UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Essays on capital markets Rahman, Nafis 2016

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata

Download

Media
24-ubc_2016_november_rahman_nafis.pdf [ 4.12MB ]
Metadata
JSON: 24-1.0314098.json
JSON-LD: 24-1.0314098-ld.json
RDF/XML (Pretty): 24-1.0314098-rdf.xml
RDF/JSON: 24-1.0314098-rdf.json
Turtle: 24-1.0314098-turtle.txt
N-Triples: 24-1.0314098-rdf-ntriples.txt
Original Record: 24-1.0314098-source.json
Full Text
24-1.0314098-fulltext.txt
Citation
24-1.0314098.ris

Full Text

  Essays on Capital Markets  by  Nafis Rahman     A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF  Doctor of Philosophy    in  THE FACULTY OF GRADUATE AND POSTDOCTORAL STUDIES (Business Administration)  The University of British Columbia (Vancouver)  August 2016  © Nafis Rahman, 2016     ii  Abstract   This thesis is a collection of three essays on capital markets.  The first essay examines how signals of reputation with non-equity stakeholders affect the market reaction to accounting restatements. Using Corporate Social Responsibility (CSR) rating as a proxy for reputation with non-equity stakeholders, I find significantly less negative market reaction to restatements for firms with better reputation. I also find that high-CSR firms experience smaller earnings-decreases and need to engage in fewer reputation restoration activities. The results suggest that a significant portion of the market value loss triggered by restatements reflects an expectation that the restating firms will face a ‘worsening of terms’ in their future transactions with the non-equity stakeholders, and CSR reputation can dampen this effect. The second essay examines the impact of accounting restatements on the information content of analyst forecast revisions (FRIC).  I find that following material restatements that are perceived to be intentional, FRIC increases significantly compared to the pre-restatement period level.  The results suggest that investors increase their reliance on analysts when there is uncertainty about the firm and the credibility of management disclosure is compromised. Additional tests reveal that the effect is greater for analysts who are less likely to have close ties with the management. The third essay studies how misaligned language between the investor and the firm contributes to the foreign investor bias.  In particular, we document a significant US institutional investor bias against firms located in Quebec relative to firms located in the Rest of Canada (ROC).  The differential bias is surprising given that Quebec and the ROC share the same country, federal law, stock exchange, accounting standards, and regulatory filings are prepared in both English and French; and given that US institutional investors are sophisticated investors at close geographic proximity to both Quebec and the ROC.  We also contrast the bias against Quebec firms with different levels of French versus English online presence, and we contrast the bias of institutional investors located in the UK versus France, to bolster our conclusion that incongruent languages are a major source of bias.   iii  Preface  The research projects in chapters 2 and 3 were identified and performed solely by the author.  The essay in chapter 4 is based on unpublished research with Russell Lundholm (The University of British Columbia) and Rafael Rogo (The University of British Columbia).  For the co-authored project, all authors worked on all aspects of the paper.  However, my contribution in the co-authored project is more concentrated in the aspects of research question identification, theoretical development, and empirical analysis, and less so in the aspect of manuscript writing. It would be appropriate to assess that I provided one-third contribution in chapter 4.               iv  Table of Contents   Abstract ......................................................................................................................................................... ii Preface ......................................................................................................................................................... iii Table of Contents ......................................................................................................................................... iv List of Tables ................................................................................................................................................ vi List of Figures ............................................................................................................................................. viii Acknowledgements ...................................................................................................................................... ix Chapter 1: Introduction ................................................................................................................................ 1 Chapter 2: Reputation with Non-Equity Stakeholders and the Equity Market Reaction to Accounting Restatements ................................................................................................................................................ 2 2.1. Introduction ....................................................................................................................................... 2 2.2 Literature Review and Hypothesis Development ............................................................................... 8 2.2.1. The Restatement Literature ........................................................................................................ 8 2.2.2 The CSR and Reputation Literature ............................................................................................ 10 2.2.3 Hypotheses Development .......................................................................................................... 12 2.3. Sample .............................................................................................................................................. 19 2.4. Variable Construction and Research Design .................................................................................... 22 2.5. Empirical Results .............................................................................................................................. 28 2.5.1. Descriptive Statistics ................................................................................................................. 28 2.5.2. Regression Analysis ................................................................................................................... 31 2.5.3. Sensitivity Checks and Additional Analysis ............................................................................... 44 2.5.4. Limitations and Ideas for Future Research ............................................................................... 46 2.6. Conclusion ........................................................................................................................................ 47 Chapter 3: Does the Information Content of Analyst Forecast Revisions Increase following Accounting Restatements? ............................................................................................................................................ 60 3.1. Introduction ..................................................................................................................................... 60 3.2. Literature Review and Hypotheses Development ........................................................................... 65 3.3. Sample .............................................................................................................................................. 69 3.4. Research Design and Variable Construction .................................................................................... 72 3.4.1. Time Line ................................................................................................................................... 72 3.4.2. Empirical Models and Variable Measurements ........................................................................ 73 v  3.5. Empirical Results .............................................................................................................................. 78 3.5.1. Descriptive Statistics ................................................................................................................. 78 3.5.2. Regression Analysis ................................................................................................................... 80 3.5.3. Additional Analysis (Untabulated) ............................................................................................ 87 3.6. Conclusion ........................................................................................................................................ 88 Chapter 4: The Foreign Investor Bias and its Linguistic Origins .................................................................. 99 4.1. Introduction ..................................................................................................................................... 99 4.2. Literature Review and Hypothesis Development .......................................................................... 102 4.2.1. Sources of Bias that Differ Between Quebec and Rest of Canada .......................................... 103 4.2.2. Does Part of the Investor Bias against Quebec have a Basis in Language? ............................ 109 4.2.3. Determinants of Institutional Holdings Unrelated to Home Bias ........................................... 114 4.3. Research Design ............................................................................................................................. 115 4.3.1. Dependent Variables ............................................................................................................... 115 4.3.2. Independent Variables ............................................................................................................ 117 4.3.3. Empirical Models ..................................................................................................................... 118 4.3.4. The Sample .............................................................................................................................. 121 4.4. Empirical Results ............................................................................................................................ 122 4.4.1. US Institutional Investor Bias against Quebec versus the Rest of Canada ............................. 122 4.4.2. Does the Investor Bias against Quebec Have a Basis in Language? ........................................ 129 4.5. Conclusion ...................................................................................................................................... 135 Chapter 5: Conclusion ............................................................................................................................... 149 References ................................................................................................................................................ 152 A   Appendices for Chapter 2 .................................................................................................................... 160 Appendix A1 .......................................................................................................................................... 160 B   Appendices for Chapter 4 .................................................................................................................... 162 Appendix B1 .......................................................................................................................................... 162 Appendix B2 .......................................................................................................................................... 164    vi  List of Tables  Table 2.1.  Variable definitions ................................................................................................................... 48 Table 2.2.  Sample ....................................................................................................................................... 52 Table 2.3.  Descriptive statistics .................................................................................................................. 53 Table 2.4.  Correlations table ...................................................................................................................... 54 Table 2.5.  Multivariate regression examining the effect of CSR on the market reaction to accounting restatements ............................................................................................................................................... 55 Table 2.6.  Multivariate regression examining the effect of CSR on the market reaction to accounting restatements (Robustness) ......................................................................................................................... 56 Table 2.7A. Multivariate regression examining the effect of CSR on decrease in earnings following restatement ................................................................................................................................................ 57 Table 2.7B. Multivariate regression examining the effect of CSR on decrease in earnings (relative to forecasted earnings) following restatement .............................................................................................. 58 Table 2.8.  Multivariate logistic regression examining the effect of CSR on management turnover following restatement ................................................................................................................................ 59 Table 3.1. Variable Definitions .................................................................................................................... 91 Table 3.2. Restatement Sample .................................................................................................................. 93 Table 3.3. Descriptive statistics of regression variables ............................................................................. 94 Table 3.4. Multivariate regression examining the change in information content of analyst forecasts following accounting restatement .............................................................................................................. 96 Table 3.5. Multivariate regression examining the information content of forecasts issued by new analysts following accounting restatement .............................................................................................................. 97 Table 3.6. Multivariate regression examining the information content of forecasts issued by new analysts following accounting restatement (Robustness Analysis) .......................................................................... 98 Table 4.1. Variable definitions .................................................................................................................. 136 Table 4.2. Descriptive statistics: US institutional holdings of Canadian firms .......................................... 139 Table 4.3. Correlations Table: US Institutional holdings of Canadian firms (Full Sample)........................ 141 Table 4.4. Regression of U.S. institutional holdings on QC dummy and control variables ....................... 142 Table 4.5. Regression of U.S. institutional holdings on QC dummy and control variables (alternate specifications and robustness) .................................................................................................................. 143 vii  Table 4.6. US investor bias investigation: regression using matched pair sample ................................... 144 Table 4.7. Descriptive statistics: US institutional holdings of Quebec firms ............................................ 145 Table 4.8. Regression of U.S. institutional holdings on FRENCHNESS and control variables ................... 146 Table 4.9. Descriptive Statistics: European institutional holdings of Canadian firms .............................. 147 Table 4.10. Regression of UK and French institutional holdings on Quebec dummy and control variables .................................................................................................................................................................. 148   viii  List of Figures  Figure 3.1. Time line of events (basic) ........................................................................................................ 90 Figure 3.2. Time line of events (split) .......................................................................................................... 90     ix  Acknowledgements  I would like to express my special gratitude and appreciation to Russell Lundholm and Rafael Rogo.  I had the wonderful privilege to work with them closely, and to get to know them as mentors, colleagues, and friends. They provided invaluable support and guidance throughout my PhD program.  I can never thank them enough.  I am very grateful to Ralph Winter for agreeing to join my committee and providing valuable insights from the perspective of someone outside the accounting discipline.  I would also like to thank all the faculty members of the accounting division for their support and helpful comments.   I am grateful to my classmates and friends at UBC for their encouragement and camaraderie.  They have been great companions in my academic journey of the past 5 years.  I would like to extend my special thanks to Elaine Cho, the Administrator of the PhD and MSc Programs, for her wonderful assistance in completing all the administrative tasks at UBC. I would like to thank my friends and family for their support and prayers.  I would especially like to thank my brother for being a great role model and inspiring me to pursue a career in academia.  Most importantly, I would like to thank my wife for loving me unconditionally and providing steadfast support through the most difficult times of my PhD program.     1  Chapter 1: Introduction  This thesis is a collection of three essays exploring the various information channels and different forces that influence the equity capital market.  The first two essays investigate the forces that can shape investors’ response to accounting restatements.  The first essay examines how reputation with non-equity stakeholders can mitigate the negative market reaction to accounting restatements.  The second essay investigates whether investors increase their reliance on analyst forecasts following accounting restatements.  While the first two essays investigate forces that arise in response to a shock to the corporate information environment, the third paper explores a particular aspect of a well-known but ill-understood investment pattern, the investor home bias.  Specifically, the third essay explores the role of language difference in foreign investor bias in Canadian equity market. Since each essay investigates a different topic, the chapters are designed to be self-contained.  I leave a more exhaustive discussion of the research question and contribution to the introduction specific to each chapter.    2  Chapter 2: Reputation with Non-Equity Stakeholders and the Equity Market Reaction to Accounting Restatements  2.1. Introduction Accounting restatements have major negative consequences, with an abnormal return around the restatement announcements ranging from −4% to −12%, depending on the sample examined (Dechow et al., 1996; Palmrose et al., 2004).  Prior research shows that only one-third of the share value loss triggered by restatements is traceable to direct causes such as reduced valuation reflecting revised earnings numbers and expected litigation costs (Karpoff et al., 2008).  The residual or unexplained portion of market value loss triggered by restatements is often attributed to reputation damage (Karpoff et al., 2008; Karpoff, 2012).   In this context, a firm’s reputation refers to stakeholders’ trust that management is willing and able to honor the firm’s explicit and implicit contracts (Karpoff, 2012).  A good reputation reduces uncertainty about the firm’s choice not to act opportunistically to the detriment of its stakeholders. The investors are more willing to hold the stock of a reputable firm; the non-equity stakeholders are more willing to do business with a reputable firm and offer more favorable terms in their transactions (Klein & Leffler, 1981; Chakravarthy et al., 2014).  Klein & Leffler (1981) define a firm’s reputation capital as the expected present value of the future trust-premiums to be received by the reputed firm.   An accounting restatement reveals the failure of management to provide truthful and accurate information about the firm’s financial position, and naturally constitutes a breach of trust between the firm and the investors. Prior research shows that restatements damage the firms’ reputation with their investors.  Following a restatement, the mandatory disclosures made 3  by the firm lose credibility (Chen, Cheng, & Lo, 2014; Wilson, 2008), and institutional investors are less willing to hold the stock of the restating firm (Hribar, Jenkins, & Wang, 2004).  It is obvious that part of the market value destruction triggered by the restatement reflects the loss of reputation with the investors.  Whether the market value loss triggered by restatement also reflects a loss of reputation with the non-equity stakeholders is a question not well explored in the prior empirical literature.   Restatement can create uncertainty about the ability and the willingness of the firm to honor its future obligations to its non-equity stakeholders since 1) it reveals that the firm has fewer financial resources than previously claimed, and 2) it raises questions about management’s intentions; this uncertainty can push the non-equity stakeholders to charge higher premiums in their future transactions with the restating firm (Chakravarthy et al., 2014).  For example, following restatements, the suppliers may be less willing to extend trade credit, creditors may charge a higher interest rate, or customers may become more suspicious about the product warranty.  Rational investors would anticipate such ‘worsening of transaction terms’ with the non-equity stakeholders in the event of a restatement announcement, and this could explain part of the share value loss triggered by restatement news.   However, if the non-equity stakeholders have reasons to continue trusting the restating firm despite the restatement announcement, then they would not significantly alter the terms of their transactions with the restating firm, and the negative consequences of restatement should be smaller for such a firm. In this paper, I investigate whether, and by how much, the signals of strong reputation with non-equity stakeholders can mitigate the negative market reaction and other adverse consequences faced by the restating firm. 4  I use a firm’s Corporate Social Responsibility (CSR) rating as a measure of its reputation with the non-equity stakeholders.  Conceptually, CSR is a measure of how ‘good’ the firm is towards its non-equity stakeholders; the increased engagement with the non-equity stakeholders can foster a reputation with them (Freeman, 1984; Jones, 1995).  Empirical research show that firms with better CSR ratings receive more favorable treatments from their non-equity stakeholders (Greening & Turban, 2000; Lev et al., 2010; Nan & Heo, 2007; Zhang et al., 2014).  Anecdotal evidence suggests there is a connection between CSR activities and firm reputation; corporate executives cite their desire to protect their firms’ reputation when asked to explain the reason behind their choice to engage in CSR reporting (Christensen, 2016; Crespin, 2012; KPMG, 2011; The Economist, 2005).  Moreover, prior research establishes the empirical validity of CSR as a proxy for reputation by showing that CSR ratings are negatively associated with opportunistic behavior towards the investors (Kim et al., 2012) and the non-equity stakeholders (Christensen, 2016). CSR ratings use performance metrics that are often outside the scope of financial reports (KPMG, 2011), and they can impart useful information about the firm’s business risk and management style that can supplement the mandatory financial disclosures (Christensen, 2016).  More than $3 trillion of invested money in socially responsible investment funds in the US (Social Investment Forum Foundation, 2010) ensures that 1) firms actively disclose and advertise their CSR performance, and 2) reputed third parties track the CSR performance of public firms.  In times of restatement, when the credibility of the financial reporting is in question, the investors can glean valuable information about operational strengths of the firm from its CSR ratings.  This can reduce uncertainty about the restating firm’s future, and assures the investors  5  that the non-equity stakeholders will not significantly worsen their terms of trade with the restating firm, which should translate into less negative market reaction to restatement news.  In this paper, I find that the market reaction to restatement news is indeed less negative for firms with good reputation with their non-equity stakeholders (high CSR firms).  My results remain significant after controlling for a battery of variables introduced in the restatement and CSR literature, and are robust to alternative specifications.  Tests based on the market reaction to first revelation dates rule out the possibility that my results arise due to restatement news of high-CSR firms systematically leaking ahead of their official announcement dates. Not all restatements are intentional, and some of them have negligible impact on the previously reported earnings.  Accordingly, investors tend to react more negatively to material restatements (Hennes et al., 2008).  The reputational impact of such material restatements should also be larger since they are likely to cause greater uncertainty about the restating firm’s ability and willingness to honor its future obligations.  The scope of the restating firm’s ‘reputation with its non-equity stakeholders’ to assuage investor concerns about the future operational prospects of the restating firm is also greater for material restatements.  My empirical analysis shows that CSR’s positive impact on the market reaction to restatement is larger for material restatements.  I find that CSR does not have a significant impact on the market reaction to errors—immaterial restatements that are unlikely to create uncertainty about the future operations of the firm.  This confirms that CSR activities protect firm value through an impact on the firm’s reputation.   One could argue that the investors’ favorable treatment of high-CSR firms in times of restatements reflects their psychological preference, and is not justifiable from a shareholder profit maximizing perspective (Bénabou & Tirole, 2010; Hong & Liskovich, 2015).  To support my argument that the observed investor reaction is consistent with economic rationale, I examine 6  the firms’ future earnings and management turnover in the post restatement period.  Prior research documents a decrease in reported earnings following restatement announcements (Murphy et al. 2009).  I show that the decrease in earnings is mitigated for high-CSR firms, and the effect is larger for material restatements.  The results support my argument that a strong reputation with the non-equity stakeholders mitigates the increased cost of doing business faced by the restating firm.  Prior research also shows that following restatements, firms often fire the senior management to restore company reputation (Chakravarthy et al., 2014; Farber, 2005; Hennes et al., 2008).  The intensity of reputation restoration activity is likely to depend on the magnitude of the reputational impact of the restatement, and the cost of the restorative actions.  Hennes et al. (2008) show that management turnover following restatement announcement is higher for material restatements.  In this paper, I find that high-CSR firms experience fewer management turnovers following restatement, suggesting that the reputation damage due to restatement is smaller for high-CSR firms.  As in the case of the market’s reaction, CSR has an economically meaningful and significant impact in lowering the probability of management turnover following material restatements, but CSR does not have a significant impact on management turnover following errors.   This paper makes several contributions to the literature. While our understanding of how restatement can damage a firm’s reputation with its investors is well developed, very few papers explore the connection between restatement and the firm’s reputation with its non-equity stakeholders.  The closest paper is Chakravarthy et al. (2014), which shows that restating firms engage in various reputation restoration activities targeted at different groups such as the investors, the employees, and the community.  But they do not test whether the market reaction 7  to restatement reflects a loss of reputation with non-equity stakeholders.  In contrast, I focus mainly on how investors assess the signals of reputation with the non-equity stakeholders when they react to restatement news.   Findings from my market reaction analysis is similar to that of Godfrey et al. (2009) and Christensen (2016), but the earlier papers investigate the impact of CSR on the market reaction to news of misconducts that are heterogeneous in terms of the injured party.  Investors would have to assess the firm-value implications for such misconducts in an abstract way.1  Both the papers argue that a firm’s CSR record creates moral capital, which results in lesser penalties for misconducts based on the legal principal of mens rea.2 In contrast, this paper finds that CSR reputation can mitigate the negative investor reaction to restatements, events where the investors themselves are the aggrieved party and the damage to equity value is readily apparent.  Since I investigate only one type of misconduct (restatements), I can control for a battery of event-specific factors that have been shown to influence the market reaction to such misconducts.  Additionally, I find that CSR’s ability to impact valuation increases with the uncertainty level (material restatement).  My research setting allows for cleaner identification of the impact of CSR.  Most importantly, I present an economic argument for why investors would view CSR positively in the wake of restatements, and substantiate my arguments using post-restatement consequence analyses. Secondly, prior research shows that investors distrust and discount the financial information disclosed by the firm after a restatement announcement (Chen et al., 2014; Wilson,                                                           1 Different types of misconduct have different share value consequences.  Equity holders might never have to bear the cost of misconducts like pollution or bribery.  It is difficult to assess the stock-price impact of such misconducts. 2 According to the mens rea principle, punishment of a crime depends on the perceived intention of the actor.  If stakeholders perceive the firm to have good intentions in general because of their CSR, they would give the firm the benefit-of-doubt in times of a scandal. 8  2008).  A natural question to ask is whether investors increase their reliance on information outside the scope of financial reports.  The results from this paper suggest that investors consider alternative sources of information to assess the firm’s prospects in the wake of restatement.  Finally, my results provide empirical evidence of the ‘insurance’ value of CSR activities hypothesized by earlier theoretical work (Peloza, 2006).  The results are consistent with the stakeholder theory of CSR, which suggests that CSR activities can create firm value (Jones, 1995; Orlitzky et al. 2003).  A different school of thought argues that CSR proxies appear to be positively associated with firm value only because successful firms have the required slack resources, and self-select to be high-CSR firms (Lys, Naughton, & Wang, 2015; Margolis, Elfenbein, & Walsh, 2007).  My results indicate that investors value CSR activities positively even in the case of material restatements, when firm’s financial health is questioned, and concern for managerial integrity and agency problem is magnified. I proceed as follows. In Section 2.2, I discuss prior literature and develop my hypotheses.  Section 2.3 describes the sample.  Section 2.4 presents the variable construction and research design.  I discuss the empirical results in section 2.5, and conclude in Section 2.6.   2.2 Literature Review and Hypothesis Development  2.2.1. The Restatement Literature The capital market functions with the general assumption that the audited financial statements contain reliable information about a firm’s economic reality.  Since industry updates, analyst reports, management guidance and other timely sources of information generally precede the earnings announcement, the importance of mandatory financial disclosures comes from their confirmatory and contracting role (Beyer et al., 2010).  These key  features of financial 9  statements are compromised when it is revealed that the previously filed financial statements, which were believed to be accurate by the investors (Bardos et al., 2011), can no longer be relied upon.  Empirical research documents that restatements have far ranging negative consequences on firm value and the corporate information environment.  Among all the corporate misconducts—bribery, product recall, antitrust violations—financial reporting violations trigger the most negative market reaction, ranging from −4% to −12% (Karpoff, 2012; Palmrose et al., 2004).  This loss in market value has been attributed to a number of factors, including revisions of expected future earnings due to the non-existence of past earnings, expected litigation costs due to restatements, uncertainty regarding managerial competence and integrity, and expectation of punishment—worsening of transaction terms— by the stakeholders (Karpoff, 2012). Using a sample of restatements involving SEC enforcement actions, Karpoff et al. (2008) estimate that only 24.5% of the share value loss triggered by restatements comes from market readjustment to the corrected financial representation, and another 8.8% reflects the expected legal penalties from regulatory fines and class action lawsuits.  The remaining 66.7% of share value destruction following financial misreporting news is attributed to reputational loss (Karpoff et al., 2008).  The large reputational loss most likely reflects the market’s expectation that the misreporting firm’s future operations will be hampered.  Consistent with the reputational argument, prior research shows that following restatements, the financial disclosures made by the firm lose credibility (Chen et al., 2014; Wilson, 2008), the restating firms experience greater cost of borrowing (Graham et al., 2008), and misreporting firms face greater litigation risk (Scholz, 2004).  Murphy et al. (2009)  finds that securities fraud leads to decreased future earnings and greater variation in stock price.  The increased cost of doing business is reflected in investor 10  expectations, and firms face higher cost of equity following restatements (Hribar & Jenkins, 2004).  Given the significant and multifaceted negative consequences of restatements, firms take multiple actions aimed at both investors and non-equity stakeholders to restore reputation, and news of such actions elicit a positive market reaction (Chakravarthy et al., 2014).  Prior research shows that fraud firms that improve their corporate governance in the post-restatement period experience superior stock performance compared to those that do not (Farber, 2005).  Hennes et al. (2008) show that CEO and CFO turnover is higher after restatements, and this effect is more pronounced for material restatements.  Following restatements, the directors are more likely to get fired, and are less likely to hold position on the board of another company (Srinivasan, 2005). While the restorative actions are costly, such actions seem to help restoring the investor trust in the restating firm’s disclosures (Chen et al., 2014). 2.2.2 The CSR and Reputation Literature The stakeholder theory of CSR argue that CSR activities can demonstrate a commitment to ‘good and ethical’ actions, and such reputation can induce the stakeholders to treat the firm favorably (Freeman, 1984; Jones, 1995).  For example, Zhang et al. (2014) find that it is easier for firms with superior CSR performance to obtain trade credit from suppliers.  Experimental research show that CSR activities can improve the effectiveness of a firm’s marketing strategy (Nan & Heo, 2007), and attract higher quality employees (Greening & Turban, 2000).  Lev et al. (2010) find that CSR activities can increase demand for the firm’s products and services   From a signalling perspective, a firm’s decision to engage in CSR activities, which are not mandatory, can signal that the firm has sufficient slack resources (Campbell, 2007).  Lys et al. (2015) find some empirical evidence consistent with the argument that managers have private information 11  about the firm’s operational health and they can signal their high type to the investors by engaging in CSR activities. Karpoff (2012) defines a firm’s reputation as the stakeholder’s trust that the firm is willing and able to honor its future commitments.  A well accepted economic explanation for why reputation creates firm value is that reputation reduces uncertainty faced by the stakeholders in their transactions with the firm (Weigelt & Camerer, 1988), and this allows the reputed firm to charge a premium in its transactions (Chakravarthy et al., 2014; Karpoff, 2012; Klein & Leffler, 1981).  Klein & Leffler (1981) develop an analytical model where the value of a firm’s reputation is the present value of the future expected reputation premiums the firm will receive from its transactions with its stakeholders.  Applying this concept in the CSR context, one can say that CSR activities generate firm value by attracting ‘favorable treatments’ or ‘reputation premiums’ from stakeholders (Jones, 1995).  Indeed, prior empirical research finds that high CSR firms enjoy superior financial performance (Orlitzky et al., 2003) and lower cost of equity (Dhaliwal et al., 2011).  The reputation of being trustworthy can be very useful in times of restatements when the management integrity is questioned. Anecdotal evidence from top executives suggests that there is a strong connection between CSR activities and reputation.  When responding to survey questions, corporate executives claim that enhancement of brand and employee morale are two major benefits of CSR (The Economist, 2005).  Large firms engage in CSR activities and demand sustainable practices from their trading partners to protect their brand (Crespin, 2012).  Firms can enhance their reputation with their CSR activities, and there is a strong trend of large companies issuing full scale corporate responsibility reports (KPMG, 2011). 12  Prior empirical research validates CSR rating as a proxy for firm’s reputation by showing that CSR rating has a significantly negative correlation with ex-ante probability of engaging in opportunistic behavior by the firm (Christensen, 2016; Kim et al., 2012).  Kim et al. (2012) show that firms with high levels of CSR activities are less likely to manage earnings through discretionary accruals, less likely to manage earnings through real operating activities, and less likely to be subject of SEC investigations.  Christensen (2016) show that CSR activities and the practice of their reporting is associated with lower probability of misconducts relating to non-equity stakeholders like bribery and discrimination.  Moreover, the market reaction to news of CSR related misconducts is less negative for high-CSR firms (Christensen, 2016).  These empirical findings are consistent with CSR activities generating a reputation with the stakeholders.  2.2.3 Hypotheses Development A restatement represents a breach of contract between the firm and an important stakeholder group—the investors.  It also shows that the firm has fewer financial resources than was previously reported.  Hence restatements can create uncertainty about both the firm’s ability and willingness to honor its commitments.  Facing such uncertainties, the non-equity stakeholders are likely to adversely alter their terms of transactions with the firm to insure themselves, which in turn would damage a firm’s reputation capital with its non-equity stakeholders.  For example, the suppliers might become more hesitant to extend trade credits, the skilled employees might try to move to another firm to protect their retirement funds, the banks might charge higher interest rates, large customers with pricing power might push down the prices of the firm’s warranty services, the local community leaders might start favoring the firm’s rival company, and the regulators might subject the firm to greater scrutiny.   13  While there has been very little empirical research exploring the impact of restatement on a firm’s non-equity stakeholders, findings of a couple of papers are relevant in this discussion.  Dechow et al. (1996) and Palmrose et al. (2004) find that analyst forecast dispersion increases following announcement of misreporting, which suggests an increase in uncertainty about the firm’s prospects.  Prior research also finds that following restatements, the restating firms experience greater cost of borrowing (Graham et al., 2008) and decreased future earnings (Murphy et al., 2009).  Rational investors are likely to anticipate that the uncertainty created by restatements will lead to a loss of reputation capital with the non-equity stakeholders, and this expectation is likely to be reflected in the market reaction to restatement announcement. If part of the share value loss trigged by the restatement announcement is caused by investors’ uncertainty about the restating firm’s ability to continue doing business with its stakeholders as before, a restating firm’s strong reputation with its non-equity stakeholders will help to curb such concerns, and holding other things constant, the market reaction to restatement announcement will be less negative for such a firm.  For example, if the company is considered to be a reliable partner by its suppliers, has developed a strong relationship with the community in which it operates and sells products, and has consistently complied with the local regulators, then these stakeholders are more likely to continue their relationships with the firm even after the restatement announcement.  Having close ties with non-equity stakeholders can be quite useful in difficult times.  For example, during the recession of 1990-1991, the US auto-giant Chrysler maintained cooperative relationships with its suppliers rather than trying to pass on its thinning margins, and in turn the suppliers became more involved in the production process and suggested ways to improve efficiency and cut costs by 10%, which allowed Chrysler to turn a profit during the economic downturn and improve their competitive position in the industry (Rigby 2001).   14  CSR performance is informative about the firm’s reputation with its non-equity stakeholders.  From an empirical perspective, 1) the existence of dedicated institutions that track CSR activities of firms, 2) the growing trend among the largest US firms to voluntarily issue CSR reports, and 3) the recent innovation of audited CSR reports (KPMG 2011) suggest that CSR activities can reveal information about the corporate values and operational strengths that can supplement the mandatory financial disclosures.  Christensen (2016) argue that CSR reports can impart valuable information about the business risk of the firm.  Dhaliwal et al. (2012) find that stand alone CSR reports are associated with lower analyst forecast error, and this association is stronger in countries with more opaque financial disclosures; their findings suggest that CSR reports help resolving uncertainties about the future profitability of firms.  CSR ratings have some appealing properties due to the way they are compiled.  Business intelligence providers such as the MSCI and Thomson Reuters have dedicated departments that diligently track the CSR performance of firms and make it available to investors.  This system of CSR performance assessment, done by third parties who focus on performance metrics that are different from the ones used in financial statements, is largely independent of the financial reporting system.  Therefore, while the restatement can cause uncertainty in the financial reporting environment, the credibility of CSR ratings are unlikely to be impacted by accounting restatement. In times of restatement, CSR ratings are likely to reduce uncertainties about the operational strengths and managerial intent of the restating firm, and give assurance to investors that the restating firm will not experience much difficulty in continuing to do business with its stakeholders.  Therefore, high CSR ratings of the restating firm are likely to mitigate the negative market reaction to restatement announcements.   15  It is plausible that firms that are ‘good’ to their non-equity stakeholders are also ‘good’ to their investors.  Kim et al. (2012) find that CSR performance (social and environmental) is negatively associated with earnings management and the probability of being subject to SEC investigations.3  If high-CSR firms are less likely to engage in behavior that leads to misreporting, then restatement announcements made by such firms are likely to create greater negative surprise.  Holding other forces constant, the market reaction to restatement announcement would be more negative for firms with better CSR ratings due to this selection-bias effect.  However, CSR performance mainly captures a firm’s greater level of engagement with its non-equity stakeholders, and does not have a direct conceptual link with the firm’s reputation not to misreport.  At the time of restatement announcement, CSR’s positive effect in assuaging the investor concerns about the restating firm’s future operations will most likely dominate the negative effect of greater shock due to selection bias effect.  I predict that the market reaction to accounting restatement is less negative for restating firms with better CSR ratings. Material restatements raises the question of management malfeasance.  In times of material restatements, investors face greater uncertainty about the firm’s financial prospects and its ability to continue its core business operations as before.  It is likely that the share value loss caused by such uncertainties is larger for the material restatements.  In that case, CSR rating’s scope to reduce such uncertainties and thereby influence the valuation process in times of                                                           3 Sustainability reports cover three types of firm-actions: environmental, social, and governance.  However, prior research in CSR generally focuses on firms’ environmental and social performance measures and leave out the governance related measures, since governance metrics generally do not reflect the firms’ ‘outreach’ to the broader society.  In the case of Kim et al. (2012), finding a negative (positive) association between better governance and earnings management (better quality reporting) would be tautological and uninteresting. Following the vast majority of papers in the CSR literature, I leave out the corporate governance metrics from my CSR proxy (Bear et al., 2010; Greening & Turban, 2000; Kim et al., 2012; Lev et al., 2010; Zhang et al., 2014).  In this paper, the term CSR performance refers to just the social and environmental performance. 16  restatement announcement should also be greater.  Hence, CSR rating’s positive impact on market reaction to restatements should be larger for material restatements.  I develop the following set of hypotheses: H1 (a). CSR has a positive (less negative) impact on the market reaction to accounting restatements. H1 (b). The impact of CSR on the market reaction to accounting restatement is larger (more positive) for material restatements.  The main argument in this paper is that when a firm restates, the investors anticipate that the firm will face a ‘worsening of transaction terms’ with its non-equity stakeholders, and this expectation is partly responsible for the large negative market reaction observed around the restatement announcement date.  However, the high CSR rating of a restating firm assures investors about the firm’s strong relationships with its non-equity stakeholders, and signals that the firm will not experience too much difficulty in its regular operations following restatement.  Hence investors act less negatively to restatement announcements of high-CSR firms.  However, investor’s preferential treatment of the high-CSR firms in times of restatement announcements can also be explained by investor’s psychological-preference or the direct-benefit theory of CSR (Bénabou & Tirole, 2010; Di Giuli & Kostovetsky, 2014; Friedman & Heinle, 2015).  According to the psychological-preference theory, some investors get psychological benefit from owning stocks in the high-CSR firms and thereby indirectly contributing to social causes (Friedman & Heinle, 2015; Hong & Liskovich, 2015).  Corporations are well positioned to pool large sums of money and engage in ‘delegated philanthropy’ on behalf of their investors (Hong & Liskovich, 2015).  For the social-welfare motivated investors, the value of a firm depends on its social performance as wells as future 17  cash-flow and risk (Bénabou & Tirole, 2010; Di Giuli & Kostovetsky, 2014; Friedman & Heinle, 2015).  Restatement of prior financial reports does not change a firm’s CSR record, and hence socially motivated investors will act less negatively to accounting restatements compared to purely profit seeking investors.  In the equilibrium, the socially motivated investors can influence prices, and this can result in less negative market reaction to restatement announcements made by the high-CSR firms.  Exploring the fundamental performance of the restating firms in the post-restatement period can confirm if the high-CSR firms suffer less operational difficulty following restatement.  Prior research shows that earnings of the restating firms decrease following restatements (Murphy et al., 2009). This finding is consistent with the notion that following restatement, the investors and non-equity stakeholders offer less favorable contracts in their transactions with the restating firm.  But such a negative impact on future earnings should be mitigated to some extent if the restating firm has developed a trusting relationship with the non-equity stakeholders, and they continue their business with the restating firm as before. According to the stakeholder-reputation theory, the decrease in earnings following restatement should be smaller for high-CSR firms.  Since material restatements have a larger scope of adversely altering the restating firm’s terms of transactions with the non-equity stakeholders, CSR reputation is likely to play a larger role in determining a firm’s profitability following material restatements.   H2 (a). CSR has a positive (less negative) impact on the change-in-earnings following accounting restatements. H2 (b). The impact of CSR on the change-in-earnings following accounting restatement is larger (more positive) for material restatements. 18  Prior research shows that following restatement news, the firms are more likely to take reputation restorative actions such as firing top management, changing brands, and initiating employee benefit programs (Chakravarthy et al., 2014).  Prior research also shows that firing the top management can help restore the credibility of financial disclosure for restating firms (Chen et al., 2014; Wilson, 2008).  Hennes et al. (2008) find that management turnover following material restatements is much higher compared to regular restatements.  Findings of prior research is consistent with the notion that the restating firms try to repair their reputations after restatement announcement, and the intensity of such reputation-repairing actions increases with the seriousness of the misreporting violation.  Management turnover is a rather costly reputation restoration activity, and firms are likely to be averse to undergo such measures when possible.  The restating firms would not have to take such drastic measures if the reputation damage due to restatement is low.  A firm’s CSR rating is expected to reduce the reputation damage due to restatements.  Hence a restating firm with high-CSR rating is likely to face less pressure to fire its top managers after restatement announcement.  As explained earlier, the damage-controlling effect of CSR is expected to be larger for material restatements.  Hence it is likely that CSR’s impact in lowering the chances of senior management turnover following restatement should be more pronounced for material restatements. Management turnover in the post restatement period may not be lower for restating firms with better CSR ratings if the corporate culture of high-CSR firms is fundamentally different and the directors of such firms are systematically less tolerant of financial misrepresentations.  Also, a restating firm can choose from multiple possible actions to restore its reputation, and different firms seem to focus on different set of restorative actions (Chakravarthy et al. 2014).  With the 19  presence of opposing forces, whether CSR reputation is negatively associated with management turnover following restatements is an empirical question.  However, in the period immediately following the restatement announcement, the concern of reputation damage is likely to be the dominant force in determining the turnover decisions.  Management turnover in the period immediately after the restatement announcement is likely to be lower for restating firms with better CSR ratings. I test the following hypotheses in the alternate form: H3 (a). CSR has a negative impact on the probability of management turnover following accounting restatements. H3 (b). The impact of CSR on the probability of management turnover following accounting restatement is larger (more negative) for material restatements.  2.3. Sample  Main Sample: Table 2.2 Panel A summarizes the sample selection procedure used in this paper. I collect restatement data for US firms from Audit Analytics Non-Reliance database, which starts coverage of restatements from 2000.  My sample covers a period from January 2000 to June 2014.  The key advantage of using the Audit Analytics (AA) database is that it covers restatements of recent years (as opposed to the GAO database with coverage ending in 2005).  The other advantage of using the AA database is that it provides many important restatement severity measures such as the total dollar impact of the restatement, whether the restatement was a fraud, whether the restatement case involved an SEC investigation, and whether there were internal control issues.   20  I collect financial data from COMPUSTAT, and market return data from CRSP.  I first match the AA restatement observations with COMPUSTAT using historical CIK numbers available from CRSP-COMPUSTAT merged database.  COMPUSTAT does not have CIK information for all the firms.  Hence I employ a second matching technique based on suggestions in the Wharton Research Database Repository.  I require that the names are a close match according to computer algorithm, the states of the headquarters match, and the total assets and total sales of the past two years reported by AA and COMPUSTAT match within 1% of each other.  I manually checked the results of my second technique and found no mismatches.  I start with 14,351 restatement observations from AA.  I can match 10,777 of them with COMPUSTAT data.  AA and COPUSTAT cover pink slip firms not traded in the major exchanges.  Requiring CRSP coverage further reduces my sample size to 6184 observations. I collect CSR ratings data from the MSCI (RiskMetrics Group before merger) KLD database, which has the largest coverage of CSR ratings.  Standardized CSR performance data from a reputed and independent organization facilitates comparison between CSR performances of firms with different reporting styles.  For my analysis, I use the lagged CSR performance data since the restating firm is likely to alter its operational and CSR behavior after the restatement announcement (Chakravarthy et al., 2014).  An accounting restatement can only change the past financial information but it can’t change the past CSR ratings.  Requiring KLD coverages limits my sample size to 2,876 observations.  After removing duplicate observations, observations with missing total assets, and observations with missing total sales, the sample size decreases to 2,754 observations.4                                                           4 Following prior literature (Scholz, 2008), I identify restatement announcements as duplicates (referring to the same underlying event) if consecutive announcements are made within 90 days of each other, or if the announcements refer to the same earnings management period beginning and ending dates.  For duplicate announcements, I keep the earlier observation in my sample. 21  For each year, KLD provides CSR ratings for the largest 3000 firms by market size.5  KLD also tracks the Domini 400 Social Index that include some small firms with exceptional CSR performance.  I exclude such small firms since selecting my sample based on outstanding CSR performance will compromise the generalizability of my results; this brings my sample size to 2,271 observations.  Not all restatements revise the previously reported earnings downward.  I remove income increasing restatements from my sample since they are not purely negative news, and it is difficult to interpret the market reaction to such restatements.  After removing such income-increasing restatements, the final sample for my main analysis has 1892 restatement observations from 1336 unique firms. First Revelation Date Hand Collected Sample: AA database records the official announcement date of the restatement, when the company issues a press release, or files the non-reliance 8K item 4.02 with the SEC, or indicates restatement or adjustment of previously reported earnings in some other document filed with the SEC.  The investors may not significantly react to the official announcement of restatement if the news of restatement leaked earlier.  This can possibly introduce a positive bias to my market reaction test if the high CSR firms systematically are early announcers.  To address this concern, I manually collect the earliest revelation date of restatements for a subset of my sample by searching in FACTIVA.6 Prior research classifies restatements into four categories based on the type of accounts being restated and shows that restatements of revenue and core-expenses (i.e., cost of goods sold, SG&A, and etc.) create more uncertainty and trigger more negative market reaction (Palmrose et                                                           5 KLD covers the S&P 500 firms in the years before 2001, the largest 1000 firms in the years 2001 and 2002, and the largest 3000 firms from 2003 onward 6 When searching in FACTIVA, I first perform a key-word search (restat, amend, investigat, fraud, irregularity, inappropriate, review, misstat, misreport) within the past one year of the official announcement date, and then read through the positive results to check if the news items refer to the restatement event in my original sample. 22  al., 2004; Scholz, 2008).  The full sample for my analysis has 1892 restatement observations with the required data.  Out of them, 1005 observations involve restatements of revenue or core-expense accounts, and trigger significant market reactions (See Table 2.2 Panel B).  Due to high cost of manual collection, I only focus on these 1005 major restatements involving core-earnings accounts.  For 730 (74.5%) out of these 1005 observations, I can locate the actual news of the restatement through my search of FACTIVA or company website, or find substantiating documentation of the restatement from the SEC filing link provided by AA.  Hence, my hand-collected sample has 730 restatement observations (Table 2.2 Panel A).      2.4. Variable Construction and Research Design  In this section, I will briefly describe the dependent variables, the variable of interest, the restatement severity controls, the CSR control variables, the general controls, the management entrenchment controls, and the type-of-news controls used for my analysis.  Table 2.1 provides detailed definitions of the variables used in this paper. Dependent Variables: My primary dependent variable in this paper is CAR, the 3-day size-adjusted cumulative abnormal return on the trading days [−1,1] relative to the restatement announcements.  The second dependant variable is ∆EARN, which is the change in earnings scaled by total assets from the year prior to the restatement announcement to the year of the restatement announcement.  Not all the firms survive long enough to report earnings at the end-of-year of restatement announcement, and removing them from my analysis would potentially introduce a survivorship bias.  To address this concern, I keep the firms that do not survive the full year, and assign an earnings figure of zero for the year of restatement, and calculate ∆EARN 23  by subtracting the previous year’s earnings from zero.  I test hypothesis 2(a) and 2(b) using both the original sample (where the missing ∆EARN observations drop from the analysis) and the survivorship-bias-free sample, and reach the same conclusion.  I use ∆EARN_UE to investigate how restatements cause future earnings to deviate from market expectations.  ∆EARN_UE is defined as the change in earnings relative to the consensus earnings-forecast prior to restatement announcement, scaled by stock price at the end of the prior fiscal year.  The fourth dependent variable used in this paper is TURNOVER, which is an indicator variable equal to one if any of the CEO or CFO resign, retire, or get dismissed in the six month period following restatement, and zero otherwise.  Variable of Interest (The CSR Proxy): The variable of interest in this paper is the CSR rating of the restating firms in the year t-1 when restatement announcement happens in the year t.  Each year, the RiskMetrics Group (now owned by MSCI), evaluates the CSR performance in roughly 80 categories along seven dimensions: community, corporate governance, diversity, employee relations, environment, human rights, and products.  Each dimension has multiple strength (positive) and concern (negative) categories, and a binary measure indicates the presence or absence of a given strength or concern.  For example, if a company “has either been notably innovative in its support for primary or secondary school education, particularly for those programs that benefit the economically disadvantaged, or the company has prominently supported job-training programs for youth,” then it gets a score of 1 in the ‘Support for Education’ category, which is a positive category in the ‘community’ dimension (The RiskMetrics Group, 2010).  Appendix A1 provides a detailed list of the CSR performance categories. 24  Prior research in accounting and management literature sum up the binary scores across the different KLD performance categories to develop parsimonious measures of a firm’s ‘total’ CSR performance (Bear et al., 2010; Dhaliwal et al., 2011; Godfrey et al., 2009; Kim et al., 2012; Mattingly & Berman, 2006).  Prior research shows that the positive and negative indicators described in the KLD data are conceptually and empirically distinct constructs and should not be combined (Mattingly & Berman, 2006).  Bear et al. (2010) show that positive CSR ratings are positively associated with the survey based reputation proxies compiled from the Fortune World’s Most Admired Companies.  Godfrey et al. (2009) show that positive CSR ratings offer some insurance in times of negative events.  For my study, the ideal empirical proxy should reflect the firm’s organizational strengths that arise due to its investments in the relationships with its non-equity stakeholders.  Following guidance from prior research involving KLD data (Bear et al., 2010; Dhaliwal et al., 2011; Godfrey et al., 2009; Mattingly & Berman, 2006), I construct my reputation proxy based on the positive CSR ratings available from KLD database.  To arrive at a summary measure of reputation with all the non-equity stakeholders, I sum up the positive CSR indicators across all the main categories except corporate governance since corporate governance is closely connect to financial reporting quality, and there is a dedicated accounting literature exclusively focusing on the role of corporate governance on firm performance.  I call this variable the CSRTotal.  One might argue that the strength indicators under the Product category can pick up an industry (nature of business) effect.  For instance, a prescription drug maker may have an inherent advantage over an alcohol producer.  While the KLD brochure argues that the product strengths are assessed in comparison to a firm’s industry (The RiskMetrics Group, 2010), it does not disclose its industry classification scheme.  To construct a variable that is robust to concerns of industry effects, I create a second variable by 25  summing up the binary strength scores across all main categories except Corporate Governance and Products, and call this variable CSRPure.   Appendix A1 shows the distribution of CSRPure (Panel 1) and CSRTotal (Panel 2); both the variables are winsorized at 1st and 99th percentile.  Both CSRPure and CSRTotal vary between 0, which indicates no significant social strength, to 10, which indicates significant strengths in at least 10 social (or environmental) indicators.  The distributions of CSRPure and CSRTotal are right tailed and look almost identical; more than 50% of the sample firms have a CSR score of 0, roughly 23% of the observations have a CSR score of 1, and there is significant variation in the top quartile. I reach the same conclusions for all my tests using both CSRPure or CSRTotal. To avoid repetition, I only report the results using CSRPure. Restatement Severity Controls: To properly examine the impact of CSR on the market reaction to restatements, I first need to control for the restatement severity variables.  The restatement literature has identified important determinants of market reaction to restatement news (Badertscher, Hribar, & Jenkins, 2011; Files, Swanson, & Tse, 2009; Hennes et al., 2008; Palmrose et al., 2004).  I control for the total dollar amount being restated (EM), and the duration of the period being restated (DURATION).  Following prior research (Badertscher et al., 2011; Hennes et al., 2008), I classify a restatement as MATERIAL if the restatement disclosure indicates fraud, or the restatement involves an SEC investigation, or a class-action lawsuits is filed within 1 year of the restatement announcement.  Prior research shows that market reaction to restatement is more negative for restatements of revenue and core-expenditure (Palmrose et al., 2004; Scholz, 2008). To control for the type of accounts being restated, I include indicator variables REVENUE and CORE-EXP.  I also include two additional severity proxies available from AA database.  First, restatements that alter the previously issued auditor opinion on the 26  internal controls of the restating firm are likely to be more serious; so I include the indicator variable WEAKNESS.  Secondly, the market should react more negatively to restatements that trigger violations of the restating firms’ loan-covenants; so I include the indicator variable COVENANT. CSR Controls: Prior literature shows that the CSR performance of companies are related to other characteristics of the firm, and these firm characteristics may also explain the market reaction to restatements.  Controlling for such firm characteristics will help me tease out the effect of reputation with the non-equity stakeholders over and above the effects of these firm attributes.  Waddock & Graves (1997) show that the size, profitability, and leverage are determinants of CSR activities.7  Prior research shows that CSR performance should be correlated with R&D expenditure and advertising intensity (Lys et al., 2015), and with the market-to-book ratio (Orlitzky et al., 2003). CSR performance also should be correlated with the level of cash, and cash from operations (Campbell, 2007). Accordingly, I control for size (LOGSIZE), profitability (ROA), leverage (LEV), book-to-market (BP), R&D expenditure (RD), advertising intensity (AD), level of cash (CASH), and cash from operations (CF). General Controls: Prior researchers have included general control variables in their model in addition to restatement severity variables to control for the general information environment of the restating firms.  Some of these general control variables (like LOGSIZE and BP) are already included in my model as part of the CSR controls.  Following prior research (Badertscher et al., 2011; Burks, 2011; Palmrose et al., 2004), I also control for the volatility index (VIX), cumulative prior return (RET) starting from 120 trading days before the restatement                                                           7 Following prior literature in CSR, I measure size as the natural log of market capitalization of the firm.  An alternative to this proxy would be the natural log of total assets.  My results are not sensitive to the choice of the size proxy. 27  announcement to 2 trading days before the announcement, number of analysts (NANALYSTS), and institutional ownership (INST_OWN).  To control for earnings announcements that coincide with the restatement announcements, I include an indicator variable (UEDUMMY), and the earnings surprise variable (UE).  UE is calculated as earnings surprise scaled by stock price 2 days before the restatement announcement.  UE is set to 0 for observations without an earnings announcement within the 3-day-window of the restatement announcement.      Management Entrenchment controls:  To proxy for management entrenchment, I use two lagged insider-ownership variables.  I include CEO_OWN, defined as the percentage of the shares outstanding owned by the CEO.  I also use INSIDER_OWN, defined as the percentage of the shares outstanding that is owned collectively by all the other top managers (excluding the CEO).  The decision to discipline managers can also be influenced by the corporate governance.  KLD data provides positive and negative governance indicators.  Since prior literature recommends against combining positive and negative indicators available from KLD database (Mattingly & Berman, 2006), I control for GOV_STR and GOV_CON separately. Type of News Controls: For my hand collected sample, I am able to collect the prompter of the restatement news.  I use indicator variables to indicate if the restatement news was prompted by the company (COMP-PROMPT), by the SEC (SEC-PROMPT), or by the auditor (AUDIT_PROMPT).  Prior research shows that the salience of the restatement announcement can impact the severity of the initial market reaction (Files et al., 2009). So, I code if the restatement news was salient.  I use the dummy variable SALIENCE to indicate whether the restatement disclosures mention restatement (or similar words) in the headline of the news rather than mentioning restatement in the body or footnote sections. 28   Industry and Year Fixed Effects: For my analyses, I include industry fixed effects since the disclosure requirements, regulatory oversight, environmental impact, and other forces that can affect the level of CSR performance are different across industries (Karpoff et al., 2005; Waddock & Graves, 1997). 8  I also control for year fixed effects since the CSR activities and reporting have become more popular in the last decade (KPMG, 2011).  Another reason for using year fixed effects is that the market reaction to restatements show a positive trend (becoming less negative) over the years (Burks, 2011).  Moreover, prior research investigating the market reaction to restatements document a spike in restatement activities during the period 2005 and 2006 mainly triggered by the SEC’s guidance about lease accounting (Scholz, 2008).  Market reaction to lease accounting issues is milder during this period.  The year fixed effects should be adequate to control for these yearly variations.  2.5. Empirical Results  2.5.1. Descriptive Statistics Table 2.2 Panel B shows that the market reaction to restatements varies considerably depending on the severity of restatements and the type of disclosures.  For official announcements, the mean market reaction to restatements involving revenue recognition issues is −4.2%, while the mean market reaction to restatements due to reclassification is only −0.6%; the restatements classified as MATERIAL invoke an average market reaction of −5.6%; the average market reaction to all restatements is −1.6%.  For the sub-sample with hand collected first                                                           8The restatement literature does not employ industry fixed effects (a handful of papers use just one indicator variable to identify financial firms (SIC 6000-6800).  CSR literature generally uses broad industry classification schemes. To be consistent with existing CSR literature, I also use a broad classification scheme, the Fama-French 12 industry model.  My conclusions remain unchanged if I use the Fama-French 48 industry model. 29  revelation dates, the average market reaction to all observations is −4.3%; the mean market reaction to REVENUE restatements is −6.9% and the mean market reaction to MATERIAL restatements is −10.1%.  Table 2.3 provides the descriptive statistics for the final sample used for testing hypotheses 1, 2, and 3 of this essay.  With the exception of indicator variables, all variables are winsorized at 1% and 99% to minimize the impact of extreme observations.  On aggregate, restatements have a mean (median) market reaction of −1.6% (−0.7%), 1st quartile of CAR is −3.7% while the 3rd quartile is 1.6%.  My findings are consistent with prior research (Scholz, 2008), which documents that more than 30% of the restatement announcements have positive market reaction.9  The results suggest that the investors do not get alarmed by all restatements.  ∆EARN has a mean of −0.013, which means that following restatement, the average restating firm experiences a substantial decrease in earnings amounting to 1.3% of the firm’s total assets.  ∆EARN_UE has a mean (median) of −0.031 (−0.002).  TURNOVER has a mean of 0.143, which means that 14.3% of the sample firms fire the CEO (or CFO) in the 6 month period following restatements.  The variable of interest in this paper is the CSR, which has a mean (median) of 1.097 (0).   Table 2.3 also provides descriptive statistics for the restatement severity variables. The mean of the indicator variables denotes the percentage of firms that fall under the categories specified by the variables.  Looking at the indicator variables, 20.2% of restating firms are classified as MATERIAL, 18.7% of the sample firms involve revenue recognition issues, and 34.6% restate core-expenditure accounts.  Mean (median) EM is 0.013 (0.001), showing that the                                                           9 This is not unexpected, since cumulative abnormal return is determined by all the contemporaneous events during the restatement announcement window, and the negative effect of some individual restatement observations would naturally be negated by contemporaneous positive news (such as earnings announcement). 30  average restatement adjusts the earnings downward by a substantial amount (1.3% of the total assets). It seems a majority of restatements involve smaller dollar amounts (P50 and P75 are .001 and .008), but the mean result is driven by high impact restatements at the top quartile.  The mean (median) duration of the restating period is 2.463 (2) years.  Moving down the table, 9.4% of the sample observations involve WEAKNESS while 1.8% of restatements trigger violation of existing loan covenants. The mean (median) market capitalization of my sample firms is $3963.8 ($1205.9) million, suggesting that I am investigating the market reaction to restatements made by fairly large firms.  The mean (median) leverage is 24.8% (21.4%) of total assets.  Similarly, the mean (median) ROA is .014 (.025), the mean (median) CF is 0.073 (0.069), and mean (median) CASH is 0.170 (0.098), indicating that the average sample firm is marginally profitable, and has positive cash-flows and cash-balance.  Of the sample firms, 3.3% have negative book value, and the mean (median) Book-to-Price ratio is 0.551 (0.479).  Results also indicate that the average sample firm spends 7.3% of net sales on research and development (RD), and 1.2% of net sales on advertisements (AD).  The distributions of both R&D and advertisement spending are right tailed, with observations from the top quartile driving the mean results.  Results from the general control variables indicate that for 30.9% of my observations, there is an earnings announcement within the 3-day-window of the restatement announcement.  The average unexpected earnings (UE) is −0.123.  The mean (median) number of analysts covering the restating firms is 9.1 (8).  On average, the institutional owners own 71.5% of the restating firm. The results from the Management Entrenchment control variables reveal that the mean (median) GOV_STR is 0.132 (0), and the mean (median) GOV_CON is 0.397 (0).  This indicates that most of the sample firms do not have noteworthy governance strengths or 31  weaknesses.  For 1826 observations in my sample, I can obtain insider ownership data. The average CEO owns 1.2% of the restating firm, with the median holding being 0%.  On average, the remaining insiders collectively own 4.3% of the restating firm, with the median collective ownership being 0.9%. Table 2.4 presents the correlations table for the final sample used in my main analysis.  As expected, CSR has a significantly positive correlation (0.051) with the market reaction.  All the restatement severity variables are significantly negatively correlated with the market reaction except CORE-EXP, with MATERIAL having the most negative correlation (-0.269).  The market reaction to restatement is also positively correlated with ROA, CF, LEVERAGE, and AD.  Counterintuitively, balance sheet cash position (CASH) is negatively correlated with market reaction.  Looking at the other dependent variables, CSR has a significantly positive correlation (0.050) with ∆EARN.  CSR has a negative but statistically insignificant correlation with TURNOVER.  Since some of the control variables are also significantly correlated with CSR, an investigation of CSR’s impact on the market reaction to restatement needs to control for them.  Multivariate regression analysis can shed light on the hypothesized relationships in this paper.  2.5.2. Regression Analysis Testing Hypothesis 1: To test hypotheses 1(a) and 1(b), I use the following specification: 𝑪𝑨𝑹 =  𝜶𝟎 +  𝜶𝟏𝑪𝑺𝑹 + 𝜶𝟐𝑪𝑺𝑹 ∗ 𝑴𝑨𝑻𝑬𝑹𝑰𝑨𝑳 + (𝒓𝒆𝒔𝒕𝒂𝒕𝒆𝒎𝒆𝒏𝒕 𝒔𝒆𝒗𝒆𝒓𝒊𝒕𝒚 𝒄𝒐𝒏𝒕𝒓𝒐𝒍𝒔)  +  (𝑪𝑺𝑹 𝒄𝒐𝒏𝒕𝒓𝒐𝒍𝒔) + (𝒈𝒆𝒏𝒆𝒓𝒂𝒍 𝒄𝒐𝒏𝒕𝒓𝒐𝒍𝒔) +(𝒊𝒏𝒅𝒖𝒔𝒕𝒓𝒚 𝑭𝑬𝒔) + (𝒚𝒆𝒂𝒓 𝑭𝑬) + 𝜺             (2.1.1) 32  Hypothesis 1(a) predicts that 𝜶𝟏 > 𝟎 and hypothesis 1(b) predicts that 𝜶𝟐 > 𝟎.  Table 2.5 reports the estimation of the reduced and full versions of equation (2.1.1). I provide 4 versions of the model, starting with column 1, which only contains the CSR score along with industry and year fixed effects.  I then add the CSR controls and the restatement severity controls to the model in column 2. Next, I add the general controls in column 3.  The results from columns 1, 2, and 3 are consistent with hypothesis 1(a).  The coefficient of CSR is 0.002 and statistically significant at the .01 level.  The share value protection afforded by an additional unit of CSR strength is economically meaningful.  Recall that the mean market reaction is −1.6% for the full sample.  A one unit (one standard deviation) increase in CSR can mitigate the market reaction to restatements by 0.2% (0.3%).   The results in the full specification without the interaction term (column 3) show that restatements involving MATERIAL restatements trigger significantly more negative (−4.1%) market reaction.  The revenue restatements also elicit 2.1% more negative market reaction and the core-expenditure restatements induce 0.7% more negative market reaction compared to the other observations in the same industry and year.  The dollar amount of restatement (EM) has a large negative coefficient. The coefficients on COVENANT and WEAKNESS are negative and significant in some of the specifications.  Among the CSR control variables, cash-flow (CF) and CASH are not statistically significant, but research and development (RD) and advertisement expenditure (AD) have significantly positive coefficients.  Moving on to general controls, earnings surprise (UE) has a significantly positive coefficient while Volatility Index (VIX) has a significantly negative coefficient.    The model in column 4 adds the interaction term of CSR and MATERIAL into the model used in column 3. The coefficient of CSR*MATERIAL is 0.005 and statistically significant at 33  .01 level. Although the coefficient on CSR becomes insignificant, the coefficients for (CSR + CSR* MATERIAL) are 0.005 and significant at .01 level [F stat =7.49]. The results are consistent with the prediction that CSR mitigates the negative market reaction to restatements, and CSR’s impact becomes more prominent for restatements involving greater uncertainty.  The coefficient of the interaction term is economically significant.  MATERIAL restatements can elicit a market reaction of −4.8% below the other observations in the same industry and year.  For MATERIAL restatements, a one unit of CSR strength can increase the market reaction by 0.5%.  Testing hypothesis 1 (Robustness Analysis):  Table 2.6 presents two different specification checks for the market reaction test.  First, since CSR has a very high correlation (0.498) with LOGSIZE, and LOGSIZE has a non-linear positive association with the market reaction to restatements (Spearman correlation is statistically significant at .1 level), I perform a size-matched analysis to ensure that results cannot be explained by any size based argument; for example, perhaps the largest firms in my sample have a fundamentally different information environment, and they may have a monopoly over high CSR ratings.   To construct a size-matched sample, I first partition my main sample between firms that have CSR ratings of zero and firms that have CSR ratings greater than zero.  For each positive-CSR firm, I pick the closest size-matched firm from the zero-CSR sample without replacement.  A matching zero-CSR firm has to be within 70%-130% of the market capitalization of the positive-CSR firm.  This matching process yields 1082 restatement observations (581 positive-CSR firms and 581 zero-CSR firms).  Untabulated results show that the mean market capitalization ($2564.832) for the zero-CSR firms is not statistically different from that of the positive CSR firms ($2599.087).  34  The first four columns of Table 2.6 report the estimation of the reduced and full versions of equation (2.1.1) using the size matched sample.  As in Table 2.5, I start with column 1, which only contains CSR score along with industry and year fixed effects.  I then add the restatement severity controls and CSR controls to the model in column 2.  Next, I add the general controls in column 3.  The results from columns 1, 2, and 3 are consistent with hypothesis 1(a).  The share value protection afforded by an additional unit of CSR strength is economically meaningful.  Untabulated results show that the mean market reaction is −1.5%.  A one unit of CSR can increase the market return by an amount of 0.3%.  The control variables behave similarly in tests using the main sample and the size-matched sample, but dollar amount of restatement (EM), COVENANT, and WEAKNESS become insignificant, while coefficient on ROA becomes positive and significant. Model 4 in Table 2.6 adds the interaction term of CSR and MATERIAL into the model used in column 3. The coefficient of CSR* MATERIAL is 0.010 and significant at .01 level. Although the coefficient on CSR becomes insignificant after adding the interaction term, the coefficients for (CSR + CSR* MATERIAL) are 0.011 and significant at .01 level [F stat =7.26].  The result supports hypothesis 1(b). The coefficient of the interaction term is quite large and economically meaningful. MATERIAL restatements can elicit a market reaction of −5.1% below the other restatements in the same industry and year.  For MATERIAL restatements, one unit CSR strength can mitigate the negative market reaction by 1.1%.  The market reaction tests using the size-matched sample support the same conclusions as that using the full sample. Results from Table 2.5 and the first four columns of Table 2.6 cannot rule out the possibility that the market reaction to restatements is less negative for high-CSR firms because they are more likely to reveal their restatement news before the official announcement date 35  captured by the Audit Analytics database.  To address concerns that the restatement news may systematically leak early for high-CSR firms causing less negative market reaction to official announcements, I test hypothesis 1(a) and 1(b) using my hand collected sample of first revelation dates.  Since the hand collected sample is a sub-sample comprising of the more serious restatements from the original sample, the distribution of the CAR is shifted to the left (more negative) compared to that of the original sample (see Table 2.2 Panel B).  Untabulated results show that the mean (median) market reaction to restatement in the sub-sample is −4.3% (−1.7%).  Comparing my manually-collected dates with the AA filing dates shows that the official announcement dates and the first revelation dates are the same for 540 (74%) restatements, and they are different by just one day for 32 (4%) restatements.  The first revelation dates vary from the official filing dates by more than one day for 22% of the sub-sample.   For the hand collected sub-sample, 29.7% of restatements are MATERIAL, and 35.2% of the observations involve revenue recognition issues.  The remaining severity proxies, firm characteristics, and general control variables are similar across the original sample and the subsample.  Untabulated results show that 74.7% of the restatement announcements are prompted by the company, 12.6% of them are prompted by the SEC, and only 4.0% of the restatement news are prompted by the auditors.  Coincidentally, exactly 50% of the restatement disclosures are SALIENT.  To test hypothesis 1(a) and 1(b) using market reaction to first revelation dates, I use the following specification: 𝑪𝑨𝑹 =  𝜶𝟎 +  𝜶𝟏𝑪𝑺𝑹 + 𝜶𝟐𝑪𝑺𝑹 ∗ 𝑴𝑨𝑻𝑬𝑹𝑰𝑨𝑳 + (𝒓𝒆𝒔𝒕𝒂𝒕𝒆𝒎𝒆𝒏𝒕 𝒔𝒆𝒗𝒆𝒓𝒊𝒕𝒚 𝒄𝒐𝒏𝒕𝒓𝒐𝒍𝒔)+  (𝑪𝑺𝑹 𝒄𝒐𝒏𝒕𝒓𝒐𝒍𝒔) + (𝒕𝒚𝒑𝒆 𝒐𝒇 𝒓𝒆𝒔𝒕𝒂𝒕𝒆𝒎𝒆𝒏𝒕 𝒏𝒆𝒘𝒔 𝒄𝒐𝒏𝒕𝒓𝒐𝒍𝒔)  +(𝒈𝒆𝒏𝒆𝒓𝒂𝒍 𝒄𝒐𝒏𝒕𝒓𝒐𝒍𝒔) +  (𝒊𝒏𝒅𝒖𝒔𝒕𝒓𝒚 𝑭𝑬𝒔) +   (𝒚𝒆𝒂𝒓 𝑭𝑬) + 𝜺            (2.1.2) 36  The last four columns of Table 2.6 reports the estimation of the reduced and full versions of equation (2.1.2).  I provide four versions of the model, starting with column 5, which only contains CSR score along with industry and year fixed effects.  I then add the restatement severity controls, the CSR controls, and type-of-news controls to the model in column 6. Next, I add the general controls in column 7.  The results from columns 5, 6, and 7 are consistent with hypothesis 1(a).  The share value protection afforded by an additional unit of CSR strength is economically meaningful.  The mean market reaction in the subsample is −4.3%.  Results from column 7 suggest that one unit of CSR can mitigate the negative market reaction by 0.4%.   Control variables behave in similar ways as they do in the main test in Table 2.5 with some notable differences.  The coefficient on REVENUE is negative but insignificant, while the coefficients for LOGSIZE and ROA are positive and significant.  Consistent with prior research (Files et al., 2009), I find that SALIENCE has a significantly negative coefficient, suggesting that more salient news trigger more negative market reactions.  The prior literature does not provide clean guidance about the predicted impact of prompters of news, and not all the researchers control each of these 3 prompters (Badertscher et al., 2011; Burks, 2011; Files et al., 2009).  In my analysis, coefficients on COMP-PROMPT, SEC-PROMPTS, and AUDIT-PROMPT are not statistically different from zero.  Model 8 in Table 2.6 adds the interaction term of CSR and MATERIAL into the model used in column 7. The coefficient of CSR* MATERIAL is 0.009 and significant at .01 level. Although the coefficient on CSR becomes insignificant after adding the interaction term, the coefficients for (CSR + CSR* MATERIAL) is 0.009 and significant at .01 level [F stat =11.89]. The coefficient of the interaction term is quite large and economically meaningful. MATERIAL restatements can elicit a market reaction of −7.9%.  For MATERIAL restatements, one unit CSR 37  strength can increase the market reaction by 0.9%.  The results based on the market reaction to first revelation dates reach exactly the same conclusions as the results based on the market reaction to official filing dates.  Results from Tables 2.5 and 2.6 provide robust evidence in support of hypotheses 1(a) and 1(b). Testing Hypothesis 2: Prior literature does not offer guidance on specific determinants of future profitability following restatement.  The economic forces driving profitability are numerous and can change over time.  For consistency, I empirically test the predictions about future profitability of restating firms using the same set of control variables that were used for the market reaction test.  The model is quite exhaustive and controls for size, ROA, leverage, cash flow and a host of other characteristics that can determine future profitability following restatement. To test hypotheses 2(a) and 2(b), I use the following specification: {∆𝑬𝑨𝑹𝑵𝑜𝑟∆𝑬𝑨𝑹𝑵_𝑼𝑬 =  𝜶𝟎 +  𝜶𝟏𝑪𝑺𝑹 + 𝜶𝟐𝑪𝑺𝑹 ∗ 𝑴𝑨𝑻𝑬𝑹𝑰𝑨𝑳   +  (𝒓𝒆𝒔𝒕𝒂𝒕𝒆𝒎𝒆𝒏𝒕 𝒔𝒆𝒗𝒆𝒓𝒊𝒕𝒚 𝒄𝒐𝒏𝒕𝒓𝒐𝒍𝒔) +  (𝑪𝑺𝑹 𝒄𝒐𝒏𝒕𝒓𝒐𝒍𝒔)                 +(𝒈𝒆𝒏𝒆𝒓𝒂𝒍 𝒄𝒐𝒏𝒕𝒓𝒐𝒍𝒔) + (𝒊𝒏𝒅𝒖𝒔𝒕𝒓𝒚 𝑭𝑬) + (𝒚𝒆𝒂𝒓 𝑭𝑬) +  𝜺 (2.2) Hypothesis 2(a) predicts that 𝜶𝟏 > 𝟎 and hypothesis 2(b) predicts that 𝜶𝟐 > 𝟎.  Table 2.7A report the estimation of equation (2.2) using the dependent variable ∆EARN. Columns 1 and 3 report the estimation of equation (2.2) without the interaction term under different settings.  Column 1 reports results from the test using all observations with one year ahead earnings available from COMPUSTAT.  In column 1, the coefficient on CSR is 0.002 and is significant at .05 level.  CSR has an economically meaningful impact on the change in earnings; recall that the mean decrease in earnings is 1.3% of total assets; a one unit increase in CSR can increase the earnings by 0.2% of total assets.  Results suggest that CSR has a considerable economic impact in determining a firm’s operational profitability following restatement. 38  Column 3 of Table 2.7A repeats the analysis done in column 1 using ∆EARN_2ndYr, defined as earnings changes calculated using two-year ahead earnings of the firms.  This setting allows for a longer time to pass between the date of the restatement and the date when future earnings are measured.  The flip side is that results are more likely to be affected by economic forces not controlled for in my model.  The coefficient of CSR is 0.003 and statistically significant at .05 level.  Results from columns 1 and 3 of Table 2.7A support hypothesis 2(a). Table 2.7A columns 2 and 4 report the estimation of equation (2) with the interaction term using the two settings described above.  In column 2, the coefficient on CSR is 0.001 and the coefficient on CSR*MATERIAL is 0.002, but none of them are statistically significant.  However, the coefficients for (CSR+CSR*MATERIL) are 0.003, and statistically significant at .10 level.  Column 4 results are similar to that in column 2.  In column 4, using the long term change-in-earnings setting, the individual coefficients on CSR and CSR*MATERIAL are not statistically different from zero, but the coefficients for (CSR +CSR*MATERIAL) are 0.05, and are statistically significant at .05 level.  The results from columns 2 and 4 support hypothesis 2(a), but fail to support hypothesis 2(b).10 Tests using ∆EARN reveal that high-CSR firms fare better than low-CSR firms in the post-restatement period, but they cannot rule out the possibility that the high-CSR firms were expected to perform better than the low-CSR firms independent of their restatement announcements.  To verify that accounting restatement caused the sample firms to experience unexpected operational difficulty, and that CSR reputation could mitigate this difficulty, I                                                           10 Fewer than 5% of the restatement firms have missing values for the next year’s earnings, and fewer than 10% of the restatement firms have missing values for the two-year ahead earnings.  In my main analysis, I define ∆EARN = − EARNt-1 for firms that fail to report next year’s earnings.  My results are unchanged when I set ∆EARN=0 (optimistic assumption) or ∆EARN= −EARN t-1  − Bookvalue t-1 (pessimistic assumption), or when I remove such observations from the analysis.  Results using ∆EARN_2ndYr are also robust for all the different type of specification checks performed on ∆EARN. 39  perform tests of hypotheses 2(a) and 2(b) using ∆EARN_UE.  ∆EARN_UE captures the deviation of firm’s earnings relative to earnings expectations that were formed prior to the restatement announcement.  Conceptually, the consensus-forecast formed prior to restatement event is representative of the earnings in a hypothetical state where the restatement does not take place, and ∆EARN_UE represents the damage caused by the restatement. Table 2.7B report the estimation of equation (2.2) using the dependent variable ∆EARN_UE.  Column 1 reports the estimation of the full model, and column 2 reports the estimation of the full model with the interaction term of CSR and MATERIAL. In Column 1, the coefficient on CSR is not significantly different from zero.  In column 2, the coefficient on CSR*MATERIAL is 0.009, and is statistically significant at .01 level.  Although the coefficient on CSR is insignificant, the coefficients of (CSR + CSR*MATERIAL) are 0.006 and statistically significant at .05 level.  The coefficient of the interaction term is economically significant.  To bring it to perspective, MATERIAL restatements can elicit a ∆EARN_UE that is −0.031 below the other restatement observations in the same industry and year.  For MATERIAL restatements, a one unit increase in CSR can raise ∆EARN_UE by 0.009.  Table 2.7B columns 3 and 4 report the estimation of equation (2.2) using the dependent variable ∆EARN_UE_ROBUST, which is calculated in the same way as ∆EARN_UE, but accounts for the possibility that the restatement news may leak earlier than the official filing date.  For ∆EARN_UE_ROBUST, the pre-restatement consensus-forecast is the mean of annual forecasts that were made at least 30 days prior to the official filing date.  While putting this restriction reduces the sample size, it minimizes the chances of pre-restatement consensus-forecast getting contaminated by forecasts that were made after the restatement news leak.  Column 3 reports the estimation of the full model, and column 4 reports the estimation of the full 40  model with the interaction term.  In column 1, the coefficient of CSR is −0.001 and not statistically different from zero.  In column 2, the coefficient on CSR*MATERIAL is 0.009 and is statistically significant at 0.05 level.  Although the coefficient on CSR is insignificant, the coefficients for (CSR+CSR*MATERIAL) are 0.006 and significant at 0.1 level.  CSR has an economically meaningful impact on ∆EARN_UE_ROBUST.  A restatement being MATERIAL causes the ∆EARN_UE_ROBUST to go down by −0.038.  In case of MATERIAL restatements, increasing CSR by one unit can increase ∆EARN_UE by 0.009.   For the analysis in Table 2.7B, I lose a large number of observations due to unavailability of forecasted earnings prior to restatement announcement.  The analysts’ choice not to cover a firm prior to restatement announcement is likely to be strategic.  Hence I am selecting the sub-sample for testing ∆EARN_UE based on non-random reasons, which can possibly introduce a sample selection bias.  I use Heckman’s (1979) two-stage approach to correct such a bias.  In the first stage, a probit analysis determines the likelihood for a restating firm to be selected in the sub-sample of 1232 observations with the required data.  This step generates a sample selection statistic, the Inverse Mills ratio (INV_MILLS).  In the second stage, I include INV_MILLS as a determinant of ∆EARN_UE.  I use an exclusion restriction in the first-stage selection model to lower the potential correlation between the errors in the selection model and the errors in the final model.  Ideally, an exclusion variable will impact the selection model, but will not influence the final model.  I choose a firm’s trading volume (VOLUME) and age (AGE) as two selection restriction variables since analysts are much more likely to cover established firms with higher trading volume, but trading volume and firm-age are unlikely to have significant impact on forecast accuracy and optimism.   41  Table 2.7.B columns 5 and 6 report the selection and final models of the Heckman two stage analysis.  In column 5, a statistically insignificant INV_MILLS suggests that sample selection bias is negligible in my setting.  Hence the unadjusted OLS estimates (column 2) should be quite similar to the adjusted OLS estimates (column 6).  In the final stage, the coefficient on CSR*MATERIAL is significant and positive.  But the coefficients on (CSR+ CSR*MATERIAL) are not statistically different from zero.  Overall the Heckman tests yield inconclusive results. Results presented in Tables 2.7A and 2.7B answer the same question using slightly different tests.  As previously discussed, tests based on ∆EARN_UE (Table 2.7B) are conceptually more refined, but their empirical implementation is more challenging.  Analyst forecasts can be mired with various biases and inaccuracies, making the consensus-forecast prior to restatement announcement a rather noisy proxy of the hypothetical earnings in a state where there were no restatement.  While the Heckman procedure suggests that the unadjusted OLS estimates are not significantly affected by any sample selection bias, the adjusted estimates do not support hypothesis 2. I acknowledge the limitations of the results in Table 2.7B.  Tests using ∆EARN support hypothesis 2(a) but fail to support hypothesis 2(b).  Tests using ∆EARN_UE show some support for both hypotheses 2(a) and 2(b).  Overall, there is substantial support for hypothesis 2(a), and limited support for hypothesis 2(b). Testing Hypothesis 3:  Hennes et al. (2008) use CEO ownership in the restating company to control for management entrenchment when they investigate whether management turnover is higher following more material restatements.  For my management turnover test, I control for a set of management entrenchment variables (including CEO_OWN) in addition to the full model introduced earlier.  To test hypotheses 3(a) and 3(b), I use the following specification: 42  𝑻𝑼𝑹𝑵𝑶𝑽𝑬𝑹 =  𝜶𝟎 +  𝜶𝟏𝑪𝑺𝑹 + 𝜶𝟐𝑪𝑺𝑹 ∗ 𝑴𝑨𝑻𝑬𝑹𝑰𝑨𝑳 +  (𝑪𝑺𝑹 𝒄𝒐𝒏𝒕𝒓𝒐𝒍𝒔)+ (𝑴𝒂𝒏𝒂𝒈𝒆𝒎𝒆𝒏𝒕 𝑬𝒏𝒕𝒓𝒆𝒏𝒄𝒉𝒎𝒆𝒏 𝒄𝒐𝒏𝒕𝒓𝒐𝒍𝒔)+ (𝒓𝒆𝒔𝒕𝒂𝒕𝒆𝒎𝒆𝒏𝒕 𝒔𝒆𝒗𝒆𝒓𝒊𝒕𝒚 𝒄𝒐𝒏𝒕𝒓𝒐𝒍𝒔) + (𝒈𝒆𝒏𝒆𝒓𝒂𝒍 𝒄𝒐𝒏𝒕𝒓𝒐𝒍𝒔) +(𝒊𝒏𝒅𝒖𝒔𝒕𝒓𝒚 𝑭𝑬𝒔) + (𝒚𝒆𝒂𝒓 𝑭𝑬) + 𝜺             (2.3) Hypothesis 3(a) predicts that 𝜶𝟏 < 𝟎 and hypothesis 3(b) predicts that 𝜶𝟐 < 𝟎.  Table 2.8 columns 1 and 2 report the estimation of equation (2.3) using logistic regressions.  Column 1 shows the full model, while column 2 shows the full model with the interaction term.  In column 1, the coefficient of CSR is −0.024 but not statistically significant.   In column 2, the coefficient of CSR*MATERIAL is −0.168 and significant at 0.05 level.  Although the coefficient on CSR is insignificant, the coefficients for (CSR+CSR*MATERIAL) is -0.144 and significant .01 level [χ2(1) =8.43].  Due to the difficulty in interpreting the interaction term in a logistic regression, I investigate the impact of CSR on management turnover following MATERIAL restatements in a separate regression.  Column 3 of Table 2.8 reports estimation of the full model after requiring that the sample observations need to be MATERIAL restatements.  The coefficient of CSR is −0.158, and is significant at .01 level.  Marginal effects calculations show that keeping all the other variables at their mean value, a one unit increase in CSR from 0 to 1 can lower the probability of TURNOVER by 2.7%.  This is a large impact considering that the unconditional probability of TURNOVER following material restatements is 22.8% (untabulated).  Hence pre-existing CSR reputation can offer considerable job security to top management following material restatements. Looking at the coefficients of the control variables in column 1 and 2, CEO_OWN has a significantly negative coefficient (decreasing chances of turnover), while restatement severity variables like MATERIAL, EM, and DURATION have significantly positive coefficients 43  (increasing chances of turnover).  In column 3, using a smaller sample of only MATERIAL restatements, the coefficients on EM, DURATION, and CEO_OWN become insignificant, but coefficient on WEAKNESS becomes positive and significant.  This suggests that having internal control issues in cases of material restatements can be particularly detrimental to the job security of the top management.   Chakravarthy et al. (2014) uses turnover of all C-Suite officers as their measure of management turnover.  In Table 2.8 columns 4, 5, and 6, I repeat the analyses performed in columns 1,2, and 3 using an alternative definition of TURNOVER.  I set TURNOVER equal to one if any C-Suite officer resigns, retires, or is dismissed in the six month period after the restatement announcement, and zero otherwise.  In column 4, the coefficient on CSR is negative but insignificant.  In column 5, the coefficient on CSR*MATERIAL is −0.146 and significant at .05 level.  Column 6 reports the results for estimation using only MATERIAL restatements. The coefficient on CSR is −0.178 and significant at .01 level.  Marginal effects calculations reveal that keeping all the other variables at their mean value, a one unit increase in CSR rating from 0 to 1 can lower the probability of TURNOVER by 4.0%.  This value is sizable when compared to the unconditional probability of C-Suite executive turnover of 29.6% (untabulated) following MATERIAL restatements.   Results from Table 2.8. show that while CSR does not have a significant impact on TURNOVER following all restatements, CSR does have a large negative impact on TURNOVER following more material restatements.  Taken together, the results are consistent with hypotheses 3(a) and 3(b).   I have few observations per year due to limited availability of CSR data coverage in the earlier period of my sample.  For years 2000, 2001, and 2003, all restatement observations yield 44  TURNOVER=1 and hence they drop out of the logit regression.  For analysis using only MATERIAL restatements in column 3, observations in a couple of industries also drop out for the same reason.  To address concerns of truncated sample, I test hypothesis 3(a) and 3(b) without using year and industry fixed effects (untabulated), and reach the same conclusions.  Results from the I choose two salient metrics for my consequence analysis: the change in earnings, which is a summary measure of firm’s fundamental performance  2.5.3. Sensitivity Checks and Additional Analysis Controlling for Negative CSR Ratings:  I use positive CSR ratings (CSR Strengths) to proxy for firm’s reputation with its non-equity stakeholders.  Prior empirical research suggests that offsetting bad press due to corporate actions that are perceived to do ‘harm’ to society is one of the motivations behind engaging in CSR activities (Kotchen & Moon, 2012).  To tease out the effect of CSR performance that is attributable to such ‘greenwashing’ (or community-washing) strategies, I run all my tests after controlling for negative CSR ratings (CSR Concerns) available from the KLD database.  My results are qualitatively and quantitatively the same after controlling for the negative ratings in all my tests. Controlling for Corporate Governance:  Environmental, Social, and Governance (ESG) metrics are the three main areas of CSR reporting.  In this paper, I define CSR score as the summation of the environmental and social ratings of a firm, and use this variable as a proxy for the firm’s reputation with its non-equity stakeholders (See Appendix A).  For completeness, I run all my market reaction and change-in-earnings tests after controlling for governance Strengths (GOV_STR) and Concerns (GOV_CON).  All my results hold after controlling for governance variables.  Interestingly, the GOV_STR has a negative impact on the market reaction to 45  restatement announcement, consistent with my earlier arguments that signals of ex-ante investor-reputation would lead to greater negative surprise at the time of restatement announcement.  The management turnover tests reported in Table 2.8 already include the governance variables as part of the management entrenchment controls.  Including or excluding governance variables as controls does not change my conclusion about CSR’s impact on management turnover. Market Reaction Test (Placebo Analysis): The central point of this paper is that accounting restatements have reputational impacts, and CSR can save firm value in times of restatement by preventing the loss of ‘reputation capital’.  If this argument is correct, then CSR ratings should not influence market reaction to negative events that do not have a reputational impact—events that do not compromise the general integrity and reliability of a firm as a business partner.  Negative earnings surprises are events that generally trigger negative market reaction, but are unlikely to have reputational impact.  As placebo analysis, I investigate whether CSR has any effect on market reaction to negative earnings surprises.   In my main analysis, I study the market reaction to 1892 restatements announced by 1336 unique firms between the years 2000 and 2014.  During this time period, these 1336 firms experienced 21,757 negative earnings surprises11.  I regress the market reaction (CAR) on earnings surprise amount (UE), an indicator variable (LOSS) to indicate if there was a net loss in the quarter, CSR, CSR controls, and the general controls.  Untabulated results show that CSR ratings have no impact on the market reaction to negative earnings surprises.  The bottom decile of negative earnings surprise observations elicit large negative market reactions just like restatement events.  But CSR has no impact on market reaction to these extremely negative                                                           11 I define a negative earnings surprise as an event when a firm announces its quarterly earnings figure that falls below the earnings number reported in the corresponding quarter of the past year (quarter t-4).   46  (bottom decile) events.  The non-result in the placebo test provides one more layer of confirmation of the central argument of my paper.  2.5.4. Limitations and Ideas for Future Research In this paper, I use a firm’s CSR rating as a proxy for its reputation with non-equity stakeholders.  The paper could have benefitted from alternative proxies of reputation.  Unfortunately the data coverage for other commonly available reputation proxies are not as broad as that of CSR ratings, and as a result, such alternative reputation proxies are available for only a small number of restatement firms.  For example, ‘Fortune Most Admired Companies’ reputation index is based on the largest 1000 firms which tend to have fewer restatements (Scholz, 2008), while ‘Fortune Best Companies to Work For’ index is available for only 100 companies.  Future research can explore whether innovative proxies that capture a firm’s commitment to honor its contracts can mitigate the negative consequences of accounting restatement.   To bolster my market reaction argument, I investigate two salient post-restatement consequences, decrease in earnings and increase in management turnover, and find that CSR can partially mitigate these effects.  It would be interesting to explore whether firms with better CSR can recover their pre-restatement earnings level faster.  The restatement literature has identified many adverse consequences of accounting restatements.  Future studies can investigate whether CSR has a mitigating impact on other types of negative consequences.      47  2.6. Conclusion  I find that restating firms’ reputation with their non-equity stakeholders can explain the variation in the market reaction to their restatement announcements.  Investors react less negatively to accounting restatements if they observe that the restating firm has strong relationships with its non-equity stakeholders.  Moreover, in the post restatement period, firms with better reputation with their non-equity stakeholders experience smaller earnings decreases, and are less likely to resort to firing their top management.  The results suggest that accounting restatements can create uncertainty about the restating firms’ ability to continue doing business with their non-equity stakeholders under favorable terms, and this uncertainty is partly responsible for the large negative market reactions triggered by restatement announcements.  The results from this paper also provide evidence consistent with the argument that investors increase their reliance on alternative channels of information when the reliability of the financial statements becomes compromised.     48  Table 2.1.  Variable definitions Variables Definitions Dependent Variables CAR Market reaction, defined as the size adjusted cumulative abnormal return over three trading days (-1,0,1) relative to the restatement announcement ∆EARN Change in income before extra-ordinary items scaled by total assets from the year prior to the restatement announcement to the year of the restatement announcement: [(IB/AT)t − (IB/AT)t− 1]  ∆EARN_UE Deviation of actual earnings from the consensus-forecast prior to the restatement announcement, scaled by stock price at the end of the previous fiscal year:  [EPSt – Forecast(EPSt)] / PRCC_Ft–1 , where Forecast(EPSt) is the mean of annual forecasts made prior to restatement announcement.  EPS information is collected from IBES. TURNOVER Indicator variable equal to one if the CEO or CFO resign, retire, or get dismissed in the six month period following the restatement announcement, and zero otherwise, available from Audit Analytics (AA) Director and Officer Turnover database. Variable of Interest  CSR Sum of the positive social and environmental ratings of a firm in year t−1 where restatement announcement happens in year t, available from MSCI KLD database (See Appendix A1) Restatement Severity Controls MATERIAL Indicator variable equal to 1 if a) fraud is mentioned in the restatement disclosure (RES_FRAUD), b) the restatement is associated with an SEC Investigation (RES_SEC_INVESTIG), and/ or c) there is a class-action law-suit filed (DAYS_TO_SECURITIES_CLASS_ACTION) within one year of the restatement filing, and 0 otherwise, available from AA Non-Reliance (restatement) database. REVENUE Indicator variable equal to 1 if restatement involves revenue recognition issue and 0 otherwise, coded from AA database (See coding details in Table 2.2 footnote) CORE-EXP Indicator variable equal to 1 if restatement involves core expense accounts and 0 otherwise, coded from AA database (See Table 2.2 footnote) EM Earnings management, defined as the total dollar impact of the restatement on reported net income, divided by the total assets in the year t-1.  The dollar impact of restatement is available from Audit Analytics from 2002.  Before 2002, this value is estimated by taking the difference between the originally reported income and the most  49  Table 2.1 (Continued) Variables Definitions  recently available income from Capital IQ for each year of the restating period, and summing up the yearly earnings management amounts. DURATION Duration of the restating period, which is the gap between the restatement ending date and restatement beginning date, available from Audit Analytics. This variable is expressed in year COVENANT Indicator variable equal to 1 if restatement involves failure to disclose violations of loan covenant or similar contingencies, or if restatement involves improper calculation of capital adequacy ratios, and 0 otherwise, available from Audit Analytics WEAKNESS Indicator variable equal to 1 if the restatement triggers opinion of ineffective internal control and 0 otherwise, available from Audit Analytics CSR Controls  LOGSIZE Natural log of the firm’s market capitalization for fiscal year t-1 when the restatement happens in year t: [Log(CSHO*PRCC_F)] LEV The debt-to-total assets ratio, calculated as [(long-term debt + debt in current liabilities)/total asset] : [(DLTT+DLC)/AT] ROA Return on asset, calculated as net income /total asset): [(NI)/AT] BP Book-to-market ratio, calculated as (book value per share/year-end market price): [(CEQ/CSHO)/PRCC_F] NEGBV Indicator variable equal to one if he book value of the firm is negative in a given year CF Cash flow from operations, calculated as operating cash flow available from the cash flow statement divided by total assets: [OANCF/AT]  CASH Cash and cash equivalents divided by the total assets: [CHE/AT] RD Research and development expense divided by net sales: [XRD/AT]  AD Advertising expense divided by net sales: [XAD/AT] General Controls  UEDUMMY Indicator variable equal to 1 if the restating firm announces earnings within the 3 day window (-1,0,1) of the restatement announcement and 0 otherwise UE Unexpected earnings, defined as the earnings per share in quarter t minus earnings per share in quarter t-4, scaled by the stock price two days before the restatement announcement (earnings announcement in case of placebo analysis).  For restatement observation for which there  50  Table 2.1 (Continued) Variables Definitions  is no coinciding earnings announcement (UEDUMMY=0), this variable is set to 0. VIX Volatility index, measured as the mean value of the Chicago Board Options Exchange’s volatility index over the 3 day window around the restatement RET Prior return, defined as the size adjusted cumulative abnormal return over days (−120 , −2) relative to the restatement announcement NANALYST log of one plus the number of analysts providing earnings forecasts during the last month of the fiscal year from I/B/E/S Summary History data INST_OWN Percentage of shares outstanding owned by the institutional investors in the year t−1, available from the Thomson Reuters S34 file (the 13-F database) VOLUME The monthly volume of shares traded prior to the restatement announcement, scaled by total shares outstanding in that month. To ensure that the measure is not affected by the market unrest around restatement announcement, I measure volume on the latest month-end that is at least 60 days prior to the restatement announcement date.  AGE Natural log of one plus the number of years the firm has existed in CRSP database Management Entrenchment Controls (Applicable only for Turnover Analysis) CEO_OWN Percentage of shares outstanding owned by the CEO of the restating firm in the year (t−1), available from CapitalIQ for all the sample firms starting from 2004, and from ExecuComp for the largest 1500 firms in years prior to 2004 INSIDER_OWN Percentage of shares outstanding collectively owned by company insiders (excluding the CEO) in the year (t−1), available from CapitalIQ (2004 onward) and Execucomp (before 2004) GOV_STR Corporate Governance Total Strengths of a firm in year t−1 where restatement announcement happens in year t, available from MSCI KLD database (see Appendix A1 footnote) GOV_CON Corporate Governance Total Concerns of a firm in year t−1 where restatement announcement happens in year t, available from MSCI KLD database (see Appendix A1 footnote) Type of News Variables (Applicable only for Robustness Analysis) COMP-PROMPT Indicator variable equal to 1 if the company prompted the restatement and 0 otherwise SEC-PROMPT Indicator variable equal to 1 if the SEC prompted the restatement and 0 otherwise 51  AUDIT-PROMPT Indicator variable equal to 1 if the auditor prompted the restatement and 0 otherwise SALIENCE Indicator variable equal to 1 if the restatement or related words appeared in the headline and 0 if the restatement news was just embedded in company news/ disclosures   52  Table 2.2.  Sample Panel A: Sample Selection Sample for Main Analysis  Observations Restatements from Audit Analytics (AA)  14351 firms without COMPUSTAT coverage  −3574 firms without CRSP coverage  −4593 firms without MSCI KLD data coverage  −3308 duplicate obs; missing return, missing total assets, missing sales,   −122 small firms that do not meet MSCI market cap criteria   −483 income increasing restatements  −379 Final Sample  1892       Sub-Sample with Hand Collected Dates for Robustness Analysis  Observations Full Sample  1892 restatements of only non-core accounts12   −467 restatements due to reclassification reasons6  −417 restatements for which the actual news/ substantiating documentation could not be traced from SEC filing link provided by Audit Analytics,  or searching Factiva and company website  −278 Final Sample  730     Panel B: Market reaction to announcements of different categories of restatement     Full Sample Official Filing Dates  Sub-Sample First Revelation Dates Type of Accounts Being Restated6  Number of Obs  Mean  CAR  Number of Obs  Mean  CAR Revenue Recognition Issues  353  −0.042  257  −0.069 Restatement of core expenses  655  −0.017  473  −0.029 Non-core accounts  467  −0.004     Reclassification issues  417  −0.006     All Restatements  1892  −.016  730  −0.043          Severity of restatements  Number of Obs  Mean  CAR  Number of Obs  Mean  CAR Material Restatements   382  −0.056  217  −0.101 Immaterial restatements/ errors  1510  −0.006  513  −0.019 All Restatements  1892  −0.016  730  −0.043  The bolded abnormal returns are negative and statistically significant at .01 level                                                            12 AA database provides granular details of the accounts that are affected by a restatement. I follow Scholz (2008) to classify my sample restatements into four categories based on the type of accounts being restated: 1) revenue recognition, 2) core expense, 3) non-core items, and 4) reclassification. Appendix A of Scholz (2008) explains how to categorize AA database restatements into these four groups. 53  Table 2.3.  Descriptive statistics   N MEAN P25 P50 P75 Dependent Variables      CAR 1892 -0.016 -0.037 -0.007 0.016 ∆EARN 1884 -0.013 -0.031 -0.001 0.017 ∆EARN_UE 1232 -0.031 -0.018 -0.002 0.004 TURNOVER 1892 0.143 0.000 0.000 0.000 Variable of Interest      CSR 1892 1.097 0.000 0.000 1.000 Restatement Severity      MATERIAL 1892 0.202 0.000 0.000 0.000 REVENUE 1892 0.187 0.000 0.000 0.000 CORE-EXP 1892 0.346 0.000 0.000 1.000 EM 1892 0.013 0.000 0.001 0.008 DURATION 1892 2.463 1.000 2.000 3.249 COVENANT 1892 0.018 0.000 0.000 0.000 WEAKNESS 1892 0.094 0.000 0.000 0.000 CSR Controls      SIZE 1892 3963.8 540.6 1205.9 3313.1 ROA 1892 0.014 0.001 0.025 0.060 BP 1892 0.551 0.281 0.479 0.725 NEGBV 1892 0.033 0.000 0.000 0.000 LEVERAGE 1892 0.248 0.049 0.214 0.378 RD 1892 0.073 0.000 0.000 0.040 AD 1892 0.012 0.000 0.000 0.012 CFO 1892 0.073 0.027 0.069 0.122 CASH 1892 0.170 0.032 0.098 0.242 General Controls      UEDUMMY 1892 0.309 0.000 0.000 1.000 UE 1892 -0.123 0.000 0.000 0.000 VIX 1892 18.06 12.96 15.55 20.02 RET 1892 -0.015 -0.155 -0.018 0.112 NANALYST^ 1892 9.1 4.0 8.0 13.0 INST_OWN 1892 71.51% 58.44% 75.60% 89.55% Entrenchment Controls      CEO_OWN 1826 1.22% 0% 0% 0.20% INSIDER_OWN 1826 4.33% 0% 0.85% 3.75% GOV_STR 1892 0.132 0.000 0.000 0.000 GOV_CON 1892 0.397 0.000 0.000 1.000       ^unlogged (original) numbers for NANALYST are reported.    54  Table 2.4.  Correlations table Pearson correlations below the diagonal and spearman correlations above the diagonal (p-values reported in italics)    CAR ∆EARNTURN OVER CSRMATERIALREVENUECORE-EXP EMDURATIONCOVENANTWEAKNESSLOG SIZE ROALEVERAGE RD AD CF CASHCAR 1 0.079 -0.068 0.024 -0.209 -0.137 -0.025 -0.198 -0.061 -0.050 -0.116 0.040 0.060 0.018 -0.076 0.060 0.052 -0.055p 0.00 0.00 0.30 0.00 0.00 0.28 0.00 0.01 0.03 0.00 0.08 0.01 0.42 0.00 0.01 0.02 0.02∆EARN 0.127 1 -0.075 0.020 -0.062 -0.036 -0.014 -0.037 -0.024 0.001 -0.050 -0.006 -0.319 0.050 -0.007 -0.026 -0.078 0.016p 0.00 0.00 0.38 0.01 0.12 0.55 0.11 0.29 0.97 0.03 0.81 0.00 0.03 0.76 0.25 0.00 0.50TURNOVER -0.103 -0.073 1 -0.007 0.122 0.049 0.056 0.114 0.110 0.023 0.086 -0.010 -0.033 -0.042 0.085 0.028 -0.018 0.066p 0.00 0.00 0.77 0.00 0.03 0.02 0.00 0.00 0.33 0.00 0.65 0.15 0.07 0.00 0.22 0.43 0.00CSR 0.051 0.050 -0.012 1 0.072 0.007 -0.013 -0.010 0.015 -0.024 -0.049 0.413 0.048 0.036 0.022 0.075 0.044 0.012p 0.03 0.03 0.62 0.00 0.75 0.58 0.66 0.51 0.30 0.03 0.00 0.04 0.12 0.33 0.00 0.05 0.60MATERIAL -0.269 -0.048 0.122 0.073 1 0.185 0.046 0.268 0.164 0.078 0.136 0.097 -0.066 0.011 0.063 0.011 -0.066 0.030p 0.00 0.04 0.00 0.00 0.00 0.04 0.00 0.00 0.00 0.00 0.00 0.00 0.64 0.01 0.65 0.00 0.20REVENUE -0.168 -0.027 0.049 0.019 0.185 1 -0.349 0.188 0.063 0.055 0.148 0.007 -0.033 0.018 0.052 0.002 -0.013 0.021p 0.00 0.24 0.03 0.42 0.00 0.00 0.00 0.01 0.02 0.00 0.77 0.16 0.44 0.02 0.92 0.58 0.35CORE-EXP -0.010 -0.006 0.056 -0.049 0.046 -0.349 1 0.324 0.267 -0.009 -0.033 -0.006 0.056 -0.088 0.038 0.065 0.116 0.067p 0.65 0.78 0.02 0.03 0.04 0.00 0.00 0.00 0.69 0.15 0.79 0.02 0.00 0.10 0.00 0.00 0.00EM -0.190 0.010 0.121 -0.053 0.226 0.130 0.120 1 0.429 0.074 0.155 -0.086 -0.005 -0.095 0.128 0.045 0.126 0.152p 0.00 0.68 0.00 0.02 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.82 0.00 0.00 0.05 0.00 0.00DURATION -0.048 0.005 0.106 -0.007 0.148 0.016 0.308 0.291 1 0.054 0.014 0.060 0.068 -0.046 0.032 0.089 0.101 0.070p 0.04 0.82 0.00 0.77 0.00 0.49 0.00 0.00 0.02 0.55 0.01 0.00 0.05 0.17 0.00 0.00 0.00COVENANT -0.070 -0.019 0.023 -0.024 0.078 0.055 -0.009 0.056 0.079 1 0.063 -0.036 -0.051 0.040 -0.007 0.052 -0.009 -0.022p 0.00 0.41 0.33 0.30 0.00 0.02 0.69 0.01 0.00 0.01 0.12 0.03 0.08 0.77 0.02 0.68 0.33WEAKNESS -0.113 -0.031 0.086 -0.030 0.136 0.148 -0.033 0.038 -0.016 0.063 1 -0.101 -0.048 -0.024 0.025 -0.022 -0.070 0.024p 0.00 0.18 0.00 0.19 0.00 0.00 0.15 0.10 0.50 0.01 0.00 0.04 0.30 0.27 0.33 0.00 0.30LOGSIZE 0.035 -0.006 -0.014 0.498 0.117 0.009 -0.015 -0.078 0.071 -0.030 -0.094 1 0.223 0.148 -0.090 0.000 0.160 -0.086p 0.13 0.80 0.55 0.00 0.00 0.71 0.52 0.00 0.00 0.20 0.00 0.00 0.00 0.00 0.99 0.00 0.00ROA 0.054 -0.371 -0.033 0.046 -0.058 -0.060 0.058 -0.185 0.066 -0.051 -0.023 0.212 1 -0.236 -0.109 0.105 0.582 0.081p 0.02 0.00 0.16 0.05 0.01 0.01 0.01 0.00 0.00 0.03 0.32 0.00 0.00 0.00 0.00 0.00 0.00LEVERAGE 0.007 0.069 -0.041 0.042 0.000 0.010 -0.074 -0.042 -0.073 0.046 -0.027 0.086 -0.123 1 -0.271 -0.061 -0.127 -0.460p 0.77 0.00 0.07 0.07 1.00 0.67 0.00 0.07 0.00 0.05 0.25 0.00 0.00 0.00 0.01 0.00 0.00RD -0.006 -0.029 0.022 -0.049 0.021 0.001 0.003 0.135 -0.014 -0.011 -0.003 -0.121 -0.436 -0.062 1 -0.036 -0.047 0.486p 0.80 0.20 0.34 0.03 0.37 0.96 0.89 0.00 0.56 0.63 0.89 0.00 0.00 0.01 0.12 0.04 0.00AD 0.065 -0.023 -0.006 0.056 0.002 0.024 0.002 -0.036 0.033 0.037 -0.016 0.029 0.062 -0.017 -0.058 1 0.135 0.077p 0.00 0.32 0.80 0.02 0.92 0.30 0.94 0.11 0.15 0.11 0.49 0.21 0.01 0.47 0.01 0.00 0.00CF 0.030 -0.073 -0.023 0.042 -0.049 -0.037 0.112 -0.011 0.100 -0.024 -0.049 0.166 0.580 -0.117 -0.423 0.116 1 0.116p 0.19 0.00 0.32 0.07 0.03 0.11 0.00 0.64 0.00 0.30 0.03 0.00 0.00 0.00 0.00 0.00 0.00CASH -0.048 -0.017 0.065 -0.013 0.053 0.022 0.071 0.208 0.100 -0.032 0.026 -0.105 -0.085 -0.337 0.441 0.057 -0.044 1p 0.04 0.46 0.00 0.58 0.02 0.34 0.00 0.00 0.00 0.17 0.25 0.00 0.00 0.00 0.00 0.01 0.0655  Table 2.5.  Multivariate regression examining the effect of CSR on the market reaction to accounting restatements Announcement date is the official filing date  MODEL    (1)  (2)  (3)  (4) VARIABLES  Prediction   CAR         CAR        CAR        CAR                      CSR  +  0.002***  0.002***  0.002***  0.000 CSR*MATERIAL  +        0.005** Restatement Severity           MATERIAL  −    -0.042***  -0.041***  -0.048*** REVENUE  −    -0.022***  -0.021***  -0.022*** CORE-EXP  −    -0.006  -0.007*  -0.007* EM  −    -0.235***  -0.235***  -0.229*** DURATION  ?    0.001  0.001  0.001 COVENANT  −    -0.022  -0.021*  -0.020 WEAKNESS  −    -0.013**  -0.012**  -0.013** CSR Controls           LOGSIZE  +    0.001  0.000  0.000 ROA  +    0.002  -0.001  -0.002 BP  ?    0.004  0.005  0.005 NEGBV  ?    -0.001  -0.000  -0.000 LEV  ?    -0.000  -0.001  -0.001 RD  ?    0.014***  0.014***  0.015*** AD  ?    0.180**  0.159**  0.158** CF  ?    0.029  0.028  0.031 CASH  ?    -0.004  -0.005  -0.005 General Controls           UEDUMMY  ?      0.003  0.002 UE  +      0.002***  0.002*** VIX  −      -0.001**  -0.001** RET  +      0.011  0.010 NANALYSTS  ?      0.002  0.002 INSTIT_OWN  ?      -0.001  -0.001            CONSTANT  ?  -0.036*  -0.025  -0.006  -0.012            Observations    1,892  1,892  1,892  1,892 R-squared    0.0278  0.1345  0.1407  0.1439 Industry FE    YES  YES  YES  YES Year FE    YES  YES  YES  YES Errors       Clustered  Clustered  Clustered  Clustered Error are clustered across firms and years.   *** p<.01, ** p<.05, * p< .1  (two tailed tests) Model (4) Test of H0, CSR+ CSR*MATERIAL= 0:       F stat =   7.49,       p-value <.01  56  Table 2.6.  Multivariate regression examining the effect of CSR on the market reaction to accounting restatements (Robustness)     Official filing dates (size matched sample)  First revelation dates (hand collected sub-sample) MODEL    (1)  (2)  (3)  (4)  (5)  (6)  (7)  (8) VARIABLES  Predic  CAR  CAR  CAR  CAR  CAR  CAR  CAR  CAR                            CSR  +  0.003**  0.003***  0.003**  0.001  0.006***  0.004***  0.004***  0.000 CSR*MATERIAL  +        0.010**        0.009*** Restatement Severity                   MATERIAL  −    -0.041***  -0.041***  -0.051***    -0.068***  -0.068***  -0.079*** REVENUE  −    -0.022***  -0.021***  -0.021***    -0.015  -0.013  -0.013 CORE-EXP  −    -0.012**  -0.012*  -0.012*         EM  −    -0.062  -0.069  -0.065    -0.171**  -0.179***  -0.171** DURATION  ?    0.001  0.001  0.001    0.001  0.001  0.001 COVENANT  −    -0.014  -0.016  -0.015    -0.017**  -0.017**  -0.020** WEAKNESS  −    -0.002  -0.001  -0.003    -0.073***  -0.070***  -0.069*** CSR Controls                   LOGSIZE  +    0.001  0.001  0.001    0.008**  0.006***  0.006** ROA  +    0.045*  0.042*  0.044*    0.099*  0.096*  0.091* BP  ?    0.013**  0.014**  0.015**    0.023**  0.026***  0.028*** NEGBV  ?    -0.025**  -0.024*  -0.024*    0.016  0.016  0.016 LEV  ?    0.013  0.014  0.013    0.018  0.020*  0.021** RD  ?    0.024**  0.024**  0.024**    0.085***  0.080***  0.078** AD  ?    0.232**  0.224**  0.221**    -0.020  -0.058  -0.057 CF  ?    -0.001  -0.002  0.000    0.023  0.020  0.027 CASH  ?    -0.019  -0.018  -0.016    0.001  0.004  0.005 Type of News                   COMP-PROMPT  ?            0.007  0.007  0.008 SEC-PROMPT  ?            0.017  0.011  0.016 AUDIT-PROMPT  ?            0.007  0.004  0.010 SALIENCE  −            -0.015**  -0.015**  -0.016*                    General Controls    NO  NO  YES  YES  NO  NO  YES  YES Industry FE    YES  YES  YES  YES  YES  YES  YES  YES Year FE    YES  YES  YES  YES  YES  YES  YES  YES Errors       Clustered   Clustered   Clustered   Clustered   Clustered   Clustered   Clustered   Clustered Observations    1,082  1,082  1,082  1,082  730  730  730  730 R-squared    0.0482  0.1388  0.1468  0.1529  0.0699  0.2741  0.2837  0.2901 Error are clustered across firms and years.  *** p<.01, ** p<.05, * p< .1 (two tailed tests). Model (4) Test of H0, CSR+ CSR*MATERIAL= 0:   F stat =7.26,    p-value <.01                 Model (8) Test of H0, CSR+ CSR*MATERIAL= 0:   F stat =11.89,   p-value <.0157   Table 2.7A. Multivariate regression examining the effect of CSR on decrease in earnings following restatement   Error are clustered across firms and years.  *** p<.01, ** p<.05, * p< .1 (two tailed tests)  ^∆EARN_2ndYr is defined using two year ahead earnings [∆EARN = EARNt+1 −  EARNt−1]. See Table 2.1 for details.  Model (2) Test of H0: CSR+ CSR*MATERIAL= 0,       F stat =   3.25               p-value <.1  Model (4) Test of H0: CSR+ CSR*MATERIAL= 0,       F stat =   3.94               p-value <.05              all observations with required data (observations for which fiscal-year-end prior to restatement announcement is on or before December 31, 2013)   all observations with required data (observations for which fiscal-year-end prior to restatement announcement is on or before December 31, 2012       (1)   (2)  (3)   (4) VARIABLES  Prediction   ∆EARN   ∆EARN  ∆EARN_ 2ndYr^  ∆EARN_ 2ndYr^                   CSR  +  0.002**  0.001  0.003**  0.003 CSR*MATERIAL  +    0.002    0.002 Restatement Severity           MATERIAL  −  -0.008  -0.011  -0.016**  -0.019** REVENUE  −  -0.007  -0.007  -0.008  -0.008 CORE-EXP  −  -0.007  -0.007  -0.011*  -0.011* EM  −  -0.187  -0.185  0.027  0.030 DURATION  +  0.002  0.001  0.003**  0.003* COVENANT  −  -0.031  -0.030  -0.007  -0.006 WEAKNESS  −  -0.003  -0.003  0.009  0.009            CSR Controls    YES  YES  YES  YES General Controls    YES  YES  YES  YES            Industry FE    YES  YES  YES  YES Year FE    YES  YES  YES  YES Errors    Clustered  Clustered  Clustered  Clustered Observations    1,884  1,884  1,761  1,761 R-squared       0.2990  0.2993  0.3558  0.3560 58   Table 2.7B. Multivariate regression examining the effect of CSR on decrease in earnings (relative to forecasted earnings) following restatement     OLS  OLS, Alternative Specification  Heckman two stage     Sample with non-zero ∆EARN_UE  Sample with non-zero ∆EARN_UE_Robust  Full Sample   Sample with non-zero ∆EARN_UE     (1)  (2)  (3)  (4)  (5)  (6) VARIABLES  Prediction  ∆EARN_UE  ∆EARN_UE  ∆EARN_UE _ Robust  ∆EARN_UE _ Robust  (selection) ∆EARN_UE  (Final) ∆EARN_UE                CSR  +  -0.001  -0.003  -0.001  -0.003  0.021  -0.003 CSR*MATERIAL  +    0.009***    0.009**  -0.054  0.008* Exclusion Restriction                      VOLUME            0.02          AGE            -0.07   Restatement Severity               MATERIAL  −  -0.019**  -0.031***  -0.027**  -0.038***  -0.082  -0.032*** REVENUE  −  0  0  0.002  0.001  -0.268***  -0.006 CORE-EXP  −  0.022*  0.022*  0.034**  0.034**  -0.018  0.022** EM  −  -0.393***  -0.377***  -0.503***  -0.487***  -2.501**  -0.432*** DURATION  −  -0.002  -0.002  -0.001  -0.001  -0.026  -0.002 COVENANT  −  0.054***  0.058***  0.057***  0.060***  0.06  0.060** WEAKNESS  −  -0.021*  -0.022*  -0.023  -0.024  -0.127  -0.025* CSR Controls    YES  YES  YES  YES  YES  YES General Controls    YES  YES  YES  YES  YES  YES Industry FE    YES  YES  YES  YES  YES  YES Year FE    YES  YES  YES  YES  YES  YES R-squared    0.1608  0.1639  0.1781  0.1806     INV_MILLS            0.045   Observations    1,232  1,232  989  989  1884  1,232 Error are clustered across firms and years in *** p<.01, ** p<.05, * p< .1 (two tailed tests),   ^To address concern that the news of restatement leaked earlier than announcement, I use ∆EARN_UE_ROBUST = [EPSt – Forecast(EPSt)] / StockPricet–1 , where Forecast(EPSt) is the mean annual earnings forecast that were made at least 30 days prior to restatement announcement.   Model (2) Test of H0: CSR+ CSR*MATERIAL= 0,          F stat =   4.27,              p-value <.05        Model (4) Test of H0: CSR+ CSR*MATERIAL= 0,          F stat =   2.97               p-value <.159   Table 2.8.  Multivariate logistic regression examining the effect of CSR on management turnover following restatement      CEO/CFO Turnover  C-Suite Officer Turnover^     Full Sample  Material Restatements  Full Sample  Material Restatements       (1)  (2)  (3)  (4)  (5)  (6) VARIABLES  Prediction  TURNOVER  TURNOVER  TURNOVER  TURNOVER  CSUITE  TURNOVER  CSUITE  TURNOVER  CSUITE                       CSR  −  -0.024  0.024  -0.158***  -0.061  -0.019  -0.178*** CSR*MATERIAL  −    -0.168**      -0.146**   Restatement Severity               MATERIAL  +  0.557***  0.740***    0.704***  0.863***   REVENUE  +  0.208  0.206  -0.106  0.061  0.058  -0.186 COREEXP  +  0.142  0.133  -0.079  -0.059  -0.068  -0.227 EM  +  4.397***  4.261***  2.366  4.464***  4.333***  2.232 DURATION  ?  0.044***  0.046***  0.056  0.048***  0.050***  0.058 COVENANT  +  0.126  0.087  -0.088  -0.016  -0.053  0.053 WEAKNESS  +  0.472  0.503*  0.876**  0.394  0.418  0.825* Entrenchment Controls              CEO_OWN  −  -0.023**  -0.023**  0.010  -0.020***  -0.020***  0.010 INSIDER_OWN  −  0.002  0.002  0.003  -0.001  -0.001  -0.000 GOV_STR  +  -0.149  -0.143  0.014  -0.262  -0.257  -0.233 GOV_CON  −  -0.183  -0.163  -0.365  -0.152  -0.134  -0.362                CSR Controls    YES  YES  YES  YES  YES  YES General Controls    YES  YES  YES  YES  YES  YES Industry FE    YES  YES  YES  YES  YES  YES Year FE    YES  YES  YES  YES  YES  YES Errors    Clustered  Clustered  Clustered  Clustered  Clustered  Clustered Observations    1,764  1,764  315  1,764  1,764  315 Pseudo R-squared     0.0789   0.0813   0.1688  0.0689   0.0709   0.1358 Error are clustered across firms and years.  *** p<.01, ** p<.05, * p< .1 (two tailed tests) ^As an alternative specification, TURNOVER is set to one when any C-Suite officer resigns, retires, or is dismissed, and zero otherwise.60  Chapter 3: Does the Information Content of Analyst Forecast Revisions Increase following Accounting Restatements?  3.1. Introduction  Accounting restatements are major corporate misconducts with far reaching consequences.  Prior research finds that in addition to triggering significant negative market reaction (Palmrose et al., 2004), restatements also create uncertainty about the future prospects of the firm (Karpoff, 2012) and reduce the credibility of disclosures made by the managers (Chen et al., 2014; Wilson, 2008).  In the uncertain times following a restatement announcement, investors are likely to seek out and rely upon information sources other than the disclosures made by the management of restating firms.  Security analysts are information intermediaries with specialized industry expertise and institutional knowledge, and analyst forecasts are an important component in the corporate information environment.  I investigate whether investors put greater reliance on information provided by analysts in the period following restatements. Prior research shows that uncertainty induced by restatements can impact the analyst forecast characteristics (Dechow et al., 1996; Palmrose et al., 2004).  Barniv & Cao (2009) find that investors put different weights on analyst characteristics that are ex-ante predictors of forecast accuracy for firms that announce restatements relative to firms without such announcements.  The literature in analyst forecast accuracy has established a number of factors (i.e. forecast horizon, size of the brokerage house of the analyst, and etc.) that are associated with forecast accuracy (Bonner et al., 2003; Clement, 1999).  Bonner et al. (2003) find that, to a certain extent, investors react to these factors in a manner consistent with how the factors are 61  related to forecast accuracy.  Barniv & Cao (2009) find that for restatement firms, investors’ implied model of forecast accuracy is better aligned with the actual statistical relationship between these factors and forecast accuracy.  However, existing literature does not answer the question whether restatements lead to an overall increase in the information content of analyst forecast revisions (how strongly investors react to forecast revisions). The forces affecting the information content of analyst forecasts following restatement can be categorized into two groups.  First, analysts have unique expertise in interpreting complex corporate events (Livnat & Zhang, 2012).  Moreover, analysts’ incentives to be optimistically biased are likely to be reduced following restatements (Hong & Kubik, 2003; Zehr & Tuck, 2003), and this can further increase the credibility of analyst information.  When credibility of a prominent information channel like management disclosures is compromised in the wake of restatement (Chen et al., 2014; Wilson, 2008), investors might increase their reliance on information provided by the analysts.  On the other hand, investors might partially blame the analysts for failing to detect the problems in the accounting numbers earlier and might discount analysts’ ability to perform independent research in the aftermath of the restatement.  Hence analyst’s value in information discovery might be compromised following restatements.  In the presence of opposing forces, whether the information content of analyst forecast revisions increases following accounting restatements is an empirical question. Prior research distinguishes between material restatements (those perceived to be intentional) and errors (those perceived to arise from honest mistakes), and finds that material restatements are associated with greater adverse consequences (Hennes et al. 2008).  Chen et al. (2014) show that the decline in information content of management disclosures following restatements is more pronounced and longer lasting for material restatements, suggesting that 62  material restatements create greater uncertainty in the information environment of the firm.  It is important to appreciate that despite their far ranging effects, accounting restatements are primarily signals about management competence and integrity.  Hence restatements are likely to have a strong and direct impact on the information content of management disclosure, but only a weak and indirect impact on the information content of other information sources like analyst forecast revisions.  Therefore, following accounting restatements, any change in the information content of analyst forecast revisions will likely be more prominent for material restatements.  Hence I particularly focus on material restatements in my analyses. In this paper, I investigate the information content of annual forecasts made during the pre- and the post- restatement periods.  I find that the information content of analyst forecast revision is indeed higher in the post-restatement period compared to that in the pre-restatement period for material restatement firms.  But for immaterial restatement (error) firms, the forecast revision information content (FRIC) in the post-restatement period stays the same as its pre-restatement level.  All my results are robust to alternative specifications and inclusion of event fixed effects, which allow me to compare a given firm’s FRIC in the post-restatement period with its own FRIC in the pre-restatement period.  The increase in FRIC after material restatements is concentrated in the first quarter following restatement announcement, and dissipates in the second quarter of the post-restatement period.  Additional analysis reveals that FRIC is not different in the pre- and post- period for a group of control firms matched on industry, year, size, and fiscal-period-ending-date, verifying that my results are not driven by any macro-economic or industry-wide change.  The results suggest that investors increase their reliance on analyst forecast revisions in the uncertain time following accounting restatements, when the credibility of management disclosure is compromised. 63  I also analyze some cross-sectional differences in FRIC in the post-restatement period for material restatement firms.  In the post-restatement period, FRIC of the new analysts who initiate coverage of the firm after the restatement announcement is higher than FRIC of the existing analysts.  The new analysts are less likely to have close ties with the management, and less likely to share the blame for failing to detect the accounting problems earlier.  Hence they are less likely to be subject to information contamination forces.  The results suggest that in the uncertain information environment following restatement announcement, the investors rely more on the new and ‘blemish-free’ analysts.  Just like the main results, the differential FRIC between the new and existing analysts is concentrated in the first quarter after restatement, and dissipates by the second quarter of the post-restatement period.  This suggests that the preference for new analysts in the first quarter following restatement is a direct result of the substitution effect working more completely for new analysts rather than new analysts having any inherent superiority over the existing analysts.  Additional analysis reveals that new analysts do not enjoy higher FRIC compared to existing analysts in the post-period for matched control firms, verifying that my results cannot be explained by a general (hypothetical) increase in investors’ interest in new analysts at the time of coverage initiation regardless of the impact of restatement announcement. This paper makes several contributions to the literature.  First, this paper provides direct evidence suggesting that analysts’ role in informing the investors becomes more prominent in the uncertain times following restatement announcement.  The finding highlights the important role analysts play as information intermediaries in the capital market, and confirms that investors consider analysts as experts who can be relied upon for information in times of uncertainty.  This paper is similar to Barniv & Cao (2009) in the sense that both papers investigate investors’ 64  response to analyst forecasts following restatements, but the two papers explore fundamentally different research questions.  Barniv & Cao (2009) investigate if investors interpret analyst characteristics related to forecast accuracy differently for restatement firms compared to non-restatement firms.  In contrast, I investigate whether market reaction sensitivity to forecast news (magnitude of the revision) increases following restatement announcement. Secondly, this paper responds to a call for new research to investigate ‘the interdependencies between the information channels that shape the corporate information environment’ (Beyer et al., 2010).  Prior research has uncovered useful insight into the analyst information production and its value to the investors (Beyer et al., 2010).  This paper investigates how the value of analyst information can change in response to shocks to management credibility that result from restatement announcements. Thirdly, this paper enhances our understanding of analyst incentives and behaviors in the aftermath of restatements.  Barniv & Cao (2009) find that analysts with superior abilities (with all-star ranking from Institutional Investor magazine) are more likely to maintain and initiate coverage of the restatement firms compared to other analysts.  They also find that a considerable number of new analysts initiate coverage of the restatement firms following the restatement announcement.  The fact that investors value analyst information more highly after the restatement announcement may be one of the primary reasons why analysts continue or initiate their coverage of restatement firms. I proceed as follows. In Section 3.2, I discuss prior literature and develop my hypotheses.  Section 3.3 describes the sample.  Section 3.4 presents the timeline, research design, and variable construction.  I discuss the empirical results in section 3.5, and conclude in Section 3.6.  65  3.2. Literature Review and Hypotheses Development  Restatement and Corporate Information Environment  Accounting restatement is a serious corporate misconduct.  Financial statements are the bedrock of corporate information environment, and the capital market operates with the general assumption that the audited financial statements contain reliable information about a firm’s economic state.  Industry updates, analyst reports, management guidance are more timely sources of information about the firm, and the importance of mandatory financial disclosures comes from their confirmatory and contracting role.  These important features of earnings news and financial statements are compromised when it is revealed that the previously filed financial statements, which were believed to be accurate by the investors (Bardos et al. 2011), were in fact inaccurate.   Revelation of misreporting can have far ranging consequences on the corporate information environment.  Prior research suggests that restatements create uncertainty about the restating firm, as evidenced by greater analyst forecast dispersion (Dechow et al. 1996; Palmrose, et al. 2004).  Chen et al. (2013) and Wilson (2008) show that the information content of the earnings announcements goes down following restatements, and this effect is greater for more material restatements, which engender greater uncertainty about the restating firm.  Gordon et al. (2014) find that following restatements, the accuracy of management guidance goes down.  Consistent with the argument that restatements create uncertainty about the firm’s future, Hribar & Jenkins (2004) find that firms face higher cost of equity following restatements. Role of Analyst Disclosures in the Corporate Information Environment Equity analysts are important information intermediaries between the firm and the investors, and they can contribute to the information environment of the firm through their 66  information discovery and information interpretation roles (Chen, Cheng, & Lo, 2010; Frankel, Kothari, & Weber, 2006).  Prior research (Livnat & Zhang, 2012) shows that the interpretation value of the analyst reports is particularly high immediately after major corporate news that are not easily quantifiable (i.e. news of officer turnover, change in policy, new trade agreement).  Prior research suggests that information value of analyst reports varies depending on the investors’ demand for analyst information, and the cost of gathering relevant information faced by the analysts (Frankel et al., 2006).  Accounting restatements can potentially impact both the demand for analyst information, and the analysts’ cost of gathering such information.  I organize the various forces that can impact the information content of analyst forecasts following accounting restatements into two competing categories: i) Information Substitution Forces and ii) Information Contamination Forces. Analyst Disclosure and the Information Substitution Forces of Restatement Restatements create uncertainty about the firm’s future and compromise the credibility of firm disclosures.  In the period immediately after restatement, the investors’ demand for analyst information is likely to increase because 1) analysts can discover relevant information and thereby reduce the uncertainty triggered by the restatement and 2) analysts can help verify and interpret the disclosures made by restating managers, whose credibility is compromised following restatements, as evidenced by prior research (Chen et al., 2014; Wilson, 2008).  But the investor demand for analyst information will increase following restatements only if the investors continue to believe that the analysts are capable of finding and analyzing relevant information, and they have the correct incentive to report without bias. It is well known that issuing negative reports can hinder the analysts’ career paths (Hong & Kubik, 2003) and invite costly lawsuits from the firm (Zehr & Tuck, 2003).  From an 67  incentive perspective, it is difficult to imagine how a restating firm reeling from the news of restatement can blame pessimistic analyst reports for its troubles.  Hence restatements are likely to curb the analyst incentive to be optimistically biased, and motivate them to report accurately.  In this paper, the overall conjecture that analysts can provide pertinent information to investors in the uncertain information environment following a restatement, leading to greater investor reliance on analyst information, is termed as the ‘information substitution’ argument.  Analyst Disclosure and the Information Contamination Forces of Restatement The investor’s perception about the ability of analysts can change following accounting restatements.  A key source of information for analysts is their access to management.  After a restatement announcement, if the investors believe that the management is incompetent or mal-intent in their disclosure policy, then they are more likely to discount the analyst’s value based on their access to the management.  Moreover, prior research finds that, in general, analysts do not change their behavior (i.e. adjust forecasts, terminate coverage) prior to fraud discovery (Griffin, 2003).  Using a sample of corporate frauds for which securities class action lawsuits were filed, Dyck et al. (2010) find that analysts were the whistle blowers for only 16.9% of the cases. The limited success of analysts in acting as whistle blowers can compromise the investors’ trust in the analysts’ ability to perform independent research and discover relevant information on their own.  If investors lose confidence in analysts’ ability to collect and disseminate useful information, then the investors’ demand for analyst produced information would go down following accounting restatements.  The information content of the analyst reports also depends on the information gathering cost function faced by the analysts since high-quality and value-relevant reports require more research.  There is some evidence suggesting that institutional holdings go down following 68  restatements (Hribar et al., 2004), which can discourage analysts from putting high effort for their research.  Moreover, a major source of information for the analysts is their relationship with the management.  Since restatements create doubts about the disclosure incentives of the managers, the ‘learning from managers’ channel may be compromised, increasing the cost of gathering information for the analysts.  In this paper, the overall conjecture that restatements cause a real or perceived loss of analysts’ ability to find and disseminate relevant information, leading to a lowering of investor reliance on analyst information, is termed as the ‘information contamination’ argument. Analyst Information Content Following Restatement According to the information substitution argument, restatements create uncertainty about the firm’s prospects, and compromise the credibility of management disclosures.  When an important information channel is compromised, investors are likely to increase their reliance on information provided by the analysts, who are the industry experts with specialized knowledge about the restating firm, and who have greater incentives to report accurately following restatements.  On the other hand, according to the information contamination argument, restatements can cause doubts about the analyst’s ability to provide relevant information to the investors, and reduce investor reliance on analyst information.  With opposing forces at work, whether the information content of analyst forecast revisions increases, decreases, or remains the same is an empirical question.  I state my first hypothesis in the null form: H1: The information content of analyst forecast revisions (FRIC) remains unchanged following accounting restatements  According to prior research (Barniv & Cao, 2009), around 30% of the analysts drop coverage in the one year following restatement announcements, but a considerable number of 69  new analysts initiate coverage during the same time.  Since these new analysts initiate coverage right after restatement announcements, they cannot be blamed for failing to detect accounting problems of the restatement firms earlier.  Moreover, the new analysts are unlikely to have close ties with the management of the restatement firms, which would help protecting their image as reliable information intermediaries.  In the post-restatement period, the new analysts are less likely to suffer from the information contamination forces compared to existing analysts.   The uncertain information environment following restatement announcements can give rise to both information substitution forces and information contamination forces.  However, the substitution forces should work for both new and existing analysts, while the contamination forces should be weak (or absent) for the new analysts.  Without the opposing effect of the contamination forces, the substitution effect experienced by the new analysts should be larger.  In the post-restatement period, the investors are likely to rely more on information provided by these new analysts compared to information provided by the existing analysts.  Stated formally: H2: In the post restatement period, the information content of analyst forecast revisions (FRIC) is greater for analysts who initiate coverage after restatement announcement  3.3. Sample  I investigate how strongly the market reacts to revisions of analysts’ earnings forecasts for restatement firms before and after the restatement announcements.  I focus on revisions of annual forecasts for the next fiscal year with horizon no larger than 365 days until the forecast period ending date13.  A significant portion of the forecasts overlap with the 3-day earnings                                                           13 Annual forecasts are the most widely available class of forecasts in IBES. 70  announcement windows.  Since the market reaction over the three days centered on the earnings announcement date is dominated by the earnings news, I exclude analyst forecast revisions that overlap with the earnings announcement window from my analysis.   Table 3.2 summarizes the selection process of the restatement firms analyzed in this paper.  I collect restatement data from Audit Analytics Non-Reliance database (AA).  AA provides restatement data from the year 2000 to 2013.  The major advantage of using the AA database is that it covers restatements of recent years (as opposed to the GAO database with coverage ending in 2005).  I collect financial data from COMPUSTAT, and market return data from CRSP.  I first match the AA restatement observations with COMPUSTAT using a combination of historical CIK and name matching.  The large majority of my sample observations are matched using this technique. COMPUSTAT does not have CIK information for all the firms.  Hence I use a second matching technique suggested in the Wharton Research Database Repository.  I require that the names match, and the total assets and total sales of the past two years reported by AA and COMPUSTAT match within 1% of each other.  I manually verified the results of my second technique and found no mismatches.  I start with 13899 restatement observations from AA database.  I can match 10,417 of them with COMPUSTAT data.  Requiring CRSP and IBES coverage further reduces my number of restatements to 5298 observations. To ensure that the restatement firms have stable information environment prior to the restatement announcement, I require that my restatement firms have non-missing book value, price, and shares outstanding data available from COMPUSTAT in the fiscal year immediately prior to the restatement announcement. Additionally, I require that the restatement firms have monthly return information from the CRSP Monthly database for at least 12 consecutive months 71  prior to the restatement announcements.  Imposing these restrictions leave 4941 observations in my sample. Some companies announce consecutive restatements over a short period of time. For my analysis, it is critical to ensure that the analyst forecast revisions in the benchmark period are not affected by previous restatements.  Hence I exclude restatements from the sample if the restating firm had filed prior restatements within the past two years of the restatement announcement date. This leaves 3781 restatements in my sample.  Some restatements are actually income increasing, and investors might view them quite differently from the income reducing restatements.  For clean interpretation, I only keep the income reducing restatements (as classified by AA database), which leaves 3182 restatement observations for analysis.  Finally, I require that my sample restatement firms have analyst forecast revisions that do not coincide with the earnings announcement window in both the pre- and the post-restatement periods.  The final sample has 1811 restatements.   Prior research shows the importance of separating material restatements/ frauds/ irregularities from simple errors (Hennes et al., 2008).  Following prior research (Badertscher et al., 2011; Hennes et al., 2008), I classify a restatement as material if i) there is an express admission of fraud in the restatement disclosure, or ii) the restatement disclosure cites SEC investigation, or iii) there is a class-action lawsuit filed within one year of the restatement filing.  The information required to identify restatements involving fraud, SEC investigation, and/ or class-action lawsuits is available from the AA database.  Out of the 1811 restatements, 367 restatements are MATERIAL restatements, which are likely to generate greater uncertainty after the restatement announcements (See Table 3.2 Panel B). My final sample has 39,089 forecast revisions of restatement firms (1811 restatements).  Since material restatements are likely to have a larger impact on the information content of 72  analyst forecast revisions, I am particularly interested in the 8675 forecast revisions of material restatement firms (367 restatements).  3.4. Research Design and Variable Construction  3.4.1. Time Line  To investigate the impact of accounting restatements on the information content of analyst forecasts, I compare the information content of annual forecast revisions in the pre- and post-restatement periods.  Over the course of a fiscal year, the forecast horizons decrease, and the analysts often become more accurate (Bonner et al., 2003) and less optimistic (Richardson et al., 2004).  Moreover, the seasonality in quarterly earnings and related market return patterns can also influence investor reaction to analyst forecasts. Hence the market reaction to annual forecasts might systematically vary across fiscal quarters. I need to account for this possible seasonality when comparing the information content of analyst forecasts in the pre- and post- restatement periods.    Figure 3.1 provides the basic timeline.  The firm’s restatement announcement takes place on date T. I define the 9 days surrounding the restatement announcement day (T−4 through T+4) as the restatement announcement window.  A relatively large announcement window is used to ensure that the market reaction to the restatement news itself does not contaminate the forecast revisions in the post restatement period.  The 6 months (180 days) of time after the restatement (T+5 through T+184) is the post-restatement period.  To address the concerns for seasonality in the analyst information forecast, I set a benchmark date (TB) that is exactly one calendar year prior to the restatement announcement date.  The 6 months (180 days) of time after the bench 73  mark event (TB+5 through TB+184) is the pre-restatement period or the benchmark period.  I investigate if FRIC in the post-restatement period is greater than FRIC in the pre-restatement period.   I investigate the first 6 months of time after restatement announcement for two reasons.  The accounting restatements are likely to have a first order impact on the information content of subsequent management disclosures since it is the managers’ past disclosures that are proven to be incorrect in the wake of restatements.  Since the uncertainty in the information environment of the firm following restatement is likely to be the greatest right after the restatement, the probability of observing an effect is the highest in the relatively short period right after restatement.  Another reason to set 6 months (180 days) time for pre- and post- restatement periods is that the news of restatement may leak earlier than the official restatement date, and therefore the days immediately prior to the official restatement announcement date cannot be considered to be in the benchmark period.  Choosing 180 days of benchmark period that ends well in advance of the official restatement announcement date minimizes the possibility that the benchmark period will be contaminated by the effect of the restatement. 3.4.2. Empirical Models and Variable Measurements Testing Hypothesis 1: I mainly adopt the research design used by Livnat & Zhang (2012), and make necessary adjustments to suit the needs of my research questions.  I use the following regression model for testing whether information content of analyst forecast revisions increases following restatements: 𝑪𝑨𝑹 =  𝜶𝟎 +  𝜶𝟏𝑹𝑹𝑬𝑽 + 𝜶𝟐𝑷𝑶𝑺𝑻 +  𝜶𝟑𝑹𝑹𝑬𝑽 ∗ 𝑷𝑶𝑺𝑻                  +  [𝑪𝒐𝒏𝒕𝒓𝒐𝒍𝒔]   +    𝑹𝑹𝑬𝑽 ∗ [𝑪𝒐𝒏𝒕𝒓𝒐𝒍𝒔]   +   𝜺   (3.1.1) 74  The dependent variable CAR is the cumulative size-adjusted abnormal return over 3 trading days (−1,0,1) centered around the forecast revision date.  For each individual analyst forecast revision, I calculate the magnitude (REVISION) as the revised forecast minus the same analyst’s previous forecast for the same forecast period deflated by stock price as of the end of the month immediately prior to the forecast revision.  I calculate REVISION for all the firms available in IBES, and then rank my REVISION variable into equal sized deciles (from 0.1 to 1) by calendar month.  I use this monthly rank variable (RREV) in my empirical analysis to prevent undue impact of outliers. The coefficient on RREV is 𝜶𝟏, which captures the information content of analyst forecast revisions in the benchmark period.  According to prior research, coefficient 𝜶𝟏should be positive and significant (Livnat & Zhang, 2012).  POST is an indicator variable equal to one if the forecast revision is made during the 180 days of post-restatement period and zero otherwise.  𝜶𝟑 is the coefficient on RREV*POST, and captures the change FRIC from the benchmark period.  According to hypothesis 1, coefficient 𝜶𝟑 should be positive and significant. Table 3.1 provides the detailed definitions for all the variables used in my empirical analyses.   Following Livnat & Zhang (2012), I control for SIZE, BM, NUMFIRMS, BROKER, EXPERIENCE, HORIZON, INNOVATIVE, and each of their interaction terms with RREV.  Additionally, I control for INFOFLOW, which captures the intensity of the analyst activities in the run up to the forecast revision.  INFOFLOW is defined as the natural log of one plus the number of days on which analyst reports were issued during the 30 days preceding the forecast revision.  SIZE, INFOFLOW, and BM capture different dimensions of the overall information environment of the restating firm.14 NUMFIRMS, BROKER, and EXPERIENCE are the analyst                                                           14 Number of analysts following (FOLLOWING) the firm is a more widely used proxy to capture the amount of analyst activity at the time of the forecast revision.  However, FOLLOWING has a very high correlation with SIZE (more than 0.6), and Livnat & Zhang (2012) do not use FOLLOWING as one of their control variables for that reason.  A restating firm may go through major changes in its information environment, and hence I include an 75  specific controls used in my empirical model.  NUMFIRMS captures the number of firms covered by the analyst, BROKER reflects the size (number of employees) of the brokerage house the analyst works for, and EXPERIENCE reflects the number of years the analyst has been in the IBES database. HORIZON and INNOVATIVE are the forecast specific controls used in the empirical model.  HORIZON is the time (in days) between the forecast revision day and the forecast period ending date.  INNOVATE is an indicator variable that is equal to one if the analyst’s current forecast deviates further from the consensus compared to analyst’s prior forecast, and zero otherwise.  Following Livnat & Zhang (2012), the log transformations of SIZE, INFOFLOW, NUMFIRMS, BROKER, EXPERIENCE, and HORIZON are used in the empirical analysis to reduce the impact of outliers in the regression variables. As discussed previously, the uncertainty in the information environment following accounting restatement is likely to be the greatest in the time period right after the restatement announcement, and that is when any change in the information content of analyst forecast revision is likely to be the largest.  I split the 180 days of post restatement period into two sub-periods of 90 days: POST1 and POST2 (See Figure 3.2).  To control for seasonality in the information content of forecast revisions, I need to compare POST1 with PRE1 and POST2 with PRE2, where PRE1 and PRE2 are two 90-day sub-periods of the pre-restatement period, as illustrated by Figure 3.2.  I expect to see a larger change in FRIC in the POST1 period compared to that in the POST2 period.  In order to separately investigate the change in FRIC in the POST1 and POST2 period, I use the following regression model:                                                           additional proxy to capture the intensity of analyst activities in my model.  The correlation of INFOFLOW and SIZE is still quite high (around 0.5), but not as extreme as that between FOLLOWING and SIZE. In my empirical analysis, I reach the same conclusion irrespective of whether I include INFOFLOW, FOLLOWING, both, or none of these variables that capture the amount of analyst activities prior to forecast revision. 76  𝑪𝑨𝑹 =  𝜶𝟎 +  𝜶𝟏𝑹𝑹𝑬𝑽 + 𝜶𝟐𝑷𝑶𝑺𝑻 +  𝜶𝟑𝑹𝑹𝑬𝑽 ∗ 𝑷𝑶𝑺𝑻 + 𝜶𝟒𝑺𝑬𝑪𝑶𝑵𝑫𝟗𝟎     + 𝜶𝟓𝑹𝑹𝑬𝑽 ∗ 𝑺𝑬𝑪𝑶𝑵𝑫𝟗𝟎 + 𝜶𝟔𝑷𝑶𝑺𝑻 ∗ 𝑺𝑬𝑪𝑶𝑵𝑫𝟗𝟎 +  𝜶𝟕𝑹𝑹𝑬𝑽 ∗ 𝑷𝑶𝑺𝑻 ∗ 𝑺𝑬𝑪𝑶𝑵𝑫𝟗𝟎                  +  [𝑪𝒐𝒏𝒕𝒓𝒐𝒍𝒔]   +    𝑹𝑹𝑬𝑽 ∗ [𝑪𝒐𝒏𝒕𝒓𝒐𝒍𝒔]   +   𝜺   (3.1.2) SECOND90 is an indicator variable that takes on the value of one if forecast revision is made during either PRE2 or POST2 period, and zero otherwise.  In equation (3.1.2), 𝜶𝟏 is the FRIC in the PRE1 period, and 𝜶𝟑 is the change in FRIC from PRE1 to POST1 period. Coefficient 𝜶𝟓 is the difference in FRIC between PRE1 and PRE2 period, and coefficient 𝜶𝟕 is the incremental change in FRIC in the POST2 period, relative to the change in FRIC in the POST1 period.  Thus, coefficients  𝜶𝟏 + 𝜶𝟓 measure the FRIC in the PRE2 period. Similarly, coefficients 𝜶𝟑 + 𝜶𝟕 capture the change in FRIC from PRE2 to POST2 period.  The interpretation of the coefficients is summarized as follows15: Interpretation of coefficients for equation (3.1.2): Forecast Revision Information Content (FRIC) First 90 days (SECOND90=0)  Second 90 days (SECOND90=1) FRIC in the PRE period 𝜶𝟏 𝜶𝟏 + 𝜶𝟓 FRIC in the POST period 𝜶𝟏 + 𝜶𝟑 𝜶𝟏 + 𝜶𝟓 + 𝜶𝟑 + 𝜶𝟕 Change in FRIC 𝜶𝟑 𝜶𝟑 + 𝜶𝟕  According to hypothesis 1, 𝜶𝟑 should be positive and significant.  Since I expect the change in FRIC to be concentrated in the first quarter of the post-restatement period, I expect 𝜶𝟑 to be larger than 𝜶𝟑 + 𝜶𝟕 (I expect 𝜶𝟕 to be negative).                                                           15 Implementing and interpreting research design models with three-way interactions can be cumbersome. Chen et al. (2014) use such models when investigating differential change in information content of earnings (ERC) following restatements for different groups, and provide an excellent discussion of their research design.  I closely follow their methodology.  77   Testing Hypothesis 2:  To test whether FRIC of new analysts who initiate coverage after restatement announcement is higher than FRIC of existing analysts in the post-restatement period, I exclusively analyze the forecast revisions made during the post-restatement period.  To determine whether the difference between ‘FRIC of new analysts’ and ‘FRIC of existing analysts’ is more prominent in the first 90 days of the POST period, I split the POST period using the SECOND90 indicator.  I use the following regression model: 𝑪𝑨𝑹 =  𝜶𝟎 +  𝜶𝟏𝑹𝑹𝑬𝑽 + 𝜶𝟐𝑵𝑬𝑾𝑨𝑵 +  𝜶𝟑𝑹𝑹𝑬𝑽 ∗ 𝑵𝑬𝑾𝑨𝑵 + 𝜶𝟒𝑺𝑬𝑪𝑶𝑵𝑫𝟗𝟎 + 𝜶𝟓𝑹𝑹𝑬𝑽 ∗ 𝑺𝑬𝑪𝑶𝑵𝑫𝟗𝟎 + 𝜶𝟔𝑵𝑬𝑾𝑨𝑵 ∗ 𝑺𝑬𝑪𝑶𝑵𝑫𝟗𝟎 + 𝜶𝟕𝑹𝑹𝑬𝑽 ∗ 𝑵𝑬𝑾𝑨𝑵 ∗ 𝑺𝑬𝑪𝑶𝑵𝑫𝟗𝟎                  +  [𝑪𝒐𝒏𝒕𝒓𝒐𝒍𝒔]   +    𝑹𝑹𝑬𝑽 ∗ [𝑪𝒐𝒏𝒕𝒓𝒐𝒍𝒔]   +   𝜺   (3.2) NEWAN is an indicator variable that equals one for the forecast revisions made by the analysts who initiate coverage after the restatement announcement (new analysts) and zero otherwise.  In equation (3.2), 𝜶𝟏 reflects FRIC of existing analysts in the POST1 period. 𝜶𝟑 captures the difference between FRIC of new analysts and FRIC of existing analysts in the POST1 period.  For the existing analysts, coefficient 𝜶𝟓 is the difference in FRIC between POST1 and POST2 period (See Figure 3.2).  𝜶𝟕 captures the incremental difference in FRIC between POST1 and POST2 period for the new analysts relative to that of the old analysts. Thus, coefficients  𝜶𝟏 + 𝜶𝟓 measure FRIC of existing analysts in the POST2 period.  Similarly, coefficients 𝜶𝟑 + 𝜶𝟕 capture the difference between FRIC of new analysts and FRIC of existing analysts in the POST2 period. The interpretation of the coefficients is summarized as follows:   78  Interpretation of coefficients for equation (3.2): Information Content (IC) First 90 days (SECOND90=0)  Second 90 days (SECOND90=1) IC for the Existing Analysts (NEWAN=0) 𝜶𝟏 𝜶𝟏 + 𝜶𝟓 IC for the New Analysts (NEWAN=1) 𝜶𝟏 + 𝜶𝟑 𝜶𝟏 + 𝜶𝟓 + 𝜶𝟑 + 𝜶𝟕 Difference in IC between New Analysts and Old Analysts 𝜶𝟑 𝜶𝟑 + 𝜶𝟕  According to hypothesis 2, FRIC of new analysts should be greater than FRIC of existing analysts.  Hence I expect 𝜶𝟑 to be positive.  If the FRIC of new analysts remains higher than the FRIC of existing analysts only for a short period, then 𝜶𝟑 + 𝜶𝟕 will be smaller than 𝜶𝟑 (𝜶𝟕 will be negative).  3.5. Empirical Results  3.5.1. Descriptive Statistics  Table 3.3 Panel A presents the descriptive statistics on the regression variables for the full sample (all restatements, all forecasts), separately for the pre- and post- restatement period.  While the log transformations of SIZE, INFOFLOW, NUMFIRMS, BROKER, HORIZON, and EXPERIENCE are used in the regression analysis, they are reported in their original (unlogged) version to facilitate interpretation.  Looking at the pre-restatement statistics, the mean (median) market reaction to forecast revisions is −0.009 (−0.004), while the mean (median) forecast revision is −0.007 (−0.001).  The mean (median) SIZE is USD 8,672.6 Million (USD 2244.1 Million), suggesting that I am investigating firms with fairly large market capitalization.  Mean (median) INFOFLOW is 5 (5.5).  Mean (median) BM is 0.635 (0.290).  On average, an analyst 79  issuing the forecast revision covers 15.7 firms, is employed by a brokerage house with 56.7 analysts, and has 8.0 years of experience. The mean (median) forecast horizon is 164.2 (174) days. Out of all the forecast revisions made during the pre-restatement period, 44.9% are INNOVATIVE.  Moving to the post-restatement period statistics, the market reaction to forecast revision and the forecast revisions themselves continue to be slightly negative.  However, the mean CAR is slightly less negative while the mean REVISION is slightly more negative compared to the pre-restatement period.  Among the forecast revisions made during the post-restatement period, 9.2% are issued by analysts who initiate coverage of the firm after the restatement announcement.  The mean (median) SIZE is USD 7657.2 Million (USD 2482.1 Million).  The mean SIZE in the post-restatement period is slightly smaller than the mean SIZE in pre-restatement period, but the median SIZE actually increases by a small amount in the post-restatement period.  On average INFOFLOW does not change significantly from pre- to post- restatement period.  Mean (median) book-to-market ratio is 0.616 (0.477).  The analyst attributes (NUMFIRMS, BROKER, and EXPERIENCE) largely remain the same, although there is a statistically significant increase in NUMFIRMS.  The average forecast horizon is 160.8, which is slightly smaller than its pre-restatement period level.  Compared to pre-restatement period, a slightly smaller proportion of the forecast revisions are INNOVATIVE (43.5%).  Table 3.3 Panel B presents the descriptive statistics on the regression variables only for the material restatements (material restatements, all forecasts), separately for the pre- and post- restatement period.  On average the material restatement firms are much larger than the overall population of restatement firms, and on average, they lose a large portion of their share value after the restatement announcement (from USD 13,671 Million in the PRE period to USD 10,962 80  Million in the POST period).  Also, the horizon of the forecast revisions for the material restatement firms is noticeably smaller than that for the full sample both in the pre- and post- restatement periods.  Apart from SIZE and HORIZON, the various other attributes of the forecast revisions in the pre- and the post-restatement periods for material restatement firms are similar to that of the full sample.  3.5.2. Regression Analysis Tables 4, 5, and 6 report the results for my regression analyses testing hypotheses 1 and 2. All the continuous variables are winsorized at 1st and 99th percentile.  Following Livnat & Zhang (2012), the errors are clustered along firm and calendar month16.  Two-tailed significance levels are reported. Regression Analysis (Testing H1): Table 3.4 presents the regression results for the change in the information content of forecast revisions after restatement announcement.  Column 1 reports the estimation of equation (3.1.1) using the full sample.  The coefficient on RREV is 0.101 and statistically significant at .01 level, suggesting that FRIC in the benchmark period is 0.101. The coefficient on RREV*POST however is not statistically different from zero.  The result suggests that in general, information content of forecast revisions remain the same following restatements.  Column 2 reports the estimation of equation (3.1.1) using just the forecast revisions of the material restatements.  In column 2, the coefficient on RREV is 0.103 and significant at .01 level, while the coefficient on RREV*POST is 0.022 and statistically significant at .10 level.                                                            16 As described previously, I rank the forecast revisions into deciles each calendar month.  The choice of clustering along calendar month is consistent with the way I rank my forecast revisions. 81  The results suggest that FRIC in the benchmark period for material restatement firms is 0.103, and the FRIC increases by 0.022 following the material restatement announcements.  The magnitude of the increase in FRIC following material restatements is fairly large (an increase of 0.022/0.103 = 21.4%). Table 3.4 column 3 reports the estimation of equation (3.1.2) using material restatements, and sheds light on whether the impact of material restatements on FRIC is more prominent in the first 90 days of the post-restatement period.  In column 3, the coefficient on RREV, which captures the FRIC in the PRE1 (benchmark) period, is 0.100.  The coefficient on RREV*POST is 0.035 and statistically significant at .10 level.  There is a 35% increase in FRIC in the POST1 period compared to the PRE1 period.  The results are different in the second 90-day portion of the timeline.  FRIC in the PRE2 period, which is captured by coefficients (RREV + RREV*SECOND90), is 0.112 and statistically significant at .01 level [F-stat = 8.24].  The change in FRIC from PRE2 to POST2 period, which is captured by coefficients (RREV*POST + RREV*POST*SECOND90), is 0.006 and not statistically significant [F-stat = 0.006].  Hence the impact of material restatements on FRIC is short lived and observable only in the first quarter of the post-restatement period. Table 3.4 column 4 repeats the analysis in column 3 after including event fixed effects.  In this specification, each restatement firm is its own control, and I analyze the within firm variation in FRIC before and after the restatement announcement.  FRIC in the PRE1 period, captured by coefficient on RREV, is 0.072 and statistically significant at 0.10 level.  The coefficient on RREV*POST is 0.035 and statistically significant at .01 level.  FRIC goes up by 0.035 (an increase of 48.6%) in the POST1 period compared to PRE1 period.  This large positive impact of material restatements on FRIC in the first 90 days of the post-restatement period does 82  not last into the second 90 days.  FRIC in the PRE2 period, which is captured by coefficients (RREV + RREV*SECOND90), is 0.093 and statistically significant at .05 level [F-stat = 5.78].  The change in FRIC from PRE2 to POST2 period, which is captured by coefficients (RREV*POST + RREV*POST*SECOND90), is -0.003 and not statistically significant [F-stat = 0.005]. Table 3.4 column 5 repeats the analysis in column 4 after removing the forecast revisions made by new analysts (who initiate coverage after restatement announcement). This specification allows me to investigate the impact of material restatements on FRIC without any influence from the effect of the new analysts.  In column 5, the coefficient of RREV is 0.067 and statistically significant at .10 level.  Coefficient on RREV*POST is 0.033 and statistically significant at .01 level.  FRIC goes up by 0.033 (an increase of 49.25%) in the POST1 period compared to PRE1 period.  The coefficient for RREV*POST*SECOND90 is negative and significant, and the coefficients (RREV*POST  +  RREV*POST*SECOND90) are not statistically significant [Fstat=-0.06], showing that FRIC in POST2 period is not different from FRIC in the PRE2 period. Across all five specifications presented in Table 3.4, the effect of control variables on FRIC is similar.  RREV*SIZE is not statistically significant while RREV*INFOFLOW is negative and significant across the board.  SIZE not having an impact on RREV is surprising, since prior literature suggests that SIZE should have a negative impact on RREV (Livnat & Zhang, 2012).  Upon closer inspection, it is revealed that the coefficient RREV*SIZE dose not load significantly because SIZE is highly correlated with INFOFLOW.  RREV*SIZE becomes significant and negative if I remove INFOFLOW and its interaction with RREV from the regression.  BM has a negative impact on FRIC, but the impact is only significant for models 83  without fixed effects.  BROKER has a positive and significant effect on FRIC, indicating that investors put more faith in analysts from larger brokerage houses.  But NUMFIRM and EXPERIENCE do not have any statistically significant impact on RREV.  The coefficient on RREV*HORIZON is negative while the coefficient on RREV*INNOVATE is positive, suggesting that forecasts with longer horizon have lower information content, while innovative or bold forecasts have higher information content. Results from Table 3.4 reveal that material restatements have a positive and significant impact on FRIC and this impact is concentrated in the first quarter of the post-restatement period.  This effect is not observable in the analysis using forecasts of all restatements, indicating that the immaterial restatements (errors) do not have any impact on the information content of forecast revisions.  While I control for a set of analyst and forecast characteristics that have been shown to impact the information content of forecasts by prior literature (Livnat & Zhang, 2012), I do not control for all possible variables that can influence the FRIC.  I acknowledge that these omitted variables might be partially responsible for the increase in FRIC in the post-restatement period.  However, the spike in FRIC coincides with the peak time for uncertainty immediately following the restatement announcement, and it is difficult to think of analyst or forecast characteristics that would deviate sharply for just for one quarter following restatement announcement, and then return to their benchmark period level.  This gives some assurance that the observed spike in FRIC is indeed caused by the uncertainty induced by the material restatements.    Regression Analysis (Testing H2): Table 3.5 presents regression results for the tests investigating whether FRIC in the post-restatement period is greater for new analysts (those who initiate coverage after restatement 84  announcement).  Since earlier results reveal that only material restatements are associated with a change in FRIC, while immaterial restatements do not have a large enough impact on the information environment of the firm, I only focus on the material restatement firms for this analysis. Column 1 of Table 3.5 reports the estimation of equation (3.2) using all the forecast revision observations available in the POST period for material restatement firms.  The indicator variable SECOND90 splits the POST period into POST1 and POST2 sub-periods (see Figure 3.2). The coefficient on RREV is 0.132 and significant at .05 level, indicating that FRIC of existing analysts in the POST1 period is 0.132.  Coefficient on RREV*NEWAN is 0.079 and significant at .05 level, suggesting that FRIC of new analysts is indeed higher than FRIC of existing analysts in the POST1 period (higher by 59.8%).  Coefficient RREV*SECOND90 is not statistically significant, but coefficient RREV*NEWAN*SECOND90 is negative and significant.  Untabulated results reveal that coefficients (RREV + RREV*SECOND90) are 0.121 and significant at .05 level [F stat = 4.97], indicating that FRIC of existing analysts in the POST2 period is 0.121.  But coefficients (RREV*NEWAN + RREV*NEWAN*SECOND90) are 0.005 and statistically insignificant [F stat = 0.09], indicating that the FRIC of new analysts is not different from FRIC of existing analysts in the POST2 period.  Recall from earlier results that the increased FRIC following restatement announcement of material restatements is concentrated in the POST1 period. New analysts only have higher FRIC compared to old analysts in the first 90 days of the post-restatement period, when the uncertainty in the information environment is probably the greatest, and when the investors are putting additional reliance on analyst information. 85  Column 2 of Table 3.5 repeats the analysis done in column 1 after including event fixed effects.  The coefficient on RREV is 0.083 but insignificant, but the coefficient on RREV*NEWAN is 0.053 and significant at .10 level.  Results indicate that FRIC of new analysts is higher than FRIC of existing analysts in the POST1 period.  Untabulated results show that coefficients (RREV + RREV*SECOND90) are 0.085 and significant at .10 level [F stat = 2.84], while the coefficients (RREV*NEWAN + RREV*NEWAN*SECOND90) are 0.007 and not statistically significant.  Hence FRIC of existing analysts is 0.085 in the POST2 period, but FRIC of new analysts is not different from FRIC of existing analysts in the POST2 period. In table 3.5, columns 3 and 4 repeat the analysis done in columns 1 and 2 respectively, but only using the forecast observations in the first 90 days of the post-restatement period (SECOND90=0).  With few observations available for the analysis, the coefficient on RREV is not statistically significant.  But the coefficient on RREV*NEWAN is 0.075 and significant at .05 level in model 3, and it is 0.059 and significant at .01 level in model 4.  Results suggest that the FRIC of new analysts is higher than FRIC of existing analysts in the POST1 period. Not all material restatement firms have forecast revisions from new analysts in the post-restatement period.  Including such firms in the analysis comparing the FRIC of new analysts with that of existing analysts may introduce selection biases.  To eliminate such concerns, I repeat the analysis in Table 3.5 using a subsample of material restatement firms who have at least one forecast revision made by new analysts and one forecast revision made by existing analysts in the post-restatement period.  I present the results in Table 3.6. Columns 1 of Table 3.6 reports the estimation of equation (2) using the restricted sample in the POST period.  Column 2 repeats the analysis in column 1 after including event fixed effects.  The results in column 1 and 2 are similar. The coefficient on RREV is positive and 86  significant, indicating that FRIC of existing analysts in the POST1 period is 0.246 (model 1) or 0.155 (model 2).  The coefficient on RREV*NEWAN is positive but insignificant in both the columns, suggesting that the FRIC of new analyst and existing analysts are not statistically different in the POST1 period.  Untabulated results reveal that the coefficients (RREV + RREV*SECOND90) are positive and significant across both columns 1 and 2, but coefficients (RREV*NEWAN + RREV*NEWAN*SECOND90) are not statistically different from zero in either model.  Results suggest that FRIC of existing analysts in the POST2 period is 0.231 (model 1) and 0.155 (model 2).  Also, FRIC of new analysts is not statistically different from FRIC of existing analysts in the POST2 period. Columns 3 and 4 of Table 3.6 repeat the analysis done in columns 1 and 2 respectively, but only using the forecast observations in the first 90 days of the post-restatement period (SECOND90=0). In column 3, coefficient on RREV is 0.185 but statistically insignificant at the conventional level.  But the coefficient on RREV*NEWAN is 0.071 and statistically significant at .10 level, suggesting that FRIC of new analysts is higher than FRIC of old analysts in the POST1 period. In column 4, coefficient on RREV is 0.136 and significant at .10 level, suggesting that FRIC of existing analysts in the POST1 period is 0.136.  The coefficient on RREV*NEWAN is 0.046 and significant at .05 level, indicating that FRIC of new analysts is higher than FRIC of existing analysts (higher by 33.8%) in the POST1 period. Results in table 3.5 and table 3.6 support the notion that for material restatement firms, new analysts have higher FRIC than existing analysts in the first 90 days following restatement announcement.  The coefficient on RREV*NEWAN is large and positive across all eight specifications, and statistically significant for six of these specifications.  Untabulated results show that the coefficient on RREV*NEWAN is positive with t-statistics close to 1.60 for the 87  remaining two specifications.  Overall, the results are consistent with hypothesis 2 of this essay.  I acknowledge that I am not able to control for all possible variables that determine FRIC, and that some of those omitted variables might systematically differ between the existing and new analysts.  However, the FRIC of new analysts differs from the FRIC of existing analysts only in the quarter immediately following material restatements, and the effect dissipates in the second quarter following restatement announcement.  This pattern of a difference-in-FRIC between the existing and new analysts is consistent with my hypothesis 2, but it cannot be explained easily by alternative explanations based on omitted variables.  3.5.3. Additional Analysis (Untabulated) Since I examine the change in FRIC over time for restatement firms, the change in macro-economic and industry specific conditions can potentially be driving the results.  To address such concerns, I repeat the analysis done in Table 3.4 using a sample of control firms matched on SIC 3 digit industry, year, size, and fiscal period ending date. I require that control firms cannot have any restatement of their own within the previous or future two years of the restatement date. I apply the same timeline for a restating firm and its matching control firm (the difference is that the control firm does not have an actual restatement announcement on its restatement date).  Untabulated results show that control firms do not experience an increase in FRIC in the post period compared to the benchmark period in any specification.  Additionally, I perform a difference-in-difference analysis, which reveals that the material restatement firms experience a larger change in their FRIC compared to the control firms in the post period but immaterial restatement (error) firms are not different from the control firms.  The analyses 88  performed using the control firms suggest that the spike in FRIC following restatement announcements cannot be explained by changes in macro-economic or industry-wide forces. Results from Table 3.5 and Table 3.6 show that FRIC of new analysts is higher compared to FRIC of existing analysts in the post-restatement period for material restatement firms.  While it is unlikely, such a result could be driven by a general spike in investors’ interest in new analysts when they initiate coverage of any firm regardless of the impact of restatement announcement.  To rule out such a possibility, I repeat the analyses presented in Table 3.5 and Table 3.6 based on forecast revisions in the post-period for the sample of matched control firms and the immaterial restatement (error) firms. The FRIC of new analysts is not different from FRIC of existing analysts in the post-period for the matching control firms (for the entire restatement sample or the material restatement sample).  The new analysts do not experience an increased FRIC compared to the existing analysts even after the announcements of immaterial restatements (errors).  Results from these additional analyses suggest that the increased FRIC of new analysts following the announcement of material restatements cannot be explained by a general increase in investor interest in the time of new coverage initiation.   3.6. Conclusion  I find that the information content of analyst forecast revision goes up in the quarter immediately following the announcement of material restatements (those restatements perceived to be intentional rather than arising from honest mistakes).  My results suggest that investors increase their reliance on analyst information when facing uncertain information environment triggered by material restatements.  This spike in the information content dissipates by the 89  second quarter of the post-restatement period, which suggests that the substitution effect is large enough to be observable only when the uncertainty surrounding the restating firm is very high.   I also find that in the first quarter following the material restatement announcement, investors react more strongly to forecast revisions made by analysts who initiate coverage after the restatement announcement.  Similar to the primary finding of a general increase in the information content of forecast revisions following restatement for all analysts, investors’ preference for the new analysts dissipate in the second quarter.  The new analysts are less likely to have close ties with the management, and they cannot be held responsible for failing to detect the fraud earlier than the official announcement date.  My results suggest that in the uncertain information environment following a material restatement, investors consider the new analysts to be relatively unbiased. The findings of this paper sheds light on the interdependence of two important information channels in the corporate information environment by showing that a shock in one channel (management disclosure) can impact investor reliance on the other channel (analyst disclosure).  The paper builds on and adds to the intersection of the literature on accounting restatement and the literature on sell-side analysts.  The results indicate that investors turn to analysts when the credibility of management disclosure is compromised.  The findings confirm the important role sell-side analysts play in the capital market.  The findings also shed light on the motivation for analysts to maintain or initiate coverage in the restatement firms.  The fact that investors value analyst information more highly after the restatement announcement is likely to be an important reason why analysts continue or initiate their coverage of material restatement firms.  90  Figure 3.1. Time line of events (basic)  Compare POST period with PRE period         Figure 3.2. Time line of events (split)  Compare POST1 with PRE1  and compare POST2 with PRE2 period     91  Table 3.1. Variable Definitions Variables Definitions Dependent Variables and Variables of Interest CAR Market reaction, defined as the cumulative abnormal return over three trading days (−1,0,1) relative to the analyst forecast date. Abnormal return is calculated by subtracting the CRSP value-weighted market return from the firm return. REVISION Revision in analyst forecast scaled by stock price at the month end immediately prior to the forecast date.      REVISIONf,i,t = [FORECASTf,i,t  − FORECASTf,i,t−1] / PRICE Where FORECASTf,i,t is the annual forecast made by analyst i at time t for the fiscal period f . Forecast information is available from IBES and price information is available from CRSP Monthly Stock File. RREV For each calendar month, forecast REVISIONs for all firms available in IBES is calculated, and then ranked into decile.  RREV is the monthly decile rank of REVISION. POST Indicator Variable equal to 1 if the forecast is made during the POST period spanning 180 days (from T+5 through T+184), and 0 otherwise (See Figure 3.1). The official restatement announcement date (T) is available from Audit Analytics (AA) Non-Reliance database, and the 9 days surrounding the restatement date (from T-4 through T+4) is set as the restatement window. SECOND90 Indicator variable that splits the 180 days of both the PRE and the POST periods into two (See Figure 3.2).   SECOND90 is equal to 1 if  i) the forecast is made during the second 90-day portion of the PRE period (from TB +95 to TB+184, where TB is the start of the benchmark period), or  ii) the forecast is made during the second 90-day portion of the POST period (from T+95 to T+184),  and 0 otherwise. Control Variables SIZE Market value of the firm equity in millions, measured as of the end of the month prior to the revision. Monthly price and shares outstanding data available from CRSP. INFOFLOW The amount of information disseminated about the firm in the run up to the forecast revision, defined as the natural log of one plus the number of days on which analyst reports were issued in the 30 days prior to the forecast revision, collected from IBES BM Book-to-market ratio as of the end of the month prior to the revision. Where Book Value is from COMPUSTAT Quarterly Fundamental file  92  Table 3.1 (Continued) Variables Definitions  at the end of the quarter prior to the revision.  For firms with negative equity, BM is set to 0. NUMFIRMS Natural log of one plus the number of firms covered by the analyst, estimated as the number of firms for which the analyst issued at least one report (of any type) in the last three months BROKER Size of the brokerage house that employs the analyst issuing the report. Size is measured as the natural log of one plus the number of analysts working in the brokerage house, estimated as the number of distinct analysts who issued at least one report in the last three months EXPERIENCE Natural log of one plus the time (in years) since the analyst began coverage in IBES, estimated as the time difference between the reporting day and the date of the earliest report issued by the analyst HORIZON Natural log of one plus the number of days from revision until the corresponding forecast period end date.  If forecast is made after the forecast period ending date, then this value is coded as 0. INNOVATIVE Indicator variable equal to 1 if the analyst’s revision moves away (relative to the previous forecast) from the consensus and 0 otherwise. Consensus is the mean analyst forecast over the prior three months, where only the most recent forecast is retained for each analyst       93  Table 3.2. Restatement Sample  Panel A: Sample Selection Restatements  Sample Size Restatements from Audit Analytics (AA) from January 2000 to December 2013  13899   Remove firms without COMPUSTAT coverage   -3482  Remove firms without CRSP coverage   -4642  Remove firms without IBES coverage   -477  Remove observations with missing book-value, price, shares outstanding data from COMPUSTAT; require that sample firms exist in CRSP for consecutive 12 months prior to restatement   -357  Remove firms that have announced restatements within past two years   -1160  Remove income increasing restatements   -599  Require that the restating firm has annual forecast revisions during both the benchmark period and the post period; require that such forecasts do not coincide with earnings announcement windows; require that such forecast revisions have the necessary return data    -1371 Final Restatement Sample  1811    Panel B: Classification of Restatements by Severity  Restatements in the Final Sample   Material Restatements^  367 Other Restatements  1444 Total  1811  ^ I classify a restatement as material if  i) there is an express admission of fraud in the restatement disclosure, or ii) the restatement disclosure cites SEC investigation, or iii) there is a class-action lawsuit filed within one year of the restatement filing. The information required for the classification of material restatements is available from Audit Analytics Non-Reliance database   94  Table 3.3. Descriptive statistics of regression variables  Panel A: All Restatements   Pre-restatement (Benchmark) Period   N  MEAN  P25  P50  P75 CAR  19736  -0.009  -0.031  -0.004  0.022 REVISION  19736  -0.007  -0.005  -0.001  0.001 SIZE  19736  8672.6  702.4  2244.1  8034.8 INFOFLOW  19736  5.4  2  5  8 BM  19736  0.635  0.290  0.476  0.783 NUMFIRMS  19736  15.7  11  15  19 BROKER  19736  56.7  16  40  88 EXPERIENCE  19736  8.0  2.9  6.5  11.2 HORIZON  19736  164.2  86  174  252 INNOVATIVE  19736  0.449  0  0  1   Post-restatement Period   N  MEAN^  P25  P50  P75 CAR  19353  -0.006***  -0.028  -0.002  0.021 REVISION  19353  -0.009***  -0.005  -0.001  0.001 NEWAN~  19353  0.092  0  0  0 SIZE  19353  7657.2***  802.8  2482.1  7278.8 INFOFLOW  19353  5.5  2  5  8 BM  19353  0.616**  0.286  0.477  0.759 NUMFIRMS  19353  15.9**  11  15  20 BROKER  19353  56.2  17  40  88 EXPERIENCE  19353  7.9  3.0  6.5  11.1 HORIZON  19353  160.8***  79  172  248 INNOVATIVE  19353  0.435**  0  0  1  ^If POST period mean is significantly different from the benchmark period mean, then significance level for the mean difference is reported: *** p<0.01, ** p<0.05, * p<0.1 (two-tailed) ~NEWAN is only meaningful in the post-restatement period.  NEWAN is zero by design in the pre-restatement period and is not reported. Note: All variables reported in their original (unlogged) version   95  Table 3.3. (Continued) Descriptive statistics of regression variables  Panel B: Material Restatements   Pre-restatement (Benchmark) Period   N  MEAN  P25  P50  P75 CAR  4284  -0.011  -0.035  -0.005  0.023 REVISION  4284  -0.010  -0.007  -0.001  0.001 SIZE  4284  13671.6  733.7  2615.9  10650.4 INFOFLOW  4284  5.1  2  5  7 BM  4284  0.651  0.262  0.457  0.771 NUMFIRMS  4284  15.0  10  14  18 BROKER  4284  59.0  16  39  92 EXPERIENCE  4284  7.7  2.8  6.0  10.9 HORIZON  4284  155.1  74  159  250 INNOVATIVE  4284  0.453  0  0  1   Post-restatement Period   N  MEAN^  P25  P50  P75 CAR  4391  -0.014  -0.041  -0.003  0.024 REVISION  4391  -0.016***  -0.009  -0.002  0.001 NEWAN~  4391  0.087***  0  0  0 SIZE  4391  10962.3***  872.3  2554.0  9014.7 INFOFLOW  4391  5.2  2  5  7 BM  4391  0.684  0.280  0.487  0.800 NUMFIRMS  4391  15.2  11  14  18 BROKER  4391  60.8  18  42  93 EXPERIENCE  4391  7.5  3.0  6.2  10.4 HORIZON  4391  147.3***  63  154  236 INNOVATIVE  4391  0.448  0  0  1   ^If POST period mean is significantly different from the benchmark period mean, then significance level for the mean difference is reported: *** p<0.01, ** p<0.05, * p<0.1 (two-tailed) ~NEWAN is only meaningful in the post-restatement period.  NEWAN is zero by design in the pre-restatement period and is not reported. Note: All variables reported in their original (unlogged) version   96  Table 3.4. Multivariate regression examining the change in information content of analyst forecasts following accounting restatement Table reports regression results from the following models:  𝐶𝐴𝑅 =  𝛼0 + 𝛼1𝑅𝑅𝐸𝑉 + 𝛼2𝑃𝑂𝑆𝑇 + 𝛼3𝑅𝑅𝐸𝑉 ∗ 𝑃𝑂𝑆𝑇 +  [𝐶𝑜𝑛𝑡𝑟𝑜𝑙𝑠]   +  𝑅𝑅𝐸𝑉 ∗ [𝐶𝑜𝑛𝑡𝑟𝑜𝑙𝑠]   +   𝜀   (3.1.1)   𝐶𝐴𝑅 =  𝛼0 + 𝛼1𝑅𝑅𝐸𝑉 + 𝛼2𝑃𝑂𝑆𝑇 + 𝛼3𝑅𝑅𝐸𝑉 ∗ 𝑃𝑂𝑆𝑇 +  𝛼4𝑆𝐸𝐶𝑂𝑁𝐷90 +  𝛼5𝑅𝑅𝐸𝑉 ∗ 𝑆𝐸𝐶𝑂𝑁𝐷90   +𝛼6𝑃𝑂𝑆𝑇 ∗ 𝑆𝐸𝐶𝑂𝑁𝐷90 + 𝛼7𝑅𝑅𝐸𝑉 ∗ 𝑃𝑂𝑆𝑇 ∗ 𝑆𝐸𝐶𝑂𝑁𝐷90  +  [𝐶𝑜𝑛𝑡𝑟𝑜𝑙𝑠]   +    𝑅𝑅𝐸𝑉 ∗ [𝐶𝑜𝑛𝑡𝑟𝑜𝑙𝑠]   +   𝜀   (3.1.2)    All Restatements; All Forecasts  Material Restatements; All Forecasts  Material Restatements; Excluding Forecasts by New Analysts    (1)  (2)  (3)  (4)  (5) VARIABLES  CAR  CAR  CAR  CAR  CAR                  RREV  0.101***  0.103***  0.100**  0.072*  0.067* POST  0.004  -0.011  -0.023*  -0.026***  -0.025*** RREV*POST  -0.002  0.022*  0.035*  0.035***  0.033** SECOND90      -0.003  -0.010  -0.010 RREV*SECOND90      0.012  0.021  0.021 POST*SECOND90      0.028*  0.033**  0.031** RREV*POST*SECOND90      -0.029  -0.038*  -0.037*            RREV*SIZE  -0.001  -0.001  -0.001  0.003  0.004 RREV*INFOFLOW  -0.025***  -0.036***  -0.039***  -0.041***  -0.042*** RREV*BM  -0.013***  -0.015**  -0.015**  -0.014  -0.013 RREV*NUMFIRM  -0.004  0.002  0.003  0.001  0.002 RREV*BROKER  0.006***  0.007**  0.008**  0.006**  0.006** RREV*EXPERIENCE  0.003  0.005  0.005  0.002  0.004 RREV*HORIZON  -0.004**  -0.007*  -0.007*  -0.006*  -0.006* RREV*INNOVATE  0.040***  0.043***  0.041***  0.031***  0.032***            Main Effects (Variables)^  YES  YES  YES  YES  YES Event Fixed Effects ~        YES  YES            Observations  39,089  8,675  8,675  8,675  8,295 R-squared  0.0846  0.1082  0.1157  0.0879  0.0856 Errors   Clustered   Clustered   Clustered   Clustered   Clustered Errors double clustered along firms and calendar months.  *** p<0.01, ** p<0.05, * p<0.1 (two-tailed) ^All variables that are interacted with RREV are included ~All variables demeaned for each restatement event instead of putting dummy variables        97  Table 3.5. Multivariate regression examining the information content of forecasts issued by new analysts following accounting restatement  Table reports regression results from the following models:  𝐶𝐴𝑅 =  𝛼0 + 𝛼1𝑅𝑅𝐸𝑉 + 𝛼2𝑁𝐸𝑊𝐴𝑁 +  𝛼3𝑅𝑅𝐸𝑉 ∗ 𝑁𝐸𝑊𝐴𝑁 + 𝛼4𝑆𝐸𝐶𝑂𝑁𝐷90 + 𝛼5𝑅𝑅𝐸𝑉 ∗ 𝑆𝐸𝐶𝑂𝑁𝐷90 +𝛼6𝑁𝐸𝑊𝐴𝑁 ∗ 𝑆𝐸𝐶𝑂𝑁𝐷90 + 𝛼7𝑅𝑅𝐸𝑉 ∗ 𝑁𝐸𝑊𝐴𝑁 ∗ 𝑆𝐸𝐶𝑂𝑁𝐷90 +  [𝐶𝑜𝑛𝑡𝑟𝑜𝑙𝑠]   +    𝑅𝑅𝐸𝑉 ∗ [𝐶𝑜𝑛𝑡𝑟𝑜𝑙𝑠]   +   𝜀   (3.2)     Material Restatements; All Forecasts in the POST period  Material Restatements; Forecasts issued in the first 90 days of the POST period     (1)  (2)  (3)  (4) VARIABLES  CAR  CAR  CAR  CAR               RREV  0.132**  0.083  0.067  0.027 NEWAN  -0.049**  -0.030*  -0.048**  -0.036*** RREV*NEWAN  0.079**  0.053*  0.075**  0.059*** SECOND90  0.021  0.009     RREV*SECOND90  -0.011  0.002     NEWAN*SECOND90  0.055**  0.038**     RREV*NEWAN*SECOND90  -0.074*  -0.046              RREV*SIZE  -0.004  0.001  0.002  0.006 RREV*INFOFLOW  -0.028  -0.043***  -0.005  -0.033 RREV*BM  -0.027***  -0.016***  -0.038***  -0.034** RREV*NUMFIRM  0.006  0.001  -0.002  -0.010 RREV*BROKER  0.009  0.005  0.010  0.008 RREV*EXPERIENCE  0.002  0.006  0.001  0.008 RREV*HORIZON  -0.006  -0.000  -0.006  0.005 RREV*INNOVATE  0.038**  0.015  0.058***  0.021*          Main Effects (Vaiables) ^  YES  YES  YES  YES Event Fixed Effects    YES    YES          Observations  4,391  4,391  2,404  2,404 R-squared  0.1292  0.1080  0.1454  0.1165 Errors   Clustered   Clustered   Clustered   Clustered Errors double clustered along firms and calendar months.  *** p<0.01, ** p<0.05, * p<0.1 (two-tailed) ^All variables that are interacted with RREV are included ~All variables demeaned for each restatement event instead of putting dummy variables 98  Table 3.6. Multivariate regression examining the information content of forecasts issued by new analysts following accounting restatement (Robustness Analysis)  Table reports regression results from the following models:  𝐶𝐴𝑅 =  𝛼0 +  𝛼1𝑅𝑅𝐸𝑉 + 𝛼2𝑁𝐸𝑊𝐴𝑁 + 𝛼3𝑅𝑅𝐸𝑉 ∗ 𝑁𝐸𝑊𝐴𝑁 + 𝛼4𝑆𝐸𝐶𝑂𝑁𝐷90 + 𝛼5𝑅𝑅𝐸𝑉 ∗ 𝑆𝐸𝐶𝑂𝑁𝐷90 +𝛼6𝑁𝐸𝑊𝐴𝑁 ∗ 𝑆𝐸𝐶𝑂𝑁𝐷90 + 𝛼7𝑅𝑅𝐸𝑉 ∗ 𝑁𝐸𝑊𝐴𝑁 ∗ 𝑆𝐸𝐶𝑂𝑁𝐷90 +  [𝐶𝑜𝑛𝑡𝑟𝑜𝑙𝑠]   +    𝑅𝑅𝐸𝑉 ∗ [𝐶𝑜𝑛𝑡𝑟𝑜𝑙𝑠]   +   𝜀   (3.2)     Material restatement firms with at least one new analyst initiating coverage in the POST period; All Forecasts in the POST period  MATERIAL Restatement firms with at least one new analyst initiating coverage in the POST (first 90 days) period; Forecasts issued in the first 90 days of the POST period    (1)  (2)  (3)  (4) VARIABLES  CAR  CAR  CAR  CAR               RREV  0.246***  0.155**  0.185  0.136* NEWAN  -0.030  -0.021  -0.037  -0.030** RREV*NEWAN  0.062  0.034  0.071*  0.046** SECOND90  0.026  0.015     RREV*SECOND90  -0.015  -0.005     NEWAN*SECOND90  0.047*  0.031*     RREV*NEWAN*SECOND90 -0.065*  -0.035              RREV*SIZE  -0.005  0.001  0.004  0.006 RREV*INFOFLOW  -0.033  -0.043**  -0.005  -0.043* RREV*BM  -0.028***  -0.017  -0.051***  -0.043*** RREV*NUMFIRM  -0.014  -0.003  -0.033**  -0.018* RREV*BROKER  0.005  0.004  0.009  0.009 RREV*EXPERIENCE  -0.006  -0.009  -0.007  -0.007 RREV*HORIZON  -0.009  -0.005  -0.012  -0.003 RREV*INNOVATE  0.037**  0.022*  0.065***  0.024*          Main Effects (Variables)^  YES  YES  YES  YES Event Fixed Effects~    YES    YES          Observations  2,905  2,905  1,555  1,555 R-squared  0.1530  0.1180  0.1789  0.1157 Errors   Clustered   Clustered   Clustered   Clustered Errors double clustered along firms and calendar months.  *** p<0.01, ** p<0.05, * p<0.1 (two-tailed) ^All the control variables that are interacted with RREV are included ~All variables demeaned for each restatement event instead of putting dummy variables  99  Chapter 4: The Foreign Investor Bias and its Linguistic Origins   Co-authored with Russell Lundholm (University of British Columbia) and Rafael Rogo (University of British Columbia)  4.1. Introduction   It is well documented that international investors have a bias that causes them to underweight foreign stocks in their portfolio, and at a rate that is increasing in the distance from their domestic country. Because financial assets are weightless, the distance between the foreign investor and the domestic firm must be a proxy for some other, more fundamental, source of bias. Country differences in accounting rules, investor protection laws, cultural norms, and language have been shown to predict the degree of bias, with varying degrees of success. Unfortunately, different countries present packages of these attributes, so that comparing the foreign investor bias across countries necessarily varies many of these factors at the same time.   In this context, foreign investment in Canada presents an interesting case. Since the Constitution Act of 1876, Canada has had two official languages, French and English, with French spoken primarily in Quebec, and English spoken primarily in the other nine provinces. However, the accounting rules are the same across the country, regulatory filings and accounting disclosures are prepared in both languages, and the geographic distance from most foreign investors is approximately the same. Thus, while some differences remain between firms located in Quebec and firms located in the Rest of Canada (ROC), many of the differences that arise when comparing firms in different countries do not exist.  100   In this study we find a surprising result: despite an almost identical information environment and very close geographic proximity, and after controlling for a host of firm-level characteristics, we find that US institutional investors have a significantly larger bias against firms located in Quebec than firms located in the ROC.  In addition, by contrasting the bias within Quebec firms based on their English versus French online presence, and by contrasting the bias of UK versus French institutional investors, we show that misaligned language between the investor and the firm is a significant cause of the differential bias between Quebec and the ROC. Although other studies have found that misaligned language may contribute to the foreign investor bias, our study does so with considerably fewer alternative explanations, and shows that even sophisticated institutional investors living in close proximity to the firms in a foreign country are greatly deterred by a foreign language.17  Quebec’s place among Canadian provinces is unique.  France colonized the region following the explorations of Jacque Cartier. France later ceded the region to the British in the Treaty of Paris in 1763; shortly thereafter the British established Quebec. From then to present day, approximately 80 percent of Quebec inhabitants speak French as their first language (Statistics Canada 2011). This contrasts to roughly two percent Francophones elsewhere in Canada.  Coincident with its unique use of the French language, Quebec evolved with many unique business characteristics and cultural differences from the ROC. Finally, Quebec is an economically important part of Canada, representing 20 percent of Canadian GDP in 2012 (Statistics Canada 2012). The question arises, how do foreign investors view firms located in Quebec relative to firms located in the ROC? And, to the extent that we discover a bias against                                                           17 We acknowledge that a region’s language is confounded with other aspects of its culture, and that it would be impossible to disentangle language from culture.  We will sometimes use the phrase ‘language/culture’ to remind the reader of this. 101  Quebec firms, can we attribute some of it to misaligned language between the foreign investor and the Quebec firm?   Our first contrast treats all the unique features of Quebec as a bundle and measures the US institutional investor bias against Quebec relative to the bias against the ROC. Using a variety of specifications, we find compelling evidence of an incremental bias against Quebec. We find that US investors hold significantly smaller percentages of Quebec firms than firms in the ROC and place significantly less weight on Quebec firms than ROC firms in their Canadian portfolio. These results hold after controlling for a battery of firm-specific characteristics, including whether the firm is cross-listed in the US, whether it has a US segment, whether it is incorporated under provincial or federal law, its size, past performance, dividend yield, and number of analysts. The magnitude of the bias against Quebec relative to the ROC is roughly comparable to the difference between being cross-listed in the US or not.  In one specification, US ownership of Quebec firms is 35.9 percent lower than ownership of ROC firms, and the relative portfolio weight on Quebec firms is roughly half the weight put on ROC firms in the Canadian portfolio.  Our next two contrasts are aimed at identifying the effect of language/culture on the size of the bias against Quebec firms. While it is impossible to isolate and quantify the contribution of each unique feature of Quebec firms that might contribute to a foreign investor bias, we control for a number of previously documented sources of bias that differ between firms within Quebec. For our first language contrast we vary attributes of the firm by creating a cross-sectional measure of each Quebec firm’s French presence on the Internet. Specifically, we measure the firm’s ‘Frenchness’ by counting the relative number of French versus English language documents resulting from a Google search of the company name. We find that a 10 102  percent increase in the firm’s Frenchness is associated with a 4.9 percent reduction in US ownership and a 79 percent reduction on the firm’s relative portfolio weight in the US investor’s Canadian portfolio.  Our second language contrast varies the native language of the foreign investor. By contrasting the bias from British institutional investors with the bias from French institutional investors, we change the alignment of language between the firm and investor but keep many other sources of bias constant. After controlling for a number of firm characteristics, we find that the difference between the British bias and the French bias is much larger for Quebec firms than for ROC firms. In one specification the British investors favour ROC firms over Quebec firms while French investors favour Quebec firms over ROC firms.  4.2. Literature Review and Hypothesis Development   French and Poterba (1991) document that, while US firms represent 49 percent of the equity capital in the largest 6 markets, US investors allocate 91 percent of their wealth to US firms. The tilt of investment portfolios away from global diversification in favour of domestic stocks, a phenomenon labeled as the ‘home bias,’ has been shown in many different countries (Tesar and Werner 1995).  With respect to US investment in Canada, Andrade and Chhaochharia (2010) show that in 2001-2006 US investors allocated one percent of their portfolio to Canadian equity, on average, while Canadian equity averaged three percent of the world portfolio over the same period. The most robust empirical predictor of the degree of bias is simply the geodesic distance between the investor’s country and the firm’s country (Portes, Rey and Oh 2001; Portes and Rey 2005).  This result, known as the ‘gravity model,’ is generally thought to be a proxy for more fundamental determinants of the bias, as financial assets do not have ‘weight.’ The search 103  for other notions of distance that capture the true obstacles to foreign investment has spawned a rich literature in accounting, economics, and finance.  We ask two broad questions: 1) do US investors exhibit a bias against Quebec firms that differs from the bias against ROC firms, and 2) does part of any incremental bias against Quebec have a basis in language and its associated cultural attributes? Accordingly, we sort the related literature into 1) differences between Quebec and the ROC that may contribute to an investor bias, but are relatively constant for firms within Quebec, 2) firm attributes or investor attributes that may cause the bias to differ between firms within Quebec or across investor locations.  We also briefly discuss forces that have been found in previous literature to contribute to an investor bias but are notably absent in our study, and forces that influence institutional holdings quite apart from a foreign investor bias.  4.2.1. Sources of Bias that Differ Between Quebec and Rest of Canada  While contrasting Quebec with the ROC holds many country features constant, there are still reasons to believe US investors will have an additional bias against Quebec firms. The bundle of differences between Quebec and the ROC collectively form the explanation for any differences we may find between the regions. 4.2.1.1. Language   The most obvious difference between Quebec and the ROC is the relative use of French; 80 percent of Quebec residents speak French as their first language, as opposed to only 2 percent in the ROC (Statistics Canada 2011). Language can proxy for economic impediments, such as the cost of gathering information, and it can also proxy for psychological impediments – a firm in a region that speaks a different language feels less familiar to a foreign investor than a firm in 104  a region that speaks the investor’s domestic language. 18  Consistent with this, in a study of 97 Finnish firms and local Finnish investors, Grinblatt and Keloharju (2001) find the Finnish-speaking household investors favour firms that publish their annual reports only in Finnish, while Swedish-speaking household investors prefer firms that publish their annual reports only in Swedish (Finland has two official languages); in contrast to our results, they find no language affect for institutional investors. In global studies, there is mixed evidence that sharing a common language impacts the flow of cross-border investment. Aviat and Coeurdacier (2007) find no evidence that a common language matters in cross-border trade, Portes and Rey (2005) and Daude and Fratzcher (2006) find mixed results for equity investments, while Lane and Milesi-Ferretti (2008), Beugelsdijk and Frijns (2010), and Aggarwal, Kearney and Lucey (2012) find that sharing a common language improves cross-border equity investment.19   An interesting feature of the information environment in Canada in this regard is that all Quebec firms file regulatory reports in SEDAR in both English and French.  While these filings do not constitute the complete information environment, the fact that annual and interim financial statements, MD&A discussions, and material change reports are all available in English certainly lowers the information processing costs, the information advantage of local investors, and the psychological distance of Quebec firms relative to firms located in the ROC. Whether this ‘levels the playing field’ for a US Anglophone investor is an empirical question.                                                           18 Cao, Han, Hirshleifer and Zhang (2011) present a model where the uncertainty created by unfamiliar assets induces pessimism in investors, resulting in less demand.  Due to differences in language and culture, it is easy to imagine that Quebec companies feel less familiar to US institutional investors than companies located in the ROC, and the model would therefore predict lower investment in Quebec companies.  In their model, as in our results, the bias can be so extreme that a status quo of zero holdings of Quebec companies is a possible equilibrium outcome.  19 While the results of these global studies are informative, we note that, unlike our study, global studies vary many things besides language when they compare across countries.  In addition, these studies aggregate investment at the country level, while we collect data at the firm level.  This allows us to control for variation in firm characteristics that have been shown to influence institutional investment, and that gets lost in country aggregation. 105  4.2.1.2. Culture  Beyond the difference in language, there are many other cultural differences between Quebec and the ROC that may influence a US investor’s perception of the firm. Two recent global studies of cross-border investment, Beugelsdijk and Frijns (2010) and Aggarwal et al. (2012), find that cultural distance is a significant deterrent to foreign investment in developed countries, even after controlling for geographic distance and the effect of sharing a common language.  First introduced by Kogut and Singh (1988), the cultural distance measure is derived from Hofstede’s four cultural dimensions, and is often used in international business research (e.g., Loree and Guisinger 1995, Barkema and Vermeulen 1997, Brouthers and Brouthers 2001).20 For each country, the Hofstede Centre gives a score ranging from 1 to 120 on the dimensions of power distance, individualism, uncertainty avoidance, and masculinity. The cultural difference between any two countries is given as the Euclidean distance between the two countries (i.e. the square root of the sum of the squared differences). Along with country scores, the Hofstede Centre gives scores for Quebec separately. The four scores for Canada are (39, 80, 52, 48), as compared to (54, 73, 45, 60) for Quebec, and (40, 91, 62, 46) for the US. Thus, the cultural distance between Canada and the US is 15 and the distance between Quebec and the US is 32. As a point of reference, the distance between England and the US is 34. Aggarwal et al. (2012) find that the impact of cultural distance is positively related to geographic distance and so the close proximity of Canada and the US may render the cultural distinction between Quebec and the ROC irrelevant. Further, they find that greater levels of power distance and masculinity in the foreign investment destination country are positively                                                           20 The cultural measures given in Hofstede, Hofstede and Minkov’s Culture and Organizations (1997) are arguably the most common proxies for a country’s culture, having been cited more than 21,000 times according to Google Scholar. 106  associated with more cross-border investment, which would favour Quebec over the ROC because Quebec scores higher on these two dimensions.  It is therefore an empirical question whether the differences in relatively “close” cultures is sufficient to generate an incremental investor bias against Quebec or if the aspects of Quebec’s culture that are more masculine with more unequally distributed power actually attracts US investment. 4.2.1.3. Political Risk  The Parti Quebecois (PQ) is a separatist political party in Quebec that advocates national sovereignty for Quebec. Such an outcome could raise significant concerns for US investors who might fear a disrupted business environment, changes in the relative power of labour markets, severe limits on mergers with firms from outside Quebec, and the possible expropriation of assets. The PQ first took power in 1976, with a failed referendum for secession in 1980.  They were defeated in 1985 but returned to power in 1994 with another referendum for secession in 1995.  The referendum failed, but by a margin of less than one percent.  The PQ was defeated in 2003, and fell to third place in the 2007 election.  They won a minority government in 2012, but after calling an election in 2014, they lost power to the liberal party, and won only 25 percent of the popular vote, the worst result since 1970.  Although the PQ’s political power ebbs and flows, their continued existence as a major party in Quebec poses a political risk to foreign investors.  Evidence that investors price the political risk associated with Quebec’s intermittent sovereignty movement can be found in Beaulieu, Cosset and Essaddam (2006) who study stock price movements of Quebec firms around the 1995 referendum. Polls prior to the vote could not reliably predict the outcome, making stock returns prior to the vote sensitive to the uncertainty of the outcome, and stock returns surrounding the vote sensitive to the resolution of that uncertainty. The authors find that in the week prior to the vote, the stock returns of Quebec firms 107  were significantly negative, and then turned significantly positive in response to the NO vote on the referendum. In addition, Tirtiroglu, Bhabra and Lel (2004) find positive stock price reactions to announcements that a firm is moving its headquarters, plants, or divisions away from Quebec. Finally, Graham, Morrill and Morrill (2005) find that firms headquartered in Quebec tend to have higher book-to-price ratios than firms headquartered elsewhere in Canada. Equity prices clearly behave as if firms in Quebec come with political risk to foreign investors.  However, prior research does not study how Quebec’s political risk, and its associated price discount, influence US investor holdings in Quebec versus the ROC.  4.2.1.4. Investor Protection  It has been argued that the foreign investor bias is partly driven by differences in the legal protection afforded foreign investors. La Porta, Silanes, Shleifer and Vishny (1997) argue that common law countries provide greater investor protection than civil law countries, both in terms of specific laws and because of better enforcement. Because Quebec is a civil law jurisdiction, while the rest of the Canadian provinces and the federal government are common law jurisdictions, this might support a foreign investor bias against Quebec. But perhaps not surprisingly, the La Porta measures of investor protection, and the conclusions drawn from them, have been disputed by legal scholars. Spamann (2010) finds that when the components of the La Porta binary coding system are augmented to a richer coding system, there is no longer a difference between common law and civil law countries in terms of investor protection. Further, Puri (2009) writes that the La Porta et al. (1997) arguments are too broad-brushed to apply simplistically to Quebec versus the ROC. The author notes that the organization of the Canadian Securities Administrators in 2003 has helped to harmonize rules across the provinces, and the 108  creation of the National Instruments has aided in setting out common regulations. The author summarizes by stating “This discussion reveals that while Quebec operates a corporate law framework within its civil law system that on the surface provides legal rules that do not offer as much protection as the federal corporate law statute (or other provincial law statues), the Quebec courts have stepped in to judicially craft remedies for shareholders. That being said, these QCA remedies are currently more difficult to access or achieve recourse under than those in the federal statutory regime.” (Puri 2009 p. 1671). We simply note that Quebec operates under a different legal system than the ROC and this may have economic implications for foreign investors. As discussed in the next section, we investigate one component of investor protection that varies across firms within Quebec by examining whether there is a difference between firms who incorporate under Quebec provincial law versus those who incorporate under federal law.21  4.2.1.5. Forces that are the same between Quebec and the Rest of Canada  Some previously documented drivers of the foreign investor bias are notably absent from our study. The accounting rules are the same across Canada, eliminating one prominent and frequently studied cause of the bias.22  And, as discussed previously, all significant regulatory disclosures – annual reports, MD&A, material changes - are produced in English and filed electronically.  In addition, even if geographic distance from the US is a relevant determinant of the foreign investor bias, Quebec and the ROC are approximately the same distance from the US. Thus, our study holds constant many factors that have been shown to have large influences on                                                           21 An investor protection issue that applies across Canada is that each province has its own security regulatory body.  However, under the passport system most provinces agree to respect each other’s registration decisions (Canadian Securities Administrators 2014). Ontario does not officially participate in the passport system, lobbying instead for a national regulatory system.  The other provinces accept Ontario’s decisions while Ontario reserves the right to makes its own decisions. 22 Bradshaw, Bushee and Miller (2004) find that foreign firms with greater degrees of US GAAP conformity have greater levels of US institutional investment.  Interestingly, the authors exclude Canada from the sample, arguing that the accounting rules are so close to US standards that there is no meaningful variation.  Similarly, Covrig, Defond and Hung (2007) find that foreign ownership of a firm is higher for IAS adopters. And in a changes design, Wahid and Yu (2012) find that international mutual fund ownership increases following adoption of IFRS. 109  the foreign investor bias, and therefore narrows the range of explanations for any bias against Quebec that we might observe. 23 4.2.1.6. Hypothesis 1  Collectively, the differences between Quebec and the ROC in language, culture, political risk, and investor protection may give rise to a US investor bias against Quebec that is over and above a bias against the ROC. Or it may be that the differences between these forces are sufficiently minor that, when combined with the uniformity of accounting rules and the small and roughly equal geographic distance from the US, there is no measurable difference on US investor behaviour.  Therefore, we empirically assess the following alternative hypothesis. H1: The US institutional investor bias against Quebec firms is greater than the bias against firms in the Rest of Canada.  4.2.2. Does Part of the Investor Bias against Quebec have a Basis in Language?   The collective differences between Quebec and the Rest of Canada, to the extent that they exert a differential impact on US investor holdings, are mutually confounded. To gain some insight into the relative contribution of different sources of bias, in this section we develop hypotheses about forces that will vary within the Quebec firm sample, and forces that differ across different foreign investor populations. Our emphasis is on investigating how differences                                                           23 Transaction costs are virtually the same for investments in Quebec versus the ROC. However, transaction costs have generally been dismissed as a viable cause for the foreign investor bias. Rowland (1999) shows that international portfolios turn over faster than similar domestic portfolios, suggesting that transaction costs are unlikely to be significantly higher for international investments. 110  in language/culture contribute to the differential foreign investor bias against Quebec firms.  We develop two hypotheses in this regard. 4.2.2.1. Frenchness  Prior research has treated language differences between the investor location and the firm location as a country-level dichotomous variable – either the investor’s country shares the same language as the firm’s country or it does not (e.g. Beugelsdijk and Frijns 2010 and Aggarwal et al. 2012, Lundholm et al. 2014).  Such an approach cannot disentangle language from other regional influences.  To sidestep this problem, we create a measure of “Frenchness” for each Quebec firm. Within Quebec, firms can have very different levels of English versus French personas. Some firms may translate every voluntary disclosure, do interviews in both languages, staff investor relations departments with both Anglophones and Francophones, while other firms may do little to help the non-Francophone investor. We create a comprehensive proxy for this tendency by measuring the relative number of firm-related documents on the web published in French or English.  We describe this measure, labeled ‘Frenchness,’ in detail in the next section, but for now it is interesting to note that it ranges from 0 to 67 percent of online documents, suggesting that there is considerable variation in the mix of language that a foreign investor might find.24 If the foreign investor bias against Quebec firms varies with the degree of Frenchness, this is an indicator that language, along with the cultural aspects that it proxies for, is a critical part of the reluctance of US investors to own Quebec stocks.                                                            24 We observe variation in the mix of French and English despite Quebec’s Bill 101, which requires all businesses with more than 50 employees to use French in signage, product labels, manuals, software, business communications, along with every employee’s right to conduct his or her work in French.  However, a second language is allowed in addition to the mandatory French as long as the French version is the most prominent.  111  4.2.2.2. Type of Incorporation  Firms within Quebec can also vary in ways that influence the degree of investor protection.  Boubraki, Bozec, Laurin and Rousseau (2011) find that many Quebec firms incorporate under federal law rather than under provincial law, arguing that federal law is more protective of investor rights.25 Of our Quebec firms, 36 percent are incorporated under Quebec provincial law, providing an interesting contrast in investor protection within Quebec.  4.2.2.3 Cross-listing  A Quebec firm can also enhance foreign investor protection, and attract US investors, by cross-listing its stock on a US exchange. In a global study Ahearne, Griever and Warnock (2004) find that US investors hold more equity in countries who have a greater fraction of their firms cross-listed in the US, arguing that this lowers the information asymmetry.  In a valuation study, King and Segal (2009) find that Tobin’s Q increases for a Canadian firm that cross-lists on a US exchange, but only if the firm’s US investor base also increases. The authors do not distinguish between Quebec firms and firms from the ROC. Further, they attribute their results to an increase in investor recognition (Merton 1987) rather than heightened investor protection. Roughly 30 percent of the Quebec firms in our sample have cross-listed shares in the US, thus providing another within-Quebec contrast on this determinant of the US investor bias.                                                            25 In 2011 Quebec replaced the Quebec Companies Act with Quebec Business Corporations Act.  The change was designed, in part, to provide better protection to shareholders under provincial law.  As this event occurred at the very end of our sample period, we do not attempt to measure its impact. 112  4.2.2.4. Differential Exposure to Political Risk  Beyond the different types of incorporation, Quebec firms can have different exposures to the political risk that Quebec presents.  Beaulieu et al. (2006) find that the stock price changes around the 1995 Quebec succession referendum were more pronounced for purely domestic Quebec firms than for multinational firms with headquarters in Quebec.  They argue that firms whose value is based largely on growth options have less exposure to Quebec’s political risk because they can move those growth options to other locations at relatively low cost.  Similarly, they argue that a Quebec firm that has foreign segments has less exposure to political risk inside Quebec. And, from a US investor prospective, the political risk may be lowest if the foreign segment is in the United States. Building on the variation in political risk that different Quebec firms pose, Graham et al. (2012) use the number of employees, the amount of revenue, and the value of assets located in and out of Quebec, as reported in an annual survey by Les Affairs, to measure the firm’s political risk. They find lower valuation multiples on the firm’s book value and earnings as the amount of assets located in Quebec increases.26 4.2.2.5. Hypothesis 2  Our treatment variable of interest is the firm’s degree of “Frenchness” on the Internet. Based on prior research, we also control for differences in the type of incorporation, cross-listing in the US, and exposure to political risk (which we proxy for with the existence of a US segment).27 We note that all of these cross-sectional differences between firms in Quebec are                                                           26 The authors of Graham et al. (2005) graciously provided their hand-collected data on the number of employees in and out of Quebec.  This variable was insignificant in our tests, probably because intersecting this data with ours greatly reduced the sample size. More importantly, none of our results where significantly affected. 27 Beaulieu et al. (2006) proxy for the political risk of Quebec firms with two variables: 1) the book-to-market ratio, which is in our study as a firm characteristic control because Bradshaw et al. (2004) show that institutional investors prefer low book-to-market ratios, and 2) an indicator for a non-Canadian segment.  We use the existence of a US segment as our political risk proxy, rather than the existence of a non-Canadian segment, because this 113  firm choices and therefore may be the product of even more fundamental influences. Also, because these attributes are relatively stable over time, we cannot attribute causation. In particular, it may be the case that US investors are repelled by firms with high Frenchness, or it may be the case that firms choose a low level of Frenchness because they have many US investors, or it may be the case that both variables are jointly determined over many years. H2: The US investor bias against Quebec firms increases with a firm’s “Frenchness.” 4.2.2.6. Hypothesis 3  In global studies of mutual fund holdings, Chan, Corvig and Ng (2005), Beugelsdijk and Frijns (2010), and Aggarwal et al. (2012) find that sharing a common language reduces the investor bias against foreign countries.  Note that these results measure cross-border investments at the country level. They do not take into account firm-level attributes, such as the firm’s location within the country, the firm choices outlined in hypothesis two, or the host of other firm attributes that have been shown to influence institutional investment (as discussed next). If misaligned language contributes to a foreign investor bias, we would expect the bias against Quebec to be stronger for investors residing in an Anglophone country than for investors residing in a Francophone country.  We choose France and the UK as the investor locations for this contrast because they each have a reasonable number of institutional investors and they are approximately the same geodesic distance from Canada. In addition, by choosing two countries that are significantly further from Canada than the US we reduce the potential level of familiarity that proximity alone may have provided the US investor. Finally, note that if any of the non-                                                          seems most relevant to US investors. Practically speaking, it makes little difference because the two indicator variables are correlated at the .99 level. 114  language features of Quebec (political risk, investor protection, etc.) are the root cause of the bias then this is the same for French and British investors, and we would find no difference. H3: The bias against Quebec firms, relative to the bias against firms in the Rest of Canada, is greater for British institutional investors than it is for French institutional investors.  4.2.3. Determinants of Institutional Holdings Unrelated to Home Bias   Kang and Stulz (1997) and Bradshaw et al. (2004) identify a number of firm-specific factors that attract institutional investors.  And, in a study of Swedish firms’ foreign ownership, Dahlquist and Robertsson (2001) find that many of the firm attributes that are attractive to institutional investors in their domestic portfolios (e.g. size, low dividend payout) are also present in the Swedish firms that are targeted by foreign investors, particularly US investors.  They argue that some previous evidence regarding a foreign investor home bias is actually just an institutional investor bias in favour of certain firm-specific attributes found in foreign firms.  For this reason, in addition to the variables discussed above, we control for the firm-specific factors found in these studies.  We also include industry fixed effects to account for the different mix of industries in Quebec versus the ROC, and we include the provincial marginal tax rate as a proxy for business costs that vary across provinces. And, to satisfy the exclusion restriction in the two-stage models that follow, we use an indicator for whether the firm is part of the S&PTSX index when modeling the decision for institutional ownership to be zero or positive.   115  4.3. Research Design 4.3.1. Dependent Variables  There are two common ways to measure the degree of foreign investor bias against a domestic stock: 1) the percent of shares held by specific foreign investors (either in the US, the UK or France), and 2) the portfolio weight of the firm in the foreign investor’s Canadian holdings relative to the weight of the firm in the Canadian market. The benchmark for both ratios is given by the international CAPM; in this model every investor should hold the market-value-weighted world portfolio. In this case the percent of shares held by specific foreign investors should be constant across all firms in Canada, regardless of whether they are in Quebec or ROC. The percent of shares held is used by Bradshaw et al. (2004) to study how accounting differences impact the home bias and by DeFond, Hu and Li (2011) to study the impact of mandatory IFRS adoption in the European Union. It is the most commonly used measure in the institutional ownership literature (see Bushee 1998 and 2001).  Similarly, under the international CAPM the relative portfolio weight should be one for all securities, regardless of their location within Canada. The relative portfolio weight is used in Dahlquist and Robertsson (2001) to study how firm characteristics influence foreign investor ownership in Sweden. It is a firm-level version of the country-level variable used in most economics studies of aggregate cross-border investment, including Dahlquist, Pinkowitz, Stulz and Williamson (2003), Chan et al. (2005), and Beugelsdijk and Frijns (2010).  Note that a general home bias against Canada will not change the relative value between Quebec and ROC for either measure; it depresses the percent of shares held equally across all Canadian firms and it has no effect on the relative portfolio weight because the allocated share of wealth is relative to the investor’s holdings in Canada. The first measure is from a firm 116  perspective – who owns the firm’s shares.  The second measure is from the investor’s perspective – how do they allocate their wealth between Canadian firms?  In both cases, the unit of observation is the firm-year because we only consider one group of foreign investors at a time.   For a given Canadian firm and year, the percent of shares held, PCT_US, is defined as the number of shares held by U.S. institutional investors, measured as of the institutional report date closest to the fiscal year end, divided by the float-adjusted shares. Float-adjusted shares are shares outstanding less the number of shares held by ‘control blocks’ at the end of the fiscal year (COMPUSTAT item CSFSM). Control blocks are assumed to be held partly for strategic and not solely for investment purposes; consequently, they are not generally available for sale to foreign investors.28 Dahlquist et al. (2003) show that investor bias estimates based on the total shares outstanding severely distort the result toward a biased measure of investment even if no bias is actually present, and Attig (2007) presents evidence that closely held shares are particularly prevalent in Quebec. We use float-adjusted shares rather than total shares outstanding to avoid biasing our ‘bias’ estimate, but as a practical matter our results are virtually the same throughout the paper if we use total shares outstanding as the denominator.  The relative portfolio weight, RPW_US, is defined as the weight of the firm in US investors’ Canadian portfolio compared to its weight in Canadian equity market.  Let 𝑀𝑉𝑗𝑈𝑆 denote the total dollar amount invested by the US institutional investors in firm j. If there are N firms in the Canadian equity market, then firm j’s weight in the US investors’ Canadian portfolio is 𝑊𝑗𝑈𝑆 = 𝑀𝑉𝑗𝑈𝑆/ ∑ 𝑀𝑉𝑗𝑈𝑆𝑁𝑗=1 .  Similarly, let 𝑀𝑉𝑗 denote the market value of firm j. Then firm                                                           28S&P computes control blocks as those shares that are a) held by another corporation; b) shares held by government entities; c) shares held by current and former officers, directors, founders or family trusts thereof; and d) shares held by trusts, pension funds, and employee stock ownership programs controlled by the company. If the sum of such shares exceeds 10 percent of shares outstanding, then this sum is subtracted from the shares outstanding to arrive at the float-adjusted number of shares.  For a more detailed description, see Standard and Poor’s (2011). 117  j’s weight in the Canadian market portfolio is 𝑊𝑗𝑀 = 𝑀𝑉𝑗/ ∑ 𝑀𝑉𝑗𝑁𝑗=1 .  The relative portfolio weight of firm j for US investors is thus defined as: RPW_US =  𝑊𝑗𝑈𝑆/ 𝑊𝑗𝑀 . As with the PCT_US, all calculations use the float-adjusted number of shares. In the second part of our analysis we contrast British and French investments in Canada – for these measures we define PCT_UK, PCT_FRANCE, RPW_UK, and RPW_FRANCE exactly as described above, but substitute in the relevant country’s institutional holdings data.   4.3.2. Independent Variables   For the first and third hypothesis our treatment variable of interest is QC, an indicator variable that takes the value of one for firms headquartered in Quebec and the value zero for firms headquartered in the ROC (as given by COMPUSTAT). For the second hypothesis the treatment variable is FRENCHNESS, which measures a firm’s online presence in French versus English (as described later).  As discussed in the previous section, we use a common set of control variables in all our models to capture firm characteristics that previous literature has found to influence institutional investment quite apart from any foreign investor bias (precise definitions are given in Table 4.1).  Following Bradshaw et al. (2004), we control for firm SIZE, firm AGE, whether the firm reports US segment sales (USSEGMENT), whether the firm is cross-listed on a US exchange (USCROSS), the number of analysts providing forecasts (NANLST), the fiscal year’s sales growth (GROWTH), the return on beginning equity (ROE), the debt-to-asset ratio (LEVERAGE), the earnings-to-price ratio (EP), the book-to-price ratio (BP), the dividend yield (DP), the raw stock return over the fiscal year (RRET), and whether the firm uses a Big 4 auditor (BIG4). All income statement variables are measured over the fiscal year and all balance sheet variables are 118  measured as of the end of the fiscal year. In addition, because Quebec may represent a different mix of industries than the rest of Canada, and institutional investors may have industry preferences, all our tests include industry fixed effects.  As discussed in the literature review, USCROSS and USSEGMENT have been used in prior valuation studies of Quebec, and so we are careful to report their effects in all our tables. In addition, we include an indicator variable PROV_INCP equal to one if the firm is incorporated under its provincial law, and equal to zero if it is incorporated under the Canadian Business Corporation Act or other federal regulation; this variable has also been used in previous studies of Quebec.  These three variables are particularly important for the within-Quebec analysis (which tests hypothesis two) in order to isolate the effect of FRENCHNESS. Finally, we include the provincial marginal tax rate (PROV_TAX) as a proxy for how friendly the province is to business, and in the models that require an exclusion restriction, we include an indicator variable equal to one if the firm is part of the S&PTSX index, and equal to zero otherwise.  Table 4.1 provides detailed definitions of all the variables used in this chapter.  4.3.3. Empirical Models  Our first dependent variable, PCT_US, is a proportion, so it lies on the unit interval, and it has a discontinuously large mass of observations at zero. In addition, the variables and relative weights that cause a firm to have zero US institutional ownership might be different from the variables and weights that determine how much US ownership a firm has, given that it is greater than zero. Consequently, a simple OLS regression is not appropriate because, with OLS a) the predicted value from a linear relationship can extend infinitely beyond the unit interval, and b) the determinants that give rise to a prediction of zero can only be slightly different from the 119  determinants that give rise to a small positive prediction. To better align our model with our data, we borrow a model that has been used to estimate the proportion of debt in a firm’s capital structure.29 Called the zero-inflated beta model, it incorporates the features above, and its application to the capital structure problem is found in Cook et al. (2008).30  Because this model is not commonly used, we provide a brief description here, and give the detailed model in Appendix B1.  We also supplement our analysis of PCT_US with traditional OLS regressions on the full sample and on the sample restricted to non-zero holdings.  The zero-inflated beta model produces two simultaneously-estimated results; the coefficients for a logit model on an indicator for whether PCT_US is zero or greater than zero and, for the observations where PCT_US is greater than zero, coefficients for a model of the level of PCT_US, where the percent is estimated as the mean parameter in a beta distribution (noting that the beta distribution is on the unit interval). To parallel the Heckman model that follows, we label the logit model as the ‘selection model’ and the beta model as the ‘final model.’ The estimation is performed using the ZOIB module in STATA, which allows for fixed effects and clustered standard errors. Finally, for some specifications we limit the data to the observations with positive PCT_US and, for these models, we estimate a simple beta model, which retains the property of being constrained on the (0,1) interval but without the selection portion of the zero-inflated beta model.  Our second dependent variable, RPW_US, has the same selection issue as PCT_US in that it also has a mass of observations at zero that are unlikely to have arrived there randomly,                                                           29 Note that the capital structure problem also models a proportion, has a large mass of observations at zero, and there is reason to believe that what causes a firm to have zero leverage might be different from what determines how levered a firm is once the value is greater than zero. 30 There is a rich statistics literature on the issue of proportions with a mass at zero.  See Kieschnick and McCullough (2003), and Ferrari and Cribari-Neto (2004) for solutions to the distributional issues and Heckman (1979) for a discussion of the selection issue that distinguishes the zero observations from the positive ones. 120  but it isn’t constrained to the unit interval. Therefore, for this dependent variable we estimate a Heckman two-stage model, where the first stage is a probit model on an indicator for whether RPW_US is zero or greater than zero and the second stage is an OLS regression that includes the inverse Mills ratio from the first stage.  We also supplement our analysis of this variable with traditional OLS regressions.   Both the zero-inflated beta model and the two-stage Heckman model are much improved if an exclusion restriction is included in the first-stage selection model; this lowers the potential correlation between the errors in the selection model and the errors in the final model.  We use an indicator for inclusion in the Standard and Poor’s TSX index as our exclusion variable (S&PTSX), reasoning that it will have a much greater impact on whether or not US institutional investors have non-zero holdings of the firm than on the level of investment, given that investment is positive.   To summarize, both the zero-inflated beta model and the two-stage Heckman model consist of a selection model and a final model.  All the control variables described above are in both the selection and final models, along with industry and year fixed effects. In addition, the S&PTSX indicator variable is in the selection models but not in the final models, and the inverse Mills ratio from the first stage is in the final stage of the Heckman model.  When the sample is all Canadian firms, the treatment variable of interest is the indicator for a Quebec firm (QC); if there is an additional bias against Quebec firms over ROC firms, then the coefficient on QC will be positive in the selection model (indicating that zero holdings are more likely), and negative in the final model (indicating that holdings are lower for Quebec firms than ROC firms).    We compute standard errors clustered at the year and industry level, allowing for correlation in residuals within years and industries.  Note that this is considerably more 121  conservative than clusters at the firm and year level.  Finally, although we typically have directional hypotheses, we report p-values for two-tailed tests.  4.3.4. The Sample  We begin by identifying all firms headquartered in Canada and listed on the Toronto Stock Exchange (COMPUSTAT EXCHG=7) between the years 2000 and 2012, resulting in an potential sample of 13,673 firm-years, representing 2094 unique firms, of which 233 are located in Quebec.  After requiring that the firm-years in our sample have the necessary data to construct our independent variables we have 9495 firm-years, and after removing companies with a stock price less than $1.00 we have 8089 firm-years, with 1249 firm-years in Quebec and 6840 firm-years in the ROC.   We collect institutional holding information from the Thomson Financial Service database available from WRDS.  For US institutional holdings, we use the S34 file (previously known as the Spectrum database).  The S34 file is based on the 13-f quarterly holdings information filed by Investment Companies with the SEC.  Rule 13(f) requires institutions managing more than $100 million in equity to report their investment positions in all the “Section 13(f) securities” with holdings greater than 10,000 shares or $200,000 in market value.  Section 13(f) securities include all exchange-traded securities, including exchange-traded foreign company securities.  It is also common for US institutional investors to voluntarily report holdings for firms not traded on US exchanges; roughly one third of our Canadian firm-years held by US investors fall into this category.  The voluntary nature of this data would only bias our results if US institutional investors systematically reported Quebec firm ownership differently than ROC firm ownership. Intersecting the Canadian firm sample with the US 122  institutional holdings yields a sample of 2956 firm-years with positive US institutional holdings, although we retain the zero holding observations for most of our tests.  For British and French institutional ownership data in Canada, we rely on the Thomson Financial Services S12 file, which reports holdings of mutual funds.31 The source data for this file is primarily the semi-annual SEC N-30D and N-30Q filings that contain reports to shareholders of the mutual fund.  Thomson also contacts mutual funds directly to update their data. The S12 file contains data for the 3000 largest global funds that have any holdings in US exchange-traded or Canadian exchange-traded firms. Intersecting this institutional holding data with the Canadian firm sample results in a British investor sample of 2778 firm-years with non-zero holdings and a French sample of 1462 firm-years with non-zero holdings (as described in Table 4.9 and discussed later).  4.4. Empirical Results  4.4.1. US Institutional Investor Bias against Quebec versus the Rest of Canada Table 4.2 panel A describes the full Canadian firm sample and panel B describes the subsample with non-zero US institutional investor holdings. Initial evidence of the US investor bias against Quebec is seen in the first row of both panels.32  For the full sample, the mean PCT_US is 3.9 percent for Quebec firms and 5.2 percent for ROC firms; for the subsample with non-zero holdings, the mean US ownership is 11.8 percent for Quebec firms and 13.9 percent for                                                           31 Because mutual funds are just one type of investment company (bank trusts, for example, are another type of investment company), the S12 file is not directly comparable to the S34 file.  Further, the holdings reported in the S34 file are aggregated across all funds under a manager that may be individually reported in the S12 file. 32 The relative sample sizes in panel A and panel B suggest that US institutional investors are more likely to have non-zero holdings in ROC firms than in Quebec firms; comparing across panels, 33 percent of Quebec firms have non-zero US investment while 37 percent of the ROC firms have non-zero US investment.   123  ROC firms.  The comparison is relatively more extreme for medians; for the non-zero holding sample the median ownership is 2.3 percent for Quebec firms and, at 4.7 percent, is more than double for ROC firms.  The second row of each panel shows that, for RPW_US, the pattern across Quebec and the ROC is the same, although the differences in means are not significant. Table 4.2 panel B shows a suggestive pattern in the distribution of relative portfolio weights in the sample of firms with positive values.  The median RPW_US for the Quebec firms is 0.572, implying that, relative to their weight in the Canadian market, Quebec firms are under-weighted by almost 50 percent in US institutional investors’ Canadian portfolios.  In contrast, the median ROC firm has a relative portfolio weight of 1.269, implying that it is over-weighted by almost 27 percent in US institutional investors’ Canadian portfolios.    Examining the control variables in panel A of Table 4.2, we see many firm-specific characteristics of Quebec firms that might deter institutional investment relative to firms in the ROC. Quebec firms are less likely to be cross-listed in the US, they have slower growth, lower past stock returns, and are less likely to employ a BIG4 auditor. These results emphasize the importance of controlling for firm-specific characteristics that might impact foreign investor holdings quite apart from the foreign investor bias (as discussed in Dahlquist and Robertsson 2001).  Interestingly, many of these differences become insignificant once the sample is limited to only non-zero institutional holdings, as seen in panel B of Table 4.2.  This suggests that US institutional investors use these variables to select the subset of Canadian firms to invest in, and this homogenizes the sample of non-zero holdings to some degree.  Notably, there is no significant difference in the mean size of the Quebec firm and the ROC firm, as seen in both panel A and panel B.  This is important because, if it was the case that Quebec firms were significantly larger than ROC firms, and if US institutional investors held equally-weighted as 124  opposed to value-weighted portfolios, then we would observe lower percentage ownership and lower relative portfolio weights in Quebec firms.   Table 4.3 gives correlations between the variables in our study.  Not surprisingly, our two dependent variables are highly correlated.  Examining the correlations of PCT_US and RPW_US with the other variables in the study shows that holdings are significantly positively correlated with whether the firm has a US segment, whether it is cross-listed in the US, firm size, and the number of analysts following the firm. There are also a number of significant correlations between the control variables. While the control variables are not central to the main message of the paper, these high correlations may cause these variables to have unexpected signs in the multivariate models.      To assess the US investor’s bias against Quebec firms versus ROC firms, while controlling for the differences in firm characteristics, we estimate the zero-inflated beta model for PCT_US and the two-stage Heckman model for RPW_US, as described above. Recall that positive coefficients in the selection models indicate that it is more likely to be an observation with zero holdings, while a positive coefficient in the final model indicates that the holdings are increasing in the variable. The first two columns of Table 4.4 give the results for the zero-inflated beta model. As seen in the first column, after controlling for many firm-specific characteristics, the probability that PCT_US=0 is significantly higher for Quebec firms than for ROC firms. The significantly negative coefficient on QC in the second column shows that, after controlling for selection, and all the firm-specific variables, Quebec firms have a significantly lower percentage of shares held by US investors than ROC firms.  125   The coefficients in the zero-inflated beta model, once raised exponentially, are the odds ratio in the selection model and the proportions ratio in the final model.33 Consequently, the 0.277 coefficient on QC in the selection model means that, after controlling for the other variables in the model, the odds of being in the zero-holding sample are 32 percent higher for a Quebec firm than for a ROC firm (exp(0.277)=1.32).  Similarly, the -0.445 coefficient on QC in the final model means that, after controlling for the other variables in the model, the proportion of US investor ownership to non-US investor ownership is 35.9 percent lower for Quebec firms than for ROC firms (exp(-0.445)=0.641, 0.641-1= -.359). In sum, the US institutional investor bias against Quebec firms causes them to invest in substantially fewer Quebec firms than ROC firms and, when they invest, to invest in substantially smaller fractions of Quebec firms than ROC firms.  The results for the two-stage Heckman model, given in the third and fourth columns of Table 4.4 mirror the results from the zero-inflated beta model.  The relative portfolio weight is significantly more likely to be zero for Quebec firms than for ROC firms and, when the weight is positive, it is significantly smaller for Quebec firms than for ROC firms.  The coefficient of -2.174 in the final model is economically significant as well.  In the sample with positive RPW_US, the relative portfolio weight for the mean Quebec firm is close to the mean ROC firm (7.016 and 7.664, respectively, as seen in Table 4.2, panel B).  However, once the firm-specific controls are added to the model, and the selection effect is controlled for with a highly significant inverse Mills ratio, the Quebec firms’ relative portfolio weights are 2.174 lower than the ROC                                                           33 As shown in Appendix 1, the coefficient on QC in the final stage model is 𝛽𝑄𝐶 = 𝑙𝑜𝑔 (𝜇𝑄(1−𝜇𝑄)⁄𝜇𝑅𝑂𝐶(1−𝜇𝑅𝑂𝐶)⁄) , where 𝜇Q is the estimated PCT_US for a Quebec firm and 𝜇ROC is the estimated PCT_US for a ROC firm, and so the exponentiated value of the coefficient is the proportions ratio. 126  firms’ relative portfolio weights.  In sum, the results for RPW_US echo the results for PCT_US; consistent with hypothesis one, US institutional investors have a bias against Quebec firms that is significantly greater than the bias against ROC firms.  Both models in Table 4.4 include indicators for provincial incorporation (PROV_INCP), US cross-listing (USCROSS), and a US segment (USSEGMENT), as well as the provincial tax rate (PROV_TAX).  These variables are primarily intended as controls when we examine within-Quebec variation in the foreign investor bias, but we include them in the full sample models for completeness.  Across the full sample of Canadian firms, US institutional investors are more likely to hold greater percentages of US cross-listed firms and firms with US segments, as seen in column 2, and they are less likely to have zero holdings in these firms, as seen in column 1. The results for the relative portfolio weights are less clear; the selection model shows US cross-listed and US-segment firms are less likely to have zero US holdings but these variables are insignificant in the final model. Finally, column 2 shows that PCT_US is significantly lower for firms in provinces with higher tax rates.  The other control variables behave largely as expected in the PCT_US model (not surprisingly because PCT_US and the control variables are both taken from Bradshaw et al. 2004).  US investor ownership increases with firm size (SIZE), with analyst coverage (NANLST), and decreases with the dividend yield (DP).  Inconsistent with Bradshaw et al. (2004), ownership decreases with the raw stock return (RRET) and with the use of a BIG4 auditor (BIG4).  Interestingly, these variables behave largely as expected for RPW_US in the first stage of the Heckman model, but are generally insignificant in the final model. Finally, the exclusion restriction variable, S&PTSX, is significant and negatively related to being a zero-holding observation, as expected. 127   Table 4.5 reports two different specification checks for each model.  First, it is possible that the negative Quebec bias is actually driven by a positive Ontario bias. The Ontario Securities Act has been interpreted as giving legal standing to buyers or sellers residing outside Ontario as long as the issuer is an Ontario firm (McCloskey 2001). Further, the Toronto Stock Exchange resides in Ontario, and its rules may afford investors additional legal protection.  Finally, Ontario is the largest Anglophone province.  If these factors cause US investors to have a positive bias towards Ontario firms, it may induce a negative bias against Quebec firms. To rule out this alternative explanation, we re-estimate our models after excluding Ontario firms from the sample. The results for PCT_US, given in columns 1 and 2, show that the coefficient estimates and significance levels of QC are very similar to those reported for the full sample. The results for RPW_US, given in Columns 5 and 6, are also consistent with the results for the full sample, although the coefficient on QC gained some significance in the selection model but lost some significance in the final model.   The second specification check in Table 4.5 is to estimate simple OLS regressions. Note that there is no selection model and the predicted coefficient on QC is negative for the full sample and the restricted sample of non-zero holdings.  As seen in column 3, PCT_US is 2.7 percent lower for Quebec firms than for ROC firms in the full sample, and is 6.3 percent lower when the sample is restricted to those firms with non-zero holdings, as seen in column 4.  Finally, the results for RPW_US are given in column 7 for the full sample and in column 8 for the non-zero holdings sample; the coefficient on QC is significantly negative in both cases.   In terms of the other firm choices that may influence institutional investment, USCROSS and USSEGMENT continue to remain significant for most of the columns in Table 4.5.  The simple OLS with the non-zero holding sample gives the easiest coefficients to interpret. In 128  column 4 we see that cross-listing increases the percent held by 8.2 percent and having a US segment increases the percent held by 7.1 percent; both values are only slightly higher than the impact of being located in Quebec (in absolute value). In terms of relative portfolio weights, column 8 shows that cross-listing increases the weight in the US investor portfolio, relative to the firm’s weight in the country by 5.168, and having a US segment increases it by 4.141.  Again, these effects are only slightly more extreme than the effect of being located in Quebec, and help to put the bias against Quebec firms in perspective.  The most significant control variable in the selection models given in Table 4.4 is firm size.  Accordingly, in Table 4.6 we match on size. Matching allows a more general relation between size and the dependent variable than the zero-inflated beta model or two-stage Heckman model afford.  For each firm-year with non-zero holdings we match the Quebec firm with the closest firm in the ROC in terms of its market value; the resulting sample has 416 observations from Quebec and 416 observations from the ROC.  As seen in the table, the coefficient for QC is -0.596, somewhat more negative than the comparable coefficient in column 2 of Table 4.4 or column 2 of Table 4.5.  The magnitude of this result is striking – a firm in Quebec has a proportion of US ownership to non-US ownership that is 45 percent lower than the size-matched firm in the ROC (exp(-0.596) = 0.55, 0.55-1 = 0.45).  For the RPW_US, the coefficient on QC is -3.879, roughly 78 percent more negative than the comparable coefficient in column 4 of Table 4.4; the portfolio weight US institutional investors put on Quebec firms, relative to their weight in the Canadian market, is 3.879 lower than the relative weight they place on a size-matched firm in the ROC.  In sum, US institutional investors exhibit a significant bias against firms located in Quebec relative to the ROC, consistent with the first hypothesis. Despite the similarities in 129  physical distance to the US, the bilingual publication of important disclosure documents, and the same accounting rules across Canadian provinces, US investors view Quebec firms less favourably than ROC firms, both in terms of the percent of shares they hold and in the relative weight they place on the firm in their Canadian portfolio.  Further, the results do not appear to be driven by differences in investor protection, insofar as controlling for provincial incorporation, US cross-listing, and removing Ontario from the sample, did not change the results.   4.4.2. Does the Investor Bias against Quebec Have a Basis in Language?   We have documented a large US investor bias against Quebec firms relative to ROC firms.  In this section we ask if any part of the US investor bias against Quebec firms is due to the use of the French language in Quebec.  We approach this problem by 1) examining the impact of variation in Quebec firms’ French internet presence on the US investor bias, and 2) by contrasting the bias of British investors and French investors.    To measure variation in language at the firm level we create a measure of a firm’s French language online presence, labeled FRENCHNESS.  We construct this measure by counting the number of French and English documents that result from an Advanced Google search of the company’s name during each fiscal year: FRENCHNESS is the ratio of French documents to the sum of French and English documents.  FRENCHNESS is a proxy for the relative gross information production about the firm in each language.  We make no effort to eliminate redundant documents in the count, reasoning that they represent the redistribution of information to different audiences.  To control for the total amount of information production, we also create the variable WEB as the log of one plus the total number of documents in both languages (i.e. the log of the denominator of FRENCHNESS).  While the idea behind FRENCHNESS is relatively 130  easy to describe, the actual Advanced Google search process is complicated, and is described in Appendix B2.  We begin with the sample of 416 Quebec firm-years with non-zero US holdings, but eliminate 30 firm-years because our search results did not yield any documents that could clearly be associated with a specific year.  The result is a sample of 386 Quebec firm-years. The median value for FRENCHNESS is 0.259, and it ranges from 0 to 0.670 (untabulated).    Table 4.7 divides the Quebec firm sample by FRENCHESS above and below the median and gives descriptive statistics for each group. The third row in the table shows that US investor’s hold 7.4 percent of the float in high FRENCHNESS firms and 13.4 percent of the float for low FRENCHNESS firms, on average, and the difference is significant. The relative portfolio weight is also lower in high FRENCHNESS firms, although the difference is not significant. Table 4.7 also shows other important differences between high and low FRENCHNESS firms. The frequency of provincial incorporation (in this case, the Quebec Company Act) is significantly higher for high FRENCHNESS firms, and the frequency of US cross-listing is significantly lower. Because these factors can influence US institutional investor holdings, but are unrelated to language, they will need to be controlled in the multivariate models. Finally, there is no significant difference in the sizes of firms with high or low FRENCHNESS.  Table 4.8 fits a beta model for PCT_US and an OLS model for RPW_US for the non-zero US investor holdings of Quebec firms (because we begin with the non-zero holdings, we do not need a first-stage selection model).34  The first row of the table shows that US holdings are significantly negatively related to a Quebec firm’s FRENCHESS.  The coefficient of -0.671 in the PCT_US implies that the ratio of US investor holdings to non-US investor holdings is 4.9                                                           34 Due to the high cost of manual data collection, we only collect information needed to construct FRENCHNESS and WEB for the firms with non-zero US investments.  We acknowledge that the results for Hypothesis 2 are applicable only for the firms with non-zero US holdings. 131  percent lower for a 10 percent increase in FRENCHNESS (exp(-0.671)=.51, .51-1=.49, times a treatment effect of .10 = 4.9 percent).  The coefficient of -7.911 in the RPW_US model is also impressive, given that the mean of RPW_US is 5.471 for low FRENCHNESS firm-years and 6.432 for high FRENCHNESS firms.  A 10 percent increase in FRENCHNESS is associated with a 79.11 percent reduction in the relative portfolio weight in the US investor’s Quebec portfolio. Of the other firm choice variables that vary within Quebec, PROV_INCP, USCROSS, and USSEGMENT, only cross-listing is significant. In sum, after controlling for all the firm-level characteristics, the degree of FRENCHNESS displayed by a Quebec firm has a significant influence on its US institutional investor holdings.  These results are consistent with our second hypothesis.  Our second approach to identifying the language/culture contribution to the foreign investor bias is to contrast the holdings of British and French institutional investors.  As discussed in the last section, this treatment provides a high contrast in the alignment of language between the investor and the Quebec firm, significantly increases the geodesic distance from Canada, but keeps the distance from Canada approximately equal for the two investor locations.35 We caution, however, that the data on institutional holdings of British and French investors is not nearly as complete the US institutional holdings data; consequently, the overall level of investment is likely to be understated.   Table 4.9 gives descriptive statistics for the samples that intersect Canadian firms with institutional investors from the UK (panel A) and from France (panel B); we only tabulate these statistics for the non-zero holdings sample but include the zero holdings data in all the models                                                           35 Obviously investors in France and the UK differ greatly in their use and understanding of the French language.  But, as we have discussed, language is tied up with culture.  Using the four Hofstede measures of culture once again, the Euclidian distance between France and Quebec is 30 while the distance between the UK and Quebec is 41.  132  that follow.  Because institutional investors are much larger in the UK than in France, it isn’t surprising that the British hold more of the float in both Quebec firms and ROC firms than the French do, as seen by comparing the first row in each panel. However, our hypothesis is that the relative bias will be smaller in France than the UK, as is indeed the case.  To see this, compare the difference in British and French holdings in Quebec (PCT_UK-PCT_FRANCE = 0.67 percent -0.14 percent = 0.53 percent) with the difference in their holdings in the ROC (PCT_UK-PCT_FRANCE = 1.21 percent –0.27 percent = 0.94 percent); the French are much closer to the British in their holdings of Quebec firms than in their holdings of ROC firms.  Interestingly, the relative portfolio weights of French investors are greater than the weights of British investors in both Quebec and the ROC – basically, French investors place smaller but more concentrated bets in Canada.  More importantly, the difference in the relative portfolio weights is greater in Quebec (RPW_UK – RPW_FR = 11.372-18.739 = -7.367) than it is in the ROC (RPW_UK – RPW_FR = 25.335 – 27.135 = -1.800).  While French investors place only slightly larger weights on ROC firms than British investors, they place substantially larger weights on Quebec firms than British investors do.  The other differences in firm characteristics between Quebec and the ROC have similar patterns across the two panels of Table 4.9. In both samples, the firms in Quebec are more likely to have a European segment (EUSEGMENT), have a lower provincial tax rate (PROV_TAX), and are less likely to be incorporated under provincial law (PROV_INCP), all of which should lead to greater institutional ownership.  Working against this, however, is the fact that in both samples, firms in Quebec are less likely to be cross-listed on a European exchange (EUCROSS).  To compare the relative bias against Quebec between the two investor groups, we begin with the full sample of 8089 Canadian firms-years, of which 2900 firm-years have non-zero 133  holdings from either British or French investors.  We then subtract the French value from the British value to create two new dependent variables, PCTDIF = PCT_UK – PCT_FR, and RPWDIF = RPW_UK – RPW_FR.   Our first model recognizes that, because the majority of observations are zero (both the British and French values are zero), we again have a selection problem. To control for this we estimate a two-stage Heckman model, where the first stage uses EUCROSS and EUSEGMENT in addition to the same set of variables that were in the first stage models given for PCT_US and RPW_US in columns 1 and 3, respectively, in Table 4.4. The first stage uses the full sample of 8089 observations. For brevity, we only report the final stage model results in Table 4.10.  As seen in column 1, the coefficient on QC is significantly negative; the difference between the percent of float held by British investors and French investors is smaller for Quebec firms than ROC firms. As a robustness test we also estimate OLS regressions on the full sample of firms, and on the subsample of non-zero holdings.  The treatment variable QC is again significantly negative in both regressions, as seen in columns 2 and 3.  Columns 5, 6, and 7 repeat the above analysis using the difference in relative portfolio weights between the two groups of investors (RPWDIF).  In all specifications the coefficient on QC is significantly negative, implying that British investors have a bigger bias against Quebec firms relative to ROC firms than French investors.   The final analysis in Table 4.10 takes a different approach to comparing the relative investor biases between British and French investors. PCTDIF and RPWDIF show the difference between French and British investors’ response to Quebec firms and ROC firms, but they don’t reveal the level of each groups’ bias against each firm location.  To investigate this issue, rather than taking the difference in their percentage holdings or relative portfolio weights, we use the 134  original dependent variables PCT_UK, PCT_FRANCE, RPW_UK and RPW_FRANCE and stack all observations with non-zero holdings from either British or French investors.  In effect, one observation in this analysis is a firm-year-investor location. We then use an indicator variable FRANCE to identify the investor location (FRANCE equals one when the investor is in France and equals zero when the investor is in the UK). Thus, different combinations of QC and FRANCE pick out the values for different firm locations and investor locations while controlling for all the previously discussed variables. Examining the first row of columns 4 and 8 in Table 4.10 shows the generic bias against Quebec; the coefficient on QC is significantly negative in both regressions. Of more interest, however, is the coefficient on the interaction between QC and FRANCE; in both models the interaction is significantly positive. Controlling for the generic bias against Quebec (QC) and for any difference in the Canada-wide level of investment by each investor location (FRANCE), French investors hold significantly more of the float and place larger portfolio weights on Quebec firms than ROC firms (QC*FRANCE).  To illustrate the results for the stacked regressions in columns 4 and 8 of Table 4.10, consider the estimated effects for different combinations of investor and firm locations.  Beginning with the values for percent holdings in column 4, label the estimated British holdings in ROC firms (FRANCE=0 and QC=0) as INTCP (basically the sum of the intercept and all the fixed effects). The estimated British holdings in Quebec firms (FRANCE=0 and QC=1) is then INTCP-0.003; British investors have a bigger bias against Quebec firms than against ROC firms. Similarly, the estimated French holdings in ROC firms is INTCP -0.010 (FRANCE=1 and QC=0), and the estimated French holdings in Quebec firms is INTCP-0.003-0.010+0.005=INTCP-0.008 (FRANCE=1 and QC=1). Thus, French investors have a smaller bias against Quebec firms than against ROC firms.  The same exercise for relative portfolio weights 135  using the estimates from column 8 gives similar results. British investors have a bigger bias against Quebec firms than ROC firms, and French investors have a bigger bias against ROC firms than Quebec firms.  In sum, comparing the behaviour of British and French investors offers a good contrast in the alignment of language between the investor and the firm while greatly increasing the distance from Canada, and yet holding it roughly equal between investor groups.  It also holds constant Quebec-specific forces, such as political risk or poor investor protection.  We find that French investors have a significantly smaller bias than British investors against Quebec firms relative to ROC firms.  These results are consistent with our third hypothesis.  4.5. Conclusion   In the search for the root cause of the foreign investor bias against domestic firms, our results indicate that foreign investors are sensitive to differences between their domestic language and the language used in the location of a foreign investment. This bias has a significant impact on firms in Quebec. US institutional investors invest significantly less in Quebec firms than in firms located elsewhere in Canada, and at a rate that varies significantly with the “Frenchness” of the Quebec firm. We also find that French investors are significantly less biased against Quebec firms than are British investors, adding further evidence to the hypothesis that linguistic differences contribute significantly to the foreign investor bias.   136  Table 4.1. Variable definitions Variables Definitions Dependent Variables PCT_US percentage ownership by U.S. institutional investors in the firm from Thomson Reuters Institutional Holdings S34 file (the 13f  data), defined as total number of shares owned by U.S. institutions divided by the float, which is shares outstanding less the number of shares held by block holders, at the end of the fiscal year RPW_US US investor relative portfolio weight is the weight of the firm in US investor’s Canadian portfolio compared to its weight in Canadian equity market.  Let 𝑀𝑉𝑗𝑈𝑆 = total dollar amount invested by the US institutional investors in firm j. If there are N firms in the Canadian equity market, then firm j’s weight in US investors’ Canadian portfolio, 𝑊𝑗𝑈𝑆 =𝑀𝑉𝑗𝑈𝑆/ ∑ 𝑀𝑉𝑗𝑈𝑆𝑁𝑗=1 .  Similarly, let 𝑀𝑉𝑗 denote the market value of firm j.  Then firm j’s weight in the Canadian market portfolio, 𝑊𝑗𝑀 =𝑀𝑉𝑗/ ∑ 𝑀𝑉𝑗𝑁𝑗=1 .  The relative portfolio weight of firm ‘j’ among US investors is defined as: 𝑊𝑗𝑈𝑆/ 𝑊𝑗𝑀 .   PCTDIF   differential ownership by the U.K. mutual funds and the French mutual funds for a given firm-year, defined as PCT_UK minus PCT_FRANCE, where PCT_UK and PCT_FRANCE are constructed in the same way as PCT_US described above.  For British and French holdings, data is available only for mutual funds, and it comes from Thomson Reuters mutual fund database (i.e. the S12 file). RPWDIF differential relative portfolio weight by the UK mutual funds and the French mutual funds for a given firm-year, defined as RPW_ UK minus RPW_FRANCE, where RPW_UK and RPW_FRANCE are constructed in the same way as RPW_US described above.  For British and French holdings, data is available only for mutual funds, and it comes from Thomson Reuters mutual fund database (i.e. the S12 file). Variables for Canada-wide Analysis QC indicator variable equal to 1 if the firm is headquartered in the Province of Quebec [LOC=Canada and STATE=Quebec] and 0 if the firm is headquartered in another Canadian province. 137  Table 4.1 (Continued) Variables Definitions S&PTSX indicator variable equal to 1 if a firm is included in the S&PTSX (previously known as TSX300) index in given year, and 0 otherwise. USCROSS indicator variable equal to 1 if the firm’s equity is traded in a major U.S. exchange or over-the-counter bulletin board in a given year as recorded in the CRSP-COMPUSTAT merged database, and 0 otherwise. USSEGMENT indicator variable equal to 1 if the firm discloses U.S. segment sales in COMPUSTAT Historical Segment data, and 0 otherwise. PROV_INCP indicator variable equal to 1 if the firm is incorporated under the provincial incorporation act, and 0 if the firm is incorporated under the Canadian Business Corporation Act or another national regulatory regime. PROVTAX provincial marginal tax rate of the province where the firm is headquartered in a given year. SIZE the logarithm of the market value of equity at the end of the fiscal year : [Log(CSHO*PRCC_F)].  Reported unlogged in descriptive tables 4.2 and 4.7. AGE firm age, defined as log of one plus the number of years since a firm’s first appearance in COMPUSTAT fundamental file. Reported unlogged in descriptive tables 4.2 and 4.7. GROWTH the one-year growth in net sales: [SALE/lag(SALE)-1] ROE  return on equity, calculated as (net income after preferred dividends /beginning common equity): [(NI-DVP)/lag(CEQ)] LEVERAGE the debt-to-total assets ratio, calculated as [(long-term debt + debt in current liabilities)/total asset] : [(DLTT+DLC)/AT] EP earnings-to-price ratio, calculated as (earnings per share/year-end market price): [EPSPX/PRCC_F] BP book-to-market ratio, calculated as (book value per share/year-end market price): [(CEQ/CSHO)/PRCC_F] DP dividend yield, calculated as (dividends per share/year-end market price): [DVPSP_F/ PRCC_F] 138  Table 4.1 (Continued) Variables Definitions RRET one year raw stock return over fiscal year NANLSYT log of one plus the number of analysts providing earnings forecasts during the last month of the fiscal year from I/B/E/S Summary History data. Reported unlogged in descriptive tables 4.2 and 4.7. BIG4 an indicator variable equal to 1 if the firm’s auditor is one of the Big 4 auditors Additional Variables for within-Quebec Analysis FRENCHNESS degree of usage of French language in the web presence of the firm in a given year, defined as the number of online French language documents generated within Canada divided by the total number of online documents (English and French language) generated within Canada about the firm during the fiscal year.  See Appendix B2 for details. WEB the size of web presence of a firm in a given year, defined as log of one plus the total number of online documents (English and French language) generated within Canada about the firm during the fiscal year Additional Variables for European Investor Ownership Analysis EUCROSS indicator variable equal to 1 if the firm’s security is traded on a French, British, or other European stock exchanges in a given year as recorded in CAPITAL IQ, and 0 otherwise EUSEGMENT indicator variable equal to 1 if the firm discloses segment sales in France, the United Kingdom, or other European countries/ regions in a given year as recorded in COMPUSTAT Historical Segment data, and 0 otherwise.    139  Table 4.2. Descriptive statistics: US institutional holdings of Canadian firms                Panel A.  Full sample                          Quebec  Rest of Canada  Difference  Count Mean P25 Median P75  Count Mean P25 Median P75  in Mean PCT_US 1249 0.039 0.000 0.000 0.003  6840 0.052 0.000 0.000 0.007  -0.012 ** RPW_US 1249 2.337 0.000 0.000 0.056  6840 2.846 0.000 0.000 0.268  -0.509  USSEGMENT 1249 0.144 0.000 0.000 0.000  6840 0.135 0.000 0.000 0.000  0.009  PROV_TAX 1249 0.101 0.089 0.090 0.119  6840 0.116 0.100 0.118 0.120  -0.016 *** PROV_INCP 1249 0.364 0.000 0.000 1.000  6840 0.672 0.000 1.000 1.000  -0.308 *** AGE 1249 14.2 6.0 11.0 19.0  6840 12.6 5.0 9.0 16.0  1.6 *** USCROSS 1249 0.300 0.000 0.000 1.000  6840 0.352 0.000 0.000 1.000  -0.052 *** SIZE 1249 1852.8 82.2 284.0 1145.2  6840 1772.5 75.9 263.2 1014.8  80.3  GROWTH 1249 0.193 -0.019 0.070 0.190  6840 0.342 -0.027 0.107 0.342  -0.149 *** ROE 1249 0.055 0.003 0.110 0.179  6840 0.071 -0.012 0.091 0.186  -0.016  LEVERAGE 1249 0.202 0.040 0.179 0.320  6840 0.214 0.033 0.181 0.328  -0.012 * EP 1249 0.001 0.002 0.056 0.085  6840 0.005 -0.008 0.046 0.084  -0.004  BP 1249 0.763 0.401 0.612 0.973  6840 0.727 0.373 0.603 0.956  0.036 * DP 1249 0.022 0.000 0.008 0.024  6840 0.032 0.000 0.005 0.044  -0.011 *** RRET 1249 0.159 -0.167 0.081 0.334  6840 0.225 -0.150 0.116 0.412  -0.066 *** NANALYST 1249 0.582 0.000 0.000 0.000  6840 0.463 0.000 0.000 0.000  0.119 * BIG4 1249 0.839 1.000 1.000 1.000  6840 0.932 1.000 1.000 1.000  -0.093 ***    140  Table 4.2 (Continued) Descriptive Statistics: US Institutional holdings of Canadian firms                Panel B.  Only firm-years with non-zero holdings                       Quebec  Rest of Canada  Difference  Count Mean P25 Median P75  Count Mean P25 Median P75  in Mean PCT_US 416 0.118 0.003 0.023 0.186  2540 0.139 0.003 0.047 0.223  -0.021 * RPW_US 416 7.016 0.056 0.572 4.933  2540 7.664 0.107 1.269 9.939  -0.648  USSEGMENT 416 0.389 0.000 0.000 1.000  2540 0.285 0.000 0.000 1.000  0.104 *** PROV_TAX 416 0.101 0.090 0.099 0.119  2540 0.115 0.100 0.118 0.120  -0.014 *** PROV_INCP 416 0.334 0.000 0.000 1.000  2540 0.596 0.000 1.000 1.000  -0.262 *** AGE 416 20.0 9.0 16.0 27.0  2540 15.7 7.0 11.0 21.0  4.3 *** USCROSS 416 0.596 0.000 1.000 1.000  2540 0.640 0.000 1.000 1.000  -0.044  SIZE 416 4384.4 275.0 1219.0 4007.0  2540 3761.4 241.9 815.5 3023.7  623.0  GROWTH 416 0.166 -0.019 0.071 0.185  2540 0.373 -0.013 0.125 0.377  -0.207 *** ROE 416 0.070 0.011 0.122 0.194  2540 0.063 -0.022 0.089 0.184  0.007  LEVERAGE 416 0.227 0.087 0.215 0.342  2540 0.201 0.036 0.168 0.308  0.026 ** EP 416 -0.004 0.009 0.052 0.078  2540 -0.001 -0.013 0.036 0.070  -0.004  BP 416 0.644 0.352 0.525 0.786  2540 0.604 0.325 0.498 0.752  0.040  DP 416 0.017 0.000 0.009 0.024  2540 0.018 0.000 0.000 0.022  -0.001  RRET 416 0.165 -0.143 0.089 0.329  2540 0.225 -0.162 0.111 0.409  -0.060  NANALYST 416 1.683 0.000 0.000 1.000  2540 1.219 0.000 0.000 0.000  0.463 ** BIG4 416 0.841 1.000 1.000 1.000  2540 0.954 1.000 1.000 1.000  -0.113 ***  Note: variable definitions are given in Table 4.1.  AGE, SIZE, and NANALYST are logged in the analysis but are reported here unlogged to aid in their interpretation. There are 416 Quebec firm-years with non-zero holdings, but only 386 of them have an Internet presence in a specific year that can be identified, resulting in two samples of 193 firm-years.  141  Table 4.3. Correlations Table: US Institutional holdings of Canadian firms (Full Sample) Spearman Correlations above the diagonal and Pearson Correlations below the diagonal (p-values reported in italic)  PCT_ US RPW_US QC US  SEG MENT PROV _TAX PROV_ INCP AGE US CROSS SIZE GROW TH ROE LEV EP BP DP RRET NAN ALYST BIG4                    PCT_US 1 0.872 -0.037 0.463 0.020 -0.163 0.200 0.521 0.227 0.001 -0.001 0.012 -0.049 -0.100 -0.136 -0.058 0.463 0.045  0.00 0.00 0.04 0.00 0.28 0.00 0.00 0.00 0.00 0.97 0.96 0.53 0.01 0.00 0.00 0.00 0.00 0.02 RPW_US 0.415 1 0.009 0.406 0.007 -0.252 0.356 0.488 0.589 0.007 0.169 0.081 0.118 -0.164 0.100 0.027 0.419 0.114  0.00 0.00 0.64 0.00 0.71 0.00 0.00 0.00 0.00 0.73 0.00 0.00 0.00 0.00 0.00 0.15 0.00 0.00 QC -0.041 0.014 1 0.079 -0.386 -0.184 0.121 -0.032 0.056 -0.090 0.045 0.072 0.069 0.029 0.059 -0.015 0.026 -0.163  0.03 0.46 0.00 0.00 0.00 0.00 0.00 0.09 0.00 0.00 0.01 0.00 0.00 0.11 0.00 0.43 0.16 0.00 USSEGMENT 0.471 0.216 0.079 1 0.007 -0.237 0.213 0.489 0.095 -0.068 -0.085 0.050 -0.090 -0.053 -0.095 -0.059 0.254 0.006  0.00 0.00 0.00 0.00 0.72 0.00 0.00 0.00 0.00 0.00 0.00 0.01 0.00 0.00 0.00 0.00 0.00 0.75 PROV_TAX -0.034 -0.059 -0.306 0.028 1 0.052 -0.100 -0.052 -0.157 0.074 -0.035 -0.054 -0.042 -0.069 -0.104 0.042 -0.142 0.031  0.06 0.00 0.00 0.13 0.00 0.00 0.00 0.00 0.00 0.00 0.06 0.00 0.02 0.00 0.00 0.02 0.00 0.10 PROV_INCP -0.174 -0.215 -0.184 -0.237 0.045 1 -0.300 -0.201 -0.258 0.115 -0.095 -0.047 -0.093 0.072 -0.097 0.018 -0.184 -0.075  0.00 0.00 0.00 0.00 0.01 0.00 0.00 0.00 0.00 0.00 0.00 0.01 0.00 0.00 0.00 0.33 0.00 0.00 AGE 0.198 0.327 0.120 0.215 -0.101 -0.301 1 0.266 0.498 -0.230 0.186 0.173 0.226 0.021 0.290 -0.039 0.159 0.142  0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.24 0.00 0.04 0.00 0.00 USCROSS 0.473 0.302 -0.032 0.489 -0.025 -0.201 0.274 1 0.199 -0.084 -0.072 0.013 -0.066 -0.035 -0.067 -0.102 0.380 0.041  0.00 0.00 0.09 0.00 0.17 0.00 0.00 0.00 0.00 0.00 0.00 0.48 0.00 0.06 0.00 0.00 0.00 0.03 SIZE 0.248 0.560 0.054 0.087 -0.161 -0.270 0.494 0.204 1 -0.009 0.373 0.177 0.321 -0.192 0.481 0.137 0.236 0.181  0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.64 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 GROWTH -0.045 -0.066 -0.069 -0.071 0.072 0.096 -0.240 -0.080 -0.091 1 0.204 -0.045 0.123 -0.169 -0.135 0.156 -0.056 -0.020  0.01 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.01 0.00 0.00 0.00 0.00 0.00 0.27 ROE -0.011 0.080 0.007 -0.083 -0.048 -0.022 0.128 -0.069 0.264 0.006 1 0.039 0.781 -0.198 0.253 0.248 0.016 0.044  0.56 0.00 0.70 0.00 0.01 0.24 0.00 0.00 0.00 0.76 0.00 0.03 0.00 0.00 0.00 0.00 0.40 0.02 LEVERAGE 0.048 0.018 0.050 0.033 -0.033 -0.019 0.118 0.000 0.100 -0.089 -0.011 1 0.068 0.053 0.332 -0.040 -0.019 -0.019  0.01 0.33 0.01 0.08 0.08 0.30 0.00 1.00 0.00 0.00 0.56 0.00 0.00 0.00 0.00 0.03 0.29 0.30 EP -0.020 0.087 -0.007 -0.078 -0.035 -0.011 0.108 -0.100 0.301 0.029 0.531 -0.055 1 0.115 0.325 0.150 0.000 0.012  0.27 0.00 0.71 0.00 0.06 0.54 0.00 0.00 0.00 0.12 0.00 0.00 0.00 0.00 0.00 0.00 0.98 0.52 BP -0.089 -0.114 0.031 -0.031 -0.029 0.054 0.002 -0.019 -0.264 -0.084 -0.069 0.004 -0.178 1 0.080 -0.335 -0.071 -0.082  0.00 0.00 0.10 0.09 0.11 0.00 0.92 0.31 0.00 0.00 0.00 0.83 0.00 0.00 0.00 0.00 0.00 0.00 DP -0.185 -0.020 -0.008 -0.134 -0.029 0.050 -0.002 -0.118 0.130 -0.081 0.089 0.283 0.088 0.051 1 -0.008 0.012 0.069  0.00 0.27 0.65 0.00 0.11 0.01 0.92 0.00 0.00 0.00 0.00 0.00 0.00 0.01 0.00 0.67 0.52 0.00 RRET -0.057 -0.027 -0.032 -0.057 0.036 0.047 -0.100 -0.098 0.030 0.105 0.116 -0.093 0.202 -0.282 -0.103 1 -0.052 -0.001  0.00 0.15 0.09 0.00 0.05 0.01 0.00 0.00 0.11 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.96 NANALYST 0.466 0.219 0.040 0.268 -0.151 -0.192 0.162 0.361 0.257 -0.047 0.000 -0.042 0.010 -0.071 -0.061 -0.052 1 0.054  0.00 0.00 0.03 0.00 0.00 0.00 0.00 0.00 0.00 0.01 0.99 0.02 0.57 0.00 0.00 0.00 0.00 0.00 BIG4 0.036 0.092 -0.163 0.006 0.003 -0.075 0.141 0.041 0.185 -0.014 0.004 -0.030 0.027 -0.088 0.021 -0.007 0.050 1  0.05 0.00 0.00 0.75 0.86 0.00 0.00 0.03 0.00 0.46 0.83 0.11 0.14 0.00 0.25 0.71 0.01 0.00 142  Table 4.4. Regression of U.S. institutional holdings on QC dummy and control variables     Zero-Inflated Beta  Heckman two stage       Full sample  Full sample     (1) (2)  (3) (4) MODEL    Selection Final  Selection Final VARIABLES  Prediction Selection, Final  PCT_US PCT_US  RPW_US RPW_US               QC  + , −  0.277*** -0.445***  0.164*** -2.174***     (2.745) (-6.546)  (2.767) (-2.979) S&PTSX  −   -0.197**   -0.125**      (-2.208)   (-2.417)  PROV_TAX  + , −  1.227 -3.983***  0.980 1.201     (0.546) (-2.973)  (0.754) (0.077) PROV_INCP  + , −  -0.080 0.150***  -0.040 0.328     (-1.079) (2.892)  (-0.941) (0.680) USCROSS  − , +  -1.140*** 0.747***  -0.679*** 0.118     (-10.329) (10.991)  (-10.695) (0.120) USSEGMENT  − , +  -0.729*** 0.485***  -0.424*** 0.474     (-5.528) (7.182)  (-5.610) (0.487) SIZE  − , +  -0.493*** 0.182***  -0.280*** -0.995**     (-14.326) (10.440)  (-14.712) (-2.136) GROWTH  − , +  -0.026 0.002  -0.014 -0.090     (-0.804) (0.146)  (-0.758) (-0.490) ROE  − , +  0.061 0.067  0.041 0.642     (0.618) (0.903)  (0.723) (0.935) AGE  ?  -0.157*** -0.010  -0.093*** -0.097     (-2.636) (-0.238)  (-2.783) (-0.220) LEVERAGE  ?  -0.180 0.422***  -0.066 8.640***     (-0.851) (2.686)  (-0.553) (4.540) EP  ?  0.680*** 0.100  0.388*** 4.312***     (4.167) (0.738)  (4.195) (2.747) BP  ?  0.133* 0.012  0.076* 2.052***     (1.777) (0.222)  (1.776) (2.984) DP  ?  9.785*** -5.058***  5.023*** 1.093     (7.302) (-7.254)  (7.317) (0.163) RRET  ?  0.058 -0.147***  0.028 -0.319     (1.006) (-4.255)  (0.864) (-0.959) NANALYST  − , +  -1.972*** 0.517***  -1.005*** -8.339***     (-8.307) (12.931)  (-9.170) (-4.942) BIG4  − , +  0.100 -0.188**  0.058 -0.220     (0.795) (-2.425)  (0.833) (-0.225) INVERSE_MILL  ?      11.673***         (6.411) CONSTANT  ?  5.518*** -4.458***  3.180*** 3.695     (12.806) (-15.836)  (13.558) (0.969) Industry FE    YES YES  YES YES Year FE    YES YES  YES YES R-squared        0.3715 ln_phi    1.739***         (41.913)     Observations       8,089 8,089   8,089 2,956 Robust t-statistics (z-statistics) in parentheses for ZOIB model (Heckman model)   *** p<0.01, ** p<0.05, * p<0.1. Errors are clustered by industry and year   143  Table 4.5. Regression of U.S. institutional holdings on QC dummy and control variables (alternate specifications and robustness)     Zero-inflated Beta  Simple OLS  Heckman two stage  Simple OLS     Sample excluding Ontario firms   Full Sample  Sample with non-zero holdings  Sample excluding       Ontario firms   Full Sample  Sample with non-zero holdings       (1) (2)  (3) (4)  (5) (6)  (7) (8) MODEL    Selection Final  Final Final  Selection Final  Final Final VARIABLES  Prediction Selection, Final  PCT_US PCT_US  PCT_US PCT_US  RPW_US RPW_US  RPW_US RPW_US                          QC  + , −  0.379*** -0.453***  -0.027*** -0.063***  0.212*** -1.605*  -1.209*** -3.191***     (2.948) (-5.971)  (-8.300) (-7.407)  (2.866) (-1.776)  (-5.296) (-4.798) S&PTSX  −   -0.166      -0.103         (-1.420)      (-1.535)     PROV_TAX  + , −  3.057 -4.135***  -0.214*** -0.423**  1.979 39.393**  -6.577 -12.640     (1.249) (-2.981)  (-3.395) (-2.330)  (1.393) (2.512)  (-1.209) (-0.799) PROV_INCP  + , −  -0.289*** 0.045  -0.003 0.010  -0.168*** -2.800***  -0.164 0.751     (-3.053) (0.677)  (-1.030) (1.299)  (-3.133) (-3.996)  (-0.924) (1.544) USCROSS  − , +  -0.907*** 0.651***  0.050*** 0.082***  -0.547*** -1.115  3.296*** 5.168***     (-6.730) (8.653)  (10.743) (10.195)  (-7.087) (-0.998)  (8.019) (7.802) USSEGMENT  − , +  -1.157*** 0.346***  0.081*** 0.071***  -0.662*** -5.836***  5.102*** 4.141***     (-5.643) (5.272)  (9.973) (7.935)  (-5.880) (-3.584)  (7.578) (5.471)                CONTROLS    YES YES  YES YES  YES YES  YES YES Industry FE    YES YES  YES YES  YES YES  YES YES Year FE    YES YES  YES YES  YES YES  YES YES R-squared       0.5157 0.5350   0.3978  0.3391 0.3895 ln_phi    1.965***               (36.246)           Observations       4,770 4,770   8,089 2,956   4,770 1,800   8,089 2,956 Robust t-statistics (z-statistics) in parentheses for ZOIB and OLS models (Heckman model)        *** p<0.01, ** p<0.05, * p<0.1. Errors are clustered by industry and year        144  Table 4.6. US investor bias investigation: regression using matched pair sample  Sample Consists of all the Quebec firm-years with non-zero US holdings, and their size matched non-zero-US-holding firms from Rest of Canada       Beta Model  OLS         (1)  (2)    VARIABLES  Prediction  PCT_US  RPW_US                 QC  −  -0.596***  -3.879***        (-5.034)  (-3.273)    PROV_TAX  −  -7.839**  -13.594        (-2.404)  (-0.399)    PROV_INCP  −  0.357***  2.905**        (3.709)  (2.224)    USCROSS  +  0.817***  5.077***        (6.943)  (3.417)    USSEGMENT  +  0.508***  4.707***        (4.168)  (3.104)              CONTROLS    YES  YES    Industry FE    YES  YES    Year FE    YES  YES              R-squared      0.4637    ln_phi    1.972***          (25.512)      Observations       832   832    Robust t-statistics in parentheses, *** p<0.01, ** p<0.05, * p<0.1, errors clustered by industry and year   145  Table 4.7. Descriptive statistics: US institutional holdings of Quebec firms   Only firm-years with non-zero holdings  Above Median FRENCHNESS  Below Median FRENCHNESS  Difference  Count Mean P25 Median P75  Count Mean P25 Median P75  in Mean FRENCHNESS 193 0.430 0.323 0.400 0.521  193 0.123 0.054 0.122 0.192  0.307 *** WEB 193 4.535 2.708 4.595 5.872  193 4.788 3.466 4.477 5.908  -0.253  FPCT_US 193 0.074 0.002 0.010 0.088  193 0.134 0.003 0.037 0.213  -0.059 *** RPW_US 193 5.471 0.034 0.270 2.267  193 6.432 0.081 0.727 3.453  -0.961  USSEGMENT 193 1.762 0.000 0.000 2.000  193 1.984 0.000 0.000 3.000  -0.223  PROV_TAX 193 0.102 0.090 0.099 0.119  193 0.102 0.090 0.099 0.119  0.000  PROV_INCP 193 0.409 0.000 0.000 1.000  193 0.290 0.000 0.000 1.000  0.119 * AGE 193 19.8 7.0 17.0 26.0  193 21.4 11.0 17.0 29.0  -1.5  USCROSS 193 0.461 0.000 0.000 1.000  193 0.684 0.000 1.000 1.000  -0.223 *** SIZE 193 3893.7 256.9 1199.5 5412.7  193 5328.1 337.4 1304.9 3854.8  -1434.4  GROWTH 193 0.242 -0.018 0.081 0.195  193 0.100 -0.005 0.079 0.173  0.142 * ROE 193 0.059 -0.017 0.113 0.188  193 0.119 0.057 0.138 0.208  -0.060  LEVERAGE 193 0.217 0.058 0.207 0.315  193 0.227 0.116 0.229 0.341  -0.010  EP 193 0.002 -0.012 0.051 0.078  193 0.018 0.031 0.056 0.076  -0.015  BP 193 0.586 0.351 0.519 0.699  193 0.647 0.348 0.482 0.771  -0.061  DP 193 0.017 0.000 0.012 0.025  193 0.018 0.000 0.009 0.024  -0.001  RRET 193 0.187 -0.131 0.098 0.331  193 0.169 -0.139 0.116 0.353  0.018  NANALYST 193 0.777 0.000 0.000 0.000  193 2.363 0.000 0.000 3.000  -1.585 *** BIG4 193 0.865 1.000 1.000 1.000  193 0.808 1.000 1.000 1.000  0.057   Note: variable definitions are given in Table 4.1.  AGE, SIZE, and NANALYST are logged in the analysis but are reported here unlogged to aid in their interpretation. There are 416 Quebec firm-years with non-zero holdings, but only 386 of them have an Internet presence in a specific year that can be identified, resulting in two samples of 193 firm-years.   146  Table 4.8. Regression of U.S. institutional holdings on FRENCHNESS and control variables            Beta Model  OLS     Only Non-Zero Holdings  Only Non-Zero Holdings       (1)  (2) VARIABLES  Prediction  PCT_US  RPW_US            FRENCHNESS  −  -0.671**  -7.911*     (-2.567)  (-1.732) WEB  ?  0.0217  0.112     (0.289)  (0.181) PROV_INCP  −  -0.0109  -1.240     (-0.0618)  (-0.850) USCROSS  +  0.863***  6.205***     (5.925)  (2.989) USSEGMENT  +  0.0698  -2.000     (0.365)  (-0.622)        CONTROLS    YES  YES Industry FE    YES  YES Year FE    YES  YES R-squared      0.528 ln_phi    2.947***       (26.12)   Observations       386   386 Robust t-statistics in parentheses, *** p<0.01, ** p<0.05, * p<0.1, errors clustered by industry and year  There are 416 Quebec firm-year observations, but only 386 of them have the required online presence in a year. See Appendix B2 for details.   147  Table 4.9. Descriptive Statistics: European institutional holdings of Canadian firms Panel A: UK investor holdings of Canadian firms (only firm-years with non-zero holdings)                 Quebec  Rest of Canada  Difference  Count Mean P25 Median P75  Count Mean P25 Median P75  in Mean PCT_UK(%) 459 0.67% 0.05% 0.18% 0.58%  2319 1.21% 0.05% 0.23% 1.17%  -0.005 *** RPW_UK 459 11.372 0.226 1.349 5.393  2319 25.335 0.455 3.226 19.079  -13.963 *** EUSEGMENT 459 0.192 0.000 0.000 0.000  2319 0.134 0.000 0.000 0.000  0.058 *** EUCROSS 459 0.414 0.000 0.000 1.000  2319 0.559 0.000 1.000 1.000  -0.145 *** PROV_TAX 459 0.103 0.090 0.099 0.119  2319 0.114 0.100 0.115 0.120  -0.011 *** PROV_INCP 459 0.351 0.000 0.000 1.000  2319 0.564 0.000 1.000 1.000  -0.213 ***                 Panel B: French investor holdings of Canadian firms (only firm-years with non-zero holdings)                 Quebec  Rest of Canada  Difference  Count Mean P25 Median P75  Count Mean P25 Median P75  in Mean PCT_FRANCE(%) 231 0.14% 0.02% 0.05% 0.11%  1231 0.27% 0.02% 0.07% 0.25%  -0.13% *** RPW_FRANCE 231 18.739 0.553 2.113 9.921  1231 27.135 1.360 5.321 29.480  -8.396 ** EUSEGMENT 231 0.260 0.000 0.000 1.000  1231 0.167 0.000 0.000 0.000  0.093 *** EUCROSS 231 0.537 0.000 1.000 1.000  1231 0.705 0.000 1.000 1.000  -0.168 *** PROV_TAX 231 0.103 0.090 0.099 0.119  1231 0.112 0.100 0.115 0.120  -0.010 *** PROV_INCP 231 0.281 0.000 0.000 1.000  1231 0.483 0.000 0.000 1.000  -0.201 ***                   148  Table 4.10. Regression of UK and French institutional holdings on Quebec dummy and control variables      Differenced  Stacked   Differenced  Stacked     Heckman  OLS  Stacked Sample (OLS)  Heckman  OLS  Stacked Sample (OLS)       (1)  (2) (3)  (4)  (5)  (6) (7)  (8)     Final-Stage  full sample non-zero holdings   non-zero holdings     full sample non-zero holdings  non-zero holdings VARIABLES  Prediction     Dif, Stacked  PCTDIF  PCTDIF PCTDIF  PCT_i  RPWDIF  RPWDIF RPWDIF  RPW_i                          QC  −  -0.002**  -0.001** -0.002**  -0.003***  -5.906**  -1.307* -5.820***  -4.257**     (-2.086)  (-2.480) (-2.204)  (-4.190)  (-2.546)  (-1.819) (-2.655)  (-2.415) FRANCE  ?       -0.010***       0.958          (-8.162)       (0.740) QC*FRANCE  +       0.005***       5.816**          (5.972)       (2.502) PROV_TAX  ? , −  0.030  0.006 0.035*  0.008  236.540***  80.547*** 256.082***  191.844***     (1.622)  (0.841) (1.804)  (0.536)  (4.010)  (3.973) (4.028)  (3.789) PROV_INCP  ? , −  -0.001*  -0.000 -0.001*  -0.001**  1.151  0.761 0.957  -1.589     (-1.817)  (-1.362) (-1.925)  (-2.572)  (0.646)  (1.118) (0.552)  (-1.011) EUCROSS  ? , +  0.005***  0.003*** 0.003***  0.003***  3.446  1.095 0.006  1.980     (5.795)  (5.469) (4.233)  (5.088)  (1.593)  (1.059) (0.003)  (1.115) EUSEGMENT  ? , +  0.003**  0.001* 0.003**  0.002**  8.248**  3.939** 8.773**  11.457***     (2.183)  (1.755) (2.467)  (2.120)  (2.417)  (2.256) (2.511)  (3.331)                  CONTROLS    YES  YES YES  YES  YES  YES YES  YES Industry FE    YES  YES YES  YES  YES  YES YES  YES Year FE    YES  YES YES  YES  YES  YES YES  YES R-squared    0.2222  0.1476 0.2078  0.2520  0.1438  0.0379 0.1297  0.2616 Observations       2,900   8,089 2,900   4,240   2,900   8,089 2,900   4,240 Robust t-statistics (z-statistics) in parentheses for OLS (Heckman), *** p<0.01, ** p<0.05, * p<0.1, errors Clustered by industry and year.  PCTDIF = PCT_UK – PCT_FRANCE and RPWDIF = RPW_UK – RPW_FRANCE.  For the Stacked Sample, one observation is a firm-year-country. When indicator variable FRANCE is 0, then PCT_i = PCT_UK, and when the indicator variable FRANCE is 1, then PCT_i = PCT_FRANCE.  RPW_i is constructed in the same way.     149  Chapter 5: Conclusion  The present thesis is a collection of three essays investigating various forces in the capital market.  Since the essays are presented in self-contained chapters, exhaustive discussion of the contribution and position of each essay in the broader accounting literature is provided in the conclusion specific to each chapter. In the first essay, I examine how reputation with non-equity stakeholders mitigate the negative consequences of accounting restatements.  Using Corporate Social Responsibility (CSR) rating as a proxy for reputation with non-equity stakeholders, I find significantly less negative market reaction to restatements for firms with better reputation. CSR’s ability to assuage investor concerns in the event of a restatement is greater for material restatements, which are likely to generate greater uncertainty about the restating firms’ credentials as reliable business partners. On average, a one standard deviation increase in CSR can mitigate 18.2% of the share value loss triggered by restatements. Post-restatement consequence analysis corroborates the results from the market reaction test. Following restatements, high-CSR firms experience smaller earnings-decreases and need to engage in fewer reputation restoration activities. My results suggest that a significant portion of the market value loss triggered by restatements reflects an expectation that the restating firms will face a ‘worsening of terms’ in their future transactions with the non-equity stakeholders, and CSR reputation can dampen this effect. In the second essay, I examine whether investors increase their reliance on analyst forecasts following accounting restatements.  I find that restatements in general do not have an impact on the information content of analyst forecast revisions (FRIC).  But for material 150  restatements that are perceived to be intentional, FRIC in the post-restatement period is significantly higher than FRIC in the pre-restatement period (20% to 50% higher depending on different specifications).  I also find that following material restatements, FRIC of new analysts who initiate coverage after the restatement announcement is larger than FRIC of existing analysts.  Both effects are short-lived and concentrated in the first quarter following restatement announcement.  My results suggest that investors increase their reliance on analysts when there is uncertainty about the firm and the credibility of management disclosure is compromised. The effect is greater for new analysts who cannot be blamed for failing to detect accounting problems earlier and who are less likely to have close ties with the management. The third essay of this thesis is bases on an unpublished co-authored paper.  We study how misaligned language between the investor and the firm contributes to the foreign investor bias. Specifically, we study the foreign investor bias against firms located in Quebec relative to firms located in the Rest of Canada (ROC). Even though Quebec shares with the ROC the same country, federal law, stock exchange, accounting standards, and Quebec firms make all regulatory filings in both French and English, we document a significant US institutional investor bias against Quebec firms relative to firms in the ROC.  We find that US investors hold a smaller percentage of Quebec firms than comparable ROC firms, and place less weight on them in their Canadian portfolio. Within Quebec we find that the size of the bias varies with the firm’s French online presence – a greater proportion of French to English documents increases the bias against the firm. We also find that British institutional investors exhibit a much larger bias against Quebec firms than French institutional investors and, in some specifications, French investors actually favour Quebec firms over firms in the ROC. These results suggest that the 151  alignment of language between the investor and the firm plays a significant role in determining the foreign investor bias.    152  References  Aggarwal, R., A. Colm, B. Kearney, and B. Lucey. 2012. Gravity and culture in foreign portfolio investment.  Journal of Banking & Finance 36:525–538. Ahearne, A., W. Griever, and F. Warnock. 2004. Information costs and home bias: an analysis of US holdings of foreign equities. Journal of International Economics 62:313-336. Andrade, S., and V. Chhaochharia. 2010. Information Immobility and Foreign Portfolio Investment. The Review of Financial Studies, 23:2429-2463. Attig, N. 2007. Excess control and the risk of corporate expropriation: Canadian evidence. Canadian Journal of Administrative Sciences 24:94–106. Aviat, A., and N. Coeurdacier. 2007. The geography of trade in goods and asset holdings. Journal of International Economics 71:22–51. Badertscher, B. A., Hribar, S. P., & Jenkins, N. T. (2011). Informed Trading and the Market Reaction to Accounting Restatements. The Accounting Review, 86(5), 1519–1547. Bardos, K. S., Golec, J., & Harding, J. P. (2011). Do Investors See through Mistakes in Reported Earnings? Journal of Financial and Quantitative Analysis, 46(06), 1917–1946. Barkema, H. and F. Vermeulen. 1997. What differences in the cultural backgrounds of partners are detrimental for international joint ventures? Journal of International Business Studies 28: 845-864. Barniv, R. R., & Cao, J. (2009). Does information uncertainty affect investors’ responses to analysts’ forecast revisions? An investigation of accounting restatements. Journal of Accounting and Public Policy, 28(4), 328–348. Bear, S., Rahman, N., & Post, C. (2010). The Impact of Board Diversity and Gender Composition on Corporate Social Responsibility and Firm Reputation. Journal of Business Ethics, 97(2), 207–221. Beaulieu, M., J. Cosset, and N. Essaddam. 2006. Political Uncertainty and Stock Market Returns: Evidence from the 1995 Quebec Referendum. The Canadian Journal of Economics 39:621-641.  Bénabou, R., & Tirole, J. (2010). Individual and Corporate Social Responsibility. Economica, 77(305), 1–19. Beugelsdijk, S., and B. Frijns. 2010. A cultural explanation of the foreign bias in international asset alocation. Journal of Banking and Finance 34:2121–2131. Beyer, A., Cohen, D. a., Lys, T. Z., & Walther, B. R. (2010). The financial reporting environment: Review of the recent literature. Journal of Accounting and Economics, 50(2-3), 296–343. Bonner, S. E., Walther, B. R., Young, S. M., Bonner, S. E., Walther, B. R., & Young, S. M. (2003). Sophistication-Related Differences in Investors’ Models of the Relative Accuracy of Analysts' Forecast Revisions. The Accounting Review, 78(3), 679–706. 153  Boubraki, N., Y. Bozec, C. Laurin, and S. Rousseau. 2011. Incorporation Law, Ownership Structure, and Firm Value: Evidence from Canada.  Journal of Empirical Legal Studies 8:358-383. Bradshaw, M., B. Bushee, and G. Miller. 2004. Accounting Choice, Home Bias, and U.S. Investment in Non-U.S. Firms. Journal of Accounting Research 42:795–841 Brouthers, K. and L. Brouthers. 2001. Explaining the national cultural distance paradox. Journal of International Business Studies 32:177-189. Burks, J. J. (2011). Are Investors Confused by Restatements after Sarbanes-Oxley? The Accounting Review, 86(2), 507–539. Bushee, B. 1998. The Influence of Institutional Investors on Myopic R&D Investment Behavior. The Accounting Review 73:305–33. Bushee, B. 2001. Do Institutional Investors Prefer Near-Term Earnings over Long-Run Value? Contemporary Accounting Research 18:207–46. Campbell, J. L. (2007). Why Would Corporations Behave in Socially Responsible Ways ? an of Corporate Theory Institutional Social Responsibility. The Academy of Management Review, 32(3), 946–967. Canadian Securities Administrators. 2014. Pan-Canadian Passport as retrieved on April 2014 from http://www.securities-administrators.ca/aboutcsa.aspx?id=96. Cao, H., B. Han, D. Hirshleifer, and H. Zhang. 2011. Fear of the Unknown: Familiarity and Economic Decisions. Review of Finance 15:173-206. Chakravarthy, J., DeHaan, E., & Rajgopal, S. (2014). Reputation Repair After a Serious Restatement. The Accounting Review, 89(4), 1329–1363. Chan, K., V. Covrig, and L. Ng. 2005. What Determines the Domestic Bias and Foreign Bias? Evidence from Mutual Fund Equity Allocations Worldwide. The Journal of Finance 60:1495-1534. Chen, X., Cheng, Q., & Lo, A. K. (2014). Is the decline in the information content of earnings following restatements short-lived? The Accounting Review, 89(1), 177–207. Chen, X., Cheng, Q., & Lo, K. (2010). On the relationship between analyst reports and corporate disclosures: Exploring the roles of information discovery and interpretation. Journal of Accounting and Economics, 49(3), 206–226. Christensen, D. M. (2016). Corporate Accountability Reporting, Assurance, and High-Profile Misconduct. The Accounting Review, 91(2), 377–399. Clement, M. B. (1999). Analyst forecast accuracy: Do ability, resources, and portfolio complexity matter? Journal of Accounting and Economics, 27(3), 285–303. Cook, D., R. Kieschnick and B. McCullough. 2008. Regression analysis of proportions in finance with self-selection. Journal of Empirical Finance 5:860–867. Covig, M., M. DeFond, and M. Hung. 2007. Home Bias, Foreign Mutual Fund Holdings, and the Voluntary Adoption of International Accounting Standards. Journal of Accounting Research 45:45-70. 154  Crespin, R. J. (2012). What Price Assurance? Crespin, R. 2012. Corporate Social Responsibility Magazine. Dahlquist, M., and G. Robertsson. 2001.  Direct foreign ownership, institutional investors, and firm characteristics. Journal of Financial Economics 59:413-440. Dahlquist, M., L. Pinkowitz, R. Stulz, and R. Williamson. 2003. Corporate governance and the home bias. Journal of Financial and Quantitative Analysis 38(1): 87-110. Daude, C., and M. Fratzscher. 2006. The Pecking Order of Cross-border Investment. European Central Bank Working Paper Series: 590, Frankfurt. Dechow, P. M., Sloan, R. G., & Sweeney, A. P. (1996). Causes and Consequences of Earnings Manipulation: An Analysis of Firms Subject to Enforcement Actions by the SEC”. Contemporary Accounting Research, 13(1), 1–36. DeFond, M., X. Hu, M. Hung, and S. Li. 2011. The impact of mandatory IFRS adoption on foreign mutual fund ownership: The role of comparability. Journal of Accounting and Economics 51(3):240-258. Dhaliwal, D. S., Li, O. Z., Tsang, A., & Yang, Y. G. (2011). Voluntary Nonfinancial Disclosure and the Cost of Equity Capital: The Initiation of Corporate Social Responsibility Reporting. The Accounting Review, 86(1), 59–100. Dhaliwal, D. S., Radhakrishnan, S., Tsang, A., & Yang, Y. G. (2012). Nonfinancial Disclosure and Analyst Forecast Accuracy: International Evidence on Corporate Social Responsibility Disclosure. The Accounting Review, 87(3), 723–759. Di Giuli, A., & Kostovetsky, L. (2014). Are red or blue companies more likely to go green? Politics and corporate social responsibility. Journal of Financial Economics, 111(1), 158–180. Dyck, A., Morse, A., & Zingales, L. (2010). Who Blows the Whistle on Corporate Fraud? Journal of Finance, 65(6), 2213–2253. Farber, D. B. (2005). Restoring Trust after Fraud: Does Corporate Governance Matter? The Accounting Review, 80(2), 539–561. Ferrari, S., and F. Cribari-Neto. 2004. Beta regression for modeling rates and proportions. Journal of Applied Statistics 31:799–815. Files, R., Swanson, E. P., & Tse, S. (2009). Stealth Disclosure of Accounting Restatements. The Accounting Review, 84(5), 1495–1520. Frankel, R., Kothari, S. P., & Weber, J. (2006). Determinants of the informativeness of analyst research. Journal of Accounting and Economics, 41(1-2), 29–54. Freeman, R. E. (1984). Strategic Management: A Stakeholder Approach. Boston: Pitman. French, K., and J. Poterba. 1991.  Investor diversification and international equity markets.  American Economic Review 81(2):222-226. 155  Friedman, H. L., & Heinle, M. S. (2015). Taste, information, and asset prices: Implications for the valuation of CSR. Working Paper. Available at SSRN: http://papers.ssrn.com/sol3/papers.cfm?abstract_id=2635057. Godfrey, P. C., Merrill, C. B., & Hansen, J. M. (2009). The Relationship between Corporate Social Responsibility and Shareholder Value: an Empirical Test of the Risk Management Hypothesis. Strategic Management Journal, 30(4), 425–455. Gordon, E. a., Henry, E., Li, X., & Sun, L. (2014). Management Guidance Pre- and Post-Restatement. Journal of Business Finance & Accounting, 41(7-8), 867–892. Graham, J., Li, S., & Qiu, J. (2008). Corporate misreporting and bank loan contracting. Journal of Financial Economics, 89(1), 44–61. Graham, R., C. Morrill, and J. Morrill. 2012. Does it matter where assets are held and income is derived? Further evidence of differential value relevance from Quebec. Journal of International Accounting, Auditing and Taxation 21(2):185–197. Greening, D., & Turban, D. (2000). Corporate Social Performance as a Competitive Advantage in Attracting a Quality Workforce. Business and Society, 39, 254–280. Griffin, P. A. (2003). A League of Their Own? Financial Analysts’ Responses to Restatements and Corrective Disclosures. Journal of Accounting, Auditing & Finance, vol. 18(4), 479–517. Grinblatt, M., and M. Keloharju. 2001. How distance, language, and culture influence stockholdings and trades. Journal of Finance 56(3):1053–1073. Heckman, J. 1979. Sample selection bias as a specification error. Econometrica 47:153–161. Hennes, K. M., Leone, A. J., & Miller, B. P. (2008). The Importance of Distinguishing Errors from Irregularities in Restatement Research: The Case of Restatements and CEO/CFO Turnover. The Accounting Review, 83(6), 1487–1519. Hofstede, G., G. J. Hofstede, and M. Minkov. 1997. Cultures and organizations. New York: McGraw-Hill. Hong, H., & Kubik, J. D. (2003). Analyzing the Analysts: Career Concerns and Biased Earnings Forecasts. Journal of Finance, 58(1), 313–351. Hong, H., & Liskovich, I. (2015). Crime, Punishment and the Halo Effect of Corporate Social Responsibility. NBER Working Papers. Available at: http://www.nber.org/papers/w21215. Hribar, P., & Jenkins, N. T. (2004). The Effect of Accounting Restatements on Earnings Revisions and the Estimated Cost of Capital. Review of Accounting Studies, 9(2/3), 337–356. Hribar, P., Jenkins, N. T., & Wang, J. (2004). Institutional Investors and Accounting Restatements. AAA 2005 FARS Meeting Paper. Available at SSRN: http://ssrn.com/abstract=591743. Jones, T. M. (1995). Instrumental Stakeholder Theory: a Synthesis of Ethics and Economics. Academy of Management Review, 20(2), 404–437. 156  Kang, J. 1997. Why is there a home bias? An analysis of foreign portfolio equity ownership in Japan. Journal of Financial Economics 46(1): 3-28. Karpoff, J. M. (2012). Does Reputation Work to Discipline Corporate Misconduct? In M. L. Barnett & T. G. Pollock (Eds.), The Oxford Handbook of Corporate Reputation. Oxford University Press. Karpoff, J. M., Lee, D. S., & Martin, G. S. (2008). The Cost to Firms of Cooking the Books The Cost to Firms of Cooking the Books. The Journal of Financial and Quantitative Analysis, 43(3), 581–611. Karpoff, J. M., Lott, Jr., J. R., & Wehrly, E. W. (2005). The Reputational Penalties for Environmental Violations: Empirical Evidence. The Journal of Law and Economics, 48, 653–675. Kieschnick, R., and B. McCullough. 2003. Regression Analysis of variates observed on (0,1): percentages, proportions, and fractions. Statistical Modeling 3:193–213. Kim, Y., Park, M. S., & Wier, B. (2012). Is Earnings Quality Associated with Corporate Social Responsibility? The Accounting Review, 87(3), 761–796. King, M., and D. Segal. 2009. The Long-Term Effects of Cross-Listing, Investor Recognition, and Ownership Structure on Valuation. The Review of Financial Studies 22(6):2393-2421. Klein, B., & Leffler, K. B. (1981). The Role of Market Forces in Assuring Contractual Performance. Journal of Political Economy, 89(4), 615–641. Kogut, B., and H. Singh. 1998. The Effect of National Culture on the Choice of Entry Mode. Journal of International Business Studies 19(3):411-432. Kotchen, M., & Moon, J. J. (2012). Corporate Social Responsibility for Irresponsibility. The B.E. Journal of Economic Analysis & Policy, 12(1), 1–21. KPMG. (2011). International Survey of Corporate Responsibility Reporting. KPMG International. La Porta, R., F. Silanes, A. Shleifer, and R. Vishny. 1997. Legal Determinants of External Finance. Journal of Finance 52:1131-1150. Lane, P., and G. Milesi-Ferretti. 2008. International investment patterns. Review of Economics and Statistics 90:538–549. Lennox, C., J. Francis, and Z. Wang. 2012. Selection Models in Accounting Research. The Accounting Review. 87:589-616. Lev, B., Petrovits, C., & Radhakrishnan, S. (2010). Is doing good good for you? how corporate charitable contributions enhance revenue growth. Strategic Management Journal, Volume 31(2), 182–200. Livnat, J., & Zhang, Y. (2012). Information interpretation or information discovery: Which role of analysts do investors value more? Review of Accounting Studies, 17(3), 612–641. Loree, D., and S. Guisinger. 1995. Policy and non-policy determinants of US equity foreign direct investment. Journal of International Business Studies 26:281-299. Lundholm, R., R. Rogo, and J. Zhang. 2014. Restoring the tower of Babel: How foreign firms communicate with US investors. The Accounting Review 89(4):1453-1485. 157  Lys, T. Z., Naughton, J. P., & Wang, C. (2015). Signaling Through Corporate Accountability Reporting. Journal of Accounting and Economics, 60(1), 56–72. Maddala, G., 1991. A perspective on the use of limited-dependent variables in accounting research. The Accounting Review 66:786–807. Margolis, J. D., Elfenbein, H. A., & Walsh, J. P. (2007). Does It Pay To Be Good? A Meta-Analysis and Redirection Of Research on the Relationship between Corporate Social and Financial Performance. Working Paper, Harvard University. Mattingly, J. E., & Berman, S. L. (2006). Measurement of corporate social responsibility. Business and Society, 20(1), 20–46. McCloskey, M. 2001. Canada: Control Blocks retrieved on April 15, 2014 from http://www.mondaq.com/canada/x/15042/securitization+structured+finance/Control+Blocks. Merton, R. 1987. A simple model of capital market equilibrium with incomplete information. The Journal of Finance 42(3):483-510. Murphy, D. L., Shrieves, R. E., & Tibbs, S. L. (2009). Understanding the Penalties Associated with Corporate Misconduct: An Empirical Examination of Earnings and Risk. Journal of Financial and Quantitative Analysis, 44(01), 55. Nan, X., & Heo, K. (2007). Consumer Responses to Corporate Social Responsibility (CSR) Initiatives: Examining the Role of Brand-Cause Fit in Cause-Related Marketing. Journal of Advertising, 36(2), 63–74.  Orlitzky, M., Schmidt, F. L., & Rynes, S. L. (2003). Corporate Social and Financial Performance: A Meta-Analysis. Organization Studies, 24(3), 403–441. Palmrose, Z.-V., Richardson, V. J., & Scholz, S. (2004). Determinants of market reactions to restatement announcements. Journal of Accounting and Economics, 37(1), 59–89. Peloza, J. (2006). Using Corporate Social Responsibility as Insurance for Financial Performance. California Management Review, 48(2), 52–73. Portes, R., and H. Rey. 2005. The determinants of cross-border equity flows. Journal of International Economics 65(2):269-296. Portes, R., H. Rey, and Y. Oh. 2001. Information and capital flows: the determinants of transactions in financial assets. European Economic Review 45:783-796. Puri, P. 2009. Legal Origins, Investor Protection, and Canada. BYU Law Review 1671-1700. Richardson, S. a., Teoh, S. H., & Wysocki, P. D. (2004). The Walk-down to Beatable Analyst Forecasts: The Role of Equity Issuance and Insider Trading Incentives. Contemporary Accounting Research, 21(4), 885–924. Rigby, D. (2001). Moving upward in a downturn. Harvard Business Review, 79, 98–105. 158  Rowland, P. 1999. Transaction costs and international portfolio diversification. Journal of International Economics 49(1):145-170.  Scholz, S. (2004). The Circumstances and Legal Consequences of Non-GAAP Reporting : Evidence from Restatements *. Contemporary Accounting Research, 21(1), 139–80. Scholz, S. (2008). The Changing Nature and Consequences of Public Company Financial Restatement: 1997 -2006. Treasury Department Report, U.S. Department of the Treasury, Washington, D.C., (April). Social Investment Forum Foundation. (2010). 2010 Report on Socially Responsible Investing Trends in the United States - Executive Summary. Spamann, H. 2010. The ‘Antidirector Rights Index’ Revisited. Review of Financial Studies 23: 467-486. Srinivasan, S. (2005). Consequences of Financial Reporting Failure for Outside Directors: Evidence from Accounting Restatements and Audit Committee Members. Journal of Accounting Research, 43(2), 291–334. Standard and Poor’s. 2011.  S&P Global 1200 Methodology, April.   Statistics Canada. 2011. French and the francophonie in Canada. Retrieved on April 15, 2014 from http://www12.statcan.gc.ca/census-recensement/2011/as-sa/98-314-x/98-314-x2011003_1-eng.cfm Statistics Canada. 2012. Gross domestic product, expenditure-based, by province and territory. Retrieved on April 15, 2014 from http://www.statcan.gc.ca/tables-tableaux/sum-som/l01/cst01/econ15-eng.htm Tesar, L., and I. Werner. 1995. Home bias  and  high  turnover. Journal of International Money and Finance 14(4):467-492. The Economist. (2005). The importance of corporate responsibility. The Economist Intelligence Unit. The RiskMetrics Group. (2010). How to Use KLD STATS & ESG Ratings Definitions. Retrieved on February 24, 2015 from http://cdnete.lib.ncku.edu.tw/93cdnet/english/lib/Getting_Started_With_KLD_STATS.pdf Tirtiroglu, D., H.  Bhabra, and U. Lel. 2004. Political uncertainty and asset valuation: evidence from business relocations in Canada. Journal of Banking and Finance 28:2237–58. Waddock, S. a., & Graves, S. B. (1997). The Corporate Social Performance - Financial Performance Link. Strategic Management Journal, 18, 303–319. Wahid, A., and G. Yu. 2014.  Accounting standards and international portfolio holdings. The Accounting Review, 89(5):1895-1930. Weigelt, K., & Camerer, C. (1988). Reputation And Corporate Strategy : A Review Of Recent Theor. Strategic Management Journal, 9, 443–454. 159  Wilson, W. M. (2008). An Empirical of the Decline in Analysis the Information of Earnings Content Restatements Following. The Accounting Review, 83(2), 519–548. Zehr, L., & Tuck, S. (2003, October 9). News Biovail threatens lawsuit against analyst. Globe and Mail Update. Zhang, M., Ma, L., Su, J., & Zhang, W. (2014). Do Suppliers Applaud Corporate Social Performance? Journal of Business Ethics, 121(4), 543–557.     160  A   Appendices for Chapter 2  Appendix A1  Social, Environmental, and Governance Performance Categories Used in the KLD Database Dimensions Strengths (Positive Categories) Concerns (Negative Categories) COMMUNITY    1. Charitable Giving, 2. Innovative Giving, 3. Non-US Charitable Giving, 4. Support for Housing, 5.Support for Education, 6.Indigenous Peoples Relations, 7. Volunteer Programs 1. Investment Controversies, 2. Negative Economic Impact, 3. Indigenous Peoples Relations, 4. Tax Disputes, 5. Other Concern  CORPORATE GOVERNANCE 1. Limited Compensation, 2. Ownership, 3. Transparency Strength, 4. Political Accountability Strength, 5. Other Strength  1. High Compensation, 2. Ownership, 3. Accounting, 4. Transparency, 5. Political Accountability, 6. Other Concern, DIVERSITY  1. Promotion, 2. CEO, 3. Board of Directors, 4. Work/Life Benefits, 5. Women & Minority Contracting, 6. Employment of the Disabled, 7. Gay & Lesbian Policies Other Strength 1. Controversies, 2. Non-Representation,  3. Other Concern   EMPLOYEE RELATIONS 1. Union Relations, 2. No-Layoff Policy, 3. Cash Profit Sharing,  4. Employee Involvement, 5. Retirement Benefits Strength, 6. Health and Safety Strength, 7. Other Strength 1. Union Relations, 2. Health and Safety Concern, 3. Workforce Reductions, 4. Retirement Benefits Concern, 5. Other Concern,   ENVIRONMENT   1. Beneficial Products and Services, 2. Pollution Prevention,  3. Clean Energy, 4. Communications, 5. Property, Plant, and Equipment, 6. Management Systems, 7. Other Strength 1. Hazardous Waste, 2. Regulatory Problems,  3. Ozone Depleting Chemicals, 4. Substantial Emissions, 5. Agricultural Chemicals, 6. Climate Change, 7. Other Concern HUMAN RIGHTS 1. Indigenous Peoples Relations Strength (added 2004),  2. Positive Record in South Africa, 3. Labor Rights Strength, 4. Other Strength,   1. South Africa, 2. Northern Ireland, 3. Burma Concern, 4. Mexico, 5. Labor Rights Concern,  6. Indigenous People Concern, 7. Other Concern  PRODUCT 1. Quality, 2. R&D/Innovation, 3. Benefits to Economically Disadvantaged, 4. Other Strength  1. Product Safety, 2. Marketing/Contracting Concern, 3. Antitrust, 4. Other Concern  Note 1: The CSR performance categories assessed by the KLD database may change over the time.  I use year fixed effects in all my analyses to control for any variation in the CSR performance rating Scheme.  Detailed definitions for individual categories are available from the KLD database brochure (The RiskMetrics Group, 2010).  161  Construction of CSRTotal and CSRPure  𝐶𝑆𝑅𝑇𝑜𝑡𝑎𝑙 = 𝐶𝑜𝑚𝑚𝑢𝑛𝑖𝑡𝑦 𝑇𝑜𝑡𝑎𝑙 𝑆𝑡𝑟𝑒𝑛𝑔ℎ𝑡𝑠 + 𝐷𝑖𝑣𝑒𝑟𝑠𝑖𝑡𝑦 𝑇𝑜𝑡𝑎𝑙 𝑆𝑡𝑟𝑒𝑛𝑔𝑡ℎ𝑠+ 𝐸𝑚𝑝𝑙𝑜𝑦𝑒𝑒 𝑅𝑒𝑙𝑎𝑡𝑖𝑜𝑛𝑠 𝑇𝑜𝑡𝑎𝑙 𝑆𝑡𝑟𝑒𝑛𝑔𝑡ℎ𝑠 + 𝐸𝑛𝑣𝑖𝑟𝑜𝑛𝑒𝑚𝑒𝑛𝑡 𝑇𝑜𝑡𝑎𝑙 𝑆𝑡𝑟𝑒𝑛𝑔𝑡ℎ𝑠+ 𝐻𝑢𝑚𝑎𝑛 𝑅𝑖𝑔ℎ𝑡𝑠 𝑇𝑜𝑡𝑎𝑙 𝑆𝑡𝑟𝑒𝑛𝑔ℎ𝑡𝑠 + 𝑃𝑟𝑜𝑑𝑢𝑐𝑡 𝑇𝑜𝑡𝑎𝑙 𝑆𝑡𝑟𝑒𝑛𝑔ℎ𝑡𝑠  𝐶𝑆𝑅𝑃𝑢𝑟𝑒 =   𝐶𝑆𝑅𝑇𝑜𝑡𝑎𝑙 − 𝑃𝑟𝑜𝑑𝑢𝑐𝑡 𝑇𝑜𝑡𝑎𝑙 𝑆𝑡𝑟𝑒𝑛𝑔𝑡ℎ𝑠   Distribution of CSRPure and CSRTotal (Final Sample for Main Analysis, 1892 obs) Panel 1       Panel 2        Note 2: For sensitivity checks discussed in section 2.5.3., I investigate the impact of 𝐶𝑆𝑅𝑇𝑜𝑡𝑎𝑙 and 𝐶𝑆𝑅𝑃𝑢𝑟𝑒 on accounting restatements after controlling for the negative-CSR performance.  I construct two summary measures of negative CSR performance, 𝐶𝑜𝑛𝑐𝑒𝑟𝑛𝑇𝑜𝑡𝑎𝑙 and  𝐶𝑜𝑛𝑐𝑒𝑟𝑛𝑃𝑢𝑟𝑒. These measures are analogous to 𝐶𝑆𝑅𝑇𝑜𝑡𝑎𝑙 and 𝐶𝑆𝑅𝑃𝑢𝑟𝑒, but are constructed by summing up the CSR concerns (negative categories) instead of CSR strengths (positive categories).   162  B   Appendices for Chapter 4  Appendix B1  This appendix borrows heavily from Cook et al. (2008) and the references therein. The zero-inflated beta model assumes the following density for the proportion yi of foreign ownership:  𝑔(𝑦𝑖; 𝛿, 𝜇, 𝜙)  = {𝛿 𝑖𝑓 𝑦𝑖 = 0(1 − 𝛿)𝑓(𝑦𝑖; 𝜇, 𝜙) 𝑖𝑓 0 < 𝑦𝑖 < 1      (1)  where 𝑓(𝑦𝑖; 𝜇, 𝜙) is a beta density defined on 0<yi<1 , given as  𝑓(𝑦𝑖; 𝑢, 𝜙) =Γ(𝜙)Γ(𝜇𝜙)Γ((1−𝜇)𝜙)𝑦𝑖𝜇𝜙−1(1 − 𝑦𝑖)(1−𝜇)𝜙−1.  (2)    where 0< 𝜇 <1 and 𝜙 >0;  𝜇 is the location parameter, with E(yi)= 𝜇 and 𝜙 is the precision parameter, with V(yi) = 𝜇(1−𝜇)𝜙+1 .  Thus, for the density g,   E(yi) = (1- 𝛿)𝜇  and V(yi) = (1 − 𝛿)𝜇(1−𝜇)𝜙+1+ 𝛿(1 − 𝛿)𝜇2.      The expression for E(yi) illustrates how failing to account for the mass of observations at zero will lead to overstated estimates of the proportion.  The expression for V(yi) illustrates that the variance increases with the mean, a distributional feature that the OLS model cannot accommodate.  The advantage of assuming the dependent variable has this density is that it allows an arbitrarily large mass point at zero through the parameter 𝛿, and it has a very flexible shape for values greater than zero as determined by the u and 𝜙 parameters.  For instance, the beta density can be uniform (𝜇 = ½, 𝜙 = 1), can descend monotonically over the interval (e.g. 𝜇 =1/4, 𝜙 = 4), or be single-peaked (e.g. 𝜇 =1/2, 𝜙 =4).36  Next we need to describe how the independent variables map into the dependent variable. There are two parts to this. First, to estimate whether yi=0 or yi>0 the model assumes a cumulative logistic function mapping the data vector Xi with weight parameters A into the probability of yi = 0, as                                                            36 One may be tempted to use a Tobit model to handle the mass of zeros.  However, the Tobit model is designed for cases where the data is censored at zero; that is, the true value is negative but the research only observes these values as being censored to zero. That is not the case for proportions. See Madalla (1991) for an extended discussion of this issue.   163  𝑃𝑟𝑜𝑏(𝑦𝑖 = 0) = 𝛿 =𝑒𝑋𝑖′𝐴1+𝑒𝑋𝑖′𝐴 .        (3)  If yi>0 the model maps the data vector Zi with coefficient weights B to the conditional mean parameter ui , as  𝐸(𝑦𝑖|𝑦𝑖 > 0) = 𝜇𝑖 =𝑒𝑍𝑖′𝐵1+𝑒𝑍𝑖′𝐵 .        (4)  Importantly, note that the two parts of the model can be estimated on different independent variable vectors Xi and Zi, and even if Xi = Zi, as will be the case in our analysis, the model allows different estimated weights A and B.   The data vectors Xi and Zi and weight vectors A and B combine to determine 𝜇𝑖 and 𝛿 in the density of yi given above.  The maximum likelihood estimates of A and B are those that maximize the joint density of the sample (yi, Xi, Zi).  The estimation is performed using the ZOIB module in STATA, which allows for fixed effects and clustered standard errors.  As with any logistic function, the exponentiated coefficients from (3) are odds ratios; similarly, the exponentiated coefficients from (4) are ratios of proportions.  That is,  𝑒𝑍𝑖′𝐵 =𝜇𝑖1−𝜇𝑖.          (5)  In the context of this study, denote the coefficient on the indicator variable QC as 𝛽𝑄𝐶 and rewrite the LHS of A5 as 𝑒𝛽𝑄𝐶𝑒𝑍𝑖′𝐵 when QC=1 (the firm is in Quebec) and 𝑒𝑍𝑖′𝐵 when QC=0 (the firm is in the Rest of Canada).  Taking the ratio gives   𝑒𝛽𝑄𝐶 =𝜇𝑄(1−𝜇𝑄)⁄𝜇𝑅𝑂𝐶(1−𝜇𝑅𝑂𝐶)⁄, or equivalently 𝛽𝑄𝐶 = 𝑙𝑜𝑔 (𝜇𝑄(1−𝜇𝑄)⁄𝜇𝑅𝑂𝐶(1−𝜇𝑅𝑂𝐶)⁄),  where the subscript on 𝜇 indicates whether the firm is in Quebec (Q) or the Rest of Canada (ROC).  As an example, in Table 4.4, column 2, the coefficient on QC is -0.445, and so the exponentiated value is 0.641.  This means that, after controlling for the other variables in the model, the ratio of US investor ownership to non-US investor ownership is 35.9 percent lower for Quebec firms than for ROC firms.   164  Appendix B2   We use Advanced Google Search to construct both WEB and FRENCHNESS for our sample of Quebec firm-years. In the search we stipulate that the source documents be a) from Canada, b) in the year that corresponds to the firm’s fiscal year, c) in the French language, and then separately, in the English language. The search string is for the exact string of the company’s name.  We first investigate how the company’s name appears on the company website and in popular news sources to develop a suitable search string (often ending with the suffix ‘Inc.’).  We require that the search string yields the company website in the first page of the search results, and manually check that the search results indeed correspond to the company.  Once the correct search string for a company is found, we identify the documents generated about the company in each language for each fiscal year. If a company changes its name over the sample period, we make appropriate changes in the search string so that we use the correct name when searching for documents in prior fiscal years. The results are then manually checked to eliminate matches that do not correspond to the firm (typically occurring on the last few pages of the search). All redundant documents are counted, reasoning that these represent rebroadcasting of information to different audiences. A minor problem with replicability occurs because the Google search algorithm is dynamic, and proprietary. To limit this problem, the search was conducted on a computer with all browser data erased after each search, and the location of the user blocked.  All searches were performed in the month of November, 2012, but the same procedure conducted at a later point in time may yield slightly different results.  The results for our sample are available on request.  

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                        
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
http://iiif.library.ubc.ca/presentation/dsp.24.1-0314098/manifest

Comment

Related Items