 1 and 2 are repeated from Table 3. Columns 3-8 restrict the sampleto merger subsets defined in Table A2, between mergers in which the press release commented on labornot being valued, the tercile of mergers with the largest increase in forecast error, and mergers after theglobal settlement. All specifications include Event-Time and Event×Analyst fixed effects transformations tocontrol for unobserved heterogeneity and to mitigate selection bias concerns. Parentheses contain p-valuescomputed from standard errors clustered at the event level.
38
Table 12: Attrition around Merger Announcements - Split by Merger Type
(1) (2) (3) (4) (5) (6) (7)DepVar: No Labor Valued? Forecast Error Global Settlement
pval in parentheses, StErr Clustered at Event Level*** p<0.01, ** p<0.05, * p<0.1
Linear probability model estimates are reported for triple-difference specifications using merger announce-ments as treatment events from 1988 to 2012. Table 12 divides the result from Table 5 by merger type. Thebinary dependent variable Separation is equal to 1 in months in which analysts separate from their brokeragehouse and 0 otherwise. The control group is restricted to include analysts with at least a 50% overlap withthe target house. Columns 2 and 3 split the sample by mergers in which the press release commented on laborbeing valued versus mergers focused on acquiring only physical assets. Columns 4 and 5 compare the tercileof mergers with the largest increase versus the largest decrease in forecast error. Columns 5 and 6 dividemergers between before and after the global analyst settlement. All specifications include Event-Time andEvent×Analyst×House fixed effects transformations to control for unobserved heterogeneity and to mitigateselection bias concerns. Parentheses contain p-values computed from standard errors clustered at the eventlevel.
39
APPENDIX
40
Table A1: Baseline Pr(Separation) outside of merger announcements
Observations 306,627 305,122 304,732 316,341 312,447 75,889R2 0.014 0.131 0.201FE Period House#Period Analys House#Period None Period House#PeriodClustering House period House period House year None Period House#Period
Linear probability model estimates and logit odds ratios are reported for analysts not impacted by mergers.The binary dependent variable Separation - No Promotion is equal to 1 in months in which analysts separatefrom their brokerage house and do not join a more prestigious house and 0 otherwise. House prestige is definedby the houses total number of analysts. Columns 1-3 are LPMs while Columns 4-6 are logits with odd ratiospresented. Spec 4 (logit no FE) is used to to generate a quality proxy.
pval in parentheses, StErr Clustered at Event Level*** p<0.01, ** p<0.05, * p<0.1
OLS estimates of single difference regressions use merger completions as treatment events from 1988-2012.The sample compares annual and quarterly estimates before merger announcement from the target andacquiring house to those of the merged entity after merger completion. Forecast Error is the absolutedeviation scaled by current stock price. Post is equal to 1 for all estimates of the merged entity and 0 forestimates of the target and acquiring house before the merger announcement. z(Timeliness), the number ofdays before the actual announcement the estimate is made, helps control for patterns in earnings estimates.Specifications differ in the length of the sample size, with (1) being 90 days before and after, (2) being oneyear before and after, and (3) and (4) being one year before and three years after.Specification 2: ForecastErrore,t,f,p = β1Postp + β2Timelinesse,t,f,p + αe,t,f
pval in parentheses, StErr Clustered at Event Level*** p<0.01, ** p<0.05, * p<0.1
Difference estimates are reported using merger announcements as treatment events from 1988 to 2012. Eachcolumn compares the full sample (columns presented in earlier tables) to estimates from two restrictedsamples: 1) mergers in which labor does not appear to be highly valued in the merger announcement pressrelease and 2) mergers after the Global Analyst Settlement. Regressions (1)-(3) include only 90 days beforeand after while columns (4)-(6) include one year before and after All specifications include Event#FPIfixed effects transformations to control for unobserved heterogeneity and to mitigate selection bias concerns.Parentheses contain p-values computed from standard errors clustered at the event level..
44
Fig
ure
A1:
Lin
ech
arts
show
fore
cast
erro
rov
erev
ent
tim
e(9
0day
per
iods)
.P
anel
(a)
splits
the
esti
mat
esb
etw
een
thos
ege
ner
ated
wit
hin
anN
BE
Rre
cess
ion,InDow
nturn
,ve
rsus
thos
eth
atw
ere
not
.P
anel
(c)
furt
her
sub
div
ides
the
esti
mat
esb
etw
eenInMerger
and
not
.P
anel
s(a
)&
(b)
show
sim
ple
aver
ages
.P
anel
s(c
)&
(d)
plo
tth
eco
effici
ents
from
the
two
lines
in(b
)(w
ith
90%
confiden
cein
terv
als)
from
diff
eren
ce-i
n-d
iffer
ence
sre
gres
sion
sw
ith
Eve
nt,
Per
iod,
and
Fis
cal
Per
iod
fixed
effec
ts.
(a)
.81
1.2
1.4
1.6
mean_acc_id
−5
05
1015
Eve
ntT
ime
("Q
uart
ers"
)
Not
Dow
ntur
n
Dow
ntur
n
(b)
.51
1.52
Average Forecast Error
−5
05
1015
Eve
ntT
ime
("Q
uart
ers"
)
InM
erge
r In
Rec
Not
Mer
ger
InR
ec
InM
erge
r N
oRec
Non
Mer
ger
NoR
ec
(c)
−.10.1.2.3
Coefficient Estimate (Forecast Error)
QN
3Q
N2
QN
1Q
1Q
2Q
3Q
4Q
5Q
6Q
7Q
8Q
9Q
10Q
11Q
12
Qua
rter
Not
in R
eces
sion
(d)
−.10.1.2.3
Coefficient Estimate (Forecast Error)
QN
3Q
N2
QN
1Q
1Q
2Q
3Q
4Q
5Q
6Q
7Q
8Q
9Q
10Q
11Q
12
Qua
rter
In R
eces
sion
45
Table A5: Estimate Level Operation Changes - Difference-in-Differences - By in Recession
(1) (2) (3) (4)DepVar: Forecast Error In Down turn Not In Down turn
Robust pval in parentheses clustered at the Event Level*** p<0.01, ** p<0.05, * p<0.1
Difference-in-Difference estimates are reported using merger announcements as treatment events from 1988-2012. The sample compares analyst quarterly earnings estimates from before the merger announcementto after merger completion split by whether the merger occurred just prior or within a recession. Allspecifications include Event, Period and Fiscal Period fixed effects transformations to control for unobservedheterogeneity and mitigate selection bias concerns. Parentheses contain p-values computed from standarderrors clustered at the event level.
46
Table A6: Redundant v Non-Redundant - Difference-in-Differences
Observations 3,212,344 6,570,582 6,163,057 407,525Adjusted R2 0.059 0.050 0.050 0.054event period FPI FE YES YES YES YES
pval in parentheses, StErr Clustered at Event Level
*** p<0.01, ** p<0.05, * p<0.1
Difference-in-Differences are reported using merger announcement and completions as treatment events from1988-2012. I compare the difference in annual & quarterly estimate forecast error prior to merger announce-ment from estimates for firms covered by both the target and acquirer prior to the merger and estimatescovered by one or the other. Results are presented for one year, three years. Forecast Error is definedas absolute deviation from actual earnings scaled by current stock price. The binary independent variableDupCov is equal to 1 for all estimates of firms covered by both the target and acquirer prior to the mergerannouncement and 0 otherwise. Post*Dupcov is the interaction of InMerger and an indicator equal to 1for all estimates after merger completion and 0 otherwise. z(Timeliness) is defined as # of days prior tothe earnings announcement the estimate is released. All specifications include Event, Period & FPI fixedeffects to control for unobserved heterogeneity. Parentheses contain p-values computed from standard errorsclustered at the event level.