Does Long-Term Earnings Guidance Mitigate Managerial Myopia?
Andrew C. Call
Arizona State University
Shuping Chen
University of Texas at Austin
Adam Esplin
University of Alberta
Bin Miao
National University of Singapore
October, 2014
Abstract: We address the policy debate on whether replacing short-term earnings guidance with
long-term earnings guidance reduces managerial myopia through reductions in accruals and real
earnings management and excess investment in fixed assets. We employ two event samples to
capture long-term guidance: a hand-collected sample of firms that issue earnings-guidance for
three to five years ahead and a sample that stops issuing quarterly guidance but continues to issue
annual guidance. Using a propensity-score matched design, we find no evidence that long-term
guidance firms manage earnings less or are more efficient in their investment decisions. Taken
together, our evidence is inconsistent with the view that long-term guidance mitigates managerial
myopia.
Preliminary. Please do not circulate. Comments welcome.
Corresponding author. We thank participants at 2014 BYU Research Conference, University of Texas at Austin
brownbag workshop, and Utah State University workshop for helpful comments. All errors are our own.
1
Does Long-Term Earnings Guidance Mitigate Managerial Myopia?
1. Introduction
Management frequently issues earnings guidance to communicate expectations of future
earnings to the firm’s stakeholders.1 This guidance is usually short-term in nature – forecasting
earnings for the upcoming quarter. However, in recent years multiple investor groups and
industry organizations argue that short-term guidance encourages managers to myopically focus
on short-term results at the expense of long-term performance and investments (e.g., CFA Center
for Financial Market Integrity and the Business Roundtable Institute for Corporate Ethics, 2006;
The Aspen Institute, 2007; the Committee for Economic Development, 2007; and the
Commission on the Regulation of U.S. Capital Markets in the 21st Century 2007). These
organizations encourage managers to cease giving short-term guidance and to instead issue long-
term earnings guidance. These sentiments are echoed by prominent academic researchers (Fuller
and Jensen 2010) and investors (e.g., Buffet 2000) alike.
We examine whether firms engage in less earnings management and make better
investment decisions, as evidenced in less excess investments, once they stop issuing short-term
earnings guidance and/or if they issue long-term guidance. This investigation is important as it
informs the policy debate on whether firms should replace short-term guidance with long-term
guidance: if critics of short-term guidance are correct in their assertion that long-term guidance
mitigates managerial myopia, replacing short-term guidance with long-term guidance should
yield less earnings management and less over- or under-investment.
To address our research question, we identify two samples of firms that issue long-term
earnings guidance from 2000 to 2012. The first sample consists of 854 firm-year observations
1 We use “management earnings forecasts”, “management earnings guidance”, and “management guidance”
interchangeably in this paper.
2
that issue long-term guidance of earnings three to five years ahead. We hand collect this sample
through key-word searches of multiple data sources and refer to this sample as the LTMF
sample. The second sample consists of firms that continue to issue annual earnings guidance
after discontinuing the issuance of quarterly earnings guidance. We identify 609 firm-year
observations using CIG/IBES guidance databases and call this sample our ANNMF sample.
These firms behave as if they are heeding the call from practitioners to provide long-term annual
guidance in lieu of quarterly guidance.2
Our empirical investigation proceeds in three steps. To tackle the self-selection inherent
in our setting, we first identify firm characteristics associated with the issuance of these long-
term guidance. Using propensity-score matched (PSM) control samples constructed from our
first-stage probit models, we then examine whether long-term guidance firms exhibit less
accruals and real earnings management than their respective control firms, and whether long-
term guidance firms exhibit less excess investment in fixed assets.
We find that the LTMF sample – firms that issue earnings guidance for 3-5 years ahead –
exhibits better stock and accounting performance, lower return volatility, a greater number of
long-term forecasts issued by analysts, and longer investment horizon by institutional investors
when compared to firms that do not issue this form of guidance. LTMF firms are also more
likely to issue dividends. In contrast, firms that continue to issue annual guidance after stopping
quarterly guidance (ANNMF firms) have experienced deteriorating performance, increasing
uncertainty in their operating and information environment, a decrease in institutional investors’
2 To more closely mirror critics’ recommendations, an ideal sample should consist of firms that issue only quarterly
guidance but move to issuing only annual guidance or even longer-term guidance. However, over our sampling
period, only 515 firms (5501 firm years) appear to issue only quarterly guidance, and of these firms, 35 firms (41
firm years) move from issuing only quarterly guidance to issuing only annual guidance. This is before requiring
these observations to have the requisite data. This small sample makes it difficult to implement a meaningful
research design.
3
holding horizons and fewer long-term growth forecasts issued by analysts when compared to
firms that continue to issue quarterly earnings guidance. These findings are largely consistent
with prior research investigating why firms cease guidance (Chen, Matsumoto, and Rajgopal
2011).
Based on these determinants of long-term guidance, we construct propensity score
matched (PSM) control samples for LTMF and ANNMF samples respectively. Our second step
of empirical analysis compares the event samples with their PSM control samples on accruals
and real earnings management. We use both signed and unsigned discretionary accruals derived
from the Jones model (1989) and the Dichow-Dichev model (2002) adjusted for economic
performance (Ball and Shivakumar 2006) and discretionary revenue (Stubben 2010) to capture
accruals earnings management. We find no evidence that our two event samples exhibit
differential accruals earnings management from their respective propensity-score matched
control samples in either a univariate or a multivariate setting. Nor do we find evidence of
differential real earnings management, proxied by abnormal operating cash flows, abnormal
discretionary expenses, and abnormal production costs based on Roychowdury (2006). Lastly,
using the McNichols and Stubben (2008) model of excess investment in fixed assets, we find no
evidence of improved investment decisions for the two event samples.
Thus, across six different measures of accruals earnings management (signed and
unsigned), three different measures of real earnings management, and two different measures of
excess investment (signed and unsigned), and using two different event samples to capture the
concept of long-term guidance, we do not find evidence of differential earnings management or
excess investment between long-term guidance and other firms. We provide no support for the
4
belief held by many practitioners that long-term guidance reduces managerial focus on short-
term results and improves investment decisions.
The evidence in this study informs the current debate on the merits of long-term guidance
as a tool to reduce managerial fixation on short-term performance and to facilitate improved
investment decisions. While some prominent practitioners argue that the issuance of long-term
guidance mitigates managerial myopia, our results do not support this argument.
We contribute to the growing literature motivated by the debate over the costs and
benefits of short-term earnings guidance, and more broadly the costs and benefits of guidance in
general. Chen, Matsumoto, and Rajgopal (2011) and Houston, Lev, and Tucker (2010) find that
when firms cease earnings guidance, analyst earnings forecast accuracy declines and forecast
dispersion increases, impairing the quality of information available to investors.3 More recently,
Call, Chen, Miao, and Tong (2014) find that firms issuing short-term quarterly guidance actually
exhibit less, not more earnings management, and Chen, Huang, and Lao (2014) find that
guidance firms report more future innovations than non-guidance firms, and that quarterly
guidance has a positive incremental impact on future innovation over annual guidance. We
extend this literature by evaluating the potential impact of one of the highly touted alternatives to
short-term guidance, and find that long-term guidance has no discernable effect on earnings
management activity or on investment decisions.4
3 We note our research question is different from the research question investigated in Chen et al. (2011) and
Houston et al. (2010): the prior two studies investigate the consequence of stopping guidance (both quarterly and
annual earnings guidance), whereas our focus is on firms that (continue to) issue long-term guidance. Their event
samples are firms that stop guidance, and our event samples are firms that continue to issue annual guidance/issue
longer-term guidance. 4 Lao (2013) utilizes the Ohlson model and finds that investors place significantly higher weight on short-term
earnings of quarterly guidance firms than on the corresponding earnings of non-guidance firms. We note, however,
that Lao (2013) focuses on investor short-termism whereas our paper focuses on managerial short-termism.
5
Lastly, the issuance of long-term guidance of earnings three to five years ahead is itself
an interesting disclosure choice that has received little attention in the academic literature. We
provide the first evidence on such long-term guidance and the characteristics of firms issuing
such guidance. Our evidence adds to the academic literature on firms’ voluntary disclosure
behavior.
We note one important caveat in interpreting our results: our two event samples, LTMF
and ANNMF, are both small. This is because in practice only a small number of firms issue
longer-term 3~5 year earnings guidance or stop quarterly guidance but continues with annual
guidance. Thus, the null results in this paper could be due to lack of power. We do not preclude
the possibility that once more firms start replacing short-term guidance with longer term
guidance, a researcher may be able to document evidence that long-term guidance indeed
mitigates managerial myopia.
The rest of the paper is organized as follows. Section two reviews relevant literature and
develops our empirical predictions. Section three describes our long-term guidance samples and
selection of our control samples. Section four presents the research designs and discusses the
respective test results. Section five concludes.
2. Background and empirical predictions
In recent years, influential practitioners and academics have been critical of the practice
of giving short-term earnings guidance. Critics allege that short-term earnings guidance fosters
managers’ myopic behavior such as earnings management and encourages fixation on short-term
earnings performance to the detriment of long-term performance and investments (Fuller and
Jensen 2010). Even regulators and politicians weigh in on this debate. Former SEC Chairman
6
William H. Donaldson cited “short-termism” as a critical issue facing corporate leaders at the
2005 CFA Institute annual conference and called upon business leaders “[to] manage business
for long-term results” (CFA Center for Financial Market Integrity 2006), and Al Gore, former
U.S. vice president, called for companies to end “… this default practice… and encourage a
long-term view of the business rather than the current focus on quarterly results” (Gore and
Blood 2012).
Multiple industry organizations similarly call for firms to issue long-term guidance in
place of short-term guidance (CFA Center for Financial Market Integrity and the Business
Roundtable Institute for Corporate Ethics, 2006; The Aspen Institute, 2008; the Committee for
Economic Development, 2007; and the Commission on the Regulation of U.S. Capital Markets
in the 21st Century 2007). Beginning in September 2005, the CFA Center for Financial Market
Integrity and the Business Roundtable Institute for Corporate Ethics co-sponsored a “Symposium
Series on Short-Termism.” One of their key recommendations was to encourage “corporate
leaders, asset managers, institutional investors, and analysts” to “[e]nd the practice of providing
quarterly guidance” and to “adopt guidance practices … that reflect overall long-term goals and
strategies.”5 Together with these two organizations, the U.S. Chamber of Commerce Center for
Capital Market Competitiveness and Committee for Economic Development recommended that,
“companies focus their communications more on their long-term strategic plans, thereby leading
and encouraging investors to do the same” (The Aspen Institute 2008).6 Another example comes
from the “Reports and Recommendations” by the Commission on the Regulation of U.S. Capital
Markets in the 21st Century (March 2007):
5 “Breaking the Short-Term Cycle: Discussions and Recommendations on How Corporate Leaders, Asset Managers,
Investors, and Analysts Can Refocus on Long-Term Value.” CFA Center for Financial Market Integrity/Business
Roundtable Institute for Corporate Ethics, 2006. 6 “Operating and Investing for the Long-Term: Best Practices in Communications, Guidance and Incentive
Structures to Create Value for the Long-Term”, the Aspen Institute 2008.
7
“All public companies should eliminate the practice of providing quarterly earnings
guidance and that companies should instead provide shareholders and Wall Street with
meaningful additional information on their long-term business strategies.”
The underlying assumption to these calls for long-term guidance is that a switch from
short-term to long-term guidance would mitigate managers’ focus on short-term results. For
example, a recent report by McKinsey & Company argues that long-term guidance can “shift
their [managers’] focus away from short-term performance and toward the drivers of long-term
company health as well as … their long-term goals.”7
We empirically investigate the assumption that long-term earnings guidance reduces
managerial myopia. Specifically, we investigate whether managers engage in less earnings
management after issuing long-term guidance. We also examine whether the investment
efficiency improves for firms that issue long-term guidance. If the assumption by investor and
industry groups is valid, we expect managers of firms that issue long-term guidance to exhibit
less earnings management and to make better investment decisions. We summarize these as
testable empirical predictions below:
P1: Long-term earnings guidance mitigates earnings management.
P2: Long-term earnings guidance improves investment decisions.
Most recommendations from investor and practitioner groups do not explicitly define the
horizon of long-term guidance. However, most criticisms of short-term guidance are focused on
the issuance of quarterly guidance. We therefore define long-term earnings guidance
alternatively as (a) management forecasts of earnings with a horizon of three to five years ahead,
or (b) management guidance for annual earnings. These empirical definitions are in line with
practitioners’ implicit proposals. For example, a report issued by the Committee for Economic
Development states: “Medium-term indicators might point toward the likelihood the company
7 “The Misguided Practice of Earnings Guidance”, McKinsey & Company publications, March 2006.
8
could maintain performance over one to five years.” (“Built to Last”, 2006). The first definition
also coincides with the definition I/B/E/S employs for long-term analysts’ growth forecasts.8
We recognize that firms issuing long-term earnings guidance may differ systematically
from firms that do not. Since these differentiating factors may also be associated with incentives
to manage earnings or investment decisions, we compare firms that issue long-term guidance to
propensity score matched control samples that exhibit similar characteristics but that do not
exhibit the same guidance behavior. We discuss our sampling process below.
3. Long-Term Guidance Samples and Control Samples for Determinant Tests
An ideal sample to address the policy debate is a sample of firms that truly replace short-
term (such as quarterly) guidance with long-term (such as annual) guidance. In other words, an
ideal sample consists of firms that were issuing only quarterly guidance before and start issuing
only annual or even longer term guidance. However, such a sample is extremely small: using
CIG/IBES guidance data base, we find that only 35 firms (41 firm-years) satisfy the sampling
criteria of issuing only annual after issuing only quarterly forecasts over the period of 2002-
2013.
Thus, we identify two samples of firms that issue long-term earnings guidance that most
closely mirror the call from practitioners which also have reasonable sample sizes for our
empirical analysis. The first sample, the LTMF sample, consists of firms that issue guidance of
earnings for three to five years ahead. We use a series of keyword searches for mentions of long-
8 The Thomson Reuters (I/B/E/S) (2009) Methodology for Estimates manual states “The long term growth rate
represents an expected annual increase in operating earnings over the company’s next full business cycle. These
forecasts refer to a period between three and five years and are expressed as a percentage.”
9
term guidance in the business press over the period January 1, 2000 to December 31, 2012.9 Our
search for long-term forecasts encompasses the Dow-Jones News Service, the Wall Street
Journal, PR News Wire, Business News Wire, and conference call transcripts via the Fair
Disclosure database. We read over 8,000 articles retrieved using the key word search to identify
our sample of long-term earnings guidance. This process yields a total of 275 unique firms and
854 firm-year observations with long-term forecasts of earnings that are three-to-five years ahead
from 2000 to 2012. The control sample pool for the LTMF event sample consists of all
CIG/IBES guidance firms that do not issue 3~5 year earnings guidance. Thus, the LTMF event
firms issue 3~5 year guidance whereas the control firms have never issued any 3~5 year earnings
guidance.10
Our second sample, the ANNMF sample, consists of firms that stop issuing quarterly
guidance while continue to issue annual guidance. We use the CIG/IBES databases to identify
these firms from 2000 to 2013. Specifically, we set an indicator variable, ANNMF, equal to one
for firms that issue at least one annual forecast without issuing any quarterly guidance for four
consecutive quarters (post period), after having issued at least 3 quarterly forecasts in the
9 The search terms we use include variations of the following string to accommodate different databases:
“(management or manager or CEO or chief executive* or CFO or chief finance* or company or firm) and
((anticipates or expect* or predict* or forecast* or see* or project* or put* or estimate) near10 (five year near3
earn*) or (three year near3 earn*) or (long term near3 earn*) or (five year near3 eps) or (three year near3 eps) or
(long term near3 eps))”. We start our search in 2000 to mitigate errors in the measurement of earnings guidance that
are not captured by the Company Issued Guidance (CIG) database prior to Regulation FD. We further augment our
sample by identifying 78 long-term forecasts captured by the CIG database and 6 long-term forecasts identified by
I/B/E/S. To ensure the completeness of our search, one author and one research assistant conducted independent
searches using various combinations of the search strings using the same data sources. 10
Again, a more ideal LTMF sample should be firms that are not subject to short-term guidance pressure, namely,
LTMF firms that do not issue quarterly earnings guidance. Imposing this data restriction leads to sample attrition of
more than 56%, resulting in a sample size of 371 observations. Given that our data requirement to perform
subsequent tests on earnings management and excess investments will lead to further data attrition, and Type II error
is a challenge in our setting, we choose to start with the larger sample of 854 observations.
10
previous four quarters (pre period).11
Figure 1 presents the timeline used in this design. These
firms are heeding practitioners’ calls to favor long-term guidance and to discontinue the practice
of issuing quarterly forecasts. This sampling procedure leads to 370 unique firms and 609 firm-
year observations from 2001 to 2012 that stopped giving quarterly earnings guidance but
continue to issue annual guidance. The control sample pool for the ANNMF event sample
consists of CIG/IBES guidance firms that have issued at least 6 quarterly forecasts in two
consecutive 8-quarter period, namely at least three quarterly forecasts in each four-quarter
interval. In other words, the ANNMF event sample issues annual guidance but does not issue
quarterly guidance in the post period, whereas the control sample issues quarterly guidance in
both the pre and post period.
Note we exclude from our control samples firms that do not appear on CIG/IBES
guidance database and therefore have not issued any guidance for two reasons. First, the debate
that motivates our study is whether long-term guidance mitigates the short-termism associated
with managers’ short-term guidance. As such, firms that issue no guidance are less relevant to
this debate. Second, a large body of literature has documented that firms issuing earnings
guidance differ systematically from those not issuing guidance (see Beyer, Cohen, Lys and
Walther 2010 for a literature review). Limiting the analysis to firms that have issued guidance
holds these differences constant and allows us to draw cleaner inferences.
Panel A of Table 1 presents descriptive statistics on the frequency of long-term guidance,
before imposing data restrictions. The issuance of 3~5 year earnings guidance (LTMF=1) is
relatively consistent through our sample period, with a peak of 81 in 2007 and a low of 34 in
11
We use the CIG database to identify the ANNMF observations for years 2000-2002 as the Thomson Reuter’s
IBES guidance database does not start its coverage of management issued guidance until 2003. For 2003-2013, we
use the combination of CIG and IBES databases to identify ANNMF observations for years 2003-2013.
11
2000.12
The incidence of firms stopping quarterly guidance while issuing annual guidance
(ANNMF=1) reaches a peak in 2006, and has tapered off in the years since.13
It is important to note that, overall, the number of firms issuing LTMF – 275 firms - and
the number of firms stopping quarterly but continuing with annual guidance (ANNMF firms) –
370 unique firms - are small compared to the total number of firms that appear in CIG/IBES
guidance database. In addition, 154 out of the 370 firms in ANNMF sample restart issuing
quarterly earnings guidance after a four-quarter period of issuing no quarterly guidance. Thus,
despite the fact that many prominent investors and important industry groups urge firms to
replace short-term guidance with longer term guidance, in practice not so many firms have
heeded to such call for longer term guidance or have done so consistently.
Panel B of Table 1 shows some industry clustering of long-term guidance, with the
LTMF observations being clustered in the Shops, Consumer Non-Durables, and Money
industries, and the ANNMF observations being clustered in the Shops and Business Equipment
industries. In our subsequent tests investigating the impact of long-term guidance, we require
that the PSM control observations to come from the same year and industry of the event
observations, and we also include industry and year fixed effects in our empirical models
investigating earnings management and investments in fixed assets.
4. Research Design and Empirical Results
4.1 What determines long-term earnings guidance?
4.1.1 Research design
12
We report only 10 LTMF observations in 2013 because we ended our search in December 2012, and some of the
firms that issued long-term guidance in 2012 have a 2013 fiscal year-end. 13
Note even though our sampling period for ANNMF observations starts in 2000 and ends in 2013, our ANNMF
starts in 2001 and ends in 2012 because we require 8 consecutive of quarters to identify these observations.
12
Our first objective is to identify factors that differentiate firms issuing long-term 3~5 year
earnings guidance (LTMF sample) from other guidance firms, and factors that lead firms to stop
issuing quarterly guidance while continuing to issue annual guidance (ANNMF sample). We
compare LTMF firms to guidance firms that do not issue 3~5 year guidance (LTMF=0), and
ANNMF firms to those that continue to issue quarterly guidance (ANNMF=0), as described in
the previous section. This comparison is important as the factors affecting firms’ voluntary
disclosure behavior can conceivably also impact firms’ earnings management behavior and
investment decisions. Addressing this question mitigates the endogeneity concern that self-
selection contributes to the observed earnings management or investment behavior. This
investigation allows us to construct propensity score matched control samples for each of our
event samples in our subsequent investigation of the impact of these practices on managerial
myopia.
The issuance of long-term 3~5 year earnings forecasts (LTMF) is a distinctly different
phenomenon from the much more prevalent practice of issuing quarterly or annual earnings
guidance. It is conceivable that firms can only issue longer-term forecasts when they are better
able to forecast the future with confidence. This reasoning and our review of the disclosure
literature (e.g., Healy and Palepu 2001; Beyer, Collins, Lys, and Walther 2010) lead to the
following conjectures: firms are more likely to issue LTMF if they (1) have solid performance
(2) have more stable operating and information environment, (3) face greater demand for long-
term information from long-term investors. We also conjecture that firms (4) undergoing
restructuring and mergers and acquisitions are less able to issue LTMF, as such changes make it
more difficult for managers to predict long term. On the other hand, (5) managers who are
historically better at forecasting are in a better position to give LTMF. Finally, our reading of the
13
press releases to identify LTMFs shows that (6) firms issuing dividends are more likely to
provide longer term information, perhaps to assure investors of the sustainability of dividends.
We use stock returns in the previous twelve months (RET-12) and the return on assets in
the prior year (ROA) to proxy for performance, daily stock return volatility (STDRET) for the prior
year and the number of long-term analyst forecasts (LTAF) in year t-1 to proxy for operating and
information environment uncertainty, respectively, and the negative of the average turnover rate
of institutional investors (CHURN) in year t-1 to proxy for demand for information from long-
term investors.14
We use an indicator variable equal to one if the firm either had a restructuring
event or a merger in year t-1 (ResMA), and capture managers’ ability to forecast earnings using a
measure of the precision of prior management earnings guidance (PRECISE), where point
estimates are considered more precise than range estimates, and range estimates are considered
more precise than qualitative forecasts. We capture dividend issuance using an indicator variable
DIV coded as one for firms issuing at least two quarterly dividends in year t-1. Detailed
definitions of all proxies are tabulated in the Appendix. We estimate the following probit
regression for our LTMF sample:
Prob (LTMF=1) it = 0 + 1RET it-1 + 2ROA it-1 + 3STDRET it-1 + 4LTAF it-1
+ 5CHURN it-1 + 6ResMA it-1 + 7PRECISE it-1 +8DIV it-1 it (1a)
Prior research has investigated why firms stop giving quarterly earnings guidance. For
example, Chen et al. (2011) and Houston et al. (2010) find that firms stop when performance
declines and uncertainty increases. Though these studies focus on the stopping decision rather
than the continuation of annual guidance, they nevertheless provide a starting point for our first
14
Specifically, we use the average turnover rate of institutional investors (CHURN) following Gaspar, Massa and
Matos (2005) to capture investment horizon. Higher CHURN rate indicates lower investment horizon. For ease of
interpretation we multiple CHURN by negative one, thus we can interpret high CHURN to indicate longer
investment horizon.
14
investigation.15
Thus, we conjecture that, compared to control firms that continue to issue
quarterly guidance, ANNMF firms experience decreasing stock and accounting performance and
increasing uncertainty. We use buy-and-hold return for the past 12 months (RET-12) and the
change in return on assets from year t-1 to year t (ROA) to proxy for performance change, and
year-over-year change in daily stock return volatility (STDRET) to capture increases in
uncertainty in the operating environment and the change in the number of IBS analysts issuing
long-term 3~5 year earnings forecasts (LTAF) to capture changes in uncertainty in the
information environment. With greater information uncertainty analysts are less able to forecast
far into the future. Prior research also finds some evidence that firms stop quarterly guidance
when facing investors with longer holding horizons, thus we include CHURN to capture this
change in demand for information. Positive CHURN indicates increases in investment horizon
of institutional investors and negative CHURN indicates decreases in investment horizons.
Detailed definitions for all variables are offered in the Appendix. We estimate the following
probit model:
Prob(ANNMF=1)it = 0 + 1RETit-1 + 2ROA it + 3STDRET it + 4LTAF it
+ 5CHURN it + it (1b)16
The changes specification captures the changes in disclosure behavior of ANNMF firms:
these firms change from giving quarterly and annual guidance to only giving annual guidance.
15
We first replicate the Chen et al. (2011) results using their sampling period. Note the Chen et al. (2011) sample is
substantially different from our sample, as their research focus is on the stopping of quarterly guidance, not the
continuation of annual guidance. Thus, their event sample, the “stopper” sample, consists of firms that stop giving
both annual and quarterly earnings guidance – these firms stop appearing in the CIG database in four adjacent
quarters after having appeared at least three quarters in the previous adjacent quarters. In addition, their sampling
period is 2002 to 2003, whereas our sampling period is much longer covering 2001-2012. Houston et al. (2010) has
similar sampling procedure as Chen et al. (2011) and their sample period is from Q12002 to Q12005, substantially
shorter than our sampling period. 16
We employ this parsimonious model in order to maximize matching on each of the individual factors and also to
avoid further data attrition, while noting that results do not change if we augment Equation (1a) with more variables
such as analyst forecast dispersion or replacing ROA with a measure based on the frequency of firms meeting or
beating analyst forecasts, which is more data-demanding.
15
We note that because the ANNMF sample is characterized by firms that change their
guidance behavior, as such we employ a changes model (1b) to generate propensity scores, while
we use a levels model in equation (1a) in an effort to differentiate firms that either issue or do not
issue long-term guidance of 3~5 year ahead earnings.17
4.1.2 Results of determinants tests
In Panel A of Table 2 we report univariate statistics for the independent variables
included in our prediction models, separately for the two event samples and their respective
control samples used in the probit regressions. Tests of difference in means and medians reveal
that LTMF firms have higher returns and ROA, lower return volatility, and more analysts issuing
long-term forecasts, are twice as likely to grant dividends and have institutional investors with
longer horizons. However, inconsistent with our predictions, these firms have more incidences of
restructuring and M&A activities and lower prior management forecast precision at the
univariate level. In contrast, ANNMF firms have experienced decreasing returns and ROA,
increasing return volatility and bigger drops in the number of analyst long-term forecasts.
ANNMF firms also exhibit an increase in institutional investors’ horizon, though the increase is
smaller than that of control firms.
Panel B of Table 2 presents our probit estimation results. The results are largely
consistent with the univariate results, with the exception that ResMA and PRECISE are not
significant in the LTMF regression. In addition, contrary to our prediction and prior research, the
coefficient on CHURN is negative, suggesting ANNMF firms have experienced a relative
decrease in investor holding horizons. It is possible that these firms stop issuing quarterly but
17
We do not include year or industry fixed effects in equations (1a) or (1b), as in our subsequent propensity score
matching procedure we require the matched firm to come from the same year and industry as the event firm.
16
continue to issue annual guidance in an effort to win back more long-term investors, instead of
responding to the demand of current investors.18
In Panel C of Table 2 we tabulate the companion of the matching variables after we
generate one-one-one matches for each observation in our two event samples. The matched
observations are from the same year and industry and have the closest propensity scores to the
event observations. Panel C shows that our matches are largely successful, with the exception of
LTAF for the LTMF matched sample and RET-12 for the ANNMF matched sample.
4.2 Do managers engage in less earnings management after the issuance of long-term guidance?
4.2.1 Research design
To investigate the potential impact of long-term guidance on managerial myopia, we
examine whether the long-term guidance firms differ in the extent to which they engage in
accrual-based and real earnings management. We compare the firms in our LTMF and ANNMF
samples to PSM control firms based on equations (1a) and (1b) generated above.
We use three proxies to capture the extent of accrual-based earnings management:
abnormal accruals from the Jones model (1991) (ABAC) and from the Dechow-Dichev model
(2002) (ABDD), both modified per Ball and Shivakumar (2006), and abnormal revenues based
on Stubben (2006, 2010) (ABREV). Detailed definitions of these variables are offered in the
Appendix.
We estimate the following regression:
EMit = α + β1ANNMF/LTMFit-1 + β2LEVit-1 + β3BTMit-1 + β4OPCYCLEit-1 + β5CAPINTit-1 + β6ROAi
+ β7SIZEit-1 + β8INSTit-1 + β9σ(CFO)it-1 + β10σ(EARN)it-1 + + ΣIND + ΣYEAR + εit (2a)
EM is one of the three abnormal accruals proxies (ABAC, ABDD, or ABREV) discussed above
18
Note alternative measures of accounting performance (the frequency of meeting or beating analyst forecasts) and
uncertainty (analyst forecast dispersion) yield the same results.
17
and outlined in the Appendix. Larger absolute values of these measures indicate more earnings
management. LTMF/ANNMF is an indicator variable set to one for LTMF/ANNMF =1
observations, and set to zero for PSM control observations.
Our control variables in Eq. (2a) are drawn from Call et al. (2014) which investigates the
impact of quarterly guidance issuance and frequency on the extent of accruals earnings
management. We include the ratio of debt to equity (LEV) to control for the effects of leverage
on earnings management (DeFond and Jiambalvo 1994; Barton and Waymire 2004). We control
for growth using the book-to-market ratio (BTM) as Skinner and Sloan (2002) find growth firms
have stronger incentives to manage earnings because the market penalizes growth firms for
negative earnings surprises. We include several variables to control for firms’ underlying
business fundamentals, such as the length of the operating cycle (OPCYCLE) and capital
intensity (CAPINT), as both have been shown to affect reported accruals (Dechow and Dichev
2002; Cohen 2008). We also control for firm performance (ROA), firm size using the natural log
of sales (SIZE), and percentage of institutional ownership (INST). We address the concerns of
correlated omitted variables raised in Hribar and Nichols (2007) by including both the standard
deviation of operating cash flows ((CFO)) and the standard deviation of earnings ((EARN)) to
control for the volatility of the firm’s operating environment. Finally, we include industry
dummies (IND) in Eq. (2a) based on Fama and French’s 12 industry groupings. We also include
year dummies (YEAR) to mitigate concerns over cross-sectional dependence (i.e., earnings
management for firm i is correlated with earnings management of firm j in a given year). We
cluster our standard errors by firm to address time-series in the residuals (Petersen 2009).
Detailed definitions of all variables are provided in the Appendix.19
19
While the extant literature has relied on a variety of empirical measures of earnings management (e.g., meeting-
or-beating benchmarks, earnings restatements, accounting frauds, etc.), we believe abnormal accruals and
18
To capture real earnings management, we employ the three proxies advanced in
Roychowdury (2006): abnormal operating cash flows (DISCFO), abnormal discretionary
expenses (DISEXP), and abnormal production costs (DISPROD). Lower values of DISCFO and
DISEXP and higher value of DISPROD indicate greater extent of real earnings management
(Roychowdury 2006). Lower DISCFO and DISEXP can stem from managers’ effort to increase
earnings through excessive price discount to generate revenue and excessive cutting of
discretionary expenses to reduce expense. Higher DISPROD can result from overproduction in
order to lower COGS. Detailed estimations of these measures are provided in the Appendix. We
estimate the following regression to investigate whether firms issuing long-term forecasts are
engaged in less real activities management than their PSM control firms:
REALEMit = α + β1ANNMF/LTMFit-1 + β2PMBAFit-1 + β3STK_ISSUEit-1 + β4LOGNAFit-1
+ β5BTMit-1 + β6LOGSHROUTit-1 + β7LOGMVit-1 + β8ROAit-1 + ΣIND + ΣYEAR + εit
(2b)
REALEM is one of the above three proxies (DISCFO, DISEXP, or DISPROD) for earnings
management through real activities. Our control variables are drawn from Cohen and Zarowin
(2010) and Zang (2012). PMBAF is firms’ percentage of meeting or beating consensus analyst
forecasts in the past 8 quarters. Habitual beaters of market expectations have a greater likelihood
of achieving this through real activities manipulations. STK_ISSUE is an indicator variable for
equity issuance as prior research finds that firms issuing stock are more likely resort to real
manipulations of earnings. Firms with more analyst following (LOGNAF) and growth firms
(BTM) are under greater pressure to manage earnings, and a greater number of shares
outstanding (LOGSHROUT) requires more earnings management to achieve a given earnings per
discretionary revenues best capture managers’ use of accounting discretion to manage earnings. In particular, critics
are concerned with “accounting shenanigans”. Since these concerns are focused on managerial discretion over
earnings, rather than on outcomes that potentially follow (e.g., meeting-or-beating benchmarks, restatements, fraud),
we assess accruals earnings management by measuring the extent of managerial intervention in the earnings process.
19
share target. We also include controls for firm size (LOGMV) and performance (ROA), as well as
industry and year fixed effects in Eq. (2b).
4.2.2 Accrual-based earnings management test results
We report the results of our accrual-based earnings management tests in Table 3
(absolute values of accruals) and Table 4 (signed accruals).
Univariate statistics in Panel A of Table 3 reveal no significant difference in accrual-
based earnings management proxies between LTMF firms and their PSM control firms.
Similarly, none of the three accruals earnings management proxies are different between
ANNMF firms and their PSM controls. Our multivariate regressions reported in Panel B reveal
similar results. Ceteris paribus, there is no significant difference in accrual-based earnings
management between our two event samples and their respective PSM control samples, using
any one of our three proxies (ABAC, ABDD, or ABREV).
The results on ANNMF event sample in Panel B is based on a cross-sectional design
comparing the magnitude of accruals after the event quarter 1. Since ANNMF firms have
changed their guidance behavior, in Panel C of Table 3 we employ a difference-in-difference
design to further examine if ANNMF firms differ from control firms in the extent of their
changes in accruals earnings management. We define a dummy variable POST coded as one for
the one year after cessation of quarterly guidance and zero for the year before, and interact POST
with ANNMF. The results in Panel C show that there is no difference in the extent of accruals
earnings management in the year before between event and PSM control samples. In the POST
period, ABDD for event firms is marginally higher than control firms (firm-clustered t=1.82),
whereas neither ABAC no ABREV are different between event and control firms.
20
The results in Table 4, based on signed accruals, yield similar inference. At the univariate
level LTMF firms exhibit higher ABDD but lower discretionary revenues ABREV (Panel A), and
in the regression in Panel B these differences persist. The opposite signs on the two accruals
earnings management measures, however, make it difficult to conclude that LTMF firms manage
earnings less than PSM controls. The signed accruals results based on ANNMF sample show that
ANNMF firms exhibit lower discretionary accruals (3 out of 6 differences are significant, Panel
A), however the difference disappears once we control for other factors that can affect the
dependent variable (Panel B), though ANNMF firms show slightly lower ABAC in the DiD
design (significant at 10% level) in Panel C.
Taken together, the results in Table 3 and Table 4 fail to yield consistent evidence that
our two long-term guidance event samples differ in discretionary accruals and discretionary
revenues from their respectively PSM control samples. We interpret our results as inconsistent
with the assumption that replacing short-term guidance with long-term guidance mitigates
accrual-based earnings management.
4.2.3 Real earnings management test results
We report the results of our real earnings management tests in Table 5. In the univariate
results presented in Panel A, we find that LTMF firms exhibit lower abnormal discretionary
expenses than their PSM controls at a univariate level, while neither DISCFO nor DISPROD are
different between the two samples. The ANNMF firms exhibit lower abnormal cash flows
(DISCFO) but higher abnormal discretionary expense (DISEXP) than their PSM control sample.
Note that real earnings management would indicate higher abnormal cash flows and lower
abnormal discretionary expense, if firms are trying to manage earnings upward to meet forecasts
as alleged by critics.
21
The results in a multivariate setting (Panel B) when controlling for other determinants of
real earnings management show that, ceteris paribus, there is no significant difference in real
earnings management (DISCFO, DISEXP, or DISPROD) proxies between the LTMF sample and
the PSM control sample. In addition, ANNMF firms exhibit higher, not lower, abnormal
discretionary expenses. This result is opposite to firms increasing earnings through aggressive
price discount or cutting discretionary expenses such as R&D.
Similar to our investigation of the accruals earnings management behavior for ANNMF
firms, we further employ a difference-in-difference design to examine the possibility that
ANNMF firms exhibit greater decrease in the extent of real earnings management than control
firms. We report these estimation results in Panel C of Table 5. The dummy variable POST is
defined the same as in Panel C of Tables 3 and 4. The results show no difference between
ANNMF observations and controls in the extent of real earnings management either before or
after the event quarter.
Taken together, these findings are inconsistent with practitioners’ assumption that long-
term guidance mitigates managerial myopia by limiting real earnings management.
4.3. Do managers make more efficient investment decisions following long-term guidance
issuance?
4.3.1 Research design
To investigate whether firms make more efficient investment decisions following the
issuance of long-term guidance, we use the models developed in McNichols and Stubben (2008)
to capture excess investment. We calculate excess investment (|XINVT|) for a seven-year period
centered on the event year (Year 0), which is defined as the year when the firm stops giving
quarterly but continues with annual guidance (ANNMF sample), or the year the firm issues 3~5
22
year ahead earnings guidance (LTMF sample). McNichols and Stubben (2008) propose two
models, a short model and a long model, to derive expected investment, and excess investment is
the residuals from these models.
The McNichols and Stubben (2008) models of expected investment are as follows:
Short Model: INVTit = a + β1Qit-1 + β2CFit-1 + eit, (3a)
Long Model: INVTit = a + β1Qit-1 + β2Q_QRT2it-1 + β3Q_QRT3it-1 +β4Q_QRT4it-1
+β5CFit-1 +β6GROWTHit-1 + β7INVTit-1 + eit, (3b)
INVT is investment in capital expenditure taken from the cash flow statements. Detailed variable
definitions are offered in the Appendix. We estimate the above two models cross-sectionally
each year by industry, with at least 20 observations for each industry-year. We interpret the
absolute values of the residuals as excess investment. We then compare the excess investment
thus derived across both the event (ANNMF or LTMF) and the PSM control samples for the
seven-year window. We also compare the signed values of XINT to gauge the extent of over- or
under-investment.
We focus our discussion of the results using the long model, and note that, with a few
exceptions, the conclusions are largely the same when using the short model. In untabulated
analysis we also compare the simple industry-adjusted means of |XINVT| for the event and
control samples and we note that the inference remains the same.
The focus of our attention is the years beginning with Year 0. For the LTMF sample year
0 is defined as the first year in the sampling period when the firm issues LTMF. For the ANNMF
sample year 0 is defined as the year the firm stops issuing quarterly guidance. Recall we require
the PSM control observations to come from the same industry and year in the matching process,
thus all the control observations have a “pseudo” event year year 0 too.
As reported in Table 6, we find limited evidence of differential subsequent excess
investment between firms that issue long-term guidance and matched control firms. For the
23
LTMF sample comparison in Panel A, we only find a significant difference in Year 2 and a very
marginal difference in year 0, when LTMF firms exhibit less excess investments than their PSM
control firms. For the ANNMF sample comparison in Panel B, we only find a significant
difference in Year 3, where ANNMF exhibit less excess investment than their PSM control
sample.
Table 7 tabulates the results based on signed values of XINT, thus we can interpret
positive values as over investment and negative values as under investment. Using the long
model, we find no difference in the extent of under- or over-investment between our two event
samples and their respective controls in any of the 7 years presented.
Collectively, while there is some evidence of lower excess investment for the LTMF
firms in Panel A of Table 6, the findings are marginal. Thus, we conclude that our evidence does
not support the view that long-term guidance firms generate less under or over- investment.
5. Conclusion
Management guidance, in particular shorter-term earnings guidance, has come under
considerable scrutiny from many investor and industry organizations in recent years (e.g., CFA
Center for Financial Market Integrity and the Business Roundtable Institute for Corporate Ethics,
2006; The Aspen Institute, 2007; the Committee for Economic Development, 2007; and the
Commission on the Regulation of U.S. Capital Markets in the 21st Century 2007). The primary
criticism of earnings guidance practice is that short-term guidance fosters managerial myopia and
negatively impacts long-term value creation. These organizations frequently encourage managers
to cease giving short-term guidance and to instead issue long-term earnings guidance. An
24
important untested belief of these advocacies is that long-term guidance can mitigate managerial
myopia.
Using a propensity score matched design, we test whether this underlying assumption
holds by examining the impact of long-term guidance on 1) the extent of accruals earnings
management and real earnings management, and on 2) the extent of excess investment.
As there is no clear definition of what constitutes long-term guidance, we construct two
samples that most closely approximate the concept of long-term guidance in practice: the first
sample of firms, the LTMF sample, has issued guidance for earnings 3~5 years ahead. We hand
collect this sample using key-word searchers over multiple databases over 2000-2013 and
identify 275 unique firms and 854 firm-year observations. The second sample of firms, the
ANNMF sample, ceases giving quarterly guidance but continues to issue annual guidance. We
obtain this sample using machine-readable data from CIG/IBES guidance databases and identify
370 firms with 609 firm-year observations from 2001-2012. Note that despite repeated calls
from practitioner groups and prominent academics for firms to replace short-term guidance with
long-term guidance, the number of firms doing so over a 12-year period is small compared to the
population of firms issuing guidance (3,279 unique firms).
To investigate the impact of long-term guidance issuance on earnings management and
investment decisions, we first estimate probit regressions to capture the determinants of long-
term guidance by comparing 1) LTMF firms with CIG/IBES guidance firms that do not issue
3~5 year ahead earnings guidance, and 2) ANNMF firms with CIG/IBES guidance firms that
continue to issue quarterly guidance.
We find LTMF firms have better performance, lower uncertainty, more long-term analyst
growth forecasts, and their institutional owners have longer investment horizons. These firms are
25
also more likely to issue dividends. In contrast, ANNMF firms have experienced declining
performance and increases in uncertainty, consistent with prior research (Chen et al. 2011;
Houston et al. 2010). We generate propensity score matched control samples based on our
determinant tests and require control observations for our two event samples to come from the
same industry and same year as the event observations.
We use three abnormal accruals measures (signed and unsigned) to capture accruals
earnings management (Jones 1991; Dechow and Dichev 2002; Stubben 2006 and 2010) and
abnormal cash flows, discretionary expenditures, and production costs advanced in
Roychowdury (2006) to capture real earnings management. None of the above measures yields
evidence that long-term guidance samples differ from their PSM control samples in the extent of
earnings management.
We capture excess investments (signed and unsigned) using the two expectations models
for investment in fixed assets advanced in McNichols and Stubben (2008), and tabulate the level
of excess investment over a seven year window, centered on the year when firms start issuing
3~5 year earnings guidance (LTMF sample) and the year when firms discontinue quarterly
guidance (for ANNMF sample). Again, we fail to find consistent evidence of a difference in
excess investment.
Taken as a whole, our evidence does not support the view that long-term guidance
mitigates managerial myopia. The evidence in this study helps inform the current policy debate
on the merits of long-term guidance as a tool to reduce managerial myopia. While some
prominent practitioners and academics believe replacing short-term guidance with long-term
guidance can mitigate managers’ focus on short-term results, our results do not support this
argument.
26
Our paper also adds to the growing literature motivated by the debate on the costs and
benefits of short-term guidance in particular and both short-term and long-term guidance in
general. Recent research shows that, contrary to popular belief, firms issuing quarterly guidance
exhibit less, not more, accruals earnings management (Call et al. 2014). In addition, concurrent
research shows that guidance firms generate a higher number of patents and patent citations
when compared to non-guiders (Chen, Huang, and Lao 2014). The greater extent of innovation
exists not only when firms give annual guidance, but also when firms give quarterly guidance.
We extend this body of research by focusing on the potential impact of long-term earnings
guidance on earnings management behavior and investment decisions. Our research provides the
first step toward a better understanding of the impact of long-term guidance on managerial short-
termism and we believe further insights can be gained from more research in this area.
27
References:
Ball, R., and L. Shivakumar. 2006. The role of accruals in asymmetrically timely gain and loss
recognition. Journal of Accounting Research 44 (2): 207-242.
Barton, J., and G. Waymire. 2004. Investor Protection under Unregulated Financial Reporting.
Journal of Accounting & Economics 38: 65-116.
Beyer, A., D. Cohen, T. Lys, and B. Walther. 2010. The financial reporting environment: A
review of recent literature. Journal of Accounting & Economics 50: 296-343.
Buffett, W. 2000. Berkshire Hathaway Inc. 2000 Annual Report. Available at
http://www.berkshirehathaway.com/2000ar/2000ar.pdf.
Call, A., S. Chen, B. Miao, and Y. Tong. 2014. Short-Term Guidance and Accrual-Based
Earnings Management. Accepted and forthcoming at the Review of Accounting Studies.
Chen, S., K. Huang and B. Lao. 2014. Earnings guidance and firm innovation. University of
Texas and Florida International University, working paper.
Chen, S., D. Matsumoto, and S. Rajgopal. 2011. Is silence golden? An empirical analysis of
firms that stop giving quarterly earnings guidance. Journal of Accounting & Economics
51: 134-150.
CFA Institute. 2006. Breaking the short-term cycle: discussion and recommendations on how
corporate leaders, asset managers, investors, and analysts can refocus on long-term value.
CFA Institute, Business Institute Roundtable for Corporate Ethics.
Cohen, D. 2008. Does Information Risk Really Matter? An analysis of the Determinants and
Economic Consequences of Financial Reporting Quality. Asia-Pacific Journal of
Accounting and Economics 15 (2): 69-90.
Cohen, D., and P. Zarowin. 2010. Accrual-based and real earnings management activities around
seasoned equity offerings. Journal of Accounting & Economics 50 (1): 2-19.
Committee for Economic Development. 2007. Built to last: Focusing corporations on long-term
performance. Available at https://www.ced.org/reports/single/built-to-last-focusing-
corporations-on-long-term-performance
Dechow, P., and I. Dichev. 2002. The quality of accruals and earnings: The role of accrual
estimation errors. The Accounting Review 77: 35-39.
DeFond, M., and J. Jiambalvo. 1994. Debt covenant violation and manipulation of accruals.
Journal of Accounting and Economics 17: 145-177.
28
Fuller, J., M. Jensen. 2010. Just Say No to Wall Street: Putting a Stop to the Earnings Game.
Journal of Applied Corporate Finance 22 (1): 59-63.
Gaspar, J.M., M. Massa, and P. Matos. 2005. Shareholder investment horizons and the market
for corporate control. Journal of Financial Economics 76: 135-165.
Healy, P.M., K.G. Palepu. 2001. A review of the empirical disclosure literature. Journal of
Accounting and Economics 31, 405–440.
Houston, J., B. Lev, and J. Tucker. 2010. To guide or not to guide? Causes and consequences of
stopping and subsequently resuming earnings guidance. Contemporary Accounting
Research 27 (1): 143–18.
Hribar, P., and D.C. Nichols. 2007. The use of unsigned earnings quality measures in tests of
earnings management. Journal of Accounting Research 45 (5): 1017-1053.
Jones, J. 1991. Earnings management during import relief investigations. Journal of
Accounting Research 29 (2): 193-228.
Lao, B. (2013). Does earnings guidance contribute to investor short-termism? University of
Texas at Austin, dissertation.
McNichols, M., and S. Stubben. 2008. Does earnings management affect firms’ investment
decisions? The Accounting Review 83 (6): 1571-1603.
Petersen, M.A. 2009. Estimating standard errors in financial panel data sets: comparing
approaches. Review of Financial Studies 22 (1): 435-480.
Roychowdhury, S. 2006. Earnings management through real activities manipulation. Journal
of Accounting and Economics 42 (3): 335-370.
Skinner, D., and R. Sloan. 2002. Earnings surprises, growth expectations, and stock returns:
do’'t let an earnings torpedo sink your portfolio. Review of Accounting Studies 7(2-3):
287-312.
Stubben, S. 2006. The use of discretionary revenues to meet earnings and revenue targets.
Doctoral dissertation, Stanford University.
Stubben, S. 2010. Discretionary revenues as a measure of earnings management. The
Accounting Review 85 (2): 695–717.
The Aspen Institute. 2007. Long-term value creation: Guiding principles for corporations and
investors. Available at http://www.aspeninstitute.org/publications/aspen-principles-long-
term-value-creation-guiding-principles-corporations-investors
29
U.S. Chamber of Commerce. 2007. Commission on the regulation of U.S. capital markets in the
21st century: Report and recommendations. Available at
http://www.centerforcapitalmarkets.com/wp-content/uploads/2014/06/Commission-on-
the-regulation-of-us-cap-markets-report-and-recommendations.pdf
Zang, A. 2012. Evidence on the tradeoff between real activities manipulation and accrual-based
earnings management. The Accounting Review 87 (2): 675-703.
30
Appendix
Definition of Variables
Variables Definition
ABAC Absolute value of the residuals based on the Jones (1991) model after controlling
for economic losses as in Ball and Shivakumar (2006). The following regression
model is estimated annually (quarterly) for each industry (based on 2-digit SIC
codes) with at least 20 observations:
ACCit = β0 + β1ΔREVit + β2NPPEit + β3INDADJ_CFOit + β4DINDit +
β5(DINDit × INDADJ_CFOit) + εit.
ACC is total accruals calculated from statement of cash flow (SCF) data as
earnings before extraordinary items minus cash flows from operations. (Quarterly
earnings before extraordinary items and quarterly cash flow from operations are
calculated using the year-to-date data items IBCY and OANCFY in Compustat.)
ΔREV is change in revenue (SALE). NPPE is net property, plant, and equipment
(PPENT) at the beginning of the year (quarter). (We use net PP&E at the
beginning of the quarter because quarterly gross PP&E (PPEGTQ) is missing for
a large number of observations in Compustat.) INDADJ_CFO is annual
(quarterly) cash flow from operations minus the median cash flow from
operations for all firms in the same industry (based on 2-digit SIC code) in the
same year (quarter). DIND is a dummy variable set to one if INDADJ_CFO is
less than zero and set to zero otherwise. All variables except DIND are deflated
by average total assets (AT), and all input variables are winsorized at the extreme
1 and 99 % level. The absolute value of the regression residuals (εi,t) is multiplied
by 100. Larger values of ABAC indicate more earnings management
ABDD Absolute value of the residuals from the Dechow-Dichev (2002) model after
controlling for economic losses as in Ball and Shivakumar (2006). The following
regression model is estimated annually (quarterly) for each industry (based on 2-
digit SIC codes) with at least 20 observations:
ACCit = β0 + β1CFOit-1 + β2CFOit + β3CFOit+1 + β4DINDit + β5(DINDit ×
INDADJ_CFOit) +εit.
ACC is total accruals calculated from statement of cash flow (SCF) data as
earnings before extraordinary items minus cash flow from operations. (Quarterly
earnings before extraordinary items and quarterly cash flow from operations are
calculated using the year-to-date data items IBCY and OANCFY in Compustat.)
INDADJ_CFO is cash from operations minus the median cash from operations
for all firms in the same industry (based on 2-digit SIC code) in the same year
(quarter). DIND is a dummy variable set to one if INDADJ_CFO is less than zero,
and set to zero otherwise. All variables except DIND are deflated by average total
assets (AT). The absolute values of the regression residuals (εi,t) are multiplied by
100. Higher values of ABDD indicate more earnings management.
ABREV Absolute value of the residuals based on the Stubben (2010) model.
The following regression model is estimated annually for each industry (based on
two-digit SIC codes) with at least 20 observations:
ΔARit = β0 + β1ΔR1_3it + β2ΔR4it + εit.
ΔAR is annual change in accounts receivables (RECT). ΔR1_3 is change in the
revenues of the first three quarters. ΔR4 is the change in revenues of the fourth
quarter.
For the quarterly tests, the following regression model is estimated quarterly for
each industry (based on two-digit SIC codes) with at least 20 observations:
ΔARiq = β0 + β1(1/ASSETiq) + β2ΔRiq + εiq.
ΔAR is quarterly change in accounts receivables (RECTQ). ΔR is the quarterly
change in revenue (SALEQ). ASSET is average total assets (ATQ). ΔR and ΔAR
are both deflated by average total assets (ATQ), and all input variables are
winsorized at the extreme 1 % and 99 % level. The absolute value of the
31
Variables Definition
regression residuals (εi,t), is multiplied by 100. Larger values of ABREV indicate
more earnings management.
ANNMF 1 if the firm issues at least 3 quarterly management forecasts in four consecutive
quarters, but 0 quarterly management forecasts and at least 1 annual management
forecast in the following four consecutive quarters. 0 if the firm issues at least 3
quarterly management forecasts in the first four consecutive quarters, and at least
3 quarterly management forecasts in the following four consecutive quarters.
BTM Ratio of book to market value of equity calculated as book value of equity (CEQ)
scaled by market value of equity (CSHO × PRCC_F).
CAPINT Capital intensity calculated as net property, plant, and equipment (PPENT)
divided by total assets (AT).
CF Total cash flow taken from the cash flow statements.
σ(CFO) Standard deviation of annual cash flow (OANCF) deflated by average total assets.
Standard deviations are calculated over the prior 10 years. A minimum of 5 years
of data is required for the calculation σCFO.
CHURN Weighted-average turnover rate of all institutional investors of the firm,
multiplied by negative one. Turnover rate is measured at the end of year t, and
defined using the methodology in Gaspar et al. (2005). Note the Gaspar et al.
(2005) measure captures the average turn-over rate of all institutional investors
for a given firm at a given time, thus it is negatively correlated with the average
investment horizon of a firm’s institutional investors. Since we multiply this
turnover rate with negative one, higher CHURN indicates longer investment
horizon of institutional investors.
ΔCHURN Year over year change in CHURN. Positive CHURN indicate increases in
long-term institutional holdings, and vice versa for negative CHURN. DISCFO Abnormal operating cash flows, measured as the residuals from the following
regression:
CFOit/Ait-1 = β0 + β1(1/Ait-1) + β2(REVit/Ait-1) + β3(ΔREVit/Ait-1) + εit.
The regression model is estimated annually for each industry (based on 2-digit
SIC codes) with at least 20 observations. CFO is cash flow from operations
(OANCF). REV is revenue (SALE). A is total assets (AT). Lower values of
DISCFO indicate greater extent of real management.
DISEXP Abnormal discretionary expenses, measured as the residuals from the following
regression:
DISEXPit/Ait-1 = β0 + β1(1/Ait-1) + β2(REVit-1/Ait-1) + εit.
The regression model is estimated annually for each industry (based on 2-digit
SIC codes) with at least 20 observations. DISEXP is SG&A expenses (XSGA).
REV is revenue (SALE). A is total assets (AT). Lower values of DISEXP indicate
greater extent of real management.
DISPROD Abnormal production costs, measured as the residuals from the following
regression:
PRODit/Ait-1 = β0 + β1(1/Ait-1) + β2(REVit/Ait-1) + β3(ΔREVit/Ait-1) + β4(ΔREVit–- 1/Ait-
1) + εit.
The regression model is estimated annually for each industry (based on 2-digit
SIC codes) with at least 20 observations. PROD is production cost, measured as
the sum of COGS and increase in inventory (COGS – INVCH). REV is revenue
(SALE). A is total assets (AT). Higher values of DISPROD indicate greater extent
of real management.
DIV Indicator variable coded as one if the firm declares at least two quarterly dividend
during the previous year, zero otherwise.
σ(EARN) Standard deviation of annual earnings before extraordinary items (IB) deflated by
average total assets. Standard deviations are calculated over the prior 10 years. A
minimum of 5 years of data is required for the calculation σEARN.
32
Variables Definition
GROWTH Natural log of total assets divided by lagged total assets.
INST Percentage institutional ownership measured at the end of year t.
ΔINST Year over year difference in INST.
INVT Investment in capital expenditures, taken from the cash flow statements.
LEV Proportion of long-term debt (DLTT) to total assets (AT).
LOGMV Natural log of market value (PRCC_F * CSHO).
LOGAF Log (1 + analyst following).
LOGSHROUT Log (number of shares outstanding).
LTAF Number of analysts issuing long-term-growth forecast for the firm during year t.
ΔLTAF Year over year change in LTAF.
LTMF An indicator variable coded as 1 if management issues long-term earnings
guidance for 3~5 year ahead during fiscal year t, 0 otherwise.
OPCYCLE Natural log of the firm’s operating cycle measured in days, based on turnover in
accounts receivable and inventory. Specifically, the firm’s operating cycle is
calculated as 360 × ((ARt + ARt-1)/SALESt + (INVt + INVt-1)/COGSt). AR is
accounts receivable (RECT). SALES is sales revenue (SALE). INV is inventory
(INVT). COGS is cost of goods sold (COGS).
PMBAF A firm’s percentage of meeting or beating earnings expectations in the 8 quarters
during year t and t-1. Expected earnings is measured as consensus analyst
forecasts before earnings announcements from IBES.
POST Indicator variable set to 1 if the firm-year observation is measured at one year
after the quarter during which short-term guidance is stopped.
PRECISE Average prediction of all management forecast issued during year t. Precision is
coded as 3 if the MF is a point estimate, 2 if the MF is a range estimate and 1 for
all other estimates.
Q Tobin’s Q, measured as market value of assets (MVE+TAbeE) divided by book
value of assets (TA).
QRT2/3/4 Indicator variable set to one if Tobin’s Q falls into the 2nd
, 3rd
, and 4th quartile of
the industry-year distribution.
RET Market-adjusted buy-and-hold returns for the 12 months beginning from month -
12 ending month -1, with month 0 being the announcement month.
ResMA Indicator set to one if the firm experiences a restructuring and/or a merger &
acquisition event, zero otherwise. Restructuring event is identified as one if the
firm reports non-zero restructuring charges (RCP ≠0) during the year. M&A
activities are identified using the SDC database.
ROA Return on assets (IB/AT).
SBEAT Indicator variable coded as 1 if 0<actual EPS – consensus forecast<=0.01
SIZE Natural log of total sales (SALE).
STDRET Standard deviation of daily raw returns over a 12-month period.
ΔSTDRET Year over year change in STDRET
SMISS Indicator variable coded as 1 if -0.01<=actual EPS – consensus forecast<0
STK_ISSUE An indicator variable coded as one if the firm increases the number of shares
outstanding in year t (CSHO) by at least 20 % and zero otherwise.
SUE Actual EPS minus consensus forecast divided by price two days before earnings
announcement.
|XINVT| Excess investment, calculated as the absolute value of the residuals from the
following two models – these models are estimated cross-sectionally for each
industry-year group with at least 20 observations for each group:
Short Model: INVTit = a + β1Qit-1 + β2CFit-1 + eit,
Long Model: INVTit = a + β1Qit-1 + β2Q_QRT2it-1 + β3Q_QRT3it-1 +β4Q_QRT4it-
1
+β5CFit-1 +β6GROWTHit-1 + β7INVTit-1 + eit, Compustat mnemonics are in parentheses.
33
Figure 1
Definition of ANNMF
Q1
3 quarterly
forecasts
Pre-
period
Q2 Q3 Q4
Post-
period
Q1 Q2 Q3 Q4
1 annual forecast and 0
quarterly forecasts
34
Table 1
Distribution of LTMF and ANNMF
Panel A: Distribution By fiscal year
Year LTMF = 1 ANNMF = 1
2000 34 -
2001 49 27
2002 75 48
2003 66 49
2004 84 60
2005 74 93
2006 75 101
2007 81 72
2008 73 72
2009 57 32
2010 51 20
2011 67 18
2012 58 17
2013 10 -
Total 854 609
Panel B: Distribution by Fama-French 12 industry
Industry LTMF = 1 ANNMF = 1
Business Equipment 38 109
Chemicals 43 18
Consumer Durables 19 20
Energy 0 10
Health 86 59
Manufacturing 59 66
Money 102 57
Consumer NonDurables 130 38
Other 55 81
Shops 172 127
Telecom 5 4
Utilities 145 20
Total 854 609
35
Table 2
Determinants of Long-Term Guidance
Panel A: Descriptive statistics
LTMF = 1
(N = 661)
LTMF = 0
(N = 16,695)
Tests of difference:
1 - 0
Mean Median Mean Median
Pred. Sign
of Diff. Mean Median
RET-12 0.111 0.063 0.110 0.001 + 0.001 0.062***
ROA 0.067 0.060 0.013 0.039 + 0.054*** 0.021***
STDRET 0.020 0.017 0.033 0.028 -0.013*** -0.011***
LTAF 4.914 4.000 2.610 2.000 + 2.304*** 2.000***
CHURN -0.271 -0.262 -0.311 -0.297 + 0.040*** 0.035***
ResMA 0.625 1.000 0.546 1.000 0.079*** 0.000***
PRECISE 2.092 2.000 2.137 2.000 + -0.046*** 0.000**
DIV 0.710 1.000 0.345 0.000 + 0.367*** 1.000***
ANNMF = 1
(N = 583)
ANNMF = 0
(N = 17,130)
Tests of difference:
1 – 0
Mean Median Mean Median
Pred. Sign
of Diff. Mean Median
RET-12 0.026 -0.063 0.138 0.034 -0.112*** -0.097***
∆ROA -0.006 -0.001 0.004 0.003 -0.010*** -0.005***
∆STDRET 0.001 0.000 -0.002 -0.002 + 0.003*** 0.002***
∆LTAF -0.250 0.000 -0.012 0.000 -0.238** 0.000**
∆CHURN 0.009 0.005 0.012 0.012 + -0.003 -0.006*
Panel B: Probit regressions
Determinants of LTMF Determinants of ANNMF
Predictors
Predicted
Sign
Coeff.
(t-statistics)
Predictors
Predicted
Sign
Coeff.
(t-statistics)
Intercept -1.100*** (-7.37) Intercept -1.816*** (-96.17)
RET-12 + 0.144*** (4.25) RET-12 -0.144*** (-3.77)
ROA + 0.901*** (2.82) ROA -0.137 (-0.62)
STDRET -30.06*** (-12.55) STDRET + 9.044*** (5.32)
LTAF + 0.080*** (15.57) LTAF -0.018** (-2.15)
CHURN + 1.647*** (4.85) CHURN + -0.527** (-2.02)
ResMA -0.009 (-0.18)
PRECISE + 0.063 (1.55)
DIV + 0.382*** (8.50)
NOBS
(Event = 1 / 0)
661 /
16,695
583 /
17,130
McFadden R2 0.172 0.012
36
Table 2 (continued)
Panel C: Comparison of matching variables
LTMF = 1
(N = 657)
LTMF = 0
(N = 657)
Tests of difference:
1 - 0
Mean Median Mean Median
Pred. Sign
of Diff. Mean Median
RET-12 0.110 0.060 0.114 0.051 + -0.004 0.010
ROA 0.067 0.060 0.062 0.051 + 0.005 0.009**
STDRET 0.020 0.017 0.020 0.018 0.000 -0.001
LTAF 4.933 4.000 4.355 3.000 + 0.578*** 1.000***
CHURN -0.270 -0.262 -0.273 -0.264 + 0.003 0.002
ResMA 0.626 1.000 0.598 1.000 0.027 0.000
PRECISE 2.093 2.000 2.073 2.000 + 0.019 0.000
DIV 0.711 1.000 0.743 1.000 + -0.032 0.000
ANNMF = 1
(N = 544)
ANNMF = 0
(N = 544)
Tests of difference:
1 – 0
Mean Median Mean Median
Pred. Sign
of Diff. Mean Median
RET-12 0.019 -0.069 0.108 0.015 -0.088** -0.083***
ROA -0.005 -0.001 0.000 0.000 -0.005 -0.001
STDRET 0.001 0.000 0.000 0.000 + 0.000 0.000
LTAF -0.252 0.000 -0.388 0.000 0.136 0.000
CHURN 0.008 0.006 0.004 0.005 + 0.005 -0.001
Note: For all variable definition please refer to the Appendix.
37
Table 3
Accruals-based Earnings Management: Absolute Value of Discretionary Accruals/Revenues
Panel A: Univariate
LTMF = 1 LTMF = 0 Difference ( 1- 0)
N Mean Median N Mean Median Mean Median
ABAC 430 0.048 0.037 421 0.047 0.031 0.001 0.006
ABDD 379 0.035 0.026 372 0.033 0.021 0.002 0.005
ABREV 429 0.013 0.008 418 0.014 0.009 -0.001 -0.001
ANNMF = 1 ANNMF = 0 Difference (1 - 0)
N Mean Median N Mean Median Mean Median
ABAC 409 0.056 0.039 404 0.057 0.035 -0.000 0.004
ABDD 361 0.049 0.032 352 0.044 0.031 0.005 0.001
ABREV 404 0.019 0.010 403 0.019 0.011 -0.001 -0.001
Panel B: Cross-sectional regressions
LTMF Regressions ANNMF regressions
ABAC ABDD ABREV ABAC ABDD ABREV
Intercept 2.449 27.241**
*
9.914*** 0.357 1.215 3.148**
(0.64) (2.90) (2.75) (0.28) (0.48) (2.67)
LTMF/ANNMF 0.283 0.164 0.289 -0.106 0.418 -0.098
(0.62) (0.35) (0.68) (-0.58) (1.29) (-0.75)
LEV -2.095 0.075 -0.571 0.496 -1.431 -0.738
(-1.20) (0.04) (-0.43) (0.78) (-0.94) (-1.46)
BTM -1.064 0.288 0.722 0.335 -0.315 -0.563
(-0.94) (0.30) (1.10) (1.33) (-0.39) (-1.51)
OPCYCLE 0.877** 0.188 -0.169 0.172 0.848*** 0.11
(2.05) (0.48) (-0.44) (1.07) (2.96) (0.92)
CAPINT 2.537 -1.821 -3.735*** -2.365*** 0.729 -1.355***
(1.53) (-1.2) (-3.14) (-3.76) (0.62) (-2.76)
ROA -12.473 -
11.045**
-6.189* -1.135 -3.766 -2.806
(-1.02) (-2.14) (-1.9) (-1.07) (-0.90) (-1.54)
SIZE -0.265 -0.17 -0.195 -0.074 -0.228* -0.122**
(-1.56) (-1.11) (-1.42) (-1.21) (-1.86) (-2.35)
INST -0.71 -3.535** -1.628 -0.296 -0.356 -0.661
(-0.50) (-2.54) (-1.31) (-0.66) (-0.39) (-1.47)
σ(CFO) 1.956 16.98 6.635 16.446*** 13.364* 4..483
(0.18) (1.30) (0.70) (3.62) (1.83) (1.20)
σ(EARN) 29.192 1.692 6.108 -4.443*** 10.575 1.129
(1.38) (0.26) (1.31) (-2.71) (1.09) (0.26)
Ind. & Yr.
Fixed Effects
Yes
Yes
Yes
Yes
Yes
Yes
N 851 813 713 807 751 847
Adj-R2 0.126 0.181 0.140 0.175 0.119 0.140
38
Table 3
Accruals-based Earnings Management: Absolute Value of Discretionary Accruals/Revenues
(continued)
Panel C: Difference-in-difference regressions for ANNMF. Post is an indicator variable coded as one for
quarters after quarter 1, the quarterly guidance cessation quarter. The sample includes one year before and one
year after the event quarter.
ABAC ABDD ABREV
Intercept 28.895*** 10.43*** -0.598
(3.15) (3.68) (-0.51)
ANNMF 0.428 -0.636 -0.067
(0.94) (-1.64) (-0.31)
POST 0.185 -0.299 0.028
(0.43) (-0.83) (0.16)
ANNMF*POST -0.28 0.992* -0.038
(-0.44) (1.82) (-0.16)
LEV 1.794 1.174 0.006
(1.24) (1.05) (0.01)
BTM -0.686 -0.271 0.102
(-0.87) (-0.50) (0.45)
OPCYCLE 0.25 -0.189 0.327**
(0.83) (-0.71) (2.14)
CAPINT -2.196* -4.415*** -2.174***
(-1.84) (-4.41) (-4.64)
ROA -5.451 -8.704*** -2.473*
(-1.18) (-2.97) (-1.95)
SIZE -0.4*** -0.275*** -0.016
(-2.98) (-2.63) (-0.25)
INST -4.185*** -2.405** -0.315
(-3.29) (-2.22) (-0.73)
σ(CFO) 15.254* 16.095** 17.344***
(1.72) (2.26) (3.73)
σ(EARN) -1.39 1.796 -3.36*
(-0.40) (0.73) (-1.70)
Industry Fixed Effects Yes Yes Yes
Year Fixed Effects Yes Yes Yes
N 1,610 1,492 1,593
Adj-R2
0.144 0.161 0.169
Note: standard errors clustered by firm. For all variable definition please refer to the Appendix.
39
Table 4
Accruals-based Earnings Management: Signed Discretionary Accruals/Revenues
Panel A: Univariate
LTMF = 1 LTMF = 0 Difference ( 1- 0)
N Mean Median N Mean Median Mean Median
ABAC 430 0.023 0.025 421 0.021 0.019 0.002 0.006
ABDD 379 0.015 0.017 372 0.005 0.003 0.010*** 0.014***
ABREV 429 0.000 -0.001 418 0.003 0.000 -0.004** -0.001**
ANNMF = 1 ANNMF = 0 Difference (1 - 0)
N Mean Median N Mean Median Mean Median
ABAC 409 0.007 0.013 404 0.023 0.020 -0.016*** -0.007*
ABDD 361 -0.004 0.007 352 0.005 0.009 -0.009* -0.002
ABREV 404 -0.002 -0.004 403 -0.001 -0.002 -0.001 -0.002
Panel B: Cross-sectional regressions
LTMF Regressions ANNMF Regressions
ABAC ABDD ABREV ABAC ABDD ABREV
Intercept -5.111 4.824 1.784 13.111 7.529 -0.225
(-1.16) (1.56) (1.24) (1.29) (1.28) (-0.13)
LTMF/ANNMF 0.195 1.009** -0.388** -0.893 -0.424 0.036
(0.4) (2.21) (-2.37) (-1.48) (-0.73) (0.15)
LEV 0.235 0.514 -0.836 -2.28 -1.482 0.043
(0.11) (0.28) (-1.24) (-1.04) (-0.78) (0.05)
BTM -1.888 -0.515 -0.357 0.071 -1.826* -0.253
(-1.64) (-0.52) (-0.84) (0.06) (-1.85) (-0.75)
OPCYCLE 0.269 -0.465 -0.089 0.185 -0.289 0.015
(0.51) (-1.08) (-0.53) (0.36) (-0.55) (0.08)
CAPINT 2.633 -5.149*** -0.61 2.779 -2.274 -0.752
(1.31) (-3.37) (-0.95) (1.47) (-1.32) (-1.13)
ROA 35.277*** 11.194* -2.395 19.898** 11.726** 1.89
(3.17) (1.93) (-0.73) (2.35) (2.11) (1.15)
SIZE -0.304 -0.247 -0.05 0.302 -0.083 -0.086
(-1.53) (-1.43) (-0.72) (1.33) (-0.37) (-1.11)
INST -2.963* -2.962** -0.104 -1.844 -2.177 -0.212
(-1.78) (-2.41) (-0.19) (-0.95) (-1.19) (-0.3)
σ(CFO) -14.224 -29.371** 5.496 -1.082 -22.358 12.231**
(-1) (-2.07) (0.95) (-0.06) (-1.42) (2.16)
σ(EARN) -15.443 -0.425 6.089 6.744 8.35 -1.566
(-0.8) (-0.04) (1.13) (0.96) (1.16) (-0.91)
Industry Fixed Effects Yes Yes Yes Yes Yes Yes
Year Fixed Effects Yes Yes yes Yes Yes Yes
N 851 751 847 813 713 807
Adj-R2 0.175 0.112 0.087 0.184 0.111 0.079
40
Table 4
Accruals-based Earnings Management: Signed Discretionary Accruals/Revenues
(continued)
Panel C: Difference-in-difference regressions for ANNMF. Post is an indicator variable coded as one for
quarters after quarter 1, the quarterly guidance cessation quarter. The sample includes one year before and one
year after the event quarter.
ABAC ABDD ABREV
Intercept 12.354 0.976 0.536
(1.26) (0.19) (0.34)
ANNMF 0.712 0.228 0.115
(1.12) (0.4) (0.42)
POST 0.414 0.341 0.018
(0.68) (0.72) (0.07)
ANNMF*POST -1.50* -0.684 -0.138
(-1.68) (-0.91) (-0.41)
LEV -2.34 -2.297 0.185
(-1.31) (-1.3) (0.24)
BTM -1.939** -3.301*** -0.17
(-2.07) (-3.71) (-0.51)
OPCYCLE 0.483 0.448 -0.002
(1.18) (1.13) (-0.01)
CAPINT 3.097** -1.639 -0.408
(2.2) (-1.11) (-0.86)
ROA 8.726** -4.174 3.043
(2.15) (-0.77) (1.26)
SIZE 0.366* 0.019 -0.051
(1.96) (0.11) (-0.66)
INST -1.362 -1.65 -0.803
(-0.87) (-1.05) (-1.52)
σ(CFO) 17.944 -7.302 13.163**
(1.48) (-0.52) (2.3)
σ(EARN) 0.798 4.299 -2.888**
(0.16) (0.71) (-2.08)
Industry Fixed Effects Yes Yes Yes
Year Fixed Effects Yes Yes Yes
N 1,610 1,492 1,593
Adj-R2
0.127 0.075 0.055
Note: standard errors clustered by firm. For all variable definition please refer to the Appendix.
41
Table 5
Real Earnings Management
Panel A: Univariate comparisons
LTMF = 1 LTMF = 0 Difference (1-0)
N Mean Median N Mean Median Mean Median
DISCFO 447 0.107 0.087 445 0.099 0.080 0.009 0.006
DISEXP 437 -0.070 -0.067 439 -0.038 -0.068 -0.032** 0.001
DISPROD 416 -0.082 -0.089 398 -0.082 -0.048 0.000 -0.041
ANNMF = 1 ANNMF = 0 Difference (1-0)
N Mean Median N Mean Median Mean Median
DISCFO 423 0.070 0.056 428 0.084 0.082 -0.013* -0.026***
DISEXP 405 -0.043 -0.063 418 -0.067 -0.077 0.024* 0.014**
DISPROD 394 -0.046 -0.036 399 -0.024 -0.019 -0.022 -0.017
Panel B: Cross-sectional regressions
LTMF Regressions ANNMF Regressions
DISCFO DISEXP DISPROD DISCFO DISEXP DISPROD
Intercept -5.085 11.407 25.336 22.093*** 18.206 -15.915
(-0.97) (0.56) (1.07) (3.11) (0.72) (-0.73)
LTMF/ANNMF -0.359 -2.992 2.755 -0.46 3.475** -3.297
(-0.42) (-1.26) (1.04) (-0.58) (2.03) (-1.65)
PMBAF -1.014 6.346 -7.966 -2.025 7.751 0.164
(-0.57) (1.42) (-1.55) (-0.87) (1.61) (0.03)
STK_ISSUE -0.428 -3.609 9.152 3.353* -2.628 2.078
(-0.18) (-0.64) (1.32) (1.79) (-0.8) (0.42)
LOGNAF 0.949 1.443 -3.79 1.955** 0.738 -2.773
(0.87) (0.69) (-1.38) (2.07) (0.37) (-1.06)
BTM -1.842 -18.15*** 18.424** -1.381 -11.557*** 8.876***
(-0.99) (-2.96) (2.42) (-1.02) (-3.74) (2.64)
LOGSHROUT -0.71 1.729 -4.75* -3.201*** 2.026 -0.578
(-0.77) (0.75) (-1.85) (-2.88) (1.16) (-0.24)
LOGMV 1.325* -4.596** 5.606** 2.727*** -3.675** 1.106
(1.68) (-2.11) (2.29) (3.06) (-2.46) (0.55)
ROA 47.58*** 21.613 -62.31*** 24.3*** 10.545 -37.353***
(6.04) (0.90) (-2.62) (2.99) (0.98) (-2.85)
Year Fixed Effects Yes Yes Yes Yes Yes Yes
Industry Fixed Effects Yes Yes Yes Yes Yes Yes
N 792 784 731 732 712 692
Adj-R2 0.562 0.203 0.291 0.416 0.164 0.174
42
Table 5 Real Earnings Management
(continued)
Panel C: Difference-in-difference regressions for ANNMF. Post is an indicator variable coded as one for
quarters after quarter 1, the quarterly guidance cessation quarter. The sample includes one year before and one
year after the event quarter.
DISCFO DISEXP DISPROD
Intercept 16.507** 24.745 -16.322
(2.53) (0.97) (-0.78)
ANNMF -0.517 2.066 -2.081
(-0.60) (1.26) (-1.04)
POST -0.591 -0.373 0.647
(-0.91) (-0.37) (0.61)
ANNMF*POST 0.329 1.643 -1.766
(0.41) (1.61) (-1.47)
PMBAF 1.247 10.143*** -5.523
(0.63) (2.7) (-1.09)
STK_ISSUE 0.372 3.402 -1.416
(0.21) (1.03) (-0.38)
LOGNAF 1.875** -1.016 -1.707
(2.14) (-0.52) (-0.73)
BTM -2.226* -13.76*** 10.781***
(-1.67) (-4.43) (3.42)
LOGSHROUT -2.716*** 1.663 -0.309
(-2.87) (0.99) (-0.14)
LOGMV 2.316*** -3.423** 1.067
(2.92) (-2.19) (0.57)
ROA 23.064*** 3.652 -41.189***
(2.74) (0.26) (-4.44)
Year Fixed Effects Yes Yes Yes
Industry Fixed Effects Yes Yes Yes
N 1,482 1,442 1,395
Adj-R2
0.390 0.176 0.170
Note: standard errors clustered by firm. For all variable definition please refer to the Appendix.
43
Table 6
Excess Investment: Absolute Value of Excess Investment
Panel A: LTMF, mean values of |XINT|
LTMF=1 LTMF=0
Test of Diff in Mean
Long
Model
Short
Model
Long
Model
Short
Model
N
Mean
|XINT|
N
Mean
|XINT| N
Mean
|XINT|
N
Mean
|XINT|
Long
Model
Short
Model
-3 586 0.110 586 0.113 577 0.106 577 0.124 -0.003 -0.011
-2 599 0.104 599 0.116 590 0.114 590 0.124 -0.012 -0.008
-1 614 0.102 614 0.114 604 0.110 604 0.127 -0.013 -0.013
0 615 0.099 615 0.107 607 0.088 607 0.102 -0.008* 0.005
1 559 0.093 559 0.108 548 0.086 548 0.098 -0.014 0.010
2 502 0.095 502 0.114 476 0.079 476 0.097 -0.020** 0.017**
3 461 0.093 461 0.114 422 0.090 422 0.101 -0.022 0.013*
Panel B: ANNMF, mean values of |XINT|
ANNMF=1 ANNMF=0
Test of Diff in Mean
Long
Model
Short
Model
Long
Model
Short
Model
N
Mean
|XINT|
N
Mean
|XINT| N
Mean
|XINT|
N
Mean
|XINT|
Long
Model
Short
Model
-3 474 0.147 474 0.159 471 0.168 471 0.180 -0.021 -0.021
-2 488 0.143 488 0.152 488 0.181 488 0.206 -0.038** -0.054***
-1 497 0.165 497 0.179 502 0.145 502 0.166 0.020 0.013
0 502 0.139 502 0.150 509 0.143 509 0.166 -0.004 -0.016
1 479 0.116 479 0.134 487 0.121 487 0.143 -0.004 -0.009
2 438 0.128 438 0.146 441 0.144 441 0.156 -0.016 -0.009
3 409 0.107 409 0.124 396 0.141 396 0.153 -0.034** -0.028**
Note: standard errors clustered by firm. For all variable definition please refer to the Appendix.
44
Table 7
Excess Investment: Signed Value of Excess Investment
Panel A: LTMF, mean values of XINT
LTMF=1 LTMF=0
Test of Diff in Mean
Long
Model
Short
Model
Long
Model
Short
Model
N
Mean
XINT
N
Mean
XINT N
Mean
XINT
N
Mean
XINT
Long
Model
Short
Model
-3 586 -0.041 586 -0.038 577 -0.027 577 -0.031 -0.014 -0.007
-2 599 -0.042 599 -0.044 590 -0.030 590 -0.036 -0.012 -0.007
-1 614 -0.028 614 -0.032 604 -0.036 604 -0.036 0.008 0.004
0 615 -0.038 615 -0.043 607 -0.041 607 -0.051 0.003 0.008
1 559 -0.037 559 -0.050 548 -0.040 548 -0.049 0.004 0.000
2 502 -0.040 502 -0.057 476 -0.034 476 -0.048 -0.006 -0.009
3 461 -0.040 461 -0.065 422 -0.029 422 -0.046 -0.011 -0.019**
Panel B: ANNMF, mean values of XINT
ANNMF=1 ANNMF=0
Test of Diff in Mean
Long
Model
Short
Model
Long
Model
Short
Model
N
Mean
XINT
N
Mean
XINT N
Mean
XINT
N
Mean
XINT
Long
Model
Short
Model
-3 474 -0.020 474 -0.011 471 -0.028 471 -0.012 0.008 0.002
-2 488 -0.020 488 -0.013 488 0.011 488 0.023 -0.031 -0.036*
-1 497 -0.003 497 0.010 502 -0.017 502 -0.016 0.014 0.025
0 502 -0.033 502 -0.032 509 -0.015 509 -0.020 -0.019 -0.012
1 479 -0.013 479 -0.036 487 -0.029 487 -0.039 0.016 0.003
2 438 -0.040 438 -0.049 441 -0.040 441 -0.038 0.001 -0.011
3 409 -0.031 409 -0.063 396 -0.028 396 -0.035 -0.003 -0.028*
Note: standard errors clustered by firm. For all variable definition please refer to the Appendix.