University of Southern California Dissertations and Theses

Can I borrow your footnotes? Learning and network benefits of footnote similarity

(USC Thesis Other)

Can I borrow your footnotes? Learning and network benefits of footnote similarity

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content Running head: CAN I BORROW YOUR FOOTNOTES? 1 Can I Borrow Your Footnotes? Learning and Network Benefits of Footnote Similarity Jeff L. McMullin University of Southern California CAN I BORROW YOUR FOOTNOTES? 2 Author Note Jeff L. McMullin, Leventhal School of Accounting, University of Southern California. Jeff McMullin is at the Kelley School of Business, Department of Accounting, Indiana University. I am grateful for the guidance of my dissertation committee: Mark DeFond (chair), Mathew D. McCubbins, Kevin J. Murphy, KR Subramanyam, and Jieying Zhang. This paper has also benefited from discussions with Suresh Nallaready, Bryce Schonberger, Karen Ton and Benjamin Whipple and from workshop participants at the University of Arizona, Indiana University, the University of Miami, the University of Minnesota, the University of Texas at Dallas, Tulane University, Vanderbilt University, and Washington University in St. Louis. I thank Lou Bloomfield for creating and making freely available the text re-use detection software. I am also grateful to Rodrigo Verdi for making available the accounting comparability data. The Perl code on Andy Leone’s website was helpful in the early stages of developing the Perl scripts I use to collect the footnotes. I thank Jake Thornock for sharing the SEC EDGAR search volume data. This paper has also benefited from my personal discussions with people who prepare footnote disclosure. A prior version of this paper was distributed with the title, “Can I borrow your footnotes? Footnote similarity and accounting comparability.” Any errors are mine. ©2013- 2014 by Jeff Lawrence McMullin. All rights reserved. Correspondence concerning this article should be addressed to Jeff McMullin, Kelley School of Business, Indiana University, Department of Accounting, 1309 E. Tenth Street, Bloomington IN 47405-1701. Contact: jmcmulli@usc.edu CAN I BORROW YOUR FOOTNOTES? 3 Abstract Regulators and the profession have long complained that “boilerplate” footnotes impair financial reporting quality, where boilerplate refers to standardized text that is similar across firms. They raise these concerns without acknowledging that the use of boilerplate may create learning and network externalities that benefit financial statement preparers and users. One channel through which boilerplate footnotes create these externalities is by affecting accounting comparability, although it is unclear whether similarity increases or decreases comparability. I investigate this question by examining the association between accounting comparability, measures of network benefits and a measure of footnote similarity derived from text re-use detection software. I begin by documenting the factors that explain footnote similarity, and find that similarity increases among firms that are in the same industry, are geographically proximate, and share the same auditor office. In answer to my primary question, I find that footnote similarity is positively associated with accounting comparability and measures of network benefits, consistent with boilerplate footnotes improving financial reporting quality, on average. In addition, the results are robust to alternative measures of accounting comparability and to several matching methods, some of which are new to the accounting literature. These findings contribute to our understanding of the role of footnotes in improving financial reporting quality, and provide new insights into the footnote creation process. This study also adds to the growing textual analysis literature by introducing the use of text re-use detection software to measure footnote similarity, and to the general accounting literature by introducing two new matching methods. Keywords: footnote similarity, boilerplate, Disclosure Framework project, geography, auditor office, plagiarism detection CAN I BORROW YOUR FOOTNOTES? 4 Table of Contents Introduction*................................................................................................................................................................*6* Motivation*.................................................................................................................................................................*12* Concerns*About*Boilerplate*Footnotes*....................................................................................................................*12* Learning*and*Network*Externalities*of*Boilerplate*............................................................................................*14* What*Contributes*to*Footnote*Similarity?*..............................................................................................................*15* SEC*EDGAR*Traffic*and*Footnote*Similarity*..........................................................................................................*19* Accounting*Comparability*and*Footnote*Similarity*...........................................................................................*20* Network*Benefits*of*Footnote*Similarity*................................................................................................................*21* Sample*and*Variable*Measurement*...............................................................................................................*23* Sample*Selection*................................................................................................................................................................*23* Measuring*Similarity*........................................................................................................................................................*24* Other*Variables*...................................................................................................................................................................*26* Measuring*Accounting*Comparability*......................................................................................................................*27* Measuring*EDGAR*Search*Volume*.............................................................................................................................*30* Measuring*of*Network*Benefits*...................................................................................................................................*31* Measuring*Economic*Similarity*..................................................................................................................................*32* Footnote*Similarity*Descriptive*Statistics*..............................................................................................................*32* Empirical*Results*...................................................................................................................................................*34* Industry,*Geographical*Location,*and*Auditor*Office*Subsample*Analysis*..............................................*34* Comparing*High*Footnote*Similarity*Firms*and*Low*Footnote*Similarity*Firms*.................................*37* Regression*Model*..............................................................................................................................................................*39* EDGAR*Search*Volume*Analysis*.................................................................................................................................*41* Accounting*Comparability*Analysis*..........................................................................................................................*42* CAN I BORROW YOUR FOOTNOTES? 5 Network*Benefits*...............................................................................................................................................................*44* Additional*Analysis*...............................................................................................................................................*46* Auditor*Changes*.................................................................................................................................................................*46* Sensitivity*Tests*.....................................................................................................................................................*54* Exclude*Loss*Firms*...........................................................................................................................................................*54* Big*N*Audited*Companies*..............................................................................................................................................*54* Arthur*Andersen*Auditor*Changes*............................................................................................................................*54* Conclusion*.................................................................................................................................................................*55* References*.................................................................................................................................................................*57* Appendix*A.*Footnote*Examples*.....................................................................................................................*62* Appendix*B.*Variable*Definitions*....................................................................................................................*64* Appendix*C.*Matching*methods:*propensity*score*matching,*genetic*matching,*and*entropy* balancing*...................................................................................................................................................................*67* Matching*Methods*.............................................................................................................................................................*67* Propensity*Score*Matching*...........................................................................................................................................*68* Genetic*Matching*and*Entropy*Balancing*...............................................................................................................*69* Comparing*Matching*Methods*.....................................................................................................................................*71* Tables*..........................................................................................................................................................................*75* CAN I BORROW YOUR FOOTNOTES? 6 Can I Borrow Your Footnotes? Learning and Network Benefits of Footnote Similarity Introduction Regulators, standard setters, investors, and accounting firms all express concerns that footnote disclosures fail to clearly communicate the information that investors need (e.g. ITAC, 2007; Radin, 2007; CIFR, 2008; FRC, 2009, 2011, and 2012; FERF and KPMG, 2011; FASB, 2012; and EFRAG, ANC, and FRC, 2012). They assert that an important reason for this failure is firms’ use of boilerplate language, where boilerplate refers to standardized text that is similar across firms. However, the legal contracts literature provides theory that implies the use of footnote boilerplate creates learning and network externalities that benefit footnote preparers and financial statement users (Kahan and Klausner, 1997). Footnote preparers benefit from the learning externality through reduced preparation costs and potential regulatory scrutiny. Financial statement users benefit from the network externality through increased comparability of accounting disclosures. The primary purpose of this study is to empirically examine the existence of both the learning and network externalities of footnote boilerplate and whether these externalities benefit financial statement users. Footnotes are an integral part of the financial statements and are meant to communicate information that either cannot be or is not on the face of the primary financial statements (FASB, 2012). Boilerplate theory suggests managers use the learning externality created by other financial statements preparers to reduce footnote preparation costs and the possibility of future regulatory scrutiny. While management is primarily responsible for preparing the financial statements, the professional literature suggests that managers may benefit from the learning externality by seeking help from two outside sources when creating the footnotes: the financial CAN I BORROW YOUR FOOTNOTES? 7 statements of industry peers, and the firm’s independent auditor (AICPA, 2012; Burke, 2002). Both outside sources are likely to increase footnote similarity across firms. For example, referring to the footnotes of industry peers increases the chance that managers will “borrow” their peer’s footnote language, and the footnote language suggested by outside auditors is likely to be “borrowed” from the auditor’s other clients. Footnote similarity may provide network benefits to financial statement users by affecting accounting comparability, an enhancing qualitative characteristic of financial reporting. The Financial Accounting Standards Board’s (FASB) Conceptual Framework states that for information to be comparable “like things must look alike and different things must look different” (FASB, 2010). However, while borrowing footnote language from these outside sources is likely to increase footnote similarity, whether this similarity benefits financial statement users is not clear. On one hand, if input from these sources leads managers to use similar footnotes to describe similar underlying economic phenomena, network benefits are created and comparability is enhanced by making like things look alike. On the other hand, if these sources lead managers to use similar footnotes to describe dissimilar underlying phenomena, network benefits are not created and comparability will be harmed by making unlike things look alike. Thus, the use of boilerplate text that increases footnote similarity may either enhance comparability, resulting in network benefits, by making like things look alike, or it may impair comparability by making different things look alike. I begin my analysis by first empirically modeling the factors that explain footnote similarity. I measure footnote similarity using text re-use detection software, a method that has CAN I BORROW YOUR FOOTNOTES? 8 not been previously used in the accounting literature. 1 Specifically, my footnote similarity measure equals the number of words in text strings, at least six words in length, which two firms use in their footnotes, scaled by the number of words in their footnotes. Firms with larger proportions of identical word strings of six or more words in length have relatively high footnote similarity, and firms with smaller proportions of identical word strings have relatively low footnote similarity. This analysis finds that footnote similarity is higher for industry peers that are in close geographical proximity and that share the same auditor office. These effects are also economically large. Specifically, when compared with firms that have low footnote similarity, firms with high footnote similarity are 5.01 times more likely to be in close geographical proximity, and 12.18 times more likely to share the same auditor office, but only 1.1 times more likely to operate in the same industry. Corroborating the association between footnote similarity and auditor office, I also find that the footnotes of firms that change auditors become more similar to the footnotes of the successor auditors’ clients, and less similar to the footnotes of the predecessor auditors’ clients. I also find that footnote similarity is relatively lower for large firms and more complex firms, consistent with firm size and complexity leading to greater entity specific differences, which are reflected in the footnote language. Additionally, I find evidence consistent with footnote “borrowing” being concentrated among relatively smaller firms who tend to borrow from their industry peers who are somewhat larger. Finally, I find that the similarity of a firm’s footnotes to other firms’ footnotes, filed after the firm has filed its 10-K with the Securities and Exchange Commission, is correlated with SEC’s Electronic Data- Gathering, Analysis, and Retrieval (EDGAR) system traffic accessing the firm’s 10-K. This 1 Text re-use detection software is also often referred to as plagiarism detection software, copy-paste detection software, and text comparison software. Although this type of software is typically used to assist in plagiarism detection, no software can conclusively detect plagiarism (Maurer, Kappe, and Zaka, 2006). Thus, I make no assertions regarding whether the textual similarities I identify constitute plagiarism. CAN I BORROW YOUR FOOTNOTES? 9 provides evidence footnote preparers search for potential language to borrow in financial statements previously filed on SEC EDGAR and take advantage of the learning externality created by financial statement preparers of previously prepared financial statements. The legal contract literature’s theory of boilerplate implies the use of footnote boilerplate creates network externalities. One of these externalities may be an increase in the comparability of accounting information. To examine this theory, I explore the relation between footnote similarity and comparability, using three measures of comparability. Two of these measures are developed in De Franco, Kothari, and Verdi (2011) and I develop the third by replacing the earnings-return regression in DeFranco et al. (2011) with the working capital accruals model developed in Dechow and Dichev (2002). While controlling for the economic similarity of the firms using a measure derived from the product description in the 10-K, the analysis of comparability finds that footnote similarity is positively associated with these three measures of accounting comparability, consistent with the use of footnote boilerplate leading to beneficial network externalities. Finally, I explore whether footnote similarity is correlated with evidence of network benefits. I search for evidence of benefits in three settings, (1) the stock price reaction to a firm’s 10-K filing, (2) sell-side equity analyst activity and forecast characteristics, and (3) the stock price reaction of one firm when a firm with similar footnotes files a 10-K or announces earnings. In each of these settings, I find evidence that financial statement users benefit from the network externality created when similar footnote language is used to describe similar underlying economic phenomena. This is interesting because footnote critics complain that boilerplate footnote language, which is likely to be correlated with my measure of footnote similarity, reduces the usefulness of the accounting information. CAN I BORROW YOUR FOOTNOTES? 10 While the main analysis examines the relation between footnote similarity and accounting quality using multivariate regression, an alternative approach involves matching methods to control for confounding factors. Thus, I repeat my primary analysis using three matching methods: (1) propensity score matching, (2) genetic matching, and (3) entropy balancing. While propensity score matching has been in the literature for many years (Rosenbaum and Rubin, 1983), genetic matching and entropy balancing are newly developed methods that seek to improve upon propensity score matching (Diamond and Sekhon, forthcoming; Hainmueller, 2012). This analysis finds that all of the study’s main findings are robust to using these three matching techniques. 2 This study makes several contributions to the literature. First, the results have implications for the FASB’s Disclosure Framework project, which attempts to improve footnote disclosure. The FASB, the SEC, and the profession blame boilerplate language for impairing footnote disclosure effectiveness (e.g., FASB, 2012; EFRAG et al., 2012). To inform this project, I test theories of boilerplate, initially developed in the legal contract literature, that argue the use of boilerplate create learning and network externalities. Contrary the arguments put forward by standard setters and regulators, my findings suggest that the use of footnote boilerplate language creates learning and network externalities. These externalities, in turn, benefit both financial statement preparers and users. Second, this study contributes to the mandatory disclosure literature by examining the externalities created by footnote similarity and how they benefit users by enhancing the comparability of the disclosure. While footnotes are an integral part of the financial statements, 2 The results are also robust to two alternative measures of comparability as captured by information transfers, which examine how firm i’s stock price changes when firm j experiences a news event—an earnings release or a 10-k filing. CAN I BORROW YOUR FOOTNOTES? 11 there is relatively little research on how footnotes affect financial reporting quality. 3 Third, this study contributes to the growing literature on accounting comparability in two ways. One way is by examining the link between management’s disclosure decisions and comparability. While most of the comparability literature examines the benefits of comparability (e.g., De Franco et al., 2011), or how factors that are outside of managers’ control impact comparability (e.g., DeFond, Hu, Hung, and Li 2011), I provide evidence that managers’ disclosure decisions can also influence comparability. Another way is by developing a new measure of comparability by substituting the Dechow and Dichev (2002) model of working capital accruals for the earnings- returns regression in the mapping framework developed by De Franco et al. 2011. Like the earnings-returns model, the Dechow and Dichev (2002) model of work capital accruals models how economic events, cash flows, are translated into accounting information, working capital accruals. This is another measure of the comparability of two firms’ accounting systems. Fourth, this study contributes to the broad accounting literature by introducing two recently developed matching methods for estimating causal effects. While accounting researchers commonly implement propensity score matching in robustness tests, the matching literature has proposed several improvements to this technique since Rosenbaum and Rubin (1983) first proposed it. I introduce two of the most recently developed methods, genetic matching (Diamond and Sekhon, forthcoming) and entropy balancing (Hainmueller, 2012), to the accounting literature. Finally, this study contributes to the rapidly growing textual analysis literature by introducing a method for measuring text similarity. Prior research uses a variety of methods to 3 In a contemporaneous working paper, Petersen et al., (2012) use the vector space model from Brown and Tucker (2011) to measure individual word choice similarity in the accounting policy footnote and find that it is negatively associated with accounting comparability. Thus, while Petersen et al., (2012) also perform textual analysis of footnotes, they use a different measure, focus on a single footnote, and find results opposite of the results I present. CAN I BORROW YOUR FOOTNOTES? 12 analyze narrative accounting disclosure. For example, Li (2008) measures readability of 10-Ks by using Gunning’s (1952) FOG index. Brown and Tucker (2011) employ a vector space model to measure modifications to MD&A disclosure. Other studies develop measures by counting the number or types of words. For example, Rogers, Van Buskirk, and Zechman (2011) measure tone using pre-defined word lists; and Li, Lundholm, and Minnis (2012) measure competition by counting the number of times the word “competition” appears in the 10-K. Li (2010) measures forward-looking statements using a Naïve Bayesian machine learning algorithm. However, no study that I am aware of in this growing literature employs text re-use detection software to measure the similarity between texts. 4 The next section discusses the motivation for this study. Section 3 discusses the sample and variable measurement. Section 4 reports the results. Additional analyses and sensitivity tests are reported in Section 5 and Section 6, respectively. Finally, Section 7 concludes. Motivation Concerns About Boilerplate Footnotes The FASB’s Statement of Financial Accounting Concepts No. 8 states, “The objective of general purpose financial reporting is to provide financial information about the reporting entity that is useful to existing and potential investors, lenders, and other creditors in making decisions about providing resources to the entity” (FASB, 2010, OB2). The balance sheet, income statement, and statement of cash flows provide the user with much of this information. However, due to the nature of these statements, they are inherently limited in the type of information they provide. Footnote disclosure provides a means to communicate information that either cannot be or is not on the face of the financial statement (FASB, 2012). 4 One notable exception is a recent dissertation, Li (2013). Li (2013) uses MOSS, a program developed to detect plagiarism in computer code, to measure footnote disclosure repeated in the MD&A section. CAN I BORROW YOUR FOOTNOTES? 13 A large number of regulators, standard setters, and professional groups have expressed concerns that footnote disclosure is inadequate and point to the use of boilerplate language as a major source of this inadequacy (Radin, 2007; CIFR, 2008; FRC, 2009, 2011, 2012; Deloitte 2010; FERF and KPMG, 2011; IAASB, 2011; ICAS and NZICA 2011; EFRAG et al., 2012; FASB, 2012). A contributing factor to the use of boilerplate language is that footnote content is typically determined on a standard-by-standard basis. The SEC’s Advisory Committee on Improvements to Financial Reporting (CIFR) argues that this piecemeal process results in “redundancies, confusion, and disorganized presentations in financial reports” (CIFR, 2008). Further, the European Financial Reporting Advisory Group (EFRAG) argues that a by-product of this process is “Auditors often use checklists or produce templates for financials statements, which can move the focus away from communication” (EFRAG et al., 2012). 5 The FASB argues that boilerplate disclosure does not provide relevant information (FASB, 2012). 6 The SEC Staff is also on record as criticizing the use of “overly vague or ‘boilerplate’ disclosure” in revenue recognition and internal controls over financial reporting disclosures (Carnall, et al. 2010). Finally, the United Kingdom’s Financial Reporting Council (FRC) writes “dense boilerplate text that is hard to follow is neither interesting nor engaging” (FRC, 2009). In response to the concerns expressed about footnote disclosure, and to a call from the FASB’s Investor’s Technical Advisory Committee (ITAC, 2007), the FASB commenced the Disclosure Framework project, the purpose of which is to improve the effectiveness of 5 Established in June 2001, the EFRAG’s purpose is to (1) provide the European Commission with technical advice about the use of international accounting standards (IAS) throughout Europe, (2) participate in the IASB’s standard setting process, and (3) within Europe, coordinate the development of views concerning IAS. 6 Ron Lott, FASB research director has stated, “I don’t personally believe [boilerplate] serves investors all that well” (Cohn 2012). CAN I BORROW YOUR FOOTNOTES? 14 disclosures in notes to financial statements (FASB, 2012). 7 A recent Financial Accounting Standards Advisory Council (FASAC) survey reports that the most commonly identify project that should be on the FASB’s agenda is the Disclosure Framework project (FASB, 2013). An underlying theme of this proposed reform is to move away from boilerplate disclosures toward more entity specific language. However, other than conjecture, little is known about the effects of boilerplate footnote disclosures on financial reporting quality. Learning and Network Externalities of Boilerplate The legal contracts literature provides a framework for thinking about the economic externalities created by the use of boilerplate language in contracts (Kahan and Klausner, 1997). This framework explains there are two externalities, a learning externality and a network externality, that arise when boilerplate contract terms are used. The initial writer of the boilerplate language and prior contract writers that use the boilerplate language confer the learning externality upon later contract writers. This learning externality benefits the writer of a contract when the writer uses boilerplate instead of drafting original language. Kahan and Klausner (1997) explain that for the individuals contracting in the later period, the learning benefit (1) increases drafting efficiency, (2) reduces uncertainty over the meaning of a term if there has been judicial involvement in determining the meaning of the term, and (3) increases the familiarity of the terms among contracting parties. In sum, the legal contracts literature argues the use of boilerplate reduces transaction and negotiation costs by improving understanding between parties to a contract and reduces uncertainty of the language used (Ahdieh, 2006). The network externality arises when boilerplate language is used in multiple contracts contemporaneously. Kahan and Klausner (1997) explain that one of the network benefits arising 7 Holzmann and Ramnath (2013) provide an summary of the FASB’s invitation to comment on the discussion paper “Disclosure Framework” (FASB, 2012). CAN I BORROW YOUR FOOTNOTES? 15 from this externality is that accountants and lawyers will provide higher quality service at lower cost. These professionals’ expertise grows as they encounter questions or disputes regarding the boilerplate language. Another benefit derived from the network externality is the number of investors and analysts of the contracts that can understand the language of the contract is greater if more boilerplate is used. Finally, a user of boilerplate benefits when there is a judicial interpretation of the language. The contracting parties benefit since the uncertainty of the language is resolved in court without having to pay the legal costs of obtaining judicial interpretation. The theory of boilerplate externalities is easily transferable to a footnote setting. Kahan and Klausner (1997) note that their theory may be relevant in non-contractual relationships as well. Below I describe how both the learning externality and the network externality create benefits for financial statement prepares and users. What Contributes to Footnote Similarity? Management is responsible for the preparation of financial statements. Apart from being management’s responsibility, management possesses the best understanding of the company’s strategy, organization, and risk exposures; and consequently, is in the best position to translate the economic phenomena underlying the firm into accounting information. The usefulness of footnote disclosure depends on how well managers perform this translation. However, the creation of accounting information is not a costless activity. The boilerplate theory predicts mangers reduce the cost of this activity by borrowing language from previously issued financials statements thus benefiting from the learning externality. Likewise, the professional literature suggests that managers are likely to reduce the costs of footnote preparation by seeking input from outside sources (AICPA, 2012; Burke, 2002). Specifically, managers are likely to prepare CAN I BORROW YOUR FOOTNOTES? 16 their footnotes based on input from industry and geographically proximate peer firms, and from guidance provided by their outside auditor. The effects of these inputs on similarity are discussed in the following sections. Industry and footnote similarity. By definition, firms within the same industry engage in similar economic activities. Given that the purpose of accounting is measure economic activity, firms in the same industry use similar accounting principles in generating accounting information. Not only will the similarity of the underlying economics lead manager to select similar accounting practices, but also managers keeping an eye on industry peers to make sure their disclosure is similar to their peer firms’ disclosure well result in similarity of accounting choices across firms in the same industry. In my personal conversations with managers that prepare financial statements, managers often expressed a desire to disclose everything that their peers disclosed and nothing that their peers did not disclose. Moreover, some investors focus on identifying troubled companies by identifying footnote disclosure that looks different from the industry norm (Leder, 2003). These reasons, along with the decrease in costs of footnote preparation due to the learning benefit lead me to expect that firms in the same industry have greater footnote similarity than firms in separate industries. Geographic proximity and footnote similarity. Using footnotes of other companies as guidance allows managers to obtain footnote language and tables already vetted by other managers, audit committees, auditors, and regulators thus benefiting from the learning externality. Not only does the use of other firms’ footnote language as guidance decrease the cost of writing new language, but it may also decreases the risk of costly litigation and regulatory investigation. Additionally, a manager concerned about revealing proprietary information in footnotes may use boilerplate disclosure to avoid disclosing this information. My conversations CAN I BORROW YOUR FOOTNOTES? 17 with several managers of public companies whose responsibilities include footnote preparation indicate that mangers searching for example footnotes are likely to start with the firm’s industry peers in the same geographic region. Industry peers’ footnotes likely contain footnote disclosure similar to what the manager is looking for since firms in the same industry experience similar underlying economic phenomenon. Similar to the way in which Generally Accepted Accounting Principles (GAAP) evolve from firms adopting accounting principles that are generally used, footnote boilerplate may also evolve from managers observing footnotes that are generally used. Contact with managers of geographically proximate peer firms is facilitated by a variety of professional and social mechanisms. These include professional organizations, educational experiences, between firm job changes, and other common day-to-day activities. Managers I spoke with often singled out one professional organization in particular that seems to be especially effective in contributing to footnote sharing, Financial Executives International (FEI). FEI was founded more than 80 years ago and has membership of more than 15,000 senior-level financial executives in more than 85 local chapters across the US (see www.financialexecutives.org). Another professional organization that facilitates footnote sharing that was cited by managers I spoke with is the SEC Professional Group. Only professionals who actively prepare and file financial reports with the SEC can join this network of professionals; and their stated purpose is to promote professional communication among its members (see www.secprofessionals.org). This organization also provides workshops and training meetings at the local chapter level. This type of dissemination of footnote disclosure language is consistent with diffusion theory (Rogers, 2003). While sharing footnote disclosures with geographically proximate peers is likely to increase footnote similarity, it is not clear what impact this relation has on accounting CAN I BORROW YOUR FOOTNOTES? 18 comparability. Exchanging information and comparing footnote disclosures among a broad set of industry peers may result in managers identifying footnotes that accurately capture their firm’s underlying economics. However, if managers are simply interested in adopting disclosures that meet the minimum standards for regulatory compliance, this exercise may result in the adoption of footnotes that do not faithfully represent the firms underlying economics. Therefore, while I expect industry peer geographic proximity to increase footnote similarity, it is unclear whether this increased similarity is likely to improve comparability. Auditor office and footnote similarity. Another common source of input in helping managers prepare footnotes is from the external auditor. However, auditor independence rules restrict the role the auditor can play in financial statement preparation. According to Regulation S-X, Rule 2-01, auditors are forbidden from assuming any managerial responsibilities (see Regulation S-X, 17 C.F.R. 210.2-01(b)). Since the preparation of financial statements, including the footnotes, is management’s responsibility, an auditor who prepares the financial statements lacks independence. The SEC’s stance on the auditor’s participation in footnote preparation was recently elaborated upon by the then Associate Chief Accountant (Burke, 2002). The SEC makes it clear that since the footnotes must be independently audited by an outside party, involvement in footnote preparation biases the auditor. However, auditors are allowed to provide their clients with editorial suggestions and comments on the footnote presentation as long as management originally proposes the presentation. Importantly, while the auditor’s input is limited by independence concerns, the SEC indicates that input from the auditor on footnote preparation is “in the public interest as it more often than not results in improved financial reporting.” CAN I BORROW YOUR FOOTNOTES? 19 The external auditor likely possesses a detailed knowledge of a large set of firms’ financial statements, overlapping to a large extent with the set of clients serviced from the auditor’s office. When providing footnote advisory services, auditors likely draw on their expertise, which is typically localized at the auditor-office level. Since auditors participate in the 10-K filing process with multiple companies, their knowledge may increase the quality of their clients’ financial reporting by suggesting footnote language that meets regulatory approval and increases the faithful representativeness of the financial statements. If this is the case, then footnote advice from auditors will lead to footnote similarity that also increases comparability. However, to the extent that auditors provide managers with disclosures that simply comply with regulatory requirement and/or fail to capture the underlying economics driving the reason for disclosure, this footnote borrowing may result in similarity that decreases comparability. Thus, while I expect footnote similarity to be relatively greater among firms that share the same outside auditor office, it is unclear whether this increased similarity improves or impairs comparability. SEC EDGAR Traffic and Footnote Similarity In conversations with financial statement preparers, an often mentioned source used to obtain examples of footnote disclosure is the SEC EDGAR system. EDGAR provides the public with free access to completed forms filed by company. One such form is form 10-K which includes companies’ annual financial statements. While prior research has examined user search volume on EDGAR from an investor perspective (e.g., Drake, Roulstone, and Thornock, 2013), I examine EDGAR user traffic from a financial statement preparers perspective. In order for a financial statement preparer to exploit the learning externality of footnote boilerplate, the preparer must have access to examples of how the boilerplate is used. Thus, I expect 10-Ks that CAN I BORROW YOUR FOOTNOTES? 20 contain greater amounts of footnote language that can be found in financial statement in the following year experience greater EDGAR search volume during the year. Accounting Comparability and Footnote Similarity Statement of Financial Accounting Concepts (SFAC) No. 8 defines comparability as "the qualitative characteristic that enables users to identify and understand similarities in, and differences among, items” (FASB, 2010, QC21). Comparable footnote disclosure enhances investor understanding of the economic phenomena the footnotes report. If two similar footnotes both accurately capture the underlying economics, then economic phenomena that is alike looks alike. Concept Statement No. 8 spells out this concept. “A faithful representation of a relevant economic phenomenon should naturally possess some degree of comparability with a faithful representation of a similar relevant economic phenomenon by another reporting entity” (FASB, 2010, QC24). The effect of footnote similarity on accounting comparability is not straightforward and depends on whether the footnotes accurately capture the underlying economic phenomena being disclosed. As stated in the Concept Statements, comparability is not enhanced by making unlike things look alike any more than it is enhanced by making like things look different (FASB, 2010, QC24). Thus, if footnotes are similar because firms use boilerplate language to describe economically similar phenomena, then footnote similarity enhances accounting comparability. Healy and Palepu (2001) point out that accounting standards regulation “potentially reduces processing costs for financial statement users by providing a commonly accepted language that managers can use to communicate with investors” (p. 412, emphasis added). However, if footnotes are similar because firms inappropriately use boilerplate language to describe economically dissimilar phenomena, then footnote similarity harms accounting comparability. CAN I BORROW YOUR FOOTNOTES? 21 Boilerplate language has also been criticized by the legal scholars. Specifically, the legal contracting literature argues that boilerplate is often included in contracts despite the drafters not understanding the boilerplate’s genesis or meaning (Gulati and Scott, 2011). Thus financial statement users only benefit from network externalities created by footnote similarity if the boilerplate accurately reflects the underlying economics of the firm. Network Benefits of Footnote Similarity An alternative way to examine whether footnote similarity improves financial reporting quality is to examine outcome measure of financial statement users. Investors and analysts are two types of financial statement users that may benefit from footnote similarity. Investors examining a firm’s footnotes are better able to process the information in the footnotes if the disclosure uses language they are familiar with. Thus, the information processing costs of financial statements that are more similar may be lower. However, if footnote boilerplate is uninformative, as regulators posit, the information in the footnotes will be useless to investors. Therefore, if footnote similarity is beneficial to investors I expect the market reaction to the filing of a 10-K to be greater if the footnotes have greater footnote similarity. On the other hand, if footnote boilerplate is not informative, I expect the market reaction to the filing of the 10-K will be smaller when the footnotes have greater similarity. Prior literature documents investors use information from peer firms when making investment decisions (e.g., Foster 1981, and Ramnath 2002). Investors using financial statements with greater footnote similarity benefit more easily use information about one firm to inform decisions about the other firm. Thus when two firms have greater footnote similarity, I expect the information transfer between firms around important information events to be greater if CAN I BORROW YOUR FOOTNOTES? 22 boilerplate use creates network benefits. However, if footnote boilerplate is less informative disclosure then this information transfer is not likely. The second group of users that may benefit from footnote similarity is sell-side equity analysts. Equity analysts are an important group of financial statement users. Prior research has examined how the characteristics of the 10-K impacts analyst following and analyst earnings forecast accuracy and dispersion. Lehavy, et al. 2011 find that greater readability improves analyst forecasts and increases analyst following. However, Brown and Tucker 2011 find that MD&A modifications are not associated with earnings forecast revisions. These analysts following multiple firms are required to process a large amount of information to generate recommendations and forecasts. Footnote similarity may impact two decisions analyst make, the decision to follow a firm and the analysts’ earnings forecasts. One factor that impacts an analyst’s decision to follow a firm is the information processing cost of following the firm. When the information processing cost required to follow a firm is lower, an analyst is more likely to follow the firm. An additional factor is whether following the firm will enhance the information set the analyst can use when examining the other firms she follows. Another reason footnote similarity may impact the decision to follow a firm is that analysts may benefit from the network externality of similar footnotes by spreading the cost to understand a given footnote over multiple firms. Thus, if footnote similarity decreases analyst information processing cost of following a firm, I expect firms with greater footnote similarity are followed by more analysts. Alternatively if greater footnote similarity impairs the informativeness of footnote disclosure, I expect firms with greater footnote similarity are followed by less analysts. CAN I BORROW YOUR FOOTNOTES? 23 Footnote similarity may also impact analyst forecasts of earnings. Analysts following firms with greater footnote similarity have more firms that can easily be compared when forecasting earnings, if footnote similarity improves the informativeness of the accounting disclosure. Thus, I expect analyst forecast accuracy to be greater and dispersion to be lower if greater footnote similarity enhances financial statement informativeness. I expect accuracy to be lower and dispersion to be higher if the use of footnote boilerplate impairs the usefulness of financial statements. Sample and Variable Measurement Sample Selection My sample consists of all firms from 1996-2010 with complete data on SEC’s EDGAR database, the Compustat annual file, and CRSP. I begin with 1996 because complete implementation of EDGAR occurred in 1996. I exclude holding companies, American Depository Receipts (ADRs) and limited partnerships to avoid matching parent and subsidiary companies and avoid examining filings of non-corporation entities. 8 Also, I exclude financial firms (SIC codes 6000-6999) and use Perl to extract the text of the footnotes from 10-Ks and remove all HTML code as well as HTML tables. 9 To construct the measure of footnote similarity, I identify a comparison group of J footnote sets for each footnote set i, where i denotes each firm-year in my sample. Computational linguistics commonly refers to this comparison group as the text corpus. Thus, hereafter I refer 8 I exclude these observations by identifying whether the company name (CONM) contains the word Holdings, Group, ADR, or LP (and commonly used abbreviations such as, Holding, Hldgs, Grp, ADR, -ADR, -LP). If at least one of these words appears in the company name, I exclude the observation. 9 To fine tune my script, I perform multiple analytical procedures over the whole sample (e.g., compare the length of the extracted notes to the prior year’s extracted notes, and examine the first and last 200 characters of the extracted text) and hand audits of over 1,000 sets of the extracted text to identify cases where my script failed. Upon identifying incorrect extractions, I either make modifications to correct the script, or extract the footnotes by hand. Also, I truncate observations at the one-half percentile and ninety nine and one-half percentile of extracted text length. CAN I BORROW YOUR FOOTNOTES? 24 to it as corpus i . The subscript denotes the corpus corresponds to footnote set i. Each corpus i consists of the most recent two years of footnotes for firms in the SIC 2-digit industry of the firm filing footnote set i. 10 By requiring the filing of the J footnote sets in corpus i to occur before the filing of footnote set i, I limit the possibility the firm filing footnote set j used footnote set i as guidance, thereby allowing me to infer the direction of text re-use. By only comparing firms within the same industry, I hold industry constant. To allow for a reasonable number of footnote sets to construct corpus i , I exclude 1996 and 1997 from the sample I analyze. This results in 40,287 sets of footnotes. Measuring Similarity I use the text re-use detection software program WCopyFind to measure footnote similarity between footnote set i and each footnote set j in corpus i . 11,12 This software calculates the number of words in text strings appearing in two documents. WCopyFind allows the user to control the parameters that define a text string match and punctuation, numbers, letter case, and words with a certain length can be ignored. Also, the user decides the minimum number of consecutive words that must match before a text string is considered a match. While selecting a large number as the minimum prevents the software from identifying commonly used text strings as re-used text, selecting a large minimum causes the software to fail to detect re-used text that is slightly edited. Another feature of WCopyFind attempts to circumvent this problem by allowing a small 10 To avoid comparing the footnotes of firm i with footnotes of firms in the same industry that have stopped filing with the SEC, I require sets of footnotes in set J to have been filed within four years prior to the footnote i filing. 11 The WCopyFind software was develop and is made freely available for download by Lou Bloomfield, a physics professor at the University of Virginia. He developed this software to detect instances of text re-use in essays submitted by students in his physics classes. See http://plagiarism.bloomfieldmedia.com/z-wordpress/software/. 12 Although there are several software programs that detect text re-use, the majority do not allow the user to control the content of the text corpus. Also, most services require the user to upload the documents for comparison on remote servers. WCopyFind allows the user to control the content of the text corpus and all comparisons occur on a local computer. CAN I BORROW YOUR FOOTNOTES? 25 number of differences in text strings while still considering the string a match. This makes the search for matching text strings less sensitive to minor edits made to re-used text while decreasing the chance of identifying commonly used phrases as instances of text re-use. I use two measures of footnote similarity, one is a raw count and the other scaled by the number of words in the footnotes. The raw count measure of footnote similarity between footnote set i and footnote set j is labeled #Text Match ij and is defined as the number of words in text strings, at least six words in length, that are in both footnote sets. The scaled measure is labeled %Text Match ij and is #Text Match ij divided by #Words in Footnotes i , where #Words in Footnotes i is the number of words in footnote set i. I select WCopyFind’s options to ignore punctuation, numbers, letter case and words longer than 20 characters. I allow a maximum of six differences in matching text string, while requiring at least 60% of the words to match exactly. Using WCopyFind I calculate #Text Match ij for comparisons between footnote set i and all J footnote sets in corpus i . This results in 17,860,665 comparisons, which equals the number of sample firms i compared to all firms in corpus i . To obtain a firm-level footnote similarity measure I select the highest value of %Text Match ij for each i and denote this measure %Text Match MAXij , and I select the highest value of #Text Match ij for each i and denote this measure #Text Match MAXij . Selecting the highest similarity comparison for each sample firm provides a measure of the likelihood that firm i has re-used text from any of its industry peers. In other words, firms with footnotes that are highly similar to an industry peer firm are more likely to have borrowed footnote text than firms with footnotes that are not very similar to an industry peer. Therefore, I expect these measures of footnote similarity to capture the likelihood of footnote borrowing. CAN I BORROW YOUR FOOTNOTES? 26 Appendix A provides two examples of footnotes that are identified by the footnote similarity measure to contain high amounts of similarity. The companies within each example are industry peers that share the same geographic proximity and share the same auditor office. The text strings that are identical matches are indicated in underlined red font, and those that are imperfect matches are in italicized green font. Comparisons across the footnotes in each group indicate that there is a significant amount of similarity in the text. For example, the Volcom, Inc. and Quicksilver, Inc. footnotes contain virtually identical text strings 90 and 94 words in length, respectively. Thus, the similarity measure appears to successfully capture the use of "boilerplate" footnotes across companies. Other Variables I obtain data on the location of the firms in my sample from Compustat’s Company File. The United States Census Bureau’s website provides data that allow me map from a firm’s county to its Metropolitan Statistical Area (MSA.) 13 I identify the location of the firms’ external auditor and obtain auditor changes data from Audit Analytics. The financial variables I use are from the Compustat Fundamental Annual File. I measure firm age as the number of years since a company first appeared on CRSP. Geographical and business segments data come from Compustat’s Segment Detail File. I obtain institutional ownership data from Thomson Financial Services. I obtain the De Franco et al., (2011) comparability variable from Rodrigo Verdi’s 13 For companies that do not have data on county location on Compustat, but do have zip code, I use a mapping from zip code to MSA provided by the Missouri Census Data Center’s website to identify these companies’ MSAs (http://mcdc.missouri.edu/websas/geocorr12.html). CAN I BORROW YOUR FOOTNOTES? 27 website. 14 I use CRSP to measure the information transfer variables. Appendix B lists the definitions for all variables used in the analysis. Measuring Accounting Comparability Accounting comparability using an earnings-return model. De Franco et al. (2011) model a firm’s accounting system as a mapping of economic events to financial statements. They construct an empirical measure of accounting comparability by first regressing quarterly net income on quarterly stock returns. Specifically, they estimate the following model over the preceding 16 quarters of data: !"#$%$&! !" =! ! +! ! !"#$%! !" +! !" (1) Where Earnings it is net income before extraordinary items divided by beginning of period market value of equity, and Return it is the quarterly stock return. The estimated coefficients ! ! and ! ! represent firm i’s accounting system that translates economic events in accounting information. De Franco et al. (2011) measure accounting comparability as how similar two firms’ accounting systems map economic events into accounting information. To measure the similarity in the mapping, they first calculate: !"#$%$&' !!" =! ! +! ! !"#$%! !" (2) !"#$%$&' !"# =! ! +! ! !"#$%! !" (3) Where !"#!"!#$ !!" is the predicted value of firm i’s earnings given firm i’s estimated α and β and firm i’s return in period t; and, !"#$%$&' !"# is the predicted value of firm i’s earnings given firm j’s estimated α and β and firm i’s return in period t. After obtaining these predicted values, 14 The data on Rodrigo Verdi’s website ends in 2009. I use the SAS code he provides to generate the accounting comparability measure for 2010 and 2011. I require through 2011 since I examine the relation between similarity at t and comparability at t+1. CAN I BORROW YOUR FOOTNOTES? 28 to measure Accounting Comparability ijt,DKV they calculate the average negative absolute difference between !"#$%$&' !!" and !"#$%$&' !"# . More specifically, !""#$!"#!$!!"#$%&%'()(*! !"#,!"# = !− 1 16 !"#$%$&' !!" −!"#$%$&' !"# ! !!!" (4) Multiplying the left hand side by negative one makes greater values represent greater accounting comparability. They estimate Accounting Comparability ijt,DKV for all ij combinations within the same two-digit SIC industry. They measure a firm-year specific measure of accounting comparability Accounting Comparability ijt,DKV as the average of the four highest values of Accounting Comparability ijt,DKV for firm i. In my analysis, I use this firm-year measure of accounting comparability. Accounting comparability using a model of working capital accruals. The main idea behind the De Franco et al., (2011) accounting comparability measure is to capture the comparability between firms’ accounting systems that maps economic events into accounting information. Therefore, a natural extension of this measure is to substitute other economic models that model accounting information as a function of economic events for the earnings- returns model (equation (1)). 15 Dechow and Dichev (2002) develop such a model of the mapping function of accruals to cash flows by modeling working capital accruals—accounting information—as a function of past, present and future cash flows—economic events. 16 The model is as follows: Δ!" !" = ! ! +! !! !"# !"!! +! !! !"# !" +! !! !"# !"!! +! !" (5) 15 De Franco et al. 2011 caveat their analysis by noting that their accounting comparability measure uses only one financial statement summary measure, earnings (See page 899). Therefore, a natural modification of their measure would be to model other pieces of the accounting system and compare these models across firms. 16 I am grateful to Ben Whipple for suggesting the Dechow and Dichev (2002) model as another model that maps economic activity into accounting information. CAN I BORROW YOUR FOOTNOTES? 29 Δ!" !" is the change in working capital—or, working capital accruals—and !"! !" is cash flow from operations. While this model is almost exclusively used by the accounting literature to capture accrual quality (See Dechow et al., 2010 for a summary), Wysocki (2008) analytically examines the model and concludes it is “quite effective in explaining total working capital accruals” (p. 9, emphasis in original). I estimate this model for each firm i over 16 quarters. I compute the predicted working capital accruals using the cash flow variables from firm i using both firm i and firm j estimated parameters in a similar fashion as described above. That is, I calculate: Δ!" !!" = ! ! +! !! !"# !"!! +! !! !"# !" +! !! !"# !"!! (6) Δ!" !"# = ! ! +! !! !"# !"!! +! !! !"# !" +! !! !"# !"!! (7) After obtaining the predicted values, in a similar manner as above I calculate the average negative absolute difference between Δ!" !!" and Δ!" !"# . More specifically !""#$%&'%(!!"#$%&%'()(*! !"#,!! = !− 1 16 Δ!" !!" −Δ!" !"# ! !!!" (8) I calculate Accounting Comparability ijt,DD for all ij comparisons within the same two-digit SIC industry. As above, I measure the firm-year measureAccounting Comparability ijt,DD as the average of the four highest values of Accounting Comparability ijt,DD for firm i. Earnings comovement. De Franco et al., (2011) develop an additional measure of comparability, earnings comovement, as an alternative to their accounting system mapping measure. They explain that this measure does not require the researcher to model the accounting system, but rather compares the relation between outputs of the two firms’ accounting systems. Earnings Comovement ijt is the adjusted-R 2 of the following model: !"#$%$&! !" = θ !!" +θ !!" !"#$%$&' !" +! !"# (9) CAN I BORROW YOUR FOOTNOTES? 30 I calculate Earnings Comovement ijt across all ij comparisions within the same two-digit SIC industry and create a firm-year measure Earnings Comovement it as the average of the four highest values of Earnings Comovement ijt for firm i. Measuring EDGAR Search Volume To explore whether financial statement preparers use other firms financial statements to decrease footnote preparation cost by exploiting the learning externality, I examine the association between EDGAR search traffic and a measure of footnote similarity. I obtain data that measures EDGAR system traffic to 10-Ks, the form containing company’s footnotes. This data has been examined in prior studies to explorer how investors search for information (Drake et al., 2013). These studies have assumed that the traffic to these forms comes from investors. I note, however, that an additional set of EDGAR users are other financial statement preparers searching for examples of how other companies have disclosed information in the footnotes. To measure traffic to the 10-K, I calculate the average number of times a day a user downloads a 10-K over the year following the 10-K’s filing. I compare this measure to the footnote similarity of the footnotes in the 10-K to the footnotes in the 10-Ks filed during the following year. Specifically, I calculate the number of words in a firm’s footnotes that can be found in other companies’ footnotes in the two years following the filing of the 10-K to capture the extent language is borrowed from a firm footnotes by other companies. I take the average of the 10 most similar footnotes. This is different from the primary measure of my study in that I compare footnotes from one firm to footnotes of firms in the same SIC 2-digit industry filed after the firm’s 10-K is filed. Since I am interested in search volume related to financial statement preparers and not investors seeking information about the firm, I exclude the first week following the filing of the 10-K to avoid confounding search traffic due to investors searching CAN I BORROW YOUR FOOTNOTES? 31 for information about the firm rather than financial statement preparers searching for examples of footnotes. Measuring of Network Benefits To measure the market reaction to the filing of the 10-K, I calculate the cumulative stock return to a firm’s stock price over three days, starting on the day the 10-K is filed and adjust by subtracting the return to a size-decile matched portfolio. Since footnote similarity has no implications for whether the news is good or bad, I take the absolute value of this return and call this variable |CARi (10-K Filing i)|. To measure the amount of information transfer between two firms, I examine the stock price return of firm i around important information events of firm j. Specifically, I measure the stock retun of firm i when firm j files its 10-K for |CARi (10-K Filing j)| and when firm j announces earnings. To identify firm j’s 10-K filing and earnings announcement dates I search for the next event after when firm i files its 10-K. I measure |CARi (Earnings Announcement j)| and |CARi (10-K Filing j)| in a similar fashion for as |CARi (10-K Filing i)| adjusting the cumulative return around the three day window by subtracting the cumlative return to the a size decile matched portfolio. I use the I/B/E/S Statistical Summary database to obtain measures of analyst following and analyst annual earnings forecast properties. Specifically, I measure Analyst Following as the number of analysts following firm i during the month following the month firm i files its 10-K. Analyst Accuracy is the squared difference between the consensus earnings forecast and I/B/E/S actual earnings of the following year. Analyst Dispersion is the standard deviation of the forecasts. To avoid confounding my these measures of analyst forecast properties with other CAN I BORROW YOUR FOOTNOTES? 32 information sources, I measure Analyst Accuracy and Analyst Dispersion for the next fiscal year’s earnings during the month following the month the current fiscal year’s 10-K is filed. Measuring Economic Similarity To measure economic similarity of a firm to the firm that is identified to have the most similar footnotes, I follow Hoberg and Phillips (2010). Specifically, I extract the company’s product description in Item 1 of form 10-K, remove all stop words, and keep only the stems of the words. Using these stemmed words I form a vector of words used in both firms product descriptions. For each firm, I create word count vectors where the element of each vector is of the number of times the firm used the corresponding word in the product description. I then calculate the cosine of the angle formed by the two vectors and call this variable Cosine Word Count Item 1. The possible values of the variables ranges from 0 to 1. Since some concern exisits that measure of accounting comparability may only capture economic similarity, I include this measure in the models of accounting comparability. Footnote Similarity Descriptive Statistics Table 1, Panel A reports descriptive statistics for the measures of footnote similarity. The means for #Text Match ij and %Text Match ij are 677.9 and 0.097, respectively. This indicates that, on average, footnote set i and any footnote set j share close to 700 words in text strings at least six words in length; and, on average, about 10% of the text in footnote set i is also in any footnote set j. The means for #Text Match MAXij and %Text Match MAXij are 1,274.6 and 0.185 respectively, which, by construction, are higher than those of all ij comparisons sample. This indicates that, on average, footnote set i and i’s most similar comparison footnote set j share over to 1,250 words in text strings at least six words in length; and, on average, over 18% of the text CAN I BORROW YOUR FOOTNOTES? 33 in footnote set i is also in footnote set j. 17 The standard deviation and the interquartile range of these variables indicate there is significant variation in the footnote similarity measures. Although I present descriptive statistics for both word count and percentage text match variables, I use %Text Match ij and %Text Match MAXij for the majority of the analysis. I do this since length of footnote is an important factor in determining #Text Match ij . %Text Match ij accounts for this by scaling by the number of words in the footnote (#Words in Footnotes i ). Panel B of Table 1 presents the yearly averages for the similarity measures and footnote length. Over the sample period, there are obvious increasing trends in the size of footnotes, as well as footnote similarity measured by number of words. However, the percentage of footnote similarity decreases slightly. These trends indicate that while the amount of similar text increases over the period, the proportion of similar text relative to the size of the footnotes slightly decreases. Thus, although many argue that firm’s use of boilerplate is increasing, to the extent that my footnote similarly measure captures boilerplate, in relative terms, it has remained relatively constant or even decreased over the sample period. Table 1, Panel B also reveals that the number of comparisons per year decreases by close to one-half over the sample period. This can be explained by a decreasing number of unique firms and a decreasing number of firms in each industry. 18 17 While these measures of similarity may seem somewhat low, it is important to note that this captures the similarity between firm i and only one firm j. Managers that borrow footnote disclosure from one firm are likely to borrow from multiple firms. Thus, this is essentially a lower bound. 18 Other papers that present descriptive on yearly samples also show a similar decreasing trend in number of firms each year (e.g., Lehavy et al., 2011). CAN I BORROW YOUR FOOTNOTES? 34 Empirical Results Industry, Geographical Location, and Auditor Office Subsample Analysis I begin my analysis by examining how industry, geographic proximity and auditor office impact footnote similarity. To examine the impact of industry, geographic proximity and sharing an auditor from the same office on footnote similarity, I use three dummy variables Same SIC 4, Same MSA and Same Auditor Office. Same SIC 4 equals one when footnote set i and footnote set j are prepared by firms operating in the same SIC 4-digit industry, zero otherwise. Same MSA equals one when footnote set i and footnote set j are prepared by firms located in the same MSA, zero otherwise. Likewise, Same Auditor Office equals one when footnote set i and footnote set j are prepared by firms that share an external auditor from the same office, zero otherwise. Table 2, Panel A reports the mean of these variables as well as their interactions. For the sample with all ij comparisons, in 21.0%, 7.3%, and 1.2% of comparisons the two firms, operate in the same SIC 4-digit industry, are located in the same MSA, and share an auditor from the same office, respectively. In this sample, 2.3% of firm comparisons are of firms located in the same MSA and operating in the same SIC 4-digit industry. Examining the interaction between Same Auditor Office with Same MSA reveals 0.92% of comparisons compare firms that share an auditor from the same office and are located in the same MSA. Likewise, the interaction between Same Auditor Office with Same SIC 4 reveals 0.42% of comparisons compare firms that share an auditor from the same office and operate in the same SIC 4-digit industry. Only in 0.35% of the comparisons are between the two firms operating in the same SIC 4-digit industry, located in the same MSA and sharing an auditor from the same office. The bottom of Panel A indicates that in the highest ij comparison sample the proportions of ij comparisons that operate in the same SIC 4-digit industry, are located in the same MSA and/or CAN I BORROW YOUR FOOTNOTES? 35 share an auditor from the same office are higher. For this sample in 36.0% of comparisons the two firms operate in the same SIC 4-digit industry, in 20.3% of comparisons the two firms are located in the same MSA, and in 10.7% of the comparisons the two firms share an auditor from the same office. In this sample, 10.1% of firm comparisons are of firms located in the same MSA and operating in the same SIC 4-digit industry. Examining the interaction between Same Auditor Office with Same MSA reveals 8.5% of comparisons compare firms that share an auditor from the same office and are located in the same MSA. Likewise, the interaction between Same Auditor Office with Same SIC 4 reveals 5.2% of comparisons compare firms that share an auditor from the same office and operate in the same SIC 4-digit industry. In 4.3% of the comparisons the two firms operate in the same SIC 4-digit industry, are located in the same MSA and share an auditor from the same office. The proportions for Same SIC 4, Same MSA, and Same Auditor Office are between 1.7 and 9.1 times higher in this sample than in the all ij comparisons sample, the highest increase being for the Same Auditor Office variable. These larger proportions of are evidence industry, geographical proximity and the auditor play a role in engendering footnote similarity. The proportions for the interactions of the dummy variables are between 4.2 and 12.3 times higher in the highest ij comparisons sample than in the all ij comparisons sample. The interactions that include Same Auditor Office are never lower than 9.24 times higher. This that the auditor plays a role in footnote borrowing. Note, also, that of these three dummy variables the increase in Same SIC 4 is the smallest, thus indicating that while industry plays a role in footnote similarity, it is not the primary role. Industry, geographic proximity and auditor office subsample analysis of all ij comparison sample. Panel B of Table 2 compares subsamples of the data containing all ij comparisons to examine the impact of geographic proximity and auditor office on footnote CAN I BORROW YOUR FOOTNOTES? 36 similarity. The sets of subsamples I compare are: (1) Same SIC 4-digit industry and Different SIC 4-digit industry; (2) Same SIC 4 and Different SIC 4-digit industry, while requiring the two firms be audited by auditors from different offices and located in different MSAs; (3) Same MSA and Different MSA; (4) Same MSA and Different MSA, while requiring the two firms be audited by auditors from different offices and operating in different SIC 4-digit industries; (5) Same Auditor Office and Different Auditor Office; (6) Same Auditor Office and Different Auditor Office, while requiring the two firms be located in different MSAs and operating in different SIC 4-digit industries; and (7) Same SIC 4-digit industry, Same MSA, Same Auditor Office and Different SIC 4-digit industry, Different MSA, Different Auditor Office. Panel B reports that firms operating in the same SIC 4-digit industry, located in the same MSA and/or sharing the same auditor office have significantly higher levels of footnote similarity than two firms that do not. 19 These differences in means correspond to percentage increases in footnote similarity ranging from 7.5% (subsample (4), ((0.125 – 0.097)/0.097) to 40.8% (subsample (7), ((0.134 – 0.095)/0.095) when going from the “Different” subsamples to the “Same” subsamples. The largest increase is in subsample (7) which compares the presence of SIC 4-digit industry Same MSA and Same auditor office with the absence of all three. These results suggest industry, geographic proximity and sharing an auditor office are associated with greater footnote similarity. Industry, geographic proximity and auditor office subsample analysis of highest ij comparison sample. Table 2, Panel C performs a similar subsample analysis as is reported in Panel B, but on the highest ij comparison sample instead of on the all ij comparison sample. This analysis also finds that firms located in the same MSA and/or sharing the same auditor from the 19 Statistical significance refers to p-values < 0.05. All tests are two-sided. CAN I BORROW YOUR FOOTNOTES? 37 same office have higher levels of footnote similarity than firms that do not. In addition, the differences in means in this sample correspond to larger increases in footnote similarity than in the sample that consists of all ij comparisons, excluding subsamples (1) and (2) that contrast Same SIC 4 and Different SIC 4. Excluding these subsamples, the percentage increases in footnote similarity range from 9.67% (subsample (4), ((0.195 – 0.178)/0.178) to 37.8% (subsample (7), ((0.245 – 0.178)/0.178)) when going from the “Different” subsample to the “Same” subsample. Again, the largest increase is in subsample (7). These results also are consistent with geographic proximity and sharing and auditor office increasing firms’ footnote similarity. Comparing High Footnote Similarity Firms and Low Footnote Similarity Firms An alternative way of examining whether operating in the same industry, being geographic proximate and sharing an auditor from the same auditor office increases similarity is to compare the proportion of firms that operate in the same SIC 4-digit industry, are located in the same MSA and/or are audited by the same auditor office as the firm’s most similar comparison firm between a subsample with low levels of footnote similarity and a subsample with high levels of footnote similarity. Table 3 reports the results of this analysis. After sorting the sample into deciles based on %Text Match MAXij , Table 3 reports that the average value of %Text Match MAXij for the lowest similarity and highest similarity deciles are 0.091 and 0.316, respectively. The average for the low similarity decile is the lower than the overall average of %Text Match ij for all ij comparisons (see Table 1 Panel A). This indicates that, while I select the highest ij comparison for each firm, the firms in the lowest decile of %Text Match MAXij on average have a level of footnote similarity less than the overall average of all ij comparisons. This is evidence that those in the lowest decile do not engage in footnote borrowing. This also indicates that for firms in the CAN I BORROW YOUR FOOTNOTES? 38 highest decile, on average, over 31.6% of the text in their footnotes can be found in an industry peer firm’s footnotes. Table 3 also reports the proportion of firms located in the same MSA and/or audited by an auditor from the same auditor office in the lowest and highest %Text Match MAXij deciles. This analysis finds that firms in the highest decile of footnote similarity are 5.01 times more likely to be located in the same MSA as the firms in the lowest decile (0.420/0.084=5.014). This analysis also finds that firms in the highest decile are 12.18 times more likely to share an auditor from the same office than firms in the lowest decile (0.356/0.029=12.186). Further, firms in the highest decile are 16.67.1 times more likely to be located in the same MSA and share an auditor office as the most similar firm than firms in the lowest decile (0.150/0.009=16.674) This analysis presents evidence that firms with high levels of similarity are more likely to be located in the same MSA and/or share an auditor office than firms with low levels of similarity. Interestingly, firms in the highest decile of footnote similarity are only 1.1 times more likely to operate in the same industry as firms in the lowest decile (0.414/0.376=1.103). The bottom of Table 3 reports the average Market Value for the “borrowing” (the i firms) and the “lending” firms (the j firms), and finds that the borrowing firms are smaller than the lending firms, on average. This is consistent with relatively smaller firms borrowing footnote disclosures from relatively larger firms, on average. However, while this result holds in the largest decile of %Text Match MAXij , it does not hold in the lowest decile. In addition, the average Market Value for firm j in the highest decile is below the mean market value for the sample. This indicates that small firms do not borrow from the very largest firms, but rather from the relatively larger firms. This is consistent with managers borrowing from relatively larger peer firms that are likely to be more similar to the smaller firms than the largest firms. Borrowing CAN I BORROW YOUR FOOTNOTES? 39 from larger firms is consistent with larger firms being more likely to have a footnote that the smaller firm needs, and with managers perceiving that larger firms have higher reporting quality. Managers may seek higher quality guidance in hopes that it will result in higher quality disclosure. Regression Model Next, I examine the impact of geographic proximity and auditor office on footnote similarity in a multivariate regression. I model footnote similarity as a function of industry, geographic proximity, auditor office, firm complexity and other financial characteristics. Table 4 Panel A reports the descriptive statistics for the sample used in the regression. The sample size drops to 20,358 observations due to data restrictions. The main reason for the drop in sample size is that the Audit Analytics data is available for the sample period 2001 to 2010. For comparison purposes, Panel A also lists the mean for each variable of all firms in the Compustat universe. Relative to the Compustat universe, the firms in my sample are larger, older, have more business and geographical segments, lower net income, higher R&D expense, and higher goodwill. However, the firms in my sample are similar to the Compustat sample in terms of leverage, working capital, and sales growth. Table 4 Panel B reports the correlations between the measures of footnote similarity, the geographical location and auditor office variables, and the control variables. 20 This analysis shows that the correlations among most of the independent variables are relatively modest. I model footnote similarity by estimating the following regression model: 20 Since some of the correlations between the independent variables are relatively high, I examine traditional multicollinearity diagnostics (variance inflation factor (VIF) or tolerance, and condition index) on the regressions in Table 4, Panel C and Table 5, Panel B. I find no evidence of multicollinearity. CAN I BORROW YOUR FOOTNOTES? 40 !""#$"#%!!"#"$%&"'(=!+! !"#$%&'(,!"#,&!!"#$%&'!!""#$%! + ! !"#$!!"#$%&'()*! +! !"#$#%"$&!!ℎ!"!#$%"&'$'#& +! (10) I use two different Footnote Similarity dependent variables, the natural log of #Text Match MAXij and %Text Match MAXij . In the regression with ln(#Text Match MAXij ) as the dependent variable, I also include #Words in Footnotes i . The other dependent variable, %Text Match MAXij , is scaled by #Words in Footnotes i . The Industry, MSA & Auditor Office variables are Same SIC 4, Same MSA, Same Auditor Office, and the interaction between these two variables. Same MSA and Same Auditor Office are as defined above. I expect the coefficients on these variables to be positive; consistent with industry, geographic proximity and auditor office contributing to footnote similarity. The Firm Complexity variables are the natural log of Market Value, Age, Business Segments, and Geographical Segments. I expected the coefficients on these variables to be negative since the underlying economics of more complex firms are more unique, thus decreasing the probability a manager can find external guidance to assist in preparing the footnotes. The Financial Characteristics variables are Return-on-Assets, Leverage, R&D over Operating Expense, Working Capital over Assets, Sales Growth, Book-to-Market, and Big N Auditor. I include these variables to control for other firm characteristics that are associated with footnote similarity. These models also include year and industry fixed effects. Table 4 Panel C reports the results of estimating these regressions. The coefficients on Same SIC 4, Same MSA and Same Auditor Office are positive and statistically significant, as predicted. 21 The coefficient on Same SIC 4 in Model (1) is 0.005 and in Model (2) is 0.033. This indicates that being in the same SIC 4-digit industry results in a 2.5% to 3.3% increase in footnote similarity. In Model (1), the impact can be assessed by dividing the coefficient by the 21 All t-statistics reported in regressions are calculated using standard errors clustered at the firm level. CAN I BORROW YOUR FOOTNOTES? 41 intercept (0.005/0.198=0.025). Since Model (2) is a log-linear model, the coefficient 0.033 can be interpreted as a percent change, or 3.3%. The coefficient on Same MSA in Model (1) is 0.010 and in Model (2) is 0.065. Interpreting these coefficients indicates that being in the same MSA results in a 5.1% to 6.5% increase in footnote similarity (0.010/0.198=0.051). . The coefficient on Same Auditor Office in Model (1) is 0.042 and in Model (2) is 0.188. Interpreting these coefficients indicates that sharing an auditor from the same office results in a 18.8% to 21.2% increase in footnote similarity (0.042/0.198=0.156). The interaction between Same Auditor Office and Same MSA is negative and significantly different from zero. Although this interaction term is negative and significant, the joint effect of Same Auditor Office and Same MSA remains positive and significant (0.042 + 0.010 - 0.014 = 0.038 p-value < 0.001). The coefficients on the three other interaction terms are not significantly different from zero. Overall, the results from these regressions corroborate the prior tests and are consistent with geographic proximity and auditor office contributing to footnote similarity. The coefficients on the Firm Complexity variables are all negative and significantly different from zero, consistent with firm complexity decreasing footnote similarity. In both regressions, five of the nine Financial Characteristics control variables are significant. The R 2 for Model (1) is 0.371 and for Model (2) is 0.547. The reason why the R 2 for Model (2) is higher is the inclusion of ln(#Words in Footnotes i ) in Model (2). As noted above, instead of including this variable as an independent variable, in Model (1) the dependent variable is scaled by #Words in Footnotes i . EDGAR Search Volume Analysis To examine whether financial statement preparers use SEC EDGAR to obtain previously used footnote language, I compare the daily average of the number of times a 10-K on EDGAR CAN I BORROW YOUR FOOTNOTES? 42 is downloaded with the extent to which the footnote language on the 10-Ks can be found in the footnotes on 10-Ks filed by industry peers during the following two years. I find that the extent to which language in a firm’s footnotes appears in the future footnotes of industry peers is positively correlated with EDGAR traffic to the 10-K containing the footnotes. Specifically, I regress ln(Average Daily EDGAR Search Volume) on mean(#Text Match j,i Top 10 ) and control variables. Results in Table 5 Panel B report the coefficient on mean(#Text Match j,i Top 10 ) is positive and significantly different from zero. This is consistent with financial statement preparers using EDGAR to search for footnote language to borrow. Accounting Comparability Analysis I examine the relation between three measures of accounting comparability and my measure of footnote similarity. The second set of analyses examine the relation between accounting comparability and footnote similarity by employing three measures of comparability Accounting Comparability i,DKV , Accounting Comparability i,DD and Earnings Comovement i . Due to data restrictions, the sample size drops to 13,114, 7,883 and 13,998 observations when using these measures of comparability, respectively. The correlations between Accounting Comparability i,DKV , and Accounting Comparability i,DD is 0.431 (p-value < 0.0001). This high correlation supports the convergent validity of these two measures of accounting comparability. The correlation between Earnings Comovement i and Accounting Comparability i,DKV is close to 0.008 (p-value = 0.315), and between Earnings Comovement i and Accounting Comparability i,DD is 0.0242 (p-value = 0.032) These low correlations question the convergent validity of Earnings Comovement i and the other two measures. Therefore, the Accounting Comparability i measures and Earnings Comovement i likely capture unrelated facets of accounting comparability. CAN I BORROW YOUR FOOTNOTES? 43 Most prior research using Accounting Comparability i,DKV to measure accounting comparability examines the benefits of comparability. For example, prior research shows this measure of comparability is associated with analyst following and forecast accuracy (De Franco et al. 2011), mediates the negative price reaction around a peer firm’s restatement announcement (Campbell and Yeung 2012), 22 and is related to the efficiency of acquisitions (Chen et al. 2012). Little if any research examines the factors within mangers’ control that impact comparability. I examine how footnote similarity, an outcome of mangers’ decisions, is related to accounting comparability. I first examine the overall relation between footnote similarity and accounting comparability. I then examine this relation with respect to factors that help explain comparability, specifically, geographic proximity and the external auditor. Table 5, Panel A provides descriptive statistics for the variables used in these regressions. To examine the overall relation between footnote similarity and accounting comparability, I estimate the following model: !""#$%&'%(!!"#$%&%'()(*! !,!!! = !+! ! %!"#$!!"#$ℎ !"#$%,! +! !"#$%"&' + !! (11) I measure the dependent variable at the end of the fiscal period following the period that corresponds to the footnotes since %Text Match MAXij can only be measured once the company has filed it is financial statements. The filing usually takes place within a couple of months after the end of the fiscal year. Table 7 reports the results from estimating equation (6) in Model (1), Model (2) and Model (3) for the three measures of comparability. This analysis find a positive and statistically significant association between two measures of accounting comparability and 22 Campbell and Yeung (2012) use an alternative specification of the De Franco et al. (2011) model. Campbell and Yeung (2012) use the Basu (1997) model that also includes a dummy variable for losses and the interaction between the loss dummy variable and quarterly return. Apart from this difference, the measure used by Campbell and Yeung (2012) is similar to the measure I use. CAN I BORROW YOUR FOOTNOTES? 44 %Text Match MAXij (for Accounting Comparability i,DKV , β 1 = 1.925, p-value. < 0.0001; for Accounting Comparability i,DD , β 1 = 0.742, p-value. < 0.001). The association between Earnings Comovement i is significant at the 10% level (Earnings Comovement i , β 1 = 0.060, p-value. = 0.061). This is evidence that footnote similarity enhances comparability, as predicted. Network Benefits Next, I examine whether footnote similarity is correlated with outcome measures that capture whether financial statements users benefit from the network externality created by the increased comparability of footnotes with more similar language. To do this I examine three variables for investors and three variables for analysts. I examine firm i’s stock price reaction to firm i’s 10-K filing. I also measure the information transferred between firm i and its industry peer firm with the most similar footnotes by examining firm i's stock price reaction when firm j files its 10-K and when firm j announces earning. Both events are the first occurrence after the date firm i has filed its 10-K. For analysts, I examine analyst following as well as the analyst earnings forecast properties dispersion and accuracy. Market Reaction to 10-K filing. Table 8 reports the unsigned market reaction of firm i’s stock when firm i filing its 10-K |CARi (10-K Filing i)|. I split the sample into five quintiles based on %Text Match MAXi,j . I find that the mean of |CARi (10-K Filing i)| is increasing in %Text Match MAXi,j . Specifically, the mean value of |CARi (10-K Filing i)| for firms in quintile two is greater than the mean value of |CARi (10-K Filing i)| for firms in quintile one (difference = 0.0054, p-value < 0.0001). Talbe 8 reports similar difference between quintiles three and two, four and three, and five and four. The difference between quintile five and one is also positive and significant (difference = 0.0164, p-value < 0.0001). This is evidence that investors find financial statements with greater footnote similarity more informative. CAN I BORROW YOUR FOOTNOTES? 45 Information transfer analysis. Accounting information that is more comparable allows investors to use information from other firms to inform decision making. That is, information about firm j informs investors’ decisions about firm i. Prior research examines information transfer around information events such as earnings announcements (e.g., Foster 1981, and Ramnath 2002) restatements (e.g., Gleason, Jenkins, and Johnson 2008), and stock splits (e.g., Tawatnuntachai and D’Mello 2002). I measure information transfer as the unsigned, market- model adjusted, cumulative abnormal return to firm i’s equity stock when firm j experiences an information event. I measure the cumulative stock return over the four-day window, minus one to plus two. The two events I employ are (1) firm j’s first earnings announcement and (2) firm j’s first 10-K filing following firm i’s filing date. For each of these measures of information transfer, I regress the unsigned return on %Text Match MAXij and controls. Table 9 reports the findings of this analysis. Panel A reports comparing the stock price reactions for quintiles formed on %Text Match MAXi,j . Panel B examines the association in a multivariate regression. Overall, I find a positive association between footnote similarity and these two measures of information transfer. This is additional evidence that the use of footnote boilerplate create network externality that benefit financial statement users. Equity analyst analysis. Following prior research, I determine whether an attribute of the disclosure in the 10-K impacts equity analysts, I regress the analyst variable on my measure of footnote similarity and control variables. Table 10 presents the results of the multivariate regressions. In Model 1, I regress Analyst Following on my measure of footnote similarity %Text Match MAXi,j and control variables used in prior research. I find a positive relation between analyst following and footnote similarity, after controlling for other factors that also impact an analyst’s decision to follow a firm. This is indicative that when firms use more similar language in CAN I BORROW YOUR FOOTNOTES? 46 footnotes analysts are more likely to follow the company. This evidence confirms my expectations footnote similarity decreases information processing costs of analysts as well as findings in a recent survey of equity analysts (Lawrence et al. 2013). The survey respondents indicated an important factor in their decision to follow a company is how similar a company is to other companies. Model 2 and 3 in Table 10 present results of regressing Analyst Accuracy and Analyst Dispersion on %Text Match MAXi,j and the standard control variables. In both models the coefficient on footnote similarity is negative. This negative coefficient indicates that as firms use more similar footnotes, analyst accuracy improves and dispersion increases. The analysis of the impact of footnote similarity on outcomes of financial statement users confirms my expectations that the use of similar language enhances financial reporting quality by decreasing information processing costs and improving the set of information available to the user. This is interesting give the conjectures of standard setters and regulators that boilerplate in footnote impairs reporting quality. Additional Analysis Auditor Changes In an attempt to corroborate my findings regarding the association between footnote similarity and same auditor office, I examine whether footnote similarity changes following auditor changes. When firms change auditors, the auditor expertise available to the firm to assist in footnote preparation also changes. To test whether this change in expertise impacts footnote similarity, I test whether a changing firm’s footnotes become more similar to the new auditor’s clients’ footnotes, and less similar to the predecessor auditor’s clients’ footnotes. CAN I BORROW YOUR FOOTNOTES? 47 For every auditor change, I measure the level of similarity between the changing firm’s footnotes and the predecessor and new auditors’ clients’ footnotes in the pre-change and post- change periods. That is, I measure the changing firm’s footnote similarity with respect to four groups of firms; (1) the predecessor auditor’s clients in the pre-change period, (2) the predecessor auditor’s clients in the post-change period, (3) the new auditor’s clients in the pre- change period, and (4) the new auditor’s clients in the post-change period. For an auditor change to be included in this analysis, I require footnote similarity measures for all four groups. Within each of the four groups, I define footnote similarity %Text Match MAXij(auditor change) as the similarity score between firm i and the auditor’s client with the most similar footnotes. To examine the impact of the change on footnote similarity, I estimate the following difference-in-differences model: %!"#$!!"#$ℎ! !"#$%(!"#$%&'!!!!"#$) =!+! ! !"#$!!ℎ!"#$+! ! !"#!!"#$%&' + !! ! !"#$!!ℎ!"#$×!"#!!"#$%&' +! !"#$%"&' + !!, (12) where Post Change is an indicator variable that equals one if the footnote similarity measure corresponds to a group in the post-change period, and where New Auditor is an indicator variable that equals one if the footnote similarity measure corresponds to the new auditor in either the pre-change or post-change periods. I include as Controls the Firm Complexity and Financial Characteristics variables from the footnote similarity regression shown in Table 4. A positive β 3 coefficient indicates that the footnotes of a firm changing auditors become more similar to the footnote of the new auditor’s client and less similar to the footnotes of the predecessor auditor’s clients. CAN I BORROW YOUR FOOTNOTES? 48 Table 11 Panel A reports the descriptive statistics of the variables for the sample used in this analysis and Panel B reports the estimation of the difference-in-differences model. The sample includes 1,220 auditor changes. I find the footnotes of a firm changing auditors become more similar to the new auditor’s clients and less similar to the predecessor auditor’s clients. The estimate for β 3 is 0.015. Given that the level of footnote similarity between the changing firm and the clients of the new auditor during the pre-change is 0.176 (untabulated), this coefficient indicates that when a firm changes auditors its footnotes become 8.5% more similar to the new auditor’s clients’ footnotes (0.015/0.176=0.085). Although the magnitude may seem small, finding any effect is surprising given that auditor participation in footnote preparation may impair auditor independence. This evidence corroborates the other analysis that shows sharing an auditor from the same office increases footnote similarity. Auditor changes: high and low pre-change footnote similarity deciles. I examine whether firms that rely on auditor guidance prior to an audit change are likely to continue to rely on the auditor for guidance after the change. The pre-change level of footnote similarity provides a measure of a firm reliance on outside sources for guidance in footnote preparation. I expect the impact of an auditor change on footnote similarity is greater for firms with higher levels of pre- change footnote similarity. To examine whether pre-change reliance on the auditor impacts the effect of an auditor change on footnote similarity, I split the auditor change sample into subsamples based pre- change levels of footnote similarity. Specifically, I identify the pre-change footnote similarity decile for each changing firm. I re-estimate the difference-in-differences regression for the highest and lowest deciles. Table 11 Panel C reports the results. I find the effect of an auditor change on footnote similarity is over three times higher for firms in the highest decile of footnote CAN I BORROW YOUR FOOTNOTES? 49 similarity when compared effect in the whole sample (0.045/0.015 = 3.00). For the highest decile sample, prior to the auditor change, the level of similarity between the changing firm and the new auditor’s clients is 0.273 (untabulated). Thus, the estimated difference-in-differences coefficient, 0.045, indicates that when a firm changes auditors its footnotes become 16.5% more similar to the new auditor’s clients (0.045/0.273=0.165). The estimated difference-in-differences coefficient for the lowest footnote similarity decile is not significantly different from zero (β 3 = 0.003, t-stat. = 0.31). This evidence is consistent with an auditor change only impacting footnote similarity when a firm relies on outside sources for guidance in footnote preparation. This also indicates if a firm did not rely on the predecessor auditor’s assistance in the pre-change period, the firm does not rely on the new auditor in the post-change period. Auditor change: big N and non-big N changes. The expertise and experience of the auditor may also influence the extent to which a manager relies on the auditor assistance in footnote preparation. Big N auditor offices typically service more and larger clients than non-Big N auditor offices; therefore, the amount of auditor expertise and experience a manager can draw on is greater when the firm’s auditor is a Big N auditor. To examine the impact of auditor expertise and experience on a firm’s reliance on an auditor, I split the sample into changes from one Big N auditor to another Big N auditor and from one non-Big N auditor to another non-Big N auditor. I exclude the Big N Auditor control variable from this regression since in each sample this variable is constant. Table 11, Panel C reports these results. I find the effect of an auditor change on footnote similarity occurs only in the Big N auditor sample. These results provide evidence that firms rely on the auditor when the auditor expertise and experience is high. Matching analysis examining relation between footnote similarity and accounting comparability. The main analysis examines the relation between footnote similarity and CAN I BORROW YOUR FOOTNOTES? 50 accounting comparability using multivariate regression to adjust for differences in covariates. However, using multivariate regression to adjust for these differences may lead biased estimates. 23 Specifically, the estimated effect is potentially sensitive to the functional form assumptions of the regression model. Also, the estimation errors may result from insufficient overlap between the treatment and control samples. The methodological literature offers matching methods as a means of circumventing these problems and enhancing the credibility of causal inferences. Although theories of causal inference clearly identify the importance of obtaining covariate balance—common support, similar means, and densities across treatment and control samples— between matched treated and control samples prior to estimating treatment effects (Rubin, 2008), the methodological literature does not agree on a method of preprocessing data to achieve this balance. Therefore, I implement three data preprocessing methods: (1) propensity score matching, (2) genetic matching, and (3) entropy balancing (Rosenbaum and Rubin 1983; Diamond and Sekhon, forthcoming; and Hainmueller, 2012). The first method is now common place in accounting research. To my knowledge, however, the last two methods not been used in the accounting literature. Appendix C discusses these methods in greater detail, as well as why genetic matching and entropy balancing offer improvements over propensity score matching. Pre-Matching covariate balance. Matching methods require a binary treatment indicator. Thus, I construct a treatment indicator Treated it which equals one if the observation’s value of %Text Match MAXij is above the annual median %Text Match MAXij , zero otherwise. 24 I 23 See the Imbens (2010) and the Spring, 2010 edition of the Journal of Economic Perspectives. Also, Armstrong et al., 2010, in Appendix B, provide a helpful discussion of the problems that can arise from functional form misspecification. 24 Although this allows for the creation of “treatment” and “control” groups, I acknowledge that splitting the sample based on the yearly median level of %Text Match_MAXi,j is perhaps the simplest way to convert a continuous CAN I BORROW YOUR FOOTNOTES? 51 classify all observations with Treated it equal to one to the treatment sample, and the rest to the control sample. Table 8, Panel A provides descriptive statistics of two samples both prior to matching and after implementing each of the matching methods. I compare the covariate distributions for these two groups by calculating standardized differences and ratio of the variance of the treated sample over the variance of the control sample. Also, I test for differences in means via t-tests, and differences in densities via bootstrapped-Kolmogorov-Smirnov (KS) tests. Testing for difference in means and difference in covariate distributions reveals that, prior to any matching, the two samples differ wildly on every covariate. In implementing these matching methods, I examine the three main measures of comparability. Due to data restrictions, the sample size reduces to 9,213, 5,584, and 9,858 when I examine Accounting Comparability i,DKV , Accounting Comparability i,DD and Earnings Comovement i , respectively. When matching, I identify a matched control sample that is similar to the treated sample. That is, since I am only constructing a matched control sample, I am estimating the average treatment effect on the treated (ATT), or the effect above median footnote similarity on accounting comparability relative to below median. Propensity score matching. I implement propensity score matching by first specifying the propensity score model. This model is similar to equation (10) except instead of using %Text Match MAXij as the dependent variable I use Treated it . I then estimate the model on the three samples that correspond to the three accounting comparability measures. Also, instead of using ordinary least squares to estimate the model, I use probit regression. The results from estimating treatment variable to a dichotomous treatment variable and may not result in homogeneous treatment levels within each group. Due to this, the magnitude of the estimated difference in outcomes is not easily interpretable. CAN I BORROW YOUR FOOTNOTES? 52 this model over the three samples are reported in Table 12 Panel B. 25 These models exhibit Pseudo-R 2 between 0.432 and 0.547. I calculate the predicted probabilities from this model and use them as the propensity scores. To obtain a matched sample, I use a local optimal algorithm to implement one-to-one caliper matching with replacement. 26 I select a caliper of 0.01. 27 Genetic matching. The purpose of genetic matching is to minimize covariate imbalance (Diamond and Sekhon, forthcoming). Genetic matching does this by using an evolutionary search algorithm to identify a version of the Generalized Mahalanobis Distance (GMD) metric that, when used to match control observations to treated observations, maximizes covariate balance. Genetic matching is a generalization of Mahalanobis distance matching and propensity score matching in that the GMD metric accommodates including in the metric any number of covariates as well as estimated propensity score. However, to limit the complexity of the optimization task, I include the propensity score and the independent variables from the propensity score model, but exclude the industry and year fixed effect indicator variables. I implement genetic matching by using the “Matching” package in the statistical software application R (Sekhon, 2011). 28 See Appendix C for expanded discussion of genetic matching. Entropy balancing. Entropy balancing approaches the task of achieving covariate balance by directly focusing on identifying weights for every observation in the control sample rather than by matching on a distance metric to identify these weights, as in propensity score matching and genetic matching (Hainmuller, 2012). Covariate balance is achieved as these 25 Note that the signs and significance level tests are not the focus of propensity score models. Rather, researchers assess the fit of the model by examining the extent to which matching on the propensity score achieves covariate balance. 26 Coca-Perraillon (2007) provides a SAS Macro that implements this algorithm. 27 This caliper size results in significant improvements in the covariate balance without discarding many observations due to lack of a match. 28 See http://cran.r-project.org/web/packages/Matching/index.html for access to this package. CAN I BORROW YOUR FOOTNOTES? 53 weights are selected to meet prespecified balance conditions. Entropy balancing uses a maximum entropy reweighting scheme to identify these weights. This results in the treatment sample and the matched control sample having exactly the same moments (e.g., mean, variance, skewness, etc.) that are specified in the balance conditions. This obviates the need for iteratively specifying either a propensity score model or selecting a version of the GMD metric and checking covariate balance. I implement entropy balancing using the R package “ebalance.” 29 See Appendix C for expanded discussion of entropy balancing. Post-matching covariate balance. Panel A of Table 12 shows the covariate balance between treated and matched control samples after applying propensity score matching, genetic matching and entropy balancing. These descriptive statistics reveal that covariate balance improves after applying any of these three data preprocessing methods. For example, the standardized difference in means for ln(Market Value) between the treated and control samples improves from -66.44 to -1.2, -7.3915 and -0.0001 for propensity score matching, genetic matching and entropy balancing, respectively. 30 Tests of differences in distributions of the continuous covariates reveal similar improvements. For example, the variance ratio of the treated and control samples of the variable ln(Market Value) improves from 0.763 to 1.051, 1.132 and 1.0792 for propensity score matching genetic matching, and entropy balancing respectively. While all three methods improve covariate balance over the unmatched sample, Table 12 Panel A shows that the treated sample and the control sample generated by entropy balancing appears to be the most balanced. 29 See http://www.mit.edu/~jhainm/ebalancepage.html for ebalance package installation instructions. 30 The genetic matching results reported in table 8 reflect applying genetic matching with a generation population size of 1000. CAN I BORROW YOUR FOOTNOTES? 54 Table 12, Panel C and Panel D report the results of univariate and multivariate tests that reexamine the relation between footnote similarity and the three main measures of accounting comparability after implementing these three matching techniques. Panel C reports that in eight of the nine univariate tests, greater footnote similarity results in significantly higher levels of accounting comparability. The multivariate model used in these tests is equation (11). Panel E reports that in all nine multivariate tests footnote similarity results in significantly higher accounting comparability. The results from examining the relation between footnote similarity and accounting comparability after implementing propensity score matching, genetic matching and entropy balancing provide evidence consistent with the results of the main analysis. That is, footnote similarity enhances accounting comparability. Sensitivity Tests Exclude Loss Firms My sample contains a slightly larger proportion of loss firms than the Compustat universe. Therefore, to ensure my results are not driven by these loss firms, I rerun all my analysis after dropping firms with negative Return-on-Assets and find similar results. Big N Audited Companies To examine whether the results reported above are driven by firms with low reporting quality or low auditing quality, I repeat the analysis only on companies that were audited by Big N auditors. On this subsample, I find similar results. Arthur Andersen Auditor Changes Auditor changes result from a complex process involving various factors. Therefore, one potential problem with examining voluntary auditor changes is that any shift in footnote CAN I BORROW YOUR FOOTNOTES? 55 similarity could be due to factors that induced the change rather than the new auditor. Prior research uses the Arthur Andersen indictment and downfall as an exogenous event to provide evidence on mandatory auditor rotations (Blouin, Grein, and Rountree, 2007). I use this setting to examine how a forced, or exogenous, auditor change impacts footnote similarity. Since I cannot observe the level of post-change similarity between the changing firm and firms audited by Arthur Andersen, I compare footnote similarity between firms audited by the new auditor office prior to the change, and firms audited by the new auditor after the change. I find consistent results. The footnotes of switching firms become more similar to footnotes of firms audited by new auditor. Interestingly, I find that this result is not dependent on the pre-change level of footnote similarity. This is further evidence that corroborates the auditor office playing a role in footnote preparation. Conclusion Regulators, standard setters, and the profession express concerns that footnote disclosure contains excessive boilerplate, where boilerplate refers to standardized text that is similar across firms; and that this boilerplate leads to footnote disclosure that fails to clearly communicate information investors need. The FASB’s conceptual framework identifies comparability as an enhancing qualitative characteristic of accounting information, and that for information to be comparable, “like things must look alike and different things must look different” (FASB, 2010). Boilerplate text that increases similarity may either enhance comparability by making like things look alike, or it may impair comparability by making different things look different. Thus, the purpose of this study is to test whether footnote similarity enhances or impairs accounting comparability. CAN I BORROW YOUR FOOTNOTES? 56 This study models factors that explain footnote similarity and finds that footnote similarity is positively correlated with geographic proximity, and auditor office; and negatively correlated with firm size, and complexity. The study then examines the relation between footnote similarity and accounting comparability and finds they are positively associated. Thus, despite the criticism of boilerplate language in footnotes, which is likely correlated with my measure of footnote similarity, I find evidence that boilerplate enhances rather than impairs comparability. I also find that footnote similarity and accounting comparability are negatively related when similarity results from auditor involvement in footnote preparation. CAN I BORROW YOUR FOOTNOTES? 57 References Ahdieh, R. B., 2006. The strategy of boilerplate. Michigan Law Review. 104, 1033-1074. American Institute of Certified Professional Accountants (AICPA), 2012. U.S. GAAP Financial Statements Presentation and Disclosure. American Institute of Certified Professional Accountants. Armstrong, C. S., Jagolinzer, A. D., Larcker, D. F., 2010. Chief executive officer equity incentives and accounting irregularities. Journal of Accounting Research. 48, 225-271 Basu, S., 1997. The conservatism principle and the asymmetric timeliness of earnings. Journal of Accounting and Economics. 24, 3-37. Blouin, J., Grein, B., Rountree, B., 2007. An Analysis of Forced Auditor Change: The Case of Former Arthur Andersen Clients . The Accounting Review 82, 621-650. Brown, S., Tucker, J., 2011. Large-sample Evidence on Firms’ Year-over-Year MD&A Modifications. Journal of Accounting Research 49, 309-346. Burke, S.L. Letter to Patrick F. Gannon. March 4, 2002. Office of the Chief Accountant: Letter to Florida Institute of Certified Public Accountants Regarding Bookkeeping. Available at: <http://www.sec.gov/info/accountants/noaction/ficpa030402.htm> Campbell, J., Yeung, E., 2012. Accounting Comparability, Investor Sophistication, and Contagion Effects. Working Paper, University of Georgia and Cornell University. Carnall, W., Jacobs, S., Mew, A., Milne, R., Bhandari, B.K., Thompson, J.,…Woody, K., 2010. SEC Staff Review of Common Financial Reporting Issues Facing Smaller Issuers. Securities and Exchange Commission. Available at: <http://www.sec.gov/news/speech/2012/spch020912co.pdf> Chen, C., Collins, D., Kravet, T., Mergenthaler, R., 2012. Financial Statement Comparability and the Efficiency of Acquisition Decisions. Working Paper, University of Iowa, and University of Texas at Dallas. Coca-Perraillon, M. (2007). Local and global optimal propensity score matching. In SAS Global Forum 2007. Paper 185-2007. Available at: <http://www2.sas.com/proceedings/forum2007/185-2007.pdf> Cohn, M., 2012. FASB Readies Framework for Financial Statement Disclosures. Accounting Today: March 29, 2012. CAN I BORROW YOUR FOOTNOTES? 58 Committee on Improvements to Financial Reporting (CIFR), 2008. Final Report of the Advisory Committee on Improvements to Financial Reporting to the United States Securities and Exchange Commission. Available at <http://www.sec.gov/about/offices/oca/acifr/acifr- finalreport.pdf> De Franco, G., Kothari, S.P., Verdi, R., 2011. The Benefits of Financial Statement Comparability. Journal of Accounting Research 49, 895-931. Dechow, P., Dichev, I., 2002. The quality of accruals and earnings: the role of accrual estimation errors. The Accounting Review 77, 35–59. Dechow, P., Ge, W., Schrand, C., 2010. Understanding earnings quality: A review of the proxies, their determinants and their consequences. Journal of Accounting and Economics 50: 344- 401. DeFond, M., Xuesong, H., Hung, M., Li, S., 2011. The impact of mandatory IFRS adoption on foreign mutual fund ownership: The role of comparability. Journal of Accounting and Economics 51, 240-258. Drake, M. S., Roulstone, D. T., Thornock, J. R., 2013. The determinants and consequences of information acquisition via EDGAR. Working Paper, Brigham Young University. Deloitte. 2010. Swimming in words, surveying narrative reporting in annual reports. Available at <http://www.deloitte.com/assets/Dcom- UnitedKingdom/Local%20Assets/Documents/Services/Audit/Corporate%20Governance/UK _Audit_Swimming_in_words.pdf> Diamond, A., Sekhon, J. S., Forthcoming. Genetic matching for estimating causal effects: A general multivariate matching method for achieving balance in observational studies. Review of Economics and Statistics. European Financial Reporting Advisory Group (EFRAG), Autorité des Normes Comptables (ANC), and Financial Reporting Council (FRC), 2012. Towards a Disclosure Framework for the Notes. Available at: <http://www.efrag.org/files/ProjectDocuments/PAAinE%20Disclosure%20Framework/1210 15_Disclosure_Framework_-_FINAL1.pdf> Fama, E., French, K., 1997. Industry costs of equity. Journal of Financial Economics 43, 153- 193. Financial Accounting Standards Board (FASB), 2010. Conceptual Framework. Statement of Financial Accounting Concepts No. 8. Financial Accounting Standards Board. CAN I BORROW YOUR FOOTNOTES? 59 Financial Accounting Standards Board (FASB), 2012. Discussion Paper. Disclosure Framework. Financial Accounting Standards Board. Available at: <http://www.fasb.org/cs/ContentServer?site=FASB&c=Document_C&pagename=FASB%2 FDocument_C%2FDocumentPage&cid=1176160160107> Financial Accounting Standards Board (FASB), 2013. FASAC issues results of 2013 FASB stakeholder survey, disclosure framework, codification improvement projects cited as priorties. [News release]. Retrieved from http://www.fasb.org/cs/ContentServer?c=FASBContent_C&pagename=FASB%2FFASBCo ntent_C%2FNewsPage&cid=1176163332716 Financial Executives Research Foundation (FERF), KPMG, 2011. Disclosure Overload and Complexity: Hidden in Plain Sight. Available at <http://www.kpmg.com/US/en/IssuesAndInsights/ArticlesPublications/Documents/disclosur e-overload-complexity.pdf> Financial Reporting Council (FRC), 2009. Louder than Words, Principles and actions for making corporate reports less complex and more relevant. Available at <http://www.frc.org.uk/FRC- Documents/FRC/Discussion-Paper-Louder-than-Words.aspx > Financial Reporting Council (FRC), 2011. Cutting Clutter, Combating clutter in annual reports. Available at < http://www.frc.org.uk/getattachment/4e747c33-cc31-469b-9173- a07a3d8f0076/Thinking-about-disclosures-in-a-broader-context-A-road-map-for-a- disclosure-framework.aspx > Financial Reporting Council (FRC), 2012. Thinking about disclosures in a broader context, A road map for a disclosure framework. Available at: <http://www.frc.org.uk/getattachment /4e747c33-cc31-469b-9173-a07a3d8f0076/Thinking-about-disclosures-in-a-broader-context- A-road-map-for-a-disclosure-framework.aspx > Foster, G. 1981. Intra-industry information transfers associated with earnings releases. Journal of Accounting and Economics 3: 201–232. Gleason, C., Jenkins, N., Johnson, W., 2008. The contagion effects of accounting restatements. The Accounting Review 83: 83-110. Gulati, M., Scott, R. E., 2011. The three and a half minute transaction: Boilerplate and the limits of contract design: forward. Hofstra Law Review. 40, 1-9. Gunning, R., 1952. The technique of clear writing. New York, NY: McGraw-Hill International Book Co. Hainmueller, J., 2012. Entropy balancing: A multivariate reweighting method to produce balanced samples in observational studies. Political Analysis 20, 25-46. CAN I BORROW YOUR FOOTNOTES? 60 Hainmueller, J., Xu, Y., 2012 Ebalance: A Stata package for entropy balancing. MIT Political Science Department Working Paper. Available at SSRN: http://ssrn.com/abstract=1943090 Healy, P. M., Palepu, K. G., 2001. Information asymmetry, corporate disclosure, and the capital markets: A review of the empirical disclosure literature. Journal of Accounting and Economics. 31, 405-440. Hoberg, G., Phillips, G., 2010. Product Market Synergies and Competition in Mergers and Acquisitions: A Text-Based Analysis. Review of Financial Studies 23 (10), 3773-38011. Holzmann, O. J., Ramnath, S., 2013. FASB’s disclosure framework project. The Journal of Corporate Accounting & Finance. 24, 79-82 Imbens, G. W., 2010. Better LATE than nothing: Some comments on Deaton (2009) and Heckman and Urzua (2009). Journal of Economic Literature. 48, 399-423. Institute of Chartered Accountants of Scotland (ICAS), New Zealand Institute of Chartered Accountants (NZICA), 2011. Losing the Excess Baggage—Reducing Disclosures in Financial Statements to What’s Important. Available at: <http://icas.org.uk/WorkArea/DownloadAsset.aspx?id=4294970656 > International Auditing and Assurance Standards Board (IAASB), 2011. The evolving nature of financial reporting: Disclosure and its audit implications. Available at: <http://www.ifac.org/sites/default/files/publications/exposure-drafts/IAASB- Disclosures_Discussion_Paper.pdf > Investors Technical Advisory Committee (ITAC) Letter to Robert Herz. December 11, 2007. Available at <http://www.fasb.org/cs/BlobServer?blobkey=id&blobwhere=1175819498550&blobheader =application%2Fpdf&blobcol=urldata&blobtable=MungoBlobs > Leder, M. 2003. Financial Fine Print, Uncovering a Company’s True Value. John Wiley & Sons, Inc., Hoboken, New Jersey. Lehavy, R., Li, F., Merkley, K., 2011. The Effect of Annual Report Readability on Analyst Following and the Properties of Their Earnings Forecasts. The Accounting Review 86, 1087- 1115. Li, F., 2008. Annual Report Readability, Current Earnings, and Earnings Persistence. Journal of Accounting and Economics 45, 221-247. Li, F., 2010. The Information Content of Forward-Looking Statements in Corporate Filings—A Naïve Bayesian Machine Learning Approach. Journal of Accounting Research. 48, 1049- 1102. CAN I BORROW YOUR FOOTNOTES? 61 Li, F., Lundholm, R., Minnis, M., 2012. A Measure of Competition Based on 10-K Filings. Journal of Accounting Research, Forthcoming. Maurer, H., Kappe, F., Zaka, B., 2006. Plagiarism-A Survey. Journal of Universal Computer Science 12, 1050-1084. Mebane, W. R., Sekhon, J. S., 1998a. Genetic Optimization Using Derivatives. Political Analysis. 7, 187-210. Software Package. http://sekhon.berkeley.edu/rgenoud/. Mebane, W. R., Sekhon, J. S., 2011. Genetic optimization using derivatives: The rgenoud package for R. Journal of Statistical Software. 42, 1-26. Petersen, K., Schmardebeck, R., Wilks, T.J., 2012. Accounting comparability and earnings attributes. Working Paper, University of Oregon, University of Arkansas, and Brigham Young University. Radin, A.J., 2007. Have We Created Financial Statement Disclosure Overload?. The CPA Journal. November 7, 2007. Ramnath, S., 2002. Investor and analyst reactions to earnings announcements of related firms: An empirical analysis. Journal of Accounting Research 40: 1351–1376. Rogers, E. M., 2003. Diffusion of Innovations, 5 th edition, Free Press. Rogers, J.L., Van Buskirk, A., Zechman, S., 2011. Disclosure Tone and Shareholder Litigation. Working Paper, University of Chicago, and Ohio State University. Rosenbaum, P. R., and Rubin, D. B., 1983. The central role of the propensity score in observational studies for causal effects. Biometrika 70: 41–55. Rosenbaum, P. R., and Rubin, D. B., 1984. Reducing bias in observational studies using subclassification on the propensity score. Journal of the American Statistical Association 79: 516–524 Rubin, D. B., 2008. For objective causal inference, design trumps analysis. Annals of Applied Statistics 2: 808–840. Sekhon, J. S., 2011. Multivariate and propensity score matching software with automated balance optimization: The matching package for R Tawatnuntachai, O., D’Mello. R., 2002. Intra-industry reactions to stock split announcements. Journal of Financial Research 25: 39–57. Wysocki, P., 2008. Assessing earnings and accruals quality: U.S. and international evidence. Working paper, MIT. CAN I BORROW YOUR FOOTNOTES? 62 Appendix A. Footnote Examples Bold, underlined, red text: exact match Italicized, green text: imperfect match Black text: no match A.1 Professional Athlete Sponsorships Volcom, Inc. Fiscal year end: December 31, 2008 10-K file date: March 16, 2009 Market value (December 31, 2008): $265.7 Million Professional Athlete Sponsorships The Company establishes relationships with professional athletes in order to promote its products and brands. The Company has entered into endorsement agreements with professional athletes in skateboarding, snowboarding, surfing and motocross. Many of these contracts provide incentives for magazine exposure and competitive victories while wearing or using the Company’s products. Such expenses are an ordinary part of the Company’s operations and are expensed as incurred. The following is a schedule of future estimated minimum payments required under such endorsement agreements as of December 31, 2008 (in thousands): Quicksilver, Inc. Fiscal year end: Oct 31, 2009 10-K file date: January 12, 2010 Market value (Oct 31, 2009): $255.91 Million Professional Athlete Sponsorships The Company establishes relationships with professional athletes in order to promote its products and brands. The Company has entered into endorsement agreements with professional athletes in sports such as surfing, skateboarding, snowboarding, bmx and motocross. Many of these contracts provide incentives for magazine exposure and competitive victories while wearing or using the Company’s products. Such expenses are an ordinary part of the Company’s operations and are expensed as incurred. The following is a schedule of future estimated minimum payments required under such endorsement agreements as of October 31, 2009 (in thousands): CAN I BORROW YOUR FOOTNOTES? 63 A.2 Stock-Based Compensation Example Pacific Sunwear California, Inc. Fiscal year end: December 31, 2002 10-K file date: March 28, 2003 Market value (Dec 31, 2002) = $900.7 Million SFAS 123, “Accounting for Stock-Based Compensation,” requires the disclosure of pro forma net income and earnings per share. Under SFAS 123, the fair value of stock-based awards to employees is calculated through the use of option-pricing models, even though such models were developed to estimate the fair value of freely tradable, fully transferable options without vesting restrictions, which significantly differ from the Company’s stock option awards. These models also require subjective assumptions, including future stock price volatility and expected time to exercise, which greatly affect the calculated values. The Company’s calculations were made using the Black-Scholes option-pricing model with the following weighted average assumptions: expected life, 5 years from option date; stock volatility, 37.1% in fiscal 2003, 55.2% in fiscal 2002 and 71.4% in fiscal 2001; risk-free interest rates, 3.2% in fiscal 2003, 3.0% in fiscal 2002 and 4.4% in fiscal 2001; and no dividends during the expected term. The Company’s calculations are based on a single-option valuation approach and forfeitures are recognized as they occur. If the computed fair values of the fiscal 2003, 2002 and 2001 awards had been amortized to expense over the vesting period of the awards, pro forma net income and earnings per share would have been reduced to the pro forma amounts indicated in the following table: The Wet Seal, Inc. Fiscal year end: December 31, 2003 10-K file date: March 30, 2004 Market value (Dec 31, 2003) = $263.4 Million SFAS No. 123, “Accounting for Stock-Based Compensation,” requires the disclosure of pro forma net income and earnings per share had the Company adopted the fair value method as of the beginning of fiscal 1995. Under SFAS No. 123, the fair value of stock-based awards to employees is calculated through the use of option-pricing models, even though such models were developed to estimate the fair value of freely tradable, fully transferable options without vesting restrictions, which significantly differ from the Company’s stock-option awards. These models also require subjective assumptions, including future stock price volatility and expected time to exercise, which greatly affect the calculated values. The Company’s calculations were made using the Black-Scholes option-pricing model with the following weighted average assumptions: The Company’s calculations are based on a valuation approach and forfeitures are recognized as they occur. If the computed fair values of the fiscal 2003, fiscal 2002 and fiscal 2001 awards had been amortized to expense over the vesting period of the awards, net income (in thousands) and earnings per share would have been reduced to the pro forma amounts indicated below: CAN I BORROW YOUR FOOTNOTES? 64 Appendix B. Variable Definitions Variable Description #Text Match ij The number of words in text strings at least six words in length found in both footnote set i and footnote set j %Text Match ij #Text Match ij divided by #Words in Footnotes i #Text Match MAXij The maximum value of #Text Match ij for footnote set i %Text Match MAXij The maximum value of %Text Match ij for footnote set i #Words in Footnotes i The total number of words in footnote set i Same SIC 4 Dummy variable equal to one if firm i and firm j operate in the same SIC 4-digit industry, zero otherwise Same MSA Dummy variable equal to one if firm i and firm j are located in the same metropolitan statistical area (MSA), zero otherwise Same Auditor Office Dummy variable equal to one if firm i and firm j are audited by the same auditor from the same office, zero otherwise Big N Auditor Dummy variable equal to one if the firm’s auditor is a Big N auditor Accounting Comparability i,DKV De Franco et al., (2011) measure. See text for description. Accounting Comparability, DD Same as Accounting Comparability, DKV , except instead of using the earnings return regression as the model of accounting system, use the Dechow and Dichev (2002) model of working capital accruals. See text for description. Earnings Comovement i The R 2 from regressing firm i’s quarterly earnings on firm j’s quarterly earnings for all ij pairs in the same two-digit industry. A firm year measure is calculated as the highest four ij measures. |CARi (10-K Filing i)| Unsigned cumulative abnormal stock return (size- decile adjusted) for firm i around firm i's 10-K filing Analyst Following Number of analysts in the first consensus annual earnings forecast following the 10-K filing Analyst Accuracy Squared difference between I/B/E/S reported earnings and the first analyst consensus annual earnings forecast issued after the 10-K filing for the fiscal period following the 10-K filing, scaled by share price at the end of the prior fiscal year. CAN I BORROW YOUR FOOTNOTES? 65 Analyst Dispersion Standard deviation of the individual analyst forecasts in the first analyst consensus annual earnings forecast after the 10-K filing for the fiscal period following the 10-K filing, scaled by share price at the end of the prior fiscal year. |CARi (Earnings Announcment j)| Unsigned cumulative abnormal stock return (size- decile adjusted) for firm i around firm j's next earnings announcement after firm i’s 10-K filing |CARi (10-K Filing j)| Unsigned cumulative abnormal stock return (size- decile adjusted) for firm i around firm j's next 10-K filing after firm i’s 10-K filing Cosine Word Count Item 1 Cosine difference of firm i and firm j’s 10-K Item 1 word count vectors of words appearing in firm i and firm j’s 10-K Item 1 Average Daily EDGAR Search Volume Average daily downloads of firm i’s 10-K over the year following the filing of the 10-K on the SEC’s EDGAR system mean(#Text Matchj j,i Top10 ) Average number of words in firm j’s footnotes that can be found in the footnotes of the 10 most similar peer firms filing their 10-K after firm j. Cash Flow Comovement i Same as Earnings Comovement i,, but instead using cash flows (OANCF) Cash Flow Volatility The standard deviation of Cash Flows over the prior three years. Market Value Market value of equity at end of fiscal year.(PRCC_F*CSHO) Age Number of days between first day the firm appears on CRSP and the filing date of the 10-K, divided by 365.25 Business Segments Number of business segments Geographical Segments Number of geographical segments Return-on-Assets Operating income after depreciation divided by total assets (OIADP/AT) Leverage Long term debt divided by total assets (DLTT/AT) R&D over Operating Expense Research and development expense divided by total operating expense (XRD/XOPR) Working Capital over Assets Working capital divided by total assets (WCAP/AT) Goodwill over Assets Goodwill divided by total assets (GDWL/AT) Sales Growth Average year-over-year percentage change in sales during the prior three years. If three years are unavailable, I use two years. If two years are unavailable, I use one year. (SALE t – SALE t- 1 )/SALE t-1 Book-to-Market Book value of equity divided by market value of equity (CEQ/(PRCC_F*CSHO)) %Institutional Ownership Percent of total share outstanding held by institutions CAN I BORROW YOUR FOOTNOTES? 66 ΔWC Quarterly change in working capital, divided by average total assets. Measured as change in current assets minus change in current liabilities, minus the change in cash plus the change in short term debt. (TCA = ΔCA – ΔCL – ΔCash + ΔSTDEBT). In terms of the Compustat variable names, the components are measured as follows: ΔCA = ACTQ t – ACTQ t-1 ; ΔCL = LCTQ t – LCTQ t-1 ; ΔCash = CHEQ t – CHEQ t-1 ; ΔSTDEBT= DLCQ t – DLCQ t-1 . CFO Quarterly cash flow from operations. Income less total accruals. In terms of the Compustat variable names, IBQ – (TCA – DPQ). CAN I BORROW YOUR FOOTNOTES? 67 Appendix C. Matching methods: propensity score matching, genetic matching, and entropy balancing Matching Methods The gold standard for estimating a causal effect is a randomized control trial (RCT) since the process of random assignment to either the treatment or control condition results in stochastically equal distributions for all covariates between the treatment and control groups. Thus, in an RCT observing a difference in the outcome measure after the intervention can be ascribed as the effect of the treatment. This is due to randomization increasing the probability the treatment and control groups differ only on the receipt of the treatment or control program. However, because of ethical concerns and/or extreme costs many causal questions cannot be examined via a RCT. As a result, these questions can only be informed by using observational (archival) data to estimate causal effects. In observational data, the assignment to treatment conditions is not controlled by the researcher and often results in treatment and control samples that have covariate distributions that differ widely. The aim of matching methods is to reduce the differences in the treatment and control sample covariate distributions, obtain overlap, and to bolster the credibility of the selection on observables assumption. Under this assumption, the assignment to treatment is independent of the observed covariates. Employing exact matching is rarely effective since the dimensionality of the matching problem increases as the number of covariates grows and is further exacerbated when covariates are continuous. To overcome these problems, researchers often match on a scalar distance metric that summarizes difference in covariates between two observations in CAN I BORROW YOUR FOOTNOTES? 68 different groups. One commonly used distance metric is the Mahalanobis distance (MD). 31 MD is calculated as: !" ! ! ,! ! = ! (! ! −! ! ) ! ! !! (! ! −! ! ). (13) ! ! and ! ! are the vectors of the observed covariates for observations i and j, respectively. ! is the sample covariance matrix of the covariates. After calculating this metric, researchers typically use a nearest neighbor matching technique to select a matched sample. Propensity Score Matching Rosenbaum and Rubin (1983) point out that MD matching performs poorly when covariates have non-ellipsoidal distributions; a problem common to many applied settings. They offer the propensity score as an alternative distance metric. The propensity score is the probability the observation received the treatment, given the covariates. While Rosenbaum and Rubin (1983) show that matching on the true propensity score results in stochastically equalized covariate distributions between the treatment and control sample, implementing this matching method is not without its problems. The main problem is that the true propensity score is unknown and, therefore, must be estimated. After estimating the propensity score—typically, the predicted probability of a probit regression where the dependent variable is a treatment group indicator and the independent variables are the covariates—as with MD matching, researchers use some form of nearest neighbor matching to identify the matched treatment and control samples. Assessing whether the estimated propensity score is a good estimate of the true propensity score is accomplished by examining the covariate balance between the treated and matched 31 This distance metric is more commonly used in statistics than in economics. CAN I BORROW YOUR FOOTNOTES? 69 control samples. Covariate balance is achieved if the covariate distributions of two samples do not differ in means (using t-tests) or distributions (using two-sample Kolmogorov-Smirnov (KS) tests.) If balance is not achieved, the researcher alters the specification of the propensity score model to include higher ordered and/or interaction terms, re-estimates the propensity scores, re- performs the matching, and then examine the covariate balance. Searching for the correct specification of the propensity score model continues through this manually iterative process until balance is achieved. Rosenbaum and Rubin (1984) offer an algorithm to use when specifying a propensity score model and recommend the iterative process continue until covariate balance is maximized. Figure C.1 diagrams the major steps in implementing propensity score matching. Genetic Matching and Entropy Balancing With the purpose of achieving covariate balance, two recently developed data preprocessing methods—genetic matching (Diamond and Sekhon, forthcoming) and entropy balancing (Hainmueller, 2012)—take advantage of decreased computational costs to avoid “manually” iterating over propensity score specifications and checking covariate balance. These two methods have the same purpose as propensity score matching, but approach the task in different ways. Genetic matching. Genetic matching uses a generalized Mahalanobis distance (GMD) metric to identify nearest neighbors. GMD adds a weight matrix ! to MD and is computed as follows: !"# ! ! ,! ! ,! = ! (! ! −! ! ) ! (! !!/! ) ! !!! !!/! (! ! −! ! ). (14) ! !!/! is the Cholesky decomposition of !; note that != !! !!/! (! !!/! ) ! . ! is a !×! matrix with all elements equaling zero except the diagonal which consists of ! parameters. Note also that k is the number of covariates included in !. Using GENOUD, an evolutionary search CAN I BORROW YOUR FOOTNOTES? 70 algorithm 32 developed by Mebane and Sekhon 1998, genetic matching identifies the ! parameters in the weight matrix to maximize covariate balance between matched treated and control observations. Covariate balance is measured by the minimum p-value of a set of balance tests, (e.g., t-tests and KS-tests comparing covariate distributions between matched treated and control samples). Genetic matching seeks to identify the ! matrix that maximizes the smallest p-value in the set of balance tests. Genetic matching is a generalization of both propensity score and Mahalanobis distance matching since both the covariates and an estimated propensity score can be included in the ! matrix. To see this, note that if balance is maximized using the propensity score alone, the algorithm zero weights all parameters except for the one corresponding to the propensity score. Likewise, if the propensity score is not useful in identifying matches, the genetic matching algorithm will zero weight the propensity score’s parameter. Figure C.2 diagrams the major steps in implementing genetic matching. Entropy balancing. Entropy balancing is similar to both propensity score matching and genetic matching in that these three methods focus on weighting the observations in the control group to maximize covariate balance between the treatment and control samples. However, the entropy balancing differs in how it approaches the maximization problem. While propensity score matching requires manually iterating through propensity score model specifications and while genetic matching automates iterating through different GMD metrics—with the purpose of identifying control sample observations’ weights—such that matching on the propensity scores or on the GMD metric maximizes covariate balance; entropy balancing approaches the problem 32 Mebane and Sekhon (2011) explain that “An evolutionary algorithm (EA) uses a collection of heuristic rules to modify a population of trial solutions in a way that each generation of trial values tends to be on average, better than its predecessor.” CAN I BORROW YOUR FOOTNOTES? 71 by estimating the control sample observations weights directly, subject to balance constraints. That is, instead of searching for a propensity score or distance metric that identifies observation weights that achieve covariate balance, entropy balancing reverses the problem and searches for the observations weights. Implementing entropy balancing requires the researcher to indicate the covariates to be balanced as well as a set of balance constraints. The balancing constraints involve requiring equality of the first, second, and/or third moments of the covariate distributions. With the constraints specified, entropy balancing searches for a set of observational weights to satisfy these constraints while also keeping weights as close as possible to the initial weights so as to keep from losing information. Although the Hainmueller (2012) notes that entropy balancing obviates checking covariate balance, he does note some limitations to entropy balancing. The main limitation relates to covariate overlap between the two samples being matched. If the two samples do not have sufficient overlap, entropy balancing may not find a set of weights to satisfy. Note that these limitations are not unique to entropy balancing, but rather pervade all matching methods. Figure C.3 diagrams the major steps in implementing Entropy balancing. Comparing Matching Methods As with all matching methods, researchers implement propensity score matching, genetic matching and entropy balancing for the same purpose, reduce covariate imbalance between two samples. However, the matching methodology literature has not reached a consensus on which matching method is the superior data preprocessing technique. This may be due to each method having apparent shortcomings. While matching on propensity scores has attractive theoretical properties, this is the case if the true propensity score it known. In applying propensity score CAN I BORROW YOUR FOOTNOTES? 72 matching, researchers iterate through propensity score models until covariate balance is achieved. This manual, iterative method is both time consuming and may also result in analyzing outcomes between groups that lack optimal covariate balance. Genetic matching makes use of modern computational resources and automates the process of obtaining covariate balance. Using an evolutionary search algorithm, genetic matching identifies a distance metric that when used in matching minimizes covariate imbalance. Unfortunately the maximization problem is usually irregular and, even with modern computational capacities, relatively costly. Moreover, propensity score matching and Mahalanobis distance matching are not guaranteed to improve balance across observed covariates (Diamond and Sekhon, forthcoming). Rather than focusing on finding a distance metric that results in covariate balance after matching, entropy balancing focuses covariate balance directly by identifying weights on control group observations that ensure the treatment sample and weighted control sample balance on specific distributional moments. Not only is covariate balance practically guaranteed, this optimization problem is far less computationally costly than genetic matching. Additionally, unlike the other methods, entropy balancing ensures the match results exhibit no worse balance the prior to matching in equal percent bias reduction (EPBR) sense (Hainmueller, 2012). Thus, while matching methodology literature does not agree about which matching method is superior; it does agree that achieving covariate balance is necessary when estimating. Thus, researchers’ attention should focus on achieving covariate balance, rather than on selecting which method is superior. CAN I BORROW YOUR FOOTNOTES? 73 Figure C.1Propensity score matching diagram This figure depicts the main steps in implementing propensity score matching as explained in Rosenbaum and Rubin, 1983. Figure C.2. Genetic matching diagram This figure depicts the main steps in implementing genetic matching as explained in Diamond and Sekhon, forthcoming. GMD is Generalized Mahalanobis Distance. See paper’s text for formula. CAN I BORROW YOUR FOOTNOTES? 74 Figure C.3. Entropy balancing diagram This figure depicts the main steps in implementing entropy balancing as explained in Hainmueller, 2012. CAN I BORROW YOUR FOOTNOTES? 75 Tables Table 1 Descriptive Statistics Panel A: Footnote Similarity Measures Mean Std. Dev. 5th Pctl. 25th Pctl. Median 75th Pctl. 95th Pctl. N All ij Comparisons #Text Match i,j 677.9 380.4 200 415 602 870 1,407 17,860,665 %Text Match i,j 0.097 0.043 0.033 0.068 0.094 0.123 0.172 17,839,616 Highest ij Comparison, for each i #Text Match MAXi,j 1,274.6 767.0 425 749 1,098 1,648 2,625 41,331 %Text Match MAXi,j 0.185 0.069 0.096 0.143 0.179 0.217 0.284 41,331 Footnote Length #Words in Footnotes i 7,590.1 5,051.2 2,309 4,115 6,294 9,763 17,064 41,331 CAN I BORROW YOUR FOOTNOTES? 76 Table 1 Descriptive Statistics Panel B: Yearly Means of Footnote Similarity Measures Year #Text Match i,j %Text Match i,j N (ij comparisons) #Text Match MAXi,j %Text Match MAXi,j Mean #Words in Footnotes i N (unique firms) Mean Number of Firms per Industry 1998 431.3 0.101 1,488,979.0 836.6 0.194 4,795.4 3,539 58.0 1999 462.2 0.103 1,593,246.0 891.6 0.197 5,009.8 3,242 53.1 2000 508.1 0.102 1,759,374.0 967.4 0.195 5,421.8 3,195 53.3 2001 572.8 0.099 1,770,116.0 1,092.3 0.191 6,330.3 3,003 50.9 2002 656.0 0.095 1,627,872.0 1,289.9 0.188 7,504.0 2,783 48.0 2003 723.5 0.097 1,405,552.0 1,408.6 0.188 8,215.9 2,579 44.5 2004 751.7 0.094 1,243,992.0 1,423.7 0.177 8,832.3 2,569 44.3 2005 809.4 0.094 1,155,150.0 1,553.4 0.177 9,548.8 2,524 43.6 2006 884.5 0.093 1,072,293.0 1,745.5 0.181 10,445.7 2,461 42.4 2007 965.0 0.097 982,328.0 1,872.5 0.185 10,885.2 2,316 40.0 2008 1,024.1 0.099 894,960.0 1,981.5 0.185 11,520.4 2,200 37.9 2009 991.9 0.093 823,221.0 1,940.6 0.177 11,863.4 2,183 37.0 2010 976.4 0.092 770,758.0 1,888.3 0.173 11,870.3 2,074 34.6 Variable definitions: See Appendix B . CAN I BORROW YOUR FOOTNOTES? 77 Table 2 Industry, Geographical Location, and Auditor Office Subsample Analysis Panel A: Descriptive Statistics of SIC 4, MSA, and Auditor Office Dummy Variables Variable Mean Std. Dev. N All ij Comparisons Same SIC 4 0.2100 0.4073 9,946,753 Same MSA 0.0752 0.2637 9,946,753 Same Auditor Office 0.0117 0.1077 9,946,753 Same MSA*Same SIC 4 0.0239 0.1526 9,946,753 Same Auditor Office*Same MSA 0.0092 0.0955 9,946,753 Same Auditor Office*Same SIC 4 0.0042 0.0650 9,946,753 Same Auditor Office*Same SIC 4*Same MSA 0.0035 0.0592 9,946,753 Highest ij Comparison, for each i Same SIC 4 0.360 0.480 21,543 Same MSA 0.203 0.402 21,543 Same Auditor Office 0.107 0.309 21,543 Same MSA*Same SIC 4 0.101 0.301 21,543 Same Auditor Office*Same MSA 0.085 0.279 21,543 Same Auditor Office*Same SIC 4 0.052 0.222 21,543 Same Auditor Office*Same SIC 4*Same MSA 0.043 0.202 21,543 CAN I BORROW YOUR FOOTNOTES? 78 Table 2 Industry, Geographical Location and Auditor Office Subsample Analysis Panel B: SIC 4, MSA, and Auditor Office Subsamples of All ij Comparisons Sample %Text Match i,j Subsamples Mean Median N (1) SIC 4 Same SIC 4 0.105 0.102 2,088,609 Different SIC 4 0.096 0.092 7,858,144 T-Stat. or Z-Stat. -287.91 291.42 P-Value <.0001 <.0001 9,946,753 (2) SIC 4 (Different Auditor Office and Different MSA) Same SIC 4 0.104 0.101 1,844,025 Different SIC 4 0.095 0.092 7,329,753 T-Stat. or Z-Stat. -253.87 259.175 P-Value <.0001 <.0001 9,173,778 (3) MSA Same MSA 0.108 0.105 747,786 Different MSA 0.097 0.094 9,198,967 T-Stat. or Z-Stat. -217.39 203.041 P-Value <.0001 <.0001 9,946,753 (4) MSA (Different Auditor Office and Different SIC 4) Same MSA 0.102 0.099 453,820 Different MSA 0.095 0.092 7,329,753 T-Stat. or Z-Stat. -111.41 109.393 P-Value <.0001 <.0001 7,783,573 [Continued] CAN I BORROW YOUR FOOTNOTES? 79 Table 2 (Continued) Industry, Geographical Location and Auditor Office Subsample Analysis Panel B: SIC 4, MSA, and Auditor Office Subsamples of All ij Comparisons Sample %Text Match i,j Subsamples Mean Median N (5) Auditor Office Same Auditor Office 0.125 0.120 116,776 Different Auditor Office 0.097 0.094 9,829,977 T-Stat. or Z-Stat. -221.93 183.066 P-Value <.0001 <.0001 9,946,753 (6) Auditor Office (Different MSA and Different SIC 4) Same Auditor Office 0.116 0.111 17,912 Different Auditor Office 0.095 0.092 7,329,753 T-Stat. or Z-Stat. -68.1 54.1452 P-Value <.0001 <.0001 7,347,665 (7) SIC 4, MSA & Auditor Office Same SIC 2, MSA & Auditor Office 0.134 0.129 34,928 Different SIC 2, MSA & Auditor Office 0.095 0.092 7,329,753 T-Stat. or Z-Stat. -173.78 142.282 P-Value <.0001 <.0001 7,364,681 CAN I BORROW YOUR FOOTNOTES? 80 Table 2 Industry, Geographical Location and Auditor Office Subsample Analysis Panel C: SIC 4, MSA, and Auditor Office Subsamples of Highest ij Comparisons Sample %Text Match MAXi,j Subsamples Mean Median N (1) SIC 4 Same SIC 4 0.188 0.182 7,764 Different SIC 4 0.184 0.180 13,779 T-Stat. or Z-Stat. -4.34 2.7788 P-Value <.0001 0.0055 21,543 (2) SIC 4 (Different Auditor Office and Different MSA) Same SIC 4 0.174 0.171 5,381 Different SIC 4 0.178 0.175 11,312 T-Stat. or Z-Stat. 4.39 -4.2278 P-Value <.0001 <.0001 16,693 (3) MSA Same MSA 0.215 0.207 4,378 Different MSA 0.178 0.175 17,165 T-Stat. or Z-Stat. -33.62 34.1729 P-Value <.0001 <.0001 21,543 (4) MSA (Different Auditor Office and Different SIC 4) Same MSA 0.195 0.195 1,286 Different MSA 0.178 0.175 11,312 T-Stat. or Z-Stat. -9.61 12.5336 P-Value <.0001 <.0001 12,598 [Continued] CAN I BORROW YOUR FOOTNOTES? 81 Table 2 (Continued) Industry, Geographical Location and Auditor Office Subsample Analysis Panel C: SIC 4, MSA, and Auditor Office Subsamples of Highest ij Comparisons Sample %Text Match MAXi,j Subsamples Mean Median N (5) Auditor Office Same Auditor Office 0.238 0.226 2,305 Different Auditor Office 0.179 0.177 19,238 T-Stat. or Z-Stat. -41.85 38.1121 P-Value <.0001 <.0001 21,543 (6) Auditor Office (Different MSA and Different SIC 4) Same Auditor Office 0.233 0.224 265 Different Auditor Office 0.178 0.175 11,312 T-Stat. or Z-Stat. -14.53 14.8182 P-Value <.0001 <.0001 11,577 (7) SIC 4, MSA & Auditor Office Same SIC 2, MSA & Auditor Office 0.245 0.226 917 Different SIC 2, MSA & Auditor Office 0.178 0.175 11,312 T-Stat. or Z-Stat. -30.19 25.6374 P-Value <.0001 <.0001 12,229 CAN I BORROW YOUR FOOTNOTES? 82 Table 3 High and Low Footnote Similarity Decile Subsample Analysis Whole Sample Bottom & Top %Text Match MAX i,j Decile Subsamples P0 - P100 P0 - P10 P90 - P100 (All Firms) (Low Footnote Similarity Firms) (High Footnote Similarity Firms) Difference Between High & Low Deciles (P-value) Variables Mean N Mean N Mean N Footnote Simiarilty Measure %Text Match MAXi,j 0.185 41,565 0.091 4,156 0.316 4,156 <.0001 SIC 4, MSA, and Auditor Office Dummy Variables Same SIC 4 0.346 41,565 0.376 4,156 0.414 4,156 0.0003 Same SIC 4 (Different MSA & Auditor Office) 0.250 21,638 0.348 1,778 0.156 1,966 <.0001 Same MSA 0.189 36,694 0.084 3,328 0.420 3,842 <.0001 Same MSA (Different SIC 4 & Auditor Office) 0.051 25,315 0.027 2,364 0.055 2,207 <.0001 Same Auditor Office 0.102 25,315 0.029 2,364 0.356 2,207 <.0001 Same Auditor Office (Different MSA & SIC 4) 0.007 36,694 0.002 3,328 0.021 3,842 <.0001 Same SIC 4, MSA & Auditor Office 0.043 21,638 0.009 1,778 0.150 1,966 <.0001 Firm Size Market Value (firm i) 2,677.8 34,454 11,341.4 2,993 629.5 3,449 Market Value (firm j) 3,883.8 34,454 6,687.3 2,993 2,375.9 3,449 Market Value (firm i) - Market Value (firm j) - 1,206.0 4,654.1 -1,746.4 Difference in Means (T-Stat.) -12.5 5.98 -11.92 P-Value <.0001 <.0001 <.0001 CAN I BORROW YOUR FOOTNOTES? 83 Table 4 Footnote Similarity Regression Panel A: Descriptive Statistics of Regression Sample Footnote Similarity Sample (N = 20,109) Compustat Mean Std. Dev. 25th Pctl. Median 75th Pctl. Mean N %Text Match MAXi,j 0.185 0.059 0.147 0.181 0.216 _ _ #Text Match MAXi,j 1,582.0 798.0 1,062.0 1,464.0 1,969.0 _ _ ln(#Text Match MAXi,j ) -1.735 0.314 -1.918 -1.708 -1.534 _ _ #Words in Footnotes) 9,141.3 4,596.2 5,771.0 8,292.0 11,555.0 _ _ ln(#Words in Footnotes) 8.990 0.539 8.661 9.023 9.355 _ _ Same SIC 4 0.356 0.479 _ _ _ _ _ Same MSA 0.205 0.404 _ _ _ _ _ Same Auditor Office 0.108 0.310 _ _ _ _ _ Same MSA*Same SIC 4 0.102 0.303 _ _ _ _ _ Same Auditor Office*Same SIC 4 0.052 0.222 _ _ _ _ _ Same Auditor Office*Same MSA 0.086 0.280 _ _ _ _ _ Same Auditor Office*Same MSA*SameSIC 4 0.042 0.201 _ _ _ _ _ Big-N Auditor 0.769 0.422 _ _ _ _ _ Market Value 2,470.6 13,003.2 60.8 250.5 1,034.9 1,742.3 110,764 ln(Market Value) 5.566 2.040 4.108 5.524 6.942 5.224 110,761 Age 15.676 14.361 6.040 11.157 20.008 13.450 78,971 ln(Age) 2.382 0.891 1.798 2.412 2.996 1.996 78,955 Business Segments 2.042 1.505 1 1 3 1.611 110,764 ln(Business Segments) 0.490 0.637 0 0 1.099 0.881 110,764 Geographical Segments 2.404 1.975 1 2 3 1.769 110,764 ln(Geographical Segments) 0.625 0.677 0 0.693 1.099 0.929 110,764 [Continued] CAN I BORROW YOUR FOOTNOTES? 84 Table 4 (continued) Footnote Similarity Regression Panel A: Descriptive Statistics of Regression Sample Footnote Similarity Sample (N = 20,109) Compustat Mean Std. Dev. 25th Pctl. Median 75th Pctl. Mean N Return-on-Assets -0.032 0.278 -0.063 0.050 0.106 0.005 110,626 Leverage 0.137 0.167 0 0.061 0.240 0.150 110,392 R&D over Operating Expense 0.111 0.189 0 0.016 0.151 0.063 110,764 Working Capital over Assets 0.303 0.257 0.098 0.276 0.493 0.287 86,801 Goodwill over Assets 0.102 0.147 0 0.023 0.161 0.058 110,764 Sales Growth 0.431 1.768 0.010 0.110 0.283 0.406 101,538 Book-to-Market 0.647 0.662 0.272 0.471 0.783 0.704 110,761 Cash Flow Volatility 0.091 0.173 0.024 0.049 0.097 0.158 104,725 CAN I BORROW YOUR FOOTNOTES? 85 Table 4 Footnote Similarity Regression Panel B: Covariance Matrix of Regression Variables (Pearson Correlations) !! !! !! !! !! Variables Column (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) %Text Match MAXi,j (1) ln(#Text Match MAXi,j ) (2) 0.95 ! ln(#Words in Footnotes) (3) -0.41 -0.46 Same SIC 4 (4) 0.05 0.02 0.07 Same MSA (5) 0.25 0.23 0.05 0.15 Same Auditor Office (6) 0.31 0.27 -0.01 0.09 0.51 Same MSA*Same SIC 4 (7) 0.19 0.17 0.07 0.45 0.66 0.33 Same Auditor Office*Same SIC 4 (8) 0.24 0.20 0.02 0.31 0.35 0.67 0.55 Same Auditor Office*Same MSA (9) 0.27 0.24 0.01 0.09 0.60 0.88 0.40 0.61 Same Auditor Office*Same MSA*SameSIC 4 (10) 0.21 0.18 0.02 0.28 0.41 0.61 0.62 0.90 0.69 Big N Auditor (11) -0.06 -0.09 0.14 0.04 0.08 0.14 0.06 0.09 0.14 0.10 ln(Market Value) (12) -0.27 -0.31 0.45 0.07 0.06 0.03 0.07 0.04 0.05 0.05 0.42 ln(Age) (13) -0.29 -0.31 0.12 -0.08 -0.14 -0.14 -0.10 -0.10 -0.13 -0.09 -0.01 ln(Business Segmentes) (14) -0.28 -0.30 0.20 -0.09 -0.13 -0.10 -0.10 -0.07 -0.10 -0.07 0.06 ln(Geographical Segments) (15) -0.08 -0.08 0.17 -0.06 0.05 0.03 0.04 0.02 0.04 0.04 0.14 Return-on-Assets (16) -0.17 -0.18 0.05 -0.01 -0.09 -0.10 -0.05 -0.05 -0.09 -0.05 0.10 Leverage (17) -0.25 -0.28 0.24 0.04 -0.06 -0.05 0.00 -0.01 -0.04 0.00 0.12 R&D over Operating Expense (18) 0.24 0.26 -0.03 0.04 0.23 0.24 0.13 0.15 0.22 0.14 0.09 Working Capital over Assets (19) 0.31 0.34 -0.29 -0.09 0.11 0.12 0.01 0.04 0.11 0.04 -0.03 Goodwill over Assets (20) -0.09 -0.07 0.18 -0.06 -0.04 -0.06 -0.04 -0.05 -0.05 -0.04 0.05 Sales Growth (21) 0.06 0.06 0.00 0.03 0.03 0.03 0.03 0.03 0.02 0.02 -0.03 Book-to-Market (22) -0.02 -0.02 -0.06 0.02 -0.05 -0.05 -0.02 -0.02 -0.05 -0.02 -0.07 Cash Flow Volatility (23) 0.13 0.14 -0.09 -0.02 0.04 0.07 0.02 0.03 0.04 0.02 -0.10 The italicized correlations are not significant. P-value > 0.01. [Continued] CAN I BORROW YOUR FOOTNOTES? 86 Table 4 (continued) ! ! ! ! ! ! ! ! ! Footnote Similarity Regression ! ! ! ! ! ! ! ! ! !! !! !! !! !! !! !! !! !! !! !! Variables Column (12) (13) (14) (15) (16) (17) (18) (19) (20) (21) (22) %Text Match MAXi,j (1) ln(#Text Match MAXi,j ) (2) ln(#Words in Footnotes) (3) Same SIC 4 (4) Same MSA (5) Same Auditor Office (6) Same MSA*Same SIC 4 (7) Same Auditor Office*Same SIC 4 (8) Same Auditor Office*Same MSA (9) Same Auditor Office*Same MSA*SameSIC 4 (10) Big N Auditor (11) ln(Market Value) (12) ln(Age) (13) 0.21 ln(Business Segmentes) (14) 0.18 0.27 ln(Geographical Segments) (15) 0.23 0.12 0.11 Return-on-Assets (16) 0.37 0.27 0.18 0.15 Leverage (17) 0.20 0.12 0.16 -0.08 0.16 R&D over Operating Expense (18) -0.05 -0.22 -0.25 -0.05 -0.49 -0.21 Working Capital over Assets (19) -0.16 -0.15 -0.22 0.06 -0.16 -0.40 0.43 Goodwill over Assets (20) 0.17 0.02 0.15 0.09 0.16 0.12 -0.14 -0.33 Sales Growth (21) -0.03 -0.16 -0.07 -0.09 -0.16 -0.01 0.14 0.04 -0.03 Book-to-Market (22) -0.41 -0.02 0.03 -0.05 -0.03 0.02 -0.16 -0.07 -0.02 -0.04 Cash Flow Volatility (23) -0.22 -0.18 -0.14 -0.11 -0.44 -0.16 0.28 0.19 -0.13 0.20 -0.06 ! ! ! ! ! CAN I BORROW YOUR FOOTNOTES? 87 Table 4 Footnote Similarity Regression Panel C: Pooled Regressions of Footnote Similarity Measures on Industry, Auditor Office, Geographic, Firm Complexity, and Financial Variables Dep. Var = %Text Match MAXi,j Dep. Var = ln(#Text Match MAXi,j ) Coeff.Estimate Coeff.Estimate Variables T-Stat T-Stat Intercept 0.198 *** 0.635 *** 24.55 5.9 MSA & Auditor Office Variables Same SIC 4 0.005 *** 0.033 *** 3.67 5.66 Same MSA 0.010 *** 0.065 *** 5.4 7.72 Same Auditor Office 0.042 *** 0.188 *** 7.78 10.41 Same MSA*Same SIC 4 0.005 * 0.020 1.73 1.46 Same Auditor Office*Same SIC 4 0.006 0.017 0.48 0.47 Same Auditor Office*Same MSA -0.014 ** -0.071 *** -2.13 -3.26 Same Auditor Office*Same MSA*Same SIC 4 0.009 0.025 0.65 0.55 Firm Complexity ln(Market Value) -0.007 *** -0.018 *** -14.55 -7.19 ln(Age) -0.006 *** -0.046 *** -6.02 -11.33 ln(Business Segments) -0.009 *** -0.035 *** -8.58 -6.66 ln(Geographical Segments) -0.008 *** -0.015 *** -6.78 -3.04 [Continued] CAN I BORROW YOUR FOOTNOTES? 88 Table 4 Footnote Similarity Regression Panel C: Pooled Regressions of Footnote Similarity Measures on Industry, Auditor Office, Geographic, Firm Complexity, and Financial Variables Dep. Var = %Text Match MAXi,j Dep. Var = ln(#Text Match MAXi,j ) Coeff.Estimate Coeff.Estimate Variables T-Stat T-Stat Intercept 0.198 *** 0.635 *** 24.55 5.9 MSA & Auditor Office Variables Same SIC 4 0.005 *** 0.033 *** 3.67 5.66 Same MSA 0.010 *** 0.065 *** 5.4 7.72 Same Auditor Office 0.042 *** 0.188 *** 7.78 10.41 Same MSA*Same SIC 4 0.005 * 0.020 1.73 1.46 Same Auditor Office*Same SIC 4 0.006 0.017 0.48 0.47 Same Auditor Office*Same MSA -0.014 ** -0.071 *** -2.13 -3.26 Same Auditor Office*Same MSA*Same SIC 4 0.009 0.025 0.65 0.55 Firm Complexity ln(Market Value) -0.007 *** -0.018 *** -14.55 -7.19 ln(Age) -0.006 *** -0.046 *** -6.02 -11.33 ln(Business Segments) -0.009 *** -0.035 *** -8.58 -6.66 ln(Geographical Segments) -0.008 *** -0.015 *** -6.78 -3.04 CAN I BORROW YOUR FOOTNOTES? 89 Table 4 (continued) Footnote Similarity Regression Panel C: Pooled Regressions of Footnote Similarity Measures on Industry, Auditor Office, Geographic, Firm Complexity, and Financial Variables Financial Characteristics Return-on-Assets 0.016 *** 0.020 * 6.71 1.9 Leverage -0.020 *** -0.019 -4.75 -0.97 R&D over Operating Expense -0.003 0.055 *** -0.61 2.76 Working Capital over Assets 0.038 *** 0.106 *** 11.36 7.33 Goodwill over Assets 0.005 0.077 *** 1.05 3.33 Sales Growth 0.000 0.002 0.5 1.43 Book-to-Market -0.007 *** -0.014 *** -6.99 -3.12 Cash Flow Volatility -0.004 -0.021 * -1.38 -1.67 Big-N Auditor 0.004 *** 0.033 *** 2.71 4.81 Footnote Length ln(#Words in Footnotes) _ -0.264 *** _ -25.37 Fixed Effects Year Yes Yes SIC 2 Industry Yes Yes R 2 0.371 0.547 N 19,020 19,020 CAN I BORROW YOUR FOOTNOTES? 90 Table 5 EDGAR Search Volume and Footnote Similarity Panel A: Descriptive Statistics(N = 5,964) Variable Mean Std. Dev. 1st Pct. 25th Pctl. Median 75th Pctl. 99th Pctl. Average Daily EDGAR Search Volume 5.88 6.99 1.24 2.67 3.89 6.38 39.34 ln(Average Daily EDGAR Search Volume) 1.46 0.70 0.22 0.98 1.36 1.85 3.67 mean(#Text Matchj j,i Top10 ) 1,620.67 606.06 280.22 1,184.39 1,594.33 2,024.22 3,161.44 Market Value 3,471.87 16,675.55 6.25 97.07 397.23 1,567.11 54,939.36 ln(Market Value) 6.02 2.00 1.83 4.58 5.98 7.36 10.91 Age in Years 18.64 16.24 1.36 8.27 13.86 24.22 80.20 ln(Age in Years) 2.55 0.92 0.31 2.11 2.63 3.19 4.38 Return-on-Assets 0.00 0.24 -1.08 -0.02 0.06 0.11 0.34 Book-to-Market 0.71 0.68 0.03 0.31 0.53 0.87 3.71 %Institutional Ownership 0.58 0.34 0.00 0.29 0.64 0.87 1.20 Business Segments 2.64 2.36 1.00 1.00 2.00 4.00 11.00 ln(Business Segments) 0.69 0.72 0.00 0.00 0.69 1.39 2.40 Geographical Segments 2.14 1.61 1.00 1.00 1.00 3.00 7.00 ln(Geographical Segments) 0.52 0.66 0.00 0.00 0.00 1.10 1.95 Price Volatility 4.04 5.46 0.22 1.32 2.55 4.84 24.45 CAN I BORROW YOUR FOOTNOTES? 91 Table 5 EDGAR Search Volume and Footnote Similarity Panel B: Regression of Footnote similarity on EDGAR Search Volume Dependent Variable = Variables ln(Average Daily EDGAR Search Volume) Intercept !0.796' *** !19.270' Variable of Interest ' mean(#Text Matchj j,i Top10 ) 0.004' *** 2.760' Control Variables ' ln(Market Value) 0.308' *** ' 42.570' ln(Age in Years) 0.084' *** ' 9.310' Return-on-Assets !0.131' *** ' !4.110' Book-to-Market 0.217' *** 19.930' %Institutional Ownership !0.036' !1.290' ln(Business Segments) !0.009' ' !0.840' ln(Geographical Segments) !0.015' !1.190' Cash Flow Volatility 0.002' 1.000' R 2 0.695 N 5,964 CAN I BORROW YOUR FOOTNOTES? 92 Table 6 Network Benefits of Footnote Similarity Analysis Descriptive Statistics Mean Std. Dev. 5th Pctl. 25th Pctl. Median 75th Pctl. 95th Pctl. N Dependent Variables Accounting Comparability i,DKV -0.637 1.076 -6.580 -0.630 -0.260 -0.120 -0.030 13,114 Accounting Comparability i,DD -0.611 0.476 -2.410 -0.770 -0.470 -0.300 -0.110 7,883 Earnings Comovement i 0.633 0.170 0.230 0.520 0.640 0.750 0.970 13,998 |CARi (10-K Filing i)| 0.045 0.050 0.000 0.012 0.028 0.057 0.255 13,208 Analyst Following 7.538 6.608 1 2 6 11 30 9,681 Analyst Accuracy 0.028 0.116 0.000 0.000 0.002 0.010 0.622 9,536 Analyst Dispersion 0.103 0.135 0.000 0.030 0.060 0.120 0.800 8,237 |CARi (Earnings Announcment j)| 0.036 0.035 0.000 0.011 0.024 0.050 0.162 11,432 |CARi (10-K Filing j)| 0.034 0.034 0.000 0.010 0.023 0.046 0.153 11,207 Variable of Interest %Text Match MAXi,j 0.183 0.054 0.075 0.145 0.179 0.214 0.355 14,128 [Continued.] CAN I BORROW YOUR FOOTNOTES? 93 Table 6 (continued) Network Benefits of Footnote Similarity Analysis Descriptive Statistics Mean Std. Dev. 5th Pctl. 25th Pctl. Median 75th Pctl. 95th Pctl. N Control Variables Cosine Word Count Item 1 0.621 0.139 0.271 0.526 0.629 0.724 0.895 14,128 Market Value 3276.9 15606.1 4.3 69.3 306.0 1344.0 59796.0 14,128 ln(Market Value) 5.767 2.116 1.467 4.238 5.724 7.203 10.999 14,128 Cash Flow Comovement 0.572 0.139 0.220 0.490 0.580 0.660 0.880 14,128 Cash Flow Volatility 0.074 0.116 0.003 0.022 0.044 0.083 0.522 14,128 Goodwill 0.105 0.146 0.000 0.000 0.034 0.169 0.602 14,128 Return-on-Assets 0.006 0.230 -0.978 -0.022 0.062 0.116 0.331 14,128 Leverage 0.144 0.166 0.000 0.000 0.087 0.247 0.655 14,128 Book-to-Market 0.613 0.587 0.034 0.268 0.458 0.757 3.027 14,128 R&D over Operating Expense 0.107 0.186 0.000 0.000 0.021 0.137 0.868 14,128 Working Capital over Assets 0.300 0.246 -0.168 0.104 0.272 0.479 0.864 14,128 Sales Growth 0.302 1.391 -0.408 0.012 0.098 0.230 5.013 14,128 Geographical Segments 2.574 2.134 1 1 2 3 10 14,128 ln(Geographical Segments) 0.695 0.680 0 0 0.693 1.099 2.303 14,128 Business Segments 2.163 1.581 1 1 1 3 7 14,128 ln(Business Segments) 0.541 0.655 0 0 0 1.099 1.946 14,128 Age in Years 17.96 15.37 3.87 7.74 12.68 22.35 79.23 14,128 ln(Age in Years) 2.606 0.730 1.353 2.047 2.540 3.107 4.372 14,128 %Institutional Ownership 0.466 0.351 0 0.108 0.475 0.770 1.156 14,128 Big-N Auditor 0.798 0.401 _ _ _ _ _ 14,128 Price Volatility 3.743 5.848 0.185 1.170 2.248 4.227 26.500 14,128 CAN I BORROW YOUR FOOTNOTES? 94 Table 7 Footnote Similarity Benefits Analysis - Accounting Comparability Pooled Regressions of Accounting Comparability on Footnote Similarity Accounting Comparability i,DKV Accounting Comparability i,DD Earnings Comovement i Model 1 Model 2 Model 3 Coeff.Estimate Coeff.Estimate Coeff.Estimate Variables Coeff. T-Stat T-Stat T-Stat Intercept α -1.647 *** -1.234 *** 0.222 *** -7.86 -10.21 12.83 Variables of Interest %Text Match MAXi,j β 1 1.925 *** 0.742 *** 0.058 * 6.17 4.95 1.8 Control Variables Cosine Word Count Item 1 β 2 0.065 0.038 0.021 ** 0.76 0.83 2.02 ln(Market Value) β 3 0.064 *** 0.028 *** 0.001 5.76 5.21 0.75 Cash Flow Comovement β 4 -0.301 ** 0.082 0.413 *** -2.25 1.44 25.11 Cash Flow Volatility β 5 -0.708 *** -1.839 *** 0.019 -4.07 -11.58 1.58 Goodwill β 6 0.125 0.187 *** -0.008 1.42 3.85 -0.62 [Continued.] CAN I BORROW YOUR FOOTNOTES? 95 Table 7 (continued) Footnote Similarity Benefits Analysis - Accounting Comparability Pooled Regressions of Accounting Comparability on Footnote Similarity Return-on-Assets β 7 0.827 *** 0.461 *** -0.017 ** 9.17 7.5 -2.27 Leverage β 8 -0.637 *** 0.191 *** 0.005 -5.97 3.71 0.44 Book-to-Market β 9 -0.362 *** 0.060 *** 0.017 *** -9.38 3.23 5.45 R&D over Operating Expense β 10 0.202 ** 0.123 ** 0.023 ** 2.38 2.21 2.04 Sales Growth β 11 -0.006 -0.003 -0.002 -0.66 -0.51 -1.61 ln(Geographical Segments) β 12 -0.057 *** -0.014 0.003 -2.69 -1.33 1.11 ln(Business Segments) β 13 0.022 0.011 -0.004 0.93 0.98 -1.47 ln(Age in Years) β 14 0.103 *** 0.054 *** -0.011 *** 4.84 5.03 -4.41 %Institutional Ownership β 15 0.172 *** 0.126 *** 0.010 3.77 4.8 1.62 Big-N Auditor β 16 0.005 0.044 ** 0.005 0.13 2.07 1.14 Fixed Effects Year Yes Yes Yes SIC 2 Industry Yes Yes Yes R 2 0.2437 0.3507 0.4319 N 13,114 7,883 13,998 CAN I BORROW YOUR FOOTNOTES? 96 Table 8 Footnote Similarity Benefits Analysis - 10-K Filing Date Market Reaction Average Market Reaction to 10-K Filing for Footnote Similarity Quintiles Footnote Similarity Quintile Variable 1 (low) 2 3 4 5 (high) |CARi (10-K Filing i)| 0.036 0.041 0.045 0.049 0.052 %Text Match MAXi,j 0.111 0.151 0.179 0.208 0.265 N 2409 2631 2823 2832 2513 |CAR i (10-K Filing i)| Difference in Means Quintiles Compared Difference t-Stat p-value 2 and 1 0.0054 *** 4.15 <.0001 3 and 2 0.0037 *** 2.75 0.006 4 and 3 0.0040 *** 2.94 0.003 5 and 4 0.0033 ** 2.24 0.025 5 and 1 0.0164 *** 11.74 <.0001 CAN I BORROW YOUR FOOTNOTES? 97 Table 9 Footnote Similarity Benefits Analysis -Information Transfer Panel A: Average Cummulative Abnormal Return of Firm i Stock Price Reaction around Firm j Event Footnote Similarity Quintile Variable 1 (low) 2 3 4 5 (high) |CAR Firm i Earn. Annc. Firm j | 0.030 0.033 0.036 0.039 0.042 N 2,075 2,279 2,427 2,479 2,172 |CAR Firm i 10-K Filing Firm j | 0.027 0.031 0.035 0.038 0.041 N 2,073 2,261 2,383 2,404 2,086 %Text Match MAXi,j 0.111 0.151 0.179 0.208 0.264 N 2,241 2,474 2,624 2,676 2,354 |CAR Firm i Earn. Annc. Firm j | Difference in Means Quintiles Compared Difference t-Stat p- value 1 and 2 -0.0026 *** -2.64 0.0084 2 and 3 -0.0038 *** -3.75 0.0002 3 and 4 -0.0024 ** -2.37 0.018 4 and 5 -0.0034 *** -3.06 0.002 1 and 5 -0.0122 *** - 11.19 <.0001 |CAR Firm i 10-K Filing Firm j | Difference in Means Quintiles Compared Difference t-Stat p- value 1 and 2 -0.0036 *** -3.86 0.0001 2 and 3 -0.0043 *** -4.44 <.0001 3 and 4 -0.0025 ** -2.49 0.013 4 and 5 -0.0030 *** -2.75 0.006 1 and 5 -0.0134 *** -12.7 <.0001 CAN I BORROW YOUR FOOTNOTES? 98 Table 9 ! ! Footnote Similarity Benefits Analysis -Information Transfer ! Panel B: Regression of Cummulative Abnormal Return of Firm i around Firm j Event on Footnote Similarity ! Dependent Variable = Dependent Variable = Variables |CAR Firm i Earn. Annc. Firm j | |CAR Firm i 10-K Filing Firm j | Intercept 0.044 *** 0.048 *** 19.01 21.4 Variable of Interest %Text Match MAXi,j 0.031 *** 0.028 *** 4.59 4.29 Control Variables ln(Market Value) -0.002 *** -0.003 *** ! -8.56 -11.42 Book-to-Market 0.000 -0.002 *** ! -0.5 -3.06 %Institutional Ownership -0.007 *** -0.007 *** ! -6.05 -6.45 Price Volatility 0.000 0.000 1.2 0.95 Cash Flow Volatility 0.019 *** 0.012 *** 5.52 3.57 R 2 0.047 0.056 N 10,270 10,270 CAN I BORROW YOUR FOOTNOTES? 99 Table 10 Footnote Similarity Benefits Analysis - Analyst Variables Pooled Regressions of Analyst Variables on Footnote Similarity Analyst Following Analyst Accuracy Analyst Dispersion Model 1 Model 2 Model 3 Coeff. Coeff. Coeff. Variables Coeff. T-Stat T-Stat T-Stat Intercept α " 17.482! *** 0.034! ** 0.086! *** " 10.100! 2.440! 5.240! Variables of Interest ! ! ! %Text Match MAXi,j β 1 3.464! ** "0.084! ** "0.099! ** 2.170! "2.530! "2.010! Control Variables ! ! ! ln(Market Value) β 2 3.032! *** "0.007! *** "0.003! * 37.990! "6.040! "1.820! Leverage β 3 1.063! ** 0.049! *** 0.061! *** 2.260! 4.310! 4.590! R&D over Operating Expense β 4 4.210! *** 0.022! ** 0.116! *** 7.180! 2.460! 5.830! Goodwill β 5 "1.154! * "0.017! "0.097! *** "1.850! "1.620! "8.280! %Institutional Ownership β 6 0.554! * "0.013! ** "0.001! 1.770! "2.130! "0.100! Sales Growth β 7 0.072! ** 0.002! 0.003! * 2.030! 1.270! 1.790! Book-to-Market β 8 1.601! *** 0.060! *** 0.054! *** 10.360! 6.910! 7.500! Price Volatility β 8 0.050! ** 0.002! *** 0.011! *** 2.120! 4.600! 11.090! Fixed Effects Year Yes Yes Yes SIC 2 Industry Yes Yes Yes R 2 0.648 0.1103 0.3154 N 9,681 9,536 8,237 CAN I BORROW YOUR FOOTNOTES? 100 Table 11 Auditor Change and Footnote Similarity Panel A: Descriptive Statistics for auditor change sample (N = 2,440) Mean Std. Dev. 25th Pctl. Median 75th Pctl. Dependent Variable %Text Match MAXi,j 0.161 0.055 0.125 0.157 0.193 Variable of Interest Post Change 0.500 0.500 0.000 0.500 1.000 New Auditor 0.500 0.500 0.000 0.500 1.000 Post Change*New Auditor 0.250 0.433 0.000 0.000 0.500 Control Variables ln(Market Value) 5.184 1.824 3.873 5.075 6.322 ln(Age in Years) 2.518 0.643 2.027 2.478 2.956 ln(Business Segments) 0.516 0.616 0.000 0.000 1.099 ln(Geographical Segments) 0.697 0.675 0.000 0.693 1.099 Return-on-Assets -0.014 0.239 -0.053 0.048 0.105 Leverage 0.120 0.161 0.000 0.034 0.208 R&D over Operating Expense 0.106 0.157 0.000 0.042 0.153 Working Capital over Assets 0.329 0.242 0.148 0.308 0.505 Goodwill 0.107 0.154 0.000 0.023 0.167 Sales Growth 0.239 0.804 0.007 0.101 0.258 Book-to-Market 0.605 0.591 0.264 0.461 0.750 Subsetting Variables Auditor Change (from Big 4 to Big 4) 0.498 0.500 0.000 0.000 1.000 Auditor Change (from non-Big 4 to non-Big 4) 0.077 0.267 0.000 0.000 0.000 CAN I BORROW YOUR FOOTNOTES? 101 Table 11 Auditor Changes Analysis Panel B: Difference in Differences Regression for Auditor Changes Sample Dep. Var. = %Text Match i,MAXj(Old/New Auditor) Variables All Changes Intercept 0.223 *** 36.420 Difference-in-Differences Variables Post Change -0.005 * -1.820 New Auditor -0.025 *** -9.110 Post Change*New Auditor 0.015 *** 3.770 Control Variables ln(Market Value) -0.007 *** -9.860 ln(Age) -0.007 *** -4.330 ln(Business Segmentes) -0.012 *** -7.150 ln(Geographical Segments) -0.001 -0.480 Return-on-Assets 0.029 *** 5.500 Leverage -0.043 *** -6.090 R&D over Operating Expense 0.028 *** 3.320 Working Capital over Assets 0.029 *** 5.630 Goodwill over Assets 0.001 0.210 Sales Growth -0.001 -0.490 Book-to-Market -0.010 *** -5.430 Big N Auditor 0.015 5.940 R 2 0.203 N 2,440 CAN I BORROW YOUR FOOTNOTES? 102 Table 11 Auditor Changes Analysis Panel C: Difference in Differences Regression for Auditor Changes Sample for Pre-Change Highest and Lowest Text Reuse Decile Subsamples Dependent Variable = %Text Match i,MAXj(Old/New Auditor) Pre-Change Footnote Similarity Measure Subsample Auditor Changes Within Same Auditor Type Top Decile Bottom Decile Big N Non-Big N Intercept 0.198 *** 0.062% *** 0.261 *** 0.194 *** 7.52 2.78% 31.44 10.92 Post Change -0.047 *** 0.016% ** -0.012 *** -0.009 -4.07 2.28% -3.27 -1.03 New Auditor -0.064 *** )0.007% -0.012 *** -0.031 *** -5.71 )1.02% -3.26 -3.46 Post Change*New Auditor 0.045 *** 0.003% 0.015 *** 0.023 * 2.81 0.31% 2.88 1.79 Controls Yes Yes Yes Yes R 2 0.207 0.333 0.279 0.340 N 272 124 1,216 188 CAN I BORROW YOUR FOOTNOTES? 103 Table 12 Matching Analysis Panel A: Covariate Balance for Accounting Comparability i,DKV Sample Treated Control Standardized Difference Difference in Means (t-tests) Variance Ratio (treated/ control) Difference in Distributions (KS-tests) Variable Mean Mean p-value B.S. p-value Same MSA Before Matching 0.060 0.019 39.46 <0.0001 After P-Score Matching 0.041 18.399 <0.0001 After Genetic Matching 0.044 15.385 <0.0001 After Entropy Balancing 0.060 <0.0001 0.999 Same Auditor Office Before Matching 0.038 0.008 33.346 <0.0001 After P-Score Matching 0.020 19.556 <0.0001 After Genetic Matching 0.026 13.093 <0.0001 After Entropy Balancing 0.038 <0.0001 0.999 Same SIC 4 Before Matching 0.078 0.044 31.008 <0.0001 After P-Score Matching 0.076 2.213 0.173 After Genetic Matching 0.059 16.88 <0.0001 After Entropy Balancing 0.078 <0.0001 0.999 ln(Market Value) Before Matching 5.2495 6.436 -66.444 <0.0001 0.763 <0.0001 After P-Score Matching 5.271 -1.2 0.524 1.051 <0.0001 After Genetic Matching 5.3815 -7.3915 <0.0001 1.132 <0.0001 After Entropy Balancing 5.2495 -0.0001 1.0000 1.0792 <0.0001 ln(Age) Before Matching 2.399 2.788 -64.236 <0.0001 0.688 <0.0001 After P-Score Matching 2.399 7.099 0.0002 1.0460 <0.0001 After Genetic Matching 2.424 -4.1429 0.0003 1.1614 <0.0001 After Entropy Balancing 2.399 -0.0001 1.0000 1.058 <0.0001 ln(Business Segments) Before Matching 0.365 0.695 -58.863 <0.0001 0.665 <0.0001 After P-Score Matching 0.376 -1.946 0.3080 0.9578 0.814 After Genetic Matching 0.3601 0.81871 0.395 1.021 0.384 After Entropy Balancing 0.365 -0.0001 1.0000 0.995 <0.0001 ln(Geographical Segments) Before Matching 0.624 0.739 -17.23 <0.0001 0.891 <0.0001 After P-Score Matching 0.55 11.061 <0.0001 1.019 <0.0001 CAN I BORROW YOUR FOOTNOTES? 104 After Genetic Matching 0.62036 0.5268 0.513 1.056 0.136 After Entropy Balancing 0.624 -0.0001 1.000 0.917 <0.0001 Return-on-Assets Before Matching -0.0412 0.03388 -27.771 <0.0001 1.900 <0.0001 After P-Score Matching -0.02089 -7.518 <0.0001 1.397 <0.0001 After Genetic Matching -0.0392 -0.747 0.148 1.115 0.006 After Entropy Balancing -0.0412 -0.0001 1.000 1.287 <0.0001 Leverage Before Matching 0.902 0.175 -56.604 <0.0001 0.768 <0.0001 After P-Score Matching 0.086 3.694 0.048 1.078 <0.0001 After Genetic Matching 0.088 2.760 <0.0001 1.048 0.088 After Entropy Balancing 0.902 -0.0001 1.000 1.046 <0.0001 R&D over Operating Expense Before Matching 0.162 0.07932 37.737 <0.0001 1.840 <0.0001 After P-Score Matching 0.17098 -3.926 0.040 1.099 <0.0001 After Genetic Matching 0.15744 2.227 0.0003 1.0240 0.074 After Entropy Balancing 0.16234 <0.0001 1.000 0.854 <0.0001 Working Capital over Assets Before Matching 0.384 0.24456 55.236 <0.0001 1.282 <0.0001 After P-Score Matching 0.42114 -14.681 <0.0001 0.916 <0.0001 After Genetic Matching 0.38826 -1.660 0.037 1.137 0.005 After Entropy Balancing 0.384 <0.0001 1.000 0.940 <0.0001 Goodwill over Assets Before Matching 0.106 0.121 -10.200 <0.0001 1.056 <0.0001 After P-Score Matching 0.100 3.466 0.090 1.051 <0.0001 After Genetic Matching 0.101 2.816 0.040 1.104 0.23 After Entropy Balancing 0.106 <0.0001 1.000 0.943 <0.0001 Sales Growth Before Matching 0.387 0.225 9.681 <0.0001 2.289 <0.0001 After P-Score Matching 0.420 -1.988 0.332 1.070 <0.0001 After Genetic Matching 0.327 3.572 <0.0001 1.507 0.028 After Entropy Balancing 0.38655 <0.0001 1.000 1.167 <0.0001 Book-to-Market Before Matching 0.566 0.587 -4.062 0.055 0.946 <0.0001 After P-Score Matching 0.542 4.486 0.027 1.142 <0.0001 After Genetic Matching 0.555 1.945 0.001 1.178 <0.0001 After Entropy Balancing 0.56552 <0.0001 1.000 1.357 <0.0001 Cash Flow Volatility Before Matching 0.093 0.05933 23.729 <0.0001 2.835 <0.0001 After P-Score Matching 0.09951 -4.995 0.010 1.282 <0.0001 After Genetic Matching 0.089 2.588 0.120 1.391 0.248 After Entropy Balancing 0.09252 <0.0001 1.000 1.391 <0.0001 CAN I BORROW YOUR FOOTNOTES? 105 Cash Flow Comovement Before Matching 0.607 0.572 29.028 <0.0001 0.808 <0.0001 After P-Score Matching 0.612 -3.853 0.054 1.025 <0.0001 After Genetic Matching 0.605 2.052 0.047 1.217 <0.0001 After Entropy Balancing 0.607 <0.0001 1.000 0.880 <0.0001 Big-N Auditor Before Matching 0.749 0.813 -14.938 <0.0001 After P-Score Matching 0.755 -1.500 0.460 After Genetic Matching 0.737 2.750 0.039 After Entropy Balancing 0.749 <0.0001 1.000 Number of Observations N N Before Matching 4,609 4,538 After P-Score Matching 4,609 4,609 After Genetic Matching 4,609 4,609 After Entropy Balancing 4,609 4,609 CAN I BORROW YOUR FOOTNOTES? 106 Table 12 Matching Analysis Panel B: Propensity Score Model Dependent Variable = Treated Sample: Accounting Comparability i,DKV Sample: Accounting Comparability i,DD Sample: Earnings Comovement i Coeff. Coeff. Coeff. Variables Wald Chi-Square Wald Chi-Square Wald Chi-Square Intercept 0.685 ** 0.384 0.413 5.055 0.256 2.227 MSA & Auditor Office Variables Same SIC 4 2.558 *** 2.665 *** 2.841 *** 145.532 94.202 183.279 Same MSA 2.958 *** 3.244 *** 3.144 *** 77.611 52.153 86.806 Same Auditor Office 7.403 *** 5.226 *** 7.682 *** 59.757 19.881 61.036 Same MSA*Same SIC 4 -0.837 -1.095 -1.238 ** 2.454 2.343 5.541 Same Auditor Office*Same SIC 4 -6.186 *** -4.315 *** -6.393 *** 32.639 9.962 33.305 Same Auditor Office*Same MSA -3.346 ** -0.306 -3.876 *** 5.775 0.025 7.566 Same Auditor Office*Same MSA*Same SIC 4 4.402 *** 1.708 4.721 *** 7.643 0.609 8.637 Firm Complexity ln(Market Value) -0.223 *** -0.250 *** -0.226 *** 353.767 253.094 381.700 ln(Age) -0.208 *** -0.235 *** -0.193 *** 63.821 49.016 57.811 ln(Business Segments) -0.192 *** -0.225 *** -0.170 *** 54.956 44.373 44.799 ln(Geographical Segments) -0.186 *** -0.177 *** -0.196 *** 50.530 27.015 59.150 Financial Characteristics Return-on-Assets 0.473 *** 0.636 *** 0.527 *** 29.842 20.674 38.228 Leverage -0.705 *** -0.761 *** -0.625 *** 42.057 28.889 35.469 R&D over Operating Expense -0.095 -0.063 -0.086 CAN I BORROW YOUR FOOTNOTES? 107 0.594 0.130 0.494 Working Capital over Assets 0.900 *** 1.055 *** 0.982 *** 106.606 76.215 132.591 Goodwill over Assets 0.458 *** 0.684 *** 0.442 *** 13.759 19.215 13.628 Sales Growth 0.020 * 0.009 0.013 3.048 0.272 1.397 Book-to-Market -0.201 *** -0.206 *** -0.211 *** 35.762 23.997 45.552 Cash Flow Volatility -0.089 -0.672 ** -0.077 0.294 4.009 0.231 CFO Comovement 0.292 ** 0.253 0.390 *** 4.428 1.894 8.772 Big-N Auditor 0.149 *** 0.085 0.113 *** 12.572 2.335 7.423 Fixed Effects Year Yes Yes Yes SIC 2 Industry Yes Yes Yes R 2 0.432 0.448 0.547 N 9,147 7,883 9,796 CAN I BORROW YOUR FOOTNOTES? 108 Table 12 Matching Analysis Panel C: Accounting Comparability Difference in Means Analysis Propensity Score Matching Genetic Matching Entropy Balancing Accounting Comparability i,DKV Mean treated -0.581 -0.583 -0.581 Mean control -0.767 -0.660 -0.951 Mean treated - Mean control 0.186 *** 0.077 ** 0.370 *** t-Stat. 6.02 2.313 6.104 N 9,218 9,218 9,218 Accounting Comparability i,DD Mean treated -0.642 -0.642 -0.642 Mean control -0.698 -0.694 -0.805 Mean treated - Mean control 0.055 *** 0.051 ** 0.162 *** t-Stat. 4.22 2.389 3.140 p-value <.0001 0.017 0.002 N 5,584 5,594 5,594 Earnings Comovement i Mean treated 0.676 0.676 0.676 Mean control 0.660 0.662 0.647 Mean treated - Mean control 0.016 *** 0.015 *** 0.029 t-Stat. 5.41 2.965 1.397 p-value <.0001 0.0030 0.163 N 9,844 9,858 9,858 CAN I BORROW YOUR FOOTNOTES? 109 Table 12 Matching Analysis Panel D: Accounting Comparability and Footnote Similarity Regression Analysis ! Propensity Score Matching Sample Genetic Matching Sample Entropy Balancing Sample ! Coeff. Coeff. Coeff. !! T-Stat T-Stat T-Stat Dependent Variable = Accounting Comparability i,DKV ! Intercept -1.011 *** -0.853 *** -1.445 *** ! -11.935 -10.629 -9.99 ! Treated i 0.168 *** 0.097 *** 0.356 *** ! 8.467 5.303 8.022 ! Control Variables Yes Yes Yes ! R 2 0.202 0.186 0.217 ! N 9,218 9,218 9,218 Dependent Variable = Accounting Comparability i,DD ! ! ! ! Intercept -1.255 *** -1.172 *** -1.396 *** ! 26.786 -25.773 -12.979 ! Treated i 0.069 *** 0.063 *** 0.142 *** ! 6.193 5.624 5.081 ! Control Variables Yes Yes Yes ! R 2 0.329 0.283 0.297 ! N 5,584 5,594 5,594 Dependent Variable = Earnings Comovement i ! ! ! ! ! Intercept 0.258 *** 0.294 *** 0.261 *** ! 28.069 31.859 18.507 ! Treated i 0.016 *** 0.011 *** 0.035 *** ! 5.979 4.165 7.489 ! Control Variables Yes Yes Yes ! R 2 0.267 0.249 0.297 !! N 9,844 9,858 9,858

Abstract (if available)

Abstract Regulators and the profession have long complained that “boilerplate” footnotes impair financial reporting quality, where boilerplate refers to standardized text that is similar across firms. They raise these concerns without acknowledging that the use of boilerplate may create learning and network externalities that benefit financial statement preparers and users. One channel through which boilerplate footnotes create these externalities is by affecting accounting comparability, although it is unclear whether similarity increases or decreases comparability. I investigate this question by examining the association between accounting comparability, measures of network benefits and a measure of footnote similarity derived from text re‐use detection software. I begin by documenting the factors that explain footnote similarity, and find that similarity increases among firms that are in the same industry, are geographically proximate, and share the same auditor office. In answer to my primary question, I find that footnote similarity is positively associated with accounting comparability and measures of network benefits, consistent with boilerplate footnotes improving financial reporting quality, on average. In addition, the results are robust to alternative measures of accounting comparability and to several matching methods, some of which are new to the accounting literature. These findings contribute to our understanding of the role of footnotes in improving financial reporting quality, and provide new insights into the footnote creation process. This study also adds to the growing textual analysis literature by introducing the use of text re‐use detection software to measure footnote similarity, and to the general accounting literature by introducing two new matching methods.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Who manages the firm matters: the incremental effect of individual managers on accounting quality

PDF

Does fair value accounting exacerbate the pro-cyclicality of bank lending?

PDF

Accrual quality and expected returns: the importance of controlling for cash flow shocks

PDF

Private disclosure contracts within supply chains and managerial learning from stock prices

PDF

Effectiveness of the SEC’s comment letters in initial public offerings

PDF

Does mandatory adoption of International Accounting Standards reduce the cost of equity capital?

PDF

Real asset liquidity and asset impairments

PDF

Does mandatory adoption of international financial reporting standards decrease the voting premium for dual-class shares: theory and evidence

PDF

Share repurchases: how important is market timing?

PDF

Protecting networks against diffusive attacks: game-theoretic resource allocation for contagion mitigation

PDF

Insider editing on Wikipedia

PDF

The effects of accounting performance and professional relationships on promotion, dismissal, and transfer decisions in a conglomerate

PDF

Executive compensation: the trend toward one size fits all

PDF

CEO reputation: who benefits -- the firm and the CEO?

PDF

Automated contracts and the lawyers who don't review them: adoption and use of machine learning technology

PDF

Essays on the firm and stakeholders relationships: evidence from mergers & acquisitions and labor negotiations

PDF

Essays on the role of entry strategy and quality strategy in market and consumer response

PDF

Leveling the playing field: unbiased tests of the relative information content of book income and taxable income

PDF

The structure of strategic communication: theory, measurement, and effects

PDF

Competing across and within platforms: antecedents and consequences of market entries by mobile app developers

Asset Metadata

Creator McMullin, Jeff Lawrence (author)

Core Title Can I borrow your footnotes? Learning and network benefits of footnote similarity

School Marshall School of Business

Degree Doctor of Philosophy

Degree Program Business Administration

Degree Conferral Date 2014-05

Publication Date 02/06/2014

Defense Date 11/22/2013

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag auditor office,boilerplate,Disclosure Framework project,footnote similarity,geography,OAI-PMH Harvest,plagiarism detection

Format application/pdf (imt)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor DeFond, Mark L. (committee chair), McCubbins, Mathew D. (committee member), Murphy, Kevin J. (committee member), Subramanyam, K.R. (committee member), Zhang, Jieying (committee member)

Creator Email mcmullin.jeff@gmail.com

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c3-361968

Unique identifier UC11296468

Identifier etd-McMullinJe-2240.pdf (filename),usctheses-c3-361968 (legacy record id)

Legacy Identifier etd-McMullinJe-2240.pdf

Dmrecord 361968

Document Type Dissertation

Format application/pdf (imt)

Rights McMullin, Jeff Lawrence

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA