the effects of financial derivatives on analyst coverage ... annual...firm’s derivatives activity...
TRANSCRIPT
The effects of financial derivatives on analyst coverage decisions
Hye Sun Chang† [email protected]
Michael Donohoe*‡
Theodore Sougiannis* [email protected]
†Singapore Management University
60 Stamford Road Singapore 178900
*University of Illinois at Urbana-Champaign 1206 S. Sixth Street, MC-706
Champaign, IL 61820
September 2017
Abstract
We investigate whether and how a firm’s use of derivatives influences analyst coverage decisions. Using a difference-in-differences design, we find that, relative to a matched control sample of non-users, analyst coverage for new derivatives users increases significantly after derivatives initiation. This increase in coverage is driven by analysts with less expertise, as reflected by career experience and All-Star status. We also find that the accuracy (dispersion) of earnings forecasts for new users decreases (increases) after derivatives initiation only when forecasts are issued by analysts with less expertise. Therefore, the evidence collectively suggests that, despite the notorious complexity of derivatives, analysts with less expertise begin covering new derivatives users, presumably to signal their talent, and, as a result, produce the less accurate and more dispersed earnings forecasts shown by prior research for these firms. Keywords: derivatives; economic complexity; reporting complexity; hedging; sell-side analysts;
earnings forecasts JEL Classification: G29; G32; M41 ‡Corresponding author. We appreciate helpful comments from Rashad Abdel-khalik, Andrew Bauer, Raluca Chiorean, Will Ciconte, Brooke Elliott, Simeon Ketterer, Marcus Kirk, Laura Li, Pete Lisowsky, Sean McGuire, Michael Mayberry, Mark Peecher, Jenny Tucker, Jim Vincent, participants at the 23rd Annual Conference of the Multinational Finance Society, the SMU/NUS/NTU Junior Faculty Research Conference, and workshop participants at the University of Illinois at Urbana-Champaign. Special thanks Mitchell Brown, Alex Menter, and Alexander Van Duch for research assistance. Donohoe and Sougiannis gratefully acknowledge financial support from the PricewaterhouseCoopers Faculty Fellowship and KPMG Professorship, respectively.
1
The effects of financial derivatives on analyst coverage decisions
1. Introduction
Derivatives are an increasingly prevalent form of corporate risk management, with nearly
two-thirds of U.S. non-financial firms participating in the $710 trillion derivatives market (Bank
for International Settlements 2013). These instruments are notoriously complex, however, as their
value can be linked to virtually any underlying asset/liability, including other derivatives, to fulfill
diverse objectives. Expert groups claim that many firms using derivatives do not apply the requisite
accounting rules correctly or consistently, making it “next to impossible” for investors to assess a
firm’s derivatives activity from its financial reports (Kawaller 2004, 29). In such cases, investors
value the services of sell-side analysts, who simplify and convey complex information (Lawrence
et al. 2014). However, because recent research finds that analysts regularly misjudge the earnings
implications of firms’ derivatives activity (Chang et al. 2016), it is unclear whether and under what
circumstances the complexity of derivatives influences analysts’ decisions to cover a firm in the
first place. We fill this void in the literature by answering these two important questions.
Analyst coverage decisions are influenced by client demand for information, opportunities
to signal talent, and economic incentives (Ramnath et al. 2008; Brown et al. 2015). The complexity
of transactions and financial information resulting from a firm’s choice to use derivatives can affect
each of these factors, and thus, coverage decisions. Further, derivatives contracts, the markets in
which they trade, and requisite reporting are extraordinarily complex (Ryan 2007). The Financial
Accounting Standards Board (FASB) has issued a series of standards to help simplify derivatives
accounting, but some of these efforts have been criticized as the “poster child of complexity”
(Leone 2007) and a “labyrinth of processes and documentation” (Valladares 2014). Prior research
considers if firm-level complexities, such as the level of intangibles (Barth et al. 2001) and annual
2
report readability (Lehavy et al. 2011), affect analyst coverage. However, empirical evidence as to
whether and when “shocks” to firm-level complexity resulting from a firm’s endogenous choice
to begin using sophisticated financial instruments influence analyst coverage is absent from the
literature. These questions are important given the interplay between financial reporting and the
information analysts provide to other market participants (Beyer et al. 2010; Chen et al. 2010).
Bhushan (1989) shows that the equilibrium number of analysts covering a firm lies at the
intersection of aggregate demand and supply curves for analyst services. Although the inherent
complexity of derivatives is very likely to increase investor demand for analyst services, the effects
on supply is an empirical question. On the one hand, analysts have financial incentives to cover
derivatives users. In particular, a firm’s use of derivatives could (1) result in more valuable advice
for analysts to sell to investors due to complex derivatives information (e.g., Lang and Lundholm
1996); (2) improve the predictability of firm performance (Zhang 2009), leading to promotions
(Hong et al. 2000); and (3) provide opportunities to earn higher trading commissions (Aretz and
Bartram 2010). Alternatively, the complexity of derivatives accounting (Ryan 2007), difficulty of
obtaining and analyzing derivatives information (Kawaller 2004), and potential for derivatives to
attract sharp criticism (Donohoe 2015a) could deter analyst coverage of derivatives users. Thus,
the overall effect depends on the competing demand and supply factors.
Using the focused setting of new derivatives users and difference-in-difference design with
a propensity score matched control sample, we find an increase in analyst coverage for firms that
begin using derivatives during 1998-2011. More specifically, analyst coverage for new derivatives
users increases by 10% after derivatives initiation relative to non-user control firms. This effect is
comparable to that of other major changes in firms’ financial reporting practices (Tan et al. 2011).
As such, these results suggest that analysts respond to the increased demand for their services due
3
to the complexity of derivatives information in firms’ financial reports.
We next consider if the increase in analyst coverage for new derivatives users is driven by
analyst expertise. Analysts with more career experience as well as those attaining “All-Star” status
possess superior skill and knowledge (Ramnath et al. 2008). Such analysts could be more likely to
respond to the increased investor demand for analyst services due to the complexity of derivatives.
However, recent research finds that analysts (on average) produce less accurate and more dispersed
earnings forecasts for new derivatives users (Chang et al. 2016). Thus, it is possible that analysts
with more expertise might instead recognize and avoid such challenges, whereas analysts with less
expertise might view the complexity of derivatives as an opportunity to signal their talent.
Consistent with the latter, we find that the increase in analyst coverage for new derivatives users
is driven by analysts with less expertise, as reflected by experience (i.e., number of years issuing
earnings forecasts) and All-Star status by Institutional Investor Magazine (Bagnoli et al. 2008).
We also examine analyst coverage at a more granular level by creating transition matrices
that trace the proportion of analysts covering new derivatives users before derivatives initiation
and the proportion ceasing or initiating coverage of new users upon and after derivatives initiation,
all by analyst expertise. The matrices identify if the observed change in coverage is due to analysts
with less expertise initiating coverage or those with more expertise ceasing coverage, which can
ultimately affect a firm’s information environment. The results confirm that the overall net change
in coverage for new derivatives users is driven by analysts with less expertise initiating coverage
rather than analysts with more expertise ceasing coverage.
In another set of tests, we examine if analyst expertise explains the less accurate and more
dispersed earnings forecasts for new derivatives users shown by Chang et al. (2016). We find that
the accuracy (dispersion) of earnings forecasts for new derivatives users decreases (increases) after
4
derivatives initiation only when forecasts are issued by analysts with low expertise. Therefore, the
evidence collectively indicates that, despite the renowned complexity of derivatives, analysts with
less expertise begin covering new derivatives users (likely to signal talent) and, in turn, produce
the less accurate and more dispersed earnings forecasts shown by Chang et al. (2016) for these
firms. Lastly, we perform structural break tests to validate that the complexity of derivatives drives
our results, as well as other analyses to mitigate other explanations such as forecast frequency,
ineffective management of risk, and firm complexity and size.
This study contributes to growing literatures on analysts (Ramnath et al. 2008; Beyer et al.
2010) and derivatives (Aretz and Bartram 2010) in numerous ways. First, anecdotal and empirical
evidence suggests that both practitioners (Holland and Glasgall 1994) and investors (Koonce et al.
2005) struggle to understand the risks and rewards of even mundane derivatives. We contribute by
examining how analysts respond to increased demand for their services due to the stark economic
and financial reporting complexity of derivatives. Second, recent research finds that, despite their
financial expertise, analysts misjudge the earnings implications of derivatives (Chang et al. 2016).
We show that the complexity of derivatives—a firm-level attribute—influences analysts’ decisions
to cover a firm in the first place. We also show that expertise—an analyst-level attribute—drives
coverage of new derivatives users and shapes analysts’ earnings forecast properties for such firms.
Finally, we answer the call for research on the interplay between financial reporting and analysts
(Beyer et al. 2010) by providing economic insight into how the institutional and regulatory setting
of derivatives interacts with market participants, and the relation between analysts and corporate
disclosures, which is critical for understanding firms’ information environments (Chen et al. 2010).
Section 2 provides a brief background, and Section 3 develops our hypothesis. Sample
selection and research design are described in Sections 4 and 5, respectively. Section 6 reports the
5
main results, Section 7 discusses additional analyses, and Section 8 concludes.
2. Background and related literature
2.1 Analyst coverage
Sell-side analysts are important information intermediaries in capital markets. As financial
experts, they produce earnings and cash flow forecasts, stock recommendations, and other reports
that help investors make resource allocation decisions. As such, analysts can help improve market
efficiency (Healy and Palepu 2001; Barth and Hutton 2004) and enhance the visibility, liquidity,
and value of firms they cover (Merton 1987; Brennan and Subrahmanyam 1995).
The drivers of analyst coverage decisions often relate to the costs and benefits accruing to
analysts and their brokerages (Ramnath et al. 2008). For instance, analysts tend to cover profitable
firms (McNichols and O’Brien 1997) and those requiring less effort to evaluate (Bhushan 1989;
Lang and Lundholm 1996; Botosan and Harris 2000; De Franco et al. 2011). They are also more
likely to cover firms with higher trading volume as such firms generate larger commissions (Irvine
2004). However, a recent survey of 365 sell-side analysts suggests client demand for information
about a firm is the primary driver of coverage decisions, with little consideration given to financial
reporting attributes (e.g., predictability) that would enhance forecast accuracy (Brown et al. 2015).
Coverage decisions are also influenced by analyst-level factors, such as affiliation (Lin and
McNichols 1998), location (Tan et al. 2011), and personal traits (Tamura 2002). One well-studied
personal trait is analyst expertise, which has been measured by prior research in a number of ways,
including the Institutional Investor All-American Research Team, the Wall Street Journal All-Star
Analyst list, “celebrity” status based on media coverage, and career length (Ramnath et al. 2008).
Several studies find that more experienced analysts have more accurate earnings forecasts (Maines
et al. 1997; Mikhail et al. 1997; Clement 1999), and more accurate earnings forecasts increase the
6
likelihood of being ranked as an All-Star analyst (Leone and Wu 2007). All-Star analysts are also
more likely to initiate coverage of firms that have a relationship with their brokerage, and cease
coverage of firms for which they produce less accurate earnings forecasts (Clarke et al. 2007).
Prior studies consider if information asymmetry and inherent uncertainty influence analyst
coverage. For instance, Barth et al. (2001) find that firms with higher levels of intangibles, which
require effort and expertise to evaluate, have higher analyst coverage. Tan et al. (2011) examine if
International Financial Reporting Standards (IFRS) reduce information gathering and processing
costs for analysts such that learning a new set of standards is not an impediment to covering foreign
firms. Lehavy et al. (2011) find greater coverage among firms with less readable annual reports.
However, prior research does not consider whether and how a substantial change in both
economic and financial reporting complexity influences analyst coverage decisions, overall or by
analyst expertise. Further, there is no empirical evidence on whether derivatives, an increasingly
common and infamously complex form of financial risk management, influences analyst coverage.
Although recent research finds a reduction in analyst earnings forecast accuracy after a firm begins
using derivatives (Chang et al. 2016), whether and how the complexity of derivatives influences
analysts’ decisions to cover a firm in the first place are open empirical questions that we consider.
2.2 Derivatives accounting and reporting
Firms routinely engage in risk management practices to insulate cash flow and earnings
from unfavorable changes in risk exposures, including interest rates, foreign exchange rates, and
commodity prices. While there are many techniques for managing risk, corporate use of derivatives
for this purpose is increasingly common (Bartram et al. 2009). A derivative is a contract or security
deriving its value based on its relation to something else, often referred to as the “underlying.” The
7
underlying is often another financial instrument or economic good, but can be almost anything.1
Derivatives are an integral part of the global economy, with estimates of notional market size
above $710 trillion and derivatives usage by nearly 66% of U.S. non-financial firms (Bank for
International Settlements 2013).
As noted by Chang et al. (2016), complexity is the “state of being difficult to understand
and apply” (U.S. Securities and Exchange Commission 2008). In our context, complexity relates
to the difficulty in understanding the mapping of economic transactions and reporting standards
into financial statements (Peterson 2012). From an economic standpoint, derivatives are complex
because their value can be linked to virtually any underlying asset/liability, including other
derivatives, to fulfill a variety of objectives. Even mundane derivatives involve elaborate contracts
with ambiguous and evolving terminology. By virtue of such economic factors, financial reporting
for derivatives is also extraordinarily complex. Firms use considerable judgment to apply elaborate
accounting rules to sophisticated transactions with widely varying details (Ryan 2012).2 As a
result, many firms inaccurately and/or inconsistently account for derivatives in their financial
reports, complicating an assessment of these firms (Kawaller 2004). Along these lines, Chang et
al. (2016) show that the complexity of derivatives hinders the accuracy and increases the dispersion
of analysts’ earnings forecasts. We, however, focus on whether and under what circumstances the
inherent complexity of derivatives influences analysts’ decisions to cover a firm in the first place.
3. Hypothesis development
Theory suggests the equilibrium number of analysts covering a firm lies at the intersection
1 Derivatives generally fall into three categories: (1) options; (2) futures and forwards; and (3) swaps. Options involve the right, but not the obligation, to buy or sell the underlying at a set price within a specified period. A futures or forward contract involves an obligation to exchange the underlying at a future date for a specific price, and swaps are agreements to exchange a stream of payments based on some underlying over a predefined period. 2 See Appendix A of Chang et al. (2016) and Donohoe (2015a) for furthers details on derivatives accounting standards.
8
of aggregate demand and supply curves for analyst services (Bhushan 1989). Numerous firm-level
factors, such as size and stock return volatility, influence the intersection of these curves (Ramnath
et al. 2008). Because the economic and reporting complexities of derivatives plague investors as
they assess firms’ financial statements (Kawaller 2004), analysts can use their financial expertise
to provide guidance and advice. Thus, the use of derivatives—a firm-level attribute—could affect
the aggregate demand and supply for analyst services.
From a demand perspective, firms provide information about derivatives in their financial
reports. While this information should aid market participants in assessing firm performance and
risk, it is not always helpful. In particular, the complex and ambiguous accounting and disclosure
practices for derivatives make it difficult for investors to evaluate the risk and reward implications
of even basic instruments. For instance, Koonce et al. (2005) provide experimental evidence that
the labels firms use to describe derivatives in their financial statements cause investors to assess
economically equivalent instruments as having different risk. But, as information intermediaries,
analysts can use their financial skills and knowledge to help investors navigate the complexity and
comprehend the economic effects of firms’ derivatives. Along these lines, research finds that the
demand for analyst information is increasing in analysts’ ability to both relay and process financial
information, implying that investors value analysts because they simplify and convey information
(Lawrence et al. 2014).3 Thus, all else equal, a firm’s use of derivatives increases the demand for
analyst services and the equilibrium number of analysts covering the firm.4
3 Chen et al. (2010) argue that financial analysts have superior information processing abilities because they have specialized training, experience, and knowledge about a firm/industry. For instance, analysts with more expertise can aid investors in understanding the meaning of certain accruals, while analysts with training in political economics can decipher the implications of international business issues, such as order backlogs from different countries. 4 The effective management of risk with derivatives can reduce noise in earnings and cash flows, potentially reducing the demand for analyst services. However, as investors struggle to comprehend even the most basic hedges and derivatives terminology (Koonce et al. 2005), the information processing role of analysts’ research is stronger and more long-lasting for firms with more complex information (Chen et al. 2010). Thus, on balance, an increase in demand for analyst services likely prevails.
9
The effect of derivatives usage on the aggregate supply of analyst coverage is not as clear.
On the one hand, analysts have at least three financial incentives to cover derivatives users. First,
greater demand for analyst services due to the complexities of derivatives can translate into more
valuable advice for analysts to sell to investors. For instance, a recent survey finds that 72% of
analysts rate investor demand for firm-level information as the most important factor in deciding
whether to cover a firm (Brown et al. 2015). Second, when a firm effectively hedges (reduces) a
risk exposure, changes in the fair value of both the derivative and hedged item are recognized in
income together if the firm elects hedge accounting (FASB 1998). Thus, a reduction in earnings
volatility is a potential byproduct of derivatives usage (Zhang 2009), which can improve analysts’
earnings forecasts (Previts et al. 1994) and lead to career promotions (Hong et al. 2000). Third, by
reducing agency conflicts (Campbell and Kracaw 1990) and corporate taxes (Donohoe 2015b),
derivatives can improve performance metrics and firm value (Aretz and Bartram 2010). Because
these (and other) benefits are attractive to investors, a firm’s use of derivatives can provide analysts
with opportunities to earn higher trading commissions. Taken together, these financial incentives
could increase the aggregate supply of analyst coverage.
On the other hand, three issues could deter analysts from covering derivatives users. First,
evaluating financial reports is far more difficult when a firm uses derivatives (Kawaller 2004). At
a minimum, analysts must determine what risk exposures exist, how much of the exposures are
hedged, and how the hedges are managed.5 However, acquiring such information from firms’
financial reports is challenging because derivatives are reported in vastly different ways among
5 For example, in the case of an agricultural company, analysts must understand what factors affect inventory, the extent to which such factors are covered by contracts (e.g., futures), and how contract prices for the purchase, production, and sale of inventory are determined, monitored, and accounted for in the financial statements.
10
the population of derivatives users (Kawaller 2004).6 Second, to accurately predict future earnings,
analysts must consider a firm’s ability to anticipate changes in its risk exposures and adjust hedge
coverage accordingly. But, assessing a static picture of derivatives positions (i.e., balance sheet)
and hedge results over a limited number of accounting periods (i.e., income statement) for this
purpose is rarely feasible. Indeed, Chang et al. (2016) find that the financial reporting complexity
of derivatives leads analysts to misjudge the effects of derivatives on earnings, resulting in less
accurate and more dispersed earnings forecasts. Third, derivatives have drawn sharp criticism from
experts, news media, and tax authorities since the mid-1990s (Donohoe 2015a). Because analysts
tend to cover firms for which they have favorable views (Hayes 1998), these negative perceptions
could deter analysts from covering derivatives users. A firm’s use of derivatives could therefore
decrease the aggregate supply of analyst coverage.
Overall, a firm’s use of derivatives will increase the demand for analyst services, but could
increase or decrease the supply of analyst coverage. Further, the effect of derivatives on analyst
coverage could vary across time as analysts’ decisions are dynamic by nature. Some analysts might
initiate coverage of derivatives users in response to investor demand, but later cease coverage if
earnings forecasts prove too difficult. Thus, we investigate the change in analyst coverage after a
firm begins using derivatives, both immediately and over time. Because analyst coverage depends
on the net effect of competing supply and demand forces, we test a non-directional hypothesis:
H1: Analyst coverage does not change after a firm initiates derivatives.
4. Data and sample selection
Following Chang et al. (2016), we begin with Compustat observations for fiscal-years 1998
6 For example, although hedge accounting matches the value effects of hedge instruments with those of the underlying hedged asset/liabilities to the same accounting period, not all firms qualify and/or choose to practice this method. Even when hedge accounting is practiced, firms rarely disclose the unhedged portion of risk exposures, which is likely more relevant than knowing how much a derivative gained or lost (Chang et al. 2016, 588).
11
to 2011 meeting four criteria: (1) publicly traded; (2) domestically incorporated; (3) non-financial,
non-utility industry; and (4) at least three years of consecutive data.7 We identify new derivatives
users by searching Form 10-Ks (extracted from the SEC’s EDGAR database) for keywords relating
to derivatives.8 Consistent with Guay (1999) and Donohoe (2015b), a firm is a New User if it does
not report a derivatives position when it first appears in the sample, but does report a position in a
later year. Firms enter the New User sample only when derivatives are first observed (after first
observing no usage). The resulting sample consists of 1,260 New Users during the sample window.
We also identify samples of derivatives users and non-users. A firm is a User if it reports
derivatives in year t and is not a new user, while firms reporting no derivatives are classified as
Non-Users. A New User firm can be a Non-User in an earlier period if it did not use derivatives
for at least two consecutive years, and enters the User sample after using derivatives for at least
two consecutive years.9 These samples consist of 17,987 Non-User and 15,584 User observations.
We then drop observations without necessary Compustat, CRSP, and I/B/E/S data to perform the
analyses that follow, resulting in 10,887 Non-User, 12,017 User, and 587 New User observations.
<INSERT TABLE 1 ABOUT HERE>
Table 1 reports characteristics of Non-Users, Users, and New Users. Panel A illustrates the
temporal distribution of each sample by derivatives reporting regime. Although the number of
Non-Users is somewhat stable, an increase in Users and New Users coincides with the enactment
7 Fiscal year 1998 is the first full financial reporting year after the phase-in of the EDGAR system. Financial (utility) firms have two-digit SIC codes 60-69 (49). We remove these firms as they are more likely to use derivatives primarily for trading purposes or act as a derivatives dealer, both of which involve different financial reporting requirements. 8 We discover a large majority of derivatives users by searching for the word, “derivative.” Other keywords include: “hedge”, “forward contract”, “futures contract”, “option”, “swap”, and “notional”. 9 To illustrate, consider a firm that did not use derivatives until 2006. From 1998 to 2005, observations for this firm are classified in the Non-User sample. In 2006, the observation is classified in the New User sample. If the firm does not continue to use derivatives in 2007, the observation for 2007 is classified in the Non-User sample. If, however, the firm continues to use derivatives in 2007, the observation for 2007 is classified in the User sample. Thus, New User designation only occurs the first time that derivatives usage is observed (after initially observing no usage). A small number of firms stop and later restart using derivatives; however, omitting them does not influence our results.
12
of SFAS Nos. 133/138 (2,602 Users; 228 New Users) and 149 (3,448 Users; 123 New Users),
which greatly altered derivatives accounting. Similarly, an increase in New Users (120) occurs
after the enhanced disclosures of SFAS No. 161. These trends are consistent with evidence that
accounting standards, namely SFAS Nos. 133/138, increased derivatives usage (Abdel-khalik and
Chen 2015). Nevertheless, we take steps to mitigate the possibility that changes in accounting
practices made existing Users appear to be New Users.10 We also examine the effects of these
standards on analysts’ coverage decisions in Section 7.1. Panel B reports the industry distribution
of each sample. Overall, firms in the manufacturing and business equipment industries comprise a
large portion of each group. Thus, where applicable, we control for industry and year fixed-effects.
5. Research design
We use the focused setting of New Users and a difference-in-differences design to test our
hypothesis. This approach offers four distinct advantages (Chang et al. 2016). First, it captures the
dynamic, multi-period setting in which analysts make decisions by evaluating changes in coverage
before and after derivatives initiation. Second, it accounts for variation in an outcome (coverage)
that is not the result of treatment exposure (derivatives initiation) by comparing the treatment
group to an untreated control group (Roberts and Whited 2013). Third, by examining the effects
of derivatives initiation, it overcomes many methodological limitations, including the possibility
of correlated omitted variables bias (Skinner 1996). Finally, it mitigates concerns that unobserved
factors drive the relation between derivatives and analyst coverage.11
10 In particular, we confirmed the initiation year for each New User by searching the entirety of Form 10-K in prior years for any evidence of derivatives usage. We also test our hypothesis (unreported) after excluding New Users in the enactment year for each standard (2001; 2003; 2008). Inferences remain the same. 11 Difference-in-differences assumes that in the absence of the treatment, average outcomes for treatment and control groups would have followed parallel paths over time. A direct test of this assumption is not possible because one cannot observe the absence of a treatment once it has occurred. However, placebo tests mitigate concerns that the parallel paths assumption is violated by falsely assuming that treatment occurs in a prior period. We conduct two (unreported) placebo tests by falsely assuming that derivatives initiation occurs either one or two years prior to
13
5.1 Untreated control group
We use propensity score matching to identify a control group of Non-Users and account
for the endogeneity in a firm’s decision to use derivatives. This nonparametric matching technique
facilitates causal inference in non-experimental settings by constructing a control group that is
similar to a treatment group (Rosenbaum and Rubin 1983). Because a firm can be a Non-User in
one year and a User in another year (Section 4), we limit potential control firms to those that do
not use derivatives at any point during the sample period (“pure” Non-Users). We then estimate
the propensity of derivatives initiation using the following probit regression model:
0 1 1 1 1Pr ,x y z k tit x it y it z it k it t it it
x y z k t
INIT RMI ACI CTRL IND YR (1)
where INIT equals 1 for New Users (i.e., initiation) and 0 for “pure” Non-User observations.
Following Donohoe (2015b), RMI is a vector of risk management incentives that explain
derivatives usage. It includes exposures to interest rate (IRISK), foreign exchange rate (FRISK),
and commodity price (CRISK) risks as surveys reveal these are the risks most often managed with
derivatives (Bodnar et al. 2003). By insulating firm value and cash flow from unfavorable changes
in risk exposures, derivatives can thwart financial distress (Mayers and Smith 1982), harmonize
financing and investment goals (Froot et al. 1993), and reduce agency conflicts (Smith and Stulz
1985). We include financial distress likelihood (ALTZ), underinvestment likelihood (USCORE),
and the sensitivity of executive compensation to firm value (ECSENS) to capture these incentives.
We also include the cash ETR (CETR) to reflect the tax planning features of derivatives (Donohoe
2015b). As derivatives substitutes, we control for convertible debt (CDEBT), preferred stock
(PSTOCK), and abnormal accruals (ABACC).12 Lastly, the volatility in cash flow (CFV) and
initiation. We find insignificant coefficients for the difference-in-differences estimator, suggesting that the parallel paths assumption holds and our reported results are associated with derivatives initiation. 12 Convertible debt includes an embedded option on firm assets, which reduces the sensitivity of equity value to changes in firm value. Preferred stock reduces the probability of financial distress by paying periodic dividends as
14
earnings (EV) reflect other basic incentives for derivatives usage (Zhang 2009). By including RMI,
the absence of derivatives among potential Non-User control firms reflects a choice not to use
them, rather than no incentives to do so. All variables are defined in the Appendix.
ACI is a vector of analyst coverage incentives.13 Consistent with prior studies (Lang and
Lundholm 1996; Barth et al. 2001; Tan et al. 2011), we include market value of equity (SIZE),
intangibles (INTANG), stock return volatility (RETVOL), and the market-to-book (MB) ratio. We
also include stock and debt issuances (ISSUE), share turnover (TURNO), and annual stock returns
(ANRET) to capture analysts’ tendency to cover firms that have more predictable performance and
provide opportunities to earn higher commissions (Hayes 1998; Lang et al. 2004). CTRL is a vector
of controls that likely influence derivatives usage and analysts’ coverage decisions. We include
geographic (GSC) and industry (ISC) sales concentration (Bushman et al. 2004) because firm-level
complexity, apart from that relating to derivatives, can influence analysts’ coverage decisions.14
We also control for profitability (ROA), foreign activity (FRGN), and mergers/acquisitions (M&A).
Finally, industry (IND) and year (YR) fixed-effects account for variation in the decision to initiate
derivatives across industries and time, respectively.15
RMI, ACI, and CTRL are lagged (t−1) to avoid simultaneity with analyst coverage in the
hypothesis tests. The predicted probabilities from Eq. (1) are the propensity scores. We match each
New User to only one “pure” Non-User as of the year before initiation by nearest propensity score,
opposed to interest. These alternatives reduce the incentive to hedge with derivatives (Nance et al. 1993). Similarly, derivatives and accruals can serve as partial substitutes for smoothing earnings (Barton 2001). 13 Because matching models do not require exclusion restrictions, the general rule is to include a comprehensive list of covariates when estimating propensity scores (Rubin 2009). Thus, while ACI is not directly related to derivatives usage, we include this vector in Eq. (1) to identify an untreated control group of Non-Users that is matched on as many relevant characteristics as possible. Nevertheless, excluding ACI from Eq. (1) yields similar inferences. 14 GSC and ISC are revenue-based Herfindahl-Hirschman indices, where smaller values indicate less geographic and industry sales concentration, respectively, and thus more complexity (see Bushman et al. 2004). The inclusion of only one of these variables in the model at a time does not influence the results. 15 In unreported tests, we replace year fixed-effects with indicators for the effective dates of derivatives accounting standards enacted during the sample period (SFAS Nos. 133/138, 149, 155, and 161). Inferences remain the same.
15
within common support, without replacement, using a caliper distance of 0.01.
5.2 Difference-in-differences specification
We compare analyst coverage of the 587 New Users to that of the 587 Non-User control
firms (identified with Eq. [1]) with the following negative binomial regression:
0 1 2 3
,
yit i it i it y it
yz k t
z it k it t it itz k t
COV NEWUSER POST NEWUSER POST ACI
+ CTRL IND YR
(2)
where COV is the number of analysts covering firm i in year t.16 NEWUSER equals 1 for New User
observations and 0 for control firm observations. POST is coded 1 for post-treatment periods (i.e.,
after initiation) for New Users and corresponding control firms (0 otherwise). The coefficient for
NEWUSER (Ψ1) captures the difference in analyst coverage between New Users and control firms
before derivatives initiation, and the coefficient for POST (Ψ2) reflects the change in analyst
coverage among control firms between pre- and post-initiation periods. Thus, the coefficient for
NEWUSER×POST (Ψ3) captures the effect of initiation on analyst coverage for New Users relative
to Non-User control firms (test of H1). ACI and CTRL are as described above. Industry (IND) and
year (YR) fixed-effects control for variation in coverage across industries and time, respectively,
and firm clustered standard errors mitigate concerns about cross-correlated residuals.17
6. Main results
6.1 Descriptive statistics and univariate tests
Table 2 reports descriptive statistics for the dependent variable, RMI, ACI, and CTRL, along
16 Negative binomial regression is common in tests of analyst coverage as it is preferred to OLS when the dependent variable ranges among nonnegative integers. The benchmark model for such count data is the Poisson (Cameron and Trivedi 2009); however, it makes two strict assumptions: (1) equal mean and variance; and (2) independence. The negative binomial distribution relaxes these assumptions and is used when count data are over-dispersed (i.e., variance exceeds the mean). For our data, diagnostic tests reveal considerable over-dispersion. 17 Following McInnis and Collins (2011) and Donohoe (2015b), we construct pseudo pre/post-initiation periods for control firms based on when the matched New User initiates. Note that we omit the vector RMI from Eq. (2) as prior research suggests that risk management incentives are more apt to explain derivatives initiation than analyst coverage (Aretz and Bartram 2010). However, in unreported tests we include RMI and find similar results.
16
with t-statistics for mean tests of differences between Non-Users and that of Users and New Users.
Mean tests of COV reveal that analyst coverage for Users and New Users is greater than that for
Non-Users. Mean tests for RMI, ACI, and CTRL indicate that Non-Users differ from Users and
New Users across many dimensions, such as interest rate risk (IRISK), distress (ALTZ), intangibles
(INTANG), stock turnover (TURNO), and profitability (ROA). Users and New Users are also larger
(SIZE) and more complex (GSC, ISC) than Non-Users. Overall, the disparity across the three
samples indicates that multivariate analyses with a matched sample will provide for more robust
tests of the link between derivatives and analyst coverage.18
6.2 Untreated control group
We examine Pearson correlations (unreported) for the variables used to identify the control
group of Non-Users (Eq. [1]). Notably, derivatives initiation (INIT) is significantly correlated with
GSC (−0.035), ISC (−0.085), and SIZE (0.104). Thus, in Section 7.4, we examine if these other
types of complexity or size drive our results. Other correlations are consistent with prior studies,
and Variance Inflation Factors (unreported) indicate no multicollinearity issues in any of our tests.
<INSERT TABLES 2–3 ABOUT HERE>
Table 3 reports the covariate balance between New Users and Non-User control firms
identified with Eq. (1). If covariates are balanced, then differences in analyst coverage can be
attributed to derivatives initiation rather than other firm attributes. Reported amounts are p-values
from tests of differences in means (t-test), medians (Wilcoxon rank-sum test), and distributions
(Kolmogorov-Smirnov test) of RMI, ACI, and CTRL between the two samples. Propensity score
matching does not require matched firms to be identical across all covariates (Caliendo and
Kopeinig 2008). Of the 24 variables, only three (FRISK, ECSENS, ROA) are statistically dissimilar
18 We use the sample of Users in a validation test in Section 7.1. Unreported tests of differences between Non-Users and potential non-user control firms (“pure” Non-Users) indicate these two samples are similar along most dimensions.
17
at a 90 percent confidence level. When all covariates are considered jointly, Hotelling’s T2-test (p
=0.998) reveals that New Users and control firms are not different from one another. In sum, these
results suggest the matching process was successful in balancing the control variables.19
6.3 Hypothesis tests
We begin our tests with Figure 1, which plots the number of analysts covering a balanced
sample of 549 New User observations relative to derivatives initiation year (t−1 to t+1). The
positive slope offers initial evidence that analyst coverage increases after derivatives initiation. In
addition, the quick reaction of analysts upon initiation (from t−1 to t) is similar to their response
to other substantial within-firm financial reporting changes (Botosan and Harris 2000).
<INSERT TABLE 4 & FIGURE 1 ABOUT HERE>
Table 4 reports estimates of Eq. (2). To examine the immediate effects of derivatives
initiation, the tests in column (1) use data for the year before (t−1) and upon (t) initiation for the
sample of 587 New Users and 587 matched control firms (2,348 firm-years). To assess the overall
effects, column (2) reports results based on all data before and after initiation (9,826 firm-years).
In column (1), the insignificant coefficient for NEWUSER indicates that analyst coverage of New
Users is similar to that of control firms before initiation in the immediate sample window, while
the negative coefficient for POST (−0.045) reveals a reduction in analyst coverage of control firms.
Across all sample years (column (2)), however, more analysts cover New Users than control firms
19 Our primary tests are susceptible to “hidden” bias if there are correlated omitted variables that are unbalanced across treatment and control groups. Although it is not possible to test if there are no unobserved variables that influence treatment selection, there are tests for assessing the sensitivity of matched-pair results to such hidden bias (Rosenbaum 2007). The central issue of these tests is how strongly an unmeasured variable must influence the selection process to undermine the matched-pair analysis (Becker and Caliendo 2007). Using the Mantel and Haenszel (1959) test statistic for binary outcomes, we find the results are insensitive to a bias that would affect the odds of derivatives initiation by a factor greater than 10. Although no strict benchmarks exist for gauging sensitivity (Armstrong et al. 2012), by comparison, if control firms were matched only by industry the results would be sensitive to a bias that less than doubled the odds of derivatives initiation. Thus, the results are robust to significant correlated omitted variables bias, largely because of the expansive controls included in the propensity score matching model (Eq. [1]). Nevertheless, as an additional robustness check, we use entropy balancing (Hainmueller 2012). Inferences remain the same.
18
before initiation (i.e., NEWUSER is positive and significant), with no significant change in control
firm coverage (i.e., POST is insignificant).
As tests of H1, the coefficients for NEWUSER×POST are positive and significant in both
columns (0.105; 0.091). These results indicate that, relative to control firms, New Users realize an
increase in analyst coverage after derivatives initiation, both immediately and overall. The increase
in coverage is consistent with analysts supplying more of their services in response to heightened
client demand for information about firms initiating complex derivatives programs (e.g., Brown et
al. 2015).20 The coefficients for other variables are consistent with prior research. For example,
the positive coefficients for SIZE and TURNO are consistent with analysts covering large firms
with higher share turnover, respectively, to obtain larger commissions (Irvine 2004). These (and
other) coefficients are statistically significant despite covariate balance with control firms (Table
3) as changes in these factors in non-match years can influence analyst coverage decisions.21
To gauge the economic effects of derivatives initiation on analyst coverage, we estimate
the percentage change in COV for New Users from the pre-initiation period to the post-initiation
period using all available data. Following Tan et al. (2011), we estimate the percentage change in
COV for New Users after initiation by computing the marginal effect of POST on COV for New
Users. The marginal effect is the partial derivative of Eq. (2) with respect to POST (NEWUSER
equals 1 and all independent variables are held at the sample mean for New Users). It indicates
20 It is not possible to further disentangle the supply and demand effects using publicly available data because observed analyst coverage depends on the net effect of these competing factors. Such distinctions are not important for our tests of H1 because either an increase in supply or demand (or both) leads to an observed increase in analyst coverage. 21 We verify that our results are robust to using strict one-to-one matching and a changes specification of differences-in-differences (Roberts and Whited 2013). That is, we regress the change in COV between the pre- and post-initiation period for New Users and control firm observations on NEWUSER and changes in ACI and CTRL. Inferences remain the same. To ensure that the results are not driven by changes in analyst coverage among control firms, we also examine whether changes in analyst coverage after initiation are statistically significantly among only New Users. We estimate Eq. (2) using the time-series of New Users, such that POST is the variable of interest. The (unreported) coefficients for POST are positive and significant, indicating that the observed increase in COV after initiation is statistically significant (p-value<0.01) among only New Users.
19
how COV changes as POST changes from 0 to 1, holding other variables constant. As a benchmark,
we compute estimates of COV in the pre-initiation period, where POST=0 and NEWUSER=1. The
ratio of the marginal effect of POST to its pre-initiation value estimates the relative percentage
change in COV for New Users after initiation. These (unreported) ratios are based on the most
conservative results in Table 4 and, thus, reflect a lower bound estimate of economic significance.
The ratios indicate that, relative to Non-User control firms, New Users experience a 10%
increase in analyst coverage (on average) after derivatives initiation. This estimate is similar to the
effects of other changes in firms’ financial reporting practices on analyst coverage. For instance,
Tan et al. (2011) find a 20% increase in analyst coverage for firms adopting IFRS. Although the
average effect of derivatives is smaller than the effect of adopting an entirely different set of
reporting standards, it is nonetheless material, suggesting that analysts do indeed respond to the
increased demand for their services due to the complexity of derivatives. However, which types of
analysts respond (or do not respond) to such demand is an open question that we address next.
6.4 Analyst expertise
6.4.1 Difference-in-differences tests
The tests in Table 4 examine whether and how the complexity of derivatives influences the
coverage decisions of analysts. We next evaluate if the observed increase in analyst coverage for
New Users after derivatives initiation is driven by analyst expertise. Analysts with more expertise
(e.g., career experience) produce more accurate earnings forecasts, often leading to media attention
and All-Star status (Ramnath et al. 2008). Because these analysts possess superior skill and
knowledge, they could be more likely than other analysts to respond to the increased demand for
analyst services due to the complexity of derivatives. However, recent research finds that analysts
(on average) produce less accurate earnings forecasts for new derivatives users (Chang et al. 2016).
20
As such, analysts with more expertise might instead recognize and avoid such challenges, while
analysts with less expertise might view the complexity of derivatives as an opportunity to signal
their talent and obtain media attention as well as career promotions. Because analyst expertise is
difficult to measure, we conduct several tests of whether the increase in coverage for New Users
differs as aspects of expertise vary. Corroborating evidence across these tests will provide stronger
inferences about the link between analyst coverage and the complexity of derivatives.
We first examine if the increase in coverage for New Users after initiation is driven by
analyst career experience. Figure 2 plots the number of analysts covering New Users relative to
initiation year (t−1 to t+1) using a balanced sample. We group analysts based on career experience,
which we define as the number of years (since 1980) that analyst j has issued forecasts for any
firm in I/B/E/S. Analysts in the top quintile of career experience are designated as high experience
(HIEXP); low experience (LOEXP) otherwise. The number of high experience analysts (solid line)
covering New Users remains stable before and after initiation (t), while the number of low
experience analysts (dashed line) increases substantially upon (t) and after initiation (t+1). This
pattern suggests that the increase in coverage after derivatives initiation is likely driven by analysts
with low career experience.
<INSERT TABLE 5 & FIGURES 2-3 ABOUT HERE>
Panel A of Table 5 reports estimates of Eq. (2), where the dependent variable is the number
of high (COV_HIEXP) and low (COV_LOEXP) career experience analysts covering firm i in year
t in columns (1) and (2), respectively, using data for the year immediately before (t−1) and upon
(t) initiation.22 The coefficient for NEWUSER×POST captures the effect of derivatives initiation
22 The sum of observations in each column do not match those in Table 4 because the dependent variable is the number of analysts with high or low career experience and our tests use firm-year data. For example, assume a New User is covered by two high-experience analysts and four low-experience analysts. The dependent variable in column (1) reflects the two high-experience analysts covering the New User, while the dependent variable in column (2) reflects
21
on the number of high (column (1)) versus low (column (2)) career experience analysts covering
New Users relative to Non-User control firms. In column (1), this coefficient is insignificant,
indicating no change in coverage for New Users from high career experience analysts after
initiation. However, the positive and significant coefficient for NEWUSER×POST (0.094) in
column (2) suggests that the increase in analyst coverage for New Users after initiation (Table 4)
is driven by analysts with low career experience. A Wald χ2-test (3.35) confirms that the interaction
coefficient in column (2) is different (p-value<0.03) from the coefficient in column (1).23
We next consider whether All-Star analysts drive the increase in analyst coverage for New
Users after derivatives initiation. All-Star analysts are selected annually by Institutional Investor
Magazine based on solicited input from buy-side managers (i.e., chief investment officers of large
money management institutions, directors of research, select analysts, and portfolio managers).24
Prior research finds that the experience level of these analysts is greater than that of less talented
analysts as the latter tend to fade away over time (Leone and Wu 2007). In our sample, All-Stars
have 6.75 years of mean career experience versus 4.75 years for non-All-Stars (unreported). The
All-Star (STAR) and high career experience (HIEXP) analyst subsamples are only moderately
correlated (0.09), however, indicating that All-Star status reflects unique analyst attributes not
otherwise captured by career experience alone.
Figure 3 plots the number of analysts covering New Users relative to initiation year (t−1 to
t+1) using a balanced sample, where analysts are grouped by All-Star (STAR) or non-All-Star
the four low-experience analysts for the same New User. Because some firms are covered by both high and low experience analysts, the sample sizes do not necessarily sum to the 2,348 observations in Table 4. 23 These results are robust to expertise definitions based on quartiles and terciles. We report quintiles as they reflect the most conservative results. 24 In particular, survey participants vote for the analysts they believe have been most helpful. Analysts are ranked by numerical scores created from the votes and weighted by the size of the voter’s institution (Bagnoli et al. 2008). Some important performance factors include industry knowledge, written reports, stock picks, earnings estimates, timely communication with investors, and responsiveness to investor reports (Leone and Wu 2007). The results are published annually in the October issue of the magazine.
22
(NONSTAR) status.25 The solid line depicting the number of All-Star analysts covering New Users
is flat before and after initiation (t), whereas the dashed line for non-All-Star analysts markedly
increases upon (t) and after (t+1) initiation. This pattern suggests that the increase in coverage after
derivatives initiation is likely driven by non-All-Star analysts.
Panel B of Table 5 reports estimates of Eq. (2), where the dependent variable is the number
of All-Star (COV_STAR) and non-All-Star (COV_NONSTAR) analysts covering firm i in year t in
columns (1) and (2), respectively, using data for the year immediately before (t−1) and upon (t)
initiation. The insignificant coefficient for NEWUSER×POST in column (1) indicates no change
in coverage for New Users from All-Star analysts after derivatives initiation. However, the positive
and significant interaction coefficient (0.087) in column (2) suggests the increase in coverage for
New Users after initiation is driven by analysts other than All-Stars. As before, the interaction
coefficient in column (2) is statistically different (p<0.02) from that in column (2).26
6.4.2 Transition tests
Table 5 reveals that the increase in analyst coverage after derivatives initiation is driven by
analysts with less expertise. However, because the difference-in-differences estimator (ψ3) reflects
the average treatment effect, the tests do not reveal if the observed change in coverage results from
analysts with less expertise initiating coverage or those with more expertise ceasing coverage of
25 I/B/E/S (detail file) includes a unique identifier for each analyst, but does not provide personal details (i.e., names). The Institutional Investor All-American Research Team file includes analyst names, but not the identifier. We use the Broker Code Key (I/B/E/S), which includes both identifiers and names, to merge the two files. However, because I/B/E/S does not provide the Broker Code Key after 2006, all subsequent observations are matched by hand. 26 It is possible that the insignificant coefficients for NEWUSER×POST in column (1) of Panels A and B of Table 5 are because our tests do not possess sufficient statistical power to detect an effect across the analyst expertise partitions. To mitigate this possibility, we conduct a post hoc power analysis, which computes achieved statistical power (1−β) as a function of significance level (α), sample size (n), and effect size (Faul et al. 2007). Following Cohen (1988), we define effect size as the incremental adjusted-R2 obtained when NEWUSER×POST is included in Eq. (2) (relative to a model with only other covariates). Using a statistical significance level of 0.10, sample sizes for tests in Table 5, and incremental adjusted-R2 (unreported), we find statistical power in excess of conventional levels (0.80). Thus, lack of statistical power is not an alternative explanation for these results.
23
New Users, which can influence the information environment.27 Thus, to ensure that less expertise
does indeed drive the results, we examine analyst coverage at a more granular level. In particular,
we create transition matrices that trace the proportion of analysts covering New Users before
derivatives initiation and the proportion ceasing or initiating coverage of New Users upon and after
derivatives initiation, across differing levels of expertise. While we expect the inferences will be
similar to those in Table 5, the matrices allow us to more closely examine the characteristics of
analysts responding to the increase in demand for their services.
Panels A and B of Table 6 trace the change in analyst coverage after derivatives initiation
by career experience and All-Star status, respectively. To construct the matrices, we sort analysts
covering a balanced sample of New Users from t−2 to t+1, where t is derivatives initiation, into
quintiles based on level of career experience and All-Star status ranking (as defined in Section
6.4.1). The balanced sample (i.e., requiring observations for each year of the t−2 to t+1 window)
helps ensure our tests focus on changes in coverage that are not due to variation in the number of
New Users across time. We then count the number of analysts ceasing or initiating coverage upon
(t) and one year after (t+1) derivatives initiation.28
<INSERT TABLE 6 ABOUT HERE>
In Panel A, for the two years before derivatives initiation (t−2 and t−1), 25.15% of analysts
covering the balanced sample of New Users are in the lowest quintile of career experience, relative
to 17%-19% in each of the other quintiles. Upon initiation (t), 670 analysts cease coverage of New
Users while 1,713 analysts begin covering these firms, resulting in a net coverage increase of 1,043
27 For example, among our sample firms, Alexion Pharmaceuticals experienced a large net increase in analyst coverage after derivatives initiation (from 12 to 17 analysts). Specifically, seven new analysts began covering the firm, and two analysts ceased coverage, after the firm initiated a derivatives program. None of the seven new analysts were in the high career experience (HI_EXP) or All-Star (STAR) groups. 28 We verify that analysts ceasing (initiating) coverage in t or t+1 did (did not) cover a New User firm in t−2 or t−1.
24
analysts. Similarly, one year later (t+1), the number of analysts beginning coverage of New Users
exceeds the number of analysts ceasing coverage by 374. However, in both post-initiation periods,
the largest proportion of analysts ceasing and initiating coverage are those with the lowest amount
of career experience. For example, upon derivatives initiation, 32.09% (38.59%) of the 670 (1,713)
analysts ceasing (initiating) coverage are in the lowest quintile of career experience, compared to
17.16% (12.38%) in the highest quintile. A similar pattern exists in Panel B, where analysts ceasing
and initiating coverage are predominately not labeled as All-Stars. Thus, while analysts in all levels
of expertise both cease and initiate coverage of New Users after derivatives initiation, the overall
net increase in coverage is driven by analysts with less expertise initiating coverage.
Panels C and D of Table 6 trace the change in coverage after derivatives initiation by career
experience and All-Star status, respectively, for analysts that already cover New User firms before
they begin using derivatives. These matrices consider how the proportion of analysts with prior
knowledge of New Users (at the firm-level) changes after derivatives are introduced. To construct
these matrices, we trace analysts already covering New Users upon derivatives initiation (i.e., not
new analysts) back two years before derivatives initiation (t−2 and t−1) and track these analysts
one year after derivatives initiation (t+1). In Panel C, the career experience of analysts that already
cover New Users before initiation (t−2 and t−1) is fairly diverse, but slightly skewed towards more
experience. This diversity in experience persists upon (t) and after (t+1) initiation, even though the
proportion of analysts in the lowest quintile of experience becomes much smaller in t+1 (14.13%).
Panel D reveals only minor changes in the ratio of All-Star analysts already covering New Users
upon and after initiation. Collectively, these matrices show that the analysts already covering New
Users before derivatives initiation have diverse levels of expertise before and after initiation. These
findings confirm the results in Panels A and B that the overall net increase in coverage is driven
25
by new analysts with less expertise initiating coverage rather than the expertise of existing analysts.
Finally, Panel E traces the career outcomes of analysts ceasing coverage of New Users. In
particular, we classify these analysts into four groups upon (t) and after (t+1) derivatives initiation:
(1) promotion (moving to a larger brokerage; defined based on a quintile rank of all brokerages in
I/B/E/S by number of analysts employed); demotion (moving to a smaller brokerage); (3) stay (no
change in brokerage size quintile); and (4) missing (analyst no longer appears in I/B/E/S). For both
t and t+1, about half of the analysts that cease coverage of New Users remain at the same brokerage
or one of similar size. A large proportion of analysts also disappear from the dataset after ceasing
coverage. While the inferences we can draw from these trends are limited, ceasing coverage of
New Users does not seemingly lead to positive or negative career outcomes in the short-term.
6.4.3 Analyst earnings forecast properties
Tables 5 and 6 indicate that the increase in analyst coverage of New Users after derivatives
initiation is not driven by analysts with more expertise, as reflected by career experience and All-
Star status. Instead, analysts with less expertise largely begin covering New Users, likely because
they believe the complexity of derivatives provides an opportunity to signal their talent, issue more
valuable reports, and earn higher commissions. These results complement Chang et al. (2016) by
providing an explanation for the earnings forecast properties they find for New Users. That is, a
large number of low expertise analysts begin covering New Users and, in turn, produce weaker
and more dispersed earnings forecasts for these firms.
To confirm this explanation, we perform difference-in-difference tests of the accuracy and
dispersion of analysts’ earnings forecasts for New Users conditional on expertise level. Using the
I/B/E/S detail file, we measure career experience consistent with the earlier tests, and then calculate
each individual analysts’ mean earnings forecast accuracy (AEFA) and dispersion (AEFD) by firm-
26
year and expertise group.29 We merge this data with the sample used to test the immediate effects
of derivatives initiation (Table 4). Finally, we calculate all of the covariates in Eq. (2) of Chang et
al. (2016), resulting in 1,271 and 1,892 analyst-firm-year observations in high and low experience
groups, respectively. Variables are defined in the Appendix.
<INSERT TABLE 7 ABOUT HERE>
Panel A (B) of Table 7 reports the results, where the dependent variable is earnings forecast
accuracy (dispersion) for high and low career experience analysts in columns (1) and (2),
respectively. In Panel A, the coefficient for NEWUSER×POST is insignificant in column (1),
indicating no change in forecast accuracy for New Users by high career experience analysts after
derivatives initiation. In contrast, the negative and significant coefficient for NEWUSER×POST
(−0.512) in column (2) reveals that, relative to control firms, the accuracy of analysts’ earnings
forecasts for New Users declines after derivatives initiation for the low experience group. In Panel
B, the coefficient for NEWUSER×POST (0.083) is positive and significant in only column (2),
suggesting analysts’ earnings forecasts for New Users are more dispersed for the low experience
group after derivatives initiation. A Wald χ2-test in each panel confirms that the coefficient for
NEWUSER×POST in column (2) is statistically different (p<0.05) from that in column (1).30 Thus,
the increase in low expertise analysts initiating coverage of New Users explains the less accurate
and more dispersed earnings forecasts shown by Chang et al. (2016) for these firms.
7. Other tests
7.1 Validation test
To validate that the complexity of derivatives does indeed influence analyst coverage, we
test the structural stability of analyst coverage over specific periods of time in the broader setting
29 Chang et al. (2016) use the consensus forecast as they do not consider analyst-level characteristics in their tests. 30 Results for tests of All-Star status (untabulated) are similar.
27
of Users. In this sense, structural stability is a statement about parameters in the context of an
econometric model. An assumption of stationarity implies model parameters (mean and variance)
are constant over time, and a structural break occurs if at least one parameter changes. Structural
breaks can occur, for example, as a result of regulatory regime changes; that is, when one mix of
reporting standards is replaced by another (Donohoe and McGill 2011; Chang et al. 2016).31
Extensive changes in reporting rules for derivatives during the sample period are likely to
influence analyst coverage. The application of fair value-based accounting under SFAS No. 133
after 2000, and its subsequent repairs/amendments by SFAS Nos. 138, 149, and 155 substantially
altered the financial reports of derivatives users. Although these standards enhanced and clarified
derivatives accounting, they permitted two distinct and basically inconsistent approaches to hedge
accounting (see Appendix A of Chang et al. (2016)). Further, SFAS No. 161, effective after 2008,
required enhanced disclosures to address concerns that SFAS No. 133 did not provide adequate
detail about how derivatives usage affects financial position, performance, and cash flow. If
analyst coverage is indeed influenced by the complexity of derivatives, we expect to see changes
in analyst coverage as the reporting practices of derivatives change.
Similar to Chang et al. (2016), we alter Eq. (2) in three ways. First, we replace NEWUSER
and POST with an indicator variable, USER, identifying the User and Non-User samples. We then
add four regime indicators equal to 1 for observations after fiscal-years June 2000 (SFAS133/138),
June 2003 (SFAS149), September 2006 (SFAS155), and November 2008 (SFAS161), respectively
(0 otherwise).32 Finally, we interact each indicator with USER such that the interaction coefficients
31 We focus on Users because the difference-in-difference design is not feasible for regime-level tests (Chang et al. 2016). By excluding New Users, structural break tests focus on the effects of ongoing derivatives usage and alleviate the potential that firm-level changes at initiation (e.g., risk exposures) drive the results. 32 Although SFAS No. 133 had an original effective date of June 1999, the FASB delayed implementation by one year. In addition, SFAS No. 138, effective June 2000, amended several aspects of SFAS No. 133. Based on their close proximity, we use only one regime indicator (SFAS133/138) to reflect these two standards.
28
capture structural breaks in analyst coverage among Users in a given reporting regime. Thus, these
tests focus on the financial reporting complexities of derivatives.33
We find a positive and significant coefficient for USER (untabulated), suggesting greater
analyst coverage for Users relative to Non-Users. The interaction coefficients also indicate that
analyst coverage for Users increases after SFAS Nos. 133/138, but declines after SFAS Nos. 149
and 155. Together, these results help validate the main results by providing further evidence that
analyst coverage is related to changes in the reporting practices of derivatives. Note, however, that
we do not infer causality between the regulatory events and coverage. Instead, these results show
a statistical relation between trends in coverage and periods of time in which the FASB was
actively changing derivatives reporting.
7.2 Forecast frequency
Table 5 suggests that analysts respond to the increased demand for their services due to the
complexity of derivatives by initiating coverage for these firms. Another potential response to such
heightened demand is for analysts to issue forecasts for the earnings of New Users more frequently.
To investigate this possibility, we estimate Eq. (2) after replacing the dependent variable with the
number of annual earnings forecasts issued by each analyst covering a firm (FREQ). We use data
for the year before (t−1) and upon (t) initiation to hold analyst coverage constant across the pre-
and post-initiation periods. We find an insignificant coefficient (unreported) for NEWUSER×
POST (p>0.34), suggesting that forecast frequency for New Users does not change after initiation.
7.3 Ineffective management of financial risk
Corporate use of derivatives is often motivated by their ability to hedge (reduce) the risk
33 Structural break analysis accounts for changes in both intercept and slope. In our model, the coefficients for the interactions (regime indicators) capture shifts in slope (intercept). We focus on interaction coefficients as they are indicative of a parameter change among derivatives users in a given reporting regime (Donohoe and McGill 2011). We are primarily interested in whether model parameters are non-stationary over time.
29
of unfavorable changes in interest rates, foreign exchange rates, and commodity prices (Guay and
Kothari 2003). However, some firms fail to hold effective hedges or intentionally speculate (e.g.,
Hentschel and Kothari 2001). As noted by Chang et al. (2016), these distinctions are important for
analysts as many of the key benefits of derivatives, such as less volatile earnings and cash flow,
only result from effective hedging (see Section 2). Thus, we consider whether analyst coverage is
influenced by how successfully firms use derivatives to manage risk exposures. On the one hand,
analysts may prefer to cover effective hedgers as their earnings are better insulated from exposures
to risk. On the other hand, analysts’ reports are potentially more valuable to investors when firms
fail to hold effective hedges or speculate in the derivatives market.
We classify New Users as effective or speculative/ineffective hedgers following Zhang
(2009).34 We estimate three models quantifying how New Users’ exposures to IRISK, FRISK, and
CRISK relate to firm features prior to derivatives usage. Using the resulting coefficients, we
compute New Users’ expected risk exposures after using derivatives. We label a New User as an
effective hedger if at least two risk exposures are less than expected after initiation. Of the 587
New Users, 499 are classified as an effective hedger; speculative/ineffective otherwise.
We then estimate Eq. (2) with two modifications for the overall effects of initiation. First,
we replace NEWUSER with (1) EH, which is equal to 1 for effective hedgers (0 otherwise), and
(2) SPIN, which is equal to 1 for speculative/ineffective hedgers (0 otherwise). Second, we interact
these two variables with POST. Consistent with our main results, the (unreported) results indicate
an increase in analyst coverage after initiation for both effective and speculative/ineffective
34 Zhang (2009) introduces a procedure for identifying firms that reduce specific risk exposures after holding all of their derivatives positions. We use this procedure because (1) it provides a reasonable ex post assessment of hedge effectiveness as few firms disclose parameters that help discern the extent to which a risk exposure is hedge (Kawaller 2004), and (2) by focusing on outcomes, it is not confounded by a firm’s discretion in hedge designation or choice to practice hedge accounting. See Zhang (2009) for details. Following Donohoe (2015b), we make two modifications to quantify factors of FRISK: (1) we use foreign income (pifo in Compustat) as an explanatory variable rather than foreign sales; and (2) due to data availability, we omit the sum of industry imports and exports as an explanatory variable.
30
hedgers. However, the increases in coverage are not statistically different between the two groups
(p>0.98), suggesting that, on average, analysts have no preference for covering firms whose
earnings are potentially better insulated from risk exposures.
We also examine whether the increase in coverage for effective and speculative/ineffective
hedgers differs by analyst expertise. For these tests, the dependent variable in the modified Eq. (2)
is the number of low career experience (COV_LOEXP) and non-All-Star (COV_NONSTAR)
analysts covering a firm. In both cases, only the coefficients for NEW_EH×POST are positive and
significant, indicating that analysts with less expertise tend to cover effective hedgers (rather than
speculative/ineffective hedgers) after initiation. One explanation is that the earnings of effective
hedgers are viewed as easier to forecast as such firms are likely insulated from risk exposures.
7.4 Firm complexity and size
The use of a difference-in-differences design with an untreated control group and numerous
covariates alleviates many alternative explanations. However, we conduct two additional tests to
further mitigate the concern that other types of complexity and/or firm size drive the main results.
First, we estimate Eq. (2) with two modifications. We replace NEWUSER with (1) an indicator
variable equal to 1 for New Users in the lowest decile of either GSC or ISC, the two complexity
measures in our model (0 otherwise), and (2) an indicator variable equal to 1 for New Users not in
the lowest decile of either measure (0 otherwise). We then interact these variables with POST.
Second, we partition New Users by the highest decile of SIZE in the same manner. For complexity,
we find that the effects of initiation on analyst coverage are not significant for geographically
complex firms (GSC). Further, while the effects are significant for industry sales concentration
(ISC) overall, they are no different among low versus high ISC partitions. For size, we also find
no difference among the partitions. Together, these (unreported) results indicate that the effects of
31
derivatives initiation on analyst coverage are not likely driven by firm complexity or size.
8. Conclusion
We examine whether and how the complexity of derivatives influences analysts’ decisions
to cover a firm. Using the focused setting of new derivatives users and a difference-in-differences
design, we find that analyst coverage increases among firms that begin using derivatives during
1998-2011. This increase in coverage is driven by analysts with low expertise, as reflected by both
career experience and All-Star status. We also find that the accuracy (dispersion) of earnings
forecasts for new derivatives users decrease (increases) after derivatives initiation only when the
forecasts are issued by analysts with less expertise. Therefore, the evidence collectively suggests
that, despite the renowned complexity of derivatives, analysts with less expertise begin covering
new derivatives users, and subsequently produce the less accurate and more dispersed earnings
forecasts shown by prior research for these firms (Chang et al. 2016).
We note three key caveats. First, a firm’s decision to use derivatives is determined, in part,
by unobservable factors, making it difficult to eliminate all alternative explanations. However, we
mitigate many of the major concerns by (1) using propensity score matching and a difference-in-
differences design; (2) validating our findings in the broader setting of derivatives users; and (3)
evaluating if ineffective risk management, firm complexity, or size drive the results. Second, we
note that our structural break tests among derivatives users preclude any inferences other than a
statistical relation. Finally, the effects of derivatives on analyst coverage could vary depending on
whether a firm hedges a majority or a small portion of its risk exposures (Ryan 2012). While
current disclosure practices limit the ability of researchers to obtain such information, we believe
this is an issue that should be addressed by future research.
32
References Abdel-khalik, A.R., Chen, P., 2015. Growth in financial derivatives: the public policy and accounting incentives. Journal of Accounting and Public Policy 34, 291-318.
Aretz, K., Bartram, S., 2010. Corporate hedging and shareholder value. Journal of Financial Research 33, 317-371.
Armstrong, C., Blouin, J., Larcker, D., 2012. The incentives for tax planning. Journal of Accounting and Economics 53, 391-411.
Bagnoli, M., Watts, S., Zhang, Y., 2008. Reg-FD and the competitiveness of all-star analysts. Journal of Accounting and Public Policy 27, 295-316.
Bank for International Settlements, 2013. OTC derivatives market activity in the second half of 2013. Monetary and Economic Department, Basel, Switzerland.
Barth, M., Hutton, A., 2004. Analyst earnings forecast revisions and the pricing of accruals. Review of Accounting Studies 9, 59-96.
Barth, M., Kasznik, R., McNichols, M., 2001. Analyst coverage and intangible assets. Journal of Accounting Research 39, 1-34.
Barton, J., 2001. Does the use of financial derivatives affect earnings management decisions? The Accounting Review 76, 1-26.
Bartram, S., Brown, G., Fehle, F., 2009. International evidence on financial derivatives usage. Financial Management 38, 185-206.
Becker, S., Caliendo, M., 2007. Sensitivity analysis for average treatment effects. The STATA Journal 7, 71-83.
Bergstresser, D., Philippon, T., 2006. CEO incentives and earnings management. Journal of Financial Economics 80, 511-529.
Beyer, A., Cohen, D.A., Lys, T.Z., Walther, B.R., 2010. The financial reporting environment: review of the recent literature. Journal of Accounting and Economics 50, 296-343.
Bhushan, R., 1989. Firm characteristics and analyst following. Journal of Accounting and Economics 11, 255-274.
Bodnar, G.M., de Jong, A., Macrae, V., 2003. The impact of institutional differences on derivatives usage: a comparative study of U.S. and Dutch firms. European Financial Management 32, 35-67.
Botosan, C., Harris, M., 2000. Motivations for a change in disclosure frequency and its consequences: an examination of voluntary quarterly segment disclosures. Journal of Accounting Research 38, 329-353.
Brennan, M., Subrahmanyam, A., 1995. Investment analysis and price formation in securities markets. Journal of Financial Economics 38, 361-381.
Brown, L., Call, A., Clement, M., Sharp, N., 2015. Inside the “black box” of sell-side fiancial analsyts. Journal of Accounting Research 53, 1-47.
Bushman, R., Chen, Q., Engel, E., Smith, A., 2004. Financial accounting information, organizational complexity and corporate governance systems. Journal of Accounting and Economics 37, 167-201.
Caliendo, M., Kopeinig, S., 2008. Some practical guidance for the implementation of propensity score matching. Journal of Economic Survey 22, 31-72.
Cameron, A.C., Trivedi, P., 2009. Microeconometrics Using Stata, 1st ed. Stata Press, College Station, TX.
Campbell, T., Kracaw, W., 1990. Corporate risk management and the incentive effects of debt. The Journal of Finance 45, 1673-1686.
Chang, H.S., Donohoe, M., Sougiannis, T., 2016. Do analysts understand the economic and reporting complexities of derivatives? Journal of Accounting and Economics 61, 584-604.
Chen, X., Cheng, Q., Lo, K., 2010. On the relationship between analyst reports and corporate disclosures: Exploring the roles of information discovery and interpretation. Journal of Accounting
33
and Economics 49, 206-226. Clarke, J., Khorana, A., Patel, A., Rau, R., 2007. The impact of all-star analyst job changes on their coverage choices and investment banking deal flow. Journal of Financial Economics 84, 713-737.
Clement, M., 1999. Analyst forecast accuracy: do ability, resources and portfolio complexity matter? Journal of Accounting and Economics 27, 285-303.
Cohen, J. 1988. Statistical Power Analysis for the Behavioral Sciences. 2nd ed. Lawrence Erlbaum Associates, Inc., Hillsdale, NJ.
De Franco, G., Kothari, S.P., Verdi, R., 2011. The benefits of financial statement comparability. Journal of Accounting Research 49, 895-931.
Donohoe, M., 2015a. Financial derivatives in corporate tax avoidance: a conceptual perspective. Journal of the American Taxation Association 37, 37-68.
Donohoe, M., 2015b. The economic effects of financial derivatives on corporate tax avoidance. Journal of Accounting and Economics 59, 1-24.
Donohoe, M., McGill, G., 2011. The effects of increased book-tax difference tax return disclosures on firm valuation and behavior. Journal of the American Taxation Association 33, 35-65.
Dyreng, S.D., Hanlon, M., Maydew, E.L., 2008. Long-run corporate tax avoidance. The Accounting Review 83, 61-82.
Faul, F., Erdfelder, E., Lang, A., Buchner, A., 2007. G*Power 3: a flexible statistical power analysis program for the social, behavioral, and biomedical sciences. Behavior Research Methods 39, 175-191.
Financial Accounting Standards Board (FASB), 1998. Accounting for Derivative Instruments and Hedging Activities. Statement of Financial Accounting Standard No. 133. FASB, Norwalk, CT.
Financial Accounting Standards Board (FASB), 2000. Accounting for Certain Derivative Instruments and Certain Hedging Activities. Statement of Financial Accounting Standard No. 138. FASB, Norwalk, CT.
Financial Accounting Standards Board (FASB), 2003. Amendment of Statement 133 on Derivative Instruments and Hedging Activities. Statement of Financial Accounting Standard No. 149. FASB, Norwalk, CT.
Financial Accounting Standards Board (FASB), 2006. Accounting for Certain Hybrid Financial Instruments. Statement of Financial Accounting Standard No. 155. FASB, Norwalk, CT.
Financial Accounting Standards Board (FASB), 2008. Disclosures about Derivatives Instruments and Hedging Activities. Statement of Financial Accounting Standard No. 161. FASB, Norwalk, CT.
Froot, K., Scharfstein, D., Stein, J., 1993. Risk management: coordinating corporate investment and financing policies. The Journal of Finance 48, 1629-1658.
Guay, W., 1999. The impact of derivatives on firm risk: an empirical examination of new derivative users. Journal of Accounting and Economics 26, 319-351.
Guay, W., Kothari, S.P., 2003. How much do firms hedge with derivatives? Journal of Financial Economics 70, 423-461.
Hainmueller, J., 2012. Entropy balancing for causal effects: a multivariate reweighting method to produce balanced samples in observational studies. Political Analysis 20, 25-46.
Hayes, R.M., 1998. The impact of trading commission incentives on analysts’ stock coverage decisions and earnings forecasts. Journal of Accounting Research 36, 299-320.
Hentschel, L., Kothari, S.P., 2001. Are corporations reducing or taking risks with derivatives? Journal of Financial and Quantitative Analysis 36, 93-118.
Healy, P., Palepu, K., 2001. Information asymmetry, corporate disclosure, and the capital markets: a review of the empirical disclosure literature. Journal of Accounting and Economics 31, 405-440.
Holland, K., Glasgall, W., 1994. A Black Hole in the Balance Sheet: A Spate of Big Losses from Derivatives Sparks a Call for Disclosure. Business Week (May 16).
34
Hong, H., Kubik, J., Solomon, A., 2000. Security analysts’ career concerns and herding of earnings forecasts. Rand Journal of Economics 31, 121-144.
Irvine, P., 2004. Analysts’ forecasts and brokerage-firm trading. The Accounting Review 79, 125-149.
Kawaller, I., 2004. What analysts need to know about accounting for derivatives. Financial Analysts Journal 60, 24-30.
Koonce, L., Lipe, M., McAnally, M., 2005. Judging the risk of financial instruments: problems and potential remedies. The Accounting Review 80, 871-895.
Lang, M., Lins, K., Miller, D., 2004. Concentrated control, analyst following, and valuation: do analysts matter most when investors are protected least? Journal of Accounting Research 42, 589-623.
Lang, M., Lundholm, R., 1996. Corporate disclosure policy and analyst behavior. The Accounting Review 71, 467-492.
Lawrence, A., Ryans, J., Sun, Y., 2014. Investors’ demand for sell-side research. University of California, Berkeley. Working Paper.
Lehavy, R., Li, F., Merkley, K., 2011. The effect of annual report readability on analyst following and the properties of their earnings forecasts. The Accounting Review 86, 1087-1115.
Leone, M., 2007. Hedge Accounting: A Matchless Future? CFO.com (May 24). Leone, A., Wu, J., 2007. What does it take to become a superstar? Evidence from institutional investor rankings of financial analysts. University of Rochester. Working Paper.
Lin, H., McNichols, M., 1998. Underwriting relationships, analysts’ earnings forecasts and investment recommendations. Journal of Accounting and Economics 25, 101-127.
Maines, L., McDaniel, L., Harris, M., 1997. Implications of proposed segment reporting standards for financial analysts’ investment judgments. Journal of Accounting Research 35, 1-24.
Mantel, N., Haenszel, W., 1959. Statistical aspects of the analysis of data from retrospective studies of disease. Journal of the National Cancer Institute 22, 719-748.
Mayers, D., Smith, C., 1982. On the corporate demand for insurance. Journal of Business 55, 281-296.
McInnis, J., Collins, D.W., 2011. The effect of cash flow forecasts on accrual quality and benchmark beating. Journal of Accounting and Economics 51, 219-239.
McNichols, M., O’Brien, P., 1997. Self-selection and analyst coverage. Journal of Accounting Research 35, 167-199.
Merton, R., 1987. A simple model of capital market equilibrium with incomplete information. The Journal of Finance 42, 483-510.
Mikhail, M., Walther, B., Willis, R., 1997. Do security analysts improve their performance with experience? Journal of Accounting Research 35, 131-157.
Nance, D., Smith, C., Smithson, C., 1993. On the determinants of corporate hedging. The Journal of Finance 48, 267-284.
Petersen, M.A., 2009. Estimating standard errors in finance panel data sets: comparing approaches. Review of Financial Studies 22, 435-480.
Peterson, K., 2012. Accounting complexity, misreporting, and the consequences of misreporting. Review of Accounting Studies 17, 72-95.
Previts, G., Bricker, R., Robinson, T., Young, S., 1994. A content analysis of sell-side financial analyst company reports. Accounting Horizons 8, 55-70.
Ramnath, S., Rock, S., Shane, P., 2008. The financial analyst forecasting literature: a taxonomy with suggestions for further research. International Journal of Forecasting 24, 34-75.
Roberts, M., Whited, T., 2013. Endogeneity in empirical corporate finance. Handbook of the Economics of Finance 2, 493-572.
35
Rosenbaum, P., 2007. Sensitivity analysis for m-estimates, tests, and confidence intervals in matched observational studies. Biometrics 63, 456-464.
Rosenbaum, P., Rubin, D., 1983. The central role of the propensity score in observational studies for causal effects. Biometrika 70, 41-55.
Rubin, D., 2009. Should observational studies be designed to allow lack of balance in covariate distributions across treatment groups? Statistics in Medicine 28, 1420-1423.
Ryan, S.G., 2012. Financial reporting for financial instruments. Foundations and Trends in Accounting 6, 187-354.
Ryan, S.G., 2007. Financial Instruments & Institutions: Accounting and Disclosure Rules, 2nd ed. John Wiley & Sons, Inc., Hoboken, NJ.
Securities and Exchange Commission (SEC), 2008. Final report of the advisory committee on improvements to financial reporting to the United States securities and exchange commissions.
Shumway, T., 2001. Forecasting bankruptcy more accurately: a simple hazard model. Journal of Business 74, 101-124.
Skinner, D., 1996. Are disclosures about bank derivatives and employee stock options ‘value relevant’? Journal of Accounting and Economics 22, 393-405.
Smith, C.W., Stulz, R.M., 1985. The determinants of firms’ hedging policies. Journal of Financial and Quantitative Analysis 20, 391-405.
Tamura, H., 2002. Individual-analyst characteristics and forecast error. Financial Analysts Journal 58, 28-35.
Tan, H., Wang, S., Welker, M., 2011. Analyst following and forecast accuracy after mandated IFRS adoptions. Journal of Accounting Research 49, 1307-1357.
Valladares, M., 2014. Derivatives Transparency Is Essential to End TBTF. The New York Times (May 13).
Zhang, H., 2009. Effect of derivative accounting rules on corporate risk management behavior. Journal of Accounting and Economics 47, 244-264.
36
Appendix Variable definitionsa
Dependent variables
COV Analyst coverage, defined as the total number of analysts covering firm i in year t (obtained from I/B/E/S detail file).
COV_HIEXP Analyst coverage from high career experience analysts, defined as the total number of analysts covering firm i in year t that are in the top quintile of the number of years (since 1980) issuing forecasts for any firm in I/B/E/S (obtained from I/B/E/S detail file).
COV_LOEXP Analyst coverage from high career experience analysts, defined as the total number of analysts covering firm i in year t that are not in the top quintile of the number of years (since 1980) issuing forecasts for any firm in I/B/E/S (obtained from I/B/E/S detail file).
COV_STAR Analyst coverage from All-Star analysts, defined as the total number of analysts covering firm i in year t that are designated as All-Stars by Institutional Investor Magazine (see Leone and Wu 2007).
COV_NOSTAR Analyst coverage from non-All-Star analysts, defined as the total number of analysts
covering firm i in year t that are not designated as All-Stars by Institutional Investor Magazine (see Leone and Wu 2007).
AEFA_HIEXP Analyst earnings forecast accuracy for high career experience analysts, defined as the absolute value of the difference between the last earnings forecast issued by analyst j and actual earnings scaled by stock price of year t for firm i. We multiply the result by −100 such that greater values indicate more accurate forecasts, and then compute the mean of individual analyst forecast accuracy by firm-year and expertise group. See Tan et al. (2011). High career experience analysts are those in the top quintile of the number of years (since 1980) issuing forecasts for any firm in I/B/E/S (obtained from I/B/E/S detail file).
AEFA_LOEXP Analyst earnings forecast accuracy for low career experience analysts, where accuracy is defined above. Low career experience analysts are those not in the top quintile of the number of years (since 1980) issuing forecasts for any firm in I/B/E/S (obtained from I/B/E/S detail file).
AEFD_HIEXP Analyst earnings forecast dispersion for high career experience analysts, defined as the absolute value of the difference between the last earnings forecast issued by analyst j and the average of forecasts made by all other analysts except analyst j covering firm i in year t. We scale the absolute value by stock price at the end of year t for firm i and multiply the result by 100. We then take the mean of individual analyst forecast deviation from the consensus forecast by firm-year and expertise group. See Hong et al. (2000). High career experience analysts are those in the top quintile of the number of years (since 1980) issuing forecasts for any firm in I/B/E/S (obtained from I/B/E/S detail file).
AEFD_LOEXP Analyst earnings forecast dispersion for low career experience analysts, where dispersion is defined above. Low career experience analysts are those not in the top quintile of the number of years (since 1980) issuing forecasts for any firm in I/B/E/S (obtained from I/B/E/S detail file).
Variables of interest
NEWUSER Indicator variable equal to 1 for all New User firm observations and 0 for all matched
37
control firm observations.
POST Indicator variable equal to 1 for both New User and matched control firm observations in periods after derivatives initiation; 0 otherwise.
Risk management incentives (RMI)
IRISK Interest rate risk exposures, defined as the absolute value of the estimated coefficient from a regression of firms’ monthly holding period stock returns on the monthly percentage change in the London Interbank Offered Rate (LIBOR) for 36 months prior to fiscal-year end. See Guay (1999), Zhang (2009), and Donohoe (2015b).
FRISK Foreign currency exchange rate risk exposures, defined as the absolute value of the estimated coefficient from a regression of firms’ monthly holding period stock returns on the monthly percentage change in the Federal Reserve Board trade-weighted U.S. dollar index for 36 months prior to fiscal-year end. See Guay (1999), Zhang (2009), and Donohoe (2015b).
CRISK Commodity price risk exposures, defined as the absolute value of the estimated coefficient from a regression of firms’ monthly holding period stock returns on the monthly percentage change in the Producer Price Index for 36 months prior to fiscal-year end. See Guay (1999), Zhang (2009), and Donohoe (2015b).
ALTZ Likelihood of entering financial distress, defined as the modified Altman-Z score based on parameter weights reported by Shumway (2001).
USCORE Likelihood of underinvestment, defined by first ranking cash flow from operations (oancf), debt-to-assets ratio (lt/at), and scores from a factor analysis of four growth opportunity measures (prior investment activity, geometric growth in market value of assets, market-to-book ratio, and research and development into deciles by year and industry. Decile ranks for debt-to-asset ratios and growth opportunity factor scores are then added to the reverse decile rank for cash flow from operations, with the result scaled by 30 (total possible points). See Donohoe (2015b).
ECSENS Sensitivity of executive compensation to firm value, defined by first computing the dollar change in value of CEO stock and option holdings that would result from a one percentage point increase in the stock price of the firm (0.01×prcc_f×[shrown_tot+ opt_unex_exer_num]). The result is then normalized by the sum of CEO salary and bonus (salary+bonus) to capture the share of total CEO compensation that would result from a one percentage point increase in firm value. Compensation data obtained from Execucomp. See Bergstresser and Philippon (2006).
CETR Cash effective tax rate (3-year), defined as the three-year sum (t to t+2) of worldwide cash taxes paid (txpd) divided by the three-year sum (t to t+2) of pre-tax book income (pi) less special items (spi). ETRs are reset to 1 (0) if greater (less) than 1 (0). See Dyreng et al. (2008).
CDEBT Convertible debt, defined as convertible debt (dcvt) divided by lagged total assets (at).
PSTOCK Preferred stock, defined as preferred stock (pstk) divided by lagged total assets (at).
ABACC Abnormal accruals, based on the performance-matched modified Jones model.
CFV Cash flow volatility, defined as the standard deviation of quarterly operating cash flows (oancfy, adjusted to reflect quarterly data) during the most recent two years.
EV Earnings volatility, defined as the standard deviation of quarterly earnings before
38
extraordinary items (ibq) during the most recent two years.
Analyst coverage incentives (ACI)
SIZE Log of equity market value (prcc_f×csho) at beginning of year t.
INTANG Ratio of intangible assets (intan) to total assets (at) at beginning of year t.
RETVOL Return volatility, defined as the standard deviation of monthly stock returns for firm i at year t−1. Ranked into deciles, where decile ranks are transformed by dividing the rank by 9 and subtracting 0.5 such that values range from −0.50 to 0.50.
MB Market to book ratio, defined as equity market value (prcc_f×csho) divided by book value of equity (at−lt−pstkl+txditc+dcvt) at beginning of year t.
ISSUE Indicator variable equal to 1 if firm issues equity or debt greater than 5 percent of total assets in year t; 0 otherwise.
TURNO Stock turnover, defined as the number of shares traded in year t, divided by the average number of shares outstanding in year t.
ANRET Annual stock return for firm i in year t−1, adjusted for contemporaneous annual market return.
Control variables (CTRL)
GSC Geographic sales concentration, defined as the sum of squares of (firm sales in each geographic segment / total firm sales) (obtained from the Compustat Segments File).
ISC Industry sales concentration, defined as the sum of squares of (firm sales in each industry segment / total firm sales) (obtained from the Compustat Segments File).
ROA Return on assets, defined as income before extraordinary items (ib) divided by total assets (at) at beginning of year t.
FRGN Indicator variable equal to 1 if foreign income or loss (pifo) is not equal to 0; 0 otherwise.
M&A Indicator variable equal to 1 if cash flow from mergers and acquisitions (aqc) is not equal to 0; 0 otherwise.
Other variables
INIT Indicator variable equal to 1 for New Users and 0 for “pure” Non-Users observations (i.e., firms that do not use derivatives at any point during the sample period).
USER Indicator variable equal to 1 if the firm reports a position in derivatives in both fiscal years t and t−1 (i.e., a User, but not a New User); 0 if the firm does not report a position in derivatives. This variable identifies the User (=1) and Non-User (=0) samples.
SURPRISE Earnings surprise, defined as the absolute value of the difference between earnings per share in year t and year t−1, divided by stock price at the beginning of year t.
SFAS133/138 Indicator variable equal to 1 for observations after fiscal-year June 2000; 0 otherwise.
SFAS149 Indicator variable equal to 1 for observations after fiscal-year June 2003; 0 otherwise.
SFAS155 Indicator variable equal to 1 for observations after fiscal-year September 2006; 0 otherwise.
SFAS161 Indicator variable equal to 1 for observations after fiscal-year November 2008; 0 otherwise.
39
FREQ Forecast frequency, defined as the number of annual earnings forecasts issued by each analyst covering firm i in year t (obtained from the I/B/E/S detail file).
EH Indicator variable equal to 1 if the firm effectively hedges (reduces) its exposure to at least two risks (interest rate (IRISK), foreign exchange rate (FRISK), or commodity price (CRISK) risks) relative to expectations after derivatives initiation; 0 otherwise. We define risk exposure reductions in the bottom decile as immaterial. See Zhang (2009) for details.
aCompustat mnemonics in parentheses.
40
Table 1 Characteristics of Non-Users, Users, and New Users
Panel A: Temporal distribution of sample observations by derivatives reporting regimea
1/1/98 to 6/15/00 to 6/30/03 to 9/15/06 to 11/15/08 to Total 6/14/00 6/29/03 9/14/06 11/14/08 12/31/11
Non-Users 1,327 3,068 2,866 1,673 1,953 10,887 Users 765 2,602 3,448 2,310 2,892 12,017 New Users 48 228 123 68 120 587 Total 2,140 5,898 6,437 4,051 4,965 23,491
Panel B: Industry distribution of sample observations
Non-Users Users New Users Industry groupb Obs. % Obs. % Obs. %
Consumer Non-Durables 483 4 925 8 29 5 Consumer Durables 336 3 430 4 27 5 Manufacturing 1,120 10 2,781 23 101 17 Energy & Extraction 215 2 793 7 15 3 Chemicals & Allied Products 145 1 538 4 10 2 Business Equipment 3,211 29 2,307 19 176 30 Telecommunications 236 2 435 4 21 4 Wholesale & Retail 1,524 14 1,286 11 79 13 Healthcare 2,028 19 946 8 59 10 Constr., Transport. & Services 1,589 15 1,576 13 70 12 Total 10,887 12,017 587
This table presents characteristics of Non-Users, Users, and New Users. Panel A illustrates the temporal distribution of sample observations, and Panel B reports industry distributions. A firm is a New User if it did not report a position in derivatives when it first appears in the sample, but did in a subsequent year. Firms enter the New User sample only when derivatives usage is first observed after initially observing no usage (see footnote 9 for an example). A firm is a User if it reports a position in derivatives at the end of both fiscal years t and t−1 (i.e., uses derivatives, but is not a new user), and a Non-User if it reports no position in derivatives at fiscal year-end. aConsistent with prior studies (Guay 1999; Donohoe 2015b), the sum of Users and New Users in a given year does not necessarily equal the amount of Users in the subsequent year because a firm can enter the sample for the first time as a User, but not a New User. bFama-French industry groups are available at: http://mba.tuck.dartmouth.edu/pages/Faculty/ken.french/datalibrary.html.
SFAS Nos. 133/138
SFAS No. 149
SFAS No. 155
SFAS No. 161
41
Table 2 Descriptive statistics
(1) (2) (3) Non-Users Users New Users Mean Median Mean Median t-stat [(1)−(2)] Mean Median t-stat [(1)−(3)]
Dependent variable COV 7.429 5.000 10.987 9.000 −33.68 8.828 7.000 −4.96
Risk management incentives (RMI) IRISK 0.004 0.002 0.002 0.001 23.59 0.004 0.002 1.97 FRISK 0.019 0.011 0.014 0.008 21.74 0.018 0.012 1.07 CRISK 0.021 0.011 0.014 0.008 21.39 0.026 0.015 −3.75 ALTZ 5.912 4.060 3.403 2.675 34.62 3.955 2.959 6.76 USCORE 0.546 0.533 0.527 0.533 9.53 0.547 0.533 −0.23 ECSENS 0.060 0.000 0.113 0.056 −28.52 0.073 0.020 −2.49 CETR 0.174 0.096 0.207 0.185 −11.81 0.204 0.164 −3.38 CDEBT 0.028 0.000 0.027 0.000 0.78 0.047 0.000 −4.83 PSTOCK 0.004 0.000 0.003 0.000 2.80 0.005 0.000 −0.48 ABACC −0.012 −0.005 −0.015 −0.008 1.60 −0.017 −0.011 0.73 CFV 0.038 0.028 0.027 0.020 26.81 0.033 0.024 3.87 EV 0.033 0.015 0.019 0.009 17.91 0.023 0.011 3.19
Analyst coverage incentives (ACI) SIZE 5.802 5.711 7.121 7.044 −59.85 6.361 6.295 −8.32 INTANG 0.124 0.050 0.173 0.111 −21.18 0.178 0.113 −7.85 RETVOL 0.117 0.167 −0.038 −0.056 39.74 0.069 0.056 3.98 MB 3.085 2.065 2.331 1.684 16.69 2.837 1.999 1.55 ISSUE 0.068 0.000 0.104 0.000 −9.67 0.063 0.000 0.47 TURNO 2.092 1.510 2.163 1.684 −3.01 2.320 1.721 −2.90 ANRET 0.139 −0.007 0.104 0.000 2.87 0.139 −0.025 0.00
Control variables (CTRL) GSC 0.768 0.890 0.643 0.595 32.51 0.721 0.759 4.23 ISC 0.813 1.000 0.692 0.687 32.38 0.732 0.844 7.56 ROA −0.026 0.035 0.028 0.042 −24.38 0.016 0.040 −4.84 FRGN 0.475 0.000 0.692 1.000 −34.09 0.650 1.000 −8.36 M&A 0.398 0.000 0.578 1.000 −27.69 0.537 1.000 −6.75
Obs. 10,887 12,017 587 This table reports descriptive statistics, along with t-statistics for mean tests of differences between Non-Users and that of Users and New Users (assuming unequal variance). Continuous variables are winsorized at 1st and 99th percentiles. Bold t-statistics denote statistical significance of at least 0.10 (two-tailed). Variables are defined in the Appendix.
42
Table 3 Covariate balance
Mean Difference Median Difference Distributional Difference p-value p-value p-value Risk management incentives (RMI) IRISK 0.309 0.588 0.102 FRISK 0.915 0.120 0.058* CRISK 0.233 0.517 0.101 ALTZ 0.762 0.960 0.214 USCORE 0.961 0.989 1.000 ECSENS 0.663 0.000*** 0.000*** CETR 0.855 0.564 0.225 CDEBT 0.677 0.574 1.000 PSTOCK 0.828 0.623 1.000 ABACC 0.408 0.507 0.535 CFV 0.995 0.848 0.973 EV 0.781 0.813 0.731
Analyst coverage incentives (ACI) SIZE 0.564 0.771 0.108 INTANG 0.509 0.625 0.842 RETVOL 0.489 0.634 0.969 MB 0.970 0.682 0.837 ISSUE 0.295 0.295 1.000 TURNO 0.397 0.384 0.337 ANRET 0.610 0.123 0.159
Control variables (CTRL) GSC 0.734 0.623 0.845 ISC 0.579 0.509 0.602 ROA 0.873 0.772 0.076* FRGN 0.908 0.908 1.000 M&A 0.697 0.697 1.000 Hotelling’s T2 0.998
This table reports the covariate balance between the 587 New Users and 587 propensity score matched control firms in the match year. Reported values are p-values for tests of differences in means (t-tests), medians (Wilcoxon rank-sum test), and distributions (Kolmogorov-Smirnov homogeneous distributions test) of matching variables (Eq. [1]). Hotelling’s T2 test is the multivariate equivalent of the two-sample t-test and considers whether the vector of all variable means differ between the two groups. *, **, and *** denote statistical significance levels of 0.10, 0.05, and 0.01, respectively (two-tailed). Variables are defined in the Appendix.
43
Table 4 Difference-in-differences tests of analyst coverage
(1) (2) COV COV Immediate Effects Overall Effects Coeff. RSE Coeff. RSE
Intercept −0.750 *** 0.147 −0.795 *** 0.118 NEWUSER 0.016 0.036 0.076 ** 0.032 POST −0.045 ** 0.026 −0.040 0.041 NEWUSER×POST 0.105 *** 0.026 0.091 ** 0.044
Analyst coverage incentives (ACI) SIZE 0.359 *** 0.012 0.319 *** 0.011 INTANG 0.352 *** 0.087 0.342 *** 0.068 RETVOL 0.357 *** 0.060 0.164 *** 0.033 MB 0.000 0.004 0.004 0.003 ISSUE −0.024 0.049 −0.047 0.040 TURNO 0.122 *** 0.010 0.134 *** 0.007 ANRET −0.183 *** 0.019 −0.164 *** 0.011
Control variables (CTRL) GSC 0.110 0.074 0.146 *** 0.055 ISC 0.221 *** 0.058 0.200 *** 0.047 ROA 0.098 0.104 0.076 0.060 FRGN −0.030 0.040 −0.008 0.030 M&A 0.053 * 0.028 0.023 0.018
Industry Included Included Year Included Included Pseduo R2 0.14 0.19 Wald χ2 (model) 2,394.24*** 3,340.25*** Observations 2,348 9,826
This table reports tests of whether the complexity of derivatives initiation influences analyst coverage (Eq. [2]). NEWUSER equals 1 for New User firm observations and 0 for matched control firm observations. POST equals 1 for periods after derivatives initiation for New Users and corresponding control firms (0 otherwise). The coefficients for NEWUSER, POST, and NEWUSER×POST reflect (a) differences in analyst coverage between New Users and control firms during the pre-initiation period, (b) the change in analyst coverage among control firms between pre- and post-initiation periods, and (c) the difference-in-differences estimator of the effect of derivatives initiation on analyst coverage for New Users relative to control firms, respectively. To assess the immediate effects of initiation, column (1) uses data for the year immediately before (t−1) and upon (t) initiation for the sample of 587 New Users and 587 matched control firms (2,348 firm-years). Column (2) reports results based on all available data before and after initiation (9,826 firm-years) to assess the overall effects of initiation. *, **, and *** denote statistical significance levels of 0.10, 0.05, and 0.01, respectively (two-tailed). Robust standard errors (RSE) are clustered by firm (Petersen 2009). Variables are defined in the Appendix.
44
Table 5 Difference-in-differences tests of analyst coverage by analyst expertise Panel A: High versus low career experience (1) (2)
COV_HIEXP COV_LOEXP High Career Experience Low Career Experience Coeff. RSE Coeff. RSE
Intercept −0.543 *** 0.182 −0.747 *** 0.144 NEWUSER −0.019 0.049 −0.031 0.036 POST 0.049 0.032 −0.028 0.022 NEWUSER×POST Ψ3 0.000 0.045 0.094 *** 0.028
Analyst coverage incentives (ACI) SIZE 0.257 *** 0.014 0.336 *** 0.012 INTANG −0.135 0.110 0.335 *** 0.083 RETVOL −0.033 0.067 0.380 *** 0.058 MB −0.006 0.005 0.003 0.004 ISSUE 0.017 0.062 −0.033 0.050 TURNO 0.038 *** 0.012 0.121 *** 0.009 ANRET −0.162 *** 0.031 −0.160 *** 0.020
Control variables (CTRL) GSC −0.043 0.096 0.042 0.072 ISC −0.072 0.071 0.253 *** 0.058 ROA 0.086 0.144 −0.005 0.095 FRGN −0.020 0.053 −0.022 0.039 M&A −0.035 0.038 0.061 ** 0.027
Industry Included Included Year Included Included Pseduo R2 0.12 0.16 Wald χ2 (model) 830.83*** 2,385.96*** Observations 1,487 2,214 Wald χ2: Ψ3 (1) = Ψ3 (2) 3.35**
45
Panel B: All-Star analysts versus non-All-Star analysts (1) (2)
COV_STAR COV_NOSTAR All-Star Analysts Other Analysts Coeff. RSE Coeff. RSE
Intercept 0.609 *** 0.228 −0.609 *** 0.135 NEWUSER −0.005 0.047 −0.012 0.034 POST 0.080 ** 0.039 −0.030 0.021 NEWUSER×POST Ψ3 −0.039 0.059 0.087 *** 0.026
Analyst coverage incentives (ACI) SIZE 0.249 *** 0.018 0.351 *** 0.011 INTANG −0.027 0.138 0.306 *** 0.081 RETVOL 0.219 *** 0.075 0.318 *** 0.056 MB 0.003 0.005 −0.000 0.004 ISSUE −0.030 0.057 −0.028 0.046 TURNO 0.080 *** 0.016 0.120 *** 0.009 ANRET −0.105 *** 0.038 −0.172 *** 0.018
Control variables (CTRL) GSC −0.077 0.122 0.074 0.071 ISC 0.103 0.082 0.210 *** 0.054 ROA 0.201 0.175 0.031 0.093 FRGN −0.009 0.057 −0.016 0.038 M&A −0.002 0.042 0.051 ** 0.026
Industry Included Included Year Included Included Pseduo R2 0.14 0.16 Wald χ2 (model) 807.15*** 2,852.57*** Observations 386 2,271 Wald χ2: Ψ3 (1) = Ψ3 (2) 4.75**
This table reports tests of whether the immediate effects of the complexity of derivatives initiation on analyst coverage are driven by analyst expertise (Eq. [2]). In Panel A, the dependent variable is the number of high (COV_HIEXP) and low (COV_LOEXP) career experience analysts covering firm i in year t in columns (1) and (2), respectively. We define career experience as the number of years (since 1980) that analyst j has issued forecasts for any firm in I/B/E/S. Analysts in the top quintile of career experience are designated as high experience; low otherwise. In Panel B, the dependent variable is the number of All-Star (COV_STAR) and non-All-Star (COV_NONSTAR) analysts covering firm i in year t in columns (1) and (2), respectively. All-Star analysts are selected annually by Institutional Investors Magazine. *, **, and *** denote statistical significance levels of 0.10, 0.05, and 0.01, respectively (two-tailed). Robust standard errors (RSE) are clustered by firm (Petersen 2009). Variables are defined in the Appendix.
46
Table 6 Transition tests of analyst coverage Panel A: Transition upon derivatives initiation (t) by career experience
Number of analysts covering New Users
t−2 and t−1 %
Car
eer
Exp
erie
nce 1 (Low) 2,069 25.15
2 1,412 17.17 3 1,633 19.85 4 1,573 19.12 5 (High) 1,539 18.71
Total 8,226 100.00 Panel B: Transition upon derivatives initiation (t) by All-Star status
Number of analysts covering New Users
t−2 and t−1 %
All
-Sta
r
NONSTAR 7,777 94.54 STAR by ranking: Rank 1 143 1.74 Rank 2 153 1.86 Rank 3 153 1.86
Total 8,226 100.00
Number of analysts ceasing coverage of New Users
t % t+1 %
Car
eer
Exp
erie
nce 1 (Low) 215 32.09 328 29.66
2 120 17.91 206 18.63 3 106 15.82 195 17.63 4 114 17.01 201 18.17 5 (High) 115 17.16 176 15.91
Total 670 100.00 1,106 100.00
Number of analysts initiating coverage of New Users
t % t+1 %
Car
eer
Exp
erie
nce 1 (Low) 661 38.59 619 41.82
2 305 17.81 271 18.31 3 278 16.23 230 15.54 4 257 15.00 196 13.24 5 (High) 212 12.38 164 11.08
Total 1,713 100.00 1,480 100.00
Number of analysts ceasing coverage of New Users
t % t+1 %
All
-Sta
r
NONSTAR 647 96.57 1,065 96.29 STAR by ranking: Rank 1 8 1.19 195 1.36 Rank 2 4 0.60 201 0.90 Rank 3 11 1.64 176 1.45
Total 670 100.00 1,106 100.00
Number of analysts initiating coverage of New Users
t % t+1 %
All
-Sta
r
NONSTAR 1,637 95.56 2,595 95.76 STAR by ranking: Rank 1 34 1.98 38 1.40 Rank 2 17 0.99 47 1.73 Rank 3 25 1.46 30 1.11
Total 1,713 100.00 2,710 100.00
47
Panel C: Transition upon derivatives initiation (t) among analysts already covering New Users by career experience
Panel D: Transition upon derivatives initiation (t) among analysts already covering Users by All-Star status
Number of analysts covering New User
firms before derivatives initiation t−2 and t−1 %
All
-Sta
r
NONSTAR 4,932 93.04 STAR by ranking: Rank 1 122 2.30 Rank 2 117 2.21 Rank 3 130 2.45
Total 5,301 100.00 Panel E: Transition upon derivatives initiation (t) among analysts ceasing coverage of New Users by outcome
Note: This table reports transition matrices that trace the proportion of analysts ceasing or initiation coverage of New Users upon and after derivatives initiation, across differing levels of expertise. Panels A and B trace the change in analyst coverage after derivatives initiation by career experience and All-Star status, respectively. To construct the matrices, we sort analysts covering a balanced sample of New Users from t−2 to t+1, where t is derivatives initiation, into quintiles based on level of career experience and All-Star status ranking (as defined in Section 6.4.1). The balanced sample (i.e., requiring observations for each year of the t−2 to t+1 window) helps ensure our tests focus on changes in coverage that are not due to variation in the number of New Users across time. We then count the number of analysts ceasing or initiating coverage upon (t) and one year after (t+1) derivatives initiation. Panels C and D trace the change in coverage after derivatives initiation by career experience and All-Star status, respectively, for analysts that already cover New Users before they initiate a derivatives program. These matrices consider how the proportion of analysts with prior knowledge of New Users (at the firm-level) changes after derivatives are introduced. To construct these matrices, we trace analysts already covering New Users upon derivatives initiation (i.e., not new analysts) back two years before derivatives initiation (t−2 and t−1) and track these analysts one year after derivatives initiation (t+1). Panel E traces the career outcomes of analysts ceasing coverage of New Users. We classify these analysts into four groups upon (t) and after (t+1) derivatives initiation: (1) promotion (moving to a larger brokerage; defined based on a quintile rank of all brokerages in I/B/E/S by number of analysts employed); demotion (moving to a smaller brokerage); (3) stay (no change in brokerage size quintile; and (4) missing (analyst no longer appears in I/B/E/S).
Number of analysts covering New Users after derivatives initiation
t % t+1 %
Car
eer
Exp
erie
nce 1 (Low) 1,007 19.00 501 14.13
2 919 17.34 602 16.98 3 1,100 20.75 788 22.23 4 1,190 22.45 867 24.46 5 (High) 1,085 20.47 787 22.20
Total 5,301 100.00 3,545 100.00
Number of analysts covering New User
firms before derivatives initiation t−2 and t−1 %
Car
eer
Exp
erie
nce 1 (Low) 1,030 19.43
2 893 16.85 3 1,118 21.09 4 1,188 22.41 5 (High) 1,072 20.22
Total 5,301 100.00
Number of analysts covering New Users after derivatives initiation
t % t+1 %
All
-Sta
r
NONSTAR 4,928 92.96 3,263 92.05 STAR by ranking: Rank 1 126 2.38 99 2.79 Rank 2 120 2.26 93 2.62 Rank 3 127 2.40 90 2.54
Total 5,301 100.00 3,545 100.00
Number of analysts ceasing coverage of New Users by career outcome t % t+1 % Promotion (to large brokerage) 8 1.19 16 1.45 Demotion (to small brokerage) 37 5.52 71 6.42 Stay (same brokerage) 318 47.46 583 52.71 Missing (disappear from I/B/E/S) 307 45.82 436 39.42 Total 670 100.00 1,106 100.00
48
Table 7 Difference-in-differences tests of analyst earnings forecast properties by analyst expertise Panel A: Analyst earnings forecast accuracy (AEFA)
(1) (2) AEFA_HIEXP AEFA_LOEXP High Career Experience Low Career Experience Coeff. RSE Coeff. RSE
Intercept −1.023 *** 0.164 −2.867 *** 0.415 NEWUSER 0.020 0.035 0.035 0.086 POST −0.018 0.022 0.096 * 0.053 NEWUSER×POST βa 0.000 0.033 −0.152 ** 0.072
Analyst coverage incentives (ACI) COV −0.001 0.003 0.011 0.007 SIZE 0.079 *** 0.016 0.160 *** 0.039 INTANG 0.281 *** 0.082 0.484 ** 0.218 RETVOL −0.019 *** 0.005 −0.068 *** 0.014 MB 0.005 0.004 0.020 * 0.012 ISSUE 0.029 0.047 0.108 0.108 TURNO 0.001 0.009 −0.041 0.025 ANRET −0.057 ** 0.024 −0.152 *** 0.054 SURPRISE −0.026 *** 0.002 −0.065 *** 0.006
Control variables (CTRL) GSC 0.048 0.070 −0.186 0.167 ISC −0.025 0.052 −0.035 0.141 ROA 0.003 ** 0.001 0.019 *** 0.003 FRGN −0.016 0.035 0.158 * 0.090 M&A −0.010 0.026 0.101 0.066
Industry Included Included Year Included Included Adjusted R2 0.37 0.39 Observations 1,271 1,892 Wald χ2: βa (1) = βa (2) 4.85**
49
Panel B: Analyst earnings forecast dispersion (AEFD)
(1) (2) AEFD_HIEXP AEFD_LOEXP High Career Experience Low Career Experience Coeff. RSE Coeff. RSE
Intercept 0.731 *** 0.110 1.673 *** 0.242 NEWUSER −0.003 0.025 −0.016 0.052 POST −0.000 0.017 −0.038 0.032 NEWUSER×POST βb 0.015 0.024 0.083 ** 0.043
Analyst coverage incentives (ACI) AFOL 0.000 0.002 −0.009 * 0.005 SIZE −0.055 *** 0.011 −0.074 *** 0.023 INTANG −0.194 *** 0.061 −0.389 *** 0.133 RETVOL 0.017 *** 0.004 0.054 *** 0.009 MB −0.006 ** 0.003 −0.020 *** 0.007 ISSUE −0.021 0.034 −0.098 0.064 TURNO −0.002 0.007 0.023 0.016 ANRET 0.026 0.017 0.055 0.036 SURPRISE 0.017 *** 0.002 0.039 *** 0.004
Control variables (CTRL) GSC −0.013 0.051 0.096 0.107 ISC 0.016 0.038 0.039 0.088 ROA −0.002 ** 0.001 −0.012 *** 0.002 FRGN 0.024 0.026 −0.097 * 0.055 M&A −0.022 0.019 −0.118 *** 0.042
Industry Included Included Year Included Included Adjusted R2 0.38 0.40 Observations 1,271 1,892 Wald χ2: βb (1) = βb (2) 2.87**
Note: This table reports tests of the accuracy (Panel A) and dispersion (Panel B) of analysts’ earnings forecasts for New Users by career experience. NEWUSER equals 1 for New User firm observations and 0 for matched control firm observations. POST equals 1 for periods after derivatives initiation for New Users and corresponding control firms (0 otherwise). The coefficients for NEWUSER, POST, and NEWUSER×POST reflect (a) differences in analyst earnings forecast properties between New Users and control firms during the pre-initiation period, (b) the change in analyst earnings forecast properties among control firms between pre- and post-initiation periods, and (c) the difference-in-differences estimator of the effect of derivatives initiation on analyst earnings forecast properties for New Users relative to control firms, respectively. To assess the immediate effects of initiation, these tests use data for the year immediately before (t−1) and upon (t) initiation for both New Users and matched control firms. *, **, and *** denote statistical significance levels of 0.10, 0.05, 0.01, respectively (two-tailed). Robust standard errors (RSE) are clustered by firm (Petersen 2009). Variables are defined in the Appendix.
50
Figure 1 Analyst coverage relative to derivatives initiation
Note: This figure plots the number of analysts covering New Users relative to the year of derivatives initiation (t−1 to t+1) for a balanced sample of 549 New Users. We restrict the sample to t−1 to t+1 to highlight the immediate effects of derivatives initiation on analyst coverage and to minimize data attrition.
4,700
4,800
4,900
5,000
5,100
5,200
5,300
-1 0 1
Num
ber
of a
naly
sts
cove
ring
New
Use
rs
Year relative to derivatives initiation (t)
51
Figure 2 Analyst coverage by analyst career experience
Note: This figure plots the number of analysts covering New Users relative to the year of derivatives initiation (t−1 to t+1) for a balanced sample of 549 New Users. We group analysts based on career experience, defined as the number of years (since 1980) that analyst j has issued forecasts for any firm in I/B/E/S. Analysts in the top quintile of career experience are designed as high experience (HIEXP), low experience (LOEXP) otherwise. We restrict the sample to t−1 to t+1 to highlight the immediate effects of derivatives initiation on analyst coverage and to minimize data attrition.
3,800
3,900
4,000
4,100
4,200
4,300
4,400
400
500
600
700
800
900
1,000
-1 0 1 Num
ber
of lo
w c
aree
r ex
peri
ence
ana
lyst
s
Num
ber
of h
igh
care
er e
xper
ienc
e an
alys
ts
Year relative to derivatives initiation (t)
HIEXP LOEXP
52
Figure 3 Analyst coverage by All-Star status
Note: This figure plots the number of analysts covering New Users relative to the year of derivatives initiation (t−1 to t+1) for a balanced sample of 549 New Users. We group analysts by All-Star (STAR) or non-All-Star (NOSTAR) status. All-Star analysts are selected annually by Institutional Investor Magazine based on solicited input from buy-side managers (i.e., chief investment officers of large money management institutions, directors of research, select analysts, and portfolio managers). We restrict the sample to t−1 to t+1 to highlight the immediate effects of derivatives initiation on analyst coverage and to minimize data attrition.
4,500
4,600
4,700
4,800
4,900
5,000
5,100
0
100
200
300
400
500
600
-1 0 1
Num
ber
of n
on-A
ll-S
tar
anal
ysts
Num
ber
of A
ll-S
tar
anal
ysts
Year relative to derivatives initiation (t)
STAR NOSTAR