Author’s Accepted Manuscript Are all analysts created equal? Industry expertise and monitoring effectiveness of financial analysts Daniel Bradley, Sinan Gokkaya, Xi Liu, Fei Xie www.elsevier.com/locate/jae PII: DOI: Reference: S0165-4101(17)30012-5 http://dx.doi.org/10.1016/j.jacceco.2017.01.003 JAE1134 To appear in: Journal of Accounting and Economics Received date: 10 November 2015 Revised date: 18 January 2017 Accepted date: 20 January 2017 Cite this article as: Daniel Bradley, Sinan Gokkaya, Xi Liu and Fei Xie, Are all analysts created equal? Industry expertise and monitoring effectiveness of financial analysts, Journal of Accounting and Economics, http://dx.doi.org/10.1016/j.jacceco.2017.01.003 This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting galley proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain. Are all analysts created equal? Industry expertise and monitoring effectiveness of financial analysts Daniel Bradley,a Sinan Gokkaya,b Xi Liu,c Fei Xied a University of South Florida, danbradley@usf.edu, 813.974.6358 b Ohio University, gokkaya@ohio.edu, 740.593.0514 c Miami University and Ohio University, liux4@ohio.edu, 740.593.2040 dUniversity of Delaware, xief@udel.edu, 302.831.3811 January 2017 ______________________________________________________________________________ Abstract We examine whether analysts’ prior industry experience influences their ability to serve as effective external firm monitors. Our analyses of firms’ financial disclosure quality, executive compensation and CEO turnover decisions portray a consistent picture that related pre-analyst industry experience is of critical importance for analysts to play an effective monitoring role. Coverage by analysts with such experience is associated with reduced earnings management, lower probability of committing financial misrepresentation, less CEO excess compensation, and higher performance sensitivity of CEO turnover. We also provide evidence on several plausible mechanisms through which industry expert analysts exert monitoring efforts and limit managerial opportunism. ______________________________________________________________________________ We thank Lauren Cohen, Russell Jame, Sebastien Michenaud, Thomas Shohfi, Rong Wang, Jian Xue, Xiaoyun Yu, Min Zhu, and seminar participants at the 2014 Conference on Financial Economics and Accounting (CFEA), the 2015 Financial Management Association, the 2015 China International Conference in Finance and Tsinghua University for helpful comments and suggestions. We are responsible for any errors. Are all analysts created equal? Industry expertise and monitoring effectiveness of financial analysts 1 “… management has not historically been the most committed to capital returns, but is seemingly willing to adjust senior ranks going forward. The current CEO's contract is down to under 10 months.” — Jamie Baker, JP Morgan analyst, on JetBlue1 1. Introduction The notion of analyst monitoring goes at least as far back as Jensen and Meckling (1976, p. 354-355), who suggest that security analysts possess comparative advantages in monitoring firm management and thus can play a large role in reducing agency costs. As the quote above illustrates, analysts can be vocal critics of management policies, which can ultimately influence firm behavior. A few months after Jamie Baker’s call, Jetblue announced that its CEO would be resigning at the end of his contract.2 In this paper, we focus on analysts’ industry expertise and examine its effect on their monitoring effectiveness. Industry expertise is a critical component of analysts’ human capital. Analysts are assigned to and specialize in few industries to take advantage of economies of scale in information production (Boni and Womack (2006), Kadan, Madureira, Wang, and Zach (2012)). Surveys of financial analysts and institutional investors consistently indicate that industry expertise is the most important attribute for analysts (Brown, Call, Clement, and Sharpe (2015) and the annual polls of Institutional Investor magazine). The limited academic attention to analysts’ industry experience has so far been confined to their earnings forecasts and stock recommendations (Boni and Womack (2006), Kadan, Madureira, Wang, and Zach (2012), Bradley, Gokkaya, and Liu (2015)). Whether industry expertise affects analysts’ effectiveness in their other functions remains unknown. We measure analysts’ industry expertise based on whether they have prior work experience in industries of the firms they cover. To do so, we manually collect analysts’ employment history from LinkedIn.com, the largest professional networking service. 3 As elaborated below, despite its positive effect on traditional metrics of analyst performance, the ramifications of industry expertise for the monitoring and corporate governance role of analysts are more complex and difficult to determine ex ante. On the one hand, it is possible that analysts with industry experience in the firms 1 Media outlets attributed the pressure to oust Jetblue’s then-CEO Dave Barger to Jamie Baker’s upgrading of the firm while noting that management turnover was likely. Source: https://www.thestreet.com/story/12885110/1/why-jetblueceo-dave-barger-was-chased-out-by-wall-street.html. 2 http://fortune.com/2014/09/18/jetblue-ceo-david-barger-to-take-off/. For more examples of analyst monitoring activities, see http://www.wsj.com/articles/noble-promises-more-transparency-1429664102 (questioning accounting methods and demanding more transparency) and http://www.broadcastingcable.com/news/news-articles/bernsteinresearch-criticizes-media-ceo-pay/110241 (criticizing CEO pay). 3 For instance, Jamie Baker, the analyst mentioned in the opening quote of this paper is an example of an analyst with related pre-analyst industry experience. According to his LinkedIn profile, he worked for two airline companies prior to becoming a sell-side analyst. See section 2 for complete details on our data collection procedure. 2 they cover can provide more effective external monitoring because their prior industry experience may allow them to develop a better understanding of the firm’s industry. Richer and more in-depth industry knowledge can enhance analysts’ ability to analyze firms’ financial information and evaluate the strategies and decisions proposed or implemented by firm management. Therefore, analysts with related industry expertise, i.e., industry expert analysts, are better equipped and thus more likely to identify and bring attention to firm policies that do not serve shareholders’ best interests. In addition, these analysts also contribute to a more transparent information environment through their more efficient information production and more accurate earnings forecasts (Bradley, Gokkaya, and Liu (2015)). As a result, industry expert analyst coverage can have the dual effects of reducing managers’ latitude and incentives to engage in self-serving behavior as well as providing an impetus for boards of directors to demand higher accountability of managers for their actions. We term this view the effective monitor hypothesis. On the other hand, prior work experience in a firm’s industry may reduce analysts’ incentive to monitor firm management in subtle, but potentially important ways. For example, having worked in the firm’s industry increases the likelihood that the analysts and the firm’s managers know each other if their career paths have crossed or they have met at work functions such as industry trade shows or conventions. Likewise, related industry work experience also increases the chance of professional connections being developed between the analysts and firm management through common friends or acquaintances from within the industry. These professional ties can potentially cloud analysts’ views, making them more likely to agree with rather than disapprove of the decisions made by managers. Together, these different channels imply that related industry experience can impair the incentives of analysts to monitor firm management, allowing corporate insiders to indulge more in activities that benefit themselves at shareholders’ expense. We term this view the impaired monitor hypothesis. We test these hypotheses by examining the effects of industry expert analyst coverage on several major corporate policies including financial disclosure quality, CEO compensation, and CEO turnover. A large part of an analyst’s job entails the perusal and analysis of financial information disclosed by firms to capital markets and the use of such disclosure as a basis to evaluate managerial decision making and forecast future performance. Therefore, we start our analysis by relating analyst industry expertise to a number of observable outcomes of the choices made by managers in firms’ financial disclosure. This is also in keeping with earlier studies by Yu (2008) and Irani and Oesch (2013) that examine the effect of analyst monitoring on corporate financial reporting quality. 3 We first examine the impact of analyst industry expertise on earnings management through discretionary accruals, which are estimated using the method suggested by Owens, Wu, and Zimmerman (2015). Earnings management is considered as a manifestation of the agency problems between managers and shareholders, because managers are able to extract various forms of private benefits and personal gains by manipulating reported financial results (e.g., Perry and Williams (1994) and Bergstresser and Philippon (2006)). Shareholders, on the other hand, bear significant costs when aggressive earnings management leads to financial misreporting that results in earnings restatements, shareholder lawsuits, and regulatory/legal sanctions against the firm (e.g., Dechow, Sloan, and Sweeney (1996) and Karpoff, Lee, and Martin (2008)). Analysts have incentives to be vigilant about aggressive earnings management. Dyck, Morse, and Zingales (2010) find that failure to detect accounting fraud at covered firms increases an analyst’s probability of being demoted. As Dichev, Graham, Harvey, and Rajgopal (2013) point out, it is difficult for outsiders to detect earnings manipulation, but large deviations from industry and peer norms can serve as a red flag for potential financial misreporting because a substantial portion of a firm’s financial reporting choices is driven by operations and economic conditions specific to its industry. Therefore, relevant industry expertise and knowledge are essential for the evaluation of many aspects of corporate financial reporting. Consistent with the effective monitoring hypothesis, we find that coverage by analysts with related industry experience is significantly and negatively associated with firms’ earnings management. In contrast, coverage by other analysts is not related to earnings management behavior. These results are robust to controlling for a wide range of analyst and firm-specific attributes. Next, we use the F-score developed by Dechow, Ge, Larson, and Sloan (2011) to capture the ex-ante probability of firms engaging in financial misreporting and earnings restatements as an ex post measure of financial misreporting. We find that both measures are significantly lower when firms are followed by more analysts with related industry expertise. As is the case with our earnings management results, coverage by other analysts is not related to the probability of financial misreporting. In further analysis, we examine firms’ CEO compensation and CEO turnover decisions. We find that coverage by industry expert analysts is associated with significantly lower levels of CEO excess compensation and higher responsiveness of firms in replacing poorly performing CEOs. Closely echoing the results from our financial disclosure analysis, we find that coverage by other analysts is not related to either CEO compensation or CEO turnover decisions. These results provide additional support for the effective monitor hypothesis. 4 A potentially serious econometric concern with respect to our analyses is identification. Because analyst coverage in general, and coverage by industry expert analysts in particular are not random, baseline OLS regressions may suffer from endogeneity biases. For example, industry expert analysts could utilize their superior industry knowledge to identify and provide coverage for firms with fewer agency problems. To mitigate endogeneity concerns, whenever possible we conduct additional analysis in which we focus on exogenous disappearances of analysts caused by brokerage house mergers or closures following Kelly and Ljungqvist (2012). Employing a difference-indifferences (DID) approach, we observe a significant increase in firms’ earnings management, ex ante probability of committing material financial misstatement (F-score), and CEO excess compensation after they lose analyst coverage due to a brokerage merger or closure. However, consistent with the baseline OLS results, this pattern only holds for the loss of coverage by analysts with related industry experience. Our collective findings indicate that industry expert analysts provide external monitoring of firm managers and improve corporate governance. However, it is not clear how these analysts play such a role. While much of analyst monitoring may take place behind the scenes in private communication with management and hence is not observable, we uncover some evidence on how they exert monitoring influence. First, using textual analysis, we find that industry expert analysts are 31% more likely to discuss governance-related issues in their reports compared to other analysts. In addition, coverage by industry expert analysts discussing corporate governance issues in their reports is associated with even a lesser degree of agency problems at covered firms. Second, we examine whether issuing cash flow forecasts is a method that industry expert analysts utilize to discourage earnings management by firms. This idea is based on the insight from McInnis and Collins (2011), who find that analysts issuing cash flow forecasts alongside earnings forecasts increase firms’ difficulties in engaging in accruals manipulation. Our analysis shows that industry expert analysts indeed are more likely to issue cash flow forecasts, and firms covered by industry expert analysts issuing cash flow forecasts display even better financial reporting quality. Finally, we examine analyst deviations from management guidance when making earnings forecasts, which may represent a less costly means for analysts to signal their disagreements with management (e.g., Feng and McVay (2010) and Louis, Sun and Urcan (2013)). We find that industry expert analysts are more likely to deviate from management guidance and their deviation is associated with improved monitoring effectiveness. Thus, we highlight several subtle mechanisms whereby industry expert analysts can improve firms’ corporate governance outcomes. 5 Overall, our evidence provides strong support for the effective monitoring hypothesis. However, certain situations may arise in which industry expert analysts’ judgement may be impaired. We examine two such situations—analysts with professional ties to firm management and analysts whose employers have investment banking relationships with firms. We find that having a professional tie to firm management emanating from overlapping industry employment experience does not hinder analysts’ monitoring ability. On the other hand, industry expert analysts with investment banking relationships with the covered firm do not provide effective monitoring, suggesting that the benefit of industry experience is offset by the incentives created by investment banking business. This is consistent with the literature on investment banking conflicts where such affiliation results in analysts catering to firm management. Our study makes three primary contributions to the literature. First, our investigation into multiple major corporate policies conveys a consistent message that industry expertise is a critical element in enabling financial analysts to perform their external monitoring function in followed firms. We complement recent studies on analysts’ monitoring role by identifying which analysts are more effective monitors. In this respect, our evidence also complements the finding by Yu (2008) that analysts working for top brokers and with longer forecasting experience have a larger effect on firms’ earnings management. These results together highlight the importance of exploring analyst heterogeneities. It is worth noting that the effect of analyst industry expertise we document is incremental to those of an analyst’s general, industry-specific, and firm-specific forecasting experience and thus represents a new and salient attribute that impacts analysts’ ability to fulfill their monitoring roles. Furthermore, unlike general, industry-specific, or firm-specific forecasting experiences, which analysts develop as outsiders looking in, the industry knowledge we focus on is obtained by analysts as insiders from previously working in certain industries and thus is capable of providing a deeper and more complete understanding of those sectors. Second, we add to the emerging literature that investigates the implications of industry expertise for a variety of economic agents in corporate and financial market settings, such as CEOs, corporate directors, and investment banks (e.g., Liu and Ritter (2011), Custodio and Mertzer (2013), Wang, Xie and Zhang (2015)). Our paper is the first to demonstrate the importance of industry expertise for the monitoring and corporate governance role of financial analysts. As such, our findings complement the evidence in Bradley, Gokkaya, and Liu (2015) that industry expertise enables analysts to make more accurate earnings forecasts as well as the evidence in Dass et al. 6 (2013), Masulis et al. (2014) and Wang, Xie, and Zhu (2015) that industry expertise facilitates the corporate governance functions of corporate boards. Finally, we advance the literature’s understanding of how sell-side analysts provide external monitoring of their coverage firms, an issue on which prior research is largely silent. We explore a large set of potential channels and demonstrate the importance of three analyst monitoring mechanisms: discussing corporate governance issues in research reports, issuing cash flow forecasts simultaneously with earnings forecasts, and deviating from management guidance in earnings forecasts. 2. Data and descriptive statistics The primary data employed in this study are constructed from a number of sources. Appendix A gives a detailed description of our sample construction process. We first pull all analysts’ annual EPS forecasts from I/B/E/S over the period of 1988 to 2011.4 This results in 17,775 unique firms and 16,766 unique analysts. We then merge this sample with CRSP/Compustat to obtain firm financial statement and stock price and return information. We next identify sell-side analysts who provided at least one annual earnings forecast and merge this sample with the I/B/E/S recommendation file to retrieve analyst last names, first name initials and brokerage house information. We eliminate analysts with missing information on first name initials and last names and also discard analyst research teams since only the last names of analyst team members can be obtained from I/B/E/S. This initial screening results in 9,275 unique analysts following 6,713 firms from 1988 to 2011. We search Zoominfo.com, an employment background indexing website to capture the surviving analysts’ full first names. We follow a very conservative approach in our web search and require that analyst last names, first name initials, and brokerage houses match the information gathered from I/B/E/S. This leaves us with 4,845 analysts covering 6,391 unique firms. For each remaining analyst in our sample, we manually collect information on employment backgrounds from LinkedIn.com, the world’s largest professional network.5 We search for the analyst’s full name along with the corresponding brokerage house’s name and collect detailed information on the names of the analyst’s pre-analyst employers and years of employment. 4 Our sample starts in 1988 because we follow Collins and Hribar (2002) in measuring accruals based on information from the cash flow statement, which is not available until 1988. 5 In section 5.4 we discuss two potential selection issues with using analysts’ LinkedIn profiles. 7 We next break down analysts’ employment experience into “related” and “unrelated” at the firm-level within their coverage portfolio. An analyst is coded as having “related industry experience” in a covered firm if the firm and the analyst’s prior employer(s) are from the same industry based on the 4-digit Global Industry Classification System (GICS) classification codes, which create 24 industry groups. Pre-analyst work experience is defined as “unrelated experience” if the GICSindustry classifications of the followed firm and the analyst’s prior employer(s) do not match. Analysts without pre-analyst industry work experience are classified as “inexperienced.” We use the information from Compustat and CRSP to identify the industry of analysts’ prior employers that are publicly traded. For former employers that are privately held, we manually obtain their business descriptions from a variety of sources, including LinkedIn, Hoover’s, Thomson Financial’s Security Data Corporation (SDC) database, the Securities and Exchange Commission’s (SEC) website, business news websites (e.g., Bloomberg and Business Week) and official corporate websites. To illustrate our data collection process, consider a sell-side analyst in our sample that previously worked at Conexant Systems (a private company) as a Process Development Engineer for 3 years and 1 month. We obtain Conexant Systems’s industry classification (semiconductors, GICS=4520) from the company’s LinkedIn page. In a given year, this analyst covers Ultratech Inc. and Cabot Microelectronics, both of which share the same primary 4-digit GICS as Conexant Systems. Therefore, he is defined as having related industry experience for these two firms. The analyst also covers Polypore International and Dynamic Materials (capital goods, GICS=2010) RTI International Metals and Titanium Metals (materials, GICS=1510), which do not have the same GICS as his prior employer. Therefore, the analyst is defined as having unrelated industry experience for his coverage in these latter cases. ***Insert Table 1 here*** In Panel A of Table 1 we provide the top five 4-digit GICS industries from which analysts obtain pre-analyst work experience as well as the top five industries with the most industry expert analysts. 6 Pre-analyst work experience most often comes from Software & Services followed by Diversified Financials. Not surprisingly, these industries also have the highest percentage of analysts 6 In the left half of the panel, we aim to show the distribution of industries from which analysts in our sample obtained any (not necessarily related) pre-analyst industry experience. For instance, of the 1,272 analysts in our sample who have pre-analyst industry experience, 162 or 12.7% (=162/1,272) obtained experience in the Software & Services industry. In the right half of the panel, we aim to show the distribution of industry expert analysts across industries. In our sample, there are 4,419 instances where in a given year an analyst covers at least one stock on which he/she has related industry expertise. We then separate these related analyst-industry-year combinations into 24 4-digit GICS industries. There are 791 related analyst-year-industry combinations in the Software & Services industry. The 17.9% figure shown in the column for this industry is computed as 791/4,419. 8 with related prior industry experience. However, the third through fifth most populated industries in which analysts have prior work experience are Commercial Services & Supplies, Banks, and Media, while Health Care Equipment & Services, Retailing, and Technology Hardware & Equipment represent the corresponding top industries with industry expert analysts. Overall, both pre-analyst work experience and industry expert analysts have a fairly wide distribution across all 24 GICS industries.7 Panel B of Table 1 presents summary statistics of key variables. Appendix B provides a detailed explanation for the construction of these variables. All variables are winsorized at the 0.5% and 99.5% tails to reduce the impact of outliers. We note that the number of firm-year observations associated with each dependent variable is different, because the construction, time period, and size of samples vary across the different analyses we conduct in the paper. For expositional convenience, the summary statistics of firm financial and analyst coverage characteristics are based on the sample used for the accrual-based earnings management analysis, which consists of 24,918 firm-year observations from 1988 to 2011. For the dependent variables, the average firm has a ratio of absolute discretionary accruals to total assets of 0.047, an F-score residual of -0.007, and total CEO compensation of $3.434 million. About 1.7% of firm-year observations witness intentional financial misreporting that leads to earnings restatements, and about 2.7% of firm-year observations experience forced CEO turnovers. In terms of firm financial characteristics, the average firm has a market value of equity (Market cap) of $4.6 billion, a market-to-book ratio (Market-to-book) of 3.2, a return on assets (ROA) of 3.4%, a growth rate in the book value of total assets (Total assets growth) of 22.9%, a standard deviation of cash flows divided by total assets (Cash flow volatility) of 22.2%, and a ratio of external financing to total assets (External financing) of 2.6%. In addition, we control for equity ownership held by dedicated and non-dedicated institutional investors in each firm.8 The average ownership by dedicated institutional investors is 8.3% and the average ownership by non-dedicated institutional investors is 55.3%. These values are in line to those reported in other studies (e.g., Bushee and Noe (2000) and Chen, Harford, and Li (2007)). Regarding analyst coverage variables, the average firm is covered by 3.97 sell-side analysts, with 0.90 having related pre-analyst industry experience (Related analysts), 1.70 having unrelated pre7 To ensure our results are not driven by the most represented industries, in untabulated analysis we limit our sample to the non-top 5 industries and continue to find similar results. 8 We follow Chen, Harford, and Li (2007) and identify dedicated institutional investors based on Bushee’s (1998) classification methodology. 9 analyst industry experience (Unrelated analysts), and 1.38 having no pre-analyst industry experience (Inexperienced analysts). 9 We separately calculate an analyst’s general and firm-specific forecasting experience as the number of years since the analyst initially appeared in I/B/E/S (Experience as analyst) and the number of years in which the analyst provided coverage for a particular firm (Experience with firm). For the average firm in our sample, the average analyst covering the firm has 6.31 years of general experience, 2.45 years of firm-specific experience, a research portfolio consisting of 12.56 unique firms (Analyst portfolio size), and 56.5% of analysts work at a top brokerage house (Analysts from top brokers) while 13.1% attain all-star status (All-star analyst). 3. Empirical results In this section we present the results from our examination of the effective monitor hypothesis versus the impaired monitor hypothesis. Sections 3.1 and 3.2 evaluate the quality of a firm’s financial disclosure policy in relation to coverage by analysts with related industry experience, with Section 3.1 based on accruals-based earnings management and Section 3.2 on measures of financial misreporting. Section 3.3 investigates the effect of analyst industry experience on CEO compensation policies. Section 3.4 examines whether the presence of analysts with related industry experience affects the sensitivity of forced CEO turnovers to firm performance. 3.1. The effect of analyst industry expertise on earnings management We examine whether the presence of industry expert analysts (Related analysts) is related to firms’ earnings management, which we measure using the absolute value of discretionary accruals (Abs_DA). Managers have incentives to manage earnings upward (e.g., to inflate the stock price prior to their stock sales, or to hit performance targets for bonuses) as well as downward (e.g., prior to option repricing or management buyouts or as an attempt to smooth earnings). Because our hypotheses are related to the magnitude rather than the direction of earnings management, we follow prior studies, such as Bergstresser and Philippon (2006) and Yu (2008), and use the absolute value of discretionary accruals scaled by total assets as our dependent variable of interest. Dopuch et al. (2011) and Owen, Wu, and Zimmerman (2015) show that measurement error in discretionary 9 Because we only include analysts with a LinkedIn profile, the number of analysts following a firm in our sample is naturally lower. An alternative sampling strategy is to include analysts without a LinkedIn profile and simply classify them as either unrelated or inexperienced analysts. Of course, doing so will misclassify some analysts with related industry experience as otherwise and therefore bias against our finding any difference in monitoring effectiveness between related analysts and the other two groups. Nevertheless, all our results continue to hold under this sampling approach. 10 accruals is related to the homogeneity of the underlying industry. Therefore, we estimate discretionary accruals based on the accrual models developed by Owens, Wu, and Zimmerman (2015) that explicitly take into account heterogeneities in firms’ economic fundamentals.10 However, to the extent that empirical proxies for firms’ business model shocks are noisy, model misspecification remains possible and the discretionary accrual estimates may still be contaminated by measurement errors (Owens et al. (2015)). We mitigate this concern by supplementing our discretionary accruals analysis with an examination of two alternative financial misreporting measures that do not require the modeling of firms’ accrual generating processes (see Section 3.2). 3.1.1. Baseline OLS regressions We first estimate OLS regressions of earnings management against analyst coverage while explicitly controlling for a battery of firm and analyst-level characteristics as discussed in Section 2. We include firm and year-fixed effects in the regressions to mitigate potential concerns about timeinvariant unobservable firm characteristics being correlated with both earnings management and analyst coverage. Our formal model is specified in equation (1). To mitigate reverse causality concerns, all firm financial characteristics included as control variables are lagged by one year. Abs_DAi,t= β0 +β1Total analystsi,t-1 [or β2Related analysts i,t-1 +β3Unrelated analysts i,t-1 +β4Inexperienced analysts i,t-1] +β5Market cap i,t-1 +β6Market-to-book i,t-1 +β7ROA i,t-1 +β8Total assets growth i,t-1 +β9Cash flow volatilityi,t-1 +β10External financing i,t-1 +β11Institutional ownership (dedicated) i,t-1 +β12Institutional ownership (non-dedicated) i,t-1 +β13Experience as analysti,t-1+β14Experience with firmi,t-1+β15Analysts from top brokersi,t-1 +β16All-star analysts i,t-1 +β17Analyst portfolio size i,t-1 +Firm and year fixed effects + ε (1) ***Insert Table 2 here*** Table 2, Panel A reports the regressions results. As in the rest of the tables, robust t-statistics based on standard errors adjusted for heteroskedasticity and firm-level clustering are in the parentheses below coefficient estimates. As a starting point of our analysis, we replicate the finding by Yu (2008) and Irani and Oesch (2013) that analyst coverage is negatively related to firms’ earnings 10 Specifically, we use the accrual model specified in equation (13) in Owens et al. (2015) and the firm-specific stock return variation measure in Chun, Kim, Morck, and Yeung (2008) to capture the heterogeneity in firms’ underlying economics and proxy for business model shocks. Our results are also robust to using the models specified in equations (14) and (15). In addition, we obtain qualitatively similar results with higher statistical significance when we use the discretionary accruals estimated from the modified Jones (1991) model in Dechow, Sloan, and Sweeney (1996) and Warfield, Wild, and Wild (1995) and the performance-adjusted discretionary accruals in Kothari, Leone, and Wasley (2005) and Gong, Louis, and Sun (2008). 11 management. Specifically, in model 1, we regress a firm’s absolute value of discretionary accruals (Abs_DA) against the total number of analysts following the firm. Consistent with analysts playing an external monitoring role, we find that total analyst coverage is significantly and negatively associated with Abs_DA. Next we decompose the total number of analysts following a firm into analysts with related industry experience (Related analysts), analysts with unrelated industry experience (Unrelated analysts), and analysts without pre-analyst industry work experience (Inexperienced analysts). We reestimate the abnormal accruals regression and present the results in model 2. We find that only coverage by analysts with related industry experience is significantly and negatively associated with a firm’s abnormal accruals. While the coefficients on the other two types of analyst coverage are also negative, they are not statistically significant. These findings suggest that the negative association between analyst coverage and earnings management documented by prior studies (Yu (2008), Irani and Oesch (2013), and Chen, Harford, and Lin (2015)) is largely driven by analysts with related industry experience. As such, they support the effective monitor hypothesis that financial analysts with related industry experience provide more effective monitoring of firms’ financial disclosure. An empirical challenge in identifying a causal relation between analyst coverage and corporate financial disclosure is the possibility that an analyst’s decision to cover a firm is associated with observable factors that could also affect the firm’s financial reporting (Healy, Hutton, and Palepu (1999)). Therefore, we follow Yu (2008) and construct residual analyst coverage to mitigate this potential endogeneity concern. Residual coverage is defined as the component of analyst coverage uncorrelated with firm-specific control variables and the main determinants of earnings management as listed in Panel B of Table 1.11 We then use residual coverage as our main proxy for analyst coverage in the remaining analysis. The regression results reported in models 3 and 4 of Table 2, Panel A are very similar to those in models 1 and 2; only coverage by analysts with related industry experience is significantly associated with firms’ earnings management behavior. 3.1.2. Brokerage mergers and closures as quasi-natural experiments 11 We estimate a series of analyst coverage regressions, where the dependent variables are the number of 1) total analysts 2) analysts with related industry experience, 3) analysts with unrelated industry experience, and 4) analysts without prior industry experience. The residuals from these regressions are then labeled as “residual coverage” for the corresponding type of analysts. Untabulated results indicate that the determinants of analyst coverage are quantitatively and qualitatively similar across these three types of analysts. For instance, analyst coverage is positively related to firm size and growth rate of assets, but negatively related to external financing activities. 12 As an alternative approach to mitigating endogeneity concerns, we exploit a unique setting where there is an unexpected exogenous shock to a firm’s analyst coverage independent of the firm’s characteristics or earnings management behavior. Specifically, this identification strategy relies on two plausibly exogenous quasi-natural experiments, namely, brokerage closures (Kelly and Ljungqvist (2012)) and brokerage mergers (Hong and Kacpercyzk (2010)). These brokerage-related events are used by a growing body of research to study the impact of financial analyst coverage on a wide array of firm policies. Kelly and Ljungqvist (2012) argue that following a merger, termination of coverage by the surviving broker could be endogenous because it chose to no longer cover the stock. Thus, we follow their approach and focus on coverage losses resulting from a merger where a stock was covered by analysts from both brokers before the merger and by only one of the analysts after the merger. We identify 17 brokerage house closures and 37 brokerage house mergers over the period of 1988 to 2008 and construct a sample of 701 treatment firms that were covered by the closed or merged brokers prior to these events.12 We employ a difference-in-differences (DID) approach where we focus on the change in a firm’s abnormal accruals from year t-1 to year t+1, with year t being the year in which the firm loses analyst coverage due to a brokerage closure or merger. We compare the change in abnormal accruals of treatment firms (treatment firm difference) with the change in abnormal accruals of control firms over the same time period (control firm difference). This allows us to control for time trends that may be common to both treatment and control firms and mitigates concerns related to reverse causality or omitted variables. Following Kelly and Ljungqvist (2012), we construct an initial sample of control firms that are covered by at least one analyst and are in the same size and book-to-market quintiles as the treatment firms in the month of June preceding the broker events. We further require that control firms do not experience analyst losses in the year of or the year before the broker event. We retain the control firm closest to the treatment firm in the number of analysts providing coverage.13 Column 1 of Panel B in Table 2 presents the results for all analyst losses. Consistent with the baseline regressions in Panel A, we find that firms losing analyst coverage emanating from brokerage 12 Brokerage closure data is from Kelly and Ljungqvist (2012) and He and Tian (2013) over 1993 and 2008. Brokerage house merger information comes from a list of M&A activities in the investment banking industry compiled by Wang, Xie, and Zhang (2015) from various sources. As discussed earlier, the sample of treatment firms is smaller than in previous studies because we limit our attention to analysts with a LinkedIn profile. However, if we include all nonLinkedIn analysts and classify them as inexperienced we obtain qualitatively similar results. 13 For robustness, we also require that treatment and control firms come from the same 4-digit GICS industry. This additional matching criterion reduces the number of treatment firms to 637. Results from the smaller sample are very similar to those reported in the paper. 13 mergers or closures engage in more aggressive earnings management behavior in the post-event year compared to the pre-event year (average treatment firm difference=0.27%, t-stat=1.88). Focusing on the difference in differences (DID) between treatment and control firms, our results indicate significantly more aggressive earnings manipulation for treatment firms (average DID=0.39%, tstat=2.15). More important for our purpose, when we differentiate among the types of analysts lost by firms in columns 2 and 3, we find that only the loss of coverage by analysts with related industry expertise precipitates a significant increase in earnings management. Specifically, in the case of industry expert analyst losses, the average DID in abnormal accruals is 1.21%, which is significant at the 1% level. It is also economically important given the mean and median values of Abs_DA in our sample are 4.7% and 2.9%, respectively. In contrast, in the case of other analyst losses, the average DID in abnormal accruals is 0.23%, insignificantly different from zero. A further comparison of the DID between firms losing industry expert analysts versus other analysts indicates that the loss of industry expert analysts leads to more pronounced increases in earnings management (difference in DID=0.98%, t-stat=-2.21). Overall, these results provide further support for the effective monitor hypothesis and suggest that related industry expertise is a valuable attribute for financial analysts to possess in order to effectively perform their monitoring function.14 3.2. Industry expert analyst coverage and financial misreporting by firms In this section, we examine the relation between industry expert analyst coverage and the likelihood of firms committing financial misreporting, a more egregious type of behavior than accrual-based earnings management. Following Wang, Xie, and Zhu (2015), we employ both ex-ante and ex-post measures of financial misreporting. We use the F-score developed by Dechow, Ge, Larson, and Sloan (2011) to measure the ex-ante probability of a firm committing financial misrepresentation. Our ex post indicator of financial misreporting is earnings restatements by firms. 3.2.1. Analyst industry expertise and F-score Dechow et al. (2011) construct a comprehensive database through careful examination of firms subject to U.S. Securities and Exchange Commission (SEC) enforcement actions as indicated in Accounting and Auditing Enforcement Releases (AAERs). They then identify a sample of firms 14 As an alternative to the DID approach, we also use the regression approach of Chen, Harford, and Lin (2015) and Irani and Oesch (2013), in which we regress changes in the absolute value of discretionary accruals against exogenous losses in analyst coverage along with changes in the control variables. We find that a firm’s abnormal accurals experience significant increases only after the loss of coverage by analysts with related industry experience. 14 that have misstated at least one of their quarterly or annual financial statements with the intent of misleading investors. The F-score is created from a model that analyzes the financial characteristics of these misstating firms. Given that part of the construct for the F-score uses accruals information, we orthogonalize the F-score against discretionary accruals and use the residuals as the dependent variable for our analysis.15 This allows us to provide evidence that is truly complementary to the discretionary accruals-based results reported in Section 3.1. We estimate regressions of F-score residuals against different types of analyst coverage, along with the same set of control variables as included in equation (1). Panel A of Table 3 presents the regression results. ***Insert Table 3 here*** The most immediate takeaway from Model 1 of Panel A is that the intensity of analyst coverage is significantly and negatively related to a firm’s F-score residual, consistent with analysts playing a monitoring role and reducing firms’ ex-ante likelihood of earnings manipulation. However, once again, we find that this result is driven by coverage by analysts with related industry experience (Model 2). Conversely, coverage by analysts with unrelated industry experience or no industry experience is not significantly associated with the ex ante probability of intentional financial misrepresentation. We next reexamine the relation between industry expert analyst coverage and the F-score residuals using exogenous shocks to a firm’s analyst coverage due to brokerage closures or mergers. As in the earnings management analysis in Section 3.1.2, we employ a DID approach around a twoyear event window surrounding the brokerage closures or mergers and examine the change in a firm’s F-score residual from year t-1 to year t+1, with year t representing the year in which the firm loses analyst coverage due to such brokerage events. Panel B of Table 3 presents these results. In column 1, we find that firms experience a significant increase in the F-score residuals when they experience a loss of analyst coverage, consistent with monitoring by analysts reducing firms’ ex ante probability of committing serious financial misreporting. In column 2, when we differentiate between the types of analyst coverage lost by firms, we find that only the loss of coverage by industry expert analysts leads to a significant increase in a firm’s F-scores. This result reaffirms our finding from Panel A about the importance of related industry experience in facilitating the monitoring role of analysts, thereby lending further support to the effective monitor hypothesis. 15 Using the original F-score produces qualitatively similar results. 15 3.2.2. Analyst industry expertise and earnings restatements We construct our sample of earnings restatements from two data sources. We start with the U.S. General Accounting Office’s (GAO) Financial Statement Restatement Database as published in 2003 and 2007. The GAO dataset is manually constructed using a Lexis-Nexis text-based search for the variants of the word “restate,” and contains restatements announced between January 1, 1997 and June 30, 2006. We then extend the GAO dataset to 2011 by using the accounting restatements obtained from the Audit Analytics restatement database. Hennes, Leone, and Miller (2008) develop a methodology that classifies a restatement as an irregularity if it satisfies at least one of the three criteria: (i) variants of the words "irregularity" or "fraud" were explicitly used in restatement announcements or relevant filings in the four years around the restatement; (ii) the misstatements came under SEC or DOJ investigations; and (iii) independent investigations were launched by the boards of directors of restatement firms. In a sample of restatements between 2002 and 2005, they demonstrate the importance and effectiveness of their classification scheme by showing that compared to error restatements, irregularity restatements are met with significantly more negative announcement returns (on average: -14% vs. -2%), are followed by shareholder class action lawsuits at a significantly higher rate, and lead to significantly more CEO/CFO turnovers. Therefore, we focus on irregularity restatements because they are more likely to represent instances of deliberate and material earnings manipulation by managers. For each irregularity restatement, we obtain information on the specific years for which financial results were restated due to earnings manipulation. We merge the restated year information with our comprehensive CRSP-CompustatI/B/E/S sample and obtain a sample of 19,772 firm-year observations from 1997 to 2011. In about 1.7% (336) firm-years, earnings were restated due to accounting irregularities.16 We estimate logistic regressions to examine whether coverage by industry expert analysts is related to the probability of a firm committing intentional financial misreporting in a given year. The dependent variable is equal to one for restated firm-years and zero otherwise. In addition to the covariates in equation (1), we also control for how prone an industry is to earnings restatements. The rationale for this control is that industries where firms have a high proclivity to commit accounting fraud and restate earnings may be considered undesirable and thus less likely to produce employees who would go on to become financial analysts in the future. This possibility creates a potentially artificial negative relation 16 There are 428 firm-years in which earnings were restated due to accounting errors. Due to their substantially less serious nature compared to accounting irregularities, we treat the error affected firm-years as non-restated years. Our results are nearly identical if we remove these observations from our analysis. 16 between a firm’s probability of earnings restatement and coverage by analysts with related industry experience.17 To mitigate this concern, we construct a measure to capture an industry’s exposure to restatements. Specifically, we first compute the percentage of firms in each industry that were associated with an irregularity restatement over the past ten years. We create a binary variable, High restate exposure, that is equal to one for industry groups with above sample median restatement frequency and zero otherwise. Table 4 reports the results. The coefficient estimates presented under Model 1 suggest that the probability of intentional financial misreporting is negatively and significantly associated with the total number of analysts following. However, as model 2 indicates, this significant and negative relation is largely attributable to industry expert analysts, and there is no evidence that coverage by other analysts is related to the probability of financial misreporting. Economically, the coefficient estimates in model 2 imply that the likelihood of intentionally misreporting financial restatements is lower by 35% per one standard deviation increase in coverage by industry expert analysts.18 Finally, it is worth noting that the High restate exposure indicator has a significantly positive coefficient, suggesting that firms in industries that historically are more fraud prone indeed have a higher probability of committing financial misrepresentation. ***Insert Table 4 here*** Overall, the evidence in Sections 3.1 and 3.2 lends strong support to the effective monitor hypothesis that industry expertise improves the effectiveness of analysts’ monitoring and curtailing of firms’ earnings manipulation activities as reflected in abnormal accruals, F-score, and irregularitydriven financial restatements.19 3.3. Analyst industry expertise and CEO compensation In the context of firms’ executive compensation policy, Chen, Harford, and Lin (2015) show that analyst coverage can serve as an additional layer of market discipline to reduce excess 17 We thank the editor and referee for suggesting this possibility. In unreported analyses, we formally evaluate whether industries that are more stable, homogeneous, and successful are more likely to produce employees that go on to become analysts of the same industry. We find no evidence that the percentage of industry expert analysts among all analysts covering a particular industry is related to the homogeneity, volatility, and performance of the industry. 18 We are not able to use brokerage closures/mergers in this setting because only two firms announced irregularity restatements after losing coverage by industry expert analysts. 19 In untabulated results, we use a firm being the target of a securities class-action lawsuit as an alternative ex post measure of financial misreporting, to the extent that earnings manipulation with an intention to mislead investors can lead to class action lawsuits (DuCharme, Malatesta, and Sefcik (2004)). We obtain a comprehensive sample of 3,566 securities class action lawsuits on 2,152 unique firms from 1996 to 2011 from the Securities Class Action Clearinghouse website (http://securities.stanford.edu/). We find that coverage by industry expert analysts are associated with a significantly lower probability of a coverage firm being targeted by a class action lawsuit, whereas coverage by other analysts is not significantly related to the probability of being sued. 17 compensation received by top executives. We argue that analysts’ industry expertise can provide them with a comparative advantage in understanding industry and firm-specific factors that have economic impact on the competitive level of executive compensation. As a result, they are more likely to recognize large discrepancies between a CEO’s compensation and her fair and competitive level of pay. Their opinions are also likely to be taken seriously by managers and boards because of analysts’ unique position to influence the stock price.20 As such, the effective monitor hypothesis predicts that firms followed by more industry expert analysts are less likely to award exorbitant compensation packages to their top executives. The impaired monitor hypothesis, however, argues that analysts with related industry experience may be connected to the managers of the firms they follow and thus have less incentive to question overly generous CEO pay. To test these predictions, we obtain CEO compensation information including salary, bonus, stock and option grants, and long-term incentive plans from Standard & Poor’s ExecuComp database for our sample. We assess the impact of analyst industry expertise on excess CEO compensation. To estimate excess CEO compensation, we regress the logarithmic transformation of total CEO compensation against a firm’s market capitalization, buy-and-hold abnormal returns over CRSP value-weighted returns, stock return volatility, as well as industry (based on two-digit SIC codes) and year fixed effects. The residuals from this estimation is our measure of excess CEO compensation (e.g., Hartzell, Ofek, and Yermack (2004), Cai, Garner, and Walkling (2009)). We estimate OLS regressions of excess CEO compensation against analyst coverage while controlling for firm financial and governance characteristics as well as firm and year-fixed effects. The regression model is specified as follows. CEO excess compensationi,t= β0 +β1Total analysts (Residual) i,t-1 [or β2Related analysts (Residual) i,t-1 +β3Unrelated analysts (Residual)i, t-1+β4Inexperienced analysts (Residual) i, t-1 +β5Market capi, t-1 +β6Tobin’s Q i, t-1 +β7Leverage i, t-1 +β8R&D/sale i, t-1 +β9Capx/sale i, t-1 +β10Xad/sale i, t-1 +β11Abnormal ROA i, t-1 +β12Abnormal stock rets i, t-1 +β13CEO tenure i, t-1 +β14Firm age i, t-1 +β15Institutional ownership (dedicated) i, t-1+β16 Institutional ownership (non-dedicated) i, t-1 +β17Experience as analyst i, t-1 +β18Experience with firm i, t-1 +β19Analysts from top brokers i, t-1 +β20 All-star analysts i, t-1 +β21Analyst portfolio size i, t-1 +Firm and year fixed effects +ε (2) ***Insert Table 5 here*** Panel A of Table 5 presents the regression results. Consistent with the effective monitor hypothesis, we find that analyst coverage is associated with significantly lower levels of excess CEO 20 According to the CFO survey by Graham, Harvey, and Rajgopal (2005), about 54.6% of respondents rank analysts among the two most important stock price setters for their companies. 18 compensation (model 1), but the relation is again driven by analysts with related industry expertise in their followed firms (model 2). In terms of economic significance, the coefficient estimates from model 2 imply that for a one standard deviation increase in the number of industry expert analysts, excess CEO compensation is lower by 4.8%, representing roughly a reduction of $165,000 in CEO compensation given the average CEO pay of $3.43 million in our sample. In further analysis, we exploit exogenous shocks to a firm’s analyst coverage created by brokerage house closures and mergers. Specifically, we focus on the change in a firm’s CEO excess compensation from year t-1 to year t+1, with year t being the year in which the firm loses analyst coverage due to a brokerage closure or merger. We then compare the changes in CEO compensation of treatment firms experiencing loss of analyst coverage to those of similar control firms over the same time period. Panel B of Table 5 presents the results. We find that the excess compensation for a treatment firm’s CEO experiences a significant increase following the loss of analyst coverage (column 1) compared to control firms. When we break down the loss of analyst coverage based on the types of analysts, we find that only the loss of industry expert analysts leads to a significant increase in CEO compensation (column 2). The difference in DID is also significant between firms losing industry expert analysts and those losing other analysts (column 4). 3.4. Industry expert analyst coverage and CEO turnover-performance sensitivity In this section we investigate whether coverage by industry expert analysts is related to the responsiveness of boards in removing poorly performing CEOs. Analysts with expertise in a firm’s related industries are likely to have a more thorough understanding of the firm’s operating environment, and the opportunities and challenges it faces. Therefore, they are likely to be able to better analyze and evaluate major corporate strategies and decision making. When firm performance disappoints, they can put more pressure on the firm to change its strategic direction and, if necessary, change its leadership. To examine if analyst industry expertise plays a role in forcing out poorly performing CEOs, we follow Jenter and Kanaan (2015) and construct our primary dataset of CEO forced turnovers from Standard & Poor’s ExecuComp database. We then use news reports collected from LexisNexis to ascertain that CEO turnover is forced, not due to any of the following reasons such as 19 death, health problems, retirement, or promotion to a higher/comparable position at another firm (e.g., Parrino (1997), Huson, Malatesta, and Parrino (2004), among others).21 We estimate a logistic model to assess the marginal impact of industry expert analysts on the CEO turnover-performance sensitivity. The dependent variable equals one for years in which a forced CEO turnover event occurs, and zero otherwise. The key independent variables include abnormal firm performance, measured by the firm’s industry-adjusted abnormal return on assets (ROA) over the previous year, and its interaction terms with the three types of analyst coverage. 22 This model specification allows CEO turnover-performance sensitivity to vary with the coverage of different types of analysts. ***Insert Table 6 here*** Table 6 reports the results. We find a significantly negative coefficient on abnormal ROA, affirming the well documented negative relation between the probability of forced CEO turnover and firm performance (see, e.g., Coughlan and Schmidt (1985), Weisbach, (1988), Fich and Shivdasani (2006)). More importantly, the interaction term between abnormal performance and coverage by industry expert analysts has a significantly negative coefficient, while the coefficients on the other two interaction terms are insignificant. This evidence is consistent with the notion that the external monitoring by industry expert analysts makes corporate boards more responsive to poor firm performance in disciplining poorly performing CEOs. Conversely, coverage by analysts without related industry expertise is not related to the CEO turnover-performance sensitivity.23 4. How do analysts monitor? So far we have provided evidence consistent with analysts with related industry experience performing an important external monitoring function for firms they cover. However, it is not entirely clear how these analysts exert their monitoring efforts. Therefore, in this section we aim to better understand the mechanisms through which industry expert analysts influence firms’ corporate 21 As described by Jenter and Kannan (2015), “all departures for which the press reports that the CEO is fired, forced out, or retires or resigns due to policy differences or pressure are classified as forced. All other departures for CEOs above and including age 60 are classified as voluntary. Departures for CEOs below age 60 are reviewed further and classified as forced if either the press does not report the reason as death, poor health, or the acceptance of another position (including the chairmanship of the board), or the press reports that the CEO is retiring, but does not announce the retirement at least six months before the succession. Finally, the cases classified as forced can be reclassified if the reports convincingly explain the departure as due to reasons that are unrelated to the firm’s activities.” 22 We obtain similar results when we measure firm performance using abnormal stock returns over the previous year. 23 Only one firm experienced a forced CEO turnover following the loss of industry expert analyst coverage, making it econometrically unreliable to analyze turnover-performance sensitivity using shocks created by brokerage closures and mergers. 20 governance and constrain managerial opportunism.24 A challenge for our analysis is that much of analysts’ monitoring activities may be unobservable to outsiders, and even the observable actions can be quite subtle because analysts that aggressively question and criticize corporate policies and strategies risk alienating prized access to firm executives. Therefore, analysts are more likely to convey their concerns and disapprovals regarding certain managerial decision making to firms through private communication. Nevertheless, we examine several observable actions of analysts to shed light on how they play a monitoring role. 4.1. Governance discussions in analyst reports We first consider discussions of governance-related issues in analyst research reports. We obtain full-text analyst research reports from Thomson Reuters Investext for our sample of analysts and extract the header information for each report. We obtain the name of the analyst writing the report, the broker, the firm and date published. We match the Thomson Reuters Investext header information with I/B/E/S broker-analyst-firm pairs and manually verify these matches. We then parse analyst reports for governance-related keywords in the following categories: ‘corporate governance,’ ‘financial restatement,’ ‘shareholder proposal,’ ‘shareholder activism,’ ‘CEO compensation,’ or ‘CEO turnover’. To come up with our list of keywords, we read 1000 randomly selected analyst reports and identify the ways in which analysts discuss corporate governance.25 We look at every bigram word combination in compiling the keyword list (see Appendix C). We then merge the parsing results with our main sample. Analyst-year observations are identified as “governance-related” for the corresponding coverage firm if one of the research reports issued by the following analyst contains one of our keywords and zero otherwise. We find (untabulated) that 9.7% of industry expert analysts discuss at least one of these governance issues for their coverage firms compared to 5.6% for non-expert analysts. These differences are statistically significant (t-stat=7.3). As a more formal way to explore the likelihood of governance discussions in research reports, we estimate a multivariate logistic regression controlling for a host of firm and analyst characteristics defined previously. The dependent variable is a binary indicator that equals one if the analyst discusses one of the governance-related issues for their coverage firms in a given year, zero otherwise. The key independent variables are two indicator variables: Related analyst, which is equal to 24 25 We greatly acknowledge many constructive comments from the referee and the editor on this section. We thank Tim Loughran for his valuable advice on the keyword identification strategy. 21 one for analysts with related industry experience in the firms they cover, and Unrelated analyst, which is equal to one for analysts with unrelated industry experience in the firms they cover. Inexperienced analysts serve as the omitted category. We present the regression results in model 1 of Panel A in Table 7. We find that Related analyst has a significant and positive coefficient, suggesting that industry expert analysts are more likely to discuss governance issues in their reports. Economically, the likelihood of such discussions is 31% higher (t-stat=4.47) for industry expert analysts relative to inexperienced analysts. On the other hand, unrelated industry work experience is not significantly associated with discussion of governance issues. ***Insert Table 7 here*** Given that industry expert analysts are more likely to discuss governance issues in their reports, we next examine if this form of monitoring impacts corporate governance outcomes. We reestimate the discretionary accruals regression in equation (1) by partitioning Related analysts into two variables based on whether an industry expert analyst discusses governance issues in research reports for a coverage firm in a given year.26 Column 1 of Panel B presents the regression results with the coefficients on control variables suppressed for brevity. We find that the negative relation between industry expert analyst coverage and discretionary accruals is more pronounced when industry expert analysts issue research reports discussing governance issues (p-value for the coefficient difference: 0.029). This suggests that corporate governance discussion in research reports represents one mechanism through which industry expert analysts monitor firm managers and exert influence over firm governance.27 4.2. Cash flow forecasts McInnis and Collins (2011) show that by issuing cash flow forecasts along with earnings forecasts, analysts may increase firms’ difficulty and costs of engaging in accruals manipulation. Building on their insight, we examine whether cash flow forecast issuance is a mechanism for industry expert analysts to improve firms’ financial reporting quality. Toward that end, we first 26 We conduct analogous tests for other outcome variables, i.e., F-score residuals, earnings restatements, CEO compensation and CEO turnover-performance sensitivity, and find qualitatively similar results (see Table IA.1 in the Internet Appendix). 27 We perform a similar analysis of analysts’ takeaway reports following analyst investor days. These are firm-hosted investor relations events where analysts are invited to meet with executives, hear about the strategy, financial position and provide an opportunity for analysts to ask questions to management. While the sample size is considerably smaller for these takeaway reports, using the same textual analysis strategy, we continue to find that industry expert analysts are more likely to discuss governance-related issues in their analyst takeaway reports published following analyst/investor days compared to non-expert analysts (5.0% versus 2.5%, respectively). 22 examine whether industry expert analysts are more likely to issue cash flow forecasts. In particular, we estimate a logistic regression where the dependent variable equals one if an analyst simultaneously provides both earnings and cash flow forecasts for the coverage firm of interest, and zero otherwise. Results in model 2 of Panel A indicate that industry expert analysts are 37.5% more likely (t-stat=5.62) to issue cash flow forecasts compared to other analysts. Next, we investigate the impact of such forecasts on coverage firms’ earnings management behavior. In particular, we distinguish between industry expert analysts based on issuance of cash flow forecasts and reestimate equation (1). In model 2 of Panel B, we find that the negative relation between industry expert analyst coverage and discretional accruals is stronger when industry expert analysts issue cash flow forecasts (p-value for the coefficient difference: 0.047).28 Thus, our analysis suggests that cash flow forecasts may represent another mechanism through which industry expert analysts perform their external monitoring function. 4.3. Deviations from management guidance In this section, we consider whether industry expert analysts deviate from management earnings guidance as a means to monitor and improve firm governance, and in particular financial reporting. Feng and McVay (2010) suggest that analysts deviate less from management earnings guidance when they have incentives to please management. Similarly, Louis, Sun, and Urcan (2013) argue that intentionally deviating from management guidance is a way in which analysts can subtly express their disagreement with management when they believe the earnings guidance contains a large managed component. Therefore, we first investigate whether industry experts deviate more from management guidance when issuing earnings forecasts. Similar to Louis, Sun, and Urcan (2013), the dependent variable of this analysis is the absolute value of the deviation of analyst i’s earnings forecast from firm j’s earnings guidance for year t scaled by firm j’s stock price on the day prior to the management guidance (abs DEVAF). To construct this variable, we use the last management guidance issued for year t and the analyst’s forecast following the guidance but prior to the actual earnings announcement. The key independent variables are Related analyst and Unrelated analyst, the binary indicators for analyst types defined in Section 4.1. In addition to the firm financial and analyst characteristics controls used in equation (1), we follow Louis, Sun, and Urcan (2013) in adding the following controls: the absolute value of discretionary accruals (Abs_DA); negative 28 In models 1 and 2 of Internet Appendix Table IA.2, we perform similar tests for F-score residuals and restatements. The results are not sensitive to how financial reporting quality is measured. 23 earnings (Loss); market-adjusted excess returns from two days following the management guidance to one day prior to the analyst forecast date (Abret); market-adjusted excess returns over the three months prior to the management guidance’s month (Abretm3); coverage firm earnings and price adjusted forecast error volatility over the past five years (Evolatility, Fvolatility, respectively); change in coverage firm’s earnings over year t-1 and year t (Che); management guidance error (Pae), calculated as the difference between the guidance and the actual earnings for year t; and the average Pae over the past two years (Appae). We also control for the time (in days) by which analyst forecasts lag the management guidance (Gap), because a longer delay by analysts in issuing their forecasts may result in larger deviations from management guidance. The results in Table 7, Panel A, model 3 suggest that compared to inexperienced analysts, industry expert analysts tend to deviate significantly more from management guidance in their earnings forecasts. On the other hand, analysts’ unrelated industry work experience does not factor into such deviations. We next investigate whether industry expert analysts’ intentional deviation from management guidance helps discourage firms’ earnings management. To do so, we re-estimate the abnormal accruals regression specified in equation (1), which we augment by differentiating among industry expert analysts based on whether the absolute value of their deviations from management guidance is above/below the sample median. The results presented in model 3 of Panel B show that firms’ abnormal accruals display a more pronounced negative relation with coverage by industry expert analysts deviating more from management guidance than with coverage by those deviating less (p-value for coefficient difference: 0.005). Models 3 and 4 in Internet Appendix Table IA.2 use F-score residuals and restatements as alternative measures of financial reporting quality and produce similar results. Therefore, our findings suggest that deviating from management guidance is a monitoring device that industry expert analysts use to limit managerial opportunism in corporate financial reporting decisions. 4.4. Earnings conference calls Another possible channel for analysts to express their opinions on corporate governance issues is through their interaction with top executives during the Q&A period of earnings conference calls. Analysts may be vocal about their disapproval of firm policies and directly question management’s actions during these calls. Thus, we examine if this is a likely conduit. We collect transcripts of earnings conference calls through Thomson Reuters Street Events and extract the Q&A section from these transcripts. We then parse the textual content of these sections for the 24 governance keywords described in section 4.1. Of the 24,971 transcripts examined, we find that only 159 conference calls mention governance-related issues. Therefore, it appears that directly questioning management’s governance practices on conference calls is not a strategy that analysts engage in. This perhaps is not entirely surprising given that managers have control over key aspects of their interactions with analysts during conference calls, including who they call on to participate and how they interact with analysts (Cohen, Lou and Malloy (2014)). Therefore, disapproving analysts may not able to get the opportunity to voice their concerns. It is, however, possible that some analysts may discuss governance-related issues during one-on-one “call-backs” immediately following the public earnings conference call, which would be consistent with the ‘behind-the-scenes’ view that at least some monitoring takes place in private communication with management. 5. Additional analyses and robustness tests 5.1. Income increasing vs. incoming decreasing earnings management The discretionary accruals analysis in Section 3.1 suggests that industry expert analysts help restrain firms’ accrual-based earnings management. In this section, we investigate whether these analysts have stronger incentives to monitor and discourage positive (income-increasing) accruals than negative (income-decreasing) accruals. We take two approaches to answer this question. In the first approach, we follow Yu (2008) and create two subsamples based on whether discretionary accruals for a firm year is positive or negative. We then reestimate the discretionary accruals regression separately in the two subsamples. The dependent variable continues to be the absolute value of discretionary accruals (Abs_DA) in each subsample regression. Table 8 presents the regression results. We find that industry expert analyst coverage has a significant and negative coefficient in both subsamples, suggesting that industry expert analysts monitor and reduce both income-increasing and income-decreasing discretionary accruals. Comparing the two coefficients, it appears that industry expert analyst coverage has a larger effect on reducing income-increasing discretionary accruals than income-decreasing ones (p-value for the coefficient difference: 0.011). ***Insert Table 8 here*** As an alternative to the subsample regression analysis, we employ a multinomial logistic regression approach in which we first sort all firm-year observations into quintiles based on the signed discretionary accruals. We then create a dependent variable that takes the value of 1 for observations in the bottom quintile (the most negative discretionary accruals), 2 for observations in the middle three quintiles (moderate discretionary accruals), and 3 for observations in the top 25 quintile (the most positive discretionary accruals). We estimate the multinomial logit regression using the full sample and therefore avoid any potential econometric concerns related to the subsample regression approach. Another advantage of this approach is that the most positive and negative discretionary accruals likely represent the most aggressive earnings management, and by focusing on these cases, we alleviate concerns about measurement errors in the raw values of discretionary accruals. Table IA.3 in the internet appendix reports the regression results, with observations in the middle three quintiles serving as the reference category. We find that the coefficient on industry expert analyst coverage is significantly negative in both columns, indicating that coverage by industry expert analysts is significantly and negatively related to the probability of a firm recording either very negative or very positive discretionary accruals. This suggests that industry expert analysts restrain both income-increasing and income-decreasing earnings management. In addition, the magnitude of the coefficient on industry expert analysts is significantly larger (p-value: 0.048) in column (2), suggesting that coverage by industry expert analysts reduces the likelihood of aggressive upward earnings management more than aggressive downward earnings management. In terms of economic significance, all else being equal, a one standard deviation increase in industry expert analyst coverage reduces the relative probability of a firm recording very negative vs. moderate discretionary accruals by 6.5%, and it reduces the relative probability of a firm recording very positive vs. moderate discretionary accruals by almost twice as much, 12.1%. Overall, results from the multinomial logit analysis reaffirm our findings from the subsample regressions above. 5.2. Analysts’ monitoring costs Our paper provides ample evidence that industry expertise aids in analysts’ monitoring of firm management. One question related to the broad literature on analyst monitoring is whether analysts bear any costs in monitoring firm management and if so, whether such costs can be severe enough to reduce analysts’ incentive to monitor in the first place. It is entirely conceivable that certain situations could arise that weaken analysts’ monitoring incentive. We explore two such possibilities in this section. First, we examine the impact of investment banking relationships on industry expert analysts’ monitoring incentives. A voluminous literature suggests that analysts affiliated with banks that have an underwriting relationship with a firm provide overly optimistic recommendations for the firm’s stock (e.g., Lin and McNichols (1998), Michaely and Womack (1999)). Such a relationship can also compromise analysts’ willingness to be effective monitors if their monitoring may potentially 26 alienate firm managers and jeopardize existing investment banking relationships. This view is confirmed by Brown et al. (2015), who survey sell-side analysts about the incentives they face. Citing one of the respondent’s answers, “Equity analysts… are very, very reluctant—even after the Spitzer rules—to upset the investment bankers, because the investment bankers bring in so much more profitability…They certainly realize that the success of their company is tied to the performance of this much higher-margin business than the business that they’re part of.” In an attempt to shed some light on this issue, we examine if investment banking pressure reduces the monitoring incentives of industry expert analysts. We classify industry expert analysts as affiliated with firm management if their employers provided equity underwriting or M&A advisory services for the firm within the past 3 years, and unaffiliated otherwise. About 15% of industry expert analysts are classified as affiliated. We treat affiliated and unaffiliated industry expert analyst coverage as separate variables and include them simultaneously in the reexamination of financial reporting, CEO compensation, and CEO turnover policies. For brevity, only the results from the abnormal accruals regression are reported in model 1 of Table 9. Regression results with the other dependent variables used throughout the paper are in the Internet Appendix. We find that only the coverage by unaffiliated industry expert analysts is significantly related to a firm’s abnormal accruals, likelihood of committing financial fraud, excess CEO compensation, and CEO turnoverperformance sensitivity. In other words, the disciplining effects of industry expert analyst coverage on these corporate policies are largely driven by unaffiliated industry expert analysts. Our findings are consistent with the view that analysts whose employers have existing investment banking relationships with the covered firm face higher costs and pressure in monitoring firm managers and thus have weaker incentives to do so.29 This evidence furthers our understanding of the impact of conflicts of interest on sell-side analysts and suggests that investment banking pressure affects more than analyst research such as earnings forecasts and stock recommendations (e.g., Lin and McNichols (1998), Michaely and Womack (1999)). ***Insert Table 9 here*** In our second inquiry, we investigate whether potential connections to firm management through overlapping prior industry work experience reduce analysts’ monitoring effectiveness. To test this conjecture, we obtain employment history of senior officers and directors in coverage firms from Boardex of Management Diagnostic Limited (Boardex). Employment connections are defined as 29 We acknowledge that the lack of statistical significance for the coverage by affiliated industry expert analysts could be due to the smaller number of these analysts. Therefore, readers should exercise caution when interpreting the results from these tests. 27 those arising from overlapping employment between the industry expert analyst and a member of management at the same firm prior to the analyst becoming a sell-side analyst. A significant advantage of pre-existing employment connections is that they are formed ex-ante, ensuring that their formation is not related to coverage firms and eliminating reverse causality concerns (Cohen et al. (2010)). An industry expert analyst is defined as professionally connected to management if she possesses at least one overlapping employment experience with a director or officer at the coverage firm (32% of industry expert analysts meet this criteria) and zero otherwise. Model 2 of Table 9 reestimates the abnormal accruals regression specified in equation (1), except that we now partition industry expert analysts based on the analyst’s professional connections to management. We find that coverage by industry expert analysts is significantly and negatively related to abnormal accruals, irrespective of professional connections between analysts and firm management. Even though the coefficient is larger in magnitude for unconnected industry expert analysts than for connected industry expert analysts, the coefficient difference is not statistically significant. The tenor of the results is similar when we repeat the analysis for other outcome variables. Therefore, it appears that professional connections do not have a material impact on industry expert analyst’s monitoring effectiveness. 5.3. When is industry expert analyst monitoring more important? In this section, we explore cross-sectional variations in the importance of analysts’ external monitoring along the dimensions of firms’ information environment and corporate governance. An important benefit of this analysis is that it can depict a more nuanced picture of the effect of analyst industry expertise by highlighting the settings in which it is especially valuable. If related industry experience enhances analysts’ monitoring effectiveness, then its effect should be more pronounced when monitoring of firm management would be more difficult without a good understanding of the firm’s industry. To test this conjecture, similar to Lee, Strong, and Zhu (2014), we construct a composite information asymmetry rank measure that is equal to the sum of quartile ranks for firm size, age, cash flow and stock return volatilities, accruals quality, analyst coverage, and analyst earnings forecast dispersion in each year. We partition our sample of firm-years into two subsamples at the median of this information asymmetry composite measure, and treat industry expert analyst coverage at high and low information asymmetry firms separately. We reestimate the abnormal accruals regression and present the results in model 3 of Table 9. We find that even though industry expert analyst coverage is significantly and negatively related to abnormal accruals at both high and 28 low information asymmetry firms, the coefficient estimate is significantly larger in magnitude in high information asymmetry firms (p-value for coefficient difference: 0.011). This evidence is consistent with our conjecture that industry expertise is more important for analysts to perform their monitoring function in informationally more opaque firms. We next investigate whether the external monitoring of industry expert analysts is more important in the absence of other strong corporate governance mechanisms limiting managers’ opportunistic behavior. Drawing upon prior literature, we use three variables to capture the strength of a firm’s corporate governance: the percentage of independent directors (Klein (2002) and Xie, Wallace, and DaDalt (2003)), the percentage of inside directors with outside directorships (Masulis and Mobbs (2011)) and the percentage of ownership held by long-term, independent institutional investors (Chen, Harford and Li (2007)). We construct a composite governance measure that is equal to the sum of the quartile ranks of each firm-year based on the above three metrics. We partition our sample of firm-years into two subsamples at the median of this composite governance measure, and treat industry expert analyst coverage at strong and weak governance firms separately. We reestimate the abnormal accruals regression and present the results in model 4 of Table 9. We find that industry expert analyst coverage is significantly and negatively related to abnormal accruals at both strong and weak governance firms. However, the negative relation is more pronounced among weak governance firms (p-value for coefficient difference: 0.045). This provides support for the conjecture that industry expert analyst monitoring is more valuable when other governance devices are lacking.30 5.4. Duration of related industry work experience To add more texture to our findings, we examine whether the length of an analyst’s related industry experience matters. Our conjecture is that longer industry work experience provides analysts with opportunities to develop more industry knowledge and expertise and thus enhances their monitoring effectiveness. To test this conjecture, we partition industry expert analysts based on whether the length of an analyst’s related industry experience is longer or shorter than the sample median and create two separate industry expert analyst coverage variables, Related analyst long 30 In untabulated analyses we perform analogous regressions for other outcome variables (i.e., the ex ante and ex post likelihood of earnings management, CEO compensation and CEO turnover-performance sensitivity) and document similar results for the impact of professional connections, investment banking affiliation, firm information environment as well as corporate governance on industry expert analysts’ monitoring effectiveness. These results are available from the authors upon request. 29 experience and Related analyst short experience. We reestimate the discretionary accrual regression and present the results in model 5 of Table 9. We find that even though both coverage variables have significant and negative coefficients, the magnitude of the coefficient is significantly larger for Related analyst long experience (p-value for coefficient difference: 0.065). We also perform analogous tests for the F-score residuals, restatements, CEO compensation and CEO turnover analyses and find qualitatively similar results. Overall, the evidence is consistent with our hypothesis and suggests that analysts with longer related industry experience play an even more effective monitoring role. 5.5. Robustness tests We conduct an assortment of robustness analyses. We first consider and control for alternative sources of analyst industry experience. Analysts can develop industry expertise simply through covering firms in the same industries over an extended period of time. To ensure that our results are not just an artifact of the effect of analyst experience in covering a particular industry, we explicitly control for the number of years over which an analyst has been issuing forecasts in the coverage firm’s industry. Model 1 of Table 10 shows that analyst experience in covering a firm’s industry has a negative but insignificant effect on earnings management. More importantly, preanalyst related industry experience continues to have a significantly negative and economically important effect on earnings management. ***Insert Table 10 here*** In a recent paper, Dass et al. (2013) focus on directors from related industries defined as executives or directors of firms from related upstream and downstream industries (i.e., DRIs) and provide empirical evidence that DRIs improve the effectiveness of their board’s monitoring. Similarly, Guan, Wong, and Zhang (2015) show that analysts simultaneously following a covered firm’s customers issue more accurate earnings forecasts for the suppliers than analysts who do not. Therefore, to mitigate the potential concern that pre-analyst industry expertise may simply proxy for analyst experience from covering firms in related industries, we include indicators for analysts following firms in related industries as well as analysts following the customers or suppliers of covered firms. The results presented in models 2 and 3 of Table 10 suggest that the effect of preanalyst industry experience on earnings management remains significantly negative. Interestingly, we find that analyst experience from covering firm’s related industries as well as customers or suppliers is also associated with less aggressive earnings management. 30 Wang, Xie, and Zhu (2015) find that audit committees with more independent directors with industry expertise, which they term as industry expert directors (IEDs), reduce firms’ earnings management behavior and probability to engage in fraud. To ensure that industry expert analysts are not simply more likely to follow firms with more IEDs on their audit committees, we control for the percentage of IEDs on a firm’s audit committee (AC_IED_Pct) in the earnings management regression. Consistent with Wang et al. (2015), we find that a greater presence of IEDs on the audit committee is significantly related to lower earnings management by firms. More importantly, the coefficient on industry expert analyst coverage remains negative and highly significant (see model 4).31 Finally, we address two potential selection issues arising from our use of analysts’ LinkedIn profiles. First, LinkedIn was not founded until 2002. Therefore, for analysts providing coverage for a firm prior to 2002 to have a LinkedIn profile, he/she must have remained active in the investment research industry or some other profession long enough (by virtue of either aptitude or age, or both) for LinkedIn to emerge. Second, having a LinkedIn profile is voluntary, so analysts who have a profile may be different from analysts that do not. To shed light on these issues, model 5 of Table 10 repeats the earnings management analysis by using firm-year observations in the post-2002 period. The results are robust. 32 We also examine whether there is any difference in monitoring ability between analysts with and without a LinkedIn profile. Toward that end, we include a firm’s LinkedIn analyst coverage and non-LinkedIn analyst coverage together as explanatory variables in the abnormal accruals regression. Results in model 6 of Table 10 show that both types of coverage have a significantly negative coefficient, but the coefficients are not statistically different from each other. In untabulated analyses, we also compare the ability or quality of these two groups of analysts as proxied by their general analyst experience, firm-specific experience, brokerage firm size, all-star analyst status and portfolio size. We do not find any significant difference. Thus, selection bias stemming from our use of LinkedIn data does not appear to be a major concern for our analysis. 6. Conclusion Jensen and Meckling (1976) first broached the idea of analysts being capable of playing a monitoring role to reduce agency problems. Consistent with their supposition, recent research has 31 All of our other results are robust to the inclusion of these additional controls as well. We do not report them in the paper for brevity, but they are available upon request. 32 Results from the F-score residuals, restatements, CEO excess compensation, and CEO turnover analyses are also robust in the post-2002 subsample. 31 produced a growing body of evidence suggesting that analyst coverage can improve corporate governance, managerial incentives, and firm decision making. In this paper we examine whether analysts’ industry experience obtained from pre-analyst employment affects their effectiveness as external monitors of followed firms. Our analyses of firms’ financial disclosure, executive compensation, and CEO turnover decisions indicate that only financial analysts with related industry experience in their followed firms can play an effective monitoring role. More specifically, we find that coverage by industry expert analysts is associated with reduced earnings management, lower likelihood of committing financial misrepresentation, less CEO excess compensation, and higher performance sensitivity of CEO turnover. In contrast, coverage by other analysts is not significantly related to these corporate policies. We further provide evidence on the channels through which industry expert analysts exert their monitoring influence, namely, by discussing corporate governance issues in research reports, issuing cash flow forecasts along with earnings forecasts, and deviating from management guidance in making earnings forecasts. Overall, our results highlight the importance of analysts’ pre-analyst industry work experience and resultant industry expertise, which suggests that not all analysts are equally effective in providing external monitoring of firms. Our study also points to a few avenues for future research. One implication of our research is that industry expert analysts are likely to have more influence over the behavior and decision making of firm management. Parallel to the analyst monitoring literature, Degeorge et al. (2013) and Derrien and Kecskés (2013) find that analyst coverage and preferences can have a causal effect on firms’ investment and financial policies. Therefore, a natural question is whether corporate policies respond more to the coverage and preferences of analysts with pertinent industry experience and to what extent the previous findings are driven by these industry expert analysts. Akin to sell-side analysts, auditors are third party agents that also play a monitoring role. Industry expertise is thought to be important in this context as well (Ferguson, Francis, and Stokes (2003), Krishnan (2003), Gul, Fung, and Jaggi (2009), McGuire, Omer, and Wang (2012), Francis, Reichelt, and Wang (2005), and Reichelt and Wang (2010)). Similar to the extant analyst literature, auditor industry expertise is typically measured by the amount of experience accumulated by an auditing firm through serving a particular industry. But we know little about the background and skill set of the individuals who actually perform the auditing and what kinds of qualification or experience at the micro level can lead to higher audit quality. While prior studies in this literature are limited to the auditing firm level, a new rule adopted by the Public Company Accounting Oversight 32 Board (PCAOB), which mandates the disclosure of the names of the participants conducting the audit, would allow these questions to be addressed at the individual auditor level. 33 References Bergstresser, Daniel, and Thomas Philippon. "CEO incentives and earnings management." Journal of Financial Economics 80.3 (2006): 511-529. Boni, Leslie, and Kent L. Womack. "Analysts, industries, and price momentum." Journal of Financial and Quantitative Analysis 41.01 (2006): 85-109. Bradley, Daniel, Sinan Gokkaya and Xi Liu, 2015. “Before an Analyst Becomes an Analyst: Does Industry Experience Matter?” The Journal of Finance, forthcoming. Brown, Lawrence D., Andrew C. Call, Michael B. Clement, and Nathan Y. Sharp. "Inside the ĀBlack Boxā of SellϋSide Financial Analysts." Journal of Accounting Research 53.1 (2015): 1-47. Bushee, Brian. J, “The influence of institutional investors on myopic R&D investment behavior.” The Accounting Review, (1998): 305-333. Bushee, Brian, J and Christopher Noe, “Corporate disclosure practices, institutional investors, and stock return volatility.” Journal of Accounting Research 38 (2000): 171-202. Cai, Jie, Jacqueline L. Garner, and Ralph A. Walkling "Electing directors." The Journal of Finance 64.5 (2009): 2389-2421. Chen, Xia, Jarrad Harford, and Kai Li. "Monitoring: Which institutions matter?" Journal of Financial Economics 86.2 (2007): 279-305. Chen, Tao, Jarrad Harford, and Chen Lin. "Do analysts matter for governance? Evidence from natural experiments." Journal of Financial Economics 115.2 (2015): 383-410. Chun, Hyunbae, Jung-Wook Kim, Randall Morck, and Bernard Yeung. "Creative destruction and firm-specific performance heterogeneity." Journal of Financial Economics 89.1 (2008): 109-135. Cohen, Lauren, Andrea Frazzini, and Christopher Malloy. "Sell-side school ties." The Journal of Finance, 65.4 (2010): 1409-1437. Cohen, Lauren, Dong Lou, and Christopher Malloy. “Playing favorites: How firms prevent the revelation of bad news.” (2014), Working paper. Hribar, Paul, and Daniel W. Collins. "Errors in estimating accruals: Implications for empirical research." Journal of Accounting Research 40. 1 (2002): 105-134. Coughlan, Anne T., and Ronald M. Schmidt. "Executive compensation, management turnover, and firm performance: An empirical investigation." Journal of Accounting and Economics 7.1 (1985): 43-66. Custódio, Cláudia, and Daniel Metzger. "How do CEOs matter? The effect of industry expertise on acquisition returns." Review of Financial Studies 26.8 (2013): 2008-2047. 34 Dass, Nishant, Omesh Kini, Vikram Nanda, Bunyamin Onal, and Jun Wang. "Board Expertise: Do Directors from Related Industries Help Bridge the Information Gap?" Review of Financial Studies 27.5 (2013): 1533-1592. Dechow, Patricia M., Weili Ge, Chad R. Larson, and Richard G. Sloan. "Predicting Material Accounting Misstatements." Contemporary Accounting Research 28.1 (2011): 17-82. Dechow, Patricia M., Richard G. Sloan, and Amy P. Sweeney. "Causes and consequences of earnings manipulation: An analysis of firms subject to enforcement actions by the sec." Contemporary Accounting Research 13.1 (1996): 1-36. Degeorge, Francois, François Derrien, Ambrus Kecskes, and Sebastien Michenaud. "Do Analysts' Preferences Affect Corporate Policies?" Available at SSRN 2253619 (2013). Derrien, François, and Ambrus Kecskés. "The real effects of financial shocks: Evidence from exogenous changes in analyst coverage." The Journal of Finance 68.4 (2013): 1407-1440. Dichev, Ilia D., John R. Graham, Campbell R. Harvey, and Shiva Rajgopal. "Earnings quality: Evidence from the field." Journal of Accounting and Economics 56.2 (2013): 1-33. Dopuch, Nicholas, Raj Mashruwala, Chandra Seethanraju and Tzachi Zach. "The impact of a heterogeneous accrual-generating process on empirical accrual models." Journal of Accounting, Auditing & Finance (2011). DuCharme, Larry L., Paul H. Malatesta, and Stephan E. Sefcik. "Earnings management, stock issues, and shareholder lawsuits." Journal of Financial Economics 71.1 (2004): 27-49. Dyck, Alexander, Adair Morse, and Luigi Zingales. "Who blows the whistle on corporate fraud?" The Journal of Finance 65.6 (2010): 2213-2253. Feng, Mei, and Sarah McVay. "Analysts' incentives to overweight management guidance when revising their short-term earnings forecasts." The Accounting Review 85.5 (2010): 1617-1646. Ferguson, Andrew, Jere R. Francis, and Donald J. Stokes. "The effects of firm-wide and office-level industry expertise on audit pricing." The Accounting Review 78.2 (2003): 429-448. Fich, Eliezer M., and Anil Shivdasani. "Are busy boards effective monitors?" The Journal of Finance 61.2 (2006): 689-724. Francis, Jere R., Kenneth Reichelt, and Dechun Wang. "The pricing of national and city-specific reputations for industry expertise in the US audit market." The Accounting Review 80.1 (2005): 113-136. Gong, Guojin, Henock Louis, and Amy X. Sun. "Earnings management, lawsuits, and stock-forstock acquirers’ market performance." Journal of Accounting and Economics 46.1 (2008): 62-77. 35 Graham, John R., Campbell R. Harvey, and Shiva Rajgopal. "The economic implications of corporate financial reporting." Journal of Accounting and Economics 40.1 (2005): 3-73. Guan, Yuyan, MH Franco Wong, and Yue Zhang. "Analyst following along the supply chain." Review of Accounting Studies 20.1 (2015): 210-241. Gul, Ferdinand A., Simon Yu Kit Fung, and Bikki Jaggi. "Earnings quality: Some evidence on the role of auditor tenure and auditors’ industry expertise." Journal of Accounting and Economics 47.3 (2009): 265-287. Hartzell, Jay C., Eli Ofek, and David Yermack. "What's in it for me? CEOs whose firms are acquired." Review of Financial Studies 17.1 (2004): 37-61. He, Jie Jack, and Xuan Tian. "The dark side of analyst coverage: The case of innovation." Journal of Financial Economics 109.3 (2013): 856-878. Healy, Paul M., Amy P. Hutton, and Krishna G. Palepu. "Stock performance and intermediation changes surrounding sustained increases in disclosure." Contemporary Accounting Research 16.3 (1999): 485-520. Hennes, Karen M., Andrew J. Leone, and Brian P. Miller. "The importance of distinguishing errors from irregularities in restatement research: The case of restatements and CEO/CFO turnover." The Accounting Review 83.6 (2008): 1487-1519. Hong, Harrison, and Marcin Kacperczyk. "Competition and bias." The Quarterly Journal of Economics 125.4 (2010): 1683-1725. Huson, Mark R., Paul H. Malatesta, and Robert Parrino. "Managerial succession and firm performance." Journal of Financial Economics 74.2 (2004): 237-275. Irani, Rustom M., and David Oesch. "Monitoring and corporate disclosure: Evidence from a natural experiment." Journal of Financial Economics 109.2 (2013): 398-418. Jensen, Michael C. and William H. Meckling. “Theory of the Firm: Managerial Behavior, Agency Costs and Ownership Structure”. Journal of Financial Economics 3.4 (1976): 305-360. Jenter, Dirk, and Fadi Kanaan. "CEO turnover and relative performance evaluation." The Journal of Finance 70.5 (2015): 2155-2184. Jones, Jennifer J. "Earnings management during import relief investigations." Journal of Accounting Research 29.2 (1991): 193-228. Kadan, Ohad, Leonardo Madureira, Rong Wang, and Tzachi Zach. "Analysts' industry expertise." Journal of Accounting and Economics 54.2 (2012): 95-120. 36 Karpoff, Jonathan M., D. Scott Lee, and Gerald S. Martin. "The consequences to managers for financial misrepresentation." Journal of Financial Economics 88.2 (2008): 193-215. Kelly, Bryan, and Alexander Ljungqvist. "Testing asymmetric-information asset pricing models." Review of Financial Studies 25.5 (2012): 1366-1413. Klein, April. "Audit committee, boar*d of director characteristics, and earnings management." Journal of Accounting and Economics 33.3 (2002): 375-400. Kothari, Sagar P., Andrew J. Leone, and Charles E. Wasley. "Performance matched discretionary accrual measures." Journal of Accounting and Economics 39.1 (2005): 163-197. Krishnan, Gopal V. "Does Big 6 auditor industry expertise constrain earnings management?." Accounting Horizons 17.1 (2003): 1-16 Lee, Edward, Norman Strong, and Zhenmei Judy Zhu. "Did Regulation Fair Disclosure, SOX, and Other Analyst Regulations Reduce Security Mispricing?" Journal of Accounting Research 52.3 (2014): 733-774. Lin, Hsiou-wei, and Maureen F. McNichols. "Underwriting relationships, analysts' earnings forecasts and investment recommendations." Journal of Accounting and Economics 25.1 (1998): 101-127. Liu, Xiaoding, and Jay R. Ritter. "Local underwriter oligopolies and IPO underpricing." Journal of Financial Economics 102.3 (2011): 579-601. Loughran, T., McDonald, B., 2011. When is a Liability not a Liability? Textual Analysis, Dictionaries, and 10-Ks. The Journal of Finance 66.1: 35-65. Louis, Henock, Amy X. Sun, and Oktay Urcan. "Do analysts sacrifice forecast accuracy for informativeness?." Management Science 59.7 (2013): 1688-1708. Masulis, Ronald W., Christian Ruzzier, Sheng Xiao, and Shan Zhao. 2014. “Do Independent Expert Directors Matter?” Working paper. Masulis, Ronald W., and Shawn Mobbs. "Are all inside directors the same? Evidence from the external directorship market." The Journal of Finance 66.3 (2011): 823-872. McGuire, Sean T., Thomas C. Omer, and Dechun Wang. "Tax avoidance: Does tax-specific industry expertise make a difference?." The Accounting Review 87.3 (2012): 975-1003. McInnis, John, and Daniel W. Collins. "The effect of cash flow forecasts on accrual quality and benchmark beating." Journal of Accounting and Economics 51.3 (2011): 219-239. Michaely, Roni, and Kent L. Womack. "Conflict of interest and the credibility of underwriter analyst recommendations." Review of Financial Studies 12.4 (1999): 653-686. 37 Owens, Edward L., Joanna Shuang Wu, and Jerold L. Zimmerman. "Business model shocks and abnormal accrual models." Available at SSRN 2365304 (2015). Parrino, Robert. "CEO turnover and outside succession a cross-sectional analysis." Journal of Financial Economics 46.2 (1997): 165-197. Perry, Susan E., and Thomas H. Williams. "Earnings management preceding management buyout offers." Journal of Accounting and Economics 18.2 (1994): 157-179. Petersen, Mitchell A. "Estimating standard errors in finance panel data sets: Comparing Approaches." Review of Financial Studies 22.1 (2009): 435-480. Reichelt, Kenneth J., and Dechun Wang. "National and office ϋ specific measures of auditor industry expertise and effects on audit quality." Journal of Accounting Research 48.3 (2010): 647-686. Wang, Cong, Fei Xie, and Kuo Zhang. 2015. Expert advice: Industry expertise of financial advisors in mergers and acquisitions. Working paper. Wang, Cong, Fei Xie, and Min Zhu. "Industry expertise of independent directors and board monitoring." Journal of Financial and Quantitative Analysis 50.05 (2015): 929-962. Warfield, Terry D., John J. Wild, and Kenneth L. Wild. "Managerial ownership, accounting choices, and informativeness of earnings." Journal of Accounting and Economics 20.1 (1995): 61-91. Weisbach, Michael S. "Outside directors and CEO turnover." Journal of Financial Economics 20 (1988): 431-460. White, Halbert. "A heteroskedasticity-consistent covariance matrix estimator and a direct test for heteroskedasticity." Econometrica (1980): 817-838. Yu, Fang Frank. "Analyst coverage and earnings management." Journal of Financial Economics 88.2 (2008): 245-271. Xie, Biao, Wallace N. Davidson, and Peter J. DaDalt. "Earnings management and corporate governance: the role of the board and the audit committee." Journal of Corporate Finance 9.3 (2003): 295-316. 38 Appendix A: Data screening and collection Firms Analysts All analysts annual EPS forecasts between 1988 to 2011 from I/B/E/S. 17,775 16,766 Merge with CRSP/COMPUSTAT for stock price data and firm characteristics. 9,684 13,255 Keep the last annual earnings forecast with a horizon between 1 month and 12 month. 6,730 12,584 Merge sample with I/B/E/S recommendation file to get analyst last name, first initial and brokerage firm estimid, which is used to identify the brokerage firm names. Remove analysts without first initial, last name or brokerage estimid, or analyst teams (those analyst names recorded as “research department” or contain two analyst last names). This is the ‘clean’ I/B/E/S sample. 6,713 9,275 Manually search Zoominfo.com for analysts’ names by matching their brokerage firm, last name and first initial. 6,391 4,845 Search for analysts’ employment history on LinkedIn.com, get industry information of pre-analyst public firms from CRSP and pre-analyst private firms from LinkedIn.com, Hoover’s, Thomson Financial’s SDC, the Securities and Exchange Commission’s (SEC) website, business news websites (e.g., Bloomberg and BusinessWeek) or company official websites. 5,542 2,450 Compute number of analysts covering subject firm for each firm year pair. Remove observations with missing value on control variables. 4,267 2,433 39 Appendix B. Variable definitions Variable Total analysts Definition Number of analysts covering subject firm j Related analysts Number of analysts with related industry experience covering subject firm j Unrelated analysts Number of analysts with unrelated pre-analyst industry work experience covering subject firm j Inexperienced analysts Number of analysts without prior industry work experience covering subject firm j Discretionary accruals (DA) Discretionary accruals are estimated from Owen, Wu, and Zimmerman’s (2015) accrual model Market cap Market capitalization of the covered firm (in $billion) by the end of the fiscal year Market-to-book Current market value of equity divided by the book value of equity in the fiscal year prior to the earnings forecast ROA Return on assets, calculated as earnings divided by total assets (#18/#6) Total assets growth Growth rate of assets is calculated by the change of assets scaled by lagged total assets Cash flow volatility The standard deviations of cash flows of a firm in the entire sample period, scaled by lagged total assets External financing The sum of net cash received from equity and debt issuance scaled by lagged total assets Institutional ownership (dedicated) The percentage of institutional holding by dedicated institutional investors. Dedicated institutional investors are classified based on Bushee (1998) and provided on his website. Institutional ownership (non-dedicated) The percentage of institutional holding by non-dedicated institutional investors calculated as total institutional investor holdings less dedicated institutional investor holdings. Experience as analyst Average number of years that analyst’s i appeared in I/B/E/S Experience with firm Average number of years since analyst’s i first earnings forecast for the covered firm j Analysts from top brokers Number of analysts working at a top decile brokerage house 40 that cover subject firm j All-star analysts Number of All-star analysts covering subject firm j Analyst portfolio size Average number of firms followed by analysts that cover the subject firm High restate exposure An indicator for industry groups with above sample median restatement frequency, which is computed as the percentage of firms in each industry that were associated with an irregularity restatement over a 10-year rolling window Tobin’s Q Market-to-book ratio of assets, where the market value of assets is computed as the book value of assets plus the market value of equity, minus the book value of equity ((#6+#199×#25-#60#74)/#6) Leverage Long-term plus short-term debt, scaled by total assets ((#9+#34)/#6) R&D/sale R&D expenditures scaled by total sales (#46/#12) Capx/sale Capital expenditures scaled by total sales (#128/#12) Xad/sale Advertising expenditures scaled by total sales (#45/#12) Abnormal ROA ROA-adjusted by industry median ROA Abnormal stock rets A firm’s annual stock returns minus its corresponding GICS industry annual stock returns CEO tenure Number of years since the CEO has been employed with the company CEO total/excess compensation CEO total compensation includes salary, bonus, stock and option grants, and long-term incentive plans from Standard & Poor’s ExecuComp database. To compute CEO excess compensation, the logarithmic transformation of total CEO compensation is regressed against a firm’s market capitalization, buy-and-hold abnormal returns over CRSP value-weighted returns, stock return volatility, as well as industry (based on twodigit SIC codes) and year fixed effects. CEO excess compensation is defined as the residuals from this estimation. Firm age Number of years since the firm’s first appearance in CRSP Past return CRSP VW-index adjusted buy-and hold abnormal returns (BHARs) over the prior 250 trading days 41 Restatement Indicator variable=1 if firm has at least one restatement in the fiscal year, 0 otherwise Abnormal ROA Firm performance by the industry-adjusted return on assets (ROA) Sales volatility Standard deviation of sales of a firm in the entire sample period, scaled by lagged assets CEO/Chairman Indicator for whether CEO is also chairman of the board Abs (Devaf) Absolute value of forecasting analyst i’s deviation from firm j’s preceding preannounced earnings for the forecasted year t scaled by stock price of firm j on the day before the preannouncement Abret Market-adjusted returns starting two days after the preannouncement date and ending one day before the analyst forecast date Loss A binary variable taking the value one if previous earnings are negative and zero otherwise Evoltil The standard deviation of earnings (as a percentage of price at the beginning of the fiscal year) over past 5 years Fvoltil Price adjusted forecast error volatility over past 5 years Che Change in earnings from year t-1 to year t Pae Management guidance minus reported earnings, deflated by price on the day prior to the earnings preannouncement Appae Average percentage management earnings forecast error over the past two years. Abretm3 Market-adjusted returns over the three months ending the month prior to the preannouncement month Gap Number of days between management forecasts and analyst forecasts. Related analyst with/without prof connections Number of industry expert analysts with/without professional connections to the covering subject firm j’s management Related analyst affiliated/unaffiliated Number of industry expert analysts working for brokers that have/haven’t provided equity underwriting or M&A advisory services for the firm within the past 3 years to the covering subject firm j’s 42 Related analyst high/low info asymmetry Number of analysts covering subject firm j with above/below sample median composite information asymmetry measure. The composite information asymmetry measure is constructed as the sum of quartile ranks based on a firm’s size, age, cash flow and stock return volatilities, accruals quality, analyst coverage, and earnings forecast dispersion. Related analyst weak/strong governance Number of analysts covering subject firm j with above/below sample median composite corporate governance measure. The composite governance measure is constructed as the sum of quartile ranks based on the percentage of independent directors, the percentage of inside directors with outside directorships, and the percentage of equity ownership held by long-term independent institutional investors Industry covering experience Average number of years that covering analysts has issued forecasts in the coverage firm’s 2-digit industry Covering related industry Indicator variable=1 if at least one of analysts also covers the coverage firm’s related industry, 0 otherwise Covering customer/supplier Indicator variable=1 if at least one of analysts also covers the coverage firm’s supplier or customer, 0 otherwise AC_IED_Pct The number of independent directors on the audit committee with industry expertise divided by the total number of independent directors on the audit committee Total LinkedIn analysts Number of LinkedIn analysts covering subject firm j Total non-LinkedIn analysts Number of non-LinkedIn analysts covering subject firm j 43 Appendix C. List of keywords used for textual analysis of analyst research reports We read 1000 randomly selected analyst reports to look for every bigram word combination that analysts use to describe governance and compiled a robust and extensive keyword listed below: Financial restatement: restate results restating results restatement of financials restatement risk material weaknesses risk of restatement accounting restatement financial misconduct financial fraud materially false materially misleading restatement announcement restatement numbers accounting shenanigans restatement of financial results restatement filings restatement activities Corporate governance: governance rules governance committee governance risk governance culture governance issues governance overhang governance record governance practices governance policies governance snapshot governance concerns governance standards governance questions governance procedures governance provisions governance problem weak governance strong governance good governance bad governance governance features governance principles governance concern governance changes governance shortcomings governance discount governance quality governance measures governance rating governance fear CEO compensation: CEO salary CEO option CEO stock CEO cash CEO bonus CEO stock option CEO restricted stock CEO other compensation CEO annual compensation CEO total compensation CEO turnover: CEO change CEO is fired CEO is laid off CEO is forced out CEO job turnover CEO has resigned CEO resigned CEO left CEO has left Shareholder proposal: activist proposal shareholder proposal Shareholder activism: activist activist hedge fund proxy fight 44 Table 1. Descriptive statistics This table reports summary statistics of our sample. Panel A provides the top five 4-digit GICS industries from which analysts obtain pre-analyst work experience as well as the top 5 industries with the most industry expert analysts. Panel B is based on the earnings management sample, which consists of 24,918 firm-year observations from 1988 to 2011. The number of observations used in different analyses varies. See Appendix B for a detailed description of these variables. Panel A: Top five four-digit GICS industries (pre-analyst work experience and industry expert analysts) Industry Software & Services Diversified Financials Commercial Services & Supplies Banks Media Pre-analyst work experience 12.74% 11.95 9.83 Industry Software & Services Diversified Financials Health Care Equipment & Services Retailing Technology Hardware & Equipment 8.18 7.63 Industry expert analysts 17.90% 8.08 6.74 6.72 6.31 Panel B: Variable descriptive statistics Dependent Variables Absolute value of discretionary accruals (Abs_DA) F-score (residuals) Restated due to irregularity CEO compensation ($ mil) Forced CEO turnover N Mean 1st quartile Median 3rd quartile Std Dev 24,918 23,233 19,772 9,428 6,623 0.047 -0.007 0.017 3.434 0.027 0.013 -0.541 0.000 1.211 0.000 0.029 -0.189 0.000 2.373 0.000 0.059 0.315 0.000 4.412 0.000 0.058 0.760 0.129 3.782 0.161 Firm Characteristics Market cap ($ bil) Market to book ROA Total assets growth Cash flow volatility External financing Institutional ownership (dedicated) Institutional ownership (non-dedicated) 24,918 24,918 24,918 24,918 24,918 24,918 24,918 24,918 4.618 3.175 0.034 0.229 0.222 0.026 0.083 0.553 0.355 1.492 0.015 0.009 0.062 -0.047 0.000 0.363 0.944 2.310 0.052 0.097 0.101 0.000 0.043 0.553 2.990 3.753 0.095 0.244 0.213 0.050 0.125 0.754 12.927 4.299 0.137 0.637 0.395 0.158 0.120 0.252 Analyst characteristics Total analysts Related analysts Unrelated analysts Inexperienced analysts Experience as analyst Experience with firm Analyst portfolio size Analysts from top brokers All-star analysts 24,918 24,918 24,918 24,918 24,918 24,918 24,918 24,918 24,918 3.972 0.896 1.698 1.377 6.312 2.454 12.561 0.565 0.131 1.000 0.000 0.000 0.000 4.000 1.000 9.333 0.333 0.000 3.000 0.000 1.000 1.000 6.000 2.000 12.000 0.600 0.000 5.000 1.000 2.000 2.000 8.250 3.500 14.545 0.900 0.200 3.524 1.499 1.987 1.387 3.362 2.137 5.724 0.350 0.244 45 Table 2. Industry expert analyst coverage and earnings management Panel A presents baseline regressions of earnings management. The sample consists of 24,918 firm-year observations from 1988 to 2011. The dependent variable is the absolute value of discretionary accruals (Abs_DA) adjusted for business model shocks using Owens, Wu and Zimmerman (2015) model. See Appendix B for a detailed description of variables. Firm and year-fixed effects are included. Coefficient values are reported as percentages and robust t-statistics based on standard errors adjusted for heteroskedasticity (White (1980)) and firm clustering (Petersen (2009)) are in parentheses. Panel B reports the effect of analyst departures from broker mergers and closures on changes in coverage firms’ earnings management. The first column provides the cross-sectional means of the mean treatment difference, mean control difference and differences-in-difference (DID) for the full sample of treatment firms affected by the loss of analyst coverage (Lost Analyst). Lost Related Experienced Analyst (Lost Other Analyst) provides results for the sample of coverage firms with terminations of coverage by related industry experienced analysts (other analysts). The last column reports the differences between these two groups. Control firms are required to be in the same size and bookto-market quintiles in the month of June with the treatment firms preceding the broker events and are covered by at least one analyst. We retain the control firm with the smallest number of difference in the number of analysts relative to that of treatment firms. *, **, and *** indicate statistical significance at the 10%, 5%, and 1%, respectively. Panel A. Baseline OLS regressions Total analysts Model 1 -0.078*** (-3.761) Model 2 Related analysts Model 3 -0.210*** (-5.074) -0.025 (-0.719) -0.037 (-0.875) Unrelated analysts Inexperienced analysts Total analysts (Residual) -0.125*** (-5.760) Related analysts (Residual) Unrelated analysts (Residual) Inexperienced analysts (Residual) Market cap Market-to-book ROA Total assets growth Cash flow volatility External financing Institutional ownership (dedicated) Institutional ownership (non-dedicated) Experience as analyst Experience with firm Model 4 -0.006 (-1.113) 0.067*** (5.505) 0.026 (0.044) -0.015 (-0.123) -0.004 (-0.019) 0.845* (1.746) -2.522*** (-4.228) -2.576*** (-7.879) 0.022 (1.138) -0.083*** -0.005 (-0.998) 0.066*** (5.453) 0.053 (0.088) -0.017 (-0.139) -0.014 (-0.062) 0.844* (1.746) -2.489*** (-4.167) -2.507*** (-7.579) 0.017 (0.895) -0.078*** 46 -0.016*** (-2.987) 0.063*** (5.151) 0.020 (0.034) -0.037 (-0.304) -0.044 (-0.194) 0.944* (1.954) -2.713*** (-4.512) -2.958*** (-9.221) 0.020 (1.072) -0.094*** -0.183*** (-4.258) -0.055 (-1.505) -0.038 (-0.941) -0.015*** (-2.742) 0.065*** (5.347) 0.078 (0.130) -0.043 (-0.354) 0.008 (0.034) 0.946* (1.964) -2.698*** (-4.482) -2.953*** (-9.213) 0.015 (0.768) -0.087*** Analysts from top brokers All-star analysts Analyst portfolio size Adj-R2 N Fixed Effects (-3.078) -0.356** (-2.395) -0.056 (1.102) 0.010 (1.102) 28.36% 24,918 Y (-2.884) -0.355** (-2.388) -0.065 (1.176) 0.011 (1.176) 28.40% 24,918 Y (-3.466) -0.365** (-2.452) -0.050 (1.298) 0.012 (1.298) 28.49% 24,918 Y (-3.197) -0.355** (-2.389) -0.027 (1.347) 0.012 (1.347) 28.42% 24,918 Y Panel B: Exogenous shocks from brokerage mergers/closures Mean ΔAbs_DA (treatment firms) Mean ΔAbs_DA (control firms) Mean DID (treatment – control) (1) Lost Analyst 0.269* (1.880) -0.126 (-1.050) 0.395** (2.150) (2) Lost Related Analyst 0.935*** (2.890) -0.274 (-0.940) 1.208*** (3.080) 47 (3) Lost Other Analyst 0.133 (0.840) -0.095 (-0.720) 0.229 (1.110) (4) Difference: (2)-(3) 0.802** (2.220) -0.178 (-0.560) 0.980** (2.210) Table 3. Industry expert analyst coverage and F-score residuals Panel A presents OLS regression results for F-score residuals from 1988 to 2011 and the change in F-score residuals around the loss of analyst coverage stemming from brokerage mergers/closures in panel B. See Appendix B for a detailed description of variables. Firm and year-fixed effects are included. Coefficient values are reported as percentages and robust t-statistics based on standard errors adjusted for heteroskedasticity (White (1980)) and firm clustering (Petersen (2009)) are in parentheses. The first column provides the cross-sectional means of the mean treatment difference, mean control difference and differences-in-difference (DID) for the full sample of treatment firms affected by the loss of analyst coverage (Lost Analyst). Lost Related Analyst (Lost Other Analyst) provides results for sample of coverage firms with terminations of coverage by related industry experienced analysts (other analysts). The last column reports the differences between these two groups. Control firms are required to be in the same size and book-to-market quintiles in the month of June with the treatment firms preceding the broker events and are covered by at least one analyst. We retain the control firm with the smallest number of difference in the number of analysts relative to that of treatment firms. *, **, and *** indicate statistical significance at the 10%, 5%, and 1%, respectively. Panel. A. Baseline OLS regressions Model 1 -0.513** (-2.137) Total Analysts (Residual) Related analysts (Residual) Unrelated analysts (Residual) Inexperienced analysts (Residual) Market cap 0.139* (1.651) 0.079 (0.703) -4.907 (-0.463) 36.190*** (5.090) -1.670 (-0.631) 0.784 (0.051) 0.985 (0.180) 4.820 (1.383) 0.409** (2.078) -1.309*** (-4.568) 0.259 (0.177) 1.614 (0.806) 0.225** (2.219) 61.85% 23,233 Y Market-to-book ROA Total assets growth Cash flow volatility External financing Institutional ownership (dedicated) Institutional ownership (non-dedicated) Experience as analyst Experience with firm Analysts from top brokers All-star analysts Analyst portfolio size Adj-R2 N Fixed Effects 48 Model 2 -1.800*** (-3.659) -0.601 (-1.548) -0.531 (-1.155) 0.128 (1.512) 0.055 (0.491) -4.635 (-0.438) 36.048*** (5.076) -2.240 (-0.846) 1.169 (0.076) 0.695 (0.127) 4.846 (1.391) 0.375* (1.850) -1.334*** (-4.651) 0.242 (0.164) 1.503 (0.750) 0.233** (2.304) 61.89% 23,233 Y Panel. B. Exogenous shocks from brokerage mergers/closures (1) (2) Lost Analyst Lost Related Analyst Mean ΔF-score residual 0.030* 0.097** (treatment firms) (1.700) (2.540) Mean ΔF-score residual -0.013 -0.058* (control firms) (-0.950) (-1.670) Mean DID 0.043* 0.155*** (treatment – control) (1.890) (2.750) 49 (3) Lost Other Analyst 0.019 (0.970) -0.006 (-0.390) 0.025 (1.000) (4) Difference (2)-(3) 0.078* (1.820) -0.052 (-1.370) 0.130** (2.110) Table 4. Industry expert analyst coverage and restatements This table presents the results from logit regressions of the probability of firms committing intentional financial misreporting that leads to earnings restatements from 1997 to 2011. The dependent variable is equal to one for firmyears whose earnings are restated due to accounting irregularities and zero otherwise. See Appendix B for a detailed description of variables. Coefficient values are reported as percentages and robust t-statistics based on standard errors adjusted for heteroskedasticity (White (1980)) and firm clustering (Petersen (2009)) are in parentheses. *, **, and *** indicate statistical significance at the 10%, 5%, and 1%, respectively. Model 1 -6.040* (-1.787) Total analysts (Residual) Related analysts (Residual) Unrelated analysts (Residual) Inexperienced analysts (Residual) Market cap 0.758 (1.289) -1.910 (-1.130) -118.620** (-2.455) 17.420*** (2.801) -106.190* (-1.900) 68.330* (1.666) 27.500 (0.319) 243.430*** (4.709) -1.490 (-0.487) 0.899 (0.195) 1.810 (0.077) 28.230 (0.930) 1.440 (0.906) 33.020** (2.032) 2.04% 19,772 Y Market-to-book ROA Total assets growth Cash flow volatility External financing Institutional ownership (dedicated) Institutional ownership (non-dedicated) Experience as analyst Experience with firm Analysts from top brokers All-star analysts Analyst portfolio size High restate exposure Pseudo-R2 N Fixed Effects 50 Model 2 -25.310*** (-3.335) 3.400 (0.615) -9.030 (-1.265) 0.540 (0.912) -1.940 (-1.128) -119.880** (-2.507) 17.430*** (2.762) -107.370* (-1.888) 70.120* (1.711) 25.790 (0.300) 240.470*** (4.642) -1.360 (-0.453) 0.900 (0.195) 2.530 (0.109) 25.880 (0.843) 1.610 (1.045) 26.730* (1.647) 2.13% 19,772 Y Table 5. Industry expert analyst coverage and excess CEO compensation This table presents regression results for CEO excess compensation (Panel A) and the change in CEO excess compensation around the loss of analyst coverage due to shocks stemming from brokerage mergers/closures (Panel B). The dependent variable is the logarithmic transformation of excess CEO compensation. See Appendix B for a detailed description of variables. Firm and year-fixed effects are included. Coefficient values are reported as percentages and robust t-statistics based on standard errors adjusted for heteroskedasticity (White (1980)) and firm clustering (Petersen (2009)) are in parentheses*, **, and *** indicate statistical significance at the 10%, 5%, and 1%, respectively. Panel A. Baseline OLS regressions Model 1 Model 2 Total analysts (Residual) -0.968** (-2.033) Related analysts (Residual) -3.190*** (-3.278) Unrelated analysts (Residual) -0.240 (-0.329) Inexperienced analysts (Residual) -0.439 (-0.431) Market cap 14.672*** 14.598*** (6.700) (6.714) Tobin’s Q 1.718 1.617 (1.392) (1.334) Leverage -0.851 -0.834 (-0.061) (-0.059) R&D/sale -1.597*** -1.595** (-2.632) (-2.593) Capx/sale 15.050*** 15.133*** (3.114) (3.092) Xad/sale 84.984 87.571 (1.165) (1.204) Abnormal ROA 78.000*** 78.897*** (5.138) (5.169) Abnormal stock rets -1.553 -1.588 (-0.841) (-0.861) CEO tenure -3.369*** -3.387*** (-7.939) (-8.004) Firm age 4.455*** 4.483*** (9.584) (9.785) Institutional ownership (dedicated) 5.807 5.232 (0.428) (0.385) Institutional ownership (non-dedicated) 4.991 5.135 (0.573) (0.591) Experience as analyst 0.189 0.116 (0.480) (0.289) Experience with firm 0.331 0.355 (0.493) (0.526) Analysts from top brokers -1.877 -2.096 (-0.567) (-0.633) All-star analysts -1.038 -0.892 (-0.211) (-0.181) Analyst portfolio size -0.040 -0.031 (-0.165) (-0.130) Adj-R2 61.52% 61.57% N 9,428 9,428 Fixed Effects Y Y 51 Panel B. Exogenous shocks from brokerage mergers/closures (1) (2) Lost Analyst Lost Related Analyst Mean ΔCEO excess comp 0.056* 0.178** (treatment firms) (1.720) (2.320) Mean ΔCEO excess comp -0.021 -0.050 (control firms) (-0.740) (-0.750) Mean DID 0.077* 0.228** (treatment – control) (1.960) (2.650) 52 (3) Lost Other Analyst 0.034 (0.970) -0.016 (-0.510) 0.051 (1.160) (4) Difference (2)-(3) 0.143* (1.690) -0.034 (-0.450) 0.177* (1.840) Table 6. Industry expert analyst coverage and CEO turnover-performance sensitivity This table presents logit regression results for forced CEO turnovers. The dependent variable is an indicator that equals 1 if there is a forced CEO turnover in year t and 0 otherwise. See Appendix B for a detailed description of variables. Coefficient values are reported as percentages and robust t-statistics based on standard errors adjusted for heteroskedasticity (White (1980)) and firm clustering (Petersen (2009)) are in parentheses. *, **, and *** indicate statistical significance at the 10%, 5%, and 1%, respectively. Model 1 Model 2 Model 3 Abnormal ROA * Total analysts (Residual) -12.210 (-0.625) Abnormal ROA * Related analysts (Residual) -165.930*** -165.390*** (-2.997) (-2.999) Abnormal ROA * Unrelated analysts (Residual) 34.300 31.350 (1.096) (0.997) Abnormal ROA * Inexperienced analysts (Residual) 28.700 26.920 (0.511) (0.475) Abnormal ROA -155.960** -219.390*** -223.910*** (-2.174) (-3.148) (-3.161) Total analysts (Residual) 0.774 (0.269) Related analysts (Residual) 7.610 7.720 (1.037) (1.047) Unrelated analysts (Residual) -3.420 -1.950 (-0.597) (-0.336) Inexperienced analysts (Residual) 1.370 0.082 (0.225) (0.013) Market cap 0.008 -0.081 -0.187 (0.014) (-0.145) (-0.321) Firm age -0.024 -0.048 -0.209 (-0.049) (-0.097) (-0.398) Leverage 13.070 13.050 13.790 (0.199) (0.200) (0.207) Tobin’s Q -5.910 -4.700 -3.740 (-0.682) (-0.550) (-0.438) Sales Volatility 34.500*** 34.790*** 34.130*** (4.040) (4.017) (3.826) CEO/Chairman -11.650 -12.040 -13.210 (-0.718) (-0.737) (-0.807) Institutional ownership (dedicated) -25.910 -37.310 -49.660 (-0.234) (-0.334) (-0.443) Institutional ownership (non-dedicated) -48.790 -48.190 -52.070 (-0.953) (-0.936) (-0.995) Abnormal stock rets -122.290*** -121.290*** -123.290*** (-5.333) (-5.283) (-5.260) Experience as analyst 2.630 (0.807) Experience with firm 5.070 (1.155) Analysts from top brokers -1.640 (-0.055) All-star analysts -8.080 (-0.200) Analyst portfolio size 1.380 (0.738) Pseudo-R2 1.98% 2.09% 2.18% N 6,623 6,623 6,623 Fixed Effects Y Y Y 53 Table 7. Industry expert analyst monitoring mechanisms This table presents the results for monitoring mechanisms by sell-side analysts. Panel A reports logit regressions of the probability of analysts discussing corporate governance related issues on coverage firms in their research reports (Model 1), issuing cash flow forecasts on coverage firms (Model 2), and OLS regression of the absolute value of analyst earnings forecast deviation from management guidance (Model 3). Panel B reports the impact of these mechanism on the earnings management of coverage firms. See Appendix B for a detailed description of variables. Coefficient values are reported as percentages and robust t-statistics based on standard errors adjusted for heteroskedasticity (White (1980)) and firm clustering (Petersen (2009)) are in parentheses. *, **, and *** indicate statistical significance at the 10%, 5%, and 1%, respectively. Model 1 Model 2 Model 3 Panel A: Monitoring mechanisms Gov Report (Logit) CF Forecast (Logit) Deviation (OLS) Related analyst (A) 27.200*** 31.850*** 0.165** (4.474) (5.617) (2.244) Unrelated analyst (A) 2.830 -4.270 -0.085 (0.528) (-0.800) (-1.377) Abs_DA 1.775*** (2.655) Market cap 12.030** 0.257** -0.051 (2.486) (2.520) (-0.664) Market-to-book -0.098 -0.080 0.002 (-0.817) (-0.213) (0.186) ROA 0.160 64.470*** -2.174*** (0.404) (4.007) (-3.192) Total assets growth 7.790 -5.510* -0.151 (0.402) (-1.813) (-1.029) Cash flow volatility -2.140 1.360* 0.003 (-0.389) (1.735) (0.007) External financing -9.040 31.660** 0.365 (-0.594) (2.160) (1.021) Institutional ownership (dedicated) -51.680*** 39.250 -0.266 (-2.862) (1.229) (-0.548) Institutional ownership (non-dedicated) 30.290 42.280*** -0.904*** (1.641) (3.260) (-2.810) Experience as analyst (A) -8.830 -1.680*** -0.009 (-0.837) (-3.225) (-1.540) Experience with firm (A) 1.770*** -1.210 0.008 (3.620) (-1.131) (0.824) Analysts from top brokers (A) 1.070 67.890*** 0.213*** (1.285) (17.015) (3.861) All-star analysts (A) 34.710*** 23.480*** 0.197** (3.852) (4.083) (2.546) Analyst portfolio size (A) 5.110*** -2.090*** 0.000 (12.649) (-6.764) (0.031) Abret -1.081*** (-7.762) Loss 0.403*** (3.163) Evoltil 0.022*** (3.561) Fvoltil 77.991*** (4.714) Che -0.090 (-0.157) Pae 6.102*** (17.776) Appae 4.920*** (6.805) 54 Abretm3 Gap 13.43% 106,287 Pseudo-R2/Adj-R2 N Fixed Effect Y Panel B: Impact on earnings management (Abs_DA) Related Analysts with Gov Report (Residual) Model 1 -0.360*** (-3.885) -0.153*** (-3.466) Related Analysts without Gov Report (Residual) Related Analysts with CF Forecast (Residual) 15.68% 94,647 Y Model 2 -0.477*** (-2.872) 0.002*** (9.458) 19.61% 23,675 Y Model 3 -0.340*** (-3.755) -0.144*** (-3.087) Related Analysts without CF Forecast (Residual) High Deviation Related Analysts (Residual) Low Deviation Related Analysts (Residual) Adj-R2 N Fixed Effect 28.41% 24,918 Y 55 28.44% 24,918 Y -0.327*** (-5.210) -0.156*** (-3.462) 28.43% 24,918 Y Table 8: Income-increasing vs income-decreasing earnings management This table presents OLS regression results of earnings management for the subsamples of income-increasing and income-decreasing discretionary accruals. The full sample consists of 24,918 firm-year observations from 1988 to 2011. The dependent variable is the absolute value of discretionary accruals (Abs_DA) in each subsample regression. *, **, and *** indicate statistical significance at the 10%, 5%, and 1%, respectively. Income-increasing DA -0.238*** (-3.582) -0.071 (-1.210) -0.043 (-0.672) -0.012 (-1.419) 0.045** (2.562) 1.822* (1.828) -0.243 (-1.236) -0.262 (-0.862) 1.379 (1.519) -1.962* (-1.823) -2.832*** (-5.693) -0.015 (-0.455) -0.085* (-1.962) -0.392* (-1.650) -0.197 (-0.668) 0.033** (2.196) 26.21% 12,015 Y Related analysts (Residual) Unrelated analysts (Residual) Inexperienced analysts (Residual) Market cap Market-to-book ROA Total assets growth Cash flow volatility External financing Institutional ownership (dedicated) Institutional ownership (non-dedicated) Experience as analyst Experience with firm Analysts from top brokers All-star analysts Analyst Portfolio size Adj-R2 N Fixed Effects 56 Income-decreasing DA -0.132** (-2.511) -0.039 (-0.823) -0.007 (-0.142) -0.018** (-2.387) 0.072*** (4.561) -1.749** (-2.244) 0.224 (1.006) 0.189 (0.605) 1.592** (2.342) -3.111*** (-4.570) -3.008*** (-7.598) 0.029 (1.140) -0.070** (-2.020) -0.260 (-1.332) 0.098 (0.424) 0.003 (0.273) 39.50% 12,903 Y Table 9. Cross-sectional variation of analyst industry expertise on earnings management This table reports results from cross-sectional regressions of earnings management. The full sample consists of 24,918 firm-year observations from 1988 to 2011. The dependent variable is the absolute value of discretionary accruals (Abs_DA) adjusted for business model shocks using Owens, Wu and Zimmerman (2015) model. See Appendix B for a detailed description of variables. Firm and year-fixed effects are included. Coefficient values are reported as percentages and robust t-statistics based on standard errors adjusted for heteroskedasticity (White (1980)) and firm clustering (Petersen (2009)) are in parentheses. *, **, and *** indicate statistical significance at the 10%, 5%, and 1%, respectively. Related analyst affiliated (residual) Related analyst unaffiliated (residual) Model 1 -0.106 (-1.376) -0.177*** (-4.129) Model 2 Related analyst with prof connections(residual) Model 3 Model 4 -0.148** (-2.172) -0.204*** (-3.900) Related analyst without prof connections(residual) Related analyst high info asymmetry (residual) -0.269*** (-4.717) -0.122** (-2.578) Related analyst low info asymmetry (residual) Related analyst weak governance(residual) -0.313*** (-3.947) -0.139*** (-2.945) Related analyst strong governance(residual) Related analyst long experience (residual) Related analyst short experience (residual) Unrelated analysts (Residual) Inexperienced analysts (Residual) Market cap Market-to-book ROA Total assets growth Cash flow volatility External financing Institutional ownership (dedicated) Institutional ownership (non-dedicated) Experience as analyst Experience with firm Analysts from top brokers All-star analysts Analyst portfolio size Adj-R2 N Fixed Effects Model 5 -0.056 (-1.530) -0.038 (-0.960) -0.015*** (-2.728) 0.065*** (5.350) 0.075 (0.126) -0.041 (-0.335) 0.014 (0.063) 0.941* (1.951) -2.731*** (-4.534) -2.972*** (-9.265) 0.015 (0.794) -0.086*** ++(-3.175) -0.355** (-2.389) -0.029 (-0.154) 0.012 (1.343) 28.41% 24,918 Y -0.056 (-1.529) -0.037 (-0.914) -0.015*** (-2.733) 0.065*** (5.353) 0.081 (0.135) -0.044 (-0.361) 0.009 (0.041) 0.946* (1.963) -2.692*** (-4.473) -2.955*** (-9.220) 0.014 (0.749) -0.087*** (-3.183) -0.356** (-2.394) -0.026 (-0.137) 0.012 (1.347) 28.42% 24,918 Y 57 -0.054 (-1.461) -0.036 (-0.901) -0.015*** (-2.704) 0.065*** (5.344) 0.079 (0.132) -0.048 (-0.391) 0.014 (0.062) 0.961** (1.987) -2.708*** (-4.504) -2.963*** (-9.256) 0.014 (0.746) -0.087*** (-3.194) -0.357** (-2.402) -0.023 (-0.121) 0.012 (1.347) 28.45% 24,918 Y -0.054 (-1.479) -0.036 (-0.911) -0.015*** (-2.711) 0.065*** (5.347) 0.091 (0.152) -0.051 (-0.416) 0.013 (0.056) 0.978** (2.021) -2.725*** (-4.518) -2.992*** (-9.340) 0.014 (0.712) -0.087*** (-3.196) -0.351** (-2.362) -0.027 (-0.146) 0.012 (1.368) 28.47% 24,918 Y -0.213*** (-4.305) -0.124** (-2.478) -0.056 (-1.538) -0.041 (-1.033) -0.016*** (-2.895) 0.065*** (5.371) 0.118 (0.198) -0.042 (-0.342) -0.002 (-0.010) 0.962** (1.993) -2.611*** (-4.346) -2.901*** (-9.048) 0.019 (0.984) -0.091*** (-3.323) -0.388*** (-2.607) -0.023 (-0.123) 0.012 (1.378) 28.42% 24,918 Y Table 10. Robustness analysis This table presents earnings management regressions with additional controls (Model 1-4), post-2002 subsample period (model 5), and for LinkedIn vs. non-LinkedIn analyst residual counts (model 6). The baseline sample consists of 24,918 firm-year observations from 1988 to 2011. The dependent variable is the absolute value of discretionary accruals (Abs_DA) adjusted for business model shocks using Owens, Wu and Zimmerman (2015) model. See Appendix B for a detailed description of variables. Firm and year-fixed effects are included. Coefficient values are reported as percentages and robust t-statistics based on standard errors adjusted for heteroskedasticity (White (1980)) and firm clustering (Petersen (2009)) are in parentheses. *, **, and *** indicate statistical significance at the 10%, 5%, and 1%, respectively. Related analysts (Residual) Unrelated analysts (Residual) Inexperienced analysts (Residual) Model 1 -0.183*** (-4.260) -0.055 (-1.506) -0.038 (-0.941) Model 2 Model 3 -0.181*** -0.183*** (-4.193) (-4.258) -0.052 -0.054 (-1.402) (-1.473) -0.033 -0.036 (-0.829) (-0.890) Model 4 -0.180*** (-4.187) -0.053 (-1.445) -0.035 (-0.882) -0.015*** (-2.740) 0.065*** (5.346) 0.077 (0.129) -0.043 (-0.352) 0.007 (0.031) 0.946* (1.964) -2.699*** (-4.480) -2.950*** (-9.215) 0.018 (0.722) -0.085*** (-3.050) -0.355** (-2.379) -0.028 (-0.148) 0.012 (1.350) -0.006 (-0.224) -0.015*** (-2.723) 0.065*** (5.385) 0.085 (0.142) -0.043 (-0.353) 0.006 (0.028) 0.950** (1.971) -2.678*** (-4.448) -2.928*** (-9.118) 0.015 (0.782) -0.087*** (-3.204) -0.357** (-2.398) -0.028 (-0.152) 0.014 (1.515) -0.014** (-2.568) 0.065*** (5.342) 0.111 (0.186) -0.044 (-0.357) 0.011 (0.048) 0.936* (1.939) -2.671*** (-4.439) -2.890*** (-8.999) 0.014 (0.745) -0.083*** (-3.030) -0.354** (-2.381) -0.033 (-0.174) 0.012 (1.359) Model 5 Model 6 -0.220*** (-3.974) 0.006 (0.158) -0.071 (-1.479) -0.103*** (-4.852) -0.093*** (-8.016) -0.008 -0.038*** (-1.041) (-5.761) 0.042*** 0.061*** (2.843) (4.970) -0.347 0.021 (-0.406) (0.035) -0.251 0.004 (-0.940) (0.033) -0.889 -0.265 (-1.433) (-1.139) 0.849 1.010** (1.202) (2.085) -0.644 -2.725*** (-0.876) (-4.552) -0.747* -3.012*** (-1.708) (-9.425) 0.025 0.018 (0.989) (0.958) -0.145*** -0.081*** (-3.728) (-2.990) 0.266 -0.359** (1.238) (-2.423) -0.084 -0.082 (-0.257) (-0.443) 0.015 0.010 (0.922) (1.081) -1.254*** (-4.123) 28.45% 24,918 Y 27.61% 13,170 Y Total LinkedIn analysts (Residual) Total non-LinkedIn analysts (Residual) Market cap Market-to-book ROA Total assets growth Cash flow volatility External financing Institutional ownership (dedicated) Institutional ownership (non-dedicated) Experience as analyst Experience with firm Analysts from top brokers All-star analysts Analyst portfolio size Industry covering experience Cover related industry -0.014*** (-2.615) 0.065*** (5.346) 0.075 (0.126) -0.042 (-0.343) 0.003 (0.013) 0.941* (1.953) -2.700*** (-4.488) -2.955*** (-9.228) 0.015 (0.756) -0.086*** (-3.180) -0.356** (-2.392) -0.026 (-0.138) 0.012 (1.340) -0.191* (-1.840) Cover customer/supplier -0.690** (-2.469) AC_IED_Pct Adj-R2 N Fixed Effects 28.41% 24,918 Y 28.43% 24,918 Y 58 28.43% 24,918 Y 28.80% 24,918 Y Internet Appendix This internet appendix accompanies, “Are all analysts created equal? Industry expertise and monitoring effectiveness of financial analysts,” submitted to the Journal of Accounting and Economics. Table IA.1 presents regression results for the analysis of whether industry expert analysts who discuss corporate governance issues in their research reports provide more effective monitoring. Table IA.2 presents regression results for the analysis of whether industry expert analysts who issue cash flow forecasts and who deviate more from management guidance provide more effective monitoring. Table IA.3 presents the multinomial logit regression results of the probability of firms’ signed discretionary accruals falling into the top quintile (the most positive discretionary accruals) or the bottom quintile (the most negative discretionary accruals). 59 Fixed Effects N Adj-R2 /Pseudo R2 Abnormal ROA * Related analysts without Gov Report (Residual) Abnormal ROA * Related analysts with Gov Report (Residual) Related analysts without Gov Report (Residual) Related analysts with Gov Report (Residual) 60 Y 23,233 Y 19,772 2.14% (-2.957) (-3.053) 61.89% -23.450*** (-1.894) -104.470* Restatements (Logit) -1.536*** (-3.659) -4.316*** F-score residual (OLS) Y 9,428 61.55% (-2.304) -2.357** (-2.238) -5.782** CEO Comp (OLS) Y 6,623 2.20% (-2.924) -176.440*** (-1.533) -232.330 CEO Turnover (Logit) This table presents excerpts of the regression results for the analysis of whether industry expert analysts who discuss corporate governance issues in their research reports provide more effective monitoring. Discussions of corporate governance-related issues on coverage firms in analyst reports are obtained from Thomson Reuters Investext. See Appendix B for a detailed description of variables. Coefficient values are reported as percentages. In parentheses below coefficient estimates are robust t-statistics based on standard errors adjusted for heteroskedasticity (White (1980)) and firm clustering (Petersen (2009)). *, **, and *** indicate statistical significance at the 10%, 5%, and 1%, respectively. Internet Appendix Table IA.1. Industry Expert Analyst Monitoring Mechanisms: Governance Discussions Fixed Effects N Adj-R2 /Pseudo R2 Low Deviation Related analysts (Residual) High Deviation Related analysts (Residual) Related analysts without CF Forecast (Residual) Related analysts with CF Forecast (Residual) Y 23,233 61 Y 19,772 2.18% (-2.099) (-2.000) 61.82% -15.320** (-2.420) -106.810** Restatements (Logit) -1.026** (-3.022) -2.692*** F-score residual (OLS) Y 23,233 61.88% Y 19,772 2.12% (-2.945) (-2.684) (-2.215) -22.150*** (-2.020) -1.298*** -38.280** Restatements (Logit) -1.879** F-score residual (OLS) This table presents excerpts of the regression results for the analysis of whether industry expert analysts who issue cash flow forecasts and who deviate more from management guidance provide more effective monitoring. See Appendix B for a detailed description of variables. Coefficient values are reported as percentages. In parentheses below coefficient estimates are robust t-statistics based on standard errors adjusted for heteroskedasticity (White (1980)) and firm clustering (Petersen (2009)). *, **, and *** indicate statistical significance at the 10%, 5%, and 1%, respectively. Internet Appendix Table IA.2. Industry Expert Analyst Monitoring Mechanisms: Cash Flow Forecasts and Divergence from Management Guidance Internet Appendix Table IA.3. Income-increasing vs income-decreasing earnings management: Multinomial logit approach This table presents the results from a multinomial logit regression. The dependent variable is a nominal variable that takes the value of one for observations with signed discretionary accruals in the bottom quintile (the most negative), 2 for observations with signed discretionary accruals in the middle three quintiles (moderate), and 3 for observations with signed discretionary accruals in the top quintile (the most positive). The sample consists of 24,918 firm-year observations from 1988 to 2011. The middle three quintiles serve as the base category. *, **, and *** indicate statistical significance at the 10%, 5%, and 1%, respectively. Related analysts (Residual) Unrelated analysts (Residual) Inexperienced analysts (Residual) Market cap Market-to-book ROA Total assets growth Cash flow volatility External financing Institutional ownership (dedicated) Institutional ownership (non-dedicated) Experience as analyst Experience with firm Analysts from top brokers All-star analysts Analyst Portfolio size Pseudo R2 N Fixed Effects Bottom Quintile Top Quintile -5.000*** -9.530*** (-2.591) (-4.990) -1.650 -0.898 (-1.012) (-0.736) 1.830 -0.146 (0.938) (-0.077) -1.810*** -1.380*** (-4.986) (-5.018) 0.717 1.670*** (1.469) (3.450) -73.670*** 125.580*** (-4.573) (6.673) 17.520*** 4.530 (3.784) (0.972) -22.320*** 4.830 (-2.773) (0.972) -6.440 51.240*** (-0.376) (2.869) -89.570*** -65.150*** (-4.384) (-3.314) -29.770** -29.880*** (-2.512) (-2.623) 2.920*** 1.510** (3.957) (2.163) -4.710*** -8.810*** (-3.925) (-7.342) -15.570*** -15.040*** (-2.621) (-2.662) -19.870** -6.820 (-2.238) (-0.824) -1.550*** -0.948** (-3.944) (-2.548) 4.98% 24,918 Y