DIVIDENDS AND EARNINGS QUALITY: THE MISSING LINK? April

DIVIDENDS AND EARNINGS QUALITY: THE MISSING LINK?
April, 2007
Jorge Farinha*
José António Moreira**
CETE / Faculty of Economics / University of Porto***
Visiting Research Fellows at ICRA-International Centre for Research in
Accounting / Lancaster University Management School
* Jorge Farinha, Faculdade de Economia da Universidade do Porto, Rua Roberto Frias, 4200464 Porto, Portugal. Tel. (351)-22-5571100, Fax (351)-22-5505050. E-mail:
[email protected] (corresponding author)
** José António Moreira, Faculdade de Economia da Universidade do Porto, Rua Roberto
Frias, 4200-464 Porto, Portugal. Tel. (351)-22-5571100, Fax (351)-22-5505050. E-mail:
[email protected].
***CETE - Research Center on Industrial, Labour and Managerial Economics. Faculty of
Economics, University of Porto. Research Center supported by Fundação para a Ciência e a
Tecnologia, Programa de Financiamento Plurianual through the Programa Operacional
Ciência, Tecnologia e Inovação (POCTI)/Programa Operacional Ciência e Inovação 2010
(POCI) of the III Quadro Comunitário de Apoio, which is financed by FEDER and Portuguese
funds.
DIVIDENDS AND EARNINGS QUALITY: THE MISSING LINK?
ABSTRACT
This papers presents, to our knowledge, the first large-scale and direct
empirical evidence in support of the hypothesis that dividends play an
important role in conveying information on the level of earnings management
by companies. Specifically, using a large sample of around 40,000 firm-year
observations in the U.S. market for the period 1987-03, and controlling for
other effects reported in the dividend literature, we document evidence of a
significant positive relationship between dividend payments and a set of
alternative measures of earnings quality. The evidence we report strongly
suggests that dividends can act as a credible signal of earnings quality, with
companies unengaged in earnings management being more likely to pay
dividends, to have higher dividend yields amongst divided payers and to
increase dividends per share. Our results breed significant new insights into
the almost 50-year old debate on dividends.
Keywords: dividend policy, earnings management, earnings quality, accruals
JEL Classification: G35, M41
1
1 Introduction and motivation
After almost 50 years of research initiated by the seminal papers of Lintner
(1956) and Miller and Modigliani (1961), dividend policy still remains, to a large
extent, a puzzle to financial economists (Black, 1976). Although numerous
theories have been proposed, particularly those relying on tax (Brennan,
1970), signalling (Miller and Rock, 1985), agency (Easterbrook, 1984), or
behavioural (Shefrin and Statman, 1984) explanations, consensus is still
elusive among researchers.
Our research is somewhere between the signalling and agency theory
paradigms, but uniquely building on the earnings management branch of the
accounting research literature. Our main hypothesis is that, given the
costliness of dividend reductions or omissions (Christie, 1994) managers
choose to pay, or increase, dividends only when they feel that their earnings
are not significantly influenced by accounting manipulations, so that a future
dividend reduction is unlikely. Therefore, in this sense, dividends would signal
not only future profitability but also earnings quality (in the sense of earnings
with a low degree of manipulation). An alternative perspective could be that,
given the presence of dividends as an effective managerial monitoring device
(Easterbrook, 1984), one would expect that managers might be less tempted
to engage in earnings manipulation for self-serving reasons. Either way,
investors would infer the quality of published earnings, all else constant, by the
relative importance of regular dividend payments to distributable earnings or
by positive dividend changes.
2
In addition, the possibility that investors may use dividends as a relevant
source of information about the likelihood of earnings manipulation by
managers is of an immense practical importance in the light of recent
accounting scandals (e.g., Enron, Worldcom, Xerox, among many others) and
the general public concern on corporate governance practices, particularly on
the part of large listed companies.
This potential role of dividends has been highlighted recently by The
Economist (2003, January 11) with the speculation that
“Investors have grown more sceptical about accounting profits
in the wake of Enron and Worldcom and now wonder if
evidence of profitability in the form of a dividend cheque might
help them to sleep more easily (….). If it becomes the norm
for most firms to pay out a large chunk of their profits as
dividends, companies posting fake results might not get away
with it for as long” (pp. 49-50).
Our results confirm such hypotheses. Specifically, using a large sample of
around 40,000 firm-year observations in the U.S. market for the period 198703, and controlling for other effects reported in the dividend and earnings
management modelling literatures, we document consistent evidence of a
positive relationship between dividend payments and several measures of
earnings quality. We also observe a similar relationship between such
measures and positive dividend changes.
Our paper proceeds as follows. The next section reviews the relevant literature
3
on dividends and earnings quality (management).1 Next, we present the
estimation models used and describe the sample. In Section 4 we show the
main results. The final section summarizes our findings.
2 Previous literature and testing hypotheses
2.1 Dividend theories and prior evidence
Lintner (1956) was the first of a string of papers that documented the stylized
fact that companies tend to pay a fixed fraction of long term earnings and to
adjust their payout ratios slowly over time to a long term level. The dividend
irrelevance proposition developed by Miller and Modigliani (1961) (M&M) was
the starting point from which numerous theories have tried to provide
explanations for why investors and companies apparently find dividends
important. By relaxing some of the assumptions in the M&M model, four major
sets of theories were developed. Noting that in most countries capital gains
taxes are lower than those on dividends and can be postponed until
realization, tax theories have suggested that not only should higher dividend
paying stocks command higher pre-tax returns but also that those stocks
could attract particular tax clienteles according to different investor tax
statuses (Brennan, 1970). Signaling theories, on the other hand, assume that
dividends can be used as a costly means to signal future profitability and
growth and investors recognize this by assigning higher values to higher
dividend paying stocks reacting favorably to the announcement of unexpected
dividend increases (Watts, 1973; Miller and Rock, 1985). More recently,
1
Throughout the paper we use indistinctively “earnings quality” and “earnings management”
with similar meaning, but opposite sense, as it tends to be common in the literature.
4
agency theory has advanced the different view that dividends may help to
reduce agency problems between managers and shareholders. This can be
done as dividends can help to dissipate free cash-flows (Jensen, 1986) and
induce more frequent monitoring by external capital markets as companies
turn to these more often to raise funds for investment purposes when
engaged in more generous dividend policies (Easterbrook, 1984). Finally,
behavioral theories suggest that individual biases such as the lack of selfcontrol or the wish to avoid regret lead to investors’ willingness to pay a
premium for dividend paying stocks (Shefrin and Statman, 1984). This in turn
leads to the prediction that, similar to tax explanations, distinct dividend
policies may be differently attractive to particular dividend clienteles induced
by behavioral considerations.
In spite of the abundance of theories, researchers have not reached a
consensus regarding the issue of what is the major force behind dividend
relevance, partly because of conflicting evidence but also because many of
the theories have similar predictions. Regarding taxes, a number of papers
(e.g., Black and Scholes, 1974; Miller and Scholes, 1982; Kalay and Michaely,
2000)
challenge
Brennan’s
(1970)
assertion,
and
Litzenberger
and
Ramaswamy’s (1979) evidence, of a positive relation between dividend yields
and pre-tax risk-adjusted returns.
However, some results that have been
reported are in accordance with dividend tax clienteles (e.g., Dhaliwal et al,
1999), although this may also be interlinked with behavioural explanations
(Graham and Kumar, 2006). In terms of the signalling perspective, recent
research has cast doubts on the idea that dividends signal future profitability
(De Angelo et al, 1996; Benartzi et al, 1997), in contrast with earlier studies
5
(e.g., Watts, 1973; Gonedes, 1978; Ofer and Siegel, 1987; Healy and Palepu,
1988). Similarly, although the evidence is generally supportive of agency
problems as an important influence on the dividend decision (e.g., Crutchley
and Hansen, 1989, Jensen et al, 1992; Agrawal and Jayaraman, 1994;
Farinha, 2003), still this has not been unchallenged (e.g., Howe et al, 1992).
Recent research has suggested that firms show a decline in their propensity to
pay dividends in the last decades (Fama and French, 2001). This fact has
been attributed by Baker and Wrugler (2004) to a declining demand for
dividends by investors, followed by managers’ adaptation to these investors’
changing preferences. Baker and Wrugler call this the “catering theory of
dividends”. At the same time, life-cycle explanations for this dividend
behaviour have been advanced by Fama and French (2001), among others.
These authors suggest that dividend policies of individual firms reflect a
balance between costs (e.g., taxes and other transaction costs; foregoing of
profitable investment opportunities) and benefits (e.g., reduction in agency
problems; signalling future profitability) whose relative importance varies
according to the life stages of a firm. Thus, in a certain way, this life-cycle
perspective is a synthesis of the major theories initially outlined (tax, agency,
signalling and behavioural).
We add to this body of literature a different explanation from those mentioned
above, based on existing evidence taken from the accounting literature
suggesting that a decline in earnings quality has going on for the past
decades (e.g. Dechow and Schrand, 2004). Specifically, we argue that
companies, other factors held constant, will pay out a larger fraction of their
income to the extent that they believe their accounting earnings to have
6
sufficient “quality”. For this purpose we define “earnings quality” as the
situation where earnings are not artificially manipulated by accounting options,
particularly those that borrow earnings from future periods or even, in a worst
case scenario, artificially inflate current results in instances where the
likelihood of sufficiently high future earnings is very low. Our reasoning is that
if dividends were to be paid when the underlying stream of current corporate
earnings is highly influenced by accounting options, thus not reflecting current
(or foreseeable) performance, managers would face the serious prospect of
cutting the dividend in the near future. This would in turn entice a strong
negative reaction of its stock price in the market (Christie, 1994). Therefore,
only in those instances where managers do not engage in active earnings
management will they risk paying substantial dividends2.
A few authors have considered the possibility of links, albeit indirect ones,
between dividends and earnings management (quality), but restricted their
analysis to special cases. Thus, Nissim and Ziv (2001) and Lie (2005) argue
that following bad news managers may take an “earnings bath” using
accounting choices to create accounting reserves that could function as a
“buffer” for future periods. Consistent with this assertion, Lie reports that
downward earnings management occurs simultaneously with dividend cuts
and omissions. One must note, however, that such relationship arises from a
common factor that influences both the dividends and the earnings
management activities, that is, the occurrence of bad news prompting the
2 Since the cost of a negative reaction of the stock price to a cut in dividend would be more
strongly felt by managers when these have stock options or high ownership stakes in the
company, we would also expect that in those cases the anticipated negative correlation
between earnings quality and dividends would be even more pronounced. Another different
possibility is that since exercise prices in stock options are frequently not adjusted by dividend
payments, managers may be willing to pay as little dividends as possible (Lambert et al, 1989)
7
earnings bath and the lack of earnings to pay out dividends (or little
willingness to offer cash payouts to investors, given the somber perspectives
for the firm). In other words, the observed relationship between dividends and
earnings management as reported by these authors is not a direct one. Thus,
that strand of research does not analyze the possibility that dividends can
signal the absence of earnings management, although it recognizes an
indirect association in the special circumstance of dividend cuts or omissions.
Following a different line of research, Mikhail et al (2003) take the perspective
that both earnings and dividends are informative about future firm cash-flows,
and that dividends are a substitute source of information when earnings
quality is low. The authors document a negative relationship between the
market reaction to dividend change announcements and earnings quality (as
measured by a single proxy3, the association between future cash-flows and
past earnings). While the authors fundamentally look at the typical signalingtheory perspective of dividend changes’ informativeness, their evidence can
be interpreted as suggesting a link between dividend policy and earnings
quality, albeit an indirect one. Our paper, in contrast, does not directly test
whether dividends are informative about future cash-flows. Instead, we
explicitly analyse whether dividend policy can be seen as a “quality seal” for
the earnings reported by managers, using several proxies for earnings quality
and focusing not on the market’s reaction to announcements but rather on the
manager’s dividend policy decision.
Also consistent with a relationship between dividend policy and a measure of
earnings accruals quality, Chen et al (2006) show evidence that the quality of
at the same time that manipulating earnings upwards so as to maximize the value of their
options . We do not, however, explicitly test in this paper these particular hypotheses.
8
earnings is priced as a risk factor when the market assesses the impact of
dividend changes, with factor loadings on such risk factor decreasing for firms
announcing dividend initiations and increases. Their results are consistent
with these firms experiencing an increase in the precision of their earnings
information. Again, these authors’ analysis is restricted to the particular cases
of dividend changes, while we look at the more general issue of the decision
to pay dividends in the first place and directly relate that decision to earnings
quality, while at the same time not ignoring the issue of dividend changes.
Therefore, no prior research that we are aware of has investigated using large
sample, the more general case where earnings quality is comparatively
analyzed for dividend payers versus non-payers or for high- versus lowdividend payers. The focus of our paper is therefore the potential role of
dividend policy as a managerial vehicle by which firms can convey to the
market a credible signal that their reported income is not being artificially
inflated by accounting options.
The argument that dividends can be a credible signal about earnings quality,
although we believe to be the first to test it in a large sample with the
methodology that will be described ahead, is relatively common in some
professional circles. Apart from the Economist’s quote mentioned earlier, also
an article in Barron’s Online (July 1, 2002) urged investors to
“Embrace stocks that pay healthy dividends. A bird in the hand is
better than two in the bush (...). Healthy dividend payments also
3
This is a limitation in the authors’ analysis, as pointed out by Nissim (2003).
9
indicate that companies are generating real earnings rather than
cooking the books.”
However, in a survey questionnaire to a Dutch investors’ panel reported by
Dong et al (2005), these authors observe that investors seem to reject the
idea that dividends signal manipulation-free “real” earnings.
In contrast with Dong et al., we focus on the supply side of dividend equation,
that is, on the decision on the part of managers, and not on the attitude shown
by investors4. Managers, in our view, make a trade-off between any short-term
gains (for their firms or themselves) arising from accounting manipulations
with the long term costs which will result from eventual dividend reductions as
real earnings are eventually uncovered.
Along those lines, the Breeden
(2003) report on corporate governance and accounting problems at Worldcom
suggests that
“Dividends are another method of gauging the reality of reported earnings.
The ability to pay dividends is dependent on the availability of cash, and
significant differences between the levels of reported earnings and cash
available for dividends would eventually be a red flag of potential
problems.” (p.126)
4
Dong et al (2005) added that although investors as a whole seemed to reject the idea that
dividends provide information on the degree of earnings manipulation, they also observed that
while younger investors categorically disagreed with such idea, older investors were
marginally in agreement with it.
10
The same report recommended Worldcom to pay out as a dividend at least
25% of reported income5 so as to ensure, among other things, the “quality” of
its earnings.
Consistent with this view, Skinner (2004) reports that those companies paying
out a larger part of their profits as dividends exhibit a greater degree of
earnings persistence in the future. This result, however, could as well be in
accordance with a traditional dividend signaling view as with our accounting
manipulation perspective where dividends suggest little manipulation of
reported income. In contrast, we test for a direct link between dividends and
measures of earnings quality which, if positive, would be inconsistent with the
traditional signaling perspective6.
A different perspective is whether dividends and earnings quality are related
due to lending covenants. Specifically, when firms are close to a debt
covenant requiring dividends to be constrained (or cut), managers might be
tempted to manipulate earnings upwards to maintain dividends.
2.2 Earnings quality: definition and measurement
Our definition of earnings quality is consistent with the one found in the
literature that considers earnings to be of high quality when accurately
5
See Breeden (2003), Recommendation 9.02, p. 128.
In fact, if dividends convey information on the future prospects of the firm which is not
already contained in current earnings, as suggested by the dividend signaling perspective,
one would expect that either no relationship between dividends and earnings manipulations
will be observed or, at best, that managers would engage in accounting practices so as to
anticipate future profits and pay higher current dividends. A similar situation might also occur if
firms have debt covenants that impose dividend constraints (Healy and Palepu, 1990). In
these last cases, dividends would be positively related to measures of income-increasing
accounting manipulation. Instead, our hypothesis is that dividends are negatively related to
such measures, in other words, that dividend payouts will be larger when earnings are less,
not more, managed by corporate insiders. It would seem contradictory, under a dividend
signaling perspective (or indeed a dividend covenant scenario), if managers were willing to
manipulate earnings downwards (a negative signal) while at the same time paying out more
dividends to signal good prospects ahead (a positive signal).
6
11
reflecting the firm’s long run performance (e.g. Schipper and Vincent, 2003;
Chan et al., 2006). Regarding the operationalization of this concept, although
there is no wide consensus, an implicit agreement has been emerging that
manipulated earnings have lower quality (e.g. Schipper and Vincent, 2003).
By manipulated earnings, or as it is better known in the literature, managed
earnings, we mean the outcome of managers’ intentional intervention on the
reporting process through accounting choices with the purpose of obtaining a
private gain, either for themselves or for their firms (Schipper, 1989).
Healy and Whalen (1999) mention three groups of incentives underlying
earnings management: i) capital market incentives, implying that firms or
managers tend to bear higher costs when reporting earnings decreases or
losses (Burgstahler and Dichev, 1997); ii) contractual incentives, due to the
existence of contracts imposing penalties if firms do not achieve given
accounting numbers (e.g. Sweeney, 1994); iii) anti-trust or government
regulation incentives, which may take many different forms and specific
motivations, like bank’s incentives to avoid overcoming liquidity ratios imposed
by the regulator (e.g. Beatty et al., 2002) or firms’ incentives to be granted
higher protection from imports (e.g. Jones, 1991).
The assumption made in the literature of a negative relationship between
earnings management and the quality of earnings implies that the measures
adopted to detect earnings management tend also to be used to detect
earnings quality (e.g. Wysocki, 2006; Schipper and Vincent, 2003).
The basic idea underlying the detection of earnings management is that
manipulation always leaves a trace behind in earnings components. Earnings
can be seen as the sum of cash flow from operations and accruals, and both
12
these components can be manipulated to achieve a given earnings target,
although the literature acknowledges that accruals are more flexible and have
lower management costs relative to cash flows (e.g. Healy, 1985; DeFond and
Jiambalvo, 1994). This explains why most of the models on this category are
built up around the accruals component.
There is a wide range of accrual models which have developed in the
literature, from a simple random-walk of total accruals (DeAngelo, 1986) to
econometrically
more
sophisticated
specifications.
Nevertheless,
the
comparative assessment of accrual estimates derived from different models
(e.g. Thomas and Zhang, 2000) does not show meaningful differences
between those from “sophisticated” and “unsophisticated” models. This is
probably the main reason why a quite simple solution, the Jones (1991)
model, has remained popular, for more than a decade, amongst the models
that deal with aggregate accruals. The structure of an accrual model of the
type Jones (1991) is based on a single linear equation of the form
(1) ACC it = α 0 + α 1Yit + ε it ,
where ACC is an aggregate measure of accruals, Y is a vector with one or
more earnings components (accrual drivers) aiming to explain the dependent
variable (for example, revenue), ε is the residual of the regression, α 0 and α 1
are parameters, t designates the specific time period, and i relates to the firm.
These parameters can be estimated by regressing the model for a given firm
using time-series data or, as it is currently more common, cross-sectionally for
an industry. It is then possible to have an estimate of the expected (normal)
accruals conditional on the realised values of Y at period t:
(2) ÂCC it = αˆ 0 + αˆ 1Yit .
13
An estimate of abnormal accruals ( DAC = ε̂ ) is given by:
(3) ACC it − ÂCC it = εˆ it = DAC it .
If the model is estimated in time-series, DAC is the residual of the regression.
Otherwise, when estimated cross-sectionally by industry, DAC can be seen as
a forecast error. Recent accrual models of this type are, for example, the
Dechow and Dichev (2002) model or the “cash flow model” introduced in Ball
and Shivakumar (2006).
It is widely accepted in the literature that the available aggregate accrual
models do have shortcomings and may not work very well in identifying
earnings management practices. Critics point out that (i) the models are
misspecified and their power is very low (Dechow et al., 1995); (ii) they can be
imprecise in estimating abnormal (discretionary) accruals (Guay et al., 1996);
(iii) a systematic error related to factors like growth, cash flow, leverage and
earnings smoothing is documented in such accruals estimates (Young, 1999);
and (iv) all of models tend to perform poorly in terms of forecasting accuracy
(Thomas and Zhang, 2000).
These limitations explain why most accounting studies tend to simultaneously
use two or more models or sometimes other solutions to detect earnings
management not directly based on accruals.
One of such solutions is the graphical methodology introduced by Burgstahler
and Dichev (1997), that is based on the analysis of the distribution of net
income (change in income) and departs from the assumption that in the
absence of earnings management such a distribution is smooth. Burgstahler
and Dichev present graphical and statistical evidence of an unusually high
frequency of firms in earnings (or earnings changes) intervals immediately to
14
the right of zero, and unusually low frequency of observations to its left, which
the authors interpret as evidence that firms manage earnings to avoid
earnings losses (or decreases). This methodology tends to be more
appropriate to detect manipulation at the centre of the earnings distribution.
Less formal constructs to detect earnings management have also been
developed in the literature. We mention two of them: (i) the existence of
“earnings restatements”, that is, the firms’ re-estimation of previous periods’
earnings following SEC investigation on earnings management practices.
Firms undertaking restatements can thus be characterized as having
manipulated earnings, and the available empirical evidence shows that these
firms have very large accruals in the years of the alleged manipulation (e.g.
Richardson et al., 2002), linking thus this construct to those mentioned above
which were directly based on accruals; (ii) the presence of extraordinary or
unusual items, seen as a measure of (low) quality. Persistent earnings that
reflect the intrinsic value of the firm are usually also referred in the literature as
“permanent” earnings. The component of reported earnings that is not
deemed to be permanent is labelled transitory, and includes, above all, those
items that are classified as extraordinary. Therefore, reported earnings are
seen as possessing lower quality the higher their component of extraordinary
and/or unusual items (e.g. Dechow and Schrand, 2004). This relationship may
explain why earnings variables used in valuation models tend to exclude those
transitory components and why analysts and investors are usually keener to
look at earnings measures above the bottom line. However, this kind of
attitude may also translate into an incentive for managers to classify as
transitory negative earnings components that are intrinsically persistent, or as
15
persistent those which are merely transitory positive ones (Ronen and Sadan,
1975).
A set of solutions aiming at directly assessing earnings quality is based on the
time-series properties of earnings and earnings components (e.g. Schipper
and Vincent, 2003). Amongst the most popular of these components used in
empirical research are: (i) “persistence”, the degree to which earnings
performance persists into the next period. This tends to be measured as the
firm-specific Pearson correlation between current and next period earnings
(Wysocki, 2006); (ii) “predictive ability”, the ability of current earnings to predict
future cash flow from operations. In a similar way as persistence, this tends to
be measured as the firm-specific Pearson correlation between current
earnings and next period cash flow (Wysocki, 2006). The main operational
limitation of these measures arises from the fact that there is a lot of noise in
the correlations when only one period length is taken. Moreover, as Dechow
and Schrand (2004) point out, persistence and predictability by themselves
are not sufficient evidence to indicate earnings quality given that such
characteristics may arise from managers’ manipulation.
2.3 Testing hypotheses
From the discussion above two main ideas are highlighted: the market tends
to penalize companies that stop or reduce dividend payments; quality
earnings tend to reflect firms’ long term underlying economic performance and
to persist through time, unlike those having lower quality, which tend to
reverse. Based on these fundamental ideas, the following hypotheses were
stated:
16
H1: The higher the firm’s earnings quality the higher the probability that it pays
dividends.
H2: Amongst dividend payers, the higher the firm’s earnings quality the higher
the probability that it pays a larger amount of dividends.
H3: The higher the firm’s earnings quality the higher the probability that it may
increase the dividend paid.
3 Reseach design and sample selection
3.1 Econometric basic models
To test the above hypotheses we use three different but close Probit models
of the type:
Ζi = f (EQ _ X , Control variables ) ,
where Ζi is a binary variable defined at a time in the following way:
-
DP, that equals 1 if the firm pays dividend in the period, and 0
otherwise;
-
DY, that equals 1 if the dividend yield is above the median, and 0
otherwise;
-
∆DPS, that equals 1 if the current period change in dividend per share
is positive, and 0 otherwise.
EQ _ X is a set of proxy variables for earnings quality, built up based on the
existing literature and defined in the following sub-section.
The set of control variables follows the literature and includes prior period
reported earnings, dividend per share and dividend yield, change in earnings
per share, capital structure (debt-to-assets ratio), and industry and yearly
control effects. From the documented empirical evidence suggesting a stylised
17
pattern of smoothing and stickiness for firm’s dividend policies (see Allen and
Michaely, 1995, for a survey), we expect the first three variables to have a
positive impact on dividends. Regarding capital structure, we leave the
expected sign as an open issue. In fact, restricting for parsimony the
discussion of the impact of capital structure on dividends within an agency
framework, while Jensen (1986) argues that dividends and debt are
alternative monitoring devices, implying a potential negative relationship
between these variables, Easterbrook (1984) also uses an agency argument
but to predict a positive association. Easterbrook argues that the payment of
dividends can intentionally prompt a greater recourse to market funding (either
through the usage of debt or equity issues) and associated market monitoring,
while at the same can having the effect of increasing debt-to-equity ratios to
counterbalance manager’s risk aversion tendency or to prevent wealth
appropriation by debtholders. Thus, the expected impact of capital structure
can either be negative or positive.
3.2 Earnings quality measurement
Given the unavailability of a true measure of earnings quality, we adopt four
different types of proxies:
- EQ _ DAC : based on discretionary accruals (DAC) and using four
different accrual models (see Exhibit 3) taken from those most
commonly used in accounting empirical research. The Jones (1991),
the “cash flow” and the Dechow and Dichev (2002) models are
sufficiently known to deserve any further comment. The “extended
Jones (1991)” model is proposed in Moreira (2002) and, more
18
recently, in Ball and Shivakumar (2006), and intends to control for the
asymmetric impact on accruals arising from conservative accounting.
In all models the dependent variable is total accruals less
depreciation, following the evidence in the literature that the
depreciation expense only tends to add noise to the estimation of
discretionary accruals (e.g. Peasnell et al., 2000).
[Exhibit 3]
- EQ _ EXTRA : it is based on the existence of transitory components in
earnings. It is defined as a dummy variable that equals -1 if the firm
has extraordinary items (Compustat #48) in the period, and 0
otherwise.
- EQ _ RESTAT : the restatement of the earnings number is the outcome
of manipulation detected by SEC. Such a manipulation is impounded
in the model through a dummy variable that equals -1 if the firm has
restated its earnings in the period (i.e. #172 - #177 ≠ 0), and 0
otherwise.
. EQ _ PER : this earnings quality proxy is based on the persistence of
reported earnings. It is defined as the value of the Pearson correlation
between current and next period operating income (#178), both
deflated by prior period total assets, estimated using a 5 year period
roll-over.
All these variables are defined in a way such that an observed increase will
correspond to a higher level of earnings quality. Thus, we predict their
coefficients in the model to be positive.
19
3.3 Graphical analysis
As mentioned above, Burgstahler and Dichev (1997) use a graphical analysis
based on the income distribution to detect signs of earnings management
around the centre of such a distribution. They document evidence suggesting
that the quality of slightly positive earnings is questionable and seems to
reflect firms’ manipulation to avoid small losses. In line with this strand of
research, we use a similar methodology to undertake a robustness test on
hypotheses 1 and 3.
To take into account that the sub-samples of firm-observations classified
under each category c [c = 1( dividend payer; positive change in dividend per
share); 2 (non-dividend payer; null or negative change in dividend per share)]
are unbalanced, we define for each earnings interval i the proportion (ppi) of
the number of observations belonging to such a category over the total
number of observations (T) in the sub-sample. Formally, such proportions are
defined as:
pp1i =
n1i
n1T
and
pp2i =
n 2i
,
n 2T
where n1i is the number of observations belonging to category 1 in interval i
and n1T is the total number of observations in the sub-sample. The graphical
analysis is performed for the difference (diff) between the proportions of each
interval:
diff i = pp1i − pp 2 i .
3.4 Sample selection and descriptive statistics
20
We use data from US listed firms for the period 1987/2003. From the 2005
Compustat disks we collect all non-financial companies, except utilities,
available in Primary, Secondary and Tertiary, Full Coverage and Research
Annual Industrial Files. After lagging and leading one period, all missing
observations in relevant variables are deleted. Observations having negative
book value of equity and earnings losses are deemed as missing. To avoid
potential biases arising from the existence of outliers, the upper and lower 1%
of earnings per share for each year is also considered as missing. The basic
sample we obtain has 44,986 observations, from which 39% are dividend
payers. Deleting observations with the persistence variable missing, the
sample size is reduced to 38,684 observations, with 43% of dividend payers.
Table 1 explains in detail the sample selection process.
[TABLE 1]
In Table 2, Panel A, we tabulate some descriptive statistics for the global
sample and for the dividend payers’ sub-sample. The dividend per share
(DIVPS) and dividend yield (DIVYL) variables show similar distributions, which
are slightly skewed to the right. The graph in Panel C suggests that these
variables experience a similar evolution throughout the period of the sample.
The capital structure (STRUCT) of dividend payers is characterized by higher
debt ratios than their counterpart of non-payers, while the earnings variable
(EARN) shows a statistically higher mean for non-dividend payers. However,
for cash-flow from operations (CFO), the opposite happens. This evidence
implies that non-dividend payers do have higher accruals, consistent with the
above discussion that the level of accruals is expected to be negatively related
21
to the quality of earnings. Panel D displays more precise and supportive
evidence on this issue, showing that dividend payers tend to have lower
discretionary accruals (higher earnings quality) and higher cash-flow from
operations, the difference being statistically significant. This preliminary
evidence thus seems to be consistent with our first hypothesis.
[TABLE 2]
Panel B shows the correlations amongst the above discussed variables. They
tend to be quite low, except for that of DIVPS vs. DIVYL, which is not
unexpected given the evolution of both variables as depicted in Panel C.
4 Empirical results
4.1 Probit analysis
Given the nature of the hypotheses to be tested, we adopted Probit models as
the appropriated econometric tool. Table 3 displays the results of the four
models that test the first hypothesis. The models have all a similar structure
and the same dependent variable (DP), a dummy that takes two categories: 1,
if the firm is a dividend payer; 0, otherwise. However, each model uses a
specific proxy for earnings quality (EQ_X), the attribute that differentiates the
models.
[TABLE 3]
In all four models the coefficients on the earnings quality variable have the
right sign and are highly significant, consistent with the expectation that the
22
probability of a company being a dividend payer increases with the quality of
its earnings. This result therefore strongly supports our first hypothesis. It
suggests that companies are more confident to pay a positive amount of
dividends when their earnings tend to reflect the long-run underlying economic
performance, because they have better conditions to keep paying cash to
shareholders in the future and thus avoid the market penalization associated
with a dividend cut or omission (Christie, 1994).
The coefficients on the control variables also follow our expectations:
companies with higher prior period earnings and those paying dividends in the
previous year are significantly more likely to be dividend payers. Moreover,
the coefficient on STRUCT, not previously predicted, is also significantly
positive consistent with Easterbrook (1984) expectation that dividend payment
may go along with an increase in firms’ debt. These results are in general
accordance with existing literature.
[TABLE 4]
Table 4 displays the results for the test of our second hypothesis. As in Smith
and Watts (1992), we use here dividend yield (or dividend-to-price ratio) as a
measure of dividend policy7. The overall evidence on the impact of earnings
quality is according to our expectations. The coefficients on DAC_X in models
1 and 3 are positive and heavily significant, supporting our prediction that
7
The alternative usage of dividend per share or dividends-to-earnings ratio as the dependent
variable proved to yield much noisier results. We did not, however, find it meaningful to rank
firms on the basis of dividends per share (as we did for dividend yield) given that this is much
contingent on the number of shares issued, the existence of stock splits, stock dividends, and
the like. In addition, the usage of a dividends-to-earnings ratio provided very volatile dividend
policy proxies given the known pattern of dividend stickiness resulting in rapidly changing
short-term payout ratios as earnings experience some degree of volatility. Finally, computing
long-term estimates for payout ratios would either require larger time-series or significantly
reduce the number of observations. Although we acknowledge that further research should
23
companies with higher earnings quality tend to pay larger dividends. In model
2, the coefficient on the proxy variable for earnings quality (EQ_EXTRA) has
the right sign but is significant at eleven percent only. In model 4, the
coefficient is negative but statistically insignificant. A tentative explanation for
this unexpected result may be the argument discussed above that the
persistence of earnings in itself is not always a good proxy for the quality of
earnings. In this specific case, almost half of the observations in the sample
show negative “persistence” (Pearson correlation between current period and
next period operating income), and we interpret this fact as the outcome of the
estimation noise in the variable.
The coefficients on the control variables are all positive and highly significant,
in general accordance with our expectations. The coefficient on STRUCT is
positive, as in Table 3. Consistent with the evidence in the literature on the
stability of firms’ dividend policies, our results show that the higher the
dividend yield of prior period the higher tends to be that of current period.
In sum, the results in Table 4 are supportive of our second hypothesis.
[TABLE 5]
Out third hypothesis is the prediction that companies with higher earnings
quality will be more likely to increase dividends. We use the (change in)
dividend per share (DIVPS) as dependent variable, instead of the dividend
yield, because changes in the former are more likely to reflect deliberate
changes in dividend policy than changes in dividend yield which can be the
dwell with this issue, we did not explore further, given the somewhat exploratory nature of our
paper, the question of the definition of an accounting-based long-term dividend policy proxy.
24
mere outcome of market behaviour. Table 5 displays evidence supportive of
this prediction.
In all four models, the coefficient on the earnings quality variable is positive,
as expected, and in three of them (models 1 to 3) they are highly significant.
The exception is again model 4, where the coefficient on EQ_PER is not
significant. As in Table 4, the proxy variable based on persistence
underperforms all other proxies. Nevertheless, the overall evidence is highly
supportive of our third hypothesis, showing that companies increasing their
dividends tend to have higher earnings quality.
The level of prior period earnings per share (EPS) and the current change in
this variable are both positively related to the probability that a company may
increase its DIVPS. A very intuitive result is that the higher prior period DIVPS
the lower the probability of a dividend increase. As before, we did not set an a
priori expectation for the coefficient on STRUCT because there is no clear
guidance in the literature for the relationship between changes in DIVPS and
companies’ capital structure. In contrast with previous results, we now find a
consistently negative and significant sign for this variable in all models.
In sum, the results in Tables 3 to 5 strongly support our testing hypotheses.
We particularly highlight the consistently positive relationship between
earnings quality and dividend payments. At a time when much of the existing
evidence suggests that companies “cook” their income numbers, our results
may be an important insight that can help users of financial information to
draw inferences about companies’ earnings quality based on their observed
dividend policies.
25
4.2 Graphical analysis
The graphical analysis we undertake is based on the empirical evidence found
in the literature that small positive earnings are of lower quality (e.g.
Burgstahler and Dichev, 1997). This analysis can be seen as complementary
to the econometric one discussed in the previous subsection. The results
depicted in Exhibit 1 seem to corroborate our previous evidence that the
payment of dividends has a positive relationship with the quality of earnings.
For small positive deflated earnings (return on assets), assumed to be of
lower quality (e.g. Burgstahler and Dichev, 1997),8 the proportion of nonpayers is higher than that of payers. As one moves to the right of the
distribution, and supposedly the quality of earnings is higher than before, the
proportion of payers becomes higher than that of non-payers. This evidence
seems thus to be consistent with our expectation and the econometric results
discussed above.
[Exhibit 1]
However, this interpretation may be criticised because it does not consider
that firms having lower earnings do have a smaller probability of paying
dividends, as the results in Table 3 show. Such criticism is justifiable, and
graphically we do not have a way of controlling for the earnings size impact.
Nevertheless, the picture does show other information that may help to
disentangle this issue. For returns on assets higher than 0.135 the proportion
8 Untabulated results available on request show that the proportion of total accruals in
earnings is at its highest for firms having deflated earnings (return on assets) lower or equal to
0.025. Given the evidence available in the literature that there is a negative relationship
26
of non-payers becomes consistently higher than that of payers. These returns
are exceptionally high and hardly can be deemed as persistent. This means
that such high earnings are not indicative of companies’ future performance
and thus, given the above discussion on the meaning of “earnings quality”, are
of low quality. Therefore, the results are consistent with our expectation, at
least in this part of the distribution: lower quality earnings go along with less
dividend payers.
To sum up, although the graphical evidence does not prove unequivocally the
expected relationship between earnings quality and dividend payment, it is
consistent with the existence of such relationship, as predicted in our first
hypothesis.
[Exhibit 2]
In Exhibit 2 we look for the relationship between earnings quality and positive
changes in dividends. Given that the market penalizes companies that reduce
or cease to pay dividends, managers in these firms ought to be very careful in
deciding whether to increase payments to shareholders or not. We argued
above that managers’ perception of earnings quality is a driving force behind
that decision. The graph depicts the difference in the proportions of
companies that increased their dividends and those who did not. The
evidence suggests that dividend increases are more frequent in companies
with higher earnings quality, this is, earnings that are neither too low nor too
high. For companies with small earnings (return on assets), assumed to be of
between the amount of accruals and earnings quality (e.g. DeAngelo, 1986), these results
seem to be another piece of evidence supporting the lower quality of small earnings.
27
lower quality, the difference in proportions is heavily negative and suggests
the existence of the discussed relationship.
4.3 Sensitivity analyses
Although not discussed in the paper, we also did a large number of sensitivity
tests to check the robustness of the results. The main ones are referred
below.
The results discussed so far for the EQ_DAC variable are based on
discretionary accruals (DAC) estimated with the Dechow and Dichev (2002)
model. We re-performed the analysis using once at a time DAC estimates
obtained from the three remaining accrual models mentioned in Exhibit 3. The
overall results are qualitatively similar to those reported above.
The EQ_PER was also estimated using current Pearson correlation between
current and next period operating income, instead of a 5 year period roll-over
reported in the paper. No meaningful differences have been detected.
The samples used are quite large and thus the models tend to be unaffected
by potential outliers the data may contain. Nevertheless, we replicated the
analysis using trimmed and untrimmed samples and the results are not
qualitatively different from those reported.
Moreover, the analyses were repeated using balanced samples matched by
the level of deflated earnings, year and industry. The overall results are
generally unaffected and qualitatively similar to those tabulated.
Finally, given the negative relationship between total accruals and earnings
quality mentioned in the literature, we also replicated the analyses using the
28
proportion of accruals in earnings as a proxy for such a quality. The results
are completely consistent with those reported in Tables 3 to 5.
5 Summary and discussion of findings
This paper presents empirical evidence on the (missing) link between
earnings quality and dividends.
We first test whether the decision to pay dividends is related to firms’ earnings
quality. The evidence we report is strongly supportive of this hypothesis as our
results show that firms with higher earnings quality do have a significantly
higher probability of being dividend payers.
We also analyse whether the quality of earnings is related to the relative
amount of dividends paid. Using the dividend yield as a proxy, the evidence
also suggests that such a relationship exists whereby firms with higher
earnings quality tend to have a higher probability of setting more generous
dividend policies.
Finally, we also test whether earnings quality affects the decision of increasing
the amount of the dividend paid. Once again, the results were generally
supportive of our hypothesis. Firms with higher earnings quality were
observed to have a higher probability of increasing the dividend payment.
These results are robust to the use of different earnings quality measures and
the control of other determinants of dividend policy which have been
acknowledged in the literature. Additionally, a graphical analysis based on
existing studies suggesting that the quality of earnings is different throughout
the distribution of deflated earnings (return on assets), is also consistent with
a positive relationships between earnings quality and dividends.
29
To our knowledge, this paper offers the first large-scale (around 40,000 firmyear observations) empirical evidence in support of the hypothesis that
dividends play an important role in conveying information on the quality of
earnings (measured using several alternative proxies) or, looking this
relationship the other way round, on the level of earnings management
undertaken by companies.
Our results thus breed significant and useful insights into the almost 50-year
old debate on dividends. Although we do not claim that the earnings quality
issue is the unique influence on dividend policy, the evidence we report in this
paper strongly suggests that dividends can act as a credible signal of earnings
quality, with companies unengaged in earnings management showing more
generous dividend policies, a possibility which has been much overlooked by
most of the existing literature. Needless to say, this is a promising field for
future research and one that will hopefully bring ever closer links between
Finance and Accounting research.
30
6 References
Agrawal, A., and N. Jarayaman, 1994, The dividend policies of all-equity firms:
a direct test of the free cash flow theory, Managerial Decision Economics
15 (2), 139-148.
Allen, F, and R. Michaely, 1995, Dividend policy, in R. Jarrow et al (eds.),
Finance, Vol. 9 of Handbooks in Operations Research and Management
Science, North Holland.
Baker, M., and J. Wurgler, 2004, A catering theory of dividends, Journal of
Finance 59 (3), 1125-1165.
Ball, R. and L. Shivakumar, 2006, The role of accruals in asymmetrically timely
gain and loss recognition, Journal of Accounting Research, vol. 44, n. 2,
pp. 207-242.
Barth, M., W. Beaver, J. Hand, and W. Lansdman 1999, Accruals, Cash flow,
and Equity Values, Review of Accounting Studies, v.3, pp. 205-229.
Beatty, A., B. Ke and K. Petroni. 2002. Earnings Management to Avoid
Earnings Declines across Publicly and Privately Held Banks, The
Accounting Review, vol.77, no.3, July, pp. 547-570.
Benartzi, S., R. Michaely, and R. Thaler, 1997, Do changes in dividends signal
the future or the past?, Journal of Finance 52 (3), 1007-1034.
Black, F., 1976, The dividend puzzle, Journal of Portfolio Management 2, 5-8.
Black, F., and M. Scholes, 1974, The effects of dividend yield and dividend
policy on common stock prices and returns, Journal of Financial
Economics 1 (1), 1-22.
Breeden, R., 2003, Restoring Trust: Report to The Hon. Jed S. Rakoff, The
United States District Court for the Southern District of New York, on
Corporate Governance for the Future of MCI, Inc.
http://news.findlaw.com/hdocs/docs/worldcom/corpgov82603rpt.pdf.
Brennan, M., 1970, Taxes, market valuation and financial policy, National Tax
Journal 23, 417-429.
Burgstahler, D. and I. Dichev. 1997, Earnings management to avoid earnings
decreases and losses, Journal of Accounting and Economics, vol.24, pp.
99-126.
Chan, K., L. Chan, N. Jegadeesh and J. Lakonishok, 2006, Earnings Quality
and Stock Returns, Journal of Business, vol. 79, n. 3, pp. 1041-1082.
Chen, S., T. Shevlin, and Y. Tong, 2006, Does the pricing of financial reporting
quality change around dividend changes?, Journal of Accounting
Research, vol. 45 (1), pp. 1-40.
Christie, W., 1994, Are dividend omissions truly the cruelest cut of all?,
Journal of Financial and Quantitative Analysis 29 (3), 459-480.
Crutchley, C., and R. Hansen, 1989, A test of the agency theory of managerial
ownership, corporate leverage and corporate dividends, Financial
Management 18, 36-76.
31
DeAngelo, H., L. DeAngelo, and D. Skinner, 1996, Reversal of fortune
dividend signalling and and the disappearance of of sustained earnings
growth, Journal of Financial Economics 40 (3), 341-471.
DeAngelo, L. 1986, Accounting numbers as market valuation substitutes: A study
of manage-ment buyouts of public stockholders, The Accounting Review,
v.61, n.3, pp. 400-420.
Dechow, P. and C. Schrand, 2004, Earnings Quality, The Research
Foundation of CFA Institute, ISBN 0-943205-68-9.
Dechow, P. and I. Dichev, 2002, The quality of accruals and earnings: the role
of accrual estimation errors, The Accounting Review, vol. 77 (Supplement),
pp. 35-59.
Dechow, R. Sloan, and A. Sweeney 1995, Detecting earnings management,
The Accounting Review, v. 70, n. 2, April, pp. 193-225.
DeFond, M. and J. Jiambalvo. 1994, Debt covenant violation and manipulation
of accruals, Journal of Accounting and Economics, v. 17, pp. 145-176.
Dhaliwal, Dan, M. Erickson, and R. Trezevant, 1999, A test of the theory of tax
clienteles for dividend policies, National Tax Journal 52, 179–194.
Dong, M., C. Robinson, and C. Veld,T, 2005, Why individual investors want
dividends, Journal of Corporate Finance 12, 121– 158.
Easterbrook, F., 1984, Two agency-cost explanations of dividends, American
Economic Review 74, 650-659.
Fama, E., and K. French, 2001, disappearing dividends: changing firm
characteristics or lower propensity to pay?, Journal of Financial Economics
60 (1), 3-43.
Farinha, J., 2003, Dividend policy, corporate governance and the managerial
entrenchment hypothesis. An empirical analysis, Journal of Business
Finance and Accounting 30 (9&10), 1173-1209.
Gonedes, N., 1978, Corporate signalling, external accounting and capital
market equilibrium: evidence on dividends, income and extraordinary
items, Journal of Accounting Research 16 (1), 26-79.
Graham, J., and A. Kumar, 2006, Do dividend clienteles exist? Evidence on
dividend preferences of retail investors, Journal of Finance 51 (3), 13051336.
Guay, W., S. Kothari, and R. Watts 1996, A market-based evaluation of
discretionary accrual models, Journal of Accounting Research, v. 34,
Supplement, pp. 83-105.
Healy, P. 1985, The Effect of Bonus Schemes on Accounting Decisions,
Journal of Accounting and Economics, v. 7, pp. 85-107.
Healy, P. and J. Wahlen, 1999, A review of the earnings management
literature and its implications for standard settings, Accounting Horizons,
vol. 13, no. 4, pp. 365-383.
Healy, P., and K. Palepu, 1988, Earnings information conveyed by dividend
initiations and omissions, Journal of Financial Economics 21, 149-175.
Healy, P., and K. Palepu, 1990, Effectiveness of accounting-based dividend
covenants, Journal of Accounting and Economics 12 (1-3), 97-124.
32
Howe, K., J. He, and G. Kao, 1992, One-time cash-flow announcements and
free cash-flow theory: share repurchases and special dividends, Journal of
Finance 47 (5), 1963-1975.
http://ssrn.com/abstract=484542
Jensen, G., D. Solberg, and T. Zorn, 1992, Simultaneous determination of
insider ownership, debt and dividend policies, Journal of Financial and
Quantitative Analysis 27 (2), 247-263
Jones, J., 1991, Earnings management during import relief investigations,
Journal of Accounting Research, v. 29, n. 2, 193-228.
Kalay, A., and R. Michaely, 2000, Dividends and taxes: a re-examination,
Financial Management 29 (2), 55-75.
Lambert, R., W. Lanen, and D. Larcker, 1989, Executive stock option plans
and corporate dividend policy, Journal of Financial and Quantitative
Analysis 24 (4), 409-425.
Lie, E., 2005, Operating performance following dividend decreases and
omissions, Journal of Corporate Finance 12 (1), 27-53.
Lintner, J., 1956, Distribution of incomes of corporations among dividends,
retained earnings and taxes, American Economic Review 46 (2), 97-113.
Litzenberger, R., and K. Ramaswamy, 1979, The effect of personal taxes and
dividends on capital asset prices: theory and empirical evidence, Journal of
Financial Economics 7 (2), 163-195.
Mikhail, M., B. Walter, and R. Willis, 2003, Reactions to dividend changes
conditional on earnings quality, Journal of Accounting, Auditing and
Finance, vol. 18 (1), pp. 121-151
Miller, M., and F. Modigliani, 1961, Dividend policy, growth and the valuation
of shares, Journal of Business 34 (4), 411-433.
Miller, M., and K. Rock, 1985, Dividend policy under asymmetric information,
Journal of Finance 40, 1031-1051.
Miller, N., and M. Scholes, 1982, Dividends and taxes: some empirical
evidence, Journal of Political Economy 90 (6), 1118-1141.
Moreira, J. A. C. 2002. Essays in links between Firm Value and Earnings
Components Under Conservative Accounting. PhD dissertation, Lancaster
University, January.
Nissim, D., 2003, Discussion – Reaction to dividend changes conditional on
earnings quality, Journal of Acounting, Auditing and Finance, vol. 18 (1),
pp. 153-161.
Nissim, D., and A. Ziv, 2001, Dividend changes and future profitability, Journal
of Finance 56 (6), 2111-2133.
Ofer, A., and D. Siegel, 1987, Corporate financial policy, information and
market expectations: an empirical investigation of dividends, Journal of
Finance 42 (4), 889-911.
Peasnell, K., P. Pope, and S. Young, 2000, Detecting earnings management
using cross-sectional abnormal accrual models, Accounting and Business
Research, v.30, no.4, pp. 313-326.
Richardson, S., I. Tuna and M. Wu, 2002, Predicting earnings management:
33
the case of earnings restatements, Unpublished working paper, October
version.
Ronen, J. and S. Sadan, 1975, Classificatory smoothing: alternative income
models, Journal of Accounting Research, vol. 13, Spring, pp.133-149.
Schipper, K. and L. Vincent, 2003, Earnings Quality, Accounting Horizons,
Supplement, pp. 97-110.
Schipper, K., 1989, Commentary on Earnings Management”, Accounting
Horizons, Vol. 3, N.º 4, pp. 91-102.
Shefrin, H., and M. Statman, 1984, Explaining investor preferences for cash
dividends, Journal of Financial Economics 13, 253-282.
Skinner, D., 2004, What do dividends tell us about earnings quality, Working
Paper, University of Michigan Business School.
Smith, C., and R. Watts, 1992, The investment opportunity set and corporate
financing, dividend and compensation policies, Journal of Financial
Economics, vol.32, pp. 262-292.
Sweeney, A., 1994, Debt Covenant Violations and Managers’ Accounting
Responses. Journal of Accounting and Economics, vol. 17, pp.281-308.
Thomas, J. and X. Zhang, 2000, Identifying unexpected accruals: a comparison
of current approaches, Journal Accounting and Public Policy, vol.19, pp.
347-376.
Watts, R., 1973, The information content of dividends, Journal of Business 46
(2), 191-211.
Wysocki, P., 2006, Assessing Earnings and Accruals Quality: U.S. and
International Evidence, Working Paper, MIT Sloan School of Management.
Young, S., 1999, Systematic Measurement Error in the Estimation of
Discretionary Accruals: An Evaluation of Alternative Modelling Procedures,
Journal of Business Finance & Accounting, v. 26, n.7-8, Sept./Oct., pp.
833-862.
34
Table 1: Sample selection. Period 1987-2003.
N. firmyears
Description
COMPUSTAT Primary, Secondary, and Tertiary, Full
Coverage, and Research Annual Industrial Files, 2005 disks
(1987-2003)
After deleting financial companies
After lagging and leading one period, and deleting missing
observations in all relevant variables
After deleting negative book value of equity and losses
After trimming dividend per share 1% top and bottom, by year
· dividend payers (39%)
· not payers (61%)
After deleting missing “persistence variable” observations
· dividend payers (43%)
· not payers (57%)
35
379,168
298,435
77,128
45,431
44,986
17,610
27,376
38,684
16,550
22,134
Table 2: Descriptive statistics. Pair-wise correlations.
Panel A: Descriptive statistics
Variables
Sample
Mean
Std
DIVPS t
Div. Payers (17,610 obs.)
0.688
0.608
0.500
DIVYL t
Div. Payers
2.907
5.337
2.136
EARN t
Global (44,986 obs.)
0.079
0.073
0.059
Div. Payers
0.073
0.057
0.059
Global
0.107
0.104
0.101
Div. Payers
0.116
0.080
0.107
Global
0.502
0.212
0.521
Div. Payers
0.533
0.189
0.560
CFO t
STRUCT t
Median
Variables’ definition: DIVPSt is current dividend per share (Compustat #26), measured in dollars; DIVYLt is current
dividend yield (#26/#199) in percentage; EARNt is earnings before extraordinary items adjusted for common shares
(#20) deflated by lagged total assets (#6); CFO is current cash flow from operating activities (#308) deflated by
lagged total assets; STRUCTt is firm capital structure defined as total debt (#6 minus book value of equity, #60) over
total assets.
Panel B: Global sample correlations: Pearson (above) / Spearman (below)
Variable
DIVPS t
DIVPS t
EARN t
-0.093
CFO t
STRUCT t
DIVYL t
0.021
0.200
0.467
0.423
-0.375
-0.059
-0.166
-0.007
EARN t
-0.032
CFO t
0.064
0.447
STRUCT t
0.163
-0.414
-0.191
DIVYL t
0.967
-0.054
0.045
0.077
0.147
Variables’ definition is per Panel A. Correlations are different from zero at less than 5%, except the one in bold
numbers.
36
Panel C: Dividend payment by year (17,610 firm-years)
2
0,45
1,8
0,4
1,6
0,35
1,4
0,3
1,2
0,25
1
0,2
0,8
0,15
0,6
0,1
0,4
0,05
0,2
0
0
1988
1989
1990
1991
1992
1993
1994
1995
DivYl
1996
1997
1998
1999
2000
2001
2002
Div_ps
Note: mean dividend yield in percentage (right axe); mean dividend per share measured in dollars.
Panel D: Univariate tests for the global sample
EQ_DACt
Class
CFOt
Mean
Median
Mean
Median
Dividend payer
-0.015
-0.010
0.116
0.107
Non-payer
-0.046
-0.026
0.102
0.096
< .0001
< .0001
< .0001
< .0001
Pr [P]=[NP]
Notes:
The table reports the median and median of EQ_DAC, an earnings quality measure based on the inverse of
discretionary accrual estimates (DAC) calculated with the Dechow and Dichev (2002) model, and of cash flow from
operations (CFO, #308), both deflated by lagged total assets. Pr is the probability associated to the test for equality of
mean (SAS TTEST) and median (SAS WILCOXON TEST).
37
Table 3: Earnings quality and the decision of paying dividends
Probit Model
DPt = α 0 + α1EQ _ X t + α 2EARN _ Dt −1 + α 3STRUCTt + α 4DIVPSt −1 + α j
Model 1
Expected
sign
Variables
EQ_DAC t
+
Model 2
∑ IND + α ∑ YEAR + e
l
Model 3
t
Model 4
Coeff.
Coeff.
Coeff.
Coeff.
[P-Value]
[P-Value]
[P-Value]
[P-Value]
2.030
[<.0001]
EQ_EXTRA t
+
0.071
[<.0001]
EQ_RESTAT t
+
0.092
[0.0003]
EQ_PER t
+
0.030
[0.0519]
EARN_D t-1
0.424
0.515
0.525
0.384
[<.0001]
[<.0001]
[<.0001]
[<.0001]
0.483
0.697
0.685
0.392
[<.0001]
[<.0001]
[<.0001]
[<.0001]
0.842
0.863
0.862
0.965
[<.0001]
[<.0001]
[<.0001]
[<.0001]
Dividend payers
17,610
17,610
17,610
16,550
Non-Payers
27,376
27,376
27,376
22,134
+
STRUCT t
?
DIVPS t-1
+
N. Observations:
Notes:
1)
2)
3)
The dependent variable (DP) is a dummy variable that takes value 1 if the firm pays dividend in the period, 0
otherwise. EQ_X is a set of variables used as proxies for earnings quality and defined in a way that an increase
corresponds to a higher level of such a quality. EQ_DAC is the inverse of discretionary accrual estimates
calculated using the Dechow and Dichev (2002) model; EQ_EXTRA is a dummy variable that takes value -1 if
the firm has extraordinary items (#48) in the period, zero otherwise; EQ_RESTAT is a dummy variable that
takes value -1 if the firm has restated its earnings in the period (i.e. #172-#177≠0), zero otherwise; EQ_PER is
the value of the Pearson correlation between current and next period operating income (#178), both deflated by
prior period total assets, estimated using a 5 year period roll-over. EARN_D is a dummy variable that takes
value 1 if prior period deflated earnings are above the median in the yearly distribution, 0 otherwise. Other
variables’ definition are per Table 2, panel A;
The results have been controlled for industry and yearly intercept effects using dummy variables. For the sake of
parsimony, these coefficients and the intercept one are not tabulated;
The subscript for firm has been dropped out in the model.
38
Table 4: Earnings quality and the magnitude of dividend yield
Probit Model
DYt = α 0 + α1EQ _ X t + α 2EARN _ Dt −1 + α 3STRUCTt + α 4DIVYLt −1 + α j
Variables
EQ_DAC t
Expected
sign
+
∑ IND + α ∑ YEAR + e
l
t
Model 1
Model 2
Model 3
Model 4
Coeff.
Coeff.
Coeff.
Coeff.
[P-Value]
[P-Value]
[P-Value]
[P-Value]
1.889
[<.0001]
EQ_EXTRA t
+
0.043
[0.1141]
EQ_RESTAT t
+
0.168
[<.0001]
EQ_PER t
+
-0.038
[0.1566]
EARN_D t-1
0.159
0.203
0.205
0.247
[<.0001]
[<.0001]
[<.0001]
[<.0001]
0.420
0.607
0.600
0.772
[<.0001]
[<.0001]
[<.0001]
[<.0001]
0.109
0.110
0.110
0.129
[<.0001]
[<.0001]
[<.0001]
[<.0001]
High dividend p.s.
8,647
8,647
8,647
8,150
Low dividend p.s.
8,655
8,655
8,655
8,161
STRUCT t
+
?
DIVYL t-1
+
N. Observations:
Notes:
1)
2)
3)
The dependent variable (DY) is a dummy variable that takes value 1 if the dividend yield (DIVYL) is above
the median, 0 otherwise. EQ_X is a set of variables used as proxies for earnings quality and defined in a
way that an increase corresponds to a higher level of such a quality. EQ_DAC is the inverse of
discretionary accrual estimates calculated using the Dechow and Dichev (2002) model; EQ_EXTRA is a
dummy variable that takes value -1 if the firm has extraordinary items (#48) in the period, zero otherwise;
EQ_RESTAT is a dummy variable that takes value -1 if the firm has restated its earnings in the period (i.e.
#172-#177≠0), zero otherwise; EQ_PER is the value of the Pearson correlation between current and next
period operating income (#178), both deflated by prior period total assets, estimated using a 5 year period
roll-over. EARN_D is a dummy variable that takes value 1 if prior period deflated earnings are above the
median, 0 otherwise. DIVYL is the dividend yield measured as a percentage. Other variables’ definition are
per Table 2, panel A;
The results have been controlled for industry and yearly intercept effects using dummy variables. For the
sake of parsimony, these coefficients and the intercept ones are not tabulated;
The subscript for firm has been dropped out in the model.
39
Table 5: Earnings quality and the change in the dividend per share paid
Probit Model
∆DPSt = α 0 + α1EQ _ X t + α 2EPSt −1 + α 3 ∆EPSt + α 4STRUCTt + α 5DIVPSt −1 + α j
Expected
sign
Variables
EQ_DAC t
+
∑ IND + α ∑ YEAR + e
l
Model 1
Model 2
Model 3
Model 4
Coeff.
Coeff.
Coeff.
Coeff.
[P-Value]
[P-Value]
[P-Value]
[P-Value]
0.619
[0.0008]
EQ_EXTRA t
+
0.074
[0.0042]
EQ_RESTAT t
+
0.117
[0.0027]
EQ_PER t
+
0.022
[0.3007]
EPS t-1
0.163
0.162
0.159
0.167
[<.0001]
[<.0001]
[<.0001]
[<.0001]
0.137
0.134
0.132
0.143
[<.0001]
[<.0001]
[<.0001]
[<.0001]
-0.247
-0.212
-0.249
-0.260
[<.0001]
[<.0001]
[<.0001]
[<.0001]
-0.294
-0.291
-0.258
-0.264
[<.0001]
[<.0001]
[<.0001]
[<.0001]
High ∆ DPS
9,457
9,457
9,457
8,716
Low ∆ DPS
7,912
7,912
7,912
7,621
+
∆EPS t
+
STRUCT t
?
DIVPS t-1
-
N. Observations:
Notes:
1)
2)
3)
The dependent variable (∆DPS) is a dummy variable that takes value 1 if the current period change in
dividend per share is positive, 0 otherwise; EPS is earnings per share of prior period (#58) and ∆EPS is
current period change in earnings per share. Other variables’ definitions are per Table 2, panel A, and
Table 3;
The results have been controlled for industry and yearly intercept effects using dummy variables. For the
sake of parsimony, these coefficients and the intercept ones are not tabulated;
The subscript for firm has been dropped out in the model.
40
t
Exhibit 1: Graphical distribution of differences in the proportions of
dividend payers/non-payers per interval of deflated earnings
0,01
0,005
0
0
0,025
0,05
0,075
0,1
0,125
0,15
0,175
0,2
0,225
0,25
0,275
0,3
-0,005
-0,01
Dif. Payer-NonPayer
Polinómio (Dif. Payer-NonPayer)
This figure shows the difference between the proportions of dividend payers and non-payers
per interval of return on assets (deflated earnings before extraordinary items, #20). The first
interval at the right of zero is [0; 0.0025[, and earnings is deflated by lagged total assets (#6).
The vertical axis represents the difference in proportions, and the proportion for each class is
defined as the number of payers/non-payers over the total number of payers/non-payers in
the sample. The trend line is of the type polynomial of fifth order. Period 1998-2002, 44,986
firm-years.
41
Exhibit 2: Graphical distribution of differences in the proportions of firms
having positive/null-negative changes in dividend per share
per interval of deflated earnings
0,01
0,005
0
0
0,025
0,05
0,075
0,1
0,125
0,15
0,175
0,2
0,225
0,25
0,275
-0,005
-0,01
-0,015
Dif. Positive-Negative change in DPS
Polinómio (Dif. Positive-Negative change in DPS)
This figure shows the difference between the proportions of firms having positive/null-negative
changes in dividend per share per interval of return on assets (deflated earnings before
extraordinary items, #20). The first interval at the right of zero is [0; 0.0025[, and earnings is
deflated by lagged total assets (#6). The vertical axis represents the difference in proportions,
and the proportion for each class is defined as the number of firms having positive/nullnegative changes in dividend per share over the total number of firms having positive/nullnegative changes in the in sample. The trend line is of the type polynomial of fourth order.
Period 1998-2002, 17,369 firm-years.
42
Exhibit 3: Accrual models’ specification.
Jones (1991) model
[
]
ACC t = γ 0 + γ 1 1
+ γ 2 ∆REVt + ε t
defl
Extended Jones (1991) model
[
]
ACC t = γ 0 + γ 1 1
+ γ 2 ∆REVt + γ 3 D1 + γ 4 RETt + γ 5 D1 _ RETt + ξ t
defl
Cash Flow model
ACC t = γ 0 + γ 1CFOt + υ t
Dechow and Dichev (2002) model
ACC t = γ 0 + γ 1CFOt −1 + γ 2CFOt + γ 3CFOt +1 + µ t
Notes:
Variables’ definitions: ACC is a measure of total accruals (#237-#308) and is defined as
TACC_D (total accruals less depreciation); defl is the deflator, i.e. the lagged total assets;
RET are market returns estimated using Compustat fiscal-year-end closing price (#199) and
dividends per share (#26); ∆REV is change in total sales (#12); D1 is a dummy variable taking
value one if RET<0, zero otherwise; CFO is cash flow from operating activities (#308). All
variables are deflated. The models have been estimated by year and industry. The industry
structure adopted is the one proposed in Barth et al. (1999).
43