Stock Recommendations for Politically Connected Firms
Elio Alfonso
School of Accounting
Florida International University
11200 SW 8th Street
Miami, Florida 33199
[email protected]
April 21, 2015
Keywords:
Political connections, analysts’ forecast accuracy, stock recommendations,
and recommendation profitability.
Data Availability: Data are available from the sources identified in the paper.
Acknowledgements: I greatly appreciate the helpful comments of an anonymous reviewer, C.S.
Agnes Cheng (dissertation chair), Dana Hollie, Wei-Ling Song, Robert Kirby Goidel, Steven
Lin, Ken Reichelt, Christine Cheng, Andrey Simonov, Robert Hogan, faculty and seminar
participants at Louisiana State University, Florida International University, University of North
Florida, The KPMG Foundation, and The PhD Project faculty and doctoral students.
Stock Recommendations for Politically Connected Firms
Abstract:
Social network connections of corporations can significantly affect operating
performance and firm valuation (Faccio 2006, Horton et al. 2012). Political
connections are one form of social networking which often manifest into improved
firm profitability as a result of political favors granted by politicians. However,
political connectedness can also lead to increased firm risk due to the unpredictability
and uncertain timing of politicians’ actions. Recent research finds that politically
connected firms have higher future earnings volatility, and that analysts have greater
difficulty forecasting the earnings of politically connected firms, than non-connected
firms (Chen et al. 2010, Cooper et al. 2010). I examine how political connections affect
analysts’ stock recommendations using a unique dataset of political contributions in
the U.S. over the period 1993 - 2012. I show that analysts’ recommendations are less
profitable for politically connected firms than for non-connected firms. I also find that
analysts are less effective in translating earnings forecasts into profitable
recommendations for firms with political connections than for firms without
connections. Overall, the findings suggest that analysts do not fully understand the
economic complexities of political connections and do not efficiently process
information on political connections into their stock recommendations.
1
I.
INTRODUCTION
Corporate executives have a vast and multidimensional array of social network
connections that they obtain through different influential sources. These social ties can arise from
university alumni networks, colleagues from previous employment, connections to politicians,
and activities such as charitable organizations, and club memberships. Social network
connections can affect the quality of financial reporting, the firm’s operating performance, and
can have extensive economic and capital market consequences (Faccio 2006, Cohen et al. 2010,
Horton et al. 2012, Fan et al. 2014). In particular, corporations often have economic interests and
important incentives to establish political connections with individual politicians in the United
States. Political connections have been shown to positively impact the value of the firm as well
as future financial performance and overall firm profitability (Faccio 2006, Faccio and Parsley
2009, Cooper et al. 2010). At the same time, these political ties can also lead to an increase in the
overall level of risk at the firm. This is because politicians often grant unobservable political
favors to firms in an unpredictable manner, which in turn disrupt the time-series pattern of
reported earnings and increase future earnings uncertainty for the firm.
Anecdotal evidence suggests that financial analysts and investors recognize that political
connections affect firm value. Analysts’ written reports often underscore the importance and
value of a firm’s political connections in highlighting the strengths of the firm for investment
purposes. For example, analyst Craig Woker writes, “Citigroup's upper management and
directors, including Sandy Weill, Robert Willumstad, Charles Prince, and Robert Rubin, are
well-respected. Credibility and political connections will be as important as Citigroup's core
fundamentals in the months ahead.” Analysts typically examine two facets of political
connections: (1) whether the firm makes political contributions and to whom the contributions
2
are made, and (2) whether officers, large shareholders, or members of the board of directors hold
a top political position. Analysts are generally optimistic in their reports about political
connections, but they are also aware of the potential for increased business risk. For example,
analyst Kevin St. Pierre writes,
“CEO Vernon Hill announced in 2003 that Commerce Bank would close down its
political action committee amidst allegations that their significant political contributions
were influencing government business in banking, bond underwriting, and insurance. …
The bank remains exposed to scrutiny of its corporate governance practices. At least 5 of
13 main board members have powerful political ties. … We believe the politically
charged element of the board stigmatizes the company, invites suspicion, and distracts
potential investors from the company’s very strong fundamentals.”
I examine whether analysts fully understand the economic complexities of political
connections and whether they efficiently process information concerning this connectedness into
their stock recommendations. In general, firms that maintain political connections with
politicians are riskier than firms with no such connections. Prior research finds that politically
connected firms have greater future earnings volatility than non-connected firms (Cooper et al.
2010). This increased earnings volatility is associated with the unpredictability of the timing of
political favors which are bestowed upon the firm through politicians. Furthermore, analysts
have greater difficulty in estimating the future earnings of politically connected firms due to
changes in the time-series pattern of earnings as a result of uncertain political benefits (Chen et
al. 2010). Therefore, to the extent that analysts are unable to efficiently process information on
firm-specific political connections, their stock recommendations will have lower profitability.
Since earnings predictability is much lower for firms with political connections, it is possible that
analysts will rely less on earnings forecasts as inputs in generating stock recommendations for
politically connected firms. Therefore, I also investigate whether political connections affect the
3
efficiency with which analysts translate their earnings forecasts into profitable stock
recommendations.
For this study, I use data on political contributions1 from the Federal Election
Commission (FEC) through the Center for Responsive Politics in the U.S. This database contains
detailed information on the political candidates to whom firms contribute as well as the amounts
contributed. Following Cooper et al. (2010), I design four proxies for political connectedness
based on the following: (1) the number of supported candidates, (2) the number of supported,
incumbent candidates weighted by the length of the firm-candidate relationship, (3) the number
of supported, incumbent candidates that hold office in the same state where the firm is
headquartered, and (4) the number of supported, incumbent candidates weighted the political
power of the candidate.
I find strong evidence that analysts’ stock recommendations are significantly less
profitable when the firms they follow are politically connected. This negative relation holds
when using the main measure of political connectedness based on the number of candidates as
well as the three more sophisticated measures which take into account the strength of the
political relationship, the ability of the politician to help the firm, and the power of the political
candidate. Recommendations continue to be less profitable for politically connected firms after
controlling for the value relevance of earnings and the poor accounting quality of politically
connected firms. I also show that for firms that have political connections, analysts’ earnings
forecast accuracy is negatively related to recommendation profitability. Overall, my findings
suggest that analysts do not fully understand the economic complexities of political connections
1
I use the terms political connections, political contributions, campaign contributions, and political donations
interchangeably throughout the study.
4
and do not efficiently process information on political connections into their stock
recommendations.
In additional analysis, I examine a subset of firms that either start up a new political
action committee (PAC) or shut down an existing PAC in order to analyze the effect of a
significant increase or decrease in political connections. For firms that start up a new PAC (close
down an existing PAC), I find that recommendation profitability is significantly lower during the
years in which the PAC is in operation compared to the years before the PAC is created (after the
PAC is shut down). Furthermore, I also show that analysts have lower effectiveness in translating
earnings forecasts into profitable recommendations during the years in which the firm is
operating the PAC as opposed to years prior to PAC creation (or years after PAC closure).
This study contributes to the literature on social network connections, corporate political
connections and political contributions, and the mapping of earnings forecasts into stock
recommendations. First, I contribute to the broader literature that examines the economic
consequences of corporate social network connections. This literature shows that a firm’s
network ties can substantially influence financial reporting quality, the operating performance of
the firm, and capital market outcomes (Faccio 2006, Cohen et al. 2010, Horton et al. 2012,
Brown and Drake 2014, Fan et al. 2014, Ishii and Xuan 2014). Specifically, many studies find
that political connectedness is directly related to the value of the firm. For example, Faccio
(2006) uses an international sample of firms and finds that there is an increase in firm value at
the announcement of officers or large shareholders entering politics. In addition, Faccio and
Parsley (2009) find a 1.7% decline in firm value for companies headquartered in the town of a
politician who has died unexpectedly. These studies provide strong evidence that political
connections matter for investors for firm valuation. I contribute to this literature by showing that
5
analysts’ stock recommendations generate significantly less profit for investors when the firms
the analysts follow are politically connected. In essence, I show that analysts’ stock
recommendations are one channel through which information related to political connections
directly affects firm valuation.
Second, I inform the literature that examines how analysts map their earnings forecasts
into stock recommendations by documenting another important factor that influences this
process. Traditionally, the financial analyst literature has focused on either analysts’ earnings
forecast accuracy or stock recommendation profitability independently of one another. Schipper
(1991) states that the process of forecasting earnings is one of many pieces of information used
to arrive at a stock recommendation. She argues that earnings forecasts are not the ultimate end
product but rather an important input into the analysts’ final output which is the stock
recommendation. Subsequently, studies have examined how analysts actually use their earnings
forecasts as inputs into generating stock recommendations. For example, Loh and Mian (2006)
show that analysts who are more accurate have more profitable recommendations. Ertimur et al.
(2007) show that this positive relation holds only when earnings are value-relevant. In essence,
earnings forecasts are less likely to be an essential input into the analysts’ valuation model if the
earnings are not an important determinant of firm value. I contribute to this line of literature by
showing that analysts who follow politically connected firms rely less on earnings forecasts as
inputs into generating stock recommendations due to the inherent riskiness and uncertainty
associated with corporate political connections.
Section II and Section III provide a review of the relevant literature and hypothesis
development. Section IV and Section V consist of the measurement of political connections and
6
research design. Section VI, Section VII, and Section VIII present the sample selection, results,
and sensitivity analyses. Section IX concludes the study.
II.
PRIOR LITERATURE
Mapping of Earnings Forecasts into Recommendations
Schipper (1991) motivated researchers to study analysts’ decision-making process in
more detail by examining how analysts’ earnings forecasts are used as inputs into analysts’
valuation models from which they calculate the intrinsic value of the firm and ultimately issue a
stock recommendation. Prior literature then began to assess the effectiveness of analysts in
transforming their earnings forecasts into stock recommendations. Loh and Mian (2006) show
that analysts who are more accurate tend to have more profitable recommendations. Ertimur et
al. (2007) extend this line of literature by showing that the positive relation between accuracy
and profitability holds only for firms whose earnings are value relevant. They also show that
analysts are more effective in translating their earnings forecasts into recommendations when
they do not have conflicts of interests stemming from investment banking activities.
Another line of literature examines the relation between earnings forecasts and stock
recommendations by using different valuation models as proxies for analysts’ private valuation
models. Bradshaw (2004) finds that there is a positive (negative) relation between stock
recommendations and simple heuristic valuation models (residual income valuation models).
This implies that analysts do not use their earnings forecasts in a sophisticated manner in
generating recommendations. Simon and Curtis (2011) extend Bradshaw’s work by showing that
the negative relation between recommendations and residual income valuations is weakest for
the most accurate analysts. Barniv et al. (2009) and Chen and Chen (2009) show that even
7
though the relation between recommendations and sophisticated earnings models is negative,
analysts are showing improvements in their translations of earnings forecasts into
recommendations in the post-Regulation FD period. Ke and Yu (2009) analyze different
explanations for why analysts do not effectively translate their earnings forecasts into
recommendations. They argue that analysts could simply be unable to make this transformation
in an efficient manner, could optimistically bias recommendations in order to obtain information
from management, or could be affected by psychological biases. I extend this line of literature by
examining a different factor pertinent to firm valuation, namely political connections, which
affects how analysts translate earnings forecasts into stock recommendations.
Political Contributions
In this study, I use political contributions to proxy for a firm’s political connections.
There are two competing theories in the economics and political science literature that seek to
explain why firms make political contributions to politicians: the theory of investment in
political capital and the theory of consumption. If political contributions are an investment in
political capital, the firm will realize returns on its investment as the politician grants favors to
the firm throughout his tenure in office.2
2
Anecdotal evidence from INVESTEXT reports is consistent with analysts assuming political contributions are
indicative of the theory of investment in political capital: (1) “Freddie Mac’s Richard Syron has held executive
positions at the Federal Home Loan Bank, the Federal Reserve, and the American Stock Exchange. In addition to his
doctorate in economics, he is known to be politically adept - something the board was searching for.” – Jim
Callahan; (2) “Among ICx Technologies' board of directors are former Energy Secretary Spencer Abraham and
former Transportation Secretary Rodney Slater. The chairman of the board, Mark Mills, has done consulting to the
White House and has worked with some of the federal research laboratories. These high-level connections may
assist ICx in winning lucrative government contracts.” – Michael Pierson; (3) “The Shaw Group's experience in
restoration work along with its political contacts - CEO Bernhard was chairman of the Louisiana Democratic Party,
and a former governor of Louisiana sits on Shaw's board - should help the company continue to win work associated
with the Gulf Coast's restoration efforts.” – John Kearney; (4) “The Catellus management team, and Mr. Rising in
particular, have strong political connections and public policy expertise. We believe this is an important attribute of
the company, as land development projects often have community and city political hurdles to overcome.” –
Jonathan Litt.
8
Stigler (1971) theorizes that government officials influence firms’ financial performance
through direct subsidies, favorable tax treatment, government contracts, and barriers to entry.
However, politicians who receive contributions can also affect policy outcomes through
unobservable actions such quid pro quo purchases of legislative votes, legislative pressure on
regulatory agencies (Snyder 1990), as well as buying access for an opportunity to make the
firm’s concerns known to legislators directly (Hall and Wayman 1990). Furthermore, Snyder
(1992) finds that political contributions are more reflective of long-term investments rather than
short-term, quid pro quo investments. He shows that firms tend to donate to the same politicians
over time, to younger representatives, and to candidates running for offices that are “stepping
stones” to higher offices which are more influential. The long-term relationships between firms
and politicians often manifest into political favors for the donating firms as opportunities arise
during the legislator’s tenure in office.
If the market views the firm’s political contributions as an investment with positive net
present value, there should be a positive relation between political contributions and future
returns. Indeed, Cooper et al. (2010) find that political contributions are positively related to
future stock returns and future operating performance. In addition, Claessens et al. (2008) find
that Brazilian firms that made political contributions experienced higher stock returns around the
1998 and 2002 elections than firms that did not make contributions. However, other studies find
an adverse effect of political contributions. For example, Aggarwal et al. (2012) find a negative
relation between the amount of campaign contributions and future stock returns.
Although the majority of studies on political contributions rely on the theory of political
investment, most studies find only weak, or no evidence, that contributions affect policy.
Tullock’s (1972) puzzle seeks to determine why there is so little money in U.S. politics. If
9
political contributions are made with the intent of earning a rate of return and the value of public
policy has significant worth to corporations, Tullock argues that firms would want to give
exponentially more money to politicians (soft money loopholes have traditionally been available
even if PAC contributions were to reach the legal limit). Hart (2001) points out that we still
know very little about why some firms choose to make campaign contributions and others do
not, and why some firms give a lot while others give a little. In addition, the investment in
political capital motivation raises the concern of an “undemocratic exchange of policy for
dollars” (Gordon et al. 2007). The authors point out that even if this exchange exists it will be
difficult to detect because the politicians will conceal their actions for fear of being exposed.
The alternative competing theory proposes that political contributions are a form of
consumption and not an investment. If this is true, the firm will not have an expectation of being
rewarded with favorable legislation in the future. Any money donated by the firm will simply be
an expression of individual political participation. Furthermore, some managers will give
because they are ideologically motivated, have personal preferences over candidates and parties,
or desire to be appointed to cabinet positions or ambassadorships (Aggarwal et al. 2012).
Ansolabehere et al. 2003 argue that the theory of consumption is more likely to explain why
firms make political contributions since all of the firms’ donations ultimately come from
individuals. They state that individuals essentially donate because they are ideologically
motivated and wish to participate in politics, not because they expect politicians to reciprocate
favors to the firm where they are employed. They further show that the amount given is usually
very small per individual and per firm and these insignificant donations are not likely to
influence politicians. They conclude that the small dollar amounts of political donations which
come from many different individuals are not being made to purchase policy.
10
III.
HYPOTHESIS DEVELOPMENT
Firms with political connections have a greater amount of risk due to the increased
uncertainty surrounding the timing of the income effects of politicians’ actions and the
legislative process. Recent research finds that politically connected firms have greater future
earnings volatility as compared to similar non-connected firms (Cooper et al. 2010). One
explanation for this increased volatility is that political favoritism is unpredictable and disrupts
the times-series pattern of the firm’s earnings. Furthermore, analysts have greater difficulty in
forecasting the future earnings of politically connected firms than those of non-politically
connected firms (Chen et al. 2010).
This evidence suggests that even sophisticated market participants who follow politically
connected firms very closely have difficulty in effectively anticipating the short-term effects of
political connections on firm valuation. Factors that can contribute to the increased future
variability of earnings include preferential access to bank financing and credit (Johnson and
Mitton 2003, Cull and Xu 2005, Leuz and Oberholzer-Gee 2006), increased awards of
government procurement contracts (Agrawal and Knoeber 2001, Goldman et al. 2013),
government bailouts in times of financial distress (Faccio et al. 2006), and benefits related to
reduced income taxation (Wu et al. 2012, Kim and Zhang 2014). To the extent that analysts use
their earnings forecasts as inputs into their private valuation models in generating stock
recommendations, the increased unpredictability inherent in the future earnings of politically
connected firms will affect the level of stock recommendation profitability.
Furthermore, prior literature finds that analysts issue more favorable stock
recommendations (and more optimistically biased earnings forecasts) in order to acquire private
information from management (Chen and Matsumoto 2006, Ke and Yu 2006). This relation is
11
even more pronounced when the firm has earnings which are hard to predict. Thus, if analysts
bias their recommendations for politically connected firms more than for non-connected firms in
order to receive useful private information from managers, it is not likely that they are using their
earnings forecasts as primary inputs to valuation models in generating stock recommendations.
Furthermore, any bias documented in stock recommendations for politically connected firms
could be related to the type of (unobservable) valuation model that the analyst uses. For example,
Bradshaw (2004) finds that analysts’ stock recommendations are more related to simple,
heuristic growth models than to sophisticated residual income valuation models. This indicates
that, on average, analysts do not use their earnings forecasts in generating recommendations,
although this relation has been reversing in the post-Reg FD period (Barniv et al. 2009, Chen and
Chen 2009). Since earnings volatility is higher for politically connected firms compared to nonconnected firms, it is likely that analysts who cover politically connected firms will not use all of
the information in their earnings forecasts in generating recommendations.
On the other hand, political relationships between firms and politicians tend to be longterm, and experienced analysts will possess many years of experience following a politically
connected firm. Analysts that have such long-standing relationships with companies will build
up very valuable, in-depth knowledge of the firm, quality of management, industry expertise, and
firm-specific, idiosyncratic information. Over an analyst’s tenure following a firm, the analyst
will experience the firm establishing political connections in some years, terminating
connections in other years, and strengthening or weakening connections in still other years.
Analysts will have the opportunity to learn from the effect of differing degrees of political
connections manifesting into political favors which influence future reported earnings. Thus, the
analyst’s specialized knowledge which has been developed over time should improve forecast
12
accuracy for these types of firms. Therefore, it is possible that analysts who have more
experience with politically connected firms, or analysts who perform more private information
search for these firms, will rely more on earnings forecasts in their private valuation models.
These competing predictions lead to the following non-directional hypotheses as stated below:
H1: Political connections are not related to stock recommendation profitability.
H2: Political connections do not affect the relation between analysts’ earnings forecast
accuracy and stock recommendation profitability.
IV.
POLITICAL CONNECTIONS MEASUREMENT
Political Contributions
Firms that would like to contribute to federal candidates and political parties are required
by the Federal Election Campaign Act of 1974 (FECA) to create political action committees
(PACs). Firms are allowed to pay for the start-up, overhead, and fundraising expenses of the
PAC but cannot give funds from the treasury to the PAC for the purpose of contributing to a
federal candidate. PAC contributions are mainly donated by the firm’s managers. In my sample,
11.8% of publicly traded firms have PACs. Approximately, one-third of all industries do not
have any firms with PACs. PACs are allowed to make direct contributions to candidates, also
called “hard money” contributions.3 In addition, PACs cannot give more than $10,000 in a 2year election cycle to any particular candidate.
Firms have also been able to make “soft money” donations directly to political parties for non-partisan partybuilding activities and issue advertising which does not specifically name political candidates. The Bipartisan
Campaign Reform Act of 2002 banned soft money contributions. However, in 2010, the ruling on Citizens United v.
Federal Election Commission ended the ban on soft money, and firms are now able to provide unlimited funds to
support or oppose political candidates to Super PACs. For issues related to campaign finance reform at the state
level see Gross and Goidel (2003).
3
13
Corporate political contributions have been used extensively in the prior literature to
measure political connections for several reasons (Myers 2005). First, firms have been required
to file publicly available, detailed information with the FEC since 1979 on their political
contributions made through their PACs. Second, since firms must report the amount of money
given by the PAC and the identity of the receiving political candidate, the firm-candidate
relationship is a direct and powerful measure of a political connection. The Lobby Reform Act of
1995 also requires firms to disclose all expenditures on executive and legislative lobbying. Even
though firms spend about ten times more on lobbying than on PAC contributions, lobbying
expenses usually cannot be attributed to any politician directly, and thus are indirect measures of
political connections. Third, political contributions are highly correlated with other forms of
political activity such as lobbying and soft money expenditures. For example, Bertrand et al.
(2011) analyze individual lobbyists’ political contributions and find a strong correlation with
their client firms’ contributions. Furthermore, they show that lobbyists systematically switch
issues as the politicians to which they make political contributions switch committee
assignments. Therefore, PAC contributions serve as a strong indicator of a political connection.
I follow Cooper et al. (2010) and design my measures of political connections based on
the number of supported candidates. They argue that if hard money contributions are correlated
with other ways in which the firm establishes relationships with politicians, the number of
politicians that a firm supports is a good proxy for a firm’s political involvement. Additionally,
Ansolabehere et al. (2003) state that only 4% of all PAC contributions are at or near the $10,000
FEC limit and the average PAC contribution is only $1,700. Therefore, since PAC contributions
are not binding and the limit is an immaterial amount to the firm, it is not the dollar amount of
the contribution that matters but rather to whom the connection is made.
14
Measures using Political Contributions
The main measure of political connections, following Cooper et al. (2010), is:
𝐽
𝐶𝐴𝑁𝐷𝐼𝐷𝐴𝑇𝐸𝑆𝑡 = ∑ 𝐶𝑎𝑛𝑑𝑖𝑑𝑎𝑡𝑒𝑠𝑗𝑡
(1)
𝑗=1
Candidatesjt is equal to one if the firm has contributed to candidate j in year t.
POLITICALCandidates equals 1 if CANDIDATES is greater than the industry-year mean, and 0
otherwise.
I also use the authors’ three alternative measures to create more refined and powerful
measures of political connections. The second proxy weighs equation (1) by factors related to the
strength of the relationship between the firm and the candidate. This is because firms tend to
build relationships with politicians over long periods of time (Snyder 1992). To capture the
strength of the relationship, I estimate the following equation:
𝐽
𝑆𝑇𝑅𝐸𝑁𝐺𝑇𝐻𝑡 = ∑ 𝐶𝑎𝑛𝑑𝑖𝑑𝑎𝑡𝑒𝑠𝑗𝑡 × 𝐼𝑗𝑡 ×
𝑗=1
𝑁𝐶𝑉𝑗𝑡
× 𝑟𝑒𝑙𝑙𝑒𝑛𝑔𝑡ℎ𝑗𝑡
𝑁𝑂𝑉𝑗𝑡
(2)
Candidatesjt is equal to one if the firm has contributed to candidate j over in year t. Incumbents
are more likely to have a greater number of corporate contributors and receive a greater dollar
amount of contributions (Snyder 1990, Gross and Goidel 2003). Therefore, I include Ijt as equal
𝑁𝐶𝑉
to one if the candidate is incumbent at time t, and 0 otherwise. The ratio 𝑁𝑂𝑉𝑗𝑡 captures the
𝑗𝑡
strength of the candidate’s party in relation to the opposing party. NCVjt is the number of votes
that candidate j’s party holds in office at time t. NCVjt is the number of votes that candidate j’s
party holds in office at time t. NOVjt is the number of votes that candidate j’s opposing party
holds in office at time t, and rellengthjt is the number of continuous months of the firm-candidate
15
relationship. POLITICALStrength equals 1 if STRENGTH is greater than the industry-year mean,
and 0 otherwise.
The third proxy weighs equation (1) by factors related to the ability of the candidate to
help the firm. Specifically, the ability of the politician to assist the firm increases significantly if
the firm is located in the same state or district (Kroszner and Stratmann 1998). To capture the
ability of the politician to help the firm, I estimate the following equation:
𝐽
𝐴𝐵𝐼𝐿𝐼𝑇𝑌𝑡 = ∑ 𝐻𝑜𝑚𝑒𝐶𝑎𝑛𝑑𝑖𝑑𝑎𝑡𝑒𝑗𝑡 × 𝐼𝑗𝑡 ×
𝑗=1
𝑁𝐶𝑉𝑗𝑡
𝑁𝑂𝑉𝑗𝑡
(3)
HomeCandidatejt is equal to one if the firm has contributed to candidate j in year t and the firm is
headquartered in the same state in which the candidate is running for office. All other variables
are defined above. POLITICALAbility equals 1 if ABILITY is greater than the industry-year mean,
and 0 otherwise.
The fourth proxy weighs equation (1) by factors related to the power of the candidate.
This is because the prior literature shows that politicians who are committee chairs or serve on
powerful committees receive more contributions than other politicians (Grier and Munger 1991).
To capture the power of the politician, I estimate the following equation:
𝐽
𝑀
𝑗=1
𝑚=1
𝑁𝐶𝑉𝑗𝑡
𝑅𝑎𝑛𝑘𝑚𝑡
𝑃𝑂𝑊𝐸𝑅𝑡 = ∑ 𝐶𝑎𝑛𝑑𝑖𝑑𝑎𝑡𝑒𝑠𝑗𝑡 × 𝐼𝑗𝑡 ×
× [∑
] (4)
𝑁𝑂𝑉𝑗𝑡
𝑀𝑒𝑑𝑖𝑎𝑛 𝑅𝑎𝑛𝑘𝑚𝑡
𝑗
Rankmt is the reciprocal of candidate j’s rank on committee m (where rank =1 for the most
important member, rank = 2 for the next important member, etc.). Median Rankmt is the median
number of members on committee m. All other variables are defined above. POLITICALPower
equals 1 if POWER is greater than the industry-year mean, and 0 otherwise.
16
V.
RESEARCH DESIGN
To examine the relation between forecast accuracy and recommendation profitability, I
follow Ertimur et al. (2007) and first estimate the following benchmark model for the pooled
sample of all stock recommendations and earnings forecasts:
PFT = α0 + α1ACCURACY + α2VR + α3FIRMEXP + α4BSIZE + α5N_FIRMS
+ α6REC_FREQ + α7LFR + α8COV + α9REG + ε
(5)
Analyst, firm, forecast, and year subscripts are omitted for the sake of brevity. I measure
forecast accuracy, ACCURACY, as the relative forecast accuracy which compares an analyst’s
absolute forecast error to the mean absolute forecast error of all analysts following the firm
(Clement 1999) as described below:
𝐴𝐶𝐶𝑈𝑅𝐴𝐶𝑌 = −1 ×
𝐴𝐹𝐸𝑖𝑗𝑘 − 𝐴𝐹𝐸
𝐴𝐹𝐸
,
where 𝐴𝐹𝐸𝑖𝑗𝑘 is the absolute forecast error for earnings forecast k that analyst i issues for firm j
and 𝐴𝐹𝐸 is the mean absolute forecast error for firm j by all analysts. Clement (1999) shows that
this measure of relative accuracy controls for firm-year effects and reduces heteroskedasticity in
forecast error distributions across firms. The equation is multiplied by negative one so that
higher values of ACCURACY connote higher levels of accuracy. I expect α1 to be positive and
significant consistent with the prior literature which finds that the positive relation between
accuracy and profitability (Loh and Mian 2006).
Recommendation profitability, PFT, is the market-adjusted return to recommendation k
made by analyst i for firm j. The buy-and-hold return is computed from the day before the
recommendation date until the earlier of 30 days or 2 days before the recommendation is revised
or reiterated. For buy (hold and sell) recommendations, $1 is invested (shorted) in the
recommended stock. I control for the value relevance of earnings since Ertimur et al. (2007)
17
show that the relation between forecast accuracy and profitability is affected by whether or not
the firm has value-relevant earnings. Following Francis et al. (2004), value relevance, VR, equals
the R2 from the following regression estimated for each firm over rolling 10-year windows:4
RET = α0 + α1EARN + α2ΔEARN + ε,
where RET is the firm's 15-month size-adjusted return ending three months after the fiscal yearend, EARN is income before extraordinary items scaled by market value at the end of the prior
fiscal year, and ΔEARN is the change in income before extraordinary items scaled by market
value at the end of the prior fiscal year. Firm experience, FIRMEXP, is the log of the number of
years through year t for which analyst i supplied forecasts for firm j. Brokerage firm size, BSIZE,
is the log of the number of analysts employed by the brokerage that analyst i works for during
year t. Number of firms, N_FIRMS, is the log of the number of firms for which analyst i issues
annual forecasts during the year in which the recommendation is made. Recommendation
frequency, REC_FREQ, is the log of the number of recommendations analyst i issues for firm j
during year t. Leader-follower ratio, LFR, is the cumulative number of days by which the
preceding two forecasts lead forecast k divided by the cumulative number of days by which the
subsequent two forecasts follow forecast k following Cooper et al. (2001). Number of analysts,
COV, is the log of the number of analysts who issue forecasts and recommendations for firm j
during year t. Lastly, regulated industry, REG, equals 1 if the firm operates in the financial
services industry (one-digit SIC code 6) or in the utilities industry (two-digit SIC code 49), and 0
otherwise.
4
The estimation requires at least 5 years of data. However, the results are not sensitive to the minimum number of
years.
18
To test the first hypothesis, I examine how political connections affect recommendation
profitability by supplementing the benchmark model with the political connections proxy using
the framework below:
PFT = β0 + β1POLITICALi + β2ACCURACY + β3VR + β6FIRMEXP + β7BSIZE
+ β8N_FIRMS + β9REC_FREQ + β10LFR + β11COV + β12REG + ε
(6)
POLITICALi is one of four proxies for political connections. Following Cooper et al.
(2010), I use the following constructs to design measures for political connectedness: (1)
CANDIDATES is the number of supported candidates, (2) STRENGTH is the number of
supported, incumbent candidates weighted by the length of the firm-candidate relationship, (3)
ABILITY is the number of supported, incumbent candidates that hold office in the same state
where the firm is headquartered, and (4) POWER is the number of supported, incumbent
candidates weighted the power of the candidate. POLITICALi equals 1 if the measure is greater
than the industry-year mean, and 0 otherwise. See Section IV for details on variable construction.
Next, I test the second hypothesis by examining how political connections affect the
relation between forecast accuracy and recommendation profitability. I supplement the baseline
model with both the political connections proxy and the interaction between political connections
and forecast accuracy:
PFT = β0 + β1POLITICALi + β2ACCURACY + β3ACCURACY*POLITICALi + β4VR
+ β5FIRMEXP + β6BSIZE + β7N_FIRMS + β8REC_FREQ + β9LFR + β10COV
+ β11REG + ε
(7)
19
VI.
SAMPLE SELECTION
The sample uses the Compustat annual database, CRSP daily stock prices, I/B/E/S detail
history file, and I/B/E/S detail recommendations file from 1993 – 2012. The sample begins in
1993 because this is the first year stock recommendations are made available in I/B/E/S. Every
recommendation is matched to an annual earnings forecast from the same analyst and the same
firm during the 30-day period prior to and including the issue date of the recommendation. Firmyears must have at least three analysts following the firm to remain in the sample because my
accuracy measure for an analyst is relative to the average accuracy of all other analysts following
the firm (Clement 1999).
For political contribution data, I obtain the database of the Federal Election Commission
(FEC) filings through the Center for Responsive Politics. This database contains all corporate
political action committee (PAC) donations to individual political candidates running for federal
office in the House of Representatives, Senate, and Presidency. I do not obtain political
contributions from individuals, labor organizations, trade organizations, party committees,
private companies, or subsidiaries of foreign firms because these sources of monetary influence
are not related to the firm’s PAC. To calculate a candidate’s committee ranking, I acquire data on
House and Senate committee assignments and rankings from Charles Stewart’s Congressional
Data Page.
I manually match the names of firms provided by the FEC database to CRSP company
names. I successfully match approximately 57% of 1,933 PAC corporate names to CRSP data
and about 82% of the dollar amount of contributions. This is consistent with Grier et al. (1994)
who match 50 - 60% of names and 80% of contributions. After using the SEC’s EDGAR
database to properly match companies with name changes, I obtain 851,813 political
20
contributions from 1,102 unique firms. The former figure includes multiple contributions to
some of the same candidates. These contributions are spread over 9,247 different political
candidates. After merging the FEC database with I/B/E/S, CRSP, and Compustat, I obtain
220,843 analyst-firm-year observations for 4,942 unique firms for the final sample.
VII.
RESULTS
Panel A and Panel B of Table 1 reports the descriptive statistics for financial analyst
characteristics for the full sample of firms. In Panel A, the mean abnormal return to stock
recommendations (PFT) is 5.8% over the 30 day measurement period. The mean (median) of
ACCURACY is -0.170 (0.109). Analysts have approximately 3.7 years of firm-specific
experience (UFIRMEXP), follow 16 different companies (UN_FIRMS), and issue about 2
recommendations per year (UREC_FREQ) for the firms they cover. The average brokerage firm
in the sample employs 65 analysts (UBSIZE). Analyst coverage (UCOV) is approximately 14
analysts per firm. Of the total 4,942 firms in the sample, approximately 524 firms (10.7%) make
political contributions and 4,418 firms (89.3%) do not make political contributions.
Panel B of Table 1 provides the descriptive statistics comparing firms that are politically
connected to firms that are not politically connected. Analysts of politically connected firms have
less profitable recommendations (PFT = 4.2%) compared to analysts of firms that are not
politically connected (6.0%). However, analysts of connected firms have greater forecast
accuracy (ACCURACY = 0.008) than analysts of non-connected firms (-0.191). Analysts that
follow firms are connected tend to have more years of firm-specific experience (UFIRMEXP =
4.7) and cover more companies (UN_FIRMS = 17.1) than analysts who follow firms that are not
connected (3.6 years and 16.2 firms, respectively). In addition, analysts who cover politically
21
connected firms tend to work at brokerage firms that employ fewer analysts (UBSIZE = 72.6)
than analysts who cover non-connected firms (64.6 analysts). Furthermore, analysts of connected
firms issue recommendations more frequently (REC_FREQ = 2.0) and more timelier earnings
forecasts (LFR = 2.2) compared to analysts of non-connected firms (1.8 and 2.3, respectively).
The number of analysts following a politically connected firm (UCOV = 21.3) is significantly
larger than the number of analysts following a non-connected firm (13.3 analysts) which is
consistent with contributing firms being the largest publicly-traded firms. Lastly, politically
connected firms tend to have significantly less negative discretionary accruals (DACC = -0.020)
and are more likely to be in a regulated industry (REG = 0.25) than non-connected firms (-0.028
and -0.028, respectively).
Panel A of Table 2 report the descriptive statistics for the political contributions that
firms make. The average firm donates $125,100 in total political contributions during any 2-year
election cycle. Total political contributions range from $5 to $3,827,600. Firms make
approximately 95 political contributions during any 2-year election cycle although the maximum
is 2,077 contributions. Firms support anywhere between 1 and 570 candidates every election
cycle, with the average firm supporting 50 candidates. Although there is significant variation in
the number of candidates that some firms support cross-sectionally, there is little variation on a
per-firm basis in the number of supported candidates over time.
Firms donate about $2,502 per candidate per election cycle ($125,100/50 candidates).
Interestingly, Cooper et al. (2010) report an average of $2,086 per candidate from total mean
contributions of only $64,694 and 31 supported candidates over the period 1979 - 2004.
Therefore, firms are still not constrained by the FEC contributions limit which is $10,000 per
candidate over an election cycle. Even though firms are contributing significantly more in total
22
dollar amounts, they are supporting a similar number of candidates per election cycle. It is
possible the larger amounts of money being donated are reflective of the benefits that firms
expect to obtain from connections to politicians which would be supportive of the theory of
investment in political capital.
Panel B of Table 2 reports political contributions by the ten largest industry contributors
based on the Fama-French (1997) 49 industry classification5. Consistent with Aggarwal et al.
(2012), the top contributors are Financial, Telecommunications, Utilities and Transportation. In
general, most of the industries have increased their political contributions over the sample
period. The Financial industry increased donations by 86% from $6.97 million to $12.95 million
over the period 1993 to 2012. The Utilities industry increased contributions by 112% from $3.95
million to $8.40 million over the same period while the Business Services industry increased
donations by 632% from $665,000 to $4.87 million.
Panel C of Table 2 reports the number of political candidates to whom the largest ten
industries contributed. The industries that contribute to the most political candidates over the full
time period are Financial, Telecommunications, Utilities and Transportation. However, over
time, the Financial, Telecommunications, and Utilities industries have actually decreased the
total number of candidates to whom they contribute. Since these industries have significantly
increased the dollar amounts of contributions, they appear to be focusing more on select
candidates using increased funding for targeted political strategies. On the other hand, the five
industries that have increased the number of candidates to whom they contributed over the
sample period include Transportation, Electronic Equipment, Pharmaceutical, Business Services,
Banking, Insurance, and Trading are all included in “Financial” because the Center for Responsive Politics
generally aggregates these industries when reporting the largest source of campaign contributions to federal
candidates and parties.
5
23
and Machinery. Since these industries have also significantly increased the amount of
contributions, they appear to be diversifying their donations more across different politicians for
strategic reasons.
Table 3 reports the Pearson and Spearman correlations for the full sample. The Pearson
correlation between political connectedness (POLITICAL) and profitability (PFT) is -0.043 and
significant at the 1% level indicating that, on average, analysts’ recommendations are less
profitable for politically connected firms. Forecast accuracy (ACCURACY) and profitability
(PFT) have a positive and significant Pearson correlation, 0.014, consistent with Loh and Mian
(2006). However, in the multiple regressions that follow, this positive relation is subsumed by
the value relevance of earnings. Value relevance (VR) and profitability (PFT) are also positively
related as evidenced by a Pearson correlation of 0.026 significant at the 1% level. There is a
positive Pearson correlation between profitability (PFT) and brokerage firm size (BSIZE) of
0.028, recommendation frequency (REC FREQ) of 0.023, and forecast timeliness (LFR) of
0.037. There is a negative Pearson correlation between profitability (PFT) and the number of
firms followed (N_FIRMS) of -0.006, analyst coverage (COV) of -0.061, and regulated industries
(REG) of -0.034.
Political connectedness (POLITICAL) and forecast accuracy (ACCURACY) are
positively and significantly related, 0.056, at the 1% level. Lastly, the correlation between
political connectedness (POLITICAL) and discretionary accruals (DACC) is 0.018 and is positive
and significant consistent with Chaney et al. (2011) who find that politically connected firms
have lower earnings quality. The majority of the correlations among the independent variables
are statistically significant but their magnitudes are not large. I investigate the issue of
multicollinearity further by calculating the Variance Inflation Factors (VIF) in all multivariate
24
regressions and find that the VIFs are all less than 2.0. This suggests that multicollinearity should
not be of serious concern.
In Table 4, I first report the results of the relation between earnings forecast accuracy and
recommendation profitability using a modified benchmark model based on the findings of
Ertimur et al. (2007). I first document that there is a positive relation between recommendation
profitability (PFT) and earnings forecast accuracy (ACCURACY = 0.0007, p = 0.0159) consistent
with Loh and Mian (2006). This indicates that superior analysts who have more accurate
earnings forecasts tend to have more profitable recommendations on average. I also control for
firms with value-relevant earnings since Ertimur et al. (2007) show that the value-relevance of
earnings affects the relation between recommendation profitability and earnings forecast
accuracy. I find that the coefficient on value relevance is positive and significant at the 1% level
(VR = 0.0129, p <.0001). In regards to control variables, I find that profitability (PFT) increases
with firm-specific experience (FIRMEXP = 0.0041), brokerage firm size (BSIZE = 0.0032),
recommendation frequency (REC_FREQ = 0.0047), and earnings forecast timeliness (LFR =
0.0010) as expected. The table also shows that profitability (PFT) decreases with analyst
coverage (COV = -0.0118) and regulated industries (REG = -0.0123).
Panel A of Table 5 presents the results which test hypothesis one. This analysis examines
the effect of political connections on recommendation profitability. In column 1, the proxy for
political connections (POLITICAL) is based on the number of supported candidates and equals 1
if the firm is politically connected, and 0 otherwise. There is a negative relation between
recommendation profitability (PFT) and political connectedness (POLITICAL = -0.0122) which
is significant at the 1% level. This suggests that analysts do not understand how to interpret the
effects of political connections into their stock recommendations when valuing the firm. The
25
coefficient on forecast accuracy remains positive and significant at the 5% level (ACCURACY =
0.0007). The results also show that there is a positive and significant relation between
profitability (PFT) and value relevance (VR = 0.0126) consistent with the findings in Ertimur et
al. 2007. In addition, analysts who have more firm-specific experience (FIRMEXP = 0.0044),
work at larger brokerage firms (BSIZE = 0.0033), issue more recommendations in a given year
(REC_FREQ = 0.0047), and have more timely forecasts relative to other analysts following the
same firm (LFR = 0.0009) also have more profitable recommendations. On the other hand, less
profitable recommendations are associated with analysts who cover firms that have a larger
analyst following (COV = -0.0103) and firms in a regulated industry (REG = -0.0117).
The main results are consistent across the other three more powerful measures of political
connections following Cooper et al. (2010). In column 2, the proxy for political connections is
based on the strength of the firm-candidate relationship (Strength). The proxy in column 3 is
based on the number of “same-state” candidates (Ability). Lastly, the proxy in column 4 is based
on the number of candidates weighted by the power of the candidates (Power). All three of the
alternative political connection proxies have a negative and significant relation with
recommendation profitability. Since these more refined measures take into consideration the
strength of the relationship, the ability of the candidate to help the firm, and the power of the
candidate, they provide additional reinforcement that analysts’ stock recommendations are less
profitable when the firms they follow have political connections.
In Panel B of Table 5, I present the results for hypothesis two. I analyze how political
connections affect the relation between forecast accuracy and recommendation profitability. I
find that there is a negative and significant relation between profitability (PFT) and the
interaction of accuracy and political connectedness (ACCURACY*POLITICAL = -0.0036). This
26
signifies that analysts rely less on earnings forecasts in private valuation models used in
generating stock recommendations for politically connected firms due to these types of firms
having greater earnings uncertainty. In addition, for all of the alternative political connection
proxies there is a negative and significant relation between recommendation profitability and the
interaction of accuracy and political connections. These more sophisticated measures provide
additional support that analysts’ earnings forecasts are less consistent with recommendations
when the firms they follow have political connections.
Panel A of Table 6 examines whether controlling for accounting quality affects the
relation between recommendation profitability and political connections. This is an important
issue since Chaney et al. (2011) find that firms with political connections have lower earnings
quality than non-connected firms. For all four political connection proxies, there is a negative
relation between recommendation profitability (PFT) and political connectedness (POLITICAL)
which is significant at the 1% level. In Panel B of Table 6, I analyze how accounting quality
affects political connections and the relation between forecast accuracy and recommendation
profitability. The table shows that there is a negative and significant relation between
profitability (PFT) and the interaction of accuracy and political connections
(ACCURACY*POLITICAL) under all four proxies for political connectedness. Overall, the level
of discretionary accruals (DACC) is either not related to, or weakly related to, recommendation
profitability (PFT). This suggests that even though politically connected firms have worse
accounting quality on average, the relation between recommendation profitability and political
connectedness (as well as forecast accuracy for politically connected firms) is not subsumed by
the overall level of earnings quality at the firm.
27
VIII. SENSITIVITY ANALYSES
The Effect of Starting up (or Shutting Down) a Political Action Committee
If a firm experiences a significant increase or decrease in its political connections in a
particular year, it is possible that the effect of political connections on recommendation
profitability and forecast accuracy will be weakened or exacerbated. To investigate this issue, I
conduct an analysis for firms that either start up or shut down a political action committee (PAC)
during the sample period. For firms that start up a new PAC, this analysis allows me to examine
how analysts’ recommendation profitability changes from the time before the formation of the
PAC when the firm has no political connections to the time after the PAC formation when the
firm has established specific connections with politicians. Likewise, for firms that close down an
existing PAC, I am able to track how recommendation profitability changes from the time the
PAC is in operation to the time when the PAC is no longer functioning. Both of these analyses
allow for a stronger test for the effect of political connections on analysts’ recommendations than
the main analysis because I am able to examine a significant change in political connections at
the firm level which potentially mitigates problems concerning correlated omitted variables and
heteroskedasticity (Brown et al. 1999, Holthausen and Watts 2001).
In Panel A of Table 7, I report the results for how recommendation profitability changes
for firms that start up a new PAC during the sample period. Following Hail et al. (2014), I
include indicator variables for the years leading up to the event (Yearst ‒ 2, t ‒ 1), the years around
the event (Yearst, t + 1), and the remaining years following the event (Yearst ≥ +2). The years before
t ‒ 2 function as the base period for the event (i.e. the creation of the new PAC). Each of these
Year indicator variables is also interacted with forecast accuracy. If recommendation profitability
decreases after the firm creates the new PAC, I expect the indicator variables representing
28
Yearst, t + 1 and Yearst ≥ +2, and their interactions with forecast accuracy, to be negative and
significant. I also expect the coefficient for Yearst ‒ 2, t ‒ 1 to be insignificant. Consistent with
expectations, I find that there is a significantly negative association between recommendation
profitability and the years following the PAC formation (Yearst ≥ +2). I also find that the
coefficients for both Yearst ‒ 2, t ‒ 1 and Yearst, t + 1 are insignificant. However, since the literature
shows that political connections are long-term and often require time to materialize into benefits
for the firm, this would likely explain why the profitability decreases in years following t + 1, but
not in year t or year t + 1. Likewise the interactions of Yearst ‒ 2, t ‒ 1 and Yearst, t + 1 with
ACCURACY are also insignificant (or weakly significant). As expected, the interaction of
Yearst ≥ +2 and ACCURACY is negative and significant at the 1% level. These results provide
strong evidence that for firms that start up a new PAC, analysts’ recommendations have lower
profitability in the years after which the PAC is created compared to the years in which the same
firm did not have a PAC in operation, and thus did not have any political connections.
In Panel B of Table 7, I examine how recommendation profitability changes for firms
that shut down an existing PAC during the sample period. If recommendation profitability is
lower during the years in which the PAC is in operation compared to the years after which the
PAC has closed operations, I expect the indicator variables representing Yearst ≤ ‒ 2 and
Yearst ‒ 1, t, and their interactions with forecast accuracy, to be negative and significant. I also
expect the coefficient for Yearst +1, t + 2 to be insignificant. Consistent with expectations, I find
that there is a significantly negative association between recommendation profitability and the
years prior to the closure of the PAC (Yearst ≤ ‒ 2). I also find that the coefficients for both
Yearst ‒ 1, t and Yearst +1, t + 2 are insignificant. Similar to the findings in Panel A, these results are
indicative of the long-term nature of political connectedness, and would explain why
29
recommendation profitability is lower in years during which the PAC is in operation (Yearst ≤ ‒ 2)
than in the year immediately before the closure (Year t ‒1) or the year of closure (Year t). As
expected, the interaction of Yearst ≤ ‒ 2 and ACCURACY is negative and significant, but the
interactions of Yearst ‒ 1, t and Yearst +1, t + 2 with ACCURACY are weakly significant. This
suggests that for firms which close down PACs, analysts have lower effectiveness in translating
earnings forecasts into profitable recommendations in the years prior to the closure of the PACs.
Overall, the results from these analyses suggest that the actual creation and termination of PACs
significantly affects the extent of political connections and analysts’ recommendation
profitability.
Propensity-Score Matching
In this section, I address potential endogeneity problems due to selection bias by using a
propensity-score matching design. Since the firm has to decide to establish a political action
committee and make political contributions, the results in the main tests have potential bias. I use
the following first-stage probit model based off Cooper et al. (2010) to estimate the probability
that a firm will make political contributions:
ACTIVE = λ0 + λ1SIZE + λ2SALES + λ3EMPLOYEES + λ4BUS_SEG + λ5GEO_SEG
+ λ6BM + λ7LEV + λ8CFO + λ9MARKET_SHARE + λ10HERF
+ λ11REG + λ12INDUSTRY_ACTIVE + ε
(8)
In model (8) above, ACTIVE equals 1 if the firm has a political action committee, 0
otherwise. SIZE is the natural logarithm of market value of equity. SALES is the natural
logarithm of sales. EMPLOYEES is the log of the number of employees. BUS_SEG is the
number of business segments. GEO_SEG is the number of geographic segments. BM is book-tomarket ratio. LEV is the sum of long-term debt and debt in current liabilities, divided by total
30
assets. CFO is operating cash flow divided by total assets. MARKET_SHARE is sales divided by
total industry sales. HERF is the Herfindahl index of industry concentration computed with net
sales. REG equals 1 if the firm operates in the financial services industry (one-digit SIC code 6)
or in the utilities industry (two-digit SIC code 49), and 0 otherwise. INDUSTRY_ACTIVE is the
number of firms in a firm's industry with an established political action committee.
Propensity scores are calculated using case-control matching on the propensity score
computed from the predicted probability of the firm having a political action committee in
Appendix B. The match algorithm does not reconsider the match once a match is made. In
addition, the algorithm uses the nearest available pair matching method. The cases are ordered
and sequentially matched to the nearest unmatched control. If more than one unmatched control
matches to a case, the control is selected at random (Parsons 2000). The propensity-score
matching procedure produces a matched sample of 30,505 control observations leading to a
combined sample of 61,010 observations.
Table 8 reports the results of the relation between political connections, forecast
accuracy, and profitability after matching on propensity scores. In column one, the table shows
support for the negative relation between profitability and political connections (POLITICAL
= -0.0071) which is significant at the 1% level. This negative relation holds when using all four
of the political connection proxies (number of candidates, strength of firm-candidate
relationship, same-state candidates, and candidate power). The table also reports a negative
coefficient on the interaction term (ACCURACY*POLITICAL = -0.0055) for all four political
connection measures consistent with the main findings. Overall, the results from the propensityscore matching tests are consistent with the main implications from the primary results.
31
The Impact of Analyst Regulatory Reforms
Bradshaw (2004) finds that analysts do not incorporate their earnings forecasts into their
stock recommendations in accordance with present value models. Specifically, he finds that
analysts’ recommendations are more related to simple heuristic models than to sophisticated
residual income models that utilize analysts’ earnings forecasts. Over the period October 2000 to
April 2003, numerous analyst and disclosure regulations came into effect including Regulation
Fair Disclosure, NASD Rule 2711, the amended NYSE Rule 472, the Sarbanes-Oxley Act, the
Global Research Analyst Settlement, and Regulation Analyst Certification. Recent research
examines whether these collective analyst regulations have had an impact on how analysts use
their earnings forecasts in generating stock recommendations (Barniv et al. 2009; Chen and Chen
2009). Overall, these papers find that analyst regulatory reforms have increased the consistency
between analysts’ earnings forecasts and stock recommendations. Therefore, in this section, I test
whether these regulations have a significant impact on the relation among political connections,
recommendation profitability and forecast accuracy. I define the Pre-Regulation period as
January 1993 to September 2000 and the Post-Regulation period from May 2003 to December
2012. For purposes of comparing the Pre- and Post-Regulation periods, the Transition period
(October 2000 – April 2003) is removed from the analysis following Lee et al. (2014).
In Table 9, I find that there is an incremental negative relation between recommendation
profitability (PFT) and political connections in the Post-Regulation period (POLITICAL*POST).
This negative relation holds under all four political connection measures (Candidates, Strength,
Ability, and Power). Furthermore, there is an incremental negative relation between
recommendation profitability (PFT) and forecast accuracy for politically connected firms in the
Post-Regulation period (ACCURACY*POLITICAL*POST). However, this incremental
32
association holds only with the Candidates and Strength measures of political connections.
Therefore, the results suggest that analysts have become less efficient in processing the firm
value implications of political connections as a result of the regulation changes. Furthermore,
analysts who follow politically connected firms have become more inefficient over time in
translating their earnings forecasts into profitable recommendations. In summary, this analysis
shows that regulatory reforms have had an impact on how analysts incorporate the effects of
political connections into recommendation profitability and earnings forecast accuracy.
IX.
CONCLUSION
This study examines whether financial analysts fully understand the economic
complexities of information regarding political connections. First, I investigate whether analysts
efficiently process information on political connections by analyzing whether analysts’ stock
recommendations are more or less profitable when the firms they cover have political
connections. Second, I investigate whether analysts effectively convert their earnings forecasts
into profitable stock recommendations for politically connected firms. Overall, I find strong
evidence that analysts’ stock recommendations are significantly less profitable when the firms
they follow have political connections. I also show that there is a significantly negative relation
between analysts’ earnings forecast accuracy and stock recommendation profitability for
politically connected firms. The main results of the paper are consistent after accounting for the
value relevance of earnings, which has been shown to drive the relation between forecast
accuracy and recommendation profitability (Ertimur et al. 2007). The results also hold after
controlling for earnings quality, a factor which has been shown to be an important characteristic
of politically connected firms. Additionally, I examine a subset of firms that either start up a new
33
political action committee (PAC) or shut down an existing PAC during the sample period. I find
that for firms that start up a new PAC (close down an existing PAC), recommendation
profitability is significantly lower in the years in which the PAC is in operation compared to
years before the PAC is created (after the PAC is discontinued). Overall, the main findings
suggest that analysts do not fully understand the economic complexities of political connections
and do not efficiently process information on political connections into their recommendations.
One limitation to the study is the endogeneity problem which arises as a result of a firm’s
choice to establish a political action committee and to make political contributions. Since it is
still not clear to researchers why firms make political contributions (Hart 2001), it is possible that
the determinants model for politically active firms contains correlated omitted variables which
induce measurement error in the matching of propensity scores. However, to the extent that I
capture the main factors affecting a firm’s decision to become politically active, the results for
the propensity-score matching model provide support for the main implications of the study.
34
REFERENCES
Aggarwal, R., Meschke, F. and T. Wang. 2012. Corporate Political Donations: Investment or
Agency? Business and Politics 14 (1) 1-40.
Agrawal, A. and C. Knoeber. 2001. Do some outside directors play a political role? Journal of
Law and Economics 44, 179-198.
Ansolabehere, S., J. M. de Figueiredo, and J. Snyder. 2003. Why is there so little money in U.S.
politics? Journal of Economic Perspectives 17 (1), 105-130.
Barniv, R., Hope, O.-K., Myring, M. J. and W.B. Thomas. 2009. Do Analysts Practice What They
Preach and Should Investors Listen? Effects of Recent Regulations. Accounting Review,
84:4, 1015-39.
Bertrand, M., Bombardini, M. and F. Trebbi. 2011. Is it Whom You Know or What You Know?
An Empirical Assessment of the Lobbying Process. Working Paper.
Bradshaw, M. T. 2004. How Do Analysts Use Their Earnings Forecasts in Generating Stock
Recommendations? Accounting Review, 79:1, 25-50.
Brown, J. and K. Drake. 2014. Network ties among low-tax firms. The Accounting Review 89 (2)
483-510.
Brown, S., Lo, K., Lys, T., 1999. Use of R2 in accounting research: measuring changes in value
relevance over the last four decades. Journal of Accounting and Economics 28, 83–115.
Chaney, P. K., Faccio, M. and D. Parsley. 2011. The quality of accounting information in
politically connected firms. Journal of Accounting and Economics, 51:1-2, 58-76.
Chen, C. and P.F. Chen. 2009. NASD Rule 2711 and Changes in Analysts' Independence in
Making Stock Recommendations. Accounting Review, 84:4, 1041-71.
Chen, C., Ding, Y. and C. Kim. 2010. High-Level Politically Connected Firms, Corruption, and
Analyst Forecast Accuracy. Journal of International Business Studies. 41, 1505-24.
Chen, S. and D. Matsumoto. 2006. Favorable versus unfavorable recommendations: The impact
on analyst access to management-provided information. Journal of Accounting Research
44 (4) 657-689.
Claessens, S., Feijen, E. and L. Laeven. 2008. Political Connections and Preferencial Access to
Finance: The Role of Campaign Contributions. Journal of Financial Economics 88: 554580.
Clement, M. B. 1999. Analyst forecast accuracy: Do ability, resources, and portfolio complexity
matter? Journal of Accounting & Economics, 27:3, 285-303.
35
Cohen, L., A. Frazzini, and C. Malloy. 2010. Sell-side school ties. The Journal of Finance, Vol.
65, No. 4, 1409-1437.
Cooper, M. J., Gulen, H. and A.V. Ovtchinnikov. 2010. Corporate Political Contributions and
Stock Returns. Journal of Finance, 65:2, 687-724.
Cooper, R. A., Day, T. E. and C.M. Lewis. 2001. Following the leader: A study of individual
analysts' earnings forecasts. Journal of Financial Economics, 61:3, 383-416.
Correira, M. 2014. Political Connections and SEC Enforcement. Journal of Accounting and
Economics 57 241-262.
Cull, R. and L. Xu. 2005. Institutions, ownership and finance: the determinants of profit
reinvestment among Chinese firms. Journal of Financial Economics 77, 117-146.
Ertimur, Y., Sunder, J. and S.V. Sunder. 2007. Measure for Measure: The Relation between
Forecast Accuracy and Recommendation Profitability of Analysts. Journal of Accounting
Research, 45:3, 567-606.
Faccio, M. 2006. Politically Connected Firms. American Economic Review 96: 369-386.
Faccio, M., Masulis, R. W. and J.J. McConnell. 2006. Political Connections and Corporate
Bailouts. Journal of Finance, 61:6, 2597-635.
Faccio, M. and D.C. Parsley. 2009. Sudden Deaths: Taking Stock of Geographic Ties. Journal of
Financial & Quantitative Analysis, 44:3, 683-718.
Fama, E. and K. French. 1997. Industry Costs of Equity. Journal of Financial Economics 43: 153193.
Fan, J., F. Guan, Z. Li, G. Yang. 2014. Relationship networks and earnings informativeness:
Evidence from corruption cases. Journal of Business, Finance & Accounting 41(7) & (8),
831-866.
Francis, J., R. LaFond, P.M. Olsson, and K. Schipper. 2004. Cost of equity and earnings attributes.
The Accounting Review 79, 967-1010.
Gross, D. A., and R. K. Goidel. 2003. The States of Campaign Finance. Columbus: Ohio State
University Press.
Goldman, E., Rocholl, J. and J. So. 2013. Politically Connected Boards of Directors and the
Allocation of Procurement Contracts. Review of Finance, 17: 1617-1648.
Gordon, S., H. Hafer, and D. Landa, 2007, Consumption or investment? On motivations
of political giving, Journal of Politics 69, 1057–107
36
Grier, K. and M. Munger. 1991. Committee assignments, constituent preferences, and campaign
contributions, Economic Inquiry 29, 24-43.
Grier, K., Munger, M. and B. Roberts. 1994. The determinants of industry political activity, 19781986, American Political Science Review 88, 911-926.
Hail, L., A. Tahoun, and C. Wang. 2014. Dividend payouts and information shocks. Journal of
Accounting Research 52, No. 2, 403-456.
Hall, B. and F. Wayman. 1990. Buying Time: Moneyed Interests and the Mobilization of Bias in
Congressional Committees. American Political Science Review. 84:3, 797-820.
Hart, D. M. 2001. Why do some firms give? Why do some firms give a lot?: High- tech pacs,
1977-1996, The Journal of Politics. 63, 1230-1249.
Holthausen, R., Watts, R., 2001. The relevance of value relevance. Journal of Accounting and
Economics 31, 3–75.
Horton, J., Y. Millo, and G. Serafeim. 2012. Resources or Power? Implications of Social Networks
on Compensation and Firm Performance. Journal of Business, Finance & Accounting, 39
(3) & (4), 399-426.
Houston, J., J. Liangliang, C. Lin, and Y. Ma. 2014. Political Connections and the Cost of Bank
Loans. Journal of Accounting Research 52, No. 1, 193-243.
Ishii, J. and Y. Xuan. 2014. Acquirer-target social ties and merger outcomes. Journal of Financial
Economics 112, 344-363.
Johnson, S. and T. Mitton. 2003. Cronyism and capital controls: evidence from Malaysia. Journal
of Financial Economics 67, 351-382.
Ke, B. and Y. Yu. 2006. The effect of issuing biased earnings forecasts on analysts’ access to
management and survival. Journal of Accounting Research 44 (5), 965-999.
Ke, B. and Y. Yu. 2009. Why Don’t Analysts Use Their Earnings Forecasts in Generating Stock
Recommendations? Working Paper, Pennsylvania State University.
Kim, C. and L. Zhang. 2014. Corporate political connections and tax aggressiveness,
Contemporary Accounting Research forthcoming.
Kroszner, R. S. and T. Stratmann. 1998. Interest-group competition and the organization of
Congress: Theory and evidence from financial services’ political action committees,
American Economic Review 88, 1163-1187.
Lee, E., N. Strong and Z. Zhu. 2014. Did Regulation Fair Disclosure, SOX, and Other Analyst
Regulations Reduce Security Mispricing? Journal of Accounting Research 52 (3) 733-774.
37
Leuz, C. and F. Oberholzer-Gee. 2006. Political relationships, global financing, and corporate
transparency: Evidence from Indonesia. Journal of Financial Economics, 81:2, 411-39.
Loh, R. K. and G.M. Mian. 2006. Do accurate earnings forecasts facilitate superior investment
recommendations? Journal of Financial Economics, 80:2, 455-83.
Mohanram, P. S. and S.V. Sunder. 2006. How has regulation FD affected the operations of
financial analysts? Contemporary Accounting Research, 23:2, 491-525.
Myers, B. W. 2005. Corporate political activity and asset pricing, Available from URL:
http://ssrn.com/abstract=950104
Parsons, L.S. 2000. Reducing bias in a propensity score matched-pair sample using greedy
matching techniques. Proceedings of the twenty-fifth annual SAS users group internal
conference, Cary, NC: SAS Institute Inc. 1166 – 1171.
Schipper, K. 1991. Analysts' Forecasts. Accounting Horizons: 105-21.
Simon, A. and A. Curtis. 2011. The Use of Earnings Forecasts in Stock Recommendations: Are
Accurate Analysts More Consistent? Journal of Business, Finance & Accounting. 38:1,
119-144.
Snyder, J. 1990. Campaign Contributions as Investments: The House of Representatives, 19801986. Journal of Political Economy. December, 98:6, 1195-227.
Snyder, J. 1992. Long-Term Investing in Politicians, Or Give Early, Give Often. Journal of Law
and Economics. April, 35:1, 15-44.
Stigler, G. J. 1971. The theory of economic regulation. Bell Journal of Economics and
Management Science, 2(1): 3-21.
Tullock, G. 1972. The Purchase of Politicians. Western Economic Journal. 10, 354-55.
Wu, W., C. Wu, C. Zhou and J. Wu. 2012. Political connections, tax benefits and firm
performance: Evidence from China. Journal of Accounting and Public Policy, 31:277-300.
38
Appendix A
Variable Definitions
POLITICALCandidates
𝐽
∑ 𝐶𝑎𝑛𝑑𝑖𝑑𝑎𝑡𝑒𝑠𝑗𝑡
𝑗=1
Candidatesjt is equal to one if the firm has contributed to
candidate j in year t. POLITICALCandidates equals 1 if greater than
industry-year mean, and 0 otherwise.
POLITICALStrength
𝐽
∑ 𝐶𝑎𝑛𝑑𝑖𝑑𝑎𝑡𝑒𝑠𝑗𝑡 × 𝐼𝑗𝑡 ×
𝑗=1
𝑁𝐶𝑉𝑗𝑡
× 𝑟𝑒𝑙𝑙𝑒𝑛𝑔𝑡ℎ𝑗𝑡
𝑁𝑂𝑉𝑗𝑡
Candidatesjt is equal to one if the firm has contributed to
candidate j in year t, and 0 otherwise. Ijt is equal to one if the
candidate is incumbent at time t, and 0 otherwise. NCVjt (NOVjt)
is the number of votes that candidate j’s party (opposing party)
holds in office at time t, and rellengthjt is the number of
continuous months of the firm-candidate relationship.
POLITICALStrength equals 1 if greater than industry-year mean,
and 0 otherwise.
POLITICALAbility
𝐽
∑ 𝐻𝑜𝑚𝑒𝐶𝑎𝑛𝑑𝑖𝑑𝑎𝑡𝑒𝑗𝑡 × 𝐼𝑗𝑡 ×
𝑗=1
𝑁𝐶𝑉𝑗𝑡
𝑁𝑂𝑉𝑗𝑡
HomeCandidatejt is equal to one if the firm has contributed to
candidate j in year t and the firm is headquartered in the same
state in which the candidate is running for office, and 0
otherwise. Ijt is equal to one if the candidate is incumbent at
time t, and 0 otherwise. NCVjt (NOVjt) is the number of votes
that candidate j’s party (opposing party) holds in office at time
t. POLITICALAbility equals 1 if greater than industry-year mean,
and 0 otherwise.
POLITICALPower
𝐽
𝑀
𝑗=1
𝑚=1
𝑁𝐶𝑉𝑗𝑡
𝑅𝑎𝑛𝑘𝑚𝑡
∑ 𝐶𝑎𝑛𝑑𝑖𝑑𝑎𝑡𝑒𝑠𝑗𝑡 × 𝐼𝑗𝑡 ×
× [∑
]
𝑁𝑂𝑉𝑗𝑡
𝑀𝑒𝑑𝑖𝑎𝑛 𝑅𝑎𝑛𝑘𝑚𝑡
𝑗
Candidatesjt is equal to one if the firm has contributed to
candidate j in year t, and 0 otherwise. Ijt is equal to one if the
candidate is incumbent at time t, and 0 otherwise. NCVjt (NOVjt)
is the number of votes that candidate j’s party (opposing party)
holds in office at time t. Rankmt is the reciprocal of candidate j’s
rank on committee m (where rank =1 for the most important
member, rank = 2 for the next important member, etc.). Median
Rankmt is the median number of members on committee m.
POLITICALPower equals 1 if greater than industry-year mean,
and 0 otherwise.
39
PFT
ACCURACY
VR
FIRMEXP
BSIZE
N_FIRMS
REC_FREQ
LFR
COV
REG
DACC
ACTIVE
SIZE
SALES
EMPLOYEES
BUS_SEG
GEO_SEG
BM
LEV
CFO
MARKET_SHARE
HERF
INDUSTRY_ACTIVE
The market-adjusted return to recommendation k made by
analyst i for firm j. The buy-and-hold return is computed from
the day before the recommendation date until the earlier of 30
days or 2 days before the recommendation is revised or
reiterated. For buy (hold and sell) recommendations, $1 is
invested (shorted) in the recommended stock.
The difference between the absolute forecast error for analyst i
for firm j at time t scaled by the mean absolute forecast error for
firm j at time t times negative one.
Value relevance is the R2 from the following regression
estimated for each firm over rolling 10-year windows:
RET = α0 + α1EARN + α2ΔEARN + ε, where RET is the firm's
15-month size-adjusted return ending three months after the
fiscal year-end, EARN is income before extraordinary items
scaled by market value at the end of the prior fiscal year, and
ΔEARN is the change in income before extraordinary items
scaled by market value at the end of the prior fiscal year.
The number of years through year t for which analyst i supplied
forecasts for firm j.
The logarithm of the number of analysts employed by the
brokerage firm that analyst i works for during year t.
The number of firms for which analyst i issues annual forecasts
during the year in which the recommendation is made.
The number of recommendations analyst i issues for firm j
during year t.
The leader-follower ratio is the cumulative number of days by
which the preceding two forecasts lead forecast k divided by the
cumulative number of days by which the subsequent two
forecasts follow forecast k.
The number of analysts who issue forecasts and stock
recommendations for firm j during year t
Equals 1 if the firm operates in the financial services industry
(one-digit SIC code 6) or in the utilities industry (two-digit SIC
code 49), and 0 otherwise.
Discretionary accruals calculated using the Modified Jones
method within each 2 digit SIC code each year.
Equals 1 if the firm has a registered PAC, 0 otherwise.
Natural log of price times shares outstanding.
Natural log of sales.
Natural log of number of employees in millions.
The number of business segments.
The number of geographic segments.
Book-to-market ratio is book equity divided by market value
The sum of long-term debt and debt in current liabilities,
divided by total assets
Operating cash flow divided by total assets
Sales divided by total industry sales.
Herfindahl index of industry concentration.
Number of firms in a firm's industry with an established PAC.
40
Appendix B
First-Stage Probit Model of Firm's Decision to be Politically Active
ACTIVE = λ 0 + λ 1 SIZE + λ 2 SALES + λ 3 EMPLOYEES + λ 4 BUS_SEG + λ 5 GEO_SEG
+ λ 6 BM + λ 7 LEV + λ 8 CFO + λ 9 MARKET_SHARE + λ 10 HERF + λ 11 REG
+ λ 12 INDUSTRY_ACTIVE + ε
Intercept
SIZE
SALES
EMPLOYEES
BUS_SEG
GEO_SEG
BM
LEV
CFO
MARKET_SHARE
HERF
REG
INDUSTRY_ACTIVE
R2
Coefficient
-7.2223 ***
0.2635 ***
0.3277 ***
0.2056 ***
0.0879 ***
-0.0810 ***
0.0066 ***
1.1788 ***
0.2534 ***
1.9192 ***
-0.4781 ***
0.6851 ***
0.0028 ***
Std. Error
0.0521
0.0066
0.0094
0.0073
0.0036
0.0032
0.0022
0.0314
0.0747
0.1650
0.1717
0.0198
0.0001
p -value
<.0001
<.0001
<.0001
<.0001
<.0001
<.0001
0.0032
<.0001
0.0007
<.0001
0.0054
<.0001
<.0001
35.95%
***, **, * indicate significance at the 1%, 5%, and 10% levels. There are 216,194 observations
estimated over 1993 - 2012. ACTIVE equals 1 if the firm has a registered political action committee, 0
otherwise. SIZE is the log of price times shares outstanding (prcc_f * csho). SALES is the natural
logarithm of sales. EMPLOYEES is the natural logarithm of the number of employees in millions (emp).
BUS_SEG is the number of business segments. GEO_SEG is the number of geographic segments.
BM, book-to-market, is stockholders' book equity divided by market value [ceq/(prcc_f*csho)]. LEV
is the sum of long-term debt and debt in current liabilities, divided by total assets [(dltt + dlc)/at]. CFO
is operating cash flow divided by total assets [(oancf - xidoc)/at]. MARKET_SHARE is sales divided
by total industry sales. HERF is the Herfindahl index of industry concentration computed with net
sales. REG equals 1 if the firm operates in the financial services industry (one-digit SIC code 6) or in
the utilities industry (two-digit SIC code 49), and 0 otherwise. INDUSTRY_ACTIVE is the number of
firms in a firm's industry with an established political action committee. All continuous variables are
winsorized at 1 and 99 percent to mitigate outliers.
41
Table 1
Descriptive Statistics
Financial Analyst Characteristics
Panel A: Descriptive Statistics for the
Variable
N
Mean
PFT
220,843
0.058
ACCURACY
220,843
-0.170
VR
220,843
0.299
UFIRMEXP
220,843
3.706
UBSIZE
220,843
65.469
UN_FIRMS
220,843
16.264
UREC_FREQ 220,843
1.833
LFR
220,843
2.309
UCOV
220,843
14.386
REG
220,843
0.192
DACC
168,969
-0.027
Pooled Sample
Std. Dev.
10%
0.130
-0.067
1.098
-1.511
0.238
0.041
3.387
1.000
63.459
9.000
10.201
7.000
1.549
1.000
4.335
0.063
8.438
5.000
0.394
0.000
0.128
-0.130
25%
0.005
-0.538
0.106
1.000
19.000
10.000
1.000
0.320
8.000
0.000
-0.069
Median
0.044
0.109
0.237
3.000
43.000
15.000
1.000
0.896
13.000
0.000
-0.019
Panel B: Descriptive Statistics by Political Involvement
Politically Connected
Non-Politically Connected
POLITICAL = 1
POLITICAL = 0
Variable
Mean
Median Std Dev
Mean
Median Std Dev
PFT
0.042
0.033
0.099
0.060
0.046
0.133
ACCURACY
0.008
0.170
0.815
-0.191
0.100
1.125
VR
0.268
0.219
0.217
0.302
0.239
0.240
UFIRMEXP
4.679
3.000
4.167
3.589
2.000
3.262
UBSIZE
72.645
55.000
66.139
64.611
42.000
63.076
UN_FIRMS
17.099
15.000
11.960
16.164
15.000
9.965
UREC_FREQ
1.964
1.000
2.124
1.818
1.000
1.465
LFR
2.159
1.000
3.759
2.327
0.886
4.399
UCOV
21.309
20.000
7.922
13.558
12.000
8.111
REG
0.249
0.000
0.432
0.185
0.000
0.389
DACC
-0.020
-0.013
0.076
-0.028
-0.020
0.132
75%
0.105
0.618
0.443
5.000
96.000
20.000
2.000
2.318
20.000
0.000
0.022
90%
0.193
0.872
0.663
8.000
140.000
26.000
3.000
5.400
26.000
1.000
0.073
Difference in Means
-0.018 ***
0.199 ***
-0.034 ***
1.090 ***
8.034 ***
0.935 ***
0.146 ***
-0.168 ***
7.750 ***
0.063 ***
0.008 ***
***, **, * indicate significance at the 1%, 5%, and 10% levels. There are 220,843 observations estimated over 1993 - 2012.
POLITICAL equals 1 if politically connected (Candidates measure #1), and 0 otherwise. PFT is the buy-and-hold, abnormal
return to the analyst's recommendation computed from the day before the recommendation date until the earlier of 30 days
or 2 days before the recommendation is revised or reiterated. For buy (hold and sell) recommendations, $1 is invested
(shorted) in the recommended stock. ACCURACY equals negative 1 times the difference between the analyst's absolute
forecast error and the mean absolute forecast error for all analysts scaled by the mean absolute forecast error for all
analysts. VR, value relevance, is the R-squared from the following regression estimated for each firm over rolling 10-year
windows: RET = α0 + α1 EARN + α2 ΔEARN + ε, where RET is the firm's 15-month return ending three months after the fiscal
year-end, EARN is income before extraordinary items scaled by market value at the end of the prior fiscal year, and ΔEARN
is the change in income before extraordinary items scaled by market value at the end of the prior fiscal year. DACC is
discretionary accruals calculated using the Modified Jones method within each 2 digit SIC code each year. UFIRMEXP is
the number of years for which the analyst supplied forecasts for the firm. UBSIZE is the number of analysts employed by
the analyst's brokerage firm. UN_FIRMS is the number of firms for which the analyst issues annual forecasts during the
year. UREC_FREQ is the number of recommendations the analyst issues for the firm during the year. LFR, leader-follower
ratio, is the cumulative number of days by which the preceding two forecasts lead the analyst's forecast divided by the
cumulative number of days by which the subsequent two forecasts follow the analyst's forecast. UCOV is the number of
analyst following during the year. REG equals 1 if the firm is in the financial services industry (one-digit SIC code 6) or
utilities industry (two-digit SIC code 49), and 0 otherwise.
42
Table 2
Descriptive Statistics for Firm Political Contributions
Panel A: Political Contributions Statistics (actuals)
Variable
Mean
Total Contributions
$125,100
Number of Contributions
95
Number of Candidates
50
Min
$5
1
1
Q1
$11,500
10
8
Median
Q3
Max
$39,700 $126,250 $3,827,600
32
104
2,077
24
64
570
Panel B: Amount of Political Contributions by Largest Industry Contributors ($000's)
1993Industry
1994
1. Financial
$6,971
2. Telecommunications
5,482
3. Utilities
3,949
4. Transportation
2,272
5. Pharmaceutical
1,722
6. Energy
3,522
7. Electronic Equipment 1,881
8. Retail
1,118
9. Business Services
665
10. Machinery
1,180
Total
$28,763
19951996
$8,554
6,420
4,182
3,233
2,144
3,579
2,183
1,699
488
1,687
$34,170
19971998
$8,312
6,507
4,512
3,462
2,600
3,555
1,925
1,723
795
1,152
$34,542
Election Cycle
1999200120032000
2002
2004
$9,435
$10,083
$11,099
6,910
7,278
6,347
5,847
6,869
6,634
5,978
6,381
6,903
3,772
4,170
5,116
3,057
2,604
3,286
2,427
2,586
3,699
2,366
3,186
3,899
1,972
2,614
3,490
962
1,186
1,491
$42,725 $46,958
$51,965
43
20052006
$11,802
6,733
7,356
6,799
6,980
3,469
4,326
3,595
3,915
2,157
$57,133
20072008
$13,642
7,677
7,867
7,312
8,106
4,466
5,441
3,820
4,494
2,980
$65,806
20092010
$10,918
7,767
8,726
6,144
8,410
5,359
5,881
4,166
4,576
3,023
$64,969
20112012
Total
$12,953 $103,769
7,735
68,854
8,404
64,347
5,388
53,873
8,179
51,200
6,943
39,839
6,272
36,621
4,444
30,015
4,866
27,878
2,805
18,624
$67,990 $495,021
Table 2 (cont.)
Descriptive Statistics for Firm Political Contributions
Panel C: Number of Candidates for Political Contributions by Largest Industry Contributors (actuals)
Industry
1. Financial
2. Telecommunications
3. Utilities
4. Transportation
5. Retail
6. Electronic Equipment
7. Pharmaceutical
8. Energy
9. Business Services
10. Machinery
Total
19931994
786
706
656
511
461
444
410
564
391
424
5,353
19951996
758
680
642
538
485
464
459
520
323
446
5,315
19971998
645
605
568
475
434
434
432
469
331
340
4,733
Election Cycle
199920012000
2002
643
658
587
622
580
578
545
537
451
489
416
432
414
394
431
383
474
455
294
317
4,835
4,865
20032004
618
583
573
544
474
447
402
405
472
324
4,842
20052006
585
547
548
541
426
420
464
393
472
341
4,737
20072008
591
584
581
572
467
481
503
411
504
418
5,112
20092010
637
634
627
596
476
494
502
459
505
456
5,386
20112012
629
621
600
582
461
481
488
428
520
429
5,239
Total
6,550
6,169
5,953
5,441
4,624
4,513
4,468
4,463
4,447
3,789
50,417
This table reports the total federal political contributions for the top 10 donating industries over the period 1993-2012. Data is obtained from the Federal Election
Commission (FEC) through the Center for Responsive Politics on political contributions to House, Senate, and Presidential elections. Data are excluded from all
noncorporate contributions, contributions from private firms and subsidiaries of foreign firms. The sample includes 851,813 contributions made by 1,102 unique
firms. The table reports firm contribution characteristics per election cycle.
44
Table 3
Pearson/Spearman Correlations
POLITICAL
POLITICAL
PFT
-0.053
ACCURACY
0.030
VR
-0.038
FIRMEXP
0.077
BSIZE
0.045
N_FIRMS
0.011
REC_FREQ
0.009
LFR
0.024
COV
0.281
REG
0.050
DACC
0.025
0.012
0.019
0.020
0.034
-0.011
0.031
0.020
-0.065
-0.051
-0.020
-0.008
-0.013
0.012
-0.022
-0.003#
0.057
0.005^
-0.014
-0.001#
-0.020
-0.017
-0.016
0.001#
-0.004!
-0.078
-0.004!
-0.029
0.043
0.254
0.113
-0.003#
0.160
0.006
0.019
0.087
-0.015
0.057
0.090
0.029
0.007
0.035
0.000#
0.046
0.171
0.058
-0.001#
0.025
-0.028
0.000#
0.052
0.011
0.009
-0.034
-0.098
PFT
-0.043
ACCURACY
0.056
0.014
VR
-0.044
0.026
-0.017
FIRMEXP
0.082
0.018
-0.020
-0.029
BSIZE
0.038
0.028
0.009
-0.016
0.039
N_FIRMS
0.019
-0.006
-0.022
-0.018
0.245
0.072
REC_FREQ
0.017
0.023
0.011
0.000#
0.110
-0.036
0.044
LFR
-0.012
0.037
0.045
0.003#
0.017
0.036
-0.001#
0.004!
COV
0.270
-0.061
-0.007
-0.083
0.164
0.095
0.058
0.029
-0.040
REG
0.050
-0.034
-0.016
-0.006
0.005!
0.032
0.162
-0.030
0.007
-0.036
DACC
0.018
-0.002#
-0.006^
-0.011
0.018
0.000#
0.031
-0.007
-0.003#
-0.077
0.070
0.042
Pearson (Spearman) correlations are below (above) the diagonal. All correlations are significant at the 1% level except those with a "#" which are not significant, "!" are significant at the
10% level, and "^" are significant at the 5% level. There are 220,843 observations estimated over 1993 - 2012. POLITICAL equals 1 if politically connected (Candidates measure #1), and 0
otherwise. PFT is the buy-and-hold, abnormal return to the analyst's recommendation computed from the day before the recommendation date until the earlier of 30 days or 2 days
before the recommendation is revised or reiterated. For buy (hold and sell) recommendations, $1 is invested (shorted) in the recommended stock. ACCURACY equals negative 1 times
the difference between the analyst's absolute forecast error and the mean absolute forecast error for all analysts scaled by the mean absolute forecast error for all analysts. VR, value
relevance, is the R-squared from the following regression estimated for each firm over rolling 10-year windows: RET = α0 + α1 EARN + α2 ΔEARN + ε, where RET is the firm's 15-month
return ending three months after the fiscal year-end, EARN is income before extraordinary items scaled by market value at the end of the prior fiscal year, and ΔEARN is the change in
income before extraordinary items scaled by market value at the end of the prior fiscal year. FIRMEXP is the log of the number of years for which the analyst supplied forecasts for the
firm. BSIZE is the log of the number of analysts employed by the analyst's brokerage firm. N_FIRMS is the log of the number of firms for which the analyst issues annual forecasts
during the year. REC_FREQ is the log of the number of recommendations the analyst issues for the firm during the year. LFR, leader-follower ratio, is the cumulative number of days by
which the preceding two forecasts lead the analyst's forecast divided by the cumulative number of days by which the subsequent two forecasts follow the analyst's forecast. COV is the
log of the number of analyst following during the year. REG equals 1 if the firm is in the financial services industry (one-digit SIC code 6) or utilities industry (two-digit SIC code 49), and
0 otherwise.
45
Table 4
Relation between Accuracy and Profitability - Benchmark Model
PFT = α 0 + α 1 ACCURACY + α 2 VR + α 3 FIRMEXP + α 4 BSIZE + α 5 N_FIRMS
+ α 6 REC_FREQ + α 7 LFR + α 8 COV + α 9 REG + ε
Intercept
ACCURACY
VR
FIRMEXP
BSIZE
N_FIRMS
REC_FREQ
LFR
COV
REG
2
Adj. R
Year Dummies
Coefficient
0.0596 ***
0.0007 **
0.0129 ***
0.0041 ***
0.0032 ***
0.0004
0.0047 ***
0.0010 ***
-0.0118 ***
-0.0123 ***
Std. Error
0.0020
0.0003
0.0012
0.0003
0.0002
0.0005
0.0005
0.0001
0.0004
0.0007
t -statistic
30.44
2.41
11.11
11.94
13.32
0.78
8.73
14.99
-27.10
-17.39
p- value
<.0001
0.0159
<.0001
<.0001
<.0001
0.4351
<.0001
<.0001
<.0001
<.0001
2.11%
Yes
***, **, * indicate significance at the 1%, 5%, and 10% levels. There are 220,843 observations
estimated over 1993 - 2012. PFT is the buy-and-hold, abnormal return to the analyst's recommendation
computed from the day before the recommendation date until the earlier of 30 days or 2 days before the
recommendation is revised or reiterated. For buy (hold and sell) recommendations, $1 is invested
(shorted) in the recommended stock. ACCURACY equals negative 1 times the difference between the
analyst's absolute forecast error and the mean absolute forecast error for all analysts scaled by the
mean absolute forecast error for all analysts. VR, value relevance, is the R-squared from the following
regression estimated for each firm over rolling 10-year windows: RET = α0 + α1 EARN + α2 ΔEARN + ε,
where RET is the firm's 15-month return ending three months after the fiscal year-end, EARN is income
before extraordinary items scaled by market value at the end of the prior fiscal year, and ΔEARN is the
change in income before extraordinary items scaled by market value at the end of the prior fiscal year.
FIRMEXP is the log of the number of years for which the analyst supplied forecasts for the firm. BSIZE
is the log of the number of analysts employed by the analyst's brokerage firm. N_FIRMS is the log of
the number of firms for which the analyst issues annual forecasts during the year. REC_FREQ is the
log of the number of recommendations the analyst issues for the firm during the year. LFR, leaderfollower ratio, is the cumulative number of days by which the preceding two forecasts lead the
analyst's forecast divided by the cumulative number of days by which the subsequent two forecasts
follow the analyst's forecast. COV is the log of the number of analyst following during the year. REG
equals 1 if the firm is in the financial services industry (one-digit SIC code 6) or utilities industry (twodigit SIC code 49), and 0 otherwise.
46
Table 5
The Effect of Political Connections on Stock Recommendation Profitability and Forecast Accuracy
Panel A: Political Connections and Stock Recommendation Profitability
PFT = β 0 + β 1 POLITICAL + β 2 ACCURACY + β 3 VR + β 4 FIRMEXP + β 5 BSIZE + β 6 N_FIRMS + β 7 REC_FREQ
+ β 8 LFR + β 9 COV + β 10 REG + ε
Intercept
POLITICAL
ACCURACY
VR
FIRMEXP
BSIZE
N_FIRMS
REC_FREQ
LFR
COV
REG
2
Adj. R
Year Dummies
Candidates
Coefficient
S.E.
0.0556 *** 0.0020
-0.0122 *** 0.0009
0.0007 **
0.0003
0.0126 *** 0.0012
0.0044 *** 0.0003
0.0033 *** 0.0002
0.0003
0.0005
0.0047 *** 0.0005
0.0009 *** 0.0001
-0.0103 *** 0.0005
-0.0117 *** 0.0007
2.18%
Yes
Strength
Coefficient
S.E.
0.0557 *** 0.0020
-0.0128 *** 0.0009
0.0007 **
0.0003
0.0125 *** 0.0012
0.0044 *** 0.0003
0.0033 *** 0.0002
0.0002
0.0005
0.0047 *** 0.0005
0.0009 *** 0.0001
-0.0103 *** 0.0005
-0.0117 *** 0.0007
2.19%
Yes
Ability
Coefficient
0.0625 ***
-0.0059 ***
0.0005 *
0.0127 ***
0.0043 ***
0.0033 ***
0.0002
0.0048 ***
0.0010 ***
-0.0109 ***
-0.0121 ***
2.13%
Yes
47
S.E.
0.0020
0.0008
0.0003
0.0012
0.0003
0.0002
0.0005
0.0005
0.0001
0.0005
0.0007
Power
Coefficient
S.E.
0.0556 *** 0.0020
-0.0124 *** 0.0009
0.0007 ** 0.0003
0.0124 *** 0.0012
0.0044 *** 0.0003
0.0033 *** 0.0002
0.0002
0.0005
0.0047 *** 0.0005
0.0009 *** 0.0001
-0.0102 *** 0.0005
-0.0117 *** 0.0007
2.19%
Yes
Table 5 (cont.)
The Effect of Political Connections on Stock Recommendation Profitability and Forecast Accuracy
Panel B: Political Connections and the Relation between Accuracy and Profitability
PFT = β 0 + β 1 POLITICAL + β 2 ACCURACY + β 3 ACCURACY*POLITICAL + β 4 VR + β 5 FIRMEXP + β 6 BSIZE
+ β 7 N_FIRMS + β 8 REC_FREQ + β 9 LFR + β 10 COV + β 11 REG + ε
Intercept
POLITICAL
ACCURACY
ACCURACY*POLITICAL
VR
FIRMEXP
BSIZE
N_FIRMS
REC_FREQ
LFR
COV
REG
R2
Year Dummies
Candidates
Coefficient
S.E.
0.0559 *** 0.0020
-0.0122 *** 0.0009
0.0009 *** 0.0003
-0.0036 *** 0.0011
0.0126 *** 0.0012
0.0044 *** 0.0003
0.0033 *** 0.0002
0.0003
0.0005
0.0047 *** 0.0005
0.0009 *** 0.0001
-0.0103 *** 0.0005
-0.0117 *** 0.0007
2.18%
Yes
Strength
Coefficient
S.E.
0.0560 *** 0.0020
-0.0128 *** 0.0009
0.0009 *** 0.0003
-0.0035 *** 0.0011
0.0125 *** 0.0012
0.0044 *** 0.0003
0.0033 *** 0.0002
0.0002
0.0005
0.0047 *** 0.0005
0.0009 *** 0.0001
-0.0103 *** 0.0005
-0.0117 *** 0.0007
2.19%
Yes
Ability
Coefficient
0.0614 ***
-0.0060 ***
0.0016 ***
-0.0021 ***
0.0126 ***
0.0043 ***
0.0033 ***
0.0002
0.0048 ***
0.0009 ***
-0.0109 ***
-0.0122 ***
2.14%
Yes
S.E.
0.0020
0.0008
0.0004
0.0005
0.0012
0.0003
0.0002
0.0005
0.0005
0.0001
0.0005
0.0007
Power
Coefficient
S.E.
0.0558 *** 0.0020
-0.0124 *** 0.0009
0.0009 *** 0.0003
-0.0026 ** 0.0011
0.0125 *** 0.0012
0.0044 *** 0.0003
0.0033 *** 0.0002
0.0002
0.0005
0.0047 *** 0.0005
0.0009 *** 0.0001
-0.0102 *** 0.0005
-0.0117 *** 0.0007
2.19%
Yes
***, **, * indicate significance at the 1%, 5%, and 10% levels. There are 220,843 observations estimated over 1993 - 2012. The four columns in this table use 4
different proxies for political connections (POLITICAL): Candidates, Strength, Ability, and Power. POLITICAL equals 1 if politically connected, 0 otherwise.
PFT is the buy-and-hold, abnormal return to the analyst's recommendation computed from the day before the recommendation date until the earlier of 30 days
or 2 days before the recommendation is revised or reiterated. For buy (hold and sell) recommendations, $1 is invested (shorted) in the recommended stock.
ACCURACY equals negative 1 times the difference between the analyst's absolute forecast error and the mean absolute forecast error for all analysts scaled by
the mean absolute forecast error for all analysts. VR, value relevance, is the R-squared from the following regression estimated for each firm over rolling 10-year
windows: RET = α0 + α1 EARN + α2 ΔEARN + ε, where RET is the firm's 15-month return ending three months after the fiscal year-end, EARN is income before
extraordinary items scaled by market value at the end of the prior fiscal year, and ΔEARN is the change in income before extraordinary items scaled by market
value at the end of the prior fiscal year. FIRMEXP is the log of the number of years for which the analyst supplied forecasts for the firm. BSIZE is the log of the
number of analysts employed by the analyst's brokerage firm. N_FIRMS is the log of the number of firms for which the analyst issues annual forecasts during
the year. REC_FREQ is the log of the number of recommendations the analyst issues for the firm during the year. LFR, leader-follower ratio, is the cumulative
number of days by which the preceding two forecasts lead the analyst's forecast divided by the cumulative number of days by which the subsequent two
forecasts follow the analyst's forecast. COV is the log of the number of analyst following during the year. REG equals 1 if the firm is in the financial services
industry (one-digit SIC code 6) or utilities industry (two-digit SIC code 49), and 0 otherwise.
48
Table 6
Accounting Quality, Political Connections, and Stock Recommendation Profitability
Panel A: The Effect of Accounting Quality on Political Connections and Recommendation Profitability
PFT = γ 0 + γ 1 POLITICAL + γ 2 ACCURACY + γ 3 VR + γ 4 DACC + γ 5 FIRMEXP + γ 6 BSIZE + γ 7 N_FIRMS
+ γ 8 REC_FREQ + γ 9 LFR + γ 10 COV + γ 11 REG + ε
Intercept
POLITICAL
ACCURACY
VR
DACC
FIRMEXP
BSIZE
N_FIRMS
REC_FREQ
LFR
COV
REG
Adj. R2
Year Dummies
Candidates
Coefficient
S.E.
0.0589 *** 0.0024
-0.0105 *** 0.0012
0.0013 *** 0.0003
0.0091 *** 0.0014
-0.0054 **
0.0023
0.0043 *** 0.0004
0.0036 *** 0.0003
0.0010 *
0.0005
0.0050 *** 0.0006
0.0010 *** 0.0001
-0.0112 *** 0.0005
-0.0187 *** 0.0015
1.83%
Yes
Strength
Coefficient
S.E.
0.0589 *** 0.0024
-0.0111 *** 0.0012
0.0012 *** 0.0003
0.0090 *** 0.0014
-0.0024
0.0026
0.0043 *** 0.0004
0.0036 *** 0.0003
0.0010 *
0.0005
0.0050 *** 0.0006
0.0010 *** 0.0001
-0.0112 *** 0.0005
-0.0185 *** 0.0015
1.82%
Yes
Ability
Coefficient
0.0672 ***
-0.0084 ***
0.0013 ***
0.0090 ***
-0.0026
0.0043 ***
0.0036 ***
0.0010 *
0.0050 ***
0.0009 ***
-0.0111 ***
-0.0193 ***
1.82%
Yes
49
S.E.
0.0024
0.0010
0.0003
0.0014
0.0026
0.0004
0.0003
0.0005
0.0006
0.0001
0.0005
0.0014
Power
Coefficient
S.E.
0.0589 *** 0.0024
-0.0104 *** 0.0012
0.0013 *** 0.0003
0.0089 *** 0.0014
-0.0026
0.0026
0.0043 *** 0.0004
0.0036 *** 0.0003
0.0010 *
0.0005
0.0050 *** 0.0006
0.0009 *** 0.0001
-0.0112 *** 0.0005
-0.0186 *** 0.0015
1.83%
Yes
Table 6 (cont.)
Accounting Quality, Political Connections, and Stock Recommendation Profitability
Panel B: The Effect of Accounting Quality on Political Connections and the Relation between Accuracy and Profitability
PFT = γ 0 + γ 1 POLITICAL + γ 2 ACCURACY + γ 3 ACCURACY*POLITICAL + γ 4 VR + γ 5 DACC + γ 6 FIRMEXP + γ 7 BSIZE
+ γ 8 N_FIRMS + γ 9 REC_FREQ + γ 10 LFR + γ 11 COV + γ 12 REG + ε
Intercept
POLITICAL
ACCURACY
ACCURACY*POLITICAL
VR
DACC
FIRMEXP
BSIZE
N_FIRMS
REC_FREQ
LFR
COV
REG
Adj. R2
Year Dummies
Candidates
Coefficient
S.E.
0.0593 *** 0.0024
-0.0105 *** 0.0012
0.0016 *** 0.0003
-0.0058 *** 0.0014
0.0092 *** 0.0014
-0.0054 **
0.0023
0.0043 *** 0.0004
0.0036 *** 0.0003
0.0010 *
0.0005
0.0050 *** 0.0006
0.0009 *** 0.0001
-0.0112 *** 0.0005
-0.0187 *** 0.0015
1.83%
Yes
Strength
Coefficient
0.0593 ***
-0.0112 ***
0.0016 ***
-0.0055 ***
0.0090 ***
-0.0024
0.0043 ***
0.0036 ***
0.0010 *
0.0051 ***
0.0010 ***
-0.0112 ***
-0.0185 ***
1.84%
Yes
S.E.
0.0024
0.0012
0.0003
0.0014
0.0014
0.0026
0.0004
0.0003
0.0005
0.0006
0.0001
0.0005
0.0015
Ability
Coefficient
0.0662 ***
-0.0083 ***
0.0021 ***
-0.0016 **
0.0089 ***
-0.0026
0.0043 ***
0.0036 ***
0.0010 *
0.0050 ***
0.0009 ***
-0.0111 ***
-0.0193 ***
1.82%
Yes
S.E.
0.0025
0.0010
0.0005
0.0007
0.0014
0.0026
0.0004
0.0003
0.0005
0.0006
0.0001
0.0005
0.0014
Power
Coefficient
0.0592 ***
-0.0104 ***
0.0015 ***
-0.0047 ***
0.0090 ***
-0.0027
0.0043 ***
0.0036 ***
0.0010 *
0.0050 ***
0.0009 ***
-0.0112 ***
-0.0186 ***
S.E.
0.0024
0.0012
0.0003
0.0014
0.0014
0.0026
0.0004
0.0003
0.0005
0.0006
0.0001
0.0005
0.0015
1.83%
Yes
***, **, * indicate significance at the 1%, 5%, and 10% levels. There are 168,969 observations estimated over 1993 - 2012. The four columns in this table use 4 different proxies for
political connections (POLITICAL): Candidates, Strength, Ability, and Power. POLITICAL equals 1 if politically connected, 0 otherwise. PFT is the buy-and-hold, abnormal return
to the analyst's recommendation computed from the day before the recommendation date until the earlier of 30 days or 2 days before the recommendation is revised or reiterated.
For buy (hold and sell) recommendations, $1 is invested (shorted) in the recommended stock. ACCURACY equals negative 1 times the difference between the analyst's absolute
forecast error and the mean absolute forecast error for all analysts scaled by the mean absolute forecast error for all analysts. VR, value relevance, is the R-squared from the
following regression estimated for each firm over rolling 10-year windows: RET = α0 + α1 EARN + α2 ΔEARN + ε, where RET is the firm's 15-month return ending three months after
the fiscal year-end, EARN is income before extraordinary items scaled by market value at the end of the prior fiscal year, and ΔEARN is the change in income before extraordinary
items scaled by market value at the end of the prior fiscal year. DACC is discretionary accruals calculated using the Modified Jones method within each 2 digit SIC code each
year. FIRMEXP is the log of the number of years for which the analyst supplied forecasts for the firm. BSIZE is the log of the number of analysts employed by the analyst's
brokerage firm. N_FIRMS is the log of the number of firms for which the analyst issues annual forecasts during the year. REC_FREQ is the log of the number of recommendations
the analyst issues for the firm during the year. LFR, leader-follower ratio, is the cumulative number of days by which the preceding two forecasts lead the analyst's forecast
divided by the cumulative number of days by which the subsequent two forecasts follow the analyst's forecast. COV is the log of the number of analyst following during the year.
REG equals 1 if the firm is in the financial services industry (one-digit SIC code 6) or utilities industry (two-digit SIC code 49), and 0 otherwise.
50
Table 7
The Effect of Starting Up (or Shutting Down) a Political Action Committee
Panel A: Firms that "Start Up" a Political Action Committee during the Sample Period
PFT = ρ 1 (Years t-2,t-1 ) + ρ 2 (Years t, t+1 ) + ρ 3 (Years t ≥ +2 ) + ρ 4 ACCURACY
+ ρ 5 (Years t-2, t-1 *ACCURACY) + ρ 6 (Year t, t+1 *ACCURACY) + ρ 7 (Years t ≥ +2 *ACCURACY)
+ ρ 8 VR + ρ 9 FIRMEXP + ρ 10 BSIZE + ρ 11 N_FIRMS + ρ 12 REC_FREQ + ρ 13 LFR
+ ρ 14 COV + ρ 15 REG + ε
Years t ‒ 2 and t ‒ 1
Years t and t + 1
Years t ≥ +2
ACCURACY
Years t ‒ 2 and t ‒ 1 *ACCURACY
Years t and t + 1*ACCURACY
Years t ≥ +2*ACCURACY
VR
FIRMEXP
BSIZE
N_FIRMS
REC_FREQ
LFR
COV
REG
Adj. R2
Coefficient
-0.0163
-0.0007
-0.0153 ***
0.0020 ***
0.0091
-0.0458 *
-0.0051 ***
0.0231 ***
0.0036 ***
0.0077 ***
0.0083 ***
0.0081 ***
0.0012 ***
-0.0022 ***
-0.0113 ***
S.E.
0.0154
0.0167
0.0008
0.0003
0.0203
0.0251
0.0010
0.0011
0.0003
0.0002
0.0004
0.0005
0.0001
0.0004
0.0007
t -statistic
-1.06
-0.04
-18.33
7.57
0.45
-1.82
-5.32
20.34
10.19
34.81
22.10
15.11
18.52
-5.63
-15.79
16.95%
F-Test for Difference Across Coefficients
Year t-2, t-1 = Year t, t+1
F = 0.56
p = 0.5696
Year t, t+1 = Year t ≥ +2
F = 168.35
p < 0.0001
Year t-2, t-1 *ACCURACY = Year t, t+1 *ACCURACY
F = 1.76
p = 0.1715
Year t, t+1 *ACCURACY = Year t ≥ +2 *ACCURACY
F = 16.15
p < 0.0001
51
p -value
0.2891
0.9650
<.0001
<.0001
0.6561
0.0682
<.0001
<.0001
<.0001
<.0001
<.0001
<.0001
<.0001
<.0001
<.0001
Table 7 (cont.)
The Effect of Starting Up (or Shutting Down) a Political Action Committee
Panel B: Firms that Shut Down a Political Action Committee during the Sample Period
PFT = ρ 1 (Years t ≤‒2 ) + ρ 2 (Years t‒1, t ) + ρ 3 (Years t+1, t+2 ) + ρ 4 ACCURACY
+ ρ 5 (Years t ≤‒2 *ACCURACY) + ρ 6 (Year t‒1, t *ACCURACY) + ρ 7 (Years t+1, t+2 *ACCURACY)
+ ρ 8 VR + ρ 9 FIRMEXP + ρ 10 BSIZE + ρ 11 N_FIRMS + ρ 12 REC_FREQ + ρ 13 LFR
+ ρ 14 COV + ρ 15 REG + ε
Years t ≤ ‒ 2
Years t ‒ 1 and t
Years t + 1 and t + 2
ACCURACY
Years t ≤ ‒ 2 *ACCURACY
Years t ‒ 1 and t*ACCURACY
Years t + 1 and t + 2 *ACCURACY
VR
FIRMEXP
BSIZE
N_FIRMS
REC_FREQ
LFR
COV
REG
Coefficient
-0.0136 ***
-0.0149
0.0400
0.0015 ***
-0.0018 **
-0.0496 *
-0.1016 *
0.0229 ***
0.0037 ***
0.0078 ***
0.0083 ***
0.0080 ***
0.0012 ***
-0.0023 ***
-0.0113 ***
Adj. R2
S.E.
0.0008
0.0167
0.0416
0.0003
0.0007
0.0251
0.0582
0.0011
0.0003
0.0002
0.0004
0.0005
0.0001
0.0004
0.0007
t -statistic
-16.66
-0.89
0.96
5.61
-2.55
-1.98
-1.75
20.19
10.65
35.13
21.98
15.01
18.34
-6.14
-15.77
p -value
<.0001
0.3747
0.3365
<.0001
0.0109
0.0479
0.0806
<.0001
<.0001
<.0001
<.0001
<.0001
<.0001
<.0001
<.0001
16.89%
F-Test for Difference Across Coefficients
Year t ≤‒2 = Year t ‒1 , t
F = 139.00
p < 0.0001
Year t ‒1 , t = Year t+1, t +2
F = 0.86
p = 0.0420
Year t ≤‒2 *ACCURACY = Year t ‒1 , t *ACCURACY
F = 5.17
p = 0.0057
Year t ‒1 , t *ACCURACY = Year t+1, t +2 *ACCURACY
F = 3.48
p = 0.0307
***, **, * indicate significance at the 1%, 5%, and 10% levels. Regressions are estimated over 1993 - 2012. Year
indicator variables signify years leading up to and following the year in which the firm started (or shut down) the
Political Action Committee. PFT is the buy-and-hold, abnormal return to the analyst's recommendation computed
from the day before the recommendation date until the earlier of 30 days or 2 days before the recommendation is
revised or reiterated. For buy (hold and sell) recommendations, $1 is invested (shorted) in the recommended stock.
ACCURACY equals negative 1 times the difference between the analyst's absolute forecast error and the mean
absolute forecast error for all analysts scaled by the mean absolute forecast error for all analysts. VR, value
relevance, is the R-adjusted from the following regression estimated for each firm over rolling 10-year windows: RET
= α0 + α1 EARN + α2 ΔEARN + ε, where RET is the firm's 15-month return ending three months after the fiscal yearend, EARN is income before extraordinary items scaled by market value at the end of the prior fiscal year, and
ΔEARN is the change in income before extraordinary items scaled by market value at the end of the prior fiscal year.
FIRMEXP is the log of the number of years for which the analyst supplied forecasts for the firm. BSIZE is the log of
the number of analysts employed by the analyst's brokerage firm. N_FIRMS is the log of the number of firms for
which the analyst issues annual forecasts during the year. REC_FREQ is the log of the number of recommendations
the analyst issues for the firm during the year. LFR, leader-follower ratio, is the cumulative number of days by which
the preceding two forecasts lead the analyst's forecast divided by the cumulative number of days by which the
subsequent two forecasts follow the analyst's forecast. COV is the log of the number of analyst following during the
year. REG equals 1 if the firm is in the financial services industry (one-digit SIC code 6) or utilities industry (two-digit
SIC code 49), and 0 otherwise.
52
Table 8
Propensity-Score Matching: The Effect of Political Connections on the Relation between Accuracy and Profitability
PFT = θ 0 + θ 1 POLITICAL + θ 2 ACCURACY + θ 3 ACCURACY*POLITICAL + θ 4 VR + θ 5 FIRMEXP + θ 6 BSIZE
+ θ 7 N_FIRMS + θ 8 REC_FREQ + θ 9 LFR + θ 10 COV + θ 11 REG + ε
Intercept
POLITICAL
ACCURACY
ACCURACY*POLITICAL
VR
FIRMEXP
BSIZE
N_FIRMS
REC_FREQ
LFR
COV
REG
Candidates
Coefficient
S.E.
0.0682 *** 0.0030
-0.0071 *** 0.0013
0.0012 *
0.0006
-0.0055 *** 0.0016
0.0101 *** 0.0020
0.0022 *** 0.0006
0.0001 *** 0.0000
-0.0010
0.0008
0.0054 *** 0.0009
0.0010 *** 0.0001
-0.0100 *** 0.0008
-0.0061 *** 0.0013
Strength
Coefficient
0.0555 ***
-0.0073 ***
0.0000
-0.0036 **
0.0099 ***
0.0020 ***
0.0045 ***
-0.0008
0.0066 ***
0.0009 ***
-0.0100 ***
-0.0071 ***
S.E.
0.0034
0.0013
0.0006
0.0016
0.0021
0.0006
0.0004
0.0008
0.0009
0.0001
0.0008
0.0014
Ability
Coefficient
0.0560 ***
-0.0044 ***
0.0002
-0.0022 *
0.0099 ***
0.0020 ***
0.0045 ***
-0.0009
0.0065 ***
0.0009 ***
-0.0100 ***
-0.0073 ***
S.E.
0.0034
0.0010
0.0007
0.0012
0.0021
0.0006
0.0004
0.0008
0.0009
0.0001
0.0008
0.0014
Power
Coefficient
S.E.
0.0556 *** 0.0034
-0.0052 *** 0.0013
0.0000
0.0006
-0.0031 ** 0.0016
0.0098 *** 0.0021
0.0020 *** 0.0006
0.0045 *** 0.0004
-0.0008
0.0008
0.0066 *** 0.0009
0.0009 *** 0.0001
-0.0102 *** 0.0008
-0.0071 *** 0.0014
***, **, * indicate significance at the 1%, 5%, and 10% levels. There are 61,010 observations estimated over 1993 - 2012. Propensity scores are calculated using
the 1st Stage Probit model in Appendix B. The four columns in this table use 4 different proxies for political connections (POLITICAL): Candidates, Strength,
Ability, and Power. POLITICAL equals 1 if politically connected, 0 otherwise. PFT is the buy-and-hold, abnormal return to the analyst's recommendation
computed from the day before the recommendation date until the earlier of 30 days or 2 days before the recommendation is revised or reiterated. For buy (hold and
sell) recommendations, $1 is invested (shorted) in the recommended stock. ACCURACY equals negative 1 times the difference between the analyst's absolute
forecast error and the mean absolute forecast error for all analysts scaled by the mean absolute forecast error for all analysts. VR, value relevance, is the Rsquared from the following regression estimated for each firm over rolling 10-year windows: RET = α0 + α1 EARN + α2 ΔEARN + ε, where RET is the firm's 15-month
return ending three months after the fiscal year-end, EARN is income before extraordinary items scaled by market value at the end of the prior fiscal year, and
ΔEARN is the change in income before extraordinary items scaled by market value at the end of the prior fiscal year. FIRMEXP is the log of the number of years
for which the analyst supplied forecasts for the firm. BSIZE is the log of the number of analysts employed by the analyst's brokerage firm. N_FIRMS is the log of
the number of firms for which the analyst issues annual forecasts during the year. REC_FREQ is the log of the number of recommendations the analyst issues for
the firm during the year. LFR, leader-follower ratio, is the cumulative number of days by which the preceding two forecasts lead the analyst's forecast divided by
the cumulative number of days by which the subsequent two forecasts follow the analyst's forecast. COV is the log of the number of analyst following during the
year. REG equals 1 if the firm is in the financial services industry (one-digit SIC code 6) or utilities industry (two-digit SIC code 49), and 0 otherwise.
53
Table 9
The Effect of Regulation Fair Disclosure, SOX, and Other Analyst Regulations
PFT = λ 0 + λ 1 POST + λ 2 POLITICAL + λ 3 POLITICAL*POST + λ 4 ACCURACY + λ 5 ACCURACY*POLITICAL + λ 6 ACCURACY*POLITICAL*POST
+ λ 7 VR + λ 8 FIRMEXP + λ 9 BSIZE + λ 10 N_FIRMS + λ 11 REC_FREQ + λ 12 LFR + λ 13 COV + λ 14 REG + ε
Intercept
POST
POLITICAL
POLITICAL*POST
ACCURACY
ACCURACY*POLITICAL
ACCURACY*POLITICAL*POST
VR
FIRMEXP
BSIZE
N_FIRMS
REC_FREQ
LFR
COV
REG
2
Adj. R
Year Dummies
Candidates
Coefficient
S.E.
0.0438 *** 0.0029
0.0134 *** 0.0024
-0.0076 *** 0.0015
-0.0086 *** 0.0019
0.0007 **
0.0003
-0.0013
0.0017
-0.0049 **
0.0022
0.0124 *** 0.0012
0.0048 *** 0.0004
0.0029 *** 0.0003
0.0006
0.0005
0.0048 *** 0.0006
0.0010 *** 0.0001
-0.0105 *** 0.0005
-0.0111 *** 0.0007
2.44%
Yes
Strength
Coefficient
S.E.
0.0440 *** 0.0029
0.0133 *** 0.0024
-0.0081 *** 0.0015
-0.0082 *** 0.0019
0.0006 **
0.0003
-0.0007
0.0017
-0.0055 **
0.0022
0.0123 *** 0.0012
0.0048 *** 0.0004
0.0030 *** 0.0003
0.0006
0.0005
0.0048 *** 0.0006
0.0010 *** 0.0001
-0.0106 *** 0.0005
-0.0112 *** 0.0007
2.44%
Yes
Ability
Coefficient
0.0444 ***
0.0245 ***
-0.0036 ***
-0.0076 ***
0.0006
-0.0013
0.0011
0.0124 ***
0.0047 ***
0.0029 ***
0.0006
0.0048 ***
0.0010 ***
-0.0109 ***
-0.0114 ***
2.40%
Yes
S.E.
0.0029
0.0026
0.0013
0.0016
0.0004
0.0015
0.0015
0.0012
0.0004
0.0003
0.0005
0.0006
0.0001
0.0005
0.0007
Power
Coefficient
0.0440 ***
0.0130 ***
-0.0078 ***
-0.0082 ***
0.0006 **
-0.0018
-0.0021
0.0123 ***
0.0048 ***
0.0029 ***
0.0006
0.0048 ***
0.0010 ***
-0.0105 ***
-0.0112 ***
S.E.
0.0029
0.0024
0.0015
0.0019
0.0003
0.0017
0.0022
0.0012
0.0004
0.0003
0.0005
0.0006
0.0001
0.0005
0.0007
2.44%
Yes
***, **, * indicate significance at the 1%, 5%, and 10% levels. There are a total of 192,804 observations (59,430 observations in the Pre-Regulation Period during January 1993 September 2000 and 133,374 observations in the Post-Regulation Period during May 2003 - December 2012). POST equals 1 if Post-Regulation Period, and 0 if Pre-Regulation
Period. The four columns in this table use 4 different proxies for political connections (POLITICAL): Candidates, Strength, Ability, and Power. POLITICAL equals 1 if politically
connected, 0 otherwise. PFT is the buy-and-hold, abnormal return to the analyst's recommendation computed from the day before the recommendation date until the earlier of 30
days or 2 days before the recommendation is revised or reiterated. For buy (hold and sell) recommendations, $1 is invested (shorted) in the recommended stock. ACCURACY
equals negative 1 times the difference between the analyst's absolute forecast error and the mean absolute forecast error for all analysts scaled by the mean absolute forecast
error for all analysts. VR, value relevance, is the R-squared from the following regression estimated for each firm over rolling 10-year windows: RET = α0 + α1 EARN + α2 ΔEARN +
ε, where RET is the firm's 15-month return ending three months after the fiscal year-end, EARN is income before extraordinary items scaled by market value at the end of the prior
fiscal year, and ?EARN is the change in income before extraordinary items scaled by market value at the end of the prior fiscal year. FIRMEXP is the log of the number of years
for which the analyst supplied forecasts for the firm. BSIZE is the log of the number of analysts employed by the analyst's brokerage firm. N_FIRMS is the log of the number of
firms for which the analyst issues annual forecasts during the year. REC_FREQ is the log of the number of recommendations the analyst issues for the firm during the year. LFR,
leader-follower ratio, is the cumulative number of days by which the preceding two forecasts lead the analyst's forecast divided by the cumulative number of days by which the
subsequent two forecasts follow the analyst's forecast. COV is the log of the number of analyst following during the year. REG equals 1 if the firm is in the financial services
industry (one-digit SIC code 6) or utilities industry (two-digit SIC code 49), and 0 otherwise.
54