Should Management Earnings Guidance be Pooled? Classifying Management Earnings Guidance into Capital Market, Opportunistic, and Disclose or Abstain Rationales: Implications for Research Edward Li
[email protected] Charles Wasley
[email protected] Jerold Zimmerman
[email protected] Simon School of Business University of Rochester
ABSTRACT Prior studies identify several motives for why firms release management earnings forecasts (MFs). A common feature of such studies is they pool MFs when drawing inferences about a specific motive. By ignoring the heterogeneous rationales managers have to issue MFs, pooling could lead to biased inferences. To address the issue, we develop an approach that classifies MFs into one of the three rationales: capital market incentives, compliance with Rule 10b-5 to disclose material nonpublic information or abstain from trading, or managerial opportunism. Our classification scheme indicates that 63% of MFs are released to lower the firm’s cost of capital, 23% are issued to comply with Rule 10b-5, and 14% are opportunistic. Four sets of tests provide construct validity of our classification scheme. These include replications of earlier studies where we find that our MF classification scheme increases power and changes prior inferences regarding MFs. Classifying MFs into our three MF categories will aid future researchers in constructing better specified and more powerful tests of the economic determinants and consequences of management’s decision to issue MFs. Current draft: January 24, 2012 We gratefully acknowledge the financial support provided by the Simon School at the University of Rochester and comments from Francois Brochet, Bill Cready, Rebecca Hann, Shane Heitzman, Chris Noe, Ed Owens, Stan Markov, Sanjog Misra, Joanna Wu, and seminar participants at KAIST, Seoul National University, Syracuse University, University of Pittsburgh, University of Texas at Dallas, the 2010 Conference on Financial Economics and Accounting at the University of Maryland, the 2010 HKUST Research Conference, the 2011 Tel Aviv University Accounting Conference, and the 2011 Journal of Contemporary Accounting & Economics Symposium.
1. Introduction The disclosure literature recognizes several different motives to explain why managers provide earnings forecasts (hereafter, MFs). These are: (i) capital market incentives that increase market liquidity and lower the firm’s cost of capital by reducing information asymmetry (see, e.g., Diamond and Verrecchia 1991, Ajinkya and Gift 1984, Graham et al. 2005, Frankel et al. 1995 and Coller and Yohn 1997); (ii) to allow managers to trade opportunistically in their firm’s securities (see, e.g., Noe 1999, Rogers and Stocken 2005, and Cheng and Lo 2006); (iii) to comply with insider trading regulations under Rule 10b-5 requiring managers with material private information to disclose or abstain from trading (DOA) (see, e.g., Loss and Seligman 2004, Cheng and Lo 2006, Rogers 2008, and Heitzman et al. 2010); and (iv) to reduce expected litigation costs (see, e.g., Skinner 1994 and Kasznik and Lev 1995). With regard to the third motive, while the securities laws are clear that if insiders want to trade in their firm’s securities they must first disclose any material nonpublic information in their possession, securities regulations are far from clear as to what constitutes material nonpublic information. The lack of such clear guidance suggests that MFs are a mechanism for managers to disclose material inside information prior to trading to comply with the securities laws. For the purpose of this study we define MFs as disclosures issued before the last three weeks of the corresponding fiscal period end. “MFs” refer collectively to “management earnings forecasts,” “management guidance,” and “management forecasts.” Disclosures issued late in a period (“warnings”) or after the period end, but before the earnings announcement (“preannouncements”) are viewed as disclosures designed to reduce a firm’s expected class action damages (iv. above). We focus on the first three motives and exclude from our tests all disclosures that are issued in the last three weeks before the end of the quarter and through the actual earnings announcement date. We do so because managers’ rationale to issue “warnings” and “preannouncements” are easy to identify, they are designed to reduce expected class action damages, and such announcements constitute less than 10% of all earnings-related forecasts. While the disclosure literature recognizes that managers face various incentives to issue guidance, most studies pool all MFs together in a single sample which implicitly assumes that just one particular motive drove managers’ disclosure decision. For example, in testing whether MFs increase transparency and lower firms’ cost of capital, researchers assume all MFs are issued for that purpose and draw conclusions based on the pooled MF sample. However, this 1
approach ignores important heterogeneity in the origins of MFs because a substantial number of MFs could be released for opportunistic or 10b-5 compliance reasons. Pooling MFs creates at least two problems. First, including MFs that are not issued to lower the firm’s cost of capital reduces power, which not only results in lower significance levels, but more seriously could lead to a spurious “no result” when an effect actually exists in the sample. Second, inclusion of MFs issued for 10b-5 compliance reasons introduces a correlated omitted-variables problem that can bias the inferences drawn if variables used to measure cost of capital effects also capture managers’ 10b-5 compliance incentives. Our study addresses the question “Should Guidance Be Pooled?” More specifically, we point out and provide supporting empirical evidence that future MF studies can construct more powerful tests of their hypotheses by focusing specifically on MFs issued for the particular motive the researcher is addressing (e.g., to lower the firm’s cost of capital or to comply with 10b-5 requirements). Before describing how we address our research question, some discussion of terminology is warranted. “COC (cost of capital) MFs” to refer to MFs issued by managers because they want to increase transparency by lowering information asymmetry and thereby satisfy investors’ and analysts’ demand for information. Managers issue “COC MFs” to increase market liquidity which facilitates access to capital markets and lowers a firm’s cost of capital. “DOA (disclose or abstain) MFs” refers to MFs issued because managers wish to trade in their firm’s securities and want to comply with Rule 10b-5, and therefore must “disclose or abstain (DOA)” from trading. Finally, “OPP (opportunistic) MFs” refers to MFs issued with the purpose of opportunistically transferring wealth from shareholders to managers. For example, managers sell (buy) shares prior to the release of bad (good) news MFs, or issue optimistic (pessimistic) MFs prior to selling (buying) shares. While all three types of MFs are “voluntary” in the sense that managers have discretion over whether or not to issue them, COC MFs and DOA MFs increase firm value, but in different ways. COC MFs increase firm value by lowering the firm’s cost of capital (i.e., denominator effect), whereas DOA MFs increase expected future cash flows by reducing expected insider trading sanctions (i.e., numerator effect).1 To address the question “Should Guidance Be Pooled?” we must first devise a method to classify MFs into mutually exclusive subsets. We do so by using properties of the data which are
1
Since COC MFs can lower the transaction costs associated with of issuing new capital they can also increase expected future cash flows. 2
designed to capture the rationales for why managers issue MFs. While managers gain utility from issuing MFs and/or trading in their firm’s stock, a manager’s utility is not directly known or observable. Thus, decisions to issue a DOA MF and/or trade in their firm’s securities are endogenous and inter-related and researchers cannot observe whether incentives to reduce the cost of capital, compliance with rule 10b-5, or opportunism is driving the decision to issue any particular MF. For example, if a manager withholds material information or releases a biased MF and then trades, the manager’s MF is ex ante opportunistic (“biased” MFs are those differing from the manager’s private information). However, if a manager discloses an unbiased COC MF, observes the market reaction, and then trades because of his/her conjecture that the market mispriced the news, the manager is behaving opportunistically, but the opportunism is ex post. This is because he/she faithfully (unbiasedly) represented his/her private information via a COC MF and by trading is simply conjecturing the market is wrong. While it is straight-forward to describe these alternative MF scenarios conceptually it is difficult (if not impossible) to empirically identify ex ante opportunism from ex post opportunism because one cannot observe manager’s private information or biased disclosures except in rare cases of admitted fraud. Not knowing managers’ exact motives to issue MFs necessitates an indirect approach to classify each MF as COC, DOA, or OPP. We first classify MFs into three mutually exclusive samples based on insider trading around the release of the MF, characteristics of the MF, the market reaction to the MF, and the MF forecast error (details later). Each sample is constructed to contain MFs most likely to have been issued based on one of the three motives we study. Using a sample of 30,876 MFs from the First Call CIG database from 1998-2010, we classify 63% as COC MFs, 23% as DOA MFs, and 14% as OPP MFs. Thus, while 63% of MFs are classified as issued to reduce cost of capital a substantial portion (37%) is not. We provide four sets of empirical tests to validate our MF classification scheme. First, we estimate a multinomial probit model of MF disclosure in a given firm quarter using variables designed to capture the cost of capital, DOA, and opportunistic incentives to issue MFs.2 If our MF classification scheme has construct validity then the variables designed to explain MFs issued for cost of capital, DOA, and opportunistic reasons will have greater explanatory power in their respective sub-sample than in the other sub-samples. As predicted, variables assumed to
2
Multinomial probit is widely used in economics and marketing to model multinomial choice problems (see, e.g., Hausman and Wise 1978, McCulloch et al. 2000 and Haaijer et al. 2000). 3
measure cost of capital, DOA, and opportunistic disclosure incentives do have greater explanatory power in their respective MF sub-sample than in the other sub-samples. Second, we replicate Cheng and Lo (2006) and find that one of their two primary inferences (i.e., managers increase the frequency of bad news MFs prior to insider purchases) holds only in the subsample of MFs that we classify as OPP MFs, but not for MFs classified as COC MFs or DOA MFs. This finding is consistent with Cheng and Lo’s (2006) hypothesis that “insiders strategically choose disclosure policies and the timing of their equity trades to maximize trading profits,” (p. 815), and it adds construct validity to our MF classification scheme. More importantly, using our OPP MF sample we overturn the other primary inference in their paper. Specifically, while Cheng and Lo (2006) conclude that managers do not adjust forecasting activity before selling shares due to litigation concerns, we find that managers do increase the frequency of good news MFs prior to insider sales. Third, we find that insider trading associated with DOA and opportunistic MFs exhibits differential predictive ability for future stock returns. If OPP MFs capture those issued by managers motivated to trade opportunistically in their firm’s stock, then such insider trading should have predictive ability for future stock returns. On the other hand, if managers issue DOA MFs to disgorge their material nonpublic information before trading, such insider trading should not predict future stock returns. The results support these predictions. Our fourth and final set of construct validity tests is based on replicating Ajinkya et al. (2005). There we find that their key results vary in predictable ways across our MF categories. Our findings have implications for the design of future MF studies as well as for the interpretation of prior research using MFs to test cost of capital disclosure incentives. First, since roughly 37% of MFs do not appear to be made to reduce firms’ cost of capital, studies treating all MFs as disclosures designed to reduce the cost of capital misclassify roughly a third of their sample, potentially leading to a loss of power. A loss of power does not call into question prior published papers documenting significant findings. However, published and unpublished MF studies that failed to document significance (and future studies) could be enhanced by limiting their samples to only those MFs (i.e., COC, DOA, or OPP) that are the particular focus of their research question. Second, prior MF papers implicitly assume the independent variables used to predict the issuance of MFs are only capturing cost of capital incentives. A problem, however (discussed later), is that variables used to proxy for cost of 4
capital incentives also capture managerial incentives to issue MFs to satisfy DOA requirements. This calls into question the reliability of the inferences drawn in prior MF studies about the predicted effects of cost of capital variables. While our MF classification scheme captures managers’ unobservable incentives to issue MFs, we acknowledge that it invariably produces some misclassifications. If misclassifications are random, they reduce the power of our construct validity tests. If they are not random, our inferences about the relative importance of the competing motives to issue MFs will be affected. This limitation is not unique to our setting. Other research areas such as earnings quality, conservatism, information content of earnings, etc., all require measures of unobservable constructs. The next section describes the relevant literatures, our MF classification scheme, the sample, and the empirical tests.
2. Literature review This section summarizes three literatures related to our study: securities regulation requiring managers to disclose material information or abstain from trading, studies of management forecasts, and papers examining the relation between MFs and insider trading. 2.1. Management’s affirmative duty to disclose material information or abstain from trading In general, managers have no affirmative duty to disclose material information or events as they occur under the securities laws unless “(1) a Commission statute or rule requires disclosure, (2) an ‘insider’ …is trading, or (3) a previous disclosure is or becomes inaccurate, incomplete, or misleading” (see Loss and Seligman 2004, p. 3510-3511). However, if insiders want to trade, Rule 10b-5 prohibits them from doing so unless they first disclose any material private information. Simply put, insiders must either disclose that information or abstain from trading (DOA).3 The firm has a legal duty to disclose material nonpublic information whenever it suspects that persons with access to material inside information may be trading in the company's stock (see Block et al. 1985). Hence, even though other managers within the firm such as divisional vice presidents do not have direct control over the decision to issue a MF,
3
The SEC routinely brings between 40 to 60 insider trading actions annually (see sec.gov/spotlight/insider trading. shtml). Besides the SEC’s insider trading enforcement, FINRA (Financial Industry Regulatory Authority) oversees U.S. stock exchanges using insider trading surveillance and investigation programs, and the Department of Justice (DOJ) pursues criminal cases. 5
knowing that such managers have material nonpublic information and plan to trade creates a legal duty for the firm to disclose the material information to comply with Rule 10b-5.4 The securities laws and SEC regulations offer little clear-cut guidance as to what constitutes material nonpublic information (see Prentice, 1999). While Staff Accounting Bulletin 99 and Regulation Full Disclosure (Reg FD) list types of information that may help managers determine materiality, Heminway (2003) argues that: “the imprecise existing legal standard defining what is ‘material’ make it difficult for those issuers, directors and officers to understand their legal obligations.” Heminway (2003) observes that while “section 10(b) and Rule 10b-5 are antifraud provisions, not mandatory disclosure rules – i.e., rules calling for specific substantive disclosure based on a transactional or periodic reporting requirement. … In the insider trading context, however, the ‘disclose or abstain’ rule has the effect of a macro mandatory disclosure rule. The existence of a duty to disclose arising out of issuer or insider securities trading activity compels disclosure of all material nonpublic information before any trade is made. … Accordingly, it is appropriate to refer to the regulation of insider trading under Rule 10b-5 as a form of disclosure regulation (emphasis added).” We argue that insiders wishing to comply with Rule 10b-5 will likely issue MFs prior to trading in their firm’s stock. Moreover, risk-averse insiders will set a low materiality threshold resulting in the disclosure of some immaterial MFs. 2.2. Research on MFs5 Managers’ incentives to disclose MFs include signaling good firm performance, reducing litigation risk, facilitating access to capital markets and reducing the cost of capital, and adjusting analysts’ and investors’ expectations (see, e.g., Coller and Yohn 1997; Frankel et al. 1995; Verrecchia and Weber 2006; Beyer et al. 2010). The theoretical disclosure literature predicts that when firms bear proprietary costs from disclosure, or when investors are uncertain about managers’ information, managers will disclose good news and withhold unfavorable news (Verrecchia 1983; Dye 1985 a and b; Verrecchia 2001, Evans and Sridhar, 2002). Early studies on MFs are consistent with these predictions (see, e.g., Penman 1980; Lev and Penman 1990). 4
Most firms have insider trading policies: prohibiting all employees from trading in the firm’s securities while in the possession of material nonpublic information, requiring pre-clearance for trades by officers and directors, and establishing normal trading windows. Bettis et al. (2000) report that roughly 75% of their 626 sample firms have policies where insider trades must be approved by the company. 5 The MF literature is vast so we make no attempt to discuss it in detail. We restrict attention to papers most closely related to our study (see Healy and Palepu, 2001, Hirst et al, 2008, and Beyer et al, 2010 for reviews). 6
Litigation risk has also been advanced as an explanation for why managers issue MFs. Healy and Palepu (2001, p.422-23) conjecture that litigation risk can affect managers’ disclosure choices in two opposing ways: “first, legal actions against managers for inadequate or untimely disclosures can encourage firms to increase voluntary disclosure. Second, litigation can potentially reduce managers’ incentives to provide disclosure, particularly of forward-looking information.” Consistent with the former, Skinner (1994) finds that MFs are more likely to preempt bad news earnings surprises to reduce expected shareholder litigation cost. Healy and Palepu’s second conjecture (i.e., litigation reduces managers’ incentives to provide forwardlooking information) is based on the premise that managers fear being (legally) penalized for forward-looking information (e.g., MFs) made in good faith that is inaccurate ex post. While this conjecture may be true, it ignores the fact that DOA rules require managers to release forwardlooking information (e.g., MFs) prior to trading in their firm’s securities.6 The DOA rationale for managers’ to issue MFs is not the same as the traditional litigation risk hypothesis examined in the literature (e.g. Skinner, 1994). Under the usual litigation risk explanation, managers issue MFs to preempt significant bad news earnings surprises to reduce the expected cost of shareholder lawsuits that would be triggered by the potential price plunge in the absence of a prior MF. Under DOA, managers issue MFs prior to trading in their firm’s securities to avoid SEC investigations and sanctions, and such MFs are not restricted to just conveying significant bad news as in Skinner (1994). In fact, to avoid allegations of insider trading the DOA motive for releasing a MF would predict that managers will disclose their private information through MFs, regardless of the sign of the news and the direction of the transaction. Moreover, as it is more likely to establish insiders’ personal gains when there are actual trades, managers would be more sensitive to the materiality level of the news and potentially lower their disclosure materiality threshold. Rogers (2008) recognizes that managers wishing to trade have a legal duty to disclose or abstain from trading. He examines three corporate disclosures (MFs, conference calls, and IPO press releases) and finds evidence that managers’ incentives affect the quality of information provided to market participants. In particular, managers provide higher quality disclosures prior to selling shares than are provided in the absence of trading. Rogers (2008) also reports lower 6
Interestingly, Skinner (1994, p. 56) recognizes (but does not pursue in detail) the possibility that the earningsrelated disclosures in his study might not be solely to reduce expected shareholder litigation costs, but rather might be disclosures to comply with SEC requirements that firms disclose material nonpublic information. 7
quality disclosures prior to purchasing shares than in the absence of trading, presumably to maintain their information advantage. While Rogers (2008) recognizes all three motives for managers to issue MFs (COC, OPP, and DOA), his focus differs from ours. He examines how the various incentives affect disclosure quality (as measured by the market reaction to the disclosure), but does not test how the various incentives affect the decision to issue a MF, which is the focus of our paper. The empirical MF literature typically assumes that firms make separate and independent guidance decisions each period (see, e.g., Hirst et al. 2008 and Tang 2011). Leuz and Verrecchia (2000, p. 94) distinguish between a “disclosure commitment” and a “disclosure” in that “the former is a decision by the firm about what it will disclose before it knows the content of the information (i.e., ex ante), whereas the latter is a decision by the management made after it observes the content (i.e., ex post).” Presumably, firms adopt ex ante MF commitment policies to reduce information asymmetry and lower the firm’s cost of capital (see Core 2001). However, managers do not disclose the existence of ex ante MF commitment policies and the literature does not provide a generally accepted method to identify these ex ante commitment MFs (Tang, 2011). To enhance comparability with prior MF research our main tests do not differentiate between ex ante MF commitment policies and ex post MFs. We recognize that ex ante commitments to issuing MFs may be those firms where disclosure theory more strongly predicts reductions in information asymmetry, and hence potential cost of capital benefits. The inclusion of both ex ante policy-determined and ex post non-policy driven MFs in our COC MF sample might reduce the power of our tests, but imparts no obvious bias. As a sensitivity test we employ Tang’s (2011) method to remove ex ante commitment policy-driven MFs from our sample and find that our inferences are robust (see section 5.3.5). 2.3. Research jointly examining MFs and insider trading Noe (1999) and Cheng and Lo (2006) address the relation between insider trading and MFs. Both document that managers buy more shares following bad news MFs than after good news MFs, and sell more share after good news MFs than after bad news MFs, results that both studies interpret as evidence of managers opportunistically timing their insider trading vis-à-vis their MF disclosure. Noe (1999) studies insider trading before and after the release of MFs to investigate whether managers are opportunistic ex ante in the sense that they trade to exploit the information revealed by the subsequent MF. The evidence is inconsistent with such ex ante 8
opportunism. Noe (1999) also investigates how managers trade after the release of MFs and finds they generally behave like contrarians – they buy if the stock price reaction to the MF is negative or sell if the reaction is positive. Noe (1999) focuses exclusively on whether managers behave strategically when they trade before or after MFs, which means he does not consider that managers could be issuing MFs to comply with DOA rules. Noe’s (1999) finding of significant trading following MFs is consistent with managers’ issuing MFs to comply with DOA rules. Under this interpretation, insiders trade after disclosing a MF because it means they have disclosed their material information prior to trading as required by rule 10b-5.7 Cheng and Lo (2006) treat MFs as a strategic decision by managers to trade opportunistically in their firm’s stock. They find that managers increase the frequency of bad news MFs when purchasing shares (presumably to reduce the purchase price), but do not alter their MF forecasting behavior when they are selling. Cheng and Lo (2006) mention DOA requirements, but argue that their findings are inconsistent with managers violating such rules. Unlike our study, however, Cheng and Lo (2006) do not comprehensively consider DOA requirements, COC incentives, and OPP managerial behavior as alternative rationales for managers’ issuance of MFs.8 2.4. Summary The literature reviewed above leads to the following conclusions. First, the literature has identified three primary reasons why managers issue MFs: cost of capital incentives, their affirmative “disclose or abstain” duty under the securities laws, and to opportunistically time their insider trading vis-à-vis the disclosure of their MFs. Second, while the securities laws are clear that if insiders want to trade in their firm’s securities they must first disclose any material nonpublic information, the regulations are far from clear as to what constitutes material nonpublic information. The lack of such guidance suggests that MFs are a mechanism for managers to disclose material inside information prior to trading to comply with the securities laws. In the next section, we develop a method to sort MFs into one of three mutually exclusive
7
Another interpretation of Noe’s (1999) results is managers are strategic ex post, but not ex ante in that they issue a MF to reduce cost of capital, and only decide to trade after they observe the market response to the MF. Here, the ex ante act of issuing the MF could be unrelated to the ex post decision to trade. 8 Other papers in the opportunistic category are Aboody and Kasznik (2000) (CEOs opportunistically time MFs to maximize stock option compensation) and Rees et al. (2008) which finds managers provide more pessimistic guidance prior to stock option awards than afterwards. 9
categories based on the manager’s likely incentives (COC, DOA, or OPP) and then describe how we validate this classification scheme. 3. Classifying MFs based on manager’s incentives To address the question (“Should Guidance Be Pooled?”) we need a method to sort MFs. The classification scheme is driven by the three explanations offered in the literature and discussed in section 2 for why managers issue MFs (cost of capital incentives, compliance with insider trading regulations, and opportunistic insider trading incentives). We begin by organizing MFs into three mutually exclusive and exhaustive samples based on properties of MFs and insider trading patterns (details in section 3.1). These samples are: (i) MFs where, a priori, the data suggest they were issued to lower the firm’s cost of capital (COC), (ii) MFs where, a priori, the data suggest that managers are likely to have had an affirmative duty to disclose material information (DOA), and (iii) those MFs where, a priori, the data suggest they were issued because the manager traded opportunistically (OPP). As in numerous other areas in the accounting literature where managers’ true motives are unobservable we must develop variables to proxy for the alternative explanations for why managers issue MFs.9 3.1
MF classification scheme: MFs issued to lower the firm’s cost of capital (COC MFs) Disclosure theory predicts that MFs are issued to improve capital market transparency,
thereby lowering a firm’s cost of capital. Candidates for MFs to include in the COC MF sample are those issued without any accompanying insider trading before or after their release. The absence of insider trading excludes (i) MFs issued for DOA reasons (because managers did not subsequently trade) and (ii) opportunistically issued MFs (so that managers could trade). Beyond MFs without insider trading before or after the MF, bad news (good news) MFs whose disclosure is preceded by insider purchasing (selling) are also more likely to have been issued to lower the firm’s cost of capital. Such MFs are not opportunistic because the buying and selling pattern would have to be reversed to be opportunistic. Such MFs also are unlikely to have been issued for DOA reasons because insider trading was before (not after) the MF’s release. Using this intuition we identify COC MFs using the following criteria: Ca: No insider trading in the 61-day window centered on the MF release date, OR 9
Settings where researchers must develop proxy variables for managers’ unobservable motives include earnings management (proxies for discretionary accruals or meet or beat benchmarks), the conservatism literature (a proxy for conservatism is required), and traditional earnings/return studies (proxies for expected earnings and expected stock returns are required). 10
3.2
Cb: Insider purchasing in the 30 days prior to release of bad news MFs (abnormal returns at MF release date < 0) or insider selling in the 30 days prior to release of good news MFs (abnormal returns at MF release date > 0). MF classification scheme: MFs issued to satisfy “disclose or abstain” requirements
(DOA MFs) Managers issue DOA MFs because they believe they are in possession of material nonpublic information and want to trade for personal liquidity reasons, to diversify, or to accumulate shares. Liquidity reasons include paying college tuition bills, income taxes, divorce settlements, buying homes, or exercising options (to pay the exercise price and taxes). Managers may also want to purchase shares because they believe the stock is undervalued and they want to credibly signal undervaluation or to meet share ownership guidelines imposed by their firm.10 To comply with securities laws managers have an affirmative duty to disclose material information prior to trading. Managers only must disclose material nonpublic information before they trade. The more “material” a manager’s private information, the more likely the manager perceives he/she has an affirmative duty to disclose, and hence is more likely to issue a MF. MFs issued to comply with 10b-5 sanctions may have large (positive or negative) stock price effects. However, as noted in section 2.1, there is little clear-cut guidance in the securities laws or SEC regulations as to what constitutes material information. This may lead risk-averse managers to set a low materiality threshold and disclose some DOA MFs that trigger small stock price effects. Based on the requirements underlying DOA rules, all MFs followed by insider trading are candidates for inclusion in the DOA sample. However, as discussed in detail below, some MFs that are followed by insider trading can reliably be classified as opportunistic (see steps Ob and Oc below). MFs in the DOA sample consist of those that are followed by insider trading not otherwise classified as opportunistic. MFs preceded by insider trading are not candidates for the DOA MF sample because DOA requires disclosure before not after trading. We assign MFs to the DOA sample based on the following criteria: Da: All MFs with insider trading (i.e., either selling or buying) in the 30 days after the MF’s release date except those classified as opportunistic (see Ob and Oc below).
10
Many firms have formal or informal policies requiring top managers to own either a fixed number of shares or a multiple of base salary (usually 1x to 5x base salary depending on rank). Managers can meet such guidelines by exercising options and holding the shares, via restricted stock grants, and/or by buying shares in the open market. 11
3.3
MF classification scheme: MFs issued opportunistically (OPP MFs) Managerial opportunism manifests itself in several ways: 1) insider selling prior to the
release of bad news; 2) insider buying prior to the release of good news; 3) issuing optimistic MFs prior to selling shares; or 4) issuing pessimistic MFs prior to buying. Based on this intuition we assign MFs to the opportunistic sample using the following criteria: Oa: Insider purchasing in the 30 days prior to release of good news MFs (abnormal returns at MF release date > 0%) or insider selling in the 30 days prior to release of bad news MFs (abnormal returns at MF release date < 0%), OR Ob: At the MF release date, the manager issues an optimistic MF (i.e., MF - Actual Earnings > 0), which triggers a large positive stock price reaction (i.e., abnormal return ≥ 5%), and there is insider selling (i.e., opportunistic selling) in the next 30 days, OR Oc: At the MF release date, the manager issues a pessimistic MF (i.e., MF - Actual Earnings < 0), triggering a large negative stock price reaction (i.e., abnormal return ≤ -5%), and there is insider buying (i.e., opportunistic buying) in the next 30 days. 11 Our MF classification scheme invariably misclassifies some MFs. For example, not all MFs classified by step Oa were necessarily opportunistic. In some cases managers may trade (ex ante) with no intention of issuing a subsequent MF because they do not believe they are in possession of material nonpublic information. However, after they trade they may learn a material fact so they disclose it via a MF and this disclosure results in a large stock price reaction. Thus, by chance the MF announcement abnormal return was positive, the insider sold, and the MF was misclassified as an OPP MF. Likewise, some COC MFs may be misclassified. Suppose a manager issued a biased MF in hopes of trading opportunistically (i.e., he/she issued an optimistic MF before selling). After issuing the MF the manager observes a market reaction that is smaller than expected, making it unprofitable to sell. Such possibilities imply that our sample of COC MFs may contain some MFs issued opportunistically ex ante, but which did not result in any insider trading ex post. Such misclassifications reduce the power of our tests to validate our MF classification scheme. Another potential source of misclassification arises from the SEC’s October 2000 Rule 10b5-1 that provides an affirmative legal defense against civil and criminal penalties to insiders 11
Replacing the ±5% cutoffs in Ob and Oc with ±3% and ±10% cutoffs does not alter the inferences. 12
who pre-plan trades (see Jagolinzer 2009). Because Thompson Financial’s Insider Trading database and other machine readable data sources do not separately identify trades executed under 10b5-1 plans, some MFs issued in the ± 30 days surrounding 10b5-1 plan trades will be incorrectly classified as DOA or OPP MFs, rather than COC MFs. However, the safe harbor protection under Rule 10b5-1 negates firms’ duty to disclose material nonpublic information prior to 10b5-1 planned trades. Hence, firms with managers who initiate 10b5-1 trading plans have less incentive to issue MFs surrounding insider trades than firms without such plans, and thereby these firms essentially have a lower chance of entering (and biasing) the MF sample. Nevertheless, 10b5-1 trading plan misclassifications can create two problems: our estimate of the frequency of COC MFs is understated, and the power of our tests to validate the classification scheme is reduced, although there is no obvious bias in our validation tests. We believe that the understatement of the frequency of COC MFs is not significant for a couple of reasons. First, Jagolinzer (2009) only identifies 1,241 firms between October 2000 and December 2005 that filed 10b5-1 plans, a relatively small number of the roughly 9,000 active firms covered by COMPUSTAT. Second, just because one executive files a 10b5-1 plan does not imply that all the firm’s executives filed a 10b5-1 plan. Hence, firms where some managers have filed 10b5-1 plans may still release MFs for COC, DOA, or OPP reasons.
4. Sample selection We start with 304,275 firm-quarters from First Call from 1998 to 2010.12 We obtain financial and stock price data from the WRDS Merged COMPUSTAT/CRSP database and lose 55,331 observations when we merge it with First Call. Next, we lose 20,729 observations by requiring coverage on Thompson Financial’s Insider Trading database, and another 158,560 observations due to missing data needed to calculate the independent variables (estimated at the firm-quarter level) underlying our multinomial probit model. We exclude 3,249 “warnings and preannouncements” leaving 66,406 firm-quarter observations that we classify into MF quarters and non-MF quarters based on whether there is at least one MF issued during the period from the last quarter’s earnings announcement date up through one day before the current quarter’s earnings announcement date. Of the 66,406 observations, 18,835 are MF quarters containing 12
Chuk et al. (2011) find that First Call is biased towards firms with more analyst coverage, more institutional ownership, and better past performance. However, since the focus our study is not to draw inferences about MF disclosure incentives using such variables we do not believe our inferences are affected by any such bias. 13
30,876 unique MFs. Since some firm-quarters contain multiple MFs, to avoid ambiguity, we drop 1,581 quarters containing MFs classified in more than one type. This final screen results in a final sample of 64,825 firm-quarter observations. Securities regulation and exchange listing requirements require prompt disclosure of material information whenever a company suspects persons with access to inside information may be trading in the company’s stock (see Block et al. 1985). Because a firm has a duty to disclose material information if it knows insiders wish to trade, we include insider trading of all top executives (not just the CEO and CFO). Our definition of insider trading includes all open market purchases or sales by the CEO, CFO, Chairman, President, Executive VP or Senior VP. We define a “good” (“bad”) news MF as one with a positive (negative) two-day (i.e., day 0 and +1) market-adjusted stock return (CAR). MFs with CARs greater than 5% in absolute value are designated “large news” MFs. We calculate MF forecast errors by subtracting the MF from actual EPS. Positive (negative) values indicate management pessimism (optimism).
5. Results 5.1
Frequencies of COC MFs, DOA MFs, and OPP MFs The results in Table 1 address our primary research question (“Should Guidance Be
Pooled?”). From the table it can be seen that our approach classifies 63.4% of all MFs as COC, 23.1% as DOA MFs, and 13.5% as OPP MFs. While our approach identifies that a majority of MFs are issued consistent with cost of capital incentives, a substantial fraction of MFs (36.6%) are not. Table 1 enumerates the steps responsible for generating the observations in each subsample. In the COC sample, Ca classifies 16,211 as COC MFs (82.9%) because there is no insider trading in the ± 30 days surrounding the MF. In the DOA sample, since there is only one step, Da, it classifies all of the sub-sample (N=7,145 or 23.1%). In the OPP sample, step Oa (insider purchasing in the 30 days prior to release of good news MFs or insider selling in the 30 days prior to release of bad news MFs) categorizes 3,343 as OPP MFs (80.2%). Since the estimated frequencies of COC MFs (63.4%) and non-COC MFs (36.6%) in Table 1 are conditional on the accuracy of our classification scheme, sections 5.3 - 5.6 provide a series of validation tests. Before reporting the results of those tests, we first provide some summary statistics about the three MF sub-samples.
14
5.2 Descriptive statistics Table 2 Panel A presents frequency distributions of the firm-quarter observations in the MF samples over the 1998-2010 period. The column “Total MFs” reveals that the number of firm-quarter observations with MFs increased from 324 in 1998 to 1,349 in 2010. Within MF samples, the percentage of MFs classified as DOA is roughly constant in the 18%-24% range while the percentage of COC MFs fluctuates between 62%-70%. The percentage of MFs classified as OPP falls from 16% in 1998 to 10% in 2010. The frequencies of all three types of MFs are relatively constant across quarters. The bottom portion of Panel A reports the frequency of “Bundled” and “Unbundled” MFs. A “Bundled” (“Unbundled”) MF is one issued (not issued) within ±1 day of an earnings announcement. The heading “Both” signifies firm-quarters containing both “Bundled” and “Unbundled” MFs. Intuitively, MFs bundled with an earnings announcement are more likely to be issued for DOA reasons because while the earnings announcement itself discloses material information about current period earnings, it does not reveal managers’ material nonpublic information about future earnings, which a concurrent MF accomplishes. Consistent with the notion that some MFs are issued to comply with DOA requirements, 80.3% of DOA MF firmquarters are issued concurrently with an earnings announcement compared to only 62.2% and 57.3% of COC and OPP MFs, respectively. Panel B of Table 2 reports descriptive statistics for each MF sample by the number of MFs as opposed to firm-quarter MFs as in Panel A. That is, while Panel A is based on firmquarter observations, Panel B is based on MF observations. The results indicate some differences across samples in terms of forecast horizon (quarterly vs. annual). For example, DOA MFs are associated with fewer annual forecasts, 55% compared to 58% and 59% for the COC and OPP MF samples, respectively. While there is a slight tendency for DOA MFs to be issued over shorter horizons, the differences are small. The results in Panel B also reveal little difference in forecast form (point, range, end, qualitative) across samples. By far, range MFs are the most common form, comprising roughly 79% of each sample. Turning to “MF News,” based on the approach used in Anilowski et al. (2007) to classify MFs into good, bad, neutral, and mixed “news,” we find that OPP MFs have a higher concentration of “bad news” (54%) than DOA MFs (44%), and COC MFs (51%). It is not surprising that OPP MFs have a higher proportion of bad news because from the insider trading literature we know that insiders sell more frequently than 15
they buy, and from Table 1 we see that about 80% of MFs classified as OPP are generated by step Oa, which classifies MFs as opportunistic if there are insider sales in the 30 days prior to the MF, and the MF is bad news. 5.3 First construct validity test of the MF classification scheme: Multinomial probit model Our first construct validity test of our MF classification scheme (others are described in sections 5.4 - 5.6) is a model designed to predict whether a particular type of MF (COC, DOA, or OPP) is issued in a firm-quarter based on the premise that throughout a given quarter managers continuously decide whether to issue a MF and whether to trade in their firm’s securities. Simply stated, managers gain utility from issuing MFs and/or trading in their firm’s stock. While the manager selects the option generating the highest utility, one cannot observe the manager’s utility for each alternative, only the outcome (i.e., the manager’s actual choice) is observed. Conceptually, manager m’s utility from selecting option k (µmk) can be expressed as: µmk = COCmβk + DOAmδk + OPPmγk + ξmk,
(1)
where COCm is a vector of variables capturing cost of capital incentives (coc1, coc2, …), DOAm is a vector of variables capturing disclose or abstain incentives (doa1, doa2, …), OPPm is a vector of variables capturing opportunism incentives (opp1, opp2, …), γk, βk, δk are vectors of coefficients to be estimated, and ξmk is an error term. Empirically, (1) is operationalized as a multinomial probit model which is appropriate in our setting because we have multiple, discrete, and unordered alternatives (issue a COC, DOA, or OPP MF, or do not issue a MF). 5.3.1
Multinomial probit model The empirical version of eq. (1) is estimated using all three MF samples. If our MF
classification scheme captures the three underlying reasons for why managers issue MFs in a meaningful way, then the coefficients on one set of independent variables (e.g., DOAm) will be larger in magnitude and have greater explanatory power in that particular MF sample (e.g., DOA MFs) than in either of the other two MF samples (e.g., COC or OPP). If the vectors of DOAm (doa1, doa2,…), COCm (coc1, coc2,…), OPPm (opp1, opp2,…) each contained a different set of variables then eq. (1) could be estimated as a simple dichotomous probit model. Such an analysis would indicate the relative frequency and importance of each of the three hypothesized motives for issuing a MF. However, as described below in sections 5.3.2 and 5.3.3, the DOA variables (doa1, doa2, …) are the same as some of the COC variables (coc1, coc2, …), hence 16
estimation of a dichotomous probit does not allow us to draw inferences about the relative importance of DOA, COC, and OPP MFs. Accordingly, we estimate the following multinomial probit model.13 MF_DUM = 0 + 1ERC + 2EARN_SURPRISE + 3SIZE + 4MB + 5RTNVOL + 6EARNVOL + 7HI_TECH + 8REGULATION + 9HABITUAL + 10RESTATE+ 11BACKDATE,
(2)
where MF_DUM takes the value of 1, 2 or 3 if a given firm-quarter contains a COC MF, DOA MF, or OPP MF, respectively, and zero for quarters without a MF. Here the dependent variable does not reflect any hierarchical measurement properties of the managers’ decision to issue MFs. It simply reflects that managers are likely to have issued a MF for COC, DOA, or OPP reasons, where the ordering of the dependent variable does not reflect any difference in the importance of the choice or whether one choice produced a higher level of utility for the manager. The independent variables are defined as follows (see Appendix A for details). ERC is the quintile rank of the firm’s estimated earnings response coefficient. EARN_SURPRISE is the quintile rank of the absolute value of the difference between the most recent consensus analyst earnings forecast issued prior to three weeks before the end of the fiscal quarter and that issued prior to the actual earnings announcement date of the prior quarter, deflated by price one day before the prior quarter’s actual earnings announcement date. In both cases quintile ranks are used to mitigate measurement error. SIZE is the natural logarithm of the market value of common equity at the beginning of the quarter. MB is the market-to-book ratio at the beginning of the quarter. RTNVOL is the standard deviation of the firm’s daily returns. EARNVOL is the standard deviation of the seasonal change in the firm’s actual quarterly EPS. HI_TECH equals one if the firm is a member of a high technology industry and zero otherwise. REGULATION equals one if the firm is a member of a regulated industry and zero otherwise. HABITUAL equals one if the firm issued a MF in at least six of the last eight quarters and zero otherwise. RESTATE equals one if the firm has at least one financial restatement during the period 19982006 per the U.S. Government Accountability Office (GAO) Financial Restatement database and zero otherwise. BACKDATE equals one if the firm is listed in The Wall Street Journal's 13
A maintained assumption of our model is that one and only one motive (COC, DOA, or OPP) explains why a particular MF was released. We recognize that more than one motive may explain some MFs. As a result the power of our tests to distinguish between alternative explanations may be reduced. 17
“Options Scorecard” on November 28, 2006 for being under scrutiny for option grant backdating and zero otherwise. Since MFs are often issued concurrently with earnings announcements, eq. (2) is estimated using: all MFs; only MFs without an earnings announcement; and only MFs bundled with an earnings announcement. Firm-quarters with only preannouncements or warnings are dropped (i.e., not counted as MF_DUM=0). 5.3.2 Determinants of MFs issued for cost of capital incentives The first nine independent variables in eq. (2), ERC, EARN_SURPRISE, SIZE, MB, RTNVOL, EARNVOL, HI_TECH, REGULATION and HABITUAL are all variables used in prior MF studies to capture or otherwise control for cost of capital disclosure incentives (see, e.g., Lennox and Park 2006). As such, they are the hypothesized determinants of COC MFs (i.e., they are COCm in eq. (1)). The intuition underlying each variable’s relation to MF issuance is as follows. ERC’s effect on MF disclosure is expected to be positive because managers of firms with greater sensitivity of stock prices to earnings-related information have stronger incentives to issue MFs for the reasons discussed in Lennox and Park 2006. EARN_ SURPRISE is a proxy for management’s private information (Kasznik and Lev, 1995). Managers wishing to maintain a good reputation with investors and satisfy analysts’ demand for timely information, issue MFs when the difference between their earnings expectations and those of analysts get large. Hence, the effect of EARN_SURPRISE on MF disclosure is expected to be positive. SIZE is used to capture the incentives of larger firms to issue more frequent MFs due to greater economies of scale in information production, greater demand for information by investors and analysts, or because they face greater litigation risk (Lang and Lundholm, 1993; Kasznik and Lev, 1995; Frankel et al. 1995). Based on these arguments, the effect of SIZE on MF disclosure is expected to be positive. Following Lennox and Park (2006), we include variables to capture firm risk: MB (market-to-book ratio), RTNVOL (return volatility), EARNVOL (earnings volatility). If MF issuance is affected by firm risk, then MB, RTNVOL, and EARNVOL should all have negative coefficients. To further control for litigation risk, as opposed to firm risk per se (Lennox and Park 2006), we use variables capturing membership in a high technology industry (HI_TECH), and membership in a regulated industry (REGULATION). If firms operating in high technology industries are more prone to litigation risk and issue MFs to reduce that risk, then HI_TECH will be positive. If firms operating in regulated industries provide more disclosure in general by virtue of being regulated, they will find it less necessary to issue MFs to communicate new 18
information to capital markets. Hence, REGULATION will be negative. Finally, HABITUAL is designed to capture the cost of capital benefits that managers perceive are associated with developing a reputation for providing MFs on a recurring basis. The coefficient on HABITUAL is expected to be positive. 5.3.3. Determinants of MFs issued to satisfy “disclose or abstain” (DOA) requirements Managers desiring to trade in their firm’s securities and also wishing to satisfy DOA rules will disclose all material nonpublic information prior to trading. The current judicial standard of materiality describes an item as material if there is a “substantial likelihood that the disclosure of the omitted fact would have been viewed by the reasonable investor as having significantly altered the ‘total mix’ of information made available” (TSC Industries v. Northway, Inc., 1976). At least two factors likely affect managers’ assessment of materiality: the extent to which the information differs from the market’s expectation and the sensitivity of a reasonable investor to a unit of earnings-related news (see. e.g., Heitzman et al. 2010). We measure the materiality of the managers’ private information as follows. EARN_SURPRISE captures the manager’s nonpublic information and ERC measures the sensitivity of a reasonable investor to a unit of earningsrelated news. We expect both to be positively related to managers’ decision to issue DOA MFs. Since we cannot observe managers’ private information about future earnings unless they disclose it, we assume that the consensus analyst forecast as of 21 days before the end of the quarter is an unbiased estimate of the manager’s expectation of the quarter’s earnings. To ensure that EARN_SURPRISE is not contaminated by “warnings” or “preannouncements” we use the consensus analyst forecast prior to the last three weeks of the quarter instead of the actual earnings for the quarter. Note that the two determinants of DOA MFs (EARN_SURPRISE and ERC) are also determinants of COC MFs (see section 5.3.2). Because we cannot identify unique determinants of DOA MFs, in order to validate our MF classification scheme we must rely on the multinomial probit model rather than a simple discrete choice model. 5.3.4 Determinants of MFs issued opportunistically Skaife et al. (2010) document that firms with weak internal controls have more significant insider trading profits compared to firms with stronger internal controls. Following Skaife et al. (2010), we conjecture that managers issuing MFs opportunistically face weaker corporate governance and control systems that serve to permit such behavior. However, attempts 19
to measure the strength of firms’ governance systems encounter both empirical and theoretical challenges (see, e.g., Larcker et al. 2007 and Brickley and Zimmerman, 2010). Consistent with such concerns, instead of using firm characteristics such as board size to identify “weak” control systems, we use outcome measures. In particular, RESTATE and BACKDATE are our opportunistic determinants of MF disclosure, and are designed to capture firms with weak control systems. Both are expected to be positively related to managers’ decision to issue opportunistic MFs and should best explain the issuance of OPP MFs. Table 3 summarizes the variables and their hypothesized effect on managers’ decision to issue MFs in each of the three samples. While COC MFs and DOA MFs share two common explanatory variables (ERC and EARN_SURPRISE), COC MFs are also expected to be explained by variables unrelated to DOA MFs. If our MF classification scheme has construct validity, then these other COC explanatory variables should have more explanatory power in the COC sample than in the other two samples. Likewise, both RESTATE and BACKDATE should only be important in the OPP sample, and, absent classification errors, should be insignificant in the COC and DOA samples. Table 4 reports the means and medians for the independent variables used in the multinomial probit model (i.e., the DOA, COC and OPP variables from the theoretical model in eq. (1)). The primary takeaway from Table 4 is that roughly two thirds of the variables used in prior MF research exhibit significantly different distributional properties across MF samples. With 11 variables per sample, there are 33 t-test and Wilcoxon tests reported in Table 4 (while both MVE and LNMVE are reported only MVE is used in the comparisons). Of that total, 26 of the t-tests and 25 of the Wilcoxon tests signify differences that are significant at the 10% level or better. The likelihood of observing 25 differences out of 33 trials where the likelihood of a difference is 10% is 0.000. If our approach to assigning MFs into samples was merely doing so at random, we would expect to observe about three significant t-tests and Wilcoxon tests at the 10% level. Observing 25 out of 33 differences is consistent with our method classifying MFs into samples with different underlying economic characteristics.14
14
Since the Wilcoxon test is a test for a difference between the distributions of two samples, the asterisks next to the median value signify a difference in the distributions rather than a difference in the medians. The mean of HABITUAL exceeds 42% in all samples and reflects the tendency of firms to persistently issue MFs as our sample period unfolds. The large mean value of RESTATE is because we assign it a value of one if the firm has a restatement anytime during the 1998-2010 period rather than just having a restatement in the current quarter. 20
5.3.5 Results of estimating the multinomial probit model Table 5 reports the results of estimating the multinomial probit model where the dependent variable is zero if the firm-quarter contains no MF, 1 if it contains only COC MFs, 2 if contains only DOA MFs, and 3 if it contains only OPP MFs. We begin by noting that interpreting a multinomial probit model is different from interpreting a dichotomous probit model. Rather than answering the question “what is the probability of a firm not issuing any MF given x set of determinants,” the multinomial probit addresses the question “what is the probability that a firm chooses to issue a COC MF (or a DOA MF or a OPP MF) as opposed to not issuing any MF given x set of determinants." In all such cases, the zero classification (i.e., no MF issued) is the baseline outcome. Model estimation produces separate coefficient estimates for each independent variable and sample. For ease of interpreting the results, we use the symbol ‡ to the right of a given coefficient to indicate whether it is consistent with the predictions in Table 3. For example, if a given variable is predicted to have a positive sign (e.g., ERC) then a ‡ denotes that the estimated coefficient is larger in that sample than in the sample where that variable is not expected to explain MF issuance. Likewise, if a given coefficient is predicted to be negative, then a ‡ denotes that the estimated coefficient is smaller in that sample than in the sample where it is not expected to explain MF disclosure.15 The findings in Table 5 Panel A reveal that the quintile rank of a firm’s ERC is significant in predicting all three types of MFs. If we were able to construct MF samples without misclassification Table 3 predicts ERC to be significant in the COC and DOA MF samples. Since we noted that misclassifications can occur (see Section 3), our interest lies in comparing the coefficients on ERC (and other independent variables) across samples. Hence, ERC should be more important in explaining MF issuance in the COC and DOA MF samples when compared to OPP sample. Similarly, Table 3 predicts the coefficient on EARN_SURPRISE to be larger in the COC and DOA MF samples when compared to the OPP sample. Consistent with this, we
Similarly, BACKDATE is assigned a value of one if a firm was alleged to have engaged in option-grant backdating anytime during the sample period, not just having a back-dating event in the current quarter. 15 Standard errors have been adjusted for heteroskedasticity and firm-specific clustering. We do not adjust for timeclustering for three reasons. First, the primary source of potential time clustering is common shocks like new regulations (e.g., Reg FD and SOX). By including year and quarter dummies we absorb such effects making oneway clustering at the firm level appropriate in our setting. That said, one-way clustering by year-quarter leads to similar findings. Second, clustering affects the standard error, but not the coefficient. Since our main focus is a cross-sample comparison of the coefficients, standard errors are less of a concern. Third, presently there is no statistical package capable of computing two-way clustered errors for multinomial probit models. 21
find that the coefficient on ERC is larger in the COC and DOA samples (0.070 and 0.096) than in the OPP sample (0.067). Also consistent with Table 3, the coefficient on EARN_SURPRISE is larger in the COC sample (0.056) than in the OPP sample (-0.009). Since the coefficient on EARN_SURPRISE is negative (as opposed to its predicted positive sign) in the DOA sample (0.001) we do not compare it to with that in the OPP sample. Table 3 predicts that the remaining seven coefficients on LNMVE, MB, RTNVOL, EARNVOL, HI_TECH, REGULATION, and HABITUAL to be more important in explaining COC MFs than either of the other two types of MFs. That is, if a given coefficient is predicted to be positive (negative) for COC MFs, it should be larger (more negative) than that for the other two samples (DOA and OPP MFs). Turning to the results we find that the coefficients on RTNVOL, EARNVOL, HI_TECH, and HABITUAL predicted to capture cost of capital disclosure incentives are consistently larger (if the coefficient is predicted to be positive) or smaller (if the coefficient is predicted to be negative) in the COC sample than in at least one of the other two samples. The last two coefficients (RESTATE and BACKDATE) capture weak governance and are expected to only be important in predicting issuance of OPP MFs. Consistent with this, the results reveal that the coefficient on RESTATE is larger in predicting OPP MFs than in predicting DOA or COC MFs, and the coefficient on BACKDATE is larger in predicting OPP MFs than in predicting COC MFs. Notice that the coefficient on RESTATE in the COC sample and the coefficients on BACKDATE in the DOA MF sample are also statistically different from zero. We interpret these significant coefficients (as well as those on other dependent variables), as an indication of some MF misclassifications in the sub-samples. With 11 variables in each sample’s model and two pair-wise comparisons to make between samples (e.g., COC MFs vs. each of the other two samples), there are 22 coefficient comparisons, where a coefficient comparison is based on the predictions in Table 3. For example, one comparison involves whether the coefficient on ERC is larger in the DOA sample than in the OPP sample. Ignoring the coefficient comparison involving EARN_SURPRISE in the DOA sample (because it has the wrong sign) leaves 21 coefficient comparisons. Of these, 14 are in the predicted direction (as denoted by a ‡ in Table 5). The likelihood of observing 14 out of 21 differences due to chance, where the probability of a difference is 0.50, is 0.095 (one-tailed test). Overall, this evidence is consistent with our MF classification scheme that predicts that
22
variables commonly used in the MF literature exhibit differential explanatory power for managers’ decision to issue MFs for COC, DOA, and OPP reasons. Panel B of Table 5 reports the results of Wald tests examining coefficients across models. If our approach is classifying MFs based on managers’ different underlying incentives, the coefficients on ERC and EARN_SURPRISE should be larger in the DOA sample than in the OPP sample. Based on a Chi-square statistic of 4.21 (p-value=0.121) we are unable to reject the null hypothesis at conventional significance levels that the coefficients on ERC and EARN_SURPRISE are equal between the DOA and OPP MF samples. The Wald test does indicate that these coefficients are different between the COC and OPP MF samples (Chi-square = 27.24, p-value < 0.001). Wald tests comparing all seven variables between the COC and DOA samples, and between the COC and OPP MF samples reveal significant differences between the COC and DOA MF samples (Chi-square statistic of 25.68, p-value < 0.001), and the COC and OPP MF samples (Chi-square statistic of 109.42, p-value < 0.001). The last two Wald tests jointly compare the two variables expected to explain OPP MFs (i.e., RESTATE and BACKDATE). The results reveal significant differences between OPP MFs DOA MFs (Chisquare statistic of 5.92, p-value < 0.052), but not between the COC MFs and OPP MFs (Chisquare statistic of 1.89, p-value =0.388). In summary, four of the six Wald tests reject the null hypothesis at the 10% level or better. We interpret this, and the other evidence in Table 5 to be generally consistent with of our MF classification scheme in the sense that a given variable’s ability to predict managers’ issuance of MFs varies across samples in predictable ways based on their various incentives for issuing earnings guidance. While the evidence in Table 5 provides some assurance that our MF classification scheme is capturing what we intend it to, the evidence in Table 5 also indicates that the approach misclassifies a number of MFs as evidenced by the coefficients on some variables differing from zero in samples where they should theoretically be zero (i.e., where they should not explain MF issuance). Even though firms adopting ex ante commitment policies to disclose MFs do so for the same economic incentives as postulated for ex post COC MFs, such firms’ MFs could differ conceptually from other MFs we classify as COC MFs. To address this concern, we redid the univariate tests in Table 4 and multinomial probit tests in Table 5 after excluding firms we identified as potentially having an ex ante commitment policy to disclose MFs. To identify such firms we adopt the methodology proposed in Tang (2011). Specifically, for a given firm in a 23
given year t, we remove all of the firm’s quarterly observations if the pattern of the firm’s MFs in year t is identical to the pattern of MFs in year t-1. For example, if in years t and t+1 a firm issued MFs in quarters 2, 3 and 4, then all 4 of the firm’s quarters in year t+1 are excluded from the sample. This led to deleting roughly 25% of the MFs classified as DOA, COC, and OPP and about 60% of the firm-quarters with no MFs. When we repeat the analysis in Tables 4 and 5 the results are similar. For example, as in our previous findings, four of the six Wald tests in Table 5 Panel B are statistically significant at the 10% level. In addition, since HABITUAL may also capture a commitment to a disclosure policy, we repeated the analysis after excluding it. Our original inferences are robust when HABITUAL is excluded. As a final robustness test we re-estimated the multinomial probit model using only bundled MF firm-quarters, and then only unbundled MF firm-quarters (firm-quarters containing BOTH “bundled” and “unbundled” MFs are excluded from these models, but were included in Table 5’s models). Un-tabulated results using only bundled MF firm-quarters reveal that three of the six Wald tests reject the null hypothesis compared to four of six originally in Table 5. Beyond that, the coefficient comparisons are similar to those reported earlier in Table 5. Untabulated results using only unbundled MF firm-quarters show that two of the six Wald tests are significant and that 13 out of 20 coefficient comparisons are in the predicted direction. While the Wald test results for unbundled MFs are weaker than those originally reported in Table 5, this is likely due to having fewer MF firm-quarters containing only unbundled MFs (3,897 unbundled MF firm-quarters vs. 12,692 MF firm-quarters in Table 5). Overall, the un-tabulated results indicate that Table 5’s results are not driven by either bundled or unbundled MFs. Since the number of correct coefficient tests is similar across models using either just bundled or unbundled MFs, we conclude that the inferences from our main tests are reasonably robust to “bundled” and “unbundled” MFs. 5.4 Second construct validity test of the MF classification scheme: Replication of Cheng and Lo (2006) Our second construct validity test of the MF classification scheme is based on a replication of Cheng and Lo (2006) (CL). We chose this study for three reasons: it is fairly recent and well-cited, it is closely related to our study in that it examines opportunistic MFs and insider trading, and we had access to the same data sources. CL (2006) find “that when managers plan to purchase shares, they increase the number of bad news forecasts to reduce the 24
purchase price … (and) do not find that managers adjust their forecasting activity when they are selling shares” (p. 815). CL (2006) recognize that various reasons exist for why managers issue MFs including: to reduce information asymmetry and the cost of capital, to reduce litigation costs, and to trade strategically in their firm’s stock. However, CL (2006) include all MFs to test whether insiders strategically disclose MFs and time their equity trades to maximize trading profits. Including all MFs, rather than just OPP MFs likely reduced the power of their tests and potentially biased their findings. The evidence below suggests that both concerns are valid. We follow CL’s (2006) sampling procedure and focus on firms with both Thompson Reuter Institutional (13F) and First Call CIG data for the sample period 1995-2002. We select all calendar firm quarters if they have insider trading data during the sample period. After obtaining data for control variables from COMPUSTAT and CRSP we have a final sample of 82,643 firm-quarters, which is slightly larger than the 77,106 in CL (2006) (possibly due to the increased coverage of the updated COMPUSTAT files). CL’s (2006) main analysis involves a two-stage regression reported in their Tables 3 and 4. To control for the endogeneity issue of reverse causality (i.e., insider trading triggered by firm disclosures) CL (2006) estimate a prediction model in the first stage where the amount of insider trading for quarter t is regressed on various determinants (e.g., size, growth option, returns, insider trading) measured from quarter t-1. In a second stage, CL (2006) regress MF frequency for quarter t on the magnitude of insider trading predicted from the first stage. Thus, insider trading in the second stage is an expected value, not the actual value. To control for any timeinvariant omitted variables they use a change specification in both stages. In un-tabulated results (available upon request) our replication of the first-stage model in CL (2006) yields results very similar to those reported Table 3 in CL (2006). Table 6 presents the results of our replication of CL’s (2006) second-stage model. Column (1) reproduces the CL (2006) findings in column (4) of their Table 4 Panel A (Zstatistics are left blank as CL (2006) did not report them). Our replication using all MFs is shown in column (2) Table 6. Similar to CL (2006), net news frequency is significantly negatively associated with predicted insider purchases (-0.024, Z-statistic = -3.07) and net news frequency is unrelated to predicted insider sales (0.0078, Z-statistic = -0.99). However, when we estimate the CL (2006) model separately using our three categories of MFs (COC MFs in column 3, DOA MFs in column 4, and OPP MFs in column 5) we find different results across the three MF 25
categories. First, the net news frequency based on either COC or DOA MFs exhibits no association with predicted insider purchases and sales. On the other hand, when we limit net news frequency to only OPP MFs we find that the coefficients on both predicted IP and predicted IS are statistically significant. This empirical evidence serves to further validate our MF classification scheme. Specifically, CL (2006) findings only exist in the sample of MFs where our approach classifies such MFs as being issued opportunistically by managers. Second and more importantly, while CL (2006) documents no effect of predicted insider sales, we find that the net news frequency of OPP MFs exhibits a significant positive relationship with insider sales (0.0207, Z-statistic = 1.67). Collectively, our findings cast doubt upon one of CL’s (2006) major conclusions “We do not find that managers adjust their forecasting activity when they are selling shares, consistent with higher litigation concerns associated with insider sales.” Using our MF classification scheme to identify OPP MFs, we find that managers do in fact adjust their forecasting activity when they are selling shares, and that this is inconsistent with higher litigation concerns associated with insider sales. Further evidence (untabulated) shows that when managers are selling shares they adjust their forecasting activities through increasing the number of good news forecasts rather than reducing the number of bad news forecasts. Finally, our findings (untabulated) are robust for unbundled MFs as in CL’s (2006) Table 4 panel B. A concern may arise that our results are mechanical because our classification scheme identifies OPP MFs primarily from step Oa in Table 1. Oa classifies a MF as opportunistic if there are insider purchases in the 30 days prior to a “good news” MF or insider sales in the 30 days prior to a “bad news” MF. However, the second-stage CL (2006) regression and our replication are based on predicted insider purchases and sales estimated from prior information. Hence, our classification scheme (based on concurrent insider trading) is unlikely to introduce a mechanical relationship. In addition, the model’s change specification further mitigates concerns about any mechanical effect. To summarize, our replication of CL (2006) yields two important conclusions. First, our MF classification scheme leads to more powerful tests of the incentives underlying specific reasons for why managers issue MFs. Second, inferences drawn from studies (e.g., CL, 2006 in particular) that assume all MFs were issued for one reason (and which ignore the other reasons) can result in incorrect inferences
26
5.5 Third construct validity test of the MF classification scheme: Predictive ability of the insider trading associated with DOA MFs and OPP MFs The third validity test of our MF classification scheme relies on prior research on the ability of insider trading to predict future stock price performance (see, e.g., Lakonishok and Lee 2001 and Seyhun 1992). This literature finds that insider trading tends to predict future firm performance in that insider purchases are generally associated with higher future stock returns than insider sales. We examine the predictive ability of DOA MFs and OPP MFs for future stock returns. If our approach to classify MFs as DOA captures MFs issued by managers motivated by a desire to release material nonpublic information prior to trading, such insider trading should exhibit no predictive ability for future stock returns. Similarly, if our approach to classify MFs as opportunistic actually captures opportunism, then the insider trading associated with such MFs should exhibit predictive ability for future stock returns. We exclude COC MFs from these predictive ability tests because 82.9% of such MFs have no insider trading in interval [-30 to +30] (see Table 1). To test our predictions, we start with all DOA and OPP MFs identified in Table 1. For each DOA and OPP MF we calculate the net insider trading during the window that was relevant for assigning that MF to its respective sample. For example, for DOA MFs, and OPP MFs in the Ob and Oc subgroups, the window is trading days +1 to +30 after the MF’s release. For OPP MFs in the Oa subgroup, the window is trading days -30 to -1 before the MF’s release. We then classify a MF as an insider purchase (sale) if the net trading during the window is purchase (sale). Future abnormal stock returns (CAR) are calculated as the raw buy-and-hold return minus the buy-and-hold return of the corresponding Fama-French 25 portfolios formed on the basis of size and book-to-market ratio. Year +1 is the first 12 months (year +2 is the first 24 months) subsequent to the [-30, +30] trading window centered on the MF’s release. We drop observations if future CAR data is unavailable and we winzorize CAR at the top and bottom 1%. We test the difference between insider purchase and sale groups using a Wilcoxon test. If insider trading associated with OPP MFs has predictive power for future performance the difference in CAR between insider purchases and sales should be positive. The results reported in Table 7 are based on a sample of 6,067 DOA MFs and 3,580 OPP MFs for the period 1998-2008. We end in 2008 due to the 24-month holding return period. Given the well-documented skewness in returns, we report median values of CAR in the first and second year subsequent to the [-30,+30] window centered on the MF’s release separately for 27
insider purchases and sales, as well as test for a difference between the two. Focusing on DOA MFs, if the direction of insider trading can predict future performance, we should observe significant positive differences in future CAR between insider purchases and sales. Contrary to this, neither the 12-month nor the 24-month return differences are significantly positive. Turning to OPP MFs, the difference in median CARs between insider purchases and sales are significantly positive in both Year +1 and Year +2 (differences = 6.41% and 9.08%, respectively). Overall, Table 7’s results provide further empirical evidence that our classification of OPP MFs captures those issued by managers motivated by a desire to trade opportunistically in their firm’s securities, whereas there is no difference in future CAR between insider purchases and sales following MFs classified as DOA. 5.6 Fourth construct validity test of the MF classification scheme: Replication of Ajinkya et al. (2005) As a final test of our MF classification scheme we replicate Ajinkya, Bhojraj, and Sengupta (2005) (ABS) to demonstrate how the results of a published study could be affected by recognizing the alternative rationales for managers to issue MFs that are examined in our study. We chose ABS (2005) because it is another fairly recent paper and well-cited paper, and we had access to the same or comparable data. ABS (2005) document that firms with better governance mechanisms, as measured by more outside directors and greater institutional ownership, are more likely to issue MFs. Following the prior literature, they motivate their tests by arguing “that managers acting in the best interests of the firm should enhance transparency by issuing more frequent,… forecasts.” (p. 348). In terms of our focus in this paper, the key aspect of ABS (2005) of interest to us is their implicit assumption that all MFs are issued to enhance transparency and increase firm value – what we refer to as the “COC” rationale for issuing earnings guidance. If our MF classification scheme is able to identify the underlying reason for issuing MFs, then the ABS (2005) results should be stronger for those MFs we classify as COC and weaker for those we classify as DOA and OPP. To replicate ABS (2005) we follow their procedure and select from First Call all firmyears from 1997 through 2002 with at least three analysts following the firm. We identify a firmyear as MF-issuing if the firm issued at least one annual MF before the fiscal year end. We match the sample with RiskMetrics and Thompson Reuter Institutional (13F) data to obtain information on outside directors and institutional ownership, respectively. Because RiskMetrics covers only S&P 1500 firms, our sample is smaller than ABS (2005) who use the more 28
comprehensive Compact Disclosure data, for which we do not have access. After obtaining data for control variables from COMPUSTAT and CRSP, our final sample contains 5,413 firm-years, which compares to a sample of 7,745 observations for the 1997-2002 time period in ABS (2005). Table 8 reports the results of our replication. Column (2) reproduces the findings in column (1) of Table 3 in ABS. Our replication using all MFs is reported in column (3). In general, most of the t-statistics on the coefficients in our probit model are reasonably consistent with those reported in ABS (2005) even though our sample size is smaller (and based on larger firms in the S&P 1500). A notable exception is INST (percentage of common shares held by institutions). ABS (2005) primary focus is on OUTDIR (percentage of the board of directors who are not officers of the firm) and INST as these are their proxies for corporate governance. Focusing on the t-statistics on OUTDIR, ABS (2005) report 4.05 compared to our 4.24. Columns 4-6 in Table 8 disaggregate all of the MFs used to estimate column (3) into our three MF categories (COC, DOA, and OPP MFs). Consistent with our conjecture, we find that the coefficient (and t-statistic) on OUTDIR is largest and the most significant in the COC MF sample (0.011 with a t-statistic of 4.63) compared to the DOA MF sample (0.006 with a tstatistic of 1.89), and the OPP MF sample (0.003 with a t-statistic of 0.94). Since ABS (2005) test the hypothesis that better governance improves transparency (i.e., lowers the firm’s cost of capital), then the MFs we identify as COC should have the strongest association with their governance proxies. This is what we find for their OUTDIR variable and this inference is robust (untabulated) to re-estimating columns (3) - (6) using data from 1997-2004 (7,429 observations), from 1997-2010 (12,747 observations), and with and without year dummies. On the other hand, we are unable to replicate ABS (2005) on the INST variable. ABS (2005) report a t-statistic on INST of 12.94, whereas using all MFs our t-statistic is only 3.30. There are several possible reasons for the difference. As mentioned above, our sample is limited to larger S&P 1500 firms. Second, Chuk et al. (2011) report that coverage on the First Call MF database is far from complete in that it is more likely to contain MFs for firms with high institutional ownership, high analyst following, and less likely to contain firms reporting recent losses. Chuk et al. (2011) conclude, “researchers should use caution if their studies are related to analyst following, institutional ownership, or firm performance.” These warnings apply to both ABS (2005) and our replication with respect to INST. Finally, we find that the First Call database of MFs is not cumulative over time. Specifically, we observe that some MFs that exist 29
on previous versions of First Call do not appear on subsequent versions. Since First Call does not document their sampling or coverage process (see Chuk et al. 2011), and given the documented biases in the First Call database, the First Call versions used by ABS (2005) and us likely differ in terms of the MFs reported.
6. Conclusions The disclosure literature identifies several primary reasons why managers issue earnings guidance: to facilitate access to capital markets at lower cost; to comply with Rule 10b-5 of the securities laws whereby managers (i.e., insiders) must disclose material nonpublic information or abstain from trading in their firm’s securities (DOA); to trade opportunistically in their firm’s securities; and to reduce the expected costs of shareholder class action litigation.16 While the literature recognizes that managers face a variety of incentives to issue guidance, most empirical studies focus on one rationale and thereby assume all MFs are disclosed for that reason. In effect, previous papers pool all MFs and assume only one reason is motivating disclosure. The primary research question we investigate is: “Should Guidance Be Pooled?” To address this question we use the primary rationales for why managers issue earnings guidance (capital market incentives, compliance with the 10b-5 DOA rules, and opportunistic trading motives) to develop an approach that classifies MFs into samples based on the likely incentive managers faced for issuing a particular MF. Our approach uses properties of the data such as insider trading around the MF’s release, characteristics of the MF, the market reaction to the MF, and the MF forecast error to classify MFs. We find that roughly 63% of MFs are issued due to capital market incentives, 23% are issued to comply with the 10b-5 DOA rule, and 14% are opportunistic. Four sets of empirical tests are used to validate our method. In the first we hypothesize that each of the three explanations to issue MFs has a set of independent variables that best explains whether that particular type of MF will be disclosed in a given firm quarter, and moreover, that those variables will differ in their explanatory power across MF classifications. We estimate a multinomial probit model to examine the factors associated with managers’ decision to issue MFs and find that variables predicted to explain a particular type of MF (e.g., 16
We do not investigate the MF disclosure rationale to reduce a firm’s expected litigation costs by issuing earnings warnings or preannouncements (see, Skinner 1994) because we limit our study to MFs issued before the last three weeks prior to the end of the quarter. 30
cost of capital MFs) have greater explanatory power for MFs in that sample vis-à-vis the other samples (i.e., the DOA and opportunistic MFs). Our second empirical test replicates Cheng and Lo (2006) where we find that their results only exist in the sample of MFs we classify as opportunistic (consistent with the focus of their study – managers disclose MFs strategically for personal gain). In addition, we overturn one of Cheng and Lo’s (2006) key inferences namely, that managers do not adjust their forecasting activity when they are selling shares. Our third empirical test documents that insider trading surrounding MFs classified as “opportunistic” is associated with future abnormal returns while MFs classified as DOA exhibit no predictive ability for future abnormal returns. Finally, a replication of Ajinkya et al. (2005) reveals that their results are stronger in a sample of MFs classified as cost of capital MFs (that is, where their hypotheses actually predict an effect) than in samples of DOA MFs or opportunistic MFs. Based on our findings we offer the simple observation that “all MFs are not created equal.” Managers have a number of economic incentives for issuing earnings guidance and no single economic explanation/rationale, such as to lower the firm’s cost of capital, explains the complex process underlying managers’ MFs decisions. While many MFs (roughly 63% by our estimate) are issued because managers want to reduce the firm’s cost of capital, a substantial fraction of MFs (37%) appear to be released because managers wish to comply with the disclose or abstain provisions of the securities laws before they trade, or because they want to profit from opportunistic insider trading. Our findings have implications for prior research using MFs to test disclosure theories and incentives as well as for the research design of future MF studies. First, many prior MF studies treated all MFs as motivated to increase capital market transparency and lower the firm’s cost of capital, when in fact some MFs were not issued for that reason. This implies that studies treating MFs as if they are all issued for capital market reasons are misclassifying roughly 37% of their sample, potentially leading to a loss of power or to incorrect inferences. While loss of power does not cause us to question prior published papers documenting significant findings, past and unpublished MF studies failing to document significance (and future studies) could be enhanced by limiting samples of MFs to only those that are the focus of the study. Second, prior MF papers implicitly assume that their independent variables are only capturing incentives to lower the firm’s cost of capital. The problem, however, is that variables used to proxy for cost of capital incentives are also natural proxies for managerial incentives to issue MFs to satisfy DOA 31
rules. This calls into question the inferences drawn in prior MF studies about the predicted effects of cost of capital variables. Finally, our analysis is a joint test of the underlying motives for why managers issue earnings guidance and our MF classification scheme. While we offer four sets of empirical evidence to provide construct validity for our MF classification scheme, we recognize that misclassifications exist as evidenced by significant coefficients in the multinomial probit model on some variables in samples where we predict they should be insignificant. An extension of our study would be to identify additional variables to explain issuance of opportunistic and DOA MFs. Such research would improve the design of future MF studies.
32
Appendix A Detailed variable definitions and measurement procedures MF_DUM takes the value of 1, 2 or 3 if a given firm-quarter is classified into the COC, DOA, or OPP MF sample, respectively, and zero for firm-quarters without a MF. The MF must be issued during the period from the earnings announcement date of the last quarter up to one day before the current quarter’s earnings announcement date. ERC is the quintile rank of the firm’s earnings response coefficient. We estimate a firm-specific ERC for each firm-quarter by regressing two-day (i.e., day 0 and +1) earnings announcement period market-adjusted returns on unexpected earnings using the 16 most recent quarters (complete data is required for all 16 quarters). Unexpected earnings is actual EPS minus the most recent consensus analyst earnings forecast issued prior to the earnings announcement date, deflated by stock price on day -1 relative to the earnings announcement date. Quintile ranks are used to mitigate measurement error. EARN_SURPRISE is the quintile rank of the absolute value of the difference between the most recent consensus analyst earnings forecast issued prior to three weeks before the end of the fiscal quarter and that issued prior to the earnings announcement date of the prior quarter, deflated by stock price on day -1 relative to the prior quarter’s earnings announcement date. Quintile ranks are used to mitigate measurement error. SIZE is the natural logarithm of the market value of common equity at the beginning of the quarter. MB is the firm’s market-to-book ratio at the beginning of the quarter. RTNVOL is the standard deviation of the firm’s daily stock return over the 250 trading days before the beginning of the quarter (a minimum of 100 trading days required). EARNVOL is the standard deviation of the seasonal change in quarterly EPS scaled by assets per share as of the beginning of the quarter based on the 16 most recent quarters of data (complete data is required for all 16 quarters). HI_TECH equals one if the COMPUSTAT SIC code at the end of the quarter is 2833–2836 (Drugs), 8731–8734 (R&D services), 7371–7379 (Programming), 3570–3577 (Computers), or 3600–3674 (Electronics), HI_TECH, and zero otherwise. REGULATION equals one if the COMPUSTAT SIC code at the end of the quarter is 4812–4813 (Telephone), 4833 (TV), 4841 (Cable), 4811–4899 (Communications), 4922–4924 (Gas), 4931 (Electricity), 4941 (Water), or 6021–6023, 6035–6036, 6141, 6311, 6321, 6331 (Financial firms), and zero otherwise. HABITUAL equals one if the firm issued a MF in at least six of the last eight quarters. RESTATE equals one if the firm has at least one financial restatement during the period 19982006 per the U.S. Government Accountability Office (GAO) Financial Restatement database and zero otherwise. BACKDATE equals one if the firm is listed in The Wall Street Journal’s “Options Scorecard” on November 28, 2006 as being under scrutiny for option grant backdating and zero otherwise.
33
References Aboody, D., and R. Kasznik, 2000, “CEO Stock Option Awards and the Timing of Corporate Voluntary Disclosures,” Journal of Accounting and Economics 29, pp. 73-100. Ajinkya, B., S. Bhojraj, and P. Sengupta, 2005, “The Association Between Outside Directors, Institutional Investors and the Properties of Management Earnings Forecasts,” Journal of Accounting Research 43, pp. 343-376. Ajinkya, B and M. Gift, 1984, “Corporate Managers’ Earnings Forecasts and Symmetrical Adjustments of Market Expectations,” Journal of Accounting Research 22, pp. 425–444. Anilowski, C., M. Feng, and D. Skinner, 2007, “Does Earnings Guidance Affect Market Returns? The Nature and Information Content of Aggregate Earnings Guidance,” Journal of Accounting and Economics 44, pp. 36-63. Baginski, S., E. Conrad, and J. Hassell, 1993, “The Effects of Management Forecast Precision on Equity Pricing and on the Assessment of Earnings Uncertainty,” Accounting Review 68, pp. 913-927. Baginski, S., J. Hassell, and M. Kimbrough, 2004, “Why Do Managers Explain Their Earnings Forecasts?" Journal of Accounting Research 42, pp. 1-29. Bettis, J., J. Coles, M. Lemmon, 2000, “Corporate Policies Restricting Trading by Insiders,” Journal of Financial Economics 57, pp. 191-220. Beyer, A., D. Cohen, T. Lys, and B. Walther, 2010, “The Financial Reporting Environment: Review of the Recent Literature,” Journal of Accounting and Economics 50, pp. 296-343. Block, D., N. Barton, and A. Garfield, 1985, “Affirmative Duty to Disclose Material Information Concerning Issuer's Financial Condition and Business Plans,” The Business Lawyer 40, pp. 1243-1265. Botosan, C., 1997, “Disclosure Level and the Cost of Equity Capital,” Accounting Review 72 (3), pp. 323-349. Botosan, C. and M. Plumlee, 2002, “A Re-examination of Disclosure Level and the Expected Cost of Equity Capital,” Journal of Accounting Research 40 (1), pp. 21-40. Brickley, J., and J. Zimmerman, 2010, “Corporate Governance Myths: Comments on Armstrong, Guay, and Weber,” Journal of Accounting & Economics 50 pp. 235-245. Cheng, Q., and K. Lo, 2006, “Insider Trading and Voluntary Disclosures,” Journal of Accounting Research 44, pp. 814-48. Chuk, E., D. Matsumoto, and G. Miller, 2011, “Assessing Methods of Identifying Management Forecasts: CIG vs. Researcher Collected,” Working paper University of Washington. Clement, M., R. Frankel, and J. Miller, 2003, “Confirming Management Earnings Forecasts, Earnings Uncertainty, and Stock Returns,” Journal of Accounting Research 41, pp. 653679. Coller, M., and T. Yohn, 1997, “Management Forecasts and Information Asymmetry: An Examination of Bid-ask Spreads,” Journal of Accounting Research 35, pp. 181-191. 34
Collins, D., and S. Kothari, 1989, “An Analysis of the Cross-sectional and Intertemporal Determinants of Earnings Response Coefficients,” Journal of Accounting and Economics 11, pp. 143-181. Core, J., 2001, “A Review of the Empirical Disclosure Literature: Discussion,” Journal of Accounting and Economics 31, pp. 441–456. Diamond, D., 1985, “Optimal Release of Information by Firms,” Journal of Finance, pp. 1071-94. Diamond, D., and R. Verrecchia, 1991, “Disclosure, Liquidity, and the Cost of Capital,” Journal of Finance 46, pp. 1325-1355. Dye, R., 1985a, “Disclosure of nonproprietary information” Journal of Accounting Research, 23, pp. 123–145. Dye, R., 1985b, “Strategic accounting choice and the effects of alternative financial reporting Requirements,” Journal of Accounting Research 23, pp. 544–574. Einhorn, E., and A. Ziv. 2008. Intertemporal dynamics of corporate voluntary disclosures. Journal of Accounting Research 46 (3): 567-589 Evans, J., and S. Sridhar, 2002, “Disclosure-disciplining Mechanisms: Capital Markets, Product Markets, and Shareholder Litigation,” The Accounting Review 77, pp. 595-626. Frankel, R., and X. Li, 2004, “Characteristics of a Firm’s Information Environment and the Information Asymmetry Between Insiders and Outsiders,” Journal of Accounting and Economics 37, pp. 229-259. Frankel, R., M. McNichols, and G. Wilson, 1995 “Discretionary Disclosure and External Financing,” Accounting Review 70, pp. 135-150. Geweke, J., and M.Keane. 2001, “Computationally Intensive Methods for Integration in Econometrics,” Handbook of Econometrics 5, pp. 3463-3568. Graham, J., C. Harvey, and S. Rajgopal, 2005, “The Economic Implications of Corporate Financial Reporting,” Journal of Accounting and Economics 40, pp. 3-73. Haaijer, R., W. Kamakura, and M. Wedel. 2000, “Response Latencies in the Analysis of Conjoint Choice Experiments,” Journal of Marketing Research 37, pp. 376-382. Hausman, J. A., and D. A. Wise, 1978. “A Conditional Probit Model for Qualitative Choice: Discrete Decisions Recognizing Interdependence and Heterogeneous Preferences,” Econometrica 46, pp. 403-426. Healy, P., A. Hutton, and K. Palepu, 1999, “Stock Performance and Intermediation Changes Surrounding Sustained Increases in Disclosure,” Contemporary Accounting Research 16, pp. 485-520. Healy, P., and K. Palepu, 2001, “Information Asymmetry, Corporate Disclosure, and the Capital Markets: A Review of the Empirical Disclosure Literature,” Journal of Accounting and Economics 31, pp. 405-440. Heitzman, S., C. Wasley, and J. Zimmerman, 2010, “The Joint Effects of Materiality Thresholds and Voluntary Disclosure Incentives on Firms’ Disclosure Decisions,” Journal of Accounting and Economics 50, pp. 127-178. 35
Heminway, J., 2003. “Materiality Guidance in the Context of Insider Trading: A Call for Action” 52 AM. U. L. REV. 1131 reprinted at 36 SEC. L. REV. 448 (2004). Hirst, D., L. Koonce, and S. Venkataraman, 2008. Management Earnings Forecast: A Review and Framework. Accounting Horizons 22, pp. 315-38. Huddart, S., B. Ke, and C. Shi, 2007, “Jeopardy, Non-public Information, and Insider Trading Around SEC 10-K and 10-Q Filings,” Journal of Accounting and Economics 43, pp. 336. Hutton, A., G. Miller, and D. Skinner, 2003, “The Role of Supplementary Statements with Management Earnings Forecasts,” Journal of Accounting Research 41, pp. 867-890. Jagolinzer, A., 2009, “SEC Rule 10b5-1 and Insiders’ Strategic Trade,” Management Science 55, pp. 224–239. Kasznik, R., and B. Lev, 1995, “To Warn or Not to Warn: Management Disclosures in the Face of an Earnings Surprise,” Accounting Review 70, pp. 113-134. Ke, B., S. Huddart and K. Petroni, 2003, “What Insiders Know about Future Earnings and How They Use It: Evidence from Insider Trades,” Journal of Accounting and Economics 35, pp. 315-346. Keane, M. 1992, “A Note on Identification in the Multinomial Probit Model,” Journal of Business and Economic Statistics 10, pp. 193-200. King, R., G. Pownall and G. Waymire, 1990, “Expectations Adjustment via Timely Management Forecasts: Review, Synthesis and Suggestions for Future Research,” Journal of Accounting Literature 9, pp. 113–144. Lakonishok, J. and I. Lee, 2001. Are insider trades informative? Review of Financial Studies 14, 79-111. Lang, M. and R. Lundholm, 1993, “Cross-sectional Determinants of Analyst Ratings of Corporate Disclosures,” Journal of Accounting Research 31, pp. 246-271. Lang, M. and R. Lundholm, 2000, “Voluntary Disclosure During Equity Offerings: Reducing Information Asymmetry or Hyping the Stock?” Contemporary Accounting Research 17, pp. 623-663. Lansford, B., B. Lev, and J. Tucker, 2009, “Causes and Consequences of Disaggregating Earnings Guidance,” Working paper Northwestern University. Larcker, D., S. Richardson, I. Tuna, 2007, “Corporate governance, accounting outcomes, and organizational performance,” Accounting Review 82, pp. 913-1008. Lennox, C., and C. Park, 2006, “The Informativeness of Earnings and Management’s Issuance of Earnings Forecasts,” Journal of Accounting and Economics 42, pp. 439-458. Leuz, C and R. Verrecchia, 2000, “The Economic Consequences of Increased Disclosure” Journal of Accounting Research 38, 91-124. Lev, B. and S. Penman, 1990, “Voluntary Forecast Disclosure, Nondisclosure, and Stock Prices,” Journal of Accounting Research 28, pp. 49-76. Loss, L., and J. Seligman, 2004, Securities Regulation, 5th ed. (Aspen Publishers). 36
McCulloch, R. E., N. G. Polson, and P. E. Rossi, 2000, “A Bayesian analysis of the multinomial probit model with fully identified parameters,” Journal of Econometrics 99, pp. 173-193. McNichols, M., 1989, “Evidence of Informational Asymmetries from Management Earnings Forecasts and Stock Returns,” Accounting Review 64, pp. 1-27. Noe, C., 1999, “Voluntary Disclosures and Insider Transactions,” Journal of Accounting and Economics 27, pp. 305-326. Penman, S., 1980, “An Empirical Investigation of the Voluntary Disclosure of Corporate Earnings Forecasts,” Journal of Accounting Research 18, pp. 132-160. Piotroski, J., and D.Roulstone. 2005, “Do Insider Trades Reflect both Contrarian Beliefs and Superior Knowledge about Future Cash Flow Realizations?” Journal of Accounting and Economics 39, pp. 55-81. Pownall, G., C. Wasley, and G. Waymire, 1993, “The Stock Price Effects of Alternative Types of Management Earnings Forecasts,” Accounting Review 68, pp. 896-912. Prentice, R., 1999, “The Internet and Its Challenges for the Future of Insider Trading Regulation” Harvard Journal of Law & Technology 12 (Winter), pp. 265-364. Rees, L., A. Srivastava, and S. Tse. 2008, “An Examination of the Accuracy and Usefulness of Management Earnings Guidance around Stock Option Grants,” Working paper Texas A&M. Rogers, J., 2008, “Disclosure Quality and Management Trading Incentives,” Journal of Accounting Research 46, pp 1265-1296. Rogers, J. and P. Stocken, 2005, “The Credibility of Management Forecasts,” Accounting Review 80, pp. 1233-1260. Rogers, J. and A. Van Buskirk, 2009, “Bundled Forecasts and Selective Disclosure of Good News,” Working Paper University of Chicago. Roulstone, D., 2003, “The Relation Between Insider-trading Restrictions and Executive Compensation,” Journal of Accounting Research 41, pp. 525-551. Skaife, H., D. Veenman, and D. Wangerin, 2010, “Internal Control over Financial Reporting and Managerial Rent Extraction: Evidence from the Profitability of Insider Trading,” Working paper University of Wisconsin. Seyhun, N., 1992, “The Effectiveness of the Insider-Trading Sanctions,” Journal of Law & Economics 35, pp. 149- 182. Skinner, D., 1994, “Why Firms Voluntarily Disclose Bad News,” Journal of Accounting Research 32, pp. 38-60. Skinner, D., 1997, “Earnings Disclosures and Stockholder Lawsuits,” Journal of Accounting and Economics 23, pp. 249-282. Tang, M., 2011 “What Guides the Guidance? An Empirical Examination of the Dynamic Nature Of Earnings Guidance in the Post Reg-FD Period,” (May 2011) working paper University of Rochester.
37
Trueman, B., 1986, “Why do managers voluntarily release earnings forecasts?” Journal of Accounting and Economics 8, pp. 53–72. Verrecchia, R., 1983, “Discretionary Disclosure,” Journal of Accounting and Economics 5, pp. 179-94. Verrecchia, R., 2001, “Essays on Disclosure,” Journal of Accounting and Economics 32, pp. 97180. Verrecchia, R., and J. Weber, 2006, “Redacted Disclosure,” Journal of Accounting Research 2006, pp. 791-814. Yermack, D.,1997, “Good Timing: CEO Stock Option Awards and Company News Announcements,” Journal of Finance 52, pp. 449-476.
38
Table 1 Steps to Classify Management Earnings Forecasts into Cost of Capital, Disclose or Abstain, and Opportunistic Samples
Sample
Observed Characteristic Ca
Cost of Capital (COC) Sample
Cb
No Insider Trading (purchases or sales) in the [-30,+30] interval. Insider purchases in the [-30,-1] interval prior to a bad news MF or insider sales in the [-30,-1] interval prior to a “good news” MF. Total COC MFs
Disclose or Abstain (DOA) Sample
Da
All MFs with insider trading (i.e., either selling or buying) in the 30 days after a MF’s release date except those classified as opportunistic MFs (see Ob and Oc below for the criteria used to identify opportunistic MFs).
Total DOA MFs
Opportunistic (OPP) Sample
Oa
Ob
Oc
Insider purchases in the 30 days prior to a “good news” MF or insider sales in the 30 days prior to a “bad news” MF. Abnormal returns on MF announcement date ≥+5% followed by insider sales in the [+1,+30] interval and MF > Actual EPS (i.e., the MF is optimistic). Abnormal returns on MF announcement date ≤ -5% followed by insider purchases in the [+1,+30] and MF < Actual EPS (i.e., the MF is pessimistic). Total OPP MFs
Overall Total
N
Percent of Sample
16,211
82.9%
3,351
17.1%
19,562
100.0%
63.4%
7,145
100.0%
23.1%
7,145
100.0%
23.1%
3,343
80.2%
601
14.4%
225
5.4%
4,169
100.0%
30,876 39
Percent of Sample
13.5%
100.0%
Table 2 Descriptive Statistics of Management Earnings Forecasts (MFs) The table reports descriptive statistics for MFs classified into Cost of Capital (COC), Disclose or Abstain (DOA), and Opportunistic (OPP) samples. Panel A (B) organizes the data by firm-quarters (by the number of MFs). The table is based on a sample of 64,825 firm-quarter observations from the period 1998-2010. Panel A presents the number of firm-quarter observations in each category by year and quarter. Quarters with (without) MFs are firmquarters that have at least one (no) MF issued during the period from the last quarter’s earnings announcement date up through one day before the current quarter’s earnings announcement. A quarter is assigned to the COC, DOA, OPP sample if all MFs issued in the quarter are classified as COC, DOA, or OPP, respectively. See section 3 and Table 1 for a description of the process used to assign MFs to the COC, DOA, and OPP samples. A quarter is classified as Bundled (Unbundled) if all MFs in the quarter are made (not made) within [-1, +1] trading days of an earnings announcement. If a quarter contains both bundled and non-bundled MFs, it is classified as Both. Panel A:
Year
Quarters with Management Earnings Forecasts (MFs) COC DOA OPP Total MFs MFs MFs MFs N % N % N % N
Quarters without MFs N
1998
212
65.2%
60
18.5%
53
16.4%
324
2,293
1999
270
61.6%
102
23.1%
67
15.2%
442
2,670
2000
404
67.4%
113
18.9%
82
13.7%
597
2,801
2001
734
67.2%
199
18.2%
158
14.4%
1,094
2,556
2002
908
68.3%
266
20.1%
153
11.5%
1,325
2,989
2003
987
67.6%
334
23.0%
139
9.6%
1,455
3,468
2004
1,153
65.4%
442
25.2%
167
9.5%
1,757
3,650
2005
1,135
63.7%
432
24.2%
213
11.9%
1,786
4,009
2006
1,234
64.6%
460
24.2%
213
11.2%
1,901
4,015
2007
1,161
63.8%
421
23.1%
238
13.1%
1,821
4,220
2008
1,257
69.5%
363
20.0%
188
10.4%
1,812
4,433
2009
1,107
69.9%
344
21.7%
133
8.4%
1,586
5,072
2010
920
68.5%
293
21.7%
133
9.9%
1,349
5,483
1
2,722
65.2%
971
23.2%
485
11.6%
4,185
11,033
2
2,858
66.8%
945
22.1%
476
11.1%
4,280
12,376
3
3,042
67.7%
951
21.2%
495
11.0%
4,487
12,153
4
2,860
66.5%
962
22.4%
481
11.2%
4,297
12,097
Bundled Only
7,143
62.2%
3,082
80.3%
1,110
57.3%
11,335
-
Unbundled Only
3,197
27.8%
636
16.6%
738
38.1%
4,571
-
Both
1,142
9.9%
118
3.1%
88
4.5%
1,348
-
Total
11,482
66.5%
3,836
22.2%
1,936
11.2%
17,254
47,571
Quarter
Timing relative to earnings announcement
40
Table 2 continued, Panel B presents the number of MFs in each sample by forecasting horizon (i.e., quarterly or annual), form (i.e., point, range, end, or qualitative estimates), and forecast news. See section 3 and Table 1 for a description of the process used to assign MFs to the COC, DOA, and OPP samples. We use the approach in Anilowski et al (2007) to classify MFs into good, bad, neutral, and mixed news. Panel B:
Categories of Management Earnings Forecasts (MFs) COC MFs
Horizon:
DOA MFs
OPP MFs
Total
N
%
N
%
N
%
N
Annual
11,384
58.2%
3,943
55.2%
2,455
58.9%
17,782
Quarterly
8,178
41.8%
3,202
44.8%
1,714
41.1%
13,094
Point
2,720
13.9%
1,054
14.8%
674
16.2%
4,448
Range
15,606
79.8%
5,643
79.0%
3,265
78.3%
24,514
End
664
3.4%
263
3.7%
115
2.8%
1,042
Qualitative
572
2.9%
185
2.6%
115
2.8%
872
Bad
10,018
51.2%
3,151
44.1%
2,257
54.1%
15,426
Good
7,169
36.6%
3,054
42.7%
1,370
32.9%
11,593
Neutral
2,301
11.8%
915
12.8%
531
12.7%
3,747
Mixed
74
0.4%
25
0.3%
11
0.3%
110
Total
19,562
63.4%
7,145
23.1%
4,169
13.5%
30,876
Form:
News:
41
Table 3 First Construct Validity Test: Variables Hypothesized to Predict Management Earnings Forecast (MF) Disclosure The table lists variables hypothesized to predict MF Disclosure in the Cost of Capital (COC), Disclose or Abstain (DOA), and Opportunistic (OPP) MF samples (predicted signs are in parentheses, see appendix A for variable definitions).
COC MF Sample ERC (+) EARN_SURPRISE (+) SIZE (+) MB (-) RTNVOL (-) EARNVOL (-) HI_TECH (+) REGULATION (-) HABITUAL (+)
DOA MF Sample ERC (+) EARN_SURPRISE (+)
42
OPP MF Sample RESTATE (+) BACKDATE (+)
Table 4 First Construct Validity Test: Descriptive Statistics for the Variables Used in the Multinomial Probit Model for Management Earnings Forecasts (MFs) Classified into Cost of Capital (COC), Disclose or Abstain (DOA), and Opportunistic (OPP) Samples. The table is based on a sample of 64,825 firm-quarter observations for the period 1998-2010. Quarters with (without) MFs are firm-quarters with at least one (no) MF issued during the period from the last quarter’s earnings announcement date up to one day before the current quarter’s earnings announcement. See section 3 and Table 1 for a description of the process used to assign MFs to the COC, DOA, and OPP samples and Appendix A for variable definitions. t-tests (Wilcoxon tests) are used to test for differences. Since the Wilcoxon test is a test for a difference between the distributions of two samples, the asterisks next to the median value signify that rather than a difference in medians. ‡’s under COC are used to indicate whether the COC sample differs from the DOA sample. *’s under DOA are used to indicate whether the DOA sample differs from the OPP sample. †’s under OPP are used to indicate whether the OPP sample differs from the COC sample. *, **, *** indicate p-values of 10%, 5%, and 1%, respectively, two-tailed tests. Quarters with MFs COC MFs (N=11,482)
Variable
DOA MFs (N=3,836)
OPP MFs (N=1,936)
ERC
Mean Median
3.376 4.000
‡‡‡ ‡‡‡
3.529 4.000
*** ***
3.393 4.000
EARN_SURPRISE
Mean Median Mean Median Mean Median Mean Median Mean Median Mean Median Mean Median Mean Median Mean Median Mean Median Mean Median
3.021 3.000 6,356 1,569 7.456 7.358 2.915 2.189 0.00093 0.00062 0.00077 0.00007 0.335 0.000 0.073 0.000 0.456 0.000 0.292 0.000 0.062 0.000
‡‡‡ ‡‡‡ ‡‡‡ ‡‡‡ ‡‡‡ ‡‡‡ ‡‡‡ ‡‡‡ ‡‡‡ ‡‡‡ ‡‡‡ ‡‡ ‡‡‡ ‡‡‡
2.797 3.000 8,117 1,855 7.670 7.526 3.160 2.452 0.00088 0.00057 0.00112 0.00008 0.376 0.000 0.065 0.000 0.520 1.000 0.266 0.000 0.082 0.000
*** *** *** *** *** *** *** ***
2.797 3.000 9,539 2,239 7.878 7.714 3.713 2.752 0.00086 0.00055 0.00093 0.00009 0.349 0.000 0.056 0.000 0.421 0.000 0.309 0.000 0.079 0.000
MVE LNMVE MB RTNVOL EARNVOL HI_TECH REGULATION HABITUAL RESTATE BACKDATE
‡‡‡ ‡‡‡ ‡‡‡ ‡‡‡ ‡‡‡ ‡‡‡ 43
* ** **
*** *** *** ***
Quarters without MFs (N=47,571) 2.843 3.000
††† ††† ††† ††† ††† ††† ††† ††† ††† ††† † †††
††† ††† ††† †††
†† †††
2.999 3.000 5,759 1,274 7.239 7.150 2.754 1.989 0.00113 0.00067 0.00115 0.00008 0.218 0.000 0.105 0.000 0.036 0.000 0.254 0.000 0.037 0.000
Table 5 First Construct Validity Test: Multinomial Probit Model of the Decision to Issue Management Earnings Forecasts (MFs) Classified into Cost of Capital, Disclose or Abstain (DOA), and Opportunistic Samples (with and without an earnings announcement). Quarters with (without) MFs are firm-quarters that have at least one (no) MFs issued during the period from last quarter’s earnings announcement date up through one day before the current quarter’s earnings announcement date. Only MFs issued at least 21 days before the forecasted fiscal period end are included. A quarter is defined as a Cost of Capital, DOA, or Opportunistic if all MFs issued in the quarter are classified as Cost of Capital, DOA, or Opportunistic, respectively. See section 3 and Table 1 for a description of the process used to assign MFs to the Cost of Capital, DOA, and Opportunistic samples and Appendix A for detailed variable definitions. ‡ denotes that the coefficient magnitude differs in the direction predicted in Table 3. Z-statistics for the multinomial probit have been adjusted for heteroskedasticity and firm-specific clustering. Year-quarter indicators are included, but suppressed. *, **, *** indicate p-values of 10%, 5%, and 1%, respectively, two-tailed tests. Panel A:
Cost of Capital
Intercept ERC
+
EARN_SURPRISE
+
LNMVE
+
MB RTNVOL EARNVOL HI_TECH REGULATION HABITUAL RESTATE BACKDATE
─ ─ ─ + ─ +
-3.037 (-18.71) 0.070 (6.59) 0.056 (5.84) 0.059 (4.58) -0.0013 (-0.09) -116.087 (-7.29) -7.793 (-1.79) 0.339 (8.33) -0.176 (-2.54) 2.516 (59.18) 0.122 (2.95) 0.129 (1.58)
DOA
*** ‡ *** ‡ *** *** ‡‡ ‡‡ *** ‡ * ‡ *** ** ‡‡ *** ***
Chi-square (p-value) N 44
+ +
-3.960 (-18.32) *** 0.096 ‡ (7.46) *** -0.001 (-0.11) 0.093 (6.39) *** -0.0005 (-0.22) -86.717 (-4.10) *** 8.483 (1.86) * 0.366 (8.00) *** -0.211 (-2.61) *** 2.466 (53.84) *** 0.041 (0.89) 0.251 (2.88) *** 7,582.19 (0.000) 64,825
Opportunistic
+ +
-4.079 (-18.75) 0.067 (5.10) -0.009 (-0.66) 0.116 (7.96) 0.0250 (4.07) -66.742 (-2.96) -8.260 (-1.81) 0.308 (6.04) -0.298 (-3.78) 2.197 (43.67) 0.154 (3.02) 0.230 (2.45)
*** ***
*** *** *** * *** *** *** ‡‡ *** ‡ **
Table 5 (continued) Panel B: Wald tests for equality of coefficients: (1) ERC, EARN_SURPRISE: DOA vs. Opportunistic Cost of Capital vs. Opportunistic
Chi-squarea
=
4.21
p-value
=
0.121
Chi-squarea
=
27.24 p-value
=
0.000
(2) LNMVE, MB, RTNVOL, EARNVOL, HI_TECH, REGULATION, HABITUAL: Cost of Capital vs. DOA Chi-squareb = 25.68 p-value = 0.001 Cost of Capital vs. Chi-squareb = 109.42 p-value = 0.000 Opportunistic (3) RESTATE, BACKDATE: Opportunistic vs. DOA Opportunistic vs. Cost of Capital a b
Chi-squarea
=
5.92
p-value
=
0.052
Chi-squarea
=
1.89
p-value
=
0.388
Chi-square test is based on 2 degrees of freedom Chi-square test is based on 7 degrees of freedom
45
Table 6 Second Construct Validity Test: Replication of Cheng and Lo (2006) and the effect of expected insider trading on management forecast frequency The table is based on a sample of 82,643 firm-calendar quarters for firms at the intersection of Thompson Reuter Institutional (13F) and First Call CIG data for the sample period 1995-2002. The dependent variable is an ordinal variable that takes value 1, 0, or -1 if the net news frequency in the current quarter is higher than, equal to, or lower than the net news frequency in the last quarter, respectively. The net news frequency is defined as the number of good news MFs minus the number of bad news MFs issued during a quarter. We classify a MF as good news or bad news based on the sign of the three-day size-adjusted return around the MF. Column (2) uses all MFs while columns (3), (4), and (5) use only COC, DOA, and OPP MFs, respectively. See section 3 and Table 1 for a description of the process used to assign MFs to the COC, DOA, and OPP samples. The predicted IP (IS) is the expected insider purchases (insider sales) estimated from a first-stage model where the amount of insider trading for the current quarter is regressed on determinants (i.e., firm size, growth opportunities, stock returns, ROE, option grants, insider trading) measured from the prior quarter (see Cheng and Lo, 2006). To be consistent with Cheng and Lo (2006) we include trades by all directors and officers. “RET Indicator” is an indicator variable that takes on a value of one if a firm’s abnormal return during the quarter is positive, and zero otherwise. “Change in RET Indicator” is the difference between the current quarter and last quarter’s RET Indicator. “Future change in RET Indicator” is the difference between the leading one quarter and current quarter’s “RET Indicator.” The table reports the mean coefficients and pseudo R2 from 30 quarterly ordered logit regressions. Z-statistics are calculated using the Fama-MacBeth approach with a Newey-West correction for serial correlation at two lags. *, **, *** indicate p-values of 10%, 5%, and 1%, respectively, one-tailed tests. Cheng and Lo (2006) Table 4 Panel A Column (4) (1) Predicted IP
-0.0365
Predicted IS
0.0013
***
Net News Frequency All MFs (2) -0.0240 (-3.07) *** 0.0078 (0.99)
Change in RET Indicator
0.4224 ***
Future Change in RET Indicator Mean pseudo R2 N
-0.0369 0.0264 77,106
Net News Frequency – COC MFs (3) 0.0042 (-0.52) 0.0022
Net News Frequency – DOA MFs (4)
Net News Frequency – OPP MFs (5)
-0.0301 (-1.17) 0.0008
-0.0593 (-2.63) *** 0.0207
(0.36)
(0.06)
0.4056 0.3403 (27.49) *** (27.81) *** -0.0505 0.0340 (-3.55) *** (-2.35) *** 0.0123
0.3351 (9.73) *** -0.0635 (-1.75) **
0.0085
0.0110 82,643
46
(1.67) ** 0.4281 (9.59) *** -0.0311 (-1.00) 0.0154
Table 7 Third Construct Validity Test: Future Stock Return Performance Subsequent to Insider Trading for Management Earnings Forecasts (MFs) Classified into Opportunistic and Disclose or Abstain (DOA) Samples. The table is based on a sample of 6,067 DOA MFs and 3,580 opportunistic MFs from the period 1998-2008. See section 3 and Table 1 for a description of the process used to assign MFs to the DOA and Opportunistic samples. The table reports future abnormal returns for the one- and two-year period subsequent to the [-30, +30] trading day window centered on the MF release date. For each MF we calculate the net insider trading during the window. Specifically, for DOA MFs, and OPP MFs in the Ob and Oc subgroups, the window is the [+1, +30] trading day window after the MF release date. For OPP MFs in the Oa subgroup, the window is the [-30, -1] trading day window before the MF release date. A MF is labeled as Insider Purchase (Insider Sale) if the net insider trading in the window is purchase (sale). Future Abnormal Return is calculated as the buy-and-hold return minus the buy-and-hold return of the corresponding Fama-French 25 portfolios formed on size and book-to-market. 12 months (24 months) refers to the first twelve (twenty-four) months subsequent to the [-30,+30] trading window centered on the MF release date. Future abnormal returns are winzorized at the top and bottom 1%. Wilcoxon tests are used to test for differences. *, **, *** indicate p-values of 10%, 5%, and 1%, respectively, one-tailed tests. Signed rank tests are used to test whether medians are significantly different from zero. †, ††, ††† indicate p-values of 10%, 5%, and 1%, respectively, one-tailed tests.
Median Future Abnormal Return in 12 and 24 Months Subsequent to DOA and Opportunistic MFs Insider Purchase
DOA MFs Insider Sale
Difference
Opportunistic MFs Insider Purchase Insider Sale
Difference
Year
(N=939)
(N=5,128)
(IP-IS)
(N=495)
(N=3,085)
(IP-IS)
12 Months
-0.03%
-1.34%†
1.31%
4.45%†††
-1.96%†††
6.41%***
24 Months
-0.78%
-2.08%††
1.30%
6.27%†††
-2.81%†††
9.08%***
47
Table 8 Fourth Construct Validity Test: Replication of Ajinkya et al. (2005) and the effect of outside directors and institutional holdings on the decision to issue management earnings forecasts The table is based on a sample of 5,413 firm-years for firms with at least three analysts. The sample period is 1997-2002. The dependent variable takes value one if a firm issues at least one annual earnings forecast before the fiscal year end, and zero otherwise. A firm-year is defined as a Cost of Capital (COC), Disclose or Abstain (DOA), or Opportunistic (OPP) if the last annual MF issued during the year is classified as DOA, OPP or COC, respectively. See section 3 and Table 1 for a description of the process used to assign MFs to the COC, DOA, and OPP groups. Following Ajinkya et al. (2005) OUTDIR is the portion of directors that are not officers of the firm, INST is the portion of institutional ownership, LMVAL is logarithm of market value of equity at the beginning of the fiscal year, AUDIT is an indicator variable for Big 5 auditors, NUMEST is the number of analysts following the firm, DISPFOR is the standard deviation of analysts’ forecasts deflated by the median forecast, LITGATE is an indicator variable for high litigation risk industries (i.e., SIC 28332836, 8731-8734, 3570-3577, 7370-7374, 3600-3674, 5200-5961), MKBK is the book to market ratio at the beginning of the fiscal year, LOSS is an indicator variable for loss firms, NEWS is an indicator variable for profit-declining firms, EARNVOL is the standard deviation of earnings over the previous 12 quarters deflated by median asset value, BETA is the equity beta estimated over the fiscal year, and FD is an indicator variable for the post-Reg FD period. As in Ajinkya et al. (2005) Z-statistics have been adjusted for heteroskedasticity and year indicators are excluded. *, **, *** indicate p-values of 10%, 5%, and 1%, respectively, two-tailed tests.
Variables
Pred. Sign (1)
Intercept OUTDIR
+
INST
+
LMVAL
+
AUDIT
+
NUMEST
+
DISPFOR
-
LITIGATE
?
MKBK
-
LOSS
-
NEWS
-
EARNVOL
-
BETA
-
FD
+
Ajinkya et al. (2005) (2) -1.580 (-15.23) 0.003 (4.05) 0.009 (12.94) 0.047 (3.642) 0.033 (0.93) 0.020 (5.64) -0.034 (-0.20) 0.191 (4.95) 0.010 (1.51) -0.420 (-9.39) 0.021 (0.69) 0.092 (0.88) -0.151 (-5.83) 0.589 (17.88)
*** *** *** ***
***
***
***
*** ***
Pseudo R2
Replication Probit (1997-2002) (3) -2.296 (-10.65) 0.006 (4.24) 0.003 (3.30) 0.157 (9.16) 0.093 (0.6) -0.005 (-1.03) 0.019 (0.65) 0.022 (0.50) -0.001 (-2.50) -0.235 (-4.33) -0.089 (-2.35) 0.528 (1.94) -0.260 (-6.83) 0.678 (18.07) 0.090
*** *** *** ***
** *** ** * *** ***
Wald Chi2 Number of observations
Multinomial Probit (1997-2002) COC (4) -3.280 (-10.34) 0.011 (4.63) 0.003 (2.20) 0.191 (7.42) 0.072 (0.32) -0.009 (-1.37) 0.047 (1.11) -0.023 (-0.34) -0.008 (-2.41) -0.315 (-3.84) -0.135 (-2.41) 0.491 (1.18) -0.346 (-6.02) 0.954 (16.95)
*** *** ** ***
** *** **
*** ***
DOA (5) -4.737 (-9.50) 0.006 (1.89) 0.006 (3.23) 0.249 (7.95) 0.530 (1.32) -0.004 (-0.54) -0.099 (-0.87) 0.082 (1.00) 0.000 (-0.32) -0.285 (-2.64) -0.055 (-0.78) 1.043 (2.54) -0.343 (-4.93) 0.762 (10.76) 514.0
7,745
5,413
5,413
48
*** * *** ***
***
** *** ***
OPP (6) -3.787 (-8.15) 0.003 (0.94) 0.005 (2.65) 0.191 (5.23) -0.074 (-0.23) 0.001 (0.13) -0.037 (-0.32) 0.145 (1.55) -0.001 (-0.99) -0.297 (-2.62) -0.110 (-1.36) 0.508 (1.18) -0.230 (-3.01) 0.698 (8.76)
***
*** ***
***
*** ***
