Friends at WSJ
Guosong Xu∗
January 28, 2020
Abstract I study the effect of the firm–journalist connections on media slant and stock returns, us- ing a dataset on the firm’s connections to Wall Street Journal reporters. When corporate events are covered by connected reporters, the news sentiment becomes markedly more favorable, leading to higher short-term stock returns and long-term price corrections. For identification, I instrument the connected coverage with the reporters’ turnover and find similar results. Connections to the media owner also matter: using Rupert Murdoch’s acquisition of the WSJ as an exogenous shock, I show that firms connected to Murdoch receive better coverage after the ownership change.
JEL classification: G14, G34, G40 Keywords: News sentiment, Mergers and acquisitions, Financial journalism, Social connections
∗Department of Finance, Erasmus University, Rotterdam School of Management. Mail: Burgemeester Oudlaan 50, 3062 PA Rotterdam, Netherlands. Email: [email protected]. For helpful comments I thank Nihat Aktas, Yakov Amihud, Bernard Black, Audra Boone, Eric de Bodt, Travers Barclay Child, Lin Willian Cong, Olivier Dessaint, Daniel Dorn, Daniel Ferreira, Eliezer Fich, Andrey Golubov, Markku Kaustia, Matti Keloharju, Cohen Lauren, Erik Loualiche (discussant), Tim Loughran (discussant), Song Ma, Mario Schabus (discussant), Ishita Sen (discussant), Paul Smeets, Noah Stoffman, Vladimir Vladimirov, Michael Weber, Burcin Yurtoglu, and participants at the 2018 European Finance Association Meeting Doctoral Tutorial, CEIBS Finance and Accounting Symposium, Helsinki Finance Summit, Paris December Finance Meeting, Aalto University, Drexel University, Erasmus University, ESCP-Europe, Fudan University School of Management, Humboldt University in Berlin, Paris-Dauphine University, and University of Strathclyde. 1 Introduction
How are media stories created? A number of studies reveal that the media have an important impact on the financial market.1 Key to this relation is an author’s persuasion, views, or bias injected into the news article (Dougal et al., 2012). However, evidence on the influences that shape financial journalists’ view in news production is limited. In this paper, I explore one particular channel of influence: that of personal social networks. Specifically, I study whether relationships between a firm and a journalist represent a mechanism for influencing the tone, or “slant,” of financial news and the consequences for asset price.
The expected effect of the journalist–firm connections is not obvious ex-ante. On the one hand, if financial media compete using accuracy for rational investors, a well-connected journalist could yield more accurate reporting by catalyzing the information flow. On the other hand, a business reporter beholden to a firm might inject his personal opinion and slant the story in favor of the firm because he relies on company management for information (Call et al., 2018) or exhibits homophily due to a shared background (Gompers et al., 2016).
To assess the effect of the journalists’ connections, I assemble a set of financial news articles published in the Wall Street Journal (WSJ) over the past 20 years. The choice of WSJ is natural because the WSJ is by far the largest print newspaper in the United States.2 In addition, the WSJ is extremely well-established among finance and investment professionals.
I focus on one type of corporate news, namely, the WSJ coverage of corporate takeovers.
Given their financial magnitude and importance, merger transactions elicit substantial investor scrutiny and attention, but the assessment of the deal synergies is subject to an individual perspective. Therefore, any distortion in the media news production may result in real wealth effects among parties to the transaction. Moreover, focusing on this homogeneous event type enables me to better control for determinants of the underlying transaction quality, motivated
1 See, for example, Huberman and Regev(2001), Tetlock(2007), Tetlock et al.(2008), Engelberg and Parsons (2011), and Peress(2014). 2 The average weekday circulation of the WSJ is above 2 million according to Alliance for Audited Media as of March 31, 2013. Alternatively, I also examine the news articles published in the New York Times.
1 by a large body of M&A literature.
Next, I obtain from Bloomberg and major professional network websites the college names of the acquirers’ CEOs and of the authors of the WSJ articles. This allows me to observe, for instance, whether John Riccitiello, the Chief Executive of Electronic Arts, attended the same college as the one attended by the WSJ reporter, Nick Wingfield, who wrote about the Electronic Arts’ acquisition of Take-Two in February 2008. Additionally, using the WSJ search engines and Factiva, I also capture the potential working relationships between a firm and a journalist by examining if a specific reporter wrote at least two exclusive non-M&A related stories about that firm during the 12 months prior to an acquisition. This approach follows Solomon(2012), which reflects the idea that if a journalist frequently covers a firm, he is more likely to build a personal relationship with that firm’s employees. Now the key question: do such relationships matter in terms of news tonality and investor reactions?
Some key pieces of evidence suggest a meaningful media slant in articles authored by journalists in a firm’s social network. First, in the cross section, news articles from a connected journalist include significantly fewer negative-sentiment words describing the firm, defined by
Loughran and McDonald.3 For example, all else equal, the use of negative words is 23.6% lower in the stories written by journalists who have working experiences with the acquirer
(t = 3.98 for the difference). Similarly, articles authored by a journalist who attended the same college as the acquirer’s CEO contain markedly 13.2% fewer negative words (t = 2.31).
Second, exploring within-journalist variations, I find that the fraction of negative words in an article about a connected firm is significantly lower than other articles written by the same author. These models reject the possibility that a journalist’s general writing styles are driving the slant in connected reports (Dougal et al., 2012).
One obvious challenge in interpreting the relation between slant and connection is that it may reflect reverse causation running from news tone to journalist connections. As Tetlock
3 See Loughran and McDonald(2011, 2016, 2017), Bodnaruk et al.(2015), and the dedicated website: http: //sraf.nd.edu/.
2 (2007) notes, “It is unclear whether the financial news media induces, amplifies, or simply reflects investors’ interpretations of stock market performance.” In this context, a connected journalist may be assigned to write about better deals after having observed market reactions.
To address this concern, in all my main tests, I include several lags of returns. Moreover,
I supplement the WSJ sample with media stories from the New York Times that cover the same event. In this pooled sample, I saturate my empirical models with highly granular merger
fixed effects to effectively control for all event-specific characteristics, including deal qualities.
Similar to the previous findings, this analysis suggests a significant less use of negative words in stories by a connected journalist relative to those by an independent journalist, even though the information content of the underlying event remains identical.
To further provide a plausibly causal interpretation, I adopt an instrument variable (IV) method. Specifically, I instrument the likelihood of a connected coverage with the turnovers of a firm’s connected journalists. The basic idea is as follows: Returning to the example of Electronic Arts, suppose the firm has access to one connected reporter; shortly before
Electronic Arts announced its bid in February 2008, the connected reporter left the WSJ; therefore, the likelihood of this bid being covered by a friendly reporter becomes virtually zero.4 Importantly, the turnover of journalists is likely due to events exogenous to the M&A deals or the firms they write about (Solomon, 2012). With the IV approach, I continue to find a significant connection effect on news tone.
Having established the link between slant and journalist connections, I examine whether this favorable coverage alters market reaction to the acquisition. On the one hand, if investors are fully aware of the biased content produced by a financial reporter, then slant should not affect stock prices. On the other hand, if investors face difficulties in figuring out potential social networks between a firm and media, there is good reason to predict that higher media sentiment in a connected report may generate better stock market reactions.
4 To avoid the potential replacement of journalist connections, I focus solely on cases where all connected journalists of a firm leave the newspaper. See details in Section 4.2.
3 Evidence suggests a strong relation between journalist connections and the abnormal stock returns in the two days following the news publication. All else being equal, stock returns are significantly higher to the acquirers whose deal is covered by a journalist with retrospective working relationships. The relation remains when I instrument the connection with journalists’ turnovers.
Exploring the mechanism underlying the connection effects on stock returns, I find that the market responses are attenuated in firms with higher stock liquidity or more rational arbitrageurs, measured by greater analysts’ coverage, larger firm size, and higher levels of institutional ownership. These findings are congruent with the predictions formulated in
Tetlock(2007). Moreover, I show that the short-term positive reaction to the optimistic news story is reversed in the long run: After 40 trading days, the documented higher returns to connected acquirers eventually converge to the similar level of those non-connected firms.5
The evidence of long-term price correction, again, clearly rejects the alternative explanation of information advantage possessed by connected journalists. Under this alternative hypothesis, a connected journalist reports higher quality deals accurately, predicting permanent pricing of information in the acquirers’ stocks rather than a long-term price correction.
In addition to the effects of social networks in the form of firm–journalist relationships, I also investigate connections to the media owner. Here, I explore an exogenous shock to the journalistic independence induced by an ownership change. In 2007, the WSJ was taken over by News Corp. This takeover represents an ideal laboratory to examine how news presentation and related stock reactions are altered for firms previously connected to the new media owner,
Rupert Murdoch (CEO of News Corp). Prior to the takeover, the WSJ itself questioned whether the newspaper could retain its editorial independence under the new owner:
“Mr. Murdoch has tended to put a strong personal imprint on papers he owns, [...]
He is known for phoning editors and even reporters about individual stories. The
5 The horizon of the price correction appears rather comforting compared to the well-documented underreaction to Friday announcements (DellaVigna and Pollet, 2009), which lasts approximately 75 days, or to the reversal caused by contrast effects (Hartzmark and Shue, 2017), which dissipates after 50 days.
4 Post’s media and business sections sometimes delight in skewering rivals, and Mr.
Murdoch’s political preferences have been clear in the news pages [...]” 6
And this is precisely what I find in a difference-in-difference analysis. Using hand-collected
firm connections to Mr. Murdoch for the subsample of acquirers who conducted M&A deals both before and after the 2007 WSJ ownership change,7 I show that firms connected to Mr.
Murdoch are better pictured in the WSJ articles following Murdoch’s takeover. Abnormal stock returns to these merger announcements also increase after the ownership change.
I conclude the analysis with a set of robustness tests. First, I show that the main results are not subject to concerns of selection in news coverage. Second, I further rule out the issue of reverse causality: while journalist connections are related to future stock reactions (day t and t + 1), they are not related to past returns (t − 2 and t − 1). The results also appear robust to alternative measures of news tone, of working relationships, or specifications that include additional control variables related to the target firm.
By examining the relations between journalists’ connections and news tone and stock re- turns, this paper contributes to two strands of burgeoning literature. Foremost, recent work reveals that firms are strategically marketing and spinning the news. For example, Solomon
(2012) examines selective publicity through the use of investor relations (IR) firms; Gurun and Butler(2012) and Golez and Karapandza(2019) report that local media positively slant the news toward local firms. Studying M&As, Ahern and Sosyura(2014) uncover that bidders originate more positive news stories during stock mergers. All these studies consistently doc- ument a real effect on the financial market.8 In contrast to active media management (e.g.,
Solomon and Soltes, 2012) or even manipulative communications (e.g., Leuz et al., 2017), this paper suggests a hidden channel of influencing the media slant via firm–reporter networks.
6 See the Wall Street Journal, “Murdoch’s Surprise Bid: $5 Billion for Dow Jones,” May 2, 2007. 7 Such connections include direct business relations with News Corp, common directors, and/or common political campaign seats. 8 Other studies that establish a link between media and financial markets include, e.g., Huberman and Regev (2001), Tetlock(2007, 2011), Tetlock et al.(2008), Fang and Peress(2009), Engelberg and Parsons(2011), Dougal et al.(2012), Peress(2014), and Solomon et al.(2014). See Tetlock(2015) for an excellent synthesis on the role of media in finance, and Goldman et al.(2019) for a thoery of financial media.
5 To the extent that my analysis does not preclude the other channels, such as connections through IR firms, the estimate documented in this paper provides a lower bound effect of the connections.
In this regard, this paper also contributes to a growing body of “network” literature. The idea of network mattering is certainly not new: In the context of M&As, Cai and Sevilir(2012) provide evidence that board connectedness creates value for acquirers; Schmidt(2015) finds mixed evidence of value brought by “friend” board members; El-Khatib et al.(2015) shows a negative relation between deal returns and CEO network centrality. Exploring the performance of venture capitals, Gompers et al.(2016) find that social ties attribute to poorer decision making. On a positive note, Cohen et al.(2008) and Engelberg et al.(2012) suggest that social networks facilitate information flow, leading to better investment or lending decisions.
Shue(2013) and Fracassi(2016) find that social networks affect firm policies. More recently,
Gurun(2016) and Di Giuli and Laux(2018) study media linkages of board members. They show that these linkages increase sentiment in news flows, leading to better financing outcomes.
Collectively, this paper provides evidence that a new form of social networks, namely those between firms and news media, could have a significant and economically meaningful impact on corporate publicity and equity prices. Hence, the results imply important considerations for both firms and financial media concerning their behavior in news production. Social connections between journalists and news sources may prove crucial in the related dabates for
financial market, as well as media, regulators.
2 Why Do Social Networks Matter for Media Slant?
The central question of this paper is whether financial journalists’ connections with firms lead to slant. Unlike political news, where newspapers have an incentive to cater to the demand of readers’ ideology (Gentzkow and Shapiro, 2010), financial news is usually assumed to cater to rational investors who request accurate information of the underlying story because they rely
6 on the accuracy of information to form expectations of the companies’ future prospects. In this case, if a well-connected journalist has an advantage of gathering informative corporate insight, news coverage of the connected firm should exhibit fewer slants. On the other hand, there is good reason to expect more slanted stories from a journalist–firm social tie. The rationale is as follows.
First, homophily in which journalists and executives share a common background implies that these journalists have similar preferences and beliefs as those of corporate executives. This could lead the journalist and the CEO to both think a merger is synergetic. Furthermore, homophily may induce insufficient desire for the journalists to consider the disadvantages of the corporate decision because they derive personal utility from the collaboration (Gompers et al., 2016).
A second explanation is that a journalist beholden to a firm relies on the firm for in- formation (Dyck and Zingales, 2002). A survey by Call et al.(2018) supports that private communications with company management constitute a major source of information for fi- nancial reporters. In addition, journalists often face backlash from the firm in response to unfavorable reporting. Though a quid pro quo implies that a non-connected reporter may also have incentives to slant towards the firm in order to establish a potential connection, such benefit is far less clear than the favoritism from a journalist in an existing relationship with the firm.
A last but non-exclusive possibility is that a journalist in the firm’s network has developed personal relationships with the firm’s CEO or employees. These personal relationships might lead to slanted reporting or, in a worse case, manipulated news dissemination.
7 3 Data Description
3.1 Sample of news articles
The analysis of this paper focuses on the WSJ reports on corporate takeovers. Prior studies
(e.g., Roll, 1988; Pound and Zeckhauser, 1990; Mitchell and Mulherin, 1994; Tetlock, 2007;
Tetlock et al., 2008; Dougal et al., 2012) reveal that WSJ coverage has a significant impact on the financial market. I narrow down my search to the takeover news because mergers and acquisitions represent the most significant corporate event and, importantly, because the interpretation of M&A synergies is highly subjective and clearly reflective of an author’s personal view.9 Focusing on a homogeneous event type also enables me to better control for the underlying event characteristics. Accordingly, I gather the sample from the following sources: the SDC M&A Database for the M&A transactions and Factiva for related WSJ articles.
I first collect all the M&A deals during 1997–2016 consisting of US public bidders and targets. The sample period begins in 1997 because the main measure of the firm–journalist connection requires a manual search on the WSJ engine, which starts in January 1997. Fol- lowing the selection methods most often used in the M&A literature, the sample excludes recapitalizations, self-tender offers, exchange offers, repurchases, partial equity stake purchas- es, acquisitions of remaining interests, privatizations, financial buyouts, as well as deals in which the target or the acquirer is a government agency.10 I further ensure that (1) the trans- action value exceeds US$10 million, (2) the acquirer owns at least 50% of the target equity after the deal completion, (3) the target is not undergoing bankruptcy proceedings, and (4) the parties are non-financial firms. These criteria yield a total number of 2,390 deals from
SDC. 9 To be precise, the synergies captured by the bidder are stochastic and subject to individual interpretations. The return to the targets is primarily determined by the offer premium (Betton et al., 2008, Chapter 15, p. 411). Therefore, this paper focuses on the bidder’s return. 10 This selection procedure is similar to, e.g., Masulis et al.(2007) and Erel et al.(2012).
8 Obviously, not all takeovers are covered by the WSJ. I search the WSJ coverage using a
Factiva algorithm that includes media sources from the WSJ US print edition. To be included in the sample, I require that the article is the first report on the transaction and appears within
2 days of the official deal announcement.
After matching the WSJ sample firms with Compustat (for financial data) and CRSP (for stock data), the final sample consists of 1,131 M&A news articles with identifiable journalists’ backgrounds. Table 1 reports the number of M&A transactions from SDC and valid WSJ sample across calendar years and months. Reassuringly, the WSJ provides a solid coverage of large transactions: Overall, 47% of all SDC deals are featured in the WSJ. The covered deals collectively account for over US$5.9 trillion in terms of transaction value and have an average deal size of US$5.2 billion. This compares to the average SDC deal size of US$2.6 billion.
The relatively large WSJ transaction value is perhaps unsurprising because these transactions attract greater media attention (e.g., Solomon and Soltes, 2012).
3.2 Journalists and connections
Measuring journalists’ social connections is complicated by the fact that individual relation- ships are not directly observed. Therefore, I follow conventions in the finance literature to proxy for such networks. For practical reasons, I focus on two types of firm–journalist con- nections, namely, working relationships and educational ties. Admittedly, this practice omits other potential social ties, such as common board affiliations and second-degree connections through IR firms. To the extent that these additional networks matter, my focus on the work and educational relationships provides a conservative estimate of connection effects.
To identify direct working relationships between a reporter and a firm, I follow Solomon
(2012) by checking whether a reporter has written at least two exclusive stories about the same acquirer in the 12 months prior to the merger. Particularly, I read each pre-deal article to ensure these stories are unrelated to the M&A deal in question. The working relationship
9 measure, CONNECT WORK, takes a value of one if any author of the M&A story has covered the acquirer at least twice in the previous year.
I construct the educational network by searching for universities that a reporter and an ac- quirer CEO attended. The WSJ provides personnel biographical information on the authors’ web page. I supplement this information with data crawled from Linkedin, a large profes- sional networking service, and Muck Rack, a public relations firm that collects journalists’ biographical information in its media database. I also gather the names of the acquirer CEO’s universities from the companies’ proxy filings on EDGAR and Bloomberg Executive Profile
Database. The measure of the educational tie, CONNECT UNIVERSITY, equals one if any author of the article attended the same school as the one attended by the acquirer CEO.11
In the top panel of Table 2, I summarize the firm–journalist connections. The measures are calculated at the news article level: For example, out of the 1,131 stories, 26% are written by a reporter that has prior working relationships with the acquirer firm. Connections through universities are far less common; the mean of CONNECT UNIVERSITY indicates that only
2.3% of the articles are written by an author who attended the same college as the CEO.
Ahern and Sosyura(2015) suggest that journalists’ experience might affect the report- ing accuracy. Along this line of findings, I collect control variables to capture a journalist’s experience and individual characteristics. These include information on each journalist’s gen- der, location (i.e., WSJ office), industry expertise, university degree, tenure (i.e., number of months at the WSJ), and total number of publications on the WSJ. The summary statistics, calculated at the article level and reported in Table 2, show that 44% of the articles have at least one female coauthor and that 23% have a reporter based in the same city as the acquirer’s headquarter. On average, the authors have worked for 5.6 years at the WSJ. In addition, 78% of the authors hold a major degree in journalism or literature, and 48% have a professional
11 The databases do not provide records of graduation years of individuals, restricting my ability to observe the connections within a cohort. I verify in my sample that the most common institutions that provide interpersonal connections are typically elite universities with strong ties among alumni, including Harvard University, Princeton University, Stanford University, and UC Berkeley.
10 specialization in the acquirer’s industry. These statistics are largely consistent with those reported by Ahern and Sosyura(2015). Table 2 also provides the summary statistics split by journalists’ connection status: for most of the reported characteristics, we do not observe a significant difference between connected and independent journalists.
3.3 News tone and stock returns
The main dependent variable, news negativity, characterizes the negative sentiment in each news article. I focus on the negative sentiment because previous studies find that negative information has a stronger impact than positive information (e.g., Rozin and Royzman, 2001;
Tetlock, 2007; Tetlock et al., 2008; Gurun and Butler, 2012). I use the negative word catego- rization of Loughran and McDonald 2016 Master Dictionary to count the number of negative words12 and calculate its fraction out of the total article word count:
News negativity = (#Negative Words/Total #Words) × 100 (1)
The average negativity expressed in an M&A article equals 1.4% (See Table 2). This compares favorably to the 1.7% negative slant reported by Gurun and Butler(2012). However, a casual observation from the right columns suggests that negativity in independent articles exceeds that in connected ones — whether this difference is statistically significant and whether it is truly driven by journalist connections remain the main question that I explore in the following section.
To capture market reaction to news, I estimate the cumulative abnormal return (CAR) for each bidder. I focus on the CAR from the WSJ publication day (which I call day 0) until the
12 The Loughran and McDonald Dictionary (Loughran and McDonald, 2011; Bodnaruk et al., 2015; Loughran and McDonald, 2016) contains sentiment words in financial applications and has been widely applied in financial context analysis (e.g., Gurun and Butler, 2012; Solomon, 2012; Ahern and Sosyura, 2015). The most frequently occurring Loughran and McDonald(2011) negative words include, e.g., loss(es), impairment, against, adverse(ly), failure, unable, doubtful, etc. To get a sense of the news tone, Internet Appendix 1 gives two examples of overall “positive” and “negative” excerpts from the WSJ articles.
11 next day. The CAR is the residual from the market model, whose parameters are estimated over a 200-day window ending 31 days before the deal announcement. This procedure requires
200 non-missing returns during the estimation window. The market portfolio is proxied by the value-weighted CRSP index. To reduce the influence of outliers, I winsorize the returns at
0.5% and 99.5% levels. Table 2 shows that the average bidder CAR is -0.7%, largely consistent with the negative bidder CAR for a sample of 1,664 public deals analyzed by Cai and Sevilir
(2012). The negative returns, however, are mainly observed in deals covered by independent journalists, as suggested in the summary of sample splits.
3.4 Additional control variables
I construct several bidder and deal related characteristics as control variables. These charac- teristics have been widely used in the M&A literature to capture potential merger qualities.
For example, I collect deal relative size, hostile attitudes, unsolicitations, diversifying bids, payment methods (e.g., cash versus stock), and completion rates from the SDC. From Com- pustat, I also gather acquirer characteristics, such as size, Tobin’s Q, leverage, cash holdings, and profitability. Additionally, to capture the acquirers’ corporate governance (e.g., Masulis et al., 2007), I obtain board-related variables, including CEO age, founder CEO status, duality, board size, and classified board, from the firms’ proxy filings prior to the M&A announcement.
Descriptive statistics for these additional variables appear in Table 2. To conserve space and avoid repetition, Appendix 1 provides the definition for all variables. In most important respects, the sample characteristics are similar to those analyzed in earlier studies: For in- stance, approximately 32% of the transactions are all-cash financed and 3% are classified as hostile bids, which is in line with the 31% and 1% reported by Cai and Sevilir(2012); At
90%, the completion rate in my sample is comparable to that of 85% in Gaspar et al.(2005).
Perhaps unsurprisingly, we observe that firms with journalist connections are on average larger in size and have higher relative deal values.
12 4 Journalist Connections and News Tone
I begin the empirical analysis by examining whether firms connected to a journalist receive more favorable coverage when their stories are featured in the newspaper. I present the empirical strategy and regression results in Section 4.1. I further take an instrument variable
(IV) approach and discuss the results in Section 4.2.
4.1 Evidence on the media slant of connected reports
To get a sense of how the journalists’ connections affect news tone, I estimate the following ordinary least squares (OLS) regression:
News negativityit = βit · CONNECTit + γi · L3(Returnt) + η · Controls + θt + ψj + it (2)
where News negativityit is the news tone of an M&A article, as defined in (1). Subscript i indexes a specific transaction, and t indicates the time (i.e., year-month) of the deal announce- ment.
The variable of interests, CONNECT , corresponds to either measure of the firm–reporter connections described in Section 3.2. L3 is an operator that transforms acquirer returns into the values of three lags (i.e., returns from −3 till −1 before the news release date). If media sentiment is influenced by the observed market reactions to the underlying event, these lagged returns will capture such factors. Across my specifications, firm-industry fixed effects, indexed by j and classified based on Fama-French 38 industries, are controlled for. In addition,
I include time fixed effects (θt) to control for unobserved time trends such as market sentiment (Baker and Wurgler, 2006).13 Additional control variables include journalist, firm, and deal characteristics. Standard errors are double-clustered by time and by industry.
I begin with the specifications that examine working relationships. Using sample arti-
13 All results are qualitatively unaltered if I use less granular year fixed effects.
13 cles from the WSJ, the regression results reported in the first column of Table 3 indicate a significant negative relation between news negativity and the journalists’ connections (CON-
NECT WORK). The point estimate suggests a substantially more favorable coverage if the author has prior working experiences with the acquirer: on average, an article authored by a connected reporter contains 23.6% fewer negative words relative to an independent article.14
Are journalists’ personal traits, such as optimism, drive the observed favorable coverage?
In coloumn 2 I additionally control for journalist fixed effects.15 Note that this is a powerful test, as it explores the variation of slant within a journalist. This means that CONNECT is identified solely from the differential news tone between a journalist’s connected and non- connected articles. As a result, journalists who authored only once in my sample are dropped, leading to a smaller number of observations. The results show that the coefficient estimate of
CONNECT WORK remains highly significant. Here, the interpretation is that the news tone in stories on the acquirers with a working relationship is approximately 19.1% less negative than the tone in the other articles authored by the same journalist. Therefore, any general journalist writing styles or time-invariant characteristics, such as the Ivy League backgrounds, cannot drive my results.
In the next set of specifications, I examine the schooling network. The results are reported in columns 3 and 4. Similar to working connections, I find that educational ties significantly reduce the negative tone in the news reports. According to column 3, the fraction of negative sentiment is 13.2% (= 0.1841/1.396) lower in stories written by a college alumnus compared to articles issued by independent journalists. This relation is robust to controlling for journalist
fixed effects: in column 4, we find that the magnitude of slant becomes even larger when we compare variations within journalists.16
14 The calculation is as follows: I divide the coefficient estimate of CONNECT WORK (-0.3299) by the mean of news negativity (1.396), and obtain 23.6%. 15 Specifically, I include the fixed effects of the first-author journalist to deal with articles with multiple coauthors. The connected journalists are always ordered as the first journalist to maximize the within- journalist variation. The model specification is otherwise similar to that in column 2. 16 Cross-sectional difference of positive tone is insignificant between connected and non-connected journalists. However, the journalists in the CEO’s alumni network use significantly more positive words in a model that
14 4.1.1 Benchmark against The New York Times
One obvious concern in the preceeding analysis is that media connections correlate with (un- observed) deal synergies, which may drive simultaneously news tone and connections. To alleviate this concern, the most stringent tests should benchmark slant from the WSJ against another media outlet, say, the New York Times (NYTimes), while controlling for merger fixed effects. Note that merger fixed effects absorb all deal- and firm-level estimates, including deal synergies. The estimate of CONNECT is solely identified from reporters covering the same event, with one in the firm’s social network.
I follow the procedure described in Section 3 to collect the NYTimes articles and journalists’ information. Out of the 1,131 WSJ reported transactions, 993 (or 88%) also appear in the
NYTimes. However, the NYTimes frequently relies on newswire services and thus does not publish the authors’ names.17 Moreover, since 2001 the NYTimes has aggregated much of its M&A news in an online newsletter under the name of DealBook, also anonymous. After excluding these anonymous NYTimes stories, I am able to obtain a sample of 506 articles.
In the last two columns of Table 3, I lump together all NYTimes and WSJ stories. Be- cause these models include highly granular merger fixed effects, I only control for journalist characteristics and newspaper dummies. The results show that estimates of both connection measures remain negative, statistically significant at 1%–5%. In other words, connected media stories exhibit higher slant than the other article that covers an identical transaction.18
These results help rule out a specific concern of endogeneity: that of reverse causality, in which a connected journalist is assigned to write about better deals. Specifically, the significant estimates of my connection measures in the presence of merger fixed effects show that deal quality cannot explain the documented slant.
includes the journalist fixed effects. These results are available as Internet Appendix Table A1. 17 For example, Bloomberg News, Bridge News, Dow Jones, Reuters, and The Associated Press account for 27% of the NYTimes sample. These newswire services do not provide the journalists’ names. 18 Additional tests support a significant and negative relation between news negativity and CONNECT in the NYTimes articles only. A table in the Internet Appendix presents the results of the analysis.
15 4.2 Instrumenting journalist connections
An immediate challenge to a causal interpretation of the previous findings is that I do not observe whether the journalists’ assignment is random. Despite the highly granular merger
fixed effects in Table 3, there might exist other omitted variables that could bias the estimates.
In this subsection, I propose an instrument variable that is correlated with the probability of a connected coverage but has no independent effect on news tone. I contend that journalists’ turnover satisfies both the relevance condition and the exclusion restriction. The rationale is as follows. The departure of a friendly reporter from the WSJ will decrease the likelihood of the retrospective firm being covered by a connected journalist in the future; however, the turnover is mostly likely driven by the reporter’s personal life events and thus is exogenous to the firms they write about (Solomon, 2012). One may worry that even in the presence of a journalist turnover, editors might assign another connected reporter as a replacement. To alleviate this concern, I focus solely on cases where a firm is connected to one reporter prior to the turnover; hence, losing one friendly reporter entails a complete loss of WSJ connections.
To identify turnovers, I rely on the dates when a journalist joined and left the WSJ.19 I start by counting the number of a firm’s connections on the date of the M&A news, including all old “friends” who wrote about the acquirer and who are still working at the WSJ. Next,
I examine the journalists’ departures in the quarter prior to the bid announcement. If a firm has only one journalist connection and that connection left the WSJ during this period, I set the turnover variable equal to one. Note that the majority of companies have only one journalist connection (conditional on being connected): the average is 1.7 and the median is one.20 Because this exercise is possible only for the firms who have conducted more than one transaction and who have at least one journalist connection, I obtain a substantially smaller sample size of 318 observations.
19 I gather these dates from the WSJ web page and other professional network websites. 20 I do not observe cases where a firm has multiple connections and all of its connections turn over. I exclude cases where only one of a firm’s several connections turn over to avoid potential replacements. However, the IV results remain similar if I include these cases.
16 Before delving into the tests, I verify in a regression framework that journalist turnovers are not related to the performance of firms that they are connected to.21 These tests support the idea that turnovers are likely exogenous to the companies that readers care about. In a further probe, I find that most reporters left the WSJ to become a founder of a company or to take a progressive new role as an editor in another media outlet. This evidence suggests that turnovers are mainly caused by personal career motives.
In column 1 of Table 4, I show that the turnover of a connected journalist significantly decreases the probability that a future report is written by a reporter in the firms’ network.
The F -statistic (34.34) on the instrument is above the critical values proposed in a Stock-Yogo weak identification test, suggesting that the estimation is efficient.
In column 2, I report the estimate of the second-stage IV regression. The effect of con- nections retain statistical significance at 1% level, pointing to the same direction of media slant as documented in Table 3. The Sargan χ2 test cannot reject the joint null that the instrument is valid. The magnitude of slant becomes somewhat larger than the OLS estimate.
There are three reasons why. First, as suggested earlier, the OLS estimate of CONNECT is likely downward biased if my measure of working relationship is correlated with an omitted
firm–journalist connection (e.g., connections through an IR firm). This biased estimate may be particularly prominent in this IV subsample where firms are on average larger (recall from
Table 2 that connected firms are larger than non-connected firms) and therefore, more likely to have other (omitted) media ties.
This leads to the second explanation for a larger IV estimate: the average treatment effect is local (Jiang, 2017). A local treatment effect makes it difficult to generalize the finding to a general population, but is an unfortunate tradeoff that a researcher faces between internal and external validities. To the extent that my purpose is to document a relation between slant and connection — rather than quantifying the slant — this exercise still offers valuable insight into the causal plausibility of my underlying hypothesis.
21 The results are unreported but available in the Internet Appendix.
17 Finally, turnovers may be measured with noise.22 To ensure that the statistical significance in the second-stage estimates is not generated by such noise, I run a reduced-form regression in column 3. In its reduced form, the regression essentially resembles a difference-in-difference comparison. The results are consistent with those in an IV model. These findings suggest that, insofar as the IV is plausible, connections cause media slant.
5 Implications for Financial Markets
The immediate message from the preceeding analysis raises a perhaps more interesting ques- tion: Does the relationship-induced slant have real effects on the financial markets? Existing studies have established a link between news tone and stock returns (e.g., Tetlock, 2007).
Based on these studies, we expect that returns to the acquirers featured by a connected jour- nalist are higher than those covered by an independent journalist. Section 5.1 analyzes the effect of connected journalist on news announcement returns. Section 5.2 and 5.3 provide additional evidence that supports the effect of connections.
5.1 Journalist connections and returns
To assess the impact on the financial markets, I estimate return regressions similarly specified as in (2). The dependent variable in these regressions is the bidder’s CAR from day 0 until day
1, where day 0 is the publication date of the WSJ story. For example, if a firm announced a merger on Monday and the WSJ covered the deal on Tuesday, the dependent variable captures the overall market reaction to the story from Tuesday until Wednesday. The main explanatory variable of interest is CONNECT.
Table 5 reports the results. As expected, the estimate of CONNECT WORK exhibits
22 To see why this matters: the OLS estimator is defined as COV {connection, negativity}/V AR{connection} and the IV estimator is defined as COV {turnover, negativity}/COV {turnover, connection}. Noise in the measure of turnovers may artificially increase the covariance between turnovers and news tone, leading to a greater estimate of connection effects in the IV regression.
18 a positive and significant sign. The coefficient estimate, based on column 1, is 0.0147 or
1.47 percentage points with a set of controls commonly used in the M&A literature. The economic magnitude here is comparable to the other studies that examine connection effects on M&A returns. For example, Cai and Sevilir(2012) show acquirer–target board connections improve acquirers’ CAR by about 2 percentage points; Schmidt(2015) similarly documents
0.8 percentage-point higher returns in acquirers whose CEO is connected to board members for advisory needs.23 To put this magnitude of reaction into perspective, given an average acquirer’s market capitalization of US$31 billion 4 days prior to the deal announcement, a
1.47% increase in return translates into a striking $456 million (= 31, 000 × 1.47%) change in value during the two-day news window.
Barber and Odean(2008) point out that investors are often subject to limited attention when they process information. If the documented stock reactions to connected stories are truly driven by news sentiment, it is likely that these effects are stronger for the news featured on the front page of the WSJ, because the front-page articles elicit most investor attention.
In particular, Fedyk(2018) shows that front page positioning of news induces higher trading volumes and larger price changes. To test this possibility, I first classify the sample M&A articles into those appearing on the front page and those appearing somewhere else inside the newspaper. I then rerun the return regressions by including an interaction term between
CONNECT and Front page to identify the incremental effect of front-page positioning.
In column 3 of Table 5, the estimate of the interaction term “CONNECT WORK×Front page” suggests that the connection effect is concentrated on the news prominently positioned on the front page. The associated magnitude of the effect also becomes slightly larger. Im- portantly, I verify in an untabulated analysis that a negative tone in the front-page and non-front-page articles is undistinguishable (t between -0.91 and 0.62 for the difference). This
23 The magnitude of return reactions is typically larger in M&A events than in other general news. Exploring the direct impact of news tone on acquirer returns in an untabulated test (where I regress CAR directly on news tone), I show that a one-standard deviation increase in negativity is associated with 32 basis point lower returns, largely matching the findings in simialr studies (e.g., Tetlock, 2007; Solomon, 2012).
19 finding speaks convincingly to the media channel moderated by investors’ attention.
Column 2 presents the estimates of college networks. We spot insignificant return dif- ferences between bidders covered by a same-college journalist and bidders without such a connection. This is perhaps because the proportion of college connections is small. Low variation in this interested variable reduces the testing power (the standard errors are high).
However, when I examine the articles positioned on the front page, I again find a positive and significant coefficient on university connections (t = 2.61 in column 4).
The last two columns of Table 4 examine the connection effect using journalist turnovers as an instrument. I employ the same subsample and turnover variable as described in Section
4.2. We see that instrumenting the working relationship yields similar results as those reported in column 1. Again, I caution against extrapolating the effect size from column 5 to a general population because the average treatment effect is likely local to the subsample, where firms are repeated acquirers and all have at least one media connection. In column 6, the reduced- form regression shows that exogenous turnovers of friendly reporters reduce the stock returns to these acquirers.
5.2 Additional evidence: Interaction effect
Theoretical models propose two possible channels through which news slant could affect stock market pricing. The models of noise traders (e.g., DeLong et al., 1990) postulate that noise traders, facing downward-sloping demand for risky assets, sell shares to rational arbitrageurs when there is a negative belief shock, pressuring prices down. Theories of liquidity traders, such as that of Campbell et al.(1993), make similar predictions, with the media sentiment proxying for a change in the levels of risk aversion.
I test these mechamisms in this subsection. First, I consider the heterogeneous responses induced by different levels of rational arbitrageurs. To characterize the level of shares held by rational traders, I obtain the analyst coverage of a firm, total asset size, and the percentage of
20 institutional ownership. Rational arbitrageurs are expected to account for a greater fraction of trades for larger firms (Kumar, 2009), firms with more analysts’ coverage (Zhang, 2006), and/or firms owned by more institutional investors (e.g., Barber and Odean, 2013). Second, I test the channel of stock liquidity. To proxy for liquidity, I adopt Amihud(2002)’s illiquidity measure, which gauges the average impact of trading volume on a stock’s absolute return.
I interact these proxies with the firm–journalist connection measure, CONNECT WORK.
Panel A in Table 6 presents these results. The first model in Panel A suggests that firms covered by more analysts are less subject to the media connection effects. To obtain a sense of the magnitude of the estimate, adding one additional analyst reduces the connection effect by approximately 12 basis points on stock returns. Consistently, column 2 and 3 show that firm size and institutional ownership mitigate the distorted stock reaction induced by connected news coverage. The last column examines the stock liquidity. Because the Amihud(2002) measure essentially captures the illiquidity of a stock, we find a significant and positive sign of the interaction coefficient. Consonant with the earlier argument, this result suggests that higher liquidity attenuates the connection effects.
Do rational traders and stock liquidity eliminate the connection effects? The answer is no.
As is clear from Panel A, the coefficient estimate of the constituent term, CONNECT WORK, remains highly significant across all specifications. In other words, even liquid firms with rational investors continue to benefit from glowing reports by their newspaper connections.
5.3 Long-run stock price correction
My next set of tests builds on the sentiment theory that predicts a price correction in the long run. In these tests, I focus on the long-term window from day two through day forty following the WSJ story. Because stocks react more positively to the connected firms during the news announcement period, we expect a more negative abnormal return to these connected firms over the long run.
21 To ease the comparison of the results between tables, I first replicate the regressions of the short-term returns in the first two columns (see Panel B1 in Table 6). Moving right to columns 3 and 4, we first observe that returns to firms with connections become significantly more negative than non-connected firms over the [2,40] window. This finding supports the long-run price correction. The post-announcement price reversal, hence, completely cancels out the initial better stock reactions to connected acquirers upon M&A stories. Indeed, combining the short- and long-run returns together, in columns 5 and 6, we find insignificant difference in returns between connected and non-connected firms. Moreover, columns 7 and
8 show that acquirer CARs calculated from one day before the deal announcement until the deal completion date for the subsample of completed transactions are also similar between connected and non-connected firms.24
A potential concern with the price reversal results in Panel B1 is that they might be due to updated information related to the transaction during the long-term window. For example, investors might react to the updated news of transaction withdrawals. To address this issue, I remove 119 cases where the deal is eventually withdrawn in Panel B2. The regressions continue to document a short-run overreaction and the subsequent price correction for the connected acquirer firms.25
The result of price correction is important because it speaks against an alternative possi- bility that connected journalists possess superior information about the covered firms. Under this alternative hypothesis, connected journalists might achieve a higher level of accuracy in their reporting. However, the fact that we observe post-announcement price reversal for the connected group clearly rejects the information hypothesis because if any, we should find a price drift for the non-connected group following the news story when additional information of these acquirers flows to the market.
24 I drop 160 completed deals in columns 7 and 8 because the completion dates of these transactions are missing. 25 Additionally, I eliminate further 342 observations with potentially confounding events during [2,40]. These confounding events are mainly driven by the acquirer and/or target earnings announcements, as identified by the I/B/E/S database. The results are qualitatively unchanged (see Internet Appendix).
22 6 Connections with Media Owner
To this point, the analysis has focused on personal social networks between reporters and
firms. In this section, I explore another form of media connection: that of the media owner.
During my sample period, the WSJ experienced a change of ownership — the acquisition by
News Corp — which provides an exogenous shock to the journalistic independence. The shock allows me to explore the media owner effect in a relatively clean setting. In the following, I
first describe the background of this ownership change.
On May 2, 2007, News Corp, under the direction of Rupert Murdoch, made an unsolicited bid for Dow Jones, which owned the WSJ. The takeover officially occurred in August 2007, ending the newspaper’s 105-year ownership by the Bancroft family. The ownership transfer triggers a fierce debate over the potentially adverse effect on the journalists’ independence.
To requote the article published on May 2, 2007 by the WSJ, “Mr. Murdoch’s bid promises to raise questions about whether the Journal would retain its editorial independence under his ownership,” because Mr. Murdoch is “known for phoning editors and even reporters about individual stories.”26
Consistent with the voiced concerns, Wagner and Collins(2014) find that Murdoch’s WSJ becomes more negative to Democrats and more positive to Republicans; Archer and Clinton
(2018) similarly document a significant change of WSJ’s political coverage following Murdoch’s acquisition.27 Like these papers, I adopt a difference-in-difference strategy that investigates whether news coverage and returns become more favorable for the firms connected to Murdoch after the WSJ takeover. To do so, I estimate the following regression:
News negativity or CARit = βit·Murdoch×P ostit+γ·Murdoch+η·Controls+θt+ψj +it (3)
26 Additional evidence includes the resignation of Dow Jones director, Leslie Hill, in opposition to this deal and the appointment of Robert Thomson, described as Murdoch’s “best friend,” as the managing editor of the WSJ and editor-in-chief of Dow Jones Newswires (Auletta, K. 2011, “Murdoch’s Best Friend,” The New Yorker). 27 See also Besley and Prat(2006) for a theoretical model of ownership and media capture.
23 where Murdoch indicates acquirers connected to Mr. Murdoch (or News Corp), and Post is a dummy variable that equals one for deal announcement years after 2007.
To identify connections with Mr. Murdoch, I search in each acquirer’s 10K and proxy filings for potential (1) business relations, such as asset purchases or supply chain links, between the two companies, (2) common directors sitting on both Boards, and/or (3) other links such as political campaign seats. Following this exercise, I find that approximately 8% of acquirers are connected to Mr. Murdoch.
Because I include year dummies (θt) in (3), the estimate of Post is not independently iden- tified. The coefficient of interest here is Murdoch×Post, which captures the average treatment effect.
Requiring acquirers to conduct more than one bid both before and after 2007 substantially reduces the sample size. Hence, this analysis relies on only 392 observations. Despite the smaller sample size, the results are striking: As Table 7 shows, after the WSJ ownership change, news coverage for acquirers connected to Mr. Murdoch contains significantly fewer negative words than independent acquirers (columns 1 and 2). On the other hand, market reactions to the takeovers by these connected firms become more positive upon their M&A news publication (columns 3 and 4). An equally interesting observation is that, prior to the takeover, the WSJ coverage of these connected firms is actually more negative, as suggested by the stand-alone positive coefficient on Murdoch. This relation, however, is completely reversed after the ownership change.
Panel B of Table 7 assesses the robustness of the preceding findings by removing 7 ac- quisitions within the Entertainment and Printing/Publishing sectors. Looking at acquisitions unrelated to these industries addresses the concern that the WSJ control change has a di- rect impact on its competitors’ acquisition synergies. The results show that the documented treatment effects are not driven by firms within the Entertainment and Printing/Publishing industries.
Given that Mr. Murdoch’s takeover of the WSJ is reasonably exogenous to the writing
24 styles of journalists or to the expected deal synergies of other acquirers, I interpret the results from this analysis causally. This exercise hence highlights the role of connections to the media owner, and complements the other studies that relate media ownership to news bias, typically observed in the political coverage (e.g., Gilens and Hertzman, 2000).
7 Alternative Explanations and Robustness Tests
To conclude the analysis, I discuss additional alternative explanations to my findings and perform a number of robustness checks.
7.1 Selection in media coverage
One potential concern is that media coverage is not random. Selective news coverage might introduce bias in my sample, which contaminates the regression estimates. I study this pos- sibility by employing a Heckman(1979) two-stage procedure to account for such selection bias.
In the first stage, I model the probability of being covered by the WSJ. Using a probit regression, I find that firm and deal size are important determinants of media coverage (see the results in Appendix 2). These findings are perhaps unsurprising because larger firms or mergers attract greater attention (Solomon and Soltes, 2012). Other factors that influence media coverage include a firm’s Q, leverage, profitability, CEO age, and CEO’s founder status.
Based on these estimates, I calculate the Heckman(1979)’s lambda (or the inverse Mills ratio) and include this in my second-stage regression to control for potential endogenous sample selection. The second-stage model is otherwise specified in a similar way to those from columns 1 through 4 in Table 3. These results appear in Panel A of Table 8.
As is clear from Panel A, explicitly incorporating sample selection barely changes the results on slant. The point estimates of CONNECT remain comparable to the baseline regression.
25 The coefficient on Heckman’s lambda yields statistical insignificant estimates across these models, suggesting that selection bias does not exert a major influence on my findings.
7.2 Reverse causality
Despite the evidence presented in Section 4.1.1 and 4.2 that is inconsistent with the possibility of reverse causality, I perform additional tests to address this concern. According to this explanation, connected journalists are assigned to report on deals that have received better market reactions, which leads them to write more favorably. In other words, these connected reports are reflective of observed stock returns, rather than driving them.
In Panel B of Table 8, I regress lagged and future returns on the connection indicator.
Columns 1 and 2 show that observed returns on t−2 and t−1 are uncorrelated with the articles written by a connected reporter, which are only published on day t. However, returns following the news publication, such as those on t and t + 1, are significantly higher to firms covered by a connected journalist. Tracing the relation between return and connection dynamically, these tests again reject the causation running from a reversed direction.
7.3 Alternative measures of news tone
In Panel C, I replace the dependent variable with different ways to measure news negativity.
In the first four columns, I proxy for the negative sentiment using the logarithm of negative word count in a story. This measure alleviates the concern that readers focus on the absolute number of negative words rather than their proportion. From columns 5 to 8, I alternatively measure the news tone with the fraction of “net” negative words. That is, the proportion of negative vocabulary minus the positive words. The results with these alternative measures are largely in line with the baseline findings, suggesting that the results are not sensitive to how the slant is defined.
26 7.4 Finer measure of working relationships
One potential concern with the connection proxy, CONNECT WORK, is that it simply charac- terizes the frequency of past reporting but fails to capture whether the journalist has actually spoken with the employees in that firm. In Panel D, I strengthen this measure by requiring that the journalist quoted the acquirer CEO in the considered M&A article. Variable CON-
NECT Work&Cite flags such cases. This refined variable ensures that the working relationship is ongoing and that the journalists are likely to have actually spoken with the CEOs. The results, both of media slant (columns 1–2) and of CAR (column 3), remain largely unchanged with this alternative connection measure.
7.5 Target characteristics
To ensure that target firm characteristics do not bias the regression estimates, in Panel E
I additionally control for the target-related variables used in the M&A literature. These variables include size, Tobin’s Q, leverage, cash holdings, profitability, and pre-announcement stock runup. To conserve space, these estimates are not tabulated. The effects of connections are robust to adding these control variables.
7.6 Correlated taste
Gentzkow and Shapiro(2010) propose a theory that (political) media may cater to their readers’ demand. DellaVigna and Hermle(2017) empirically show that readers’ tastes can explain media slant using a large set of movie reviews. Under this hypothesis, newspapers may cater to their readers’ passion about a specific set of firms and therefore give them positive slant. However, readers’ demand for slant is unlikely to explain the news bias related to the journalist connections: Based on the results in columns 5 and 6 of Table 3, the merger fixed effects in these models should absorb all firm (or event) specific factors including the general public interests in that firm. Therefore, concerns over the demand side of slant are eliminated.
27 8 Discussion: Do Media Connections Matter to Firms?
In contrast to the common opinion that reporters should act the role of a gatekeeper (Shoe- maker and Vos, 2009), the analysis in this paper suggests that journalists’ social networks can shape media slant. Do these findings matter after all? For investors, news bias certainly hinders efficient information flow between the firms and the financial market. For firms, one may suspect that some firms reap actual benefits from being connected. In the context of this study, for example, do connected acquirers experience a better merger outcome? I ex- amine the effect of journalist connections on real deal outcomes. The results do not suggest that these connections are related to the merger premium, target returns, or the actual deal completion.28 However, the short-term distortion in the acquirer returns could still cause significant real wealth transfer among parties to the transaction: For example, in case of a
floating ratio stock merger, for a pre-determined deal premium, the connected acquirer can use their (temporarily) inflated shares to issue fewer new shares to match the bid price, thus avoiding dilution. Moreover, prior studies show that the acquirer CEO faces lower risks of litigation (Gong et al., 2008) or being fired (Lehn and Zhao, 2006) if the deal announcement returns are higher. Hence, media slant can have real consequences for the firms.
9 Conclusion
This paper builds on a burgeoning field of studies that document the very existence of media effects on the financial markets, especially those discovering predictable stock returns from journalists’ writing styles (Dougal et al., 2012). However, what influences a journalist’s tone in reporting? A potential answer to this question is the journalist’s social networks with the
firm. I focus on two types of social connections in this paper: direct working relationships and alumni networks with the CEO. 28 These results are available in the Internet Appendix.
28 Specifically, I ask two questions: (1) do journalists’ connections bias the news tone when a connected firm is covered, and (2) if so, do the connections have an impact on the firm’s stock returns? Using merger news articles published in the WSJ from 1997 through 2016, I find that the answer to both questions is “yes.” Such connections are associated with significantly less use of negative-sentiment words in the news article, and the upbeat optimism of friendly reporters is related to markedly better stock reactions, especially when the firms are featured on the front-page of the newspaper.
To credibly make a causal claim, I instrument the propensity of being covered by a con- nected journalist with the turnover of the connected reporter. The two-stage regression results yield a consistent relation of connected coverage, news tone, and stock returns. Thus, the im- mediate takeaway is that favorable media slant and the associated better stock reactions seem to be caused by the firm–journalist connections.
While prior studies have investigated deliberate marketing and spinning of news (e.g.,
Solomon, 2012), my findings suggest an indirect channel of influencing media through social networks. However, caution is needed in interpreting these results: while my analysis points to a clear media slant, it does not necessarily indicate media manipulation. Indeed, biased reporting might be a product of unconscious psychological working; various factors, such as institutional environment, could potentially affect journalistic independence. In this later regard, I show that an exogenous shock to the journalistic independence, induced by the News
Corp’s takeover of the WSJ, leads to more favorable coverage of firms that were connected to the new media owner and subsequent better stock returns to these connected firms.
Overall, the evidence suggests that social networks have a greater impact on the firms’ stock performance than one might have initially thought. Do firms take advantage of their media connections to achieve even manipulative communications? I leave this question to future research.
29 References
Ahern, K. R., Sosyura, D., 2014. Who writes the news? Corporate press releases during merger negotiations. Journal of Finance 69, 241–291.
Ahern, K. R., Sosyura, D., 2015. Rumor has it: Sensationalism in financial media. Review of Financial Studies 28, 2050–2093.
Amihud, Y., 2002. Illiquidity and stock returns: Cross-section and time-series effects. Journal of Financial Markets 5, 31–56.
Archer, A. M., Clinton, J., 2018. Changing owners, changing content: Does who owns the news matter for the news? Political Communication 35, 353–370.
Baker, M., Wurgler, J., 2006. Investor sentiment and the cross-section of stock returns. Journal of Finance 61, 1645–1680.
Barber, B. M., Odean, T., 2008. All that glitters: The effect of attention and news on the buying behavior of individual and institutional investors. Review of Financial Studies 21, 785–818.
Barber, B. M., Odean, T., 2013. The behavior of individual investors. Chapter 22 in the Handbook of the Economics of Finance, Volume 2, Part B, Elsevier.
Besley, T., Prat, A., 2006. Handcuffs for the grabbing hand?: Media capture and government ac- countability. American Economic Review 96, 720–736.
Betton, S., Eckbo, B. E., Thorburn, K. S., 2008. Corporate takeovers. In B. Espen Eckbo (ed.), Handbook of Corporate Finance, Empirical Corporate Finance, vol. 2, North–Holland: Elsevier.
Bodnaruk, A., Loughran, T., McDonald, B., 2015. Using 10-K text to gauge financial constraints. Journal of Financial and Quantitative Analysis 50, 623–646.
Cai, Y., Sevilir, M., 2012. Board connections and M&A transactions. Journal of Financial Economics 103, 327–349.
Call, A. C., Emett, S. A., Maksymov, E., Sharp, N. Y., 2018. Meet the press: Survey evidence on financial journalists as information intermediaries. Arizona State University working paper, available at SSRN: https://ssrn.com/abstract=3279453 .
Campbell, J. Y., Grossman, S. J., Wang, J., 1993. Trading volume and serial correlation in stock returns. Quarterly Journal of Economics 108, 905–939.
Cohen, L., Frazzini, A., Malloy, C., 2008. The small world of investing: Board connections and mutual fund returns. Journal of Political Economy 116, 951–979.
DellaVigna, S., Hermle, J., 2017. Does conflict of interest lead to biased coverage? Evidence from movie reviews. Review of Economic Studies 84, 1510–1550.
DellaVigna, S., Pollet, J. M., 2009. Investor inattention and Friday earnings announcements. Journal of Finance 64, 709–749.
DeLong, J. B., Shleifer, A., Summers, L. H., Waldmann, R. J., 1990. Noise trader risk in financial markets. Journal of Political Economy 98, 703–738.
30 Di Giuli, A., Laux, P. A., 2018. Access to public capital markets and board members’ media linkages. Working paper, available at SSRN: https://ssrn.com/abstract=2742312 .
Dougal, C., Engelberg, J. E., Garcia, D., Parsons, C. A., 2012. Journalists and the stock market. Review of Financial Studies 25, 639–679.
Dyck, A., Zingales, L., 2002. The corporate governance role of the media. In Peter K. Cornelius and Bruce Kogut (ed.), Corporate Governance and Capital Flows in a Global Economy, Oxford University Press.
El-Khatib, R., KathyFogel, Jandik, T., 2015. CEO network centrality and merger performance. Journal of Financial Economics 116, 349–382.
Engelberg, J. E., Gao, P., Parsons, C. A., 2012. Friends with money. Journal of Financial Economics 103, 169–188.
Engelberg, J. E., Parsons, C. A., 2011. The causal impact of media in financial markets. Journal of Finance 66, 67–97.
Erel, I., Liao, R. C., Weisbach, M. S., 2012. Determinants of cross-border mergers and acquisitions. Journal of Finance 67, 1045–1082.
Fang, L., Peress, J., 2009. Media coverage and the cross-section of stock returns. Journal of Finance 64, 2023–2052.
Fedyk, A., 2018. Front page news: The effect of news positioning on financial mar- kets. Working paper, available at https://scholar.harvard.edu/fedyk/publications/ front-page-news-effect-news-positioning-financial-markets .
Fracassi, C., 2016. Corporate finance policies and social networks. Management Science 63, 2420– 2438.
Gaspar, J.-M., Massa, M., Matos, P., 2005. Shareholder investment horizons and the market for corporate control. Journal of Financial Economics 76, 135–165.
Gentzkow, M., Shapiro, J. M., 2010. What drives media slant? Evidence from U.S. daily newspapers. Econometrica 78, 35–71.
Gilens, M., Hertzman, C., 2000. Corporate ownership and news bias: Newspaper coverage of the 1996 Telecommunications Act. Journal of Politics 62, 369–386.
Goldman, E., Martel, J., Schneemeier, J., 2019. A theory of financial media. Working paper, available at SSRN: https://ssrn.com/abstract=3457591 .
Golez, B., Karapandza, R., 2019. Home-Country Media Slant and Equity Prices. Working paper, available at SSRN: https://ssrn.com/abstract=3396726 .
Gompers, P. A., Mukharlyamov, V., Xuan, Y., 2016. The cost of friendship. Journal of Financial Economics 119, 626–644.
Gong, G., Louis, H., Sun, A. X., 2008. Earnings management, lawsuits, and stock-for-stock acquirers’ market performance. Journal of Accounting and Economics 46, 62–77.
31 Gurun, U. G., 2016. Benefits of publicity. Working paper, available at SSRN: https://ssrn.com/ abstract=2706969 .
Gurun, U. G., Butler, A. W., 2012. Don’t believe the hype: Local media slant, local advertising, and firm value. Journal of Finance 67, 561–598.
Hartzmark, S. M., Shue, K., 2017. A tough act to follow: Contrast effects in financial markets. Journal of Finance forthcoming.
Heckman, J., 1979. Sample slection bias as a specification error. Econometrica 47, 153–161.
Huberman, G., Regev, T., 2001. Contagious speculation and a cure for cancer: A nonevent that made stock prices soar. Journal of Finance 56, 387–396.
Jiang, W., 2017. Have instrumental variables brought us closer to the truth. Review of Corporate Finance Studies 6, 127–140.
Kumar, A., 2009. Hard-to-value stocks, behavioral biases, and informed trading. Journal of Financial and Quantitative Analysis 44, 1375–1401.
Lehn, K. M., Zhao, M., 2006. CEO turnover after acquisitions: Are bad bidders fired? Journal of Finance 61, 1759–1811.
Leuz, C., Meyer, S., Muhn, M., Soltes, E., Hackethal, A., 2017. Who falls prey to the Wolf of Wall Street? Investor participation in market manipulation. NBER Working Paper No. 24083 .
Loughran, T., McDonald, B., 2011. When is a liability not a liability? Textual analysis, dictionaries, and 10-Ks. Journal of Finance 66, 35–65.
Loughran, T., McDonald, B., 2016. Textual analysis in accounting and finance: A survey. Journal of Accounting Research 54, 1187–1230.
Loughran, T., McDonald, B., 2017. The use of EDGAR filings by investors. Journal of Behavioral Finance 18, 231–248.
Masulis, R. W., Wang, C., Xie, F., 2007. Corporate governance and acquirer returns. Journal of Finance 62, 1851–1889.
Mitchell, M. L., Mulherin, J. H., 1994. The impact of public information on the stock market. Journal of Finance 49, 923–950.
Peress, J., 2014. The media and the diffusion of information in financial markets: Evidence from newspaper strikes. Journal of Finance 69, 2007–2043.
Pound, J., Zeckhauser, R. J., 1990. Clearly heard on the Street: The effect of takeover rumors on stock prices. Journal of Business 63, 291–308.
Roll, R., 1988. R-squared. Journal of Finance 43, 541–566.
Rozin, P., Royzman, E. B., 2001. Negativity bias, negativity dominance, and contagion. Personality and Social Psychology Review 5, 296–320.
32 Schmidt, B., 2015. Costs and benefits of friendly boards during mergers and acquisitions. Journal of Financial Economics 117, 424–447.
Shoemaker, P. J., Vos, T., 2009. Gatekeeping Theory. 1st Edition, New York: Routledge.
Shue, K., 2013. Executive networks and firm policies: Evidence from the random assignment of MBA peers. Review of Financial Studies 26, 1401–1442.
Solomon, D. H., 2012. Selective publicity and stock prices. Journal of Finance 67, 599–638.
Solomon, D. H., Soltes, E., 2012. Managerial control of business press coverage. Working paper, available at SSRN: https://ssrn.com/abstract=1918138 .
Solomon, D. H., Soltes, E., Sosyura, D., 2014. Winners in the spotlight: Media coverage of fund holdings as a driver of flows. Journal of Financial Economics 113, 53–72.
Tetlock, P. C., 2007. Giving content to investor sentiment: The role of media in the stock market. Journal of Finance 62, 1139–1168.
Tetlock, P. C., 2011. All the news that’s fit to reprint: Do investors react to stale information? Review of Financial Studies 24, 1481–1512.
Tetlock, P. C., 2015. “The role of media in finance” in S.P. Anderson, D. Stromberg, and J. Waldfogel (Eds.). Handbook of Media Economicse, Vol 1B, Chapter 18, 701–721. Oxford: Elsevier.
Tetlock, P. C., Saar-Tsechansky, M., Macskassy, S., 2008. More than words: Quantifying language to measure firms’ fundamentals. Journal of Finance 63, 1437–1467.
Wagner, M. W., Collins, T. P., 2014. Does ownership matter?: The case of Rupert Murdoch’s purchase of the Wall Street Journal. Journalism Practice 8, 758–771.
Zhang, X. F., 2006. Information uncertainty and stock returns. Journal of Finance 61, 105–137.
33 Table 1: Sample M&A Articles by Calendar Time
This table lists the count of M&A news articles that appeared on the Wall Street Journal by year and by month. The column “SDC” presents the total number of M&A deals recorded by SDC which fulfill the sample selection criteria described in Section 2. The col- umn “WSJ sample” presents the number of transactions covered by the Wall Street Journal. The column “Percent” presents the percentage of the ”WSJ sample” out of the SDC sample.
SDC WSJ Sample Percent WSJ/SDC Panel A: Year 1997 226 88 0.39 1998 252 100 0.40 1999 285 128 0.45 2000 206 95 0.46 2001 158 58 0.37 2002 80 14 0.18 2003 102 28 0.27 2004 97 41 0.42 2005 114 56 0.49 2006 107 55 0.51 2007 102 48 0.47 2008 79 38 0.48 2009 71 42 0.59 2010 81 46 0.57 2011 57 35 0.61 2012 58 34 0.59 2013 56 32 0.57 2014 73 59 0.81 2015 101 74 0.73 2016 85 60 0.71 Total 2,390 1,131 0.47 Panel B: Month January 162 64 0.40 February 190 92 0.48 March 187 88 0.47 April 200 94 0.47 May 233 113 0.48 June 234 115 0.49 July 205 107 0.52 August 199 93 0.47 September 157 73 0.46 October 228 98 0.43 November 199 97 0.49 December 196 97 0.49
34 Table 2: Summary Statistics
This table describes the summary statistics for the main variables used in the empirical tests. Definitions of these variables are in Appendix 1.
Connected Connected All sample Independent (University) (Work) N Mean S.D. Mean S.D. Mean S.D. Mean S.D. Journalist connections: CONNECT WORK 1131 0.261 0.439 CONNECT UNIVERSITY 1131 0.023 0.150 Deal characteristics: Bidder CAR[0,1] 1131 -0.007 0.070 0.006 0.099 0.005 0.058 -0.011 0.073 All cash deal 1131 0.316 0.465 0.231 0.430 0.363 0.482 0.301 0.459 All stock deal 1131 0.265 0.442 0.269 0.452 0.237 0.426 0.276 0.447 Diversifying deal 1131 0.332 0.471 0.308 0.471 0.369 0.483 0.318 0.466 Relative deal size 1131 0.538 0.707 0.547 0.521 0.325 0.571 0.611 0.738 Hostile 1131 0.030 0.171 0.077 0.272 0.020 0.141 0.033 0.179 Unsolicited 1131 0.047 0.211 0.077 0.272 0.041 0.198 0.048 0.213 Completed 1131 0.895 0.307 0.846 0.368 0.929 0.258 0.884 0.320 Journalist and news variables: News negativity (%) 1131 1.396 0.868 1.393 0.819 1.269 0.712 1.441 0.914 Same city 1131 0.227 0.419 0.269 0.452 0.319 0.467 0.195 0.397 Female 1131 0.441 0.497 0.538 0.508 0.478 0.500 0.424 0.495 Jounalist tenure (month) 1131 67 59 52 44 73 58 65 60 Number WSJ articles (log.) 1131 5.292 1.218 5.257 1.330 5.626 0.896 5.177 1.288 Industry expert 1131 0.477 0.500 0.385 0.496 0.600 0.491 0.437 0.496 Jounalism degree 1131 0.781 0.414 0.769 0.430 0.803 0.398 0.774 0.418 Article word count (log.) 1131 6.100 0.500 6.220 0.580 6.333 0.486 6.331 0.491 Firm characteristics: Firm size (log.) 1131 8.592 1.761 8.605 1.752 9.764 1.523 8.171 1.651 Tobin’s Q 1131 2.764 2.949 3.309 4.359 3.057 3.128 2.651 2.810 Firm leverage 1131 0.149 0.134 0.143 0.152 0.122 0.114 0.158 0.138 Firm cash 1131 0.157 0.182 0.209 0.220 0.158 0.166 0.158 0.188 Firm profitability 1131 0.043 0.119 0.032 0.089 0.051 0.127 0.041 0.117 CEO age 1131 54 7.312 54 9 55 7 54 7 CEO founder 1131 0.131 0.337 0.192 0.402 0.108 0.312 0.139 0.346 Dual 1131 0.656 0.475 0.615 0.496 0.695 0.461 0.644 0.479 Board size (log.) 1131 2.280 0.278 2.241 0.238 2.405 0.242 2.236 0.277 Classified board 1131 0.447 0.497 0.346 0.485 0.339 0.474 0.488 0.500
35 Table 3: Journalist Connections and News Tone
OLS regressions of news negativity. The dependent variable is News Negativity, measured as the fraction of negative words in the text of a news article. Columns 1 to 4 examine all news articles that appear in the WSJ; Columns 5 to 6 pool merger stories covered by both the WSJ and NYTimes. Definitions for all variables are in Appendix 1. Models in columns 1 through 4 include time (i.e., merger year-month), firm industry, state, and the WSJ office location fixed effects; columns 2 and 4 additionally control for journalist fixed effects. In models 5 and 6, newspaper and deal fixed effects are included. Standard errors, double-clustered by deal year-month and by industry, are reported in parentheses. Symbols ***, **, and * indicate statistical significance at the 1%, 5%, and 10% level, respectively.
WSJ Sample WSJ + NYTimes 1 2 3 4 5 6
CONNECT WORK -0.3299*** -0.2665** -0.2301*** (0.083) (0.126) (0.081) CONNECT UNIVERSITY -0.1841** -0.3067** -0.4710** (0.080) (0.150) (0.238) Stock returns before publication date: L1 -0.3921 -0.4566 -0.5020 -0.4136 – – (0.426) (0.679) (0.434) (0.704) L2 -1.2281 -2.5026* -1.0194 -2.2552* – – (1.228) (1.345) (1.276) (1.341) L3 -0.0339 0.3476 -0.1782 0.1662 – – (0.879) (1.211) (0.806) (1.165) Same city -0.0598 -0.1451 -0.0865 -0.1688 0.0764 0.0603 (0.088) (0.159) (0.093) (0.163) (0.107) (0.107) Female -0.0209 0.1844 -0.0366 0.1888 -0.0742 -0.0564 (0.050) (0.189) (0.063) (0.187) (0.064) (0.064) Jounalist tenure -0.0004 0.0021 -0.0004 0.0018 -0.0006 -0.0005 (0.001) (0.001) (0.001) (0.001) (0.000) (0.000) Number WSJ articles 0.0516 -0.0446 0.0346 -0.0398 0.0389 0.0277 (0.038) (0.076) (0.040) (0.077) (0.030) (0.030) Industry expert -0.0915 -0.2614** -0.1183 -0.2776** -0.0832 -0.1026* (0.085) (0.109) (0.086) (0.113) (0.060) (0.060) Jounalism degree 0.0382 0.2518 0.0417 0.2264 -0.0014 0.0017 (0.063) (0.223) (0.059) (0.235) (0.073) (0.074) Relative deal size 0.0622 0.1567** 0.0613 0.1564* – – (0.050) (0.078) (0.052) (0.081) Toehold 0.0075 0.0023 0.0063 0.0001 – – (0.006) (0.007) (0.006) (0.008) Hostile 0.7448*** 0.7521*** 0.7963*** 0.7962*** – – (0.197) (0.283) (0.208) (0.283) Unsolicited 0.5873*** 0.4513** 0.5934*** 0.4716** – – (0.137) (0.186) (0.146) (0.190) Diversifying deal 0.0347 -0.0867 0.0453 -0.0788 – – (0.072) (0.100) (0.074) (0.101) Continued on next page
36 All cash deal 0.1150 0.1957 0.1399* 0.2015 – – (0.077) (0.126) (0.080) (0.130) All stock deal 0.0352 0.1191 0.0593** 0.1375* – – (0.037) (0.083) (0.030) (0.076) Article length 0.0003*** 0.0004* 0.0003*** 0.0004* – – (0.000) (0.000) (0.000) (0.000) Firm size 0.0516 0.0990 0.0100 0.0655 – – (0.036) (0.072) (0.037) (0.064) Tobin’s Q -0.0062 -0.0250* -0.0116 -0.0277* – – (0.021) (0.015) (0.020) (0.015) Firm leverage 0.2078 -0.4250 0.3129 -0.3556 – – (0.263) (0.364) (0.261) (0.369) Firm cash 0.2762 0.9165** 0.2594 0.9009** – – (0.384) (0.412) (0.419) (0.420) Firm profitability -0.6992* -1.1037** -0.5781 -0.9853** – – (0.403) (0.489) (0.389) (0.462) CEO age 0.0054 0.0111* 0.0055 0.0114* – – (0.004) (0.006) (0.004) (0.007) CEO founder -0.0588 0.0204 -0.0546 0.0040 – – (0.081) (0.116) (0.076) (0.120) Board size -0.0479 -0.0579 -0.0514 -0.0539 – – (0.171) (0.225) (0.170) (0.222) Dual -0.1015 -0.1896** -0.1041 -0.2039** – – (0.080) (0.090) (0.084) (0.093) Classified board -0.0585 -0.0458 -0.0623 -0.0463 – – (0.075) (0.096) (0.074) (0.094) Constant 2.1467* 0.8035 3.0311** 1.4064 1.3017*** 1.3282** (1.206) (1.357) (1.217) (1.343) (0.178) (0.238) Time FE × × × × Firm industry FE × × × × Firm state FE × × × × WSJ office FE × × × × Journalist FE × × Newspaper FE × × Deal FE × × Observations 1,131 979 1,131 979 1,012 1,012 R2 0.406 0.632 0.393 0.627 0.677 0.674
37 Table 4: 2-Stage Least Squares Regressions
Two-stage least squares (2SLS) regressions where CONNECT WORK is instrumented with connected journalists’ turnovers. The sample includes bidders who have conducted at least 2 M&A deals and who have at least one WSJ journalist connection measured with working relationships. In the first stage (columns 1), CONNECT WORK is instrumented by the connected-journalist turnover in the quarter of M&A announcement. In the second stage (columns 2), news negativity is regressed on the instrumented CONNECT WORK. Columns 3 report the reduced form regressions, where news negativity is regressed directly on the instrument variable. Control variables include: acquirer stock returns from t − 3 till -1 (where day 0 is the news publication date), deal relative size, toehold, hostile, unsolicited, diversifying deal, all cash deal, all stock deal, firm size, Tobin’s Q, leverage, cash, profitability, CEO age, CEO founder, dual, board size, classified board. Definitions for all variables are in Appendix 1. In all regressions, announcement year and industry dummies are included. Standard errors, which are reported in parentheses, are double-clustered by by industry and by time. Symbols ***, **, and * indicate statistical significance at the 1%, 5%, and 10% level, respectively.
1 2 3 1st stage 2SLS reduced form
Turnover -0.6520*** 0.4648** (0.111) (0.228) CONNECT WORK -0.7129*** (0.251) Controls × × × Time FE × × × Firm industry FE × × × Observations 318 318 318 R2 0.199 0.327 0.332 F-statistic 34.34 Sargen χ2 statistic (p-val) 1.68 (0.19)
38 Table 5: Journalist Connections and Stock Returns
Regressions of bidders’ cumulative abnormal returns (CAR) from day 0 until day 1, where day 0 is the WSJ article publication date. The sample covers the WSJ news articles described in Table 2. CARs are calculated with a one-factor market model where the market returns are proxied by the value-weighted CRSP index. Market model parameters are estimated over a 200-day non-missing-value window ending 31 days before the announcement date. Columns 1 to 4 employ OLS regressions, and columns 5 to 6 utilize 2SLS regressions. Column 5 reports the second stage 2SLS results where CONNECT WORK is instrumented with connected journalists’ turnover (the first stage results appear in Table 4). Column 6 reports the reduced form regression. Definitions for all variables are in Appendix 1. In all regressions, time (i.e., announcement year-month (columns 1–4) or year (columns 5–6)) and industry dummies are included. Standard errors, which are reported in parentheses, are double-clustered by time and by industry. Symbols ***, **, and * indicate statistical significance at the 1%, 5%, and 10% level, respectively.
OLS IV 1 2 3 4 5 6 2SLS reduced form
CONNECT WORK 0.0147** -0.0067 0.0408** (0.007) (0.008) (0.019) CONNECT UNIVERSITY 0.0031 -0.0438* (0.021) (0.026) CONNECT WORK×Front page 0.0306*** (0.011) CONNECT UNIVERSITY×Front page 0.0893*** (0.034) Instrument: Turnover -0.0266* (0.014) Front page -0.0051 0.0007 (0.006) (0.006) Relative deal size -0.0116 -0.0114 -0.0112 -0.0114 0.0000 0.0036 (0.008) (0.008) (0.008) (0.008) (0.009) (0.012) Toehold -0.0007 -0.0006 -0.0008 -0.0006 -0.0010** -0.0007 (0.001) (0.001) (0.000) (0.001) (0.000) (0.001) Hostile -0.0315*** -0.0325*** -0.0312*** -0.0346*** -0.0241** -0.0245*** (0.011) (0.012) (0.011) (0.012) (0.011) (0.009) Unsolicited 0.0079 0.0082 0.0057 0.0072 -0.0220 -0.0210 (0.013) (0.014) (0.013) (0.014) (0.023) (0.030) Diversifying deal -0.0024 -0.0027 -0.0020 -0.0030 0.0022 0.0011 (0.006) (0.006) (0.006) (0.007) (0.004) (0.006) All cash deal 0.0134* 0.0120 0.0136* 0.0123 0.0104 0.0079 (0.007) (0.007) (0.007) (0.008) (0.011) (0.011) All stock deal -0.0065 -0.0072 -0.0059 -0.0071 0.0023 0.0007 (0.007) (0.006) (0.006) (0.007) (0.002) (0.005) Continued on next page
39 Firm size 0.0012 0.0032 0.0013 0.0032 -0.0009 0.0015 (0.002) (0.002) (0.002) (0.002) (0.004) (0.004) Tobin’s Q -0.0000 0.0002 -0.0000 0.0004 0.0024 0.0029* (0.001) (0.001) (0.001) (0.001) (0.002) (0.002) Firm leverage 0.0223 0.0169 0.0251 0.0189 0.0213 0.0122 (0.035) (0.034) (0.034) (0.034) (0.075) (0.089) Firm cash -0.0458*** -0.0434*** -0.0437*** -0.0402** -0.0087 -0.0045 (0.016) (0.016) (0.017) (0.017) (0.026) (0.034) Firm profitability -0.0142 -0.0188 -0.0119 -0.0194 -0.1082*** -0.1208*** (0.034) (0.035) (0.033) (0.036) (0.027) (0.033) CEO age -0.0001 -0.0001 -0.0001 -0.0001 -0.0005 -0.0004 (0.000) (0.000) (0.000) (0.000) (0.000) (0.001) CEO founder -0.0072 -0.0076 -0.0063 -0.0076 -0.0058 -0.0093 (0.013) (0.012) (0.013) (0.012) (0.010) (0.009) Board size -0.0069 -0.0062 -0.0058 -0.0055 0.0019 -0.0065 (0.011) (0.011) (0.011) (0.010) (0.020) (0.018) Dual -0.0069 -0.0070 -0.0070 -0.0068 -0.0019 -0.0009 (0.006) (0.006) (0.005) (0.006) (0.005) (0.006) Classified board -0.0008 -0.0009 -0.0003 0.0001 -0.0195*** -0.0187** (0.004) (0.004) (0.004) (0.004) (0.004) (0.009) Stock returns before publication date: L1 -0.1084*** -0.1039*** -0.1124*** -0.1013** -0.0049 0.0233 (0.039) (0.039) (0.039) (0.039) (0.063) (0.089) L2 -0.0831 -0.0872 -0.0884 -0.0991 -0.0168 -0.0807 (0.089) (0.091) (0.088) (0.091) (0.164) (0.214) L3 0.0749 0.0803 0.0711 0.0866 0.1278 0.1536 (0.091) (0.094) (0.091) (0.094) (0.185) (0.221) Constant 0.0293 0.0039 0.0220 -0.0102 0.0332 0.0604 (0.032) (0.028) (0.025) (0.020) (0.033) (0.043) Time FE × × × × × × Firm industry FE × × × × × × Observations 1,131 1,131 1,131 1,131 318 318 R2 0.302 0.298 0.308 0.305 0.198 0.204
40 Table 6: Additional Support for Connection Effects
This table provides further evidence of connection effects on stock returns. Panel A reports the results on interaction effects between journalist connection (CONNECT WORK) and proxies for the arbitrage opportunity. The dependent variable is bidders’ cumulative abnormal returns (CAR) from the article publication day (day 0) until the following day. Proxies for the arbitrage opportunity include: analysts, firm size, institution ownership, and illiquidity using Amihud(2002)’s measure. Panel B shows the relation between CONNECT WORK and long-run return reactions, using both the full sample described in Table 2 (Panel B.1) and the completed merger subsample (Panel B.2). Columns 1 and 2 in Panel B replicate the results of bidder’s CAR over [0,1]. The dependent variable in columns 3 and 4 is bidders’ CAR over [2,40]. The dependent variable in columns 5 and 6 is bidders’ CAR over [0,40]. The dependent variable in columns 7 and 8 is CAR from the day before the deal announcement until the deal completion date (for the subsample of completed transactions). Control variables include: relative deal size, toehold, hostile, unsolicited, diversifying deal, all cash deal, all stock deal, firm size, Tobin’s Q, leverage, cash, profitability, CEO age, CEO founder, dual, board size, classified board, and three lagged returns. Definitions for all variables are in Appendix 1. In all regressions, time (i.e., year-month) and industry dummies are included. Standard errors are reported in parentheses and are double-clustered by deal year-month and by industry. Symbols ***, **, and * indicate statistical significance at the 1%, 5%, and 10% level, respectively.
Panel A. Interaction effects CAR[0,+1] 1 2 3 4
CONNECT WORK×Analysts -0.0012*** (0.000) CONNECT WORK×Firm size -0.0154*** (0.004) CONNECT WORK×Institution ownership -0.0529** (0.025) CONNECT WORK×Iliquidity 0.1252*** (0.013) CONNECT WORK 0.0375*** 0.1592*** 0.0464*** 0.0119* (0.012) (0.041) (0.017) (0.006) Analysts 0.0003 (0.000) Firm size 0.0044* (0.003) Institution ownership 0.0142 (0.011) Iliquidity -0.0038 (0.003) Controls × × × × Time FE × × × × Firm industry FE × × × × Joint effects of CONNECT WORK: ∂f()/∂CONNECT WORK 0.0187*** 0.0264*** 0.0135** 0.0191*** (0.007) (0.007) (0.006) (0.006) Observations 1,131 1,131 1,131 1,131 R2 0.308 0.318 0.307 0.318 41 Table 6 (continued): Additional Support for Connection Effects
Panel B. Long-run price correction
Panel B.1: Full sample CAR[0,1] CAR[2,40] CAR[0,40] CAR[-1,complete] 1 2 3 4 5 6 7 8
CONNECT WORK 0.0189*** 0.0147** -0.0223** -0.0351** -0.0008 -0.0175 -0.0110 0.0018 (0.006) (0.007) (0.011) (0.014) (0.013) (0.017) (0.020) (0.025) Controls × × × × Time FE × × × × × × × × Firm industry FE × × × × × × × × Observations 1,131 1,131 1,131 1,131 1,131 1,131 852 852 R2 0.265 0.302 0.247 0.278 0.229 0.262 0.281 0.301
Panel B.2: Excluding withdrawn transactions CAR[0,1] CAR[2,40] CAR[0,40] CAR[-1,complete] 1 2 3 4 5 6 7 8
CONNECT WORK 0.0149** 0.0131* -0.0330** -0.0404*** -0.0150 -0.0243 -0.0110 0.0018 (0.006) (0.007) (0.013) (0.014) (0.015) (0.016) (0.020) (0.025) Controls × × × × Time FE × × × × × × × × Firm industry FE × × × × × × × × Observations 1,012 1,012 1,012 1,012 1,012 1,012 852 852 R2 0.280 0.317 0.257 0.282 0.242 0.272 0.281 0.301
42 Table 7: Connections with Media Owner: WSJ’s Acquisition by News Corp
This table reports the results of difference-in-difference analysis, using a sample of repetitive acquirers who have conducted M&A deals before and after 2007. The dependent variable in columns 1 and 2 is News negativity calculated with equation (1). The dependent variable in columns 3 and 4 is bidder’s CAR[0, +1], where day 0 is the WSJ article publication date. Variable Murdoch is a dummy variable that equals one if a bidder is connected to News Corp or Rupert Murdoch. Variable Post takes a value of one for years from 2008 onward. Panel A includes all deals. Panel B excludes transactions in the Entertainment and Publishing (Printing) sectors (based on Fama-French 48 industries). Control variables include deal relative size, toehold, hostile, unsolicited, diversifying deal, all cash deal, all stock deal, firm size, Tobin’s Q, leverage, cash, profitability, CEO age, CEO founder, dual, board size, classified board. Definitions for all variables are in Appendix 1. In all regressions, time (i.e., announcement year), industry, acquirer state, and WSJ office dummies are included. Standard errors appear in parentheses and are double-clustered by year and by industry. ***, **, and * indicate statistical significance at the 1%, 5%, and 10% level, respectively.
Panel A. All available transactions News Negativity CAR[0,+1] 1 2 3 4
Murdoch × Post -0.8578*** -0.9240*** 0.0235* 0.0318** (0.228) (0.186) (0.013) (0.016) Murdoch 0.4998*** 0.4477** 0.0020 -0.0103 (0.182) (0.191) (0.009) (0.016) Controls × × Time FE × × × × Firm industry FE × × × × Firm state FE × × × × WSJ office FE × × × × Observations 392 392 392 392 R2 0.311 0.389 0.304 0.363
Panel B. Excluding M&As in Entertainment and Publishing (Printing) sectors News Negativity CAR[0,+1] 1 2 3 4
Murdoch × Post -0.9128*** -0.9332*** 0.0321*** 0.0373*** (0.200) (0.142) (0.002) (0.014) Murdoch 0.5604*** 0.4634** -0.0028 -0.0123 (0.130) (0.182) (0.009) (0.018) Controls × × Time FE × × × × Firm industry FE × × × × Firm state FE × × × × WSJ office FE × × × × Observations 385 385 385 385 R2 0.312 0.389 0.281 0.345
43 Table 8: Robustness Tests
Robustness tests for the results on news tone and for the results on bidder returns. Panel A uses the Heckman (1979) two-stage procedure to account for possible sample selection bias. The dependent variable is News nega- tivity. Heckman’s lambda (or the inverse Mills ratio) and the same set of control variables and fixed effects as in columns 1 through 4 of Table 3 are controlled for. Panel B examines the possibility of reverse causality. Bidder stock returns from t−2 until t+1 (where t = 0 is the WSJ publication date) are regressed on CONNECT WORK. The same set of control variables and fixed effects as in column 1 of Table 5 are included. Panel C uses alternative measures of news negativity as the dependent variable, where log(# Negative words) (columns 1–4) is the natural logarithm of the number of negative words in a news article, and Net negativity (columns 5–8) is the percentage of negative words minus the percentage of positive words. Panel D alternatively defines CONNECT WORK as the journalists who have a working relationship with the firm and who have also cited the acquirer CEO in their reports. The variable CONNECT WORK&CITE flags such cases. Panel E additionally control for target firm characteristics. Definitions for all variables are in Appendix 1. Fixed effects are indicated at the end of each panel. Standard errors are reported in parentheses and are double-clustered by deal year-month and by industry. Symbols ***, **, and * indicate statistical significance at the 1%, 5%, and 10% level, respectively.
Panel A. Selection in news coverage (Heckman’s model) News negativity 1 2 3 4
CONNECT WORK -0.3320*** -0.2631** (0.083) (0.124) CONNECT UNIVERSITY -0.1846** -0.3013** (0.079) (0.149) Heckman’s lambda 1.3828 -1.3300 0.8415 -1.9730 (2.248) (4.103) (2.172) (4.250) Controls × × × × Time FE × × × × Industry and state FE × × × × WSJ office FE × × × × Journalist FE × × Observations 1,131 979 1,131 979 R2 0.406 0.632 0.394 0.627
Panel B. Reverse causality Return t − 2 Return t − 1 Return t Return t + 1 1 2 3 4
CONNECT WORK -0.0016 0.0066 0.0128* 0.0049* (0.003) (0.004) (0.008) (0.003) Controls × × × × Time FE × × × × Firm industry FE × × × × Observations 1,131 1,131 1,131 1,131 R2 0.223 0.279 0.293 0.203 Continued on next page
44 Panel C. Alternative measures of news tone log(#Negative words) Net negativity 1 2 3 4 5 6 7 8
CONNECT WORK -0.2613*** -0.2123** -0.2606*** -0.1810 (0.054) (0.095) (0.076) (0.129) CONNECT UNIVERSITY -0.1808* -0.2110** -0.1380 -0.5802*** (0.097) (0.102) (0.137) (0.184) Controls × × × × × × × × Time FE × × × × × × × × Industry and state FE × × × × × × × × WSJ office FE × × × × × × × × Journalist FE × × × × Observations 1,087 944 1,087 944 1,131 979 1,131 979 R2 0.669 0.778 0.660 0.774 0.385 0.627 0.380 0.628
Panel D. Work connections & cites of CEOs News negativity CAR[0,1] 1 2 3
CONNECT Work & Cite -0.3204*** -0.2349* 0.0118* (0.082) (0.142) (0.007) Controls × × × Time FE × × × Firm industry and state FE × × × WSJ office FE × × Journalist FE × Observations 1,131 979 1,131 R2 0.405 0.630 0.343
Panel E. Additional controls for target characteristics News negativity CAR[0,1] 1 2 3 4 5 6
CONNECT WORK -0.3144*** -0.2109* 0.0122* (0.084) (0.115) (0.007) CONNECT UNIVERSITY -0.1677* -0.3123** 0.0014 (0.094) (0.152) (0.020) Target-firm controls × × × × × × Acquirer and deal controls × × × × × × Time FE × × × × × × Firm industry and state FE × × × × × × WSJ office FE × × × × Journalist FE × × Observations 1,115 964 1,115 964 1,115 1,115 R2 0.415 0.642 0.404 0.640 0.358 0.355
45 Appendix
1. Variable definition Panel A: Journalist and news related variables News negativity The fraction of negative words in the text of a news article. The negative words are defined in the Loughran and McDonald 2016 Master Dictionary. (Source: http://sraf.nd.edu/) CONNECT WORK Indicator variable that identifies a direct working relation between the jour- nalist and the firm. This variable equals one if at least one journalist has written two or more stories (unrelated to the considered merger) about the acquirer during the previous calendar year. (Source: WSJ) CONNECT UNIVERSITY Indicator variable that equals one if the acquirer CEO and (at least one) journalist graduated from the same university. (Source: proxy filings, Linkedin, Bloomberg, WSJ, Muck Rack) Same city Indicator variable that equals one if the acquirer CEO and (at least one) journalist are located in the same metropolitan area. (Source: proxy filings, Linkedin, WSJ, Muck Rack) Female Indicator variable that equals one if the news article has at least one female journalist. (Source: Linkedin, WSJ) Journalist tenure The average number of months that a coauthor works at the Wall Street Journal. (Source: Linkedin, WSJ) Number WSJ articles The mean of total number of articles that a journalist writes for the Wall Street Journal until the date of the acquisition announcement. (Source: WSJ) Industry expert Indicator variable that equals one if any coauthor of the article has a primary focus on the industry of the acquirer firm. (Source: Linkedin, WSJ) Journalism degree Indicator variable that equals one if any journalist has graduated with a major in journalism or in literature. (Source: Linkedin) Article length Total number of words in the news article. (Source: WSJ) Front page Dummy variable equals one if the news article appears on the front page of the Wall Street Journal. (Source: Factiva) Panel B: M&A deal related variables Bidder CAR [-1,+1] Bidder’s three-day cumulative abnormal return around announcement date calculated using the one-factor market model. The market model parame- ters are estimated over the (-230, -31) trading days prior to the announce- ment date with value-weighted CRSP market index. (Source: CRSP) Relative deal size Deal value reported by SDC scaled by the bidder’s market value of equity four days prior to the announcement. (Source: SDC, CRSP) Unsolicited Dummy variable equals one if the transaction is classified as unsolicited in the SDC, zero otherwise. (Source: SDC) Hostile Dummy variable equals one if the transaction is classified as hostile in the SDC, zero otherwise. (Source: SDC) Toehold Bidder’s ownership in the target prior to the merger announcement. (Source: SDC) Continued on next page
46 All cash deal Dummy variable equals one for purely cash-financed transactions, zero oth- erwise. (Source: SDC) All stock deal Dummy variable equals one for purely equity-financed transactions, zero otherwise. (Source: SDC) Diversifying deal Dummy variable equals one if the acquirer and the target are not in the same Fama-French 49 industry, zero otherwise. (Source: SDC) Panel C: Firm level variables Firm size The logarithm of book value of total assets. (Source: Compustat) Tobin’s Q Market value of assets (book value of assets minus book value of equity plus market value of equity) over book value of assets. (Source: CRSP, Compustat) Firm leverage Book value of debts over market value of total assets. (Source: CRSP, Compustat) Firm cash Cash or cash equivalent scaled by the book value of total assets. (Source: Compustat) Firm profitability Net income scaled by the book value of total assets. (Source: Compustat) Analysts Number of analysts who make quarterly earnings forecasts for a given bidding firm in the quarter prior to the M&A announcement. (Source: I/B/E/S) Institutional ownership Percent of shares owned by institutional investors. (Source: Thomson- Reuters Institutional Holdings Database) Illiquidity Illiquidity measure defined following Amihud(2002): 1 /N P |Ri,t| , V OLDi,t t∈[m−2,m] where R is the daily stock return, VOLD is the daily dollar trading volume, N is the number of days with available trading data in the three months up to the M&A announcement month (month m). (Source: CRSP) CEO age Age of the acquirer CEO as reported in the proxy filing in the year of M&A announcement. (Source: proxy filings) CEO founder Indicator variable that equals one if the acquirer CEO is the founder of the firm. (Source: proxy filings) Dual Indicator variable that equals one if the acquirer CEO is also the Chairman of the Board. (Source: proxy filings) Board size The total number of directors reported in the annual proxy filing prior to the M&A announcement. (Source: proxy filings) Classified board Indicator variable that equals one if the bidder’s board is staggered. (Source: proxy filings)
47 2. Determinants of WSJ Coverage
Probit regression that estimates the first stage of Heckman (1979). The dependent variable is a dummy variable that indicates the WSJ coverage as described in Table 2. Definitions for all variables are list- ed in Appendix 1. A constant term is estimated but not reported. Standard errors are reported in parentheses and are double-clustered by industry and by time. Symbols ***, **, and * indicate statistical significance at the 1%, 5%, and 10% level, respectively.
Coeff. Marginal effects Firm size 0.3067** 0.0029 (0.119) Relative deal size 0.6601*** 0.0062 (0.134) Diversifying deal 0.0574 0.0005 (0.213) All cash deal -0.0977 -0.0009 (0.408) All stock deal -0.2746 -0.0026 (0.288) #Analysts -0.0097 -0.0001 (0.009) Tobin’s Q 0.0501** 0.0005 (0.023) Firm leverage -2.1583*** -0.0203 (0.314) Firm cash -0.4773 -0.0045 (0.512) Firm profitability -3.0404*** -0.0287 (0.730) NYC headquarter -0.2556 -0.0024 (0.363) California headquarter -0.0135 -0.0001 (0.129) Past returns -17.9329 -0.1691 (18.205) CEO age 0.0146** 0.0001 (0.006) CEO founder -0.2461* -0.0023 (0.145) Board size -0.5473 -0.0052 (0.977) Classified board -0.0966 -0.0009 (0.198) Observations 2,390 R2 0.142
48 Internet Appendix for Friends at WSJ
Guosong Xu
1 1. Examples of positive versus negative news coverage:
The following excerpts illustrate two WSJ articles that carry an overall “positive” tone: — FPL Group’s acquisition of Constellation Energy announced on December 18, 2005 (news titled: FPL, Constellation Reach Agreement On $11 Billion Deal): “The deal enables Constellation to benefit from FPL’s stronger, A-grade credit rating and lower risk profile, which should help it continue to expand a power-trading business that increasingly competes against the A-rated operations of investment banks including Goldman Sachs Group Inc., Bear Stearns Cos. and Morgan Stanley. FPL, in turn, will garner more unregulated assets able to capitalize on growing profit margins in lucrative markets including California, Texas and New England.” — Brooks Automation’s acquisition of Helix Technology announced on July 11, 2005 (news titled: Brooks to Acquire Helix For $454 Million in Stock): “Brooks Automation Inc.’s plans to acquire Helix Technology Corp. for about $454 million in stock are designed to bolster business from semiconductor makers by offering them a broader array of manufacturing tools from a single supplier, Edward Grady, Brooks’ president and chief executive, told Dow Jones in an interview. “The whole premise of this is very strategic,” Mr. Grady said, adding that the combination of the two companies is “totally complementary.”” The next two excerpts, on the contrary, have an overall “negative” tone: — Gannett’s acquisition of Central Newspapers announced on June 28, 2000 (news titled: Gannett Agrees To Buy Central Newspapers): “Unlike Gannett’s other deals, which have been broadly welcomed by Wall Street, the proposed acquisition of Central poses some risks. Because of the deal’s accounting structure, it will reduce Gannett’s earnings per share by 7% in 2001, something Gannett executives like to avoid when crafting acquisitions because it upsets the company’s typically conservative investors. Not including the accounting charges, Gannett says the deal will add to its cash earnings this year.” — Pfizer’s acquisition of Wyeth announced on January 26, 2009 (news titled: Pfizer Deal to Buy Wyeth Leaves Doubts): “Pfizer Inc. hailed its planned $68 billion takeover of rival Wyeth as an ideal combination, but analysts say the deal will only partially solve some of the New York drug giant’s long-term problems. The revenues generated by Wyeth’s most attractive products, such as pediatric vaccine Prevnar, won’t be sufficient to make up for the loss of $12.4 billion in annual revenues Pfizer faces when the patent on its anticholesterol drug Lipitor expires in 2011, analysts said. And some of them expressed doubts about how the newly created behemoth, with a combined $71 billion in revenues, would discover enough new products to generate growth. “Moving that needle is going to be extraordinarily difficult,” said Timothy Anderson, a health-care analyst at Sanford Bernstein.”
2 Table A1. Journalist Connections and Positive Tone
OLS regressions of positive tone. The dependent variable is Positive tone, measured as the fraction of positive words in the text of a news article. The sample includes merger articles published by the WSJ as described in Table 2. Control variables are the same as in Table 3 columns 1–4. Definitions for all variables are in Appendix 1. All models include time (i.e., merger year-month), firm industry, state, and the WSJ office location fixed effects; columns 2 and 4 additionally control for journalist fixed effects. Standard errors, double-clustered by deal year-month and by industry, are reported in parentheses. Symbols ***, **, and * indicate statistical significance at the 1%, 5%, and 10% level, respectively.
WSJ sample: Positive tone 1 2 3 4
CONNECT WORK -0.0693 -0.0855 (0.042) (0.055) CONNECT UNIVERSITY -0.0462 0.2735** (0.135) (0.121) Controls as in Table 3 × × × × Time FE × × × × Firm industry FE × × × × Firm state FE × × × × WSJ office FE × × × × Journalist FE × × Observations 1,131 979 1,131 979 R2 0.362 0.603 0.361 0.604
3 Table A2. Journalist Connections and Negative Tone:
Articles from The New York Times
OLS regressions of negative news tone. The dependent variable is News negativity, measured as the fraction of negative words in the text of a news article. The sample includes merger articles published by the New York Times as described in Section 4.1.1. Control variables are the same as in Table 3 columns 1–4. Definitions for all variables are in Appendix 1. All models include time, firm industry, state, and the NYTimes office location fixed effects; columns 2 and 4 additionally control for journalist fixed effects. Standard errors, double-clustered by deal year-month and by industry, are reported in parentheses. Symbols ***, **, and * indicate statistical significance at the 1%, 5%, and 10% level, respectively.
NYTimes sample: News negativity 1 2 3 4
CONNECT WORK -0.0375** -0.0074 (0.018) (0.078) CONNECT UNIVERSITY -0.3425* -0.6508*** (0.200) (0.185) Controls as in Table 3 × × × × Time FE × × × × Firm industry FE × × × × Firm state FE × × × × NYTimes office FE × × × × Journalist FE × × Observations 506 506 506 506 R2 0.357 0.516 0.359 0.524
4 Table A3. Are Journalist Turnovers Related to Firm Performance?
This table examines whether journalist turnovers are related to various characteristics of firms that they cover. The sample includes the firms that have conducted more than one acquisition and that have at least one journalist connections. The dependent variable is a dummy variable that equals one if a firm’s connection leaves the WSJ, and zero otherwise. Columns 1 uses a probit regression without fixed effects. Marginal effects from the probit model are reported in column 2. Column 3 uses an OLS model with year and firm-industry fixed effects. A constant term is estimated in all models but not reported. Definitions for all variables are in Appendix 1. Standard errors are reported in parentheses and are double- clustered by year and by industry. Symbols ***, **, and * indicate statistical significance at the 1%, 5%, and 10% level, respectively.
Probit OLS 1 2 3 Coeff. Marginal effects Coeff.
Firm size 6.2086 0.1078 0.0063 (4.010) (0.008) Institution ownership -20.6873 -0.3591 -0.0553 (20.887) (0.060) #Analysts -0.2291 -0.0040 -0.0007 (0.243) (0.001) Tobin’s Q -5.2893 -0.0918 0.0009 (4.199) (0.002) Firm leverage 6.8742 0.1193 -0.0538 (19.374) (0.062) Firm cash 2.9403 0.0510 -0.0162 (10.205) (0.027) Firm profitability 52.9439 0.9189 0.0055 (39.444) (0.060) NYC headquarter 16.5038*** 0.2864 0.0336 (5.945) (0.025) California headquarter 7.9967** 0.1388 -0.0005 (3.830) (0.009) Past returns -390.1422 -6.7715 -1.5699 (269.703) (1.286) CEO age -0.0261 -0.0005 0.0005 (0.246) (0.001) CEO founder 18.6778 0.3242 0.0092 (12.016) (0.019) Board size 17.6919 0.3071 0.0109 (12.670) (0.034) Classified board — — 0.0027 (0.007) Year FE × Firm industry FE × Observations 321 321 (Pseudo) R2 0.663 0.146 5 Table A4: Additional Robustness Tests of Long-run Price Corrections
Additional tests that show the relation between CONNECT WORK and long-run return reactions. Starting from the full sample as described in Table 2, I exclude withdrawn transactions and firms with confounding news events over the window [2,4], where 0 is the news publication date. Columns 1 and 2 replicate the results of bidder’s CAR over [0,1]. The dependent variable in columns 3 and 4 is bidders’ CAR over [2,40]. The dependent variable in columns 5 and 6 is bidders’ CAR over [0,40]. The dependent variable in columns 7 and 8 is CAR from the day before the deal announcement until the deal completion date. Control variables include: relative deal size, toehold, hostile, unsolicited, diversifying deal, all cash deal, all stock deal, firm size, Tobin’s Q, leverage, cash, profitability, CEO age, CEO founder, dual, board size, classified board, and three lagged returns. Definitions for all variables are in Appendix 1. In all regressions, time (i.e., year-month) and industry dummies are included. Standard errors are reported in parentheses and are double-clustered by deal year-month and by industry. Symbols ***, **, and * indicate statistical significance at the 1%, 5%, and 10% level, respectively.
Sample that excludes confounding events over [2,40] CAR[0,1] CAR[2,40] CAR[0,40] CAR[-1,complete] 1 2 3 4 5 6 7 8
CONNECT WORK 0.0178*** 0.0168* -0.0494*** -0.0558*** -0.0252 -0.0330 -0.0178 -0.0176 (0.007) (0.010) (0.017) (0.019) (0.020) (0.022) (0.039) (0.040) Controls × × × × Time FE × × × × × × × × Firm industry FE × × × × × × × × Observations 670 670 670 670 670 670 670 670 R2 0.408 0.438 0.342 0.383 0.343 0.375 0.377 0.408
6 Table A5: Journalist Connections and Deal Outcomes
This table examines whether journalist connections are related to various M&A deal outcomes, using the sample described in Table 2. Columns 1 to 2 examine targets’ cumulative abnormal returns (CAR) over [-1,1]. Target CARs are similarly calculated as bidder CARs described in Table 4. Columns 3 and 4 examine 4-week premium, defined as offer price over target’s market capitalization 4 weeks prior to the announcement. Columns 5 and 6 examine return premium, defined as target CAR over [-42,5]. Columns 7 and 8 examine the consummation of the deal, and columns 9 and 10 examine whether the deal is rumored. Regressions 1 through 6 are estimated with OLS, and 7 through 10 are estimated with logit models. In columns 7 to 10, marginal effects of the coefficient estimates are reported in the brackets. Control variables include: relative size, toehold, hostile, unsolicited, diversifying deal, all cash deal, all stock deal, firm size, Tobin’s Q, leverage, cash, profitability, CEO age, CEO founder, dual, board size, classified board. Columns 1 and 2 control for target characteristics (Target size, Tobin’s Q, leverage, cash, profitability, target stock runup). Definitions for all variables are in Appendix 1. In all regressions, time (i.e., announcement year-month) and industry dummies are included. Standard errors are reported in parentheses and are double-clustered by deal year-month and by industry. Symbols ***, **, and * indicate statistical significance at the 1%, 5%, and 10% level, respectively.
7 Return premium: Target CAR[-1,1] 4-week premium Consummation Rumored Target CAR[-42,5] 1 2 3 4 5 6 7 8 9 10 OLS OLS OLS OLS OLS OLS Logit Logit Logit Logit
CONNECT WORK 0.0251 1.1286 0.0207 -0.0076 0.8073 (0.017) (5.386) (0.017) (0.540) (0.533) [-0.001] [0.099] CONNECT UNIVERSITY 0.0046 -13.6002 0.0189 -0.6179 0.8608 (0.063) (9.895) (0.064) (0.710) (1.697) [-0.044] [0.103] Deal & firm controls × × × × × × × × × × Time FE × × × × × × × × × × Firm industry FE × × × × × × × × × × Observations 1,115 1,115 1,129 1,129 1,115 1,115 517 517 426 426 R2 0.427 0.426 0.375 0.376 0.419 0.419 0.527 0.527 0.359 0.352