News Bias in Financial Journalists’ Social Networks

2.1 BUSINESS MEDIA AS AN INFORMATION INTERMEDIARY AND MEDIA SLANT

Price discovery does not happen instantly (Lee [2001]). Value-relevant information is often disseminated to the capital markets by information intermediaries, such as sell-side analysts and the business media. Research in accounting and finance has established two key roles of financial media. First, they facilitate the dissemination of firms’ news. For example, Kross, Ro, and Schroeder [1990] find that Wall Street Journal coverage improves analysts’ forecasts. Bushee et al. [2010] and Rogers, Skinner, and Zechman [2016] show that media coverage reduces information asymmetry. As a result, journalists’ activities have the potential to directly influence the pricing of information in the stock market (e.g., Tetlock [2007], Fang and Peress [2009], Kothari, Li, and Short [2009], Engelberg and Parsons [2011]).

The second role of media is information creation. Recently, Guest [2021] has shown that the Wall Street Journal’s editorial content provides new information that facilitates investors’ understanding of earnings news. New information also helps investors monitor firms’ managers. For example, Frankel and Li [2004], Dyck, Volchkova, and Zingales [2008], Liu and McConnell [2013], and Dai, Parwada, and Zhang [2015] find that the media help discipline managers and enhance corporate governance. Miller [2006] highlights the role of the media as a watchdog for accounting fraud. Core, Guay, and Larcker [2008] reveal that media influence CEO pay. Bushman, Williams, and Wittenberg-Moerman [2016] suggest that media stories affect syndicated loan origination.

Given the importance of the business media, a burgeoning literature investigates news spin. Perhaps most prominently, Reuter and Zitzewitz [2006] and Gurun and Butler [2012] provide evidence that firms use advertising expenditures to influence the tone of financial news coverage. Solomon [2012] shows that investor relations services help firms spin the news. Cagé et al. [2024] document the influence of media owners on media bias. These studies raise questions about the independence of media. More broadly, the journalism literature examines the demand- and supply-side determinants of media bias. For example, Ekstrom et al. [2013] show that journalistic values and interviewee characteristics both relate to interrogation bias. Ameli and Molaei [2020] and Joa and Yun [2022] find that the partisan affiliation of the media is significantly associated with news sentiment. Yamamoto and Nah [2018] suggest that community-level factors, such as social and political trust, contribute to newspaper credibility. Conway-Silva, Ervin, and Kenski [2020] reveal that cable news uses different sourcing patterns to produce biased coverage. This literature primarily focuses on political bias in media coverage. In contrast, my work examines financial media coverage and explores the potential media slant introduced by personal connections between journalists and management of the firms they cover.

2.2 MEDIA COVERAGE DURING M&A NEGOTIATIONS

M&As create large incentives for firms to manage media coverage. Ahern and Sosyura [2014] show that favorable news coverage affects acquirers’ stock prices and deal terms. Hossain and Javakhadze [2020] examine directors who work in media firms. They find that media directors influence takeover premia and announcement returns by improving acquirers’ media coverage before deal announcements.

Although these papers focus on the strategic manipulation of media coverage during M&A negotiations, which can directly affect deal terms, my study focuses on whether journalists’ activities after a deal's announcement (e.g., through analyzing transaction synergies) are related to subsequent M&A outcomes. Analysis of M&A deal synergies provides a good setting for testing media bias because M&A synergies are not directly observable and thus create incentives for news spin (Gentzkow and Shapiro [2006]). Moreover, journalists’ analyses could influence the perception of the deal postannouncement and may influence subsequent deal processes, such as competition and completion, which are important for the shareholders of the transacting parties. Finally, from an identification perspective, focusing on one homogeneous event type (i.e., M&As) provides a clean setting because the tone of news coverage is primarily determined by the underlying story. Combining various events—say, scandals and acquisitions—would make it difficult to control for underlying event characteristics and could lead to biased estimates if these characteristics were correlated with a journalist's connectedness. To assess the external validity of the tests, I extend the analysis to a sample of financial fraud news in section 4.4.

2.3 INSTITUTIONAL BACKGROUND AND HYPOTHESES DEVELOPMENT

I focus on two of the most prominent financial newspapers in this study, the Wall Street Journal and Financial Times. The Journal has a longstanding reputation as a business newspaper with a primary focus on the U.S. market, whereas the Financial Times is known to have a global focus, covering news from around the world. Both are well established among finance professionals.

In the production of news, editors are primarily responsible for selecting the stories to publish, whereas journalists write and develop assigned stories. Reporters may either be general assignment reporters or specialize in particular industries for business news reporting.4

However, there are also reasons to expect that journalists introduce bias when they have social ties with the firms they cover. First, financial journalists rely on company management for information, which creates an implicit incentive for favorable bias (Dyck and Zingales [2003]). A survey by Call et al. [2022] finds that private communications with firms constitute a major source of information for financial reporters. Intriguingly, their survey also reveals that journalists often face backlash from firms in response to unfavorable reporting.

Second, due to homophily—the tendency for individuals with similar backgrounds to associate with each other—journalists with ties to corporate executives may share similar preferences and beliefs as those executives. This could lead a journalist and a CEO to both view a corporate decision through the same lens, for instance, as value creating even though it is not. As Uzzi [1996, p. 678] notes, such a social tie “disposes one to interpret favorably another's intentions and actions.”

Finally, a journalist in a firm's network is more likely to have developed personal relationships with the firm's management due to the frequent contact between individuals in the same network (Barrios et al. [2022]). These personal relationships could lead to favoritism in which journalists depict their connections more positively. Meanwhile, due to more positive slant in the connected coverage, investors may respond more favorably to the related information (Tetlock [2007]). This happens because noise or liquidity traders face downward-sloping demand for risky assets (DeLong et al. [1990]) or because investors have limited attention (Barber and Odean [2008]). Importantly, although initial mispricing could occur, the stock price should ultimately be corrected in the long run as markets are expected to see through biased information (De Bondt and Thaler [1985]).

In the following sections, I examine the empirical question of whether articles written by connected journalists are more accurate or biased.

3.1 SAMPLE OF NEWS ARTICLES

I first collect all of the M&A bids from 1997 to 2016 involving U.S. public firms. These data come from Securities Data Company (SDC) Platinum. The sample period begins in 1997 because the main measure of the journalists’ social networks requires a manual search on the Wall Street Journal engine, which starts in 1997. Following the sample selection methods commonly used in the M&A literature (see, e.g., Masulis, Wang, and Xie [2007]), I ensure that (1) the transaction value exceeds US$10 million, (2) the target is not undergoing bankruptcy proceedings, and (3) the parties are nonfinancial firms. These criteria yield 2,390 deals.

Not all M&As are covered by newspapers. I search the media coverage using the Wall Street Journal and Financial Times Web sites and Factiva, a media research tool owned by Dow Jones. To ensure that I am not selecting stories about M&A rumors, I require the news article to be the first report on the transaction following the official deal announcement. Additionally, I focus only on articles that provide author information. This leads me to drop newswire articles.5 Following these selection criteria, I obtain 1,370 articles from the Wall Street Journal and 666 articles from Financial Times, corresponding to 57% and 28% coverage rates, respectively. Among these articles, 651 unique deals are covered by both newspapers, representing an overlap of 1,302 articles (i.e., 651 articles from each outlet). I refer to this panel data set of 1,302 articles as the “pooled sample,” upon which I conduct media slant analysis.

In some of my cross-sectional analyses, such as those of deal announcement returns and deal processes, I require additional information on stock performance and firm characteristics. I obtain firms’ financial data from Compustat and stock data from CRSP. After matching the covered firms with Compustat and CRSP, I am left with 1,131 deals covered by the Wall Street Journal and 650 deals covered by the Financial Times. Figure 1 shows the number of M&A bids across years. The number of bids is lower after the dot-com bubble crash in 2001 and again during the financial crisis in 2008. Overall the Wall Street Journal provides more solid coverage of transactions than the Financial Times: The covered deals in the Journal (Financial Times) collectively account for over US$5.9 trillion (5.1 trillion) in transaction value and have a median size of US$1.5 billion (2.7 billion). This compares to the median SDC deal size of US$0.4 billion. The relatively large transaction value is perhaps unsurprising, as big transactions attract greater media attention. Most of these M&A stories (over 95%) appear within two days following the official press release by the company.

3.2 JOURNALIST-FIRM NETWORKS

Measuring journalists’ social networks is challenging, as individual networks are not directly observable. Therefore, I adopt conventions used in the accounting and finance literatures to construct two proxies: working relationships and educational ties. In addition, I use social media pairs, drawing from the journalism literature, as another proxy for journalist-firm ties. Admittedly, this practice omits other potential social networks, such as second-degree ties. To the extent that these omitted networks matter for media slant, my focus on the work, educational, and social media ties conservatively estimate network effects.

I identify working relationships following the logic of Solomon [2012]. Specifically, I check whether a reporter has written multiple exclusive stories about the same company within the 12 months prior to the merger.6 For example, Betsy McKay, the author who covered PepsiCo's acquisition of Quaker Oats in December 2000, wrote more than 10 stories about PepsiCo in 2000, featuring several interviews with the firm's top executives. It is likely that Betsy had established a working relationship with the firm. My measure of the working relationship, CONNECT_WORK, takes a value of one if any author of the M&A story covered the bidding firm at least twice in the previous year.7 Admittedly, working relationships may be a product of specific journalist skills. Regardless of their cause, I argue that they could influence slant if a social tie is formed. I discuss the related endogeneity issue in the following section.

To construct the educational networks, I search for the universities that the journalists and the CEOs attended. I use a variety of sources. The newspaper Web sites provide personnel biographies of the authors. I supplement this information with data gathered from LinkedIn, a professional networking service. I gather the CEOs’ universities from the companies’ proxy filings on the Securities and Exchange Commission's (SEC's) EDGAR database and Bloomberg. Finally, if I cannot determine a reporter's or CEO's university with the primary sources, I perform general Web searches to collect additional information. Because the databases do not provide records of graduation years, I cannot systematically observe the school ties within a cohort. Nevertheless, this measure of educational ties is consistent with prior work showing that general alumni networks have significant impacts on financial markets (e.g., Cohen, Frazzini, and Malloy [2008], Ru et al. [2020], Chen et al. [2021]). The measure of educational ties, CONNECT_UNIVERSITY, equals one if any author of the article attended the same school as that attended by the CEO.

Finally, I create a variable, CONNECT_TWITTER, to capture the existence of a CEO-journalist pair on social media (specifically Twitter). The indicator variable equals one if a CEO and a journalist follow each other on Twitter and zero otherwise. Research on journalism suggests that interactions on social media, such as Twitter, can indicate both formal and informal social connections (Barnidge et al. [2020]). These social interactions in turn may influence the (perceived) media bias (de Zuniga, Diehl, and Ardevol-Abreu [2018]).

Table 1 reports the summary statistics of journalist-firm networks. Approximately 27% of articles across the Wall Street Journal and Financial Times are authored by journalists with a working relationship with the firm. This aligns with Li's [2015] remark that business reporters seldom cover the same firm more than once a year. As a comparison, I create a variable, Industry expert, that identifies whether the authors specialize in the relevant industry. Table 1 shows that about 46% of articles are written by reporters who are industry experts. This comparison suggests that repeated coverage goes beyond general expertise and indicates a relationship with the firm. However, social ties via schooling and on Twitter are infrequent. Only 2.3% of journalists have attended the same university as the firm's CEO, and less than 1% of journalist-CEO pairs follow each other on Twitter. Note that this latter statistic is influenced by the fact that many CEOs do not have Twitter accounts.

Although a social tie from the same alma mater is arguably exogenous to the concurrent economic fundamentals of a corporate event, social ties based on working relationships or social media are not. To account for journalists’ experience and personal traits, I construct several variables, namely, Local journalist, Tenure, Industry expert, and Female, to control for the general influence of journalist characteristics. These variables are described in online appendix 1. Furthermore, online appendix table A1 reports the correlations between journalist-firm ties and journalist characteristics. We see that a working relationship is positively correlated with a journalist being local or an industry expert. However, the variable CONNECT_UNIVERSITY is uncorrelated with other journalist traits or event characteristics, suggesting that a schooling tie formed years ago is likely exogenous to a firm's current performance.

3.3 MEDIA SLANT

The average negativity expressed in an M&A article equals 1.3%−1.4% (see table 1). This compares favorably to the 1.7% negative slant reported by Gurun and Butler [2012]. To provide a sense of the slant, online appendix 2 offers several examples of articles (excerpts). I obtain similar results using a measure of net slant, namely, the fraction of negative words minus positive words, in the robustness tests.

3.4 DEAL AND FIRM CHARACTERISTICS

I construct several firm and deal characteristics as control variables. These characteristics are motivated by the M&A literature and have been widely used to capture merger qualities (e.g., Cai and Sevilir [2012], Kim, Verdi, and Yost [2020]). For example, I collect the deal's relative size, hostile attitudes, nonsolicitations, cross-industry bids, payment methods, and toeholds from SDC. From Compustat, I gather information about firm size, Tobin's Q, leverage, cash holdings, and profitability. Institutional ownership data come from Thomson/Refinitiv. Analyst coverage comes from the Institutional Brokers’ Estimate System. Additionally, I obtain CEO- and board-related variables—including CEO age, duality, and classified board—from the firms’ proxy filings in EDGAR.9 Online appendix 1 reports the definitions of all variables. Table 1 provides the summary statistics.

3.5 ENDOGENEITY OF MEDIA COVERAGE

I start by discussing the concern that a sample selection bias exists. Specifically, it would be worrisome if news coverage were correlated with reporters’ social networks. To investigate this possibility, I use a probit estimator to predict media coverage using the full list of M&A bids.10 The analysis reveals that journalists’ social networks do not have a significant impact on news coverage. Instead coverage is primarily driven by firm and deal size as well as by sensational aspects, such as hostile and unsolicited transactions. This evidence is consistent with the argument that individual reporters do not control which stories are ultimately published.

4.1 EMPIRICAL STRATEGY AND MAIN RESULTS

The main variable of interest, CONNECT, corresponds to one of the three measures of journalist-firm ties described in section 3. To test whether ties relate to slant, I estimate whether β differs statistically from zero. Equation (2) includes deal fixed effects (D_i), which allow me to benchmark slant across media while holding the information content of the event constant. Because deal fixed effects subsume all deal-level estimates, the regression does not control for any (time-varying or invariant) deal and firm characteristics.11 As a result, the only controls I include are journalist characteristics, Z_ij. Here, Z includes a dummy variable indicating whether the journalist is in the same city as the firm's headquarters, the journalist's gender, tenure, and industry expertise. I use media fixed effects (M_m) to control for the general writing style of the media outlet. Alternatively, I include journalist fixed effects (J_j), which control for a journalist's general slant. Standard errors are clustered by event (i.e., a deal) to account for the possibility that media slant or news sentiment may be correlated within an M&A event.12 In this pooled sample, each M&A event is covered by two articles. The mean percentage of articles written by journalists with a working relationship, university tie, or social media connection is 27%, 2%, and 0.8%, respectively.

Panel A of table 2 reports the results. Across the columns, the coefficients on CONNECT are all negative and statistically significant. The negative point estimates suggest less negative coverage in connected publications. Based on the estimates with deal and media outlet fixed effects (see the odd-numbered columns), an article authored by a journalist with a working relationship contains 36.7% fewer negative words, relative to an independent article on the same event. In the case of a school or social media connection, the use of negative words is 19.8% and 58.9% lower, respectively.13 Because school ties and social media connections are rare, I refrain from drawing strong inferences from these estimates. However, these results support the evidence derived from the working relationship, suggesting that connections are associated with less negative coverage.

Since media sentiment is influenced by the interplay between companies and journalists, I also estimate a model that includes more saturated journalist fixed effects (see the even-numbered columns). This model essentially compares journalists’ tone in excess of their usual tone. It addresses the concern that journalists with more corporate ties might be generally more optimistic. The results remain consistent with previous findings.

Call et al. [2022] indicate that financial journalists often rely on company disclosures when crafting their stories. Therefore, as another way of comparing slant, I benchmark the language in the newspapers to that of the press release issued by the bidding companies. These results are reported in table A3 of the online appendix. The results continue to show a network bias that favors the firm. Here, the economic effects are that corporate ties through a working relationship or a school or social media tie are associated with 22.1%, 23.9%, and 47.4% less negativity relative to firm-initiated press releases, respectively.

To get a sense of magnitudes, I compare these results to the well-documented advertising bias: Gurun and Butler [2012] show that spending US$200,000 per month on advertising in a national newspaper is associated with a 36% increase in slant measured in the same way as my study. Thus, the impacts of social networks appear to be similar in magnitude, if not larger, than the impact of advertising bias at the US$200,000-per-month level.

With respect to the control variables, there is little evidence that expertise, as captured by Tenure and Industry expert, is systematically associated with slant. This is perhaps unsurprising because the estimates are conditional on the information content of the event.

4.2 INSTRUMENTING JOURNALIST TIES

Despite the use of various fixed effects in table 2, panel A, endogeneity concerns remain. The primary concern is the possibility of reverse causality. For example, executives might have stronger incentives to connect with journalists if their company's information asymmetry is high. Alternatively, if the firm's information environment is poor, journalists may feel that they can make more of an impact by developing relationships with the company's executives and by providing favorable (i.e., less negative) press coverage.

To address this concern, I propose an instrumental variable that is correlated with the probability of connected coverage but has no independent effect on slant. I contend that journalists’ turnover timing (i.e., the specific time at which they leave the media outlet) satisfies both the relevance condition and the exclusion restriction. The rationale is as follows: The departure of a friendly reporter today will decrease the likelihood of the firm being covered by a connected journalist in the future. However, the exact timing of turnover (say, one month before a merger announcement) is unlikely to be determined by the covered firms and is thus exogenous (Solomon [2012]).

Two issues may confound the plausibility of using turnover timing as an instrumental variable. First, one may worry that, even in the presence of turnover, editors might assign other connected reporters as a replacement (thus violating the relevance condition). To alleviate this concern, I focus solely on cases where a firm is connected to one reporter prior to the M&A.14

Another concern is that journalist turnovers could be driven by their past slant and thus would not be exogenous. To mitigate this concern, I manually search to identify the reasons behind each turnover and only include cases where a reporter leaves to take on a more senior role, such as editor, at another media outlet. If a journalist is promoted to a higher position, this is not a turnover triggered by abnormal slant in the journalist's reporting (Call et al. [2022]). Conversely, if a journalist leaves due to inaccurate reporting (slant), it is more likely that that person has been demoted or has quit journalism. In my analysis, I focus on the timing of promotions that occur in the six months leading up to an M&A announcement. If a firm's sole connected journalist turns over during this period, I set the dummy variable Turnover (promotion) equal to one. Although testing for the exclusion restriction is inherently infeasible, I perform untabulated tests and find that (1) firm performance is uncorrelated with the connected journalists’ turnover and (2) the average slant in the journalists’ past publications is not associated with Turnover (promotion).

To construct the sample of turnovers, I consider the following criteria: First, given that a firm's connectedness is endogenous, it would not be logical to compare firms with zero journalist ties to those with ties. Therefore, I focus on firms that have at least one journalist tie so that the firms are more comparable. Second, because promotion timing generates exogenous variation of journalist ties in a time series, the exercise is possible only for firms that have conducted more than one transaction (see online appendix 3 for a description). These restrictions result in a reduction of observations. Therefore, I perform this analysis only with the connectedness measure of working relationships. Using the new sampling criteria, I obtain 320 articles from both newspapers covering the same event, with 248 authored by connected journalists. In total, I observe 10 turnover events preceding the M&A announcement.15

Panel B of table 2 reports the 2SLS results. Similar to the preceding OLS analysis, the regression relies on a pooled data set with deal fixed effects. The first-stage results, presented in columns 1 and 2, indicate that a recent turnover of a connected journalist significantly reduces the likelihood of a future report being written by a connected reporter in the firms’ network. The F-statistic on the instrument is above the critical values proposed by Stock-Yogo, suggesting that the estimation is efficient.

Columns 3 and 4 report the estimate of the second-stage regression. The network effect is highly statistically significant, pointing in the same direction of slant as that documented in panel A of table 2. The Sargan χ² test cannot reject the joint null that the instrument is valid. The magnitude of the slant is close to the corresponding OLS estimate reported in panel A. This suggests that the estimates are likely to be accurate.16

Jiang [2017] notes that, in an instrumental variable test, the average treatment effect is local, which inhibits generalizing the findings to a larger population. I acknowledge that the local effect in this smaller sample is the cost of achieving greater internal validity. To the extent that my purpose is to detect a network bias, rather than quantify the slant, this exercise still offers valuable insight into the plausibility of the underlying hypothesis. However, when interpreting the results, one should be cautious about underlying assumptions, such as the exclusion restriction.

4.3 ADDITIONAL EVIDENCE: QUOTES OF CEOs, ACCURACY, AND SPILLOVER EFFECTS

In this section, I provide additional evidence supporting the effect of social ties between journalists and firms. These tests examine (1) verbatim quotes of connected CEOs, (2) alternative textual features in news stories, and (3) spillover effects on a CEO's social network of firms.

I first assess the extent to which the mergers are ascribed to specific individuals, such as the acquiring CEO. The rationale for this test is that, in addition to the generally favorable coverage of an M&A deal, a reporter with social ties may be particularly friendly to a senior executive of the firm. In the first three columns of panel A of table 3, I study this conjecture. The dependent variable is the number of quotes of the CEO in a news article. The regression specification is otherwise similar to that in equation (2).

The results show that connected journalists are indeed more likely to directly quote the acquiring CEO. For example, an article authored by a journalist with a working relationship with the firm contains 15% more CEO's verbatim quotes than other articles covering the same deal. Likewise, a journalist who attended the same university as the CEO cites 38% more of the CEO's quotes. These findings reveal a novel strategy in which connected journalists attribute deal qualities to specific corporate executives.

I next examine the textual characteristics of news reports that pertain to accuracy. Note that a less negative tone does not necessarily imply that the news content is inaccurate. Under the information transmission hypothesis, connected journalists with access to private information should provide more accurate statements in their analyses. However, under the favoritism hypothesis, journalists within a firm's social network may attempt to obfuscate the true nature of the transaction by using a less negative but more ambiguous or equivocal tone. To achieve this, they may avoid using precise numerical analyses and instead rely on ambiguous language, such as uncertainty words and weak modal words (Kim [2018]). In columns 4−12 of panel A of table 3, I use three different dependent variables that capture the accuracy (or ambiguity) of a news article, namely, the percentage of numerical predictions, the percentage of uncertainty words, and the percentage of weak modal words.17 These regressions are specified based on equation (2).

The results suggest that journalists with a working relationship or a Twitter tie with the CEO use fewer numerical analyses in their reports than independent journalists. They also use significantly more ambiguous language in their writing. Specifically, CONNECT_WORK (TWITTER) is associated with a 1% (3%) decrease in numerical analyses, an 11% (37%) increase in uncertainty words, and a 10% (25%) increase in weak modal words. However, the effect of university ties is statistically insignificant. These findings provide suggestive evidence that journalist ties are related to a less accurate portrayal of underlying news events.

Shani and Westphal [2016] suggest that journalists who engage in negative coverage of a firm tend to distance themselves from other executives in that firm's social network. As a result, these journalists are more likely to write negatively about these other connected firms. Their finding suggests a potential spillover of journalist coverage, in which a journalist's ties to a firm positively influence reporting about other firms in the social network of the connected firm. To probe this possibility, I expand my analysis and examine the news coverage of firms in the network of the acquirer to which a journalist is connected. Specifically, I focus on the network of directors of the acquirer. From BoardEx, I gather data on all outside directors who sit on the board of the acquirer at the time of the transaction. I then collect names of the (public) firms where these directors are currently working. Next, I use Factiva to collect all news articles published in the Wall Street Journal and Financial Times about the directors’ firms (i.e., second-degree connected firms) within one year of the M&A announcement. Because the number of M&As conducted by the directors’ firms is small and overlaps with the M&A news sample, I include other business news in this analysis. These other events include (1) earnings news, (2) corporate governance events (e.g., compensation, executive turnovers, accounting fraud, shareholder voting), (3) product market events, and (4) other major financial news such as debt restructuring or bankruptcies.18

Panel B of table 3 reports the analysis of spillovers. I regress Negative slant (of the articles concerning the directors’ firms) on a dummy variable, CONNECT_Indirect, which flags journalists who have indirect ties to the director's company through their network with the acquirer firm. These tests include media outlet fixed effects and firm × year-month fixed effects. The latter fixed effects ensure that I make comparisons of slant within the same director firm's events in the same year-month. The coefficient on CONNECT_Indirect thus identifies the spillover effect of a second-degree social tie. Column 1 in Panel B shows that the estimate of CONNECT_Indirect is negative and significant at the 5% level. Relative to other articles about the same firm's events in the same year-month, indirectly connected journalists use 39% fewer negative words when describing the directors’ firms in the acquirer's network.

In the second column of panel B, I compare the direct (first-degree) network effect to the spillover. This regression is similar to that in column 1 of panel B but additionally includes a dummy variable, CONNECT_Direct, which identifies journalists who have direct social ties to the director's company. CONNECT_Direct equals one if the article is written by journalists with a direct working relationship, university tie, or Twitter connection to the company management, and zero otherwise. The result confirms the existence of both direct and spillover effects. The direct effect exceeds the spillover effect, with a 43% decrease in negative slant associated with a direct social tie compared to a 33% decrease with a second-degree tie. Although a smaller effect, the spillover is still economically meaningful and statistically significant.

4.4 EXTERNAL VALIDITY: EVIDENCE FROM ACCOUNTING FRAUDS

To evaluate the external validity and generalizability of my findings, I analyze a different type of financial news: accounting fraud. Firms charged with fraud often face intense public scrutiny. The sentiment of news coverage in these cases can either exacerbate or alleviate the public's response to these crises. Therefore, I examine differences in media coverage of accounting fraud by independent versus connected journalists.

I follow prior studies to focus on the SEC's investigations of alleged securities law violations (e.g., Files [2012]). To identify fraudulent firms, I search the SEC's Accounting and Auditing Enforcement Releases between 1997 and 2016. For media coverage, I use Factiva to collect publications from the Wall Street Journal and the Financial Times. Like in the M&A news analysis, I use a pooled event panel where the same fraud firm is covered by both newspapers. This leads to a final sample of 229 and 149 articles in the Wall Street Journal and Financial Times, respectively.19 Unsurprisingly, these articles are much more negative than the M&A stories: The mean negative slant is 5%. As in the main analysis, I use equation (2) to test for the network effect on media slant.

Table 3, panel C, reports the results. As in the M&A news, journalist ties, whether through working relationships or university networks, are associated with significantly fewer negative words in reporting accounting fraud.20 These findings corroborate the evidence of network effects observed from the M&A news reporting.21

5.1 JOURNALIST NETWORKS AND STOCK MARKET RETURNS

5.1.3 Long-Run Stock Returns

To further distinguish between the hypotheses of information transmission (H1) and favoritism (H2), I study acquirers’ long-term returns. Intuitively, if the higher short-run return to connected news is driven by bias rather than information, a stock price correction is expected in the long run (Tetlock [2007]). This intuition is supported by figure 2, which depicts the univariate comparison of the daily CAR around media publications. It offers several insights. First, there is no significant difference in returns between connected and unconnected firms on each single day before publication day t. Second, consistent with the findings presented in table 4, returns to connected articles are significantly higher immediately upon the news release. Finally, and most importantly, the higher abnormal returns to connected firms gradually disappear in the long term.

To formalize this pattern, I use a multivariate regression framework in table 5, panel B. To facilitate comparison between tables, I first replicate the short-term returns regressions in column 1. The results in column 2 suggest that firms with journalist ties are associated with significantly more negative returns over the [2,40] window, supporting the long-run price correction conjecture (H2).25 The postannouncement price reversal completely offsets the initially more favorable responses to connected firms. Indeed, when the short- and long-run returns are combined, column 3 reveals that the difference in returns between connected and independent firms is indistinguishable from zero. Moreover, column 4 examines acquirer CAR from one day before the deal announcement until the deal completion date. Studying acquirer returns over this longer event window allows me to evaluate differences in stock price revaluation between connected and unconnected firms (Malmendier, Opp, and Saidi [2016]). The result again suggests similar overall returns between connected and unconnected firms.

Taken together, the evidence rejects the hypothesis that favorable coverage in the connected news stories represents value-relevant information. Although studies suggest that media disseminate and create information (e.g., Bushee et al. [2010]), my findings offer a caveat: Whereas independent media facilitate price discovery, biased media coverage due to a connection effect may impede this process.

5.2 JOURNALIST NETWORKS AND ATTRIBUTES OF THE BIDDING PROCESS

In this section, I explore the relation between journalist ties and M&A bidding. Recent studies suggest that individuals often rely on heuristics in decision making, even during competitive auctions (Lacetera, Pope, and Sydnor [2012]). Additionally, models of inattention postulate that salience can distort how individuals react to information (DellaVigna [2009]). For example, Dutordoir, Strong and Sun [2022] show that people tend to home in on M&A announcement returns to infer investment synergies while disregarding long-term price correction. Similarly, Fich and Xu [2023] find that such a bias distorts capital allocation.

To investigate whether connected media coverage relates to attributes of the bidding process, I first examine bid competition in the wake of a Wall Street Journal publication.26 If M&A competitors react to a connected positive article, there may be a positive correlation between connected deals and bid competition. In column 1 of table 6, I use a probit model to predict public bid competition. The dependent variable is an indicator for competing bids received by the target firm over [0,40] following the news publication. The regression specification is otherwise similar to that in equation (3). The result suggests that connected articles are correlated with a significant increase in future bids. In column 2, I instrument connected coverage with journalist Turnover. In this subsample of the analysis, I obtain comparable results. Here, journalist ties are associated with an 18% increase in the probability of competing bids.

The increased competition in M&A deals covered by connected journalists may relate to subsequent positive bid price revisions or deal withdrawals from the initial bidders. Columns 3 through 6 of table 6 test these hypotheses. In columns 3 and 4, I examine whether the initial bidders in deals covered by connected reporters are more likely to raise their bids. The results confirm this positive relation. The economic effect is large, with the results in column 4 revealing that the likelihood of a bid revision more than doubles from 11% (at sample means) to 28%.

In columns 5 and 6, I examine deal consummation. With the probit model in column 5, I do not find statistically significant evidence that a connected bid is associated with deal withdrawal. However, when I instrument journalist ties with Turnover, I find that connected bids are significantly correlated with a lower likelihood of completion in this subsample (see column 6).

A distinction between my findings and prior ones is that mine imply a connection effect through noninformation channels. For example, Hossain and Javakhadze [2020] note that companies’ media directors are positively associated with deal initiations and negatively correlated with takeover premiums. Unlike journalists, media directors are directly involved in M&A decision making. Therefore, their influence on deal outcomes likely stems from their ability to affect decision making via information channels.

Finally, my findings of higher bidding contests and offer prices do not necessarily suggest a causal effect of media slant. For example, favorable coverage by connected journalists may prompt target firms to initiate more (unobservable) takeover bids, which may explain the large effects on bid competition and price revision following media article publications. In addition, my findings do not imply that managers are being harmed by their journalist ties. In fact, studies have shown that managers benefit from favorable short-run market responses, which are associated with their media networks. For example, Lehn and Zhao [2006] find that bidding CEOs are less likely to be fired when the initial deal announcement returns are higher. Similarly, Gong, Louis, and Sun [2008] show that better short-term returns help reduce litigation risks for bidding managers.

5.3 JOURNALIST NETWORKS AND FUTURE CAREER TRAJECTORY

To better understand the potential impact of company ties on financial journalists’ careers, in this section, I examine journalist turnover. As highlighted by Call et al. [2022], accuracy and credibility are crucial factors that affect the performance of financial reporters. If media slant is perceived as harming a reporter's credibility, connected journalists may be more likely to exit journalism in the long term. Meanwhile, social networks may provide these journalists with more outside opportunities, particularly within their connected industries.

To examine journalists’ careers, I create five outcome variables using data sourced from Factiva and LinkedIn. Specifically, I collect information on (1) cases of journalists changing their industry focus,27 (2) instances of departure from the journalist's media outlet, (3) cases of quitting journalism entirely, (4) instances of leaving for a nonjournalism industry position, and (5) instances of departure for a role in the journalist's previously connected industry.28 These career outcomes are measured at least one year after the journalist's latest publication date in my sample. The results remain similar when I restrict the outcome variable to the three-year period following the latest publication. For example, it is common to observe a journalist who is connected to a tech company leave a newspaper to work in public relations within the tech industry.

Following Kempf [2020], I study journalists’ long-term career paths by regressing the aforementioned outcome variables on CONNECT, an indicator variable for journalists who have social ties (of any type) with a company. In these regressions, I control for journalist characteristics, such as gender, tenure, number of publications (in logarithm), and journalism degree. I also include media outlet fixed effects and journalist location (city) fixed effects.

Table 7 presents the results. The findings confirm that social ties relate to financial journalists’ long-term career paths. The first column shows that connected journalists are 26.9% more likely to change their industry coverage compared to independent reporters. The effect is significant at the 1% level. In the second column, I examine whether connected journalists are more likely to leave their employer. The results reveal that these journalists are 18.5% more likely to leave than independent journalists. Moreover, column 3 shows that these journalists are 13.3% more likely to quit journalism altogether.

Next, I present the results for departures to an industry job. The findings suggest that connected journalists are 10.7% more likely to transition to a nonjournalism job compared to independent reporters (column 4). Furthermore, column 5 reveals that journalists with social networks in a particular industry are 16.8% more likely to enter that industry. These results are economically significant and consistent with the findings of Kempf [2020], who shows that connected credit analysts are 33% more likely to be hired by an investment bank. In online appendix table A6, I show that these results remain robust when I measure the career outcomes within a three-year window after the journalists’ latest publication.

Taken together, the evidence implies different career paths between connected and independent financial reporters. Keep in mind, however, that these results do not suggest connections cause journalist turnover, nor do they distinguish between the credibility and external opportunity channels. A more precise welfare analysis that compares the costs and benefits of journalist connections remains an interesting avenue for future research.