The Relationship Between Information Asymmetry and Dividend PolicyPap

Finance and Economics Discussion Series
Divisions of Research & Statistics and Monetary Affairs

Federal Reserve Board, Washington, D.C.
The Relationship Between Information Asymmetry and Dividend

Policy
Cindy M. Vojtech
2012-13
NOTE: Staff working papers in the Finance and Economics Discussion Series (FEDS) are preliminary
materials circulated to stimulate discussion and critical comment. The analysis and conclusions set forth
are those of the authors and do not indicate concurrence by other members of the research staff or the
Board of Governors. References in publications to the Finance and Economics Discussion Series (other than
acknowledgement) should be cleared with the author(s) to protect the tentative character of these papers.
The Relationship Between Information Asymmetry

and Dividend Policy
Cindy M. Vojtech
March 2012
Abstract
This paper examines how the quality of firm information disclosure affects shareholders use of dividends to mitigate agency problems. Managerial
compensation is linked to firm value. However, because the manager and
shareholders are asymmetrically informed, the manager can manipulate the
firms accounting information to increase perceived firm value. Dividends
can limit such practices by adding to the cost faced by a manager manipulating earnings. Empirical tests match model predictions. Dividend-paying
firms show less evidence of earnings management. Furthermore, nondividend
payers changed earnings announcement behavior more than dividend payers
Division of Monetary Affairs, [email protected]. I owe special thanks to my adviser
Roger Gordon for his guidance and encouragement. The paper has benefitted from comments and
helpful feedback from Silke Forbes, Nikolay Halov, Garey Ramey, and seminar participants at the
University of California, San Diego and at the 2010 Southwest Finance Association Conference.
All remaining errors are my own. The views expressed in this paper are solely the responsibility
of the author and should not be interpreted as reflecting the views of the Board of Governors of
the Federal Reserve System or of anyone else associated with the Federal Reserve System.
2
following the SarbanesOxley Act, a law that increased financial disclosures.
Keywords: Dividends, Earnings Management, Information Asymmetry,
SarbanesOxley Act, Financial Disclosure
JEL classifications: G30, G35, G38, K22, M41, M43
Introduction
The use of dividends is a common practice by U.S. public firms, totaling around
$690 billion in 2010.1 From a tax perspective, paying dividends is inefficient because managers can use the same cash to invest in firm growth, generating capital
gains.2 This paper provides an explanation of dividend behavior by showing how
dividend policy helps mitigate agency problems. Dividend policy limits a managers
discretion over accounting reports; dividends therefore make reported earnings more
informative.
A manager of a public company makes many investment decisions that are not
seen by shareholders. Shareholders do not generally see the individual projects
adopted or specific assets purchased by a manager nor can shareholders see all of
the investment opportunities available to a manager. Financial reports are a primary
source of information about the performance of firm investments, but the manager
influences that information.
Starting with Easterbrook (1984) and Jensen (1986), researchers began explaining
dividend policy as a result of agency problems. Agency problems arise because the
manager has incentives beyond simply maximizing shareholder value. Dividends
pull free cash flow out of the firm so that the manager has less funds to misinvest
1
2
This figure is based on firms that are in the COMPUSTAT database.

The dividend tax rate has generally been higher than the capital gains tax rate that applies
when shareholders sell their shares. Dividends also create a tax event for all taxable shareholders.
3
(Jensen, 1986).
Gordon and Dietz (2006) and Chetty and Saez (2007) developed incentive conflict
models and showed that agency models perform better than other types of dividend
models by having predictions that were more aligned with the empirical data. These
agency models can predict behavior around tax changes, explain the heterogeneity
of payout policies across firms, and explain how high levels of ownership by the
management and the board of directors (hereafter, board) influence payout policy.
This paper contributes to the literature by showing theoretically and empirically how
information asymmetry interacts with mechanisms that mitigate agency problems.
In order to align the incentives of managers with shareholder interests, managerial
compensation is linked to firm value. However, the manager and shareholders are
asymmetrically informed. As a result, the manager can manipulate the firms accounting information to increase perceived firm value. Because the board selects
the dividend, my model shows how dividends can induce managers to reveal more
information in accounting reports. Dividends lower the available funds for new
investment, which raises the marginal product of firm capital. Because earnings
manipulation is assumed to have real cost, any manipulation reduces funds further,
causing a drop in future profits proportional to the marginal product of capital.
Dividends make the manipulations more expensive, inducing more-accurate reporting.
I test my model by examining how proxies of earnings management (EM) are affected by dividend policy. EM is the purposeful movement of earnings from one
period to another for a private benefit.3 More EM is possible when there is more
information asymmetry. According to my model, dividend payers should use less
EM; the empirical tests match this prediction. Dividend payers have less evidence
of EM than nondividend payers. This paper is the first to empirically test for EM
by U.S. firms across dividend policy type and to document a difference in the size
3
This definition is based on Schipper (1989).
4
of apparent EM behavior.4
The model predictions also hold when there is an exogenous shock to financial
disclosure. The SarbanesOxley Act of 2002 (SOX) was designed to decrease the size
of the information gap between the manager and shareholders by increasing financial
disclosures and by establishing severe penalties for managers if reports do not fairly
represent the financial condition of the firm. Tests using EM proxies show that
nondividend payers appear to have changed earnings announcement behavior more
than dividend payers following the passage of SOX. This behavior suggests that
dividends had indeed been useful in limiting earnings management.
The proxies for EM used in this paper rely on a large accounting and finance literature. Prior researchers have developed ways to proxy for EM by estimating
discretionary accruals (DA). Positive (negative) DA indicate inflating (deflating)
earnings. Because there is no consensus on the best method for estimating DA,
several methods for measuring EM are used in this paper.
The next section reviews the stylized facts of dividend behavior and EM behavior and
provides more background on SOX. Section 3 develops a model with several testable
predictions regarding the interaction among dividends, information asymmetry, and
EM. Section 4 tests these predictions by first estimating DA with various methods
and then using these EM proxies in regressions. Section 5 provides the conclusion.
4
Researchers have looked at EM behavior by dividend payers. Kasanen, Kinnunen and Niskanen
(1996) look at dividend payers in Finland and find evidence that firms use EM to meet dividendbased targets for earnings. Daniel, Denis and Naveen (2008) look at dividend payers in the U.S.
and find evidence that dividend payers use EM to meet debt covenant targets so that dividends
can be paid. Chaney and Lewis (1995) mention in a footnote that dividends could be used as a
cost to over-reporting but do not model or test the idea. Savov (2006) uses a sample of German
companies to test the relationship among EM, investment, and dividends. Regressions of EM
proxies on dividends and other firm characteristics show a negative relationship between dividends
and EM, but results are not statistically significant.
2
2.1
Stylized Facts and Background

Dividend Behavior
In an effort to explain dividend behavior, researchers have proposed several theories. Many theories can be categorized as explaining dividend payment as either
an agency cost or as a signal of quality (manager or earnings). Overall, the predictions of agency models better match empirical data than those of signaling models
(Gordon and Dietz [2006] and Chetty and Saez [2007]). Agency models predict dividend increases after dividend tax decreases, when firms are likely to start paying
dividends, and can explain the heterogeneity of payment policies across firms.
The application of signaling models is varied. Both Bhattacharya (1979) and Miller
and Rock (1985) develop theoretical models that connect dividends with future earnings; however, empirical support of this relationship is weak. DeAngelo, DeAngelo
and Skinner (1996) did not find evidence that dividends could identify firms with
superior earnings. This result should not be surprising given that dividends tend
to be stable while earnings are more volatile. Brav, Graham, Harvey and Michaely
(2005) also reject the signaling explanation based on their survey data.
Generally, dividends can be paid because the company has a history of being profitable. DeAngelo, DeAngelo and Stulz (2006) find that established firms with high
retained earnings-to-equity ratios are more likely to pay dividends. Fama and French
(2001) find that dividend payers tend to be large, highly profitable, slow growth,
and established firms.
However, signaling models can explain some important relationships seen in the
data. Dividends are not only backward-looking. This paper contributes to the
literature by building upon agency models and showing how dividends can signal
true earnings. My model and results are related to the empirical findings of Skinner
and Soltes (2011). These authors show that dividends provide information about
6
longer-run expected earnings, which reported earnings are permanent.
My model is also related to that of John and Knyazeva (2006), who explain payout
policies from the perspective of agency problems but frame payout policies as a
type of precommitment. Because firms with weak governance have potentially large
agency problems, managers commit to dividend payments to satisfy the market.
Dividends impose a major commitment given the negative market reaction to a
dividend omission or decrease. My model uses dividends as a commitment device,
but the board makes the commitment.
2.2
Reported Earnings Behavior
A large amount of the information that current and prospective shareholders receive
about a firm comes from financial statements. Generally Accepted Accounting Principles (GAAP) are used for financial statements of U.S. public firms. Under GAAP,
the manager has the flexibility to influence such things as when bad customer credit
is written off, how inventory is expensed, how capital goods are depreciated, and
how to value pension liabilities. The manager can also influence the timing of real
transactions through means such as deciding when new investments are made and
pushing through large-volume sales near the end of a reporting period.
Earnings management (EM) involves any combination of these tactics with the purpose of achieving an earnings target. Given managerial incentives, the earnings target is the one that maximizes the combined value of such things as bonuses, stock
options, and share holdings. Notice that managers and shareholders incentives are
only aligned in the last item, assuming that both the manager and shareholders sell
their shares at the same time.
Many methods of EM are not illegal, and researchers generally believe that EM
is utilized in varying degrees by many firms. Manager decisions regarding EM are
motivated by capital market events such as initial public offerings (IPOs), secondary
7
offerings, or management buyouts (Perry and Williams [1994], Teoh, Welch, and
Wong [1998], and Teoh, Wong, and Rao [1998]). EM decisions are also influenced
by the use of options and firm value in managerial compensation packages and by the
managers desire to remain employed. These decisions in turn affect how managers
inform shareholders about the firms financial performance (Healy [1985], Chaney
and Lewis [1995], Aboody and Kasznik [2000], and Healy and Palepu [2001]).
Chaney and Lewis (1995) have a model similar to the one presented here. Managers
have compensation tied to firm value, have private information on firm value, and
can announce earnings away from true earnings. Chaney and Lewis do not allow
dividends, but the authors recognize that dividends could be used as a cost of EM.
2.3
Background on the SarbanesOxley Act of 2002
SOX significantly increased the reporting requirements of U.S. public firms. The
stated motivation behind SOX was to improve the quality of information disclosed
to investors. According to the title page of the act, SOX is an act to protect
investors by improving the accuracy and reliability of corporate disclosures made
pursuant to the securities laws, and for other purposes (U.S. Congress, 2002).
To improve corporate disclosures, SOX implemented several changes, including a
requirement for the manager to certify financial statements, a requirement that all
audit committee members of the board be independent, and a requirement for firms
to disclose details of their internal controls.5 This paper will not test the separate
features of SOX but will assume that overall the law decreased the information
asymmetry between the manager and shareholders.6
5
SOX also mandated the creation of a quasi-government agency to oversee the audit industry,
but on June 28, 2010, the Supreme Court ruled 5-to-4 that this mandate was unconstitutional. The
ruling only affected that agency, the Public Company Accounting Oversight Board (PCAOB), and
directed the PCAOB to be placed under the control of the Securities and Exchange Commission.
6
See Coates IV (2007) for a more detailed discussion of the various components of SOX.
8
A key component of SOX, section 302, requires CEOs and CFOs to certify firm
financial reports. If the certification is proven to be incorrect, the officers are liable for a $5 million fine or 20 years in jail.7 While the language of the law only
prohibits untrue statements and requires fair presentation, the severity of the
punishments and the uncertainty of enforcement could make managers push for
more conservative estimates in the publishing of financial reports. Securities and
Exchange Commission litigation is more likely when earnings are overstated (Watts,
2003a,b). This asymmetry in enforcement and overall uncertainty could lead to a
significant change in reported earnings behavior.
President George W. Bush signed SOX into law on July 30, 2002, in the midst of
several corporate financial restatements and several allegations of fraud. The uproar
over these announcements could have also suppressed aggressive accounting. Furthermore, the dissolution of Arthur Anderson may have led the remaining auditors
to be more assertive in their auditing work.
Prior research and the tests reported in this paper suggest that there has been a
change in reported earnings around the time SOX was passed. Earnings management behavior decreased. Cohen, Dey and Lys (2008) find an increase in the absolute
value of discretionary accruals before SOX followed by a reversal of the trend after SOX. Lobo and Zhou (2006) focus on the managers choice to lower earnings
after the passage of SOX and find evidence that managers significantly decreased
discretionary accruals in the post-SOX period, suggesting less inflation of earnings.
7
U.S. Congress (2002) Sec. 906.
Model
3.1
Overview and Set-up
In this model, there are three periods (0, 1, 2) and two players: the manager and
the shareholders. All players are risk neutral.
The managers objective is to maximize the value of the managers compensation
package. The shareholders are represented by the board. The board and the shareholders are considered as the same player because the board and the shareholders
have the same objective of maximizing the value of the firm. The board helps
monitor the manager by setting the firm dividend policy.
In period zero, the board and the manager establish a contract covering the next
two periods. The contract specifies an allocation of nM shares for the manager to be
paid at the end of the first period. A portion of these shares will vest and will be
sold after the announcement of first-period earnings.8 The balance of shares (1 )
cannot be sold until the end of the second period, when the firm is liquidated. All
shares are assumed to retain dividend rights.9 The terms of the managers contract
are public knowledge and cannot be renegotiated.
Once the contract is set, the board commits to a dividend policy. The policy desig8
Because all managers sell these shares, this event does not provide shareholders with additional
information. Empirically, managers tend to hold a large amount of equity ownership. While there
are restrictions about when trades can be executed, managers are able to sell options, and they
are generally free to sell shares. However, Bettis, Coles and Lemmon (2000) find that many firms
have explicit blackout periods. Other firm-level policies may include ownership requirements that
mandate a minimum ownership level. Firms may also place restrictions on the size of transactions
or have an approval process.
9
Similar assumptions are adopted by Miller and Rock (1985) and Chaney and Lewis (1995).
Managerial compensation is linear in the value of the firm with exogenous weights. The expected
value of the early vesting shares can be considered as the labor market price the firm must pay for
the manager. It is set equal to an outside option the manager has when signing the contract.
10
nates a specific level of dividends.
The managers contract also includes the assignment of an initial capital stock K0 ,
which determines the distribution of earnings in period one. Only the manager
sees true earnings. The manager has the option of using firm cash flows for productive investment or for EM, announcing earnings different from true earnings.
Announcing higher earnings can potentially raise the value of the shares sold after
the earnings announcement. However, the board has already established a dividend
policy. Because dividends are paid out of firm cash flows, dividends limit the resources available for EM and therefore limiting the amount of price manipulation
that managers can exert on firm value. Figure 1 shows the order of events for this
model.
All manager compensation is equity-based. The manager can only earn more by
increasing firm value.10 While this form of compensation aligns the interests of
the manager with shareholders, the agency problems are not entirely solved. The
manager has more information about the true performance of firm operations and
has control over the release of firm information. This information asymmetry could
allow the manager to push the market value away from true value.11 While this
model only uses shares for compensation, the incentives are similar to managers
with option portfolios. Managers will want to improve firm valuations around the
time when options are exercisable.12
10
It is important to note that manager compensation is not linked to an effort or ability type.
All hired managers are equally capable of identifying new investments for the firm.
11
This model is not designed to find the optimal contract for shareholder wealth maximization.
Rather, the managerial compensation design is meant to mimic compensation structures seen
in the data. Actual contracts tie pay to performance or to long-term results much less than
optimal contract models suggest. Based on contract theory, the optimal contract for a risk-neutral
agent with unobserved actions is to sell the firm to the manager. Lucas and McDonald (1990)
offer several explanations why contracts can limit but not eliminate problems associated with
information asymmetry, including timing considerations. For a comprehensive survey of managerial
compensation practices see Murphy (1999).
12
The use of options in compensation also changes the risk profile of the compensation package.
This model has no incentive or mechanism for the manager to increase or decrease risk.
11
To simplify the analysis, this model does not allow for the possibility of share repurchases and does not allow for further financing from debt or equity. This model
generally follows the new view modeling assumption that investment is done primarily out of retained earnings.13 For purposes of modeling dividend behavior, this
funding assumption follows the empirical evidence that dividend payers tend to be
large, highly profitable, and established firms.
3.2
True Earnings, Earnings Announcements, and Firm Valuation
Shareholders develop expectations of firm earnings based on observing industry performance and knowing initial capital. Their unconditional expectation of earnings
can be denoted as f (Kt1 ), where Kt1 is the level of capital in period t 1 and
f () is the production function of the firm.
The true earnings of the firm are only known by the manager. True earnings for
period one and period two, and the change in capital over time, are
1 = f (K0 ) + 1
2 = f (K1 )
K1 = (1 )K0 + I1 ,
(1)
where [0, 1) is the depreciation rate of capital and 1 is a production shock

seen only by the manager. When period two starts, only the manager knows the
amount of additional investment in capital I1 . Only production from period one
capital determines period two earnings. At the end of the second period, the firm
is liquidated.
The production function has the following properties: f C ; f (K) 0; f (0) = 0;
f 0 > 0; and f 00 < 0. The production shock has two possible values: 1,H (high) and
1,L (low). The probability of a low shock is .
13
The new view model is described in Auerbach (1979) and Bradford (1981).
12
After seeing true earnings in the first period, the manager must announce a level of
earnings a1 . The announcement can be different than the true earnings, but there
is a cost of lying. The relationship between announced earnings and true earnings
for the two periods can be written as
a1 = 1 +
= f (K0 ) + 1 +
a2 = 2
= f (K1 ) .
Empirically, the manager has flexibility in controlling reported earnings through
accounting rules and the timing of real transactions. As suggested by the formulas
above, many of these practices just change how things are counted so the timing of
earnings moves from one period to another. The inflation (deflation) in period
one is reversed in period two. However, these efforts distract the manager from
identifying optimal projects, creating real costs.
These real costs lower the amount of investment which, in turn, lowers period two
earnings.14 The cost has the following properties: c C ; c() 0; c(0) = 0;
c(x) = c(x); and c00 > 0. Notice that the same cost is incurred whether the
manager is inflating or deflating earnings.15 The cost of changing earnings also
increases at an increasing rate.
The cash flow generated by the firm is assumed to be equal to the true earnings
of the firm minus the taxes payable based on announced earnings. The cashflow
constraint is therefore
D1 + I1 + c() = f (K0 ) + 1 a1 ,
(2)
where D1 is the dividend paid in the first period and is the corporate tax rate.
14
The cost of EM can also be understood as using programs such as volume discounts to improve
sales, which undercut future sales or incurring extra fees to get additional capacity on line.
15
In reality, it may be cheaper to deflate earnings because auditors may be less worried about
conservative practices (Watts, 2003a,b).
13
Cash flow can be used for the dividend, for investment, or for covering the costs
associated with inflating (or deflating) earnings.16
The model assumes that reported earnings are taxable. While not explicitly true
for U.S. firms, this assumption avoids the potential problem that outsiders can
use the information in GAAP accounting statements and tax statements to better
understand the level of EM.17
Given the order of the decisions, investment is a residual. Period one capital can be
calculated by combining equation 1 and equation 2:
K1 = (1 )K0 + f (K0 ) + 1 a1 c[a1 f (K0 ) 1 ] D1 .
(3)
If shareholders have perfect information about firm earnings, firm value in period
zero is determined by the present value of the firms expected payouts. To simplify
the analysis, the differential tax treatment between dividends and capital gains are
dropped. Let d be the discount factor based on the net-of-tax rate of return an
investor can get on a similar risk asset. Under perfect information, the firm value
in period zero equals
h
i
V0 = E0 dD1 + d2 V2
"

E0 [V2 ] = E0 (1 )f (1 )K0 + f (K0 ) + 1 a1 c() D1
#
+ .
16
(4)
Some methods of EM may speed up the receipt of cash, for instance, volume sales near the end
of the period. However, most of the earnings gains from volume sales come in the form of credit
sales, which provide no cash. EM methods such as changing inventory methods, writing off debt,
or changing the composition of depreciated assets do not provide cash except to the extent that
taxes change.
17
See Erickson, Hanlon and Maydew (2004) for a more extensive discussion of tax earnings
versus GAAP earnings. These authors study firms that restated earnings when original reports
were higher. They find evidence that firms overstating earnings paid higher taxes. However, these
cases are tied to allegations of fraud. The use of fraud is outside the scope of this paper.
14
Equation (4) shows that even if shareholders were perfectly informed, managers will
report earnings that differ from true earnings in order to minimize the present value
of corporate tax payments. Let a1 = a1, be the optimal announcement strategy
given a production shock (high or low) and perfect information. The first order
condition is
V2
= (1 )f 0 (K1 )[ + c0 ] + = 0.
a1
Deflating earnings by one dollar will increase the value of the firm if the after-tax
marginal product of more dollars of capital from tax savings covers both the
marginal cost of EM from moving that dollar and the delayed tax payment. If there
was no corporate tax and if shareholders have perfect information, there would be
no reason to misreport. See appendix A for details on the characteristics of the
solution for the perfect information problem.
However, shareholders do not have perfect information about true earnings. They
will use the announced earnings to update their beliefs about whether the firm
received a low (
) or high (1 ) production shock.18
V1 (a1 ) = D1 + V1,L (a1 ) + (1 )V1,H (a1 ),
where V1, (a1 ) is the firm value in period one given that the manager reports earnings
of a1 and has a -type production shock.
Shareholders can value each production shock type firm independently,
V1,0 (a1 ) = dV2,0
h

= d (1 )f (1 )K0 + f (K0 ) + 1,0 a1 c[a1 f (K0 ) 1,0 ] D1
i
+ [a1 f (K0 ) 1,0 ] ,
where 0 is the production shock type that shareholders infer. Notice that earnings
announcement depends on the production shock 1 .
18
Equilibrium types will be discussed later, but note that if there is a pooling equilibrium, nothing
is learned from the earnings announcement. If there is a separating equilibrium, the announcement
reveals the production shock.
15
3.3
Managerial Incentives
The manager uses the earnings announcement to maximize the payoff from the
compensation package. The optimal earnings announcement for a manager with
-type production shock will depend on the type 0 shareholders infer. There will
be two levels of earnings announcements in a separating equilibrium, and one announcement in a pooling equilibrium. Shareholders will value a firm assuming a low
production shock for any off-the-equilibrium-path announcements. The managers
maximization formula is

U1, (a1 , 0 ) = nM D1 + nM V1,0 (a1 ) + d(1 )V2, (a1 ) .
Perfect Information
Define a
1, as a managers announcement strategy. If shareholders know the production shock 1 perfectly, every manager will have an announcement strategy that
is optimal for tax purposes a
1, = a1, . There is no incentive to change earnings
further.
Imperfect Information
If shareholders do not know the value of the production shock, managers will want
to use the first-best announcement strategy.19
Low Production Shock Manager
Under a separating equilibrium, managers with a low production shock know that
shareholders will correctly infer from the earnings announcement that there was a
low production shock. These managers will therefore optimize firm value conditional
on a low shock.
U1,L (
a1,L , L) U1,L (a1,L , L).
The best strategy in this case is to choose the tax optimizing announcement a
1,L =
a1,L .
19
Attention focuses on stable equilibria by restricting out-of-equilibrium beliefs, which eliminate
many unintuitive equilibria. This solution strategy is discussed by Cho and Kreps (1987).
16
A separating equilibrium is supported only if a manager facing a low production
shock cannot do better by mimicking the announcement made by a high-shock
manager. The low-shock manager will not imitate as long as
U1,L (a1,L , L) U1,L (
a1,H , H).
(5)
High Production Shock Manager

Because U1,L (
a1,H , H) is a declining function in a
1,H , this incentive constraint (equation 5) defines the minimum value that the high-shock manager can announce that
will prevent mimicking. This level is denoted as amin
1,H .
Ignoring this incentive constraint, a manager with a high-type shock can choose the
tax optimizing announcement and not worry about the low type mimicking. Any
further exaggeration of earnings would lower investment, lowering firm value.
a
1,H = a1,H
However, supporting the separating equilibrium requires that
U1,H (amin
1,H , H) U1,H (a1,H , L).
The announcement by a high-shock manager then satisfies

a
1,H = max U1,H (amin
,
H),
U
(a
,
L)
.
1,H 1,H
1,H
There are also pooling equilibria if the incentive constraints do not hold. A manager
with a high-type shock will make the tax optimizing earnings announcement, and
that announcement can be mimicked by a manager with a low-type shock. See
appendix A for more details on the model solution.
3.4
Board Dividend Policy
Because the board is trying to maximize firm value, the dividend policy is designed
to optimize ex ante value. The optimal dividend depends on the value of the firm
17
in the expected equilibrium: separating or pooling. The dividend is set to help
minimize exaggerated earnings announcements.
Complete Information
Optimal investment depends on the marginal product of firm capital. Under perfect
information (equation 4), the first order condition for the boards maximization
problem becomes20
V0
: 1 = d(1 )E0 [f 0 (K1 )].
D1
Dividends will be used to pull cash out of the firm if the discounted after-tax marginal
return is less than one. Let r equal the after-tax rate of return available for a similar
risk asset. Then the board will use dividends to manage firm capital such that
E0 [f 0 (K1 )] 1 + r.
(6)
This equation will hold as an inequality when all earnings are left in the firm to
be invested in capital. This relationship is well established in the dividend model
literature.21 Given the cash flow constraint, the only role of the dividend is to
determine the level of new investment.
Incomplete Information
Because the dividend policy is announced before the earnings announcement, the
board will set the policy using the initial probabilities of the production shock. To
optimize dividends, the board will recognize how the dividend affects the nature of
the equilibrium.
The effect of dividends on firm value has two channels. There is direct effect and
an indirect effect through a change in earnings announcement.
V1 = D1 + V1,L (a1,L (D1 ), D1 ) + (1 )V1,H (a1,H (D1 ), D1 )
V a
V a
V1
V1,L
V1,H
1,L
1,L
1,H
1,H
= 1+
+
+ (1 )
+
= 0 (7)
D1
a1,L D1
D1
a1,H D1
D1
20
21
To simplify the analysis, depreciation is not assumed to have different tax treatment.
See Gordon and Dietz (2006) and Chetty and Saez (2007) for further discussion.
18
If there is a pooling equilibrium, the board will simply use equation 6 to set dividends, where there is a strict inequality when D1 = 0. Because the board maximizes
ex ante firm value, the ex post investment will be too high if the production shock
is high, and the ex post investment will be too low if the production shock is low.
If a separating equilibrium exists such that both manager types announce the taxoptimizing level of earnings, the board will again use equation 6 to set the dividend,
where there is a strict inequality when D1 = 0. Notice that the first term in each
set of parenthesis (equation 7) equals zero due to the envelope rule.
If a separating equilibrium is supported by high types announcing exaggerated earnhigher than the optimal for tax purposes, the value of the firm
ings amin
1,H , earnings

min
V1,H a1,H
is not optimized amin D1 > 0 . In these cases, dividends have an added benefit
1,H
on firm value. Dividends lower announcements, lowering EM costs.

U1,L (a1,L , L) U1,L (amin
1,H , H)
amin
1,H
=
D1
U1,L (a1,L ,L)

U1,L (amin
1,H ,H)
D1
D1
U1,L (amin
1,H ,H)
<0
amin
1,H
Proof of this relationship is in appendix A. Because low types have a higher marginal
product of capital, mimicking low types face higher costs to exaggerating earnings
and are harmed more by dividends. A higher dividend causes the incentive constraint to hold at a lower value for amin
1,H . Less value is lost because the self-selection
constraint becomes less binding.
3.5
Sudden Decrease in the Size of Information Asymmetry
Now assume there is a large decrease in the amount of information asymmetry

between the manager and shareholders. Auditors could have become instantaneously
more vigilant or law changes could make EM more costly. SOX and the overall
change in the corporate environment in the early 2000s have aspects of these two
19
pressures. These forces would cause the EM cost function to increase to c such that:
c(x) c(x), x; c C ; c(0) = 0; c(x) = c(x); and c00 > 0.
Under this new information regime, EM is more expensive. The new regulations
force more reporting and make it harder to change the timing of earnings. As a
result, managers report earnings closer to the truth.
The information shock affects all firms, but managers at dividend paying firms were
already being constrained by board dividend policy. As shown in the last subsection
(3.4), a manager at a nondividend paying firm has more freedom to manage earnings.
As a result, the information regime change is more likely to constrain the earnings
announcement of managers at nondividend paying firms than dividend paying firms.
3.6
Predictions
Based on the model described above, the following relationships are predicted. These
relationships will be tested in the next section (4).
P1: If dividends help limit the use of EM, managers at dividend paying firms will
show less EM behavior than those at nondividend paying firms.
P2: If SOX or the overall change in the accounting environment increased the
amount of financial disclosure in company financial statements, the information
asymmetry between shareholders and the manager should have decreased. Given
that EM is a proxy for the size of the information gap, the amount of EM should
have decreased following SOX.
P3: Given that managers at dividend paying firms are more constrained in their
use of EM, the drop in EM will be less for dividend paying firms than nondividend
paying firms following the passage of SOX.
20
Testing for Earnings Management
4.1
Data
The data for all of the analyses in this paper come from the COMPUSTAT North
America Fundamentals Annual database and the Center for Research in Security
Prices (CRSP) database available through Wharton Research Data Services. The
COMPUSTAT database contains market and financial data on public U.S. firms.
The CRSP database has daily stock price and dividend data for U.S. firms. Only
data on the public firms trading on the NYSE, AMEX, or NASDAQ are used for
this paper. In statistical terms, the general data set is an unbalanced panel because
firms enter and leave the data set as firms get listed on these exchanges, delist, go
bankrupt, or are acquired.22
Following past research, the samples exclude financial companies and utilities because these industries have regulations on capital. These regulations influence earnings motives and the ability to return earnings to shareholders through dividends
(Chetty and Saez [2005]).23
4.2
Estimates of Discretionary Accruals
An extensive amount of accounting research has focused on ways to model or detect EM. Because modeling techniques can only proxy for actual EM, using these
methods is a test of both the detection model and the use of EM. Despite these
limitations, expected accrual models are widely adopted by researchers.24 Because
22
See appendix B for more details on the data sets and on the specific COMPUSTAT and CRSP
variables used.
23
The specific SIC codes excluded are 4900-4949 and 6000-6999, matching Fama and French
(2001); Chetty and Saez (2005); and DeAngelo, DeAngelo and Stulz (2006).
24
See Dechow, Sloan and Sweeney (1995); Dechow and Dichev (2002); Dechow and Schrand
(2004); Kothari, Leone and Wasley (2005); and Cohen, Dey and Lys (2008) for more discussion.
21
there is no consensus on the best model to use, several methods of proxying for EM
are used in this paper.
The seminal work by Jones (1991) showed a method to estimate the amount of
manipulation by comparing a firms reported accruals to expected accruals. Several
papers since then have improved upon this method. Four primary models are reported in this paper: three models are variations of the Jones model and one model
is a performance matching model suggested by Kothari, Leone and Wasley (2005).
The overall goal of these expected accrual models is to obtain a measure of discretionary accruals (DA), accruals that are more easily controlled by managers. Any
change in total accruals (TA) comes from changes in DA and normal accruals (NA),
accruals that come about through standard firm operations and that are less open
to control.
T At = (DAt DAt1 ) + (N At N At1 )
Jones modeled expected accruals based on observable firm characteristics. The first
model in this paper will follow her basic technique and include a constant in the
regression to help reduce heteroskedasticity not handled by deflating the variables
with lagged assets and to help control for problems related to an omitted scale
variable (Brown, Lo, and Lys [1999] and Kothari, Leone, and Wasley [2005]). The
primary regression is25

REV (12)
P P E (7)
T Ai,t
1
i,t
i,t
= 0 + 1
+ 1
+ 2
+ i,t . (8)
Ai,t1 (6)
Ai,t1 (6)
Ai,t1 (6)
Ai,t1 (6)
The level of total accruals required by firm i depends on firm size measured by lagged
total assets (A), on the change in firm revenues (REV ), and on the firms fixed
capital, proxied by property, plant, and equipment (P P E). Everything is scaled by
lagged total assets. Regressions are run at the industry level (two-digit SIC) for each
25
The parenthetical numbers in the formulas for this section are the COMPUSTAT annual data
numbers.
22
year. Each year-industry regression must have at least ten firm-year observations to
be included in this analysis.26
Total accruals for all the models reported in this paper are defined following Dechow,
Sloan and Sweeney (1995) and Kothari, Leone and Wasley (2005) (KLW).27
T A = [Current Assets(4) Cash(1)]
[Current Liabilities(5) Current Maturities of LT Debt(34)]
Depreciation and Amortization Expense(14)
DA are estimated by taking the difference between reported accruals and expected
accruals.
i,t = i,t
DA
1
REV
P P E
T Ai,t
i,t
i,t
0
1
1
2
Ai,t1
Ai,t1
Ai,t1
Ai,t1
Positive DA are evidence of inflating earnings. Negative DA are evidence of deflating
earnings.
The second Jones model (Modified Jones) works the same
except the regression

REVi,t (12)RECi,t (2)
formula (equation 8) has a change in the revenue term to 1
,
Ai,t1 (6)
where REC is accounts receivable (Dechow, Sloan, and Sweeney [1995]). By taking
out the change in receivables, this form of the model assumes that changes in credit
sales are discretionary. This type of model is better suited to detect EM achieved
through methods such as volume sales near the end of a reporting period.28
26
Due to this constraint, analysis is limited to firms that have a fiscal year end date of December
31.
27
Further EM testing that is not reported here include using the total accrual definition used
by Cohen, Dey and Lys (2008): Earnings before extraordinary minus operating cash flows. I also
test alternative definitions of industries: three-digit SIC or the 49 industry definitions created by
Eugene Fama and Kenneth French (see http://mba.tuck.dartmouth.edu/pages/faculty/ken.
french/data_library.html). These alternative testing results are qualitatively similar to those
reported in this paper.
28
This EM tactic is also known as channel stuffing.
23
The third Jones model (Jones with ROA) is another variation of the total accruals
regression formula. It includes return on assets (ROA) on the right-hand side.29
KLW argue that including ROA helps improve specifications where there are periods
of abnormal returns. However, KLW also point out that there are many reasons to
expect that ROA does not affect accruals linearly. According to their tests, a model
matching on performance (current year ROA) performs the best.
This performance-matching model (Performance KLW) is the fourth DA measure
reported in this paper. The Performance KLW DA for firm i in year t is defined as
the Jones DA for firm i in year t minus the Jones DA for the firm with the closest
ROA in the same 2-digit SIC code and the same year.30 Effectively, this proxy of
EM defines DA relative to a firms closest industry peer by ROA.
The dividend model in this paper predicted that dividend paying firms will have
lower levels of EM than nondividend paying firms and that EM will fall after SOX.
Because DA is positive or negative depending on whether the manager is inflating or
deflating earnings, evidence of earnings management will be proxied by the absolute
value of DA.
Given a hypothesis about managing earnings for a specific event, some researchers
use an alternative strategy of modeling accruals by using a pre-event estimation
period. SOX is a testable event, but it also changed the disclosure rules. Using a preevent estimation technique assumes a non-time varying relationship between normal
accruals and firm characteristics. It is likely that SOX changed these relationships,
making the results from a pre-event estimation strategy biased. Comparing firm
accrual behavior within the year will help identify those firms that had abnormal
changes in accruals versus peers in the same period.
29
This model is also tested but not reported using the Modified Jones variation. The results are
similar to those reported in this paper.

30
This model is also tested but not reported using the Modified Jones variation. The results are
similar to those reported in this paper.
24
4.3
Initial Results of Discretionary Accruals Testing
Figure 2 shows how the median absolute value of DA has changed between 1980
and 2008 using the four primary models described in section 4.2. Before taking
the absolute value, each DA measure has been winsorized at the top and bottom 1
percent of the distribution. Across all models, nondividend payers consistently have
a higher level of absolute DA. Nondividend payers show more evidence of EM than
dividend payers. This difference remains the case as the composition of firms in the
sample changes, as shown in the bottom panel.31
Figure 3 focuses on the time period around the passage of SOX. The vertical line
separates the pre-SOX and post-SOX periods. Recall that SOX was passed in July
2002. Given that all the firms in this sample have a December fiscal year end,
December 2002 was the first financial report under the new law. Not all of the
aspects of SOX were phased in by this point, but the officers did have to certify
their financial reports.
For each model, dividend payers have roughly the same median absolute value of
DA throughout the period, but there is a drop in the same measure for nondividend
payers between the pre- and post-SOX periods. The graphs suggest that the peak of
EM behavior was around 2000, well before SOX. This early change in behavior may
have been in response to a changing environment. The stock market had peaked,
and the Arthur AndersenEnron case was unfolding. According to all of the models,
median absolute DA for nondividend payers fell both in 2001 and in 2002. These
results are consistent with other research on DA and SOX (Cohen, Dey, and Lys
[2008] and Lobo and Zhou [2006]).
31
Due to the required data to run the DA regressions, many firm-year observations are dropped
from the original database. These excluded firms are generally smaller and younger firms that
are more likely to be nondividend payers. The exclusions create a nonrepresentative sample of
the market, including a relatively high composition of dividend payers. Given the prediction of
the model that nondividend payers are more likely to use EM, results from tests using a sample
composed of more established firms are likely to be conservative.
25
Overall, this initial DA evidence supports the three predictions developed from the
theoretical model. Dividend payers use less EM than nondividend payers (P1).
Firms use less EM after SOX (P2). SOX changed the behavior of nondividend
payers more than dividend payers (P3).
4.4
Regressions with Discretionary Accruals
To further examine a possible behavior change following SOX and the relationship
between dividends and EM, this section reports regressions of absolute DA on payout
policy and firm characteristics.
Table 1 shows the summary statistics of the data used in the DA models and for
the regressions in this section. The time window is narrowed to the four annual
reports before and after SOX (data from December 1998 to December 2005). Table
2 shows the DA measures broken down for all firms, for nondividend payers, and for
dividend payers. Notice that the means of the discretionary accruals for all firms
(first four rows in table 2) are near zero. These means should be zero by design
because the discretionary accrual is the regression residual. However, these data
have been winsored at the bottom and top 1 percent.
As expected based on the prior graphs, the mean and median of absolute DA for
nondividend payers are higher than those of dividend payers. The mean/standard
deviation ratio is shown in the last column of table 2. According to all models,
dividend payers have lower relative variation in the absolute level of DA.
The baseline regression is constructed to test the three predictions.
abs(DAi,t ) = + 1 Dividend payeri,t + 2 SOXt Dividend payeri,t
+ year dummies
(9)
Dividend payeri,t and SOXt are dummy variables. Dividend payeri,t equals one if
the firm i paid a dividend in year t and equals zero otherwise. SOXt equals one for
26
all periods after the passage of SOX, where December 2002 is the first such period.
The baseline model specification (equation 9) also includes year dummies.
According to the theoretical model, dividend payers use less EM. Because absolute
DA is a proxy for EM, dividend payers should have lower absolute DA (P1). The
expected sign of the coefficient on Dividend payeri,t is negative. The theoretical
model shows that managers should use less EM after SOX due to its higher cost.
Absolute DA should be lower following SOX (P2). Therefore, the coefficients for
the year dummies should be lower for years following SOX than for those pre-SOX.
Given that dividend payers are already constrained by the dividend, SOX should not
affect DA behavior as much as for nondividend payers (P3). The interacted term
should counteract the year dummies, making the expected sign on the interacted
term positive.
Table 3 reports the results of the baseline regression. The successive columns separately test the four primary DA measures. The signs of all the primary coefficients
are as expected and statistically significant, and the year dummy coefficients follow
the expected pattern. The difference between the 2001 and 2002 dummy variables
is 1 percent across all models, and tests of whether the coefficients are equal are
rejected. Also notice that absolute DA fell by roughly 2 percent between 2000 and
2001. However, dividend payers did not experience the same drop. As expected, the
coefficient on the interacted term offsets the SOX decrease. The coefficient is about
3 percent for all of the models, more than offsetting the SOX drop. This excess may
be due to the previously mentioned 2-percent drop between 2000 and 2001. Recall
that the DA measures are scaled by assets. The units are DA as a share of assets.
These regression results suggest that nondividend payers shrank the composition of
their assets consisting of DA by 1 percent after SOX.
The theoretical model also showed that the amount of earnings management depends
on the marginal product of firm capital. The marginal product of capital can be
proxied by firm characteristics such as size, life cycle, and profitability. The next set
of specifications use the four firm characteristic variables Fama and French (2001)
27
included in their study on dividend payers.32
abs(DAi,t ) = + 1 Dividend payeri,t + 2 SOXt Dividend payeri,t
Valuei,t
+3 NYSE market capitalizationi,t + 4
Assetsi,t1
Earningsi,t
+5 Asset growthi,t + 6
Assetsi,t1
+ year dummies
(10)
NYSE market capitalization is a proxy of size. It is equal to the percentage of NYSE

firms that have the same or a lower market capitalization (takes a value between
zero and one). The Value/Lagged assets measure, also known as the market-to-book
ratio, is similar to Tobins q. Young firms that are expected to grow and become
more profitable in the future are highly valued by the market. These firms tend to
trade at a higher ratio than older firms; it is a proxy for life cycle. Asset growth
is another proxy for life cycle under the assumption that younger firms grow faster
than older firms. Earnings/Lagged assets is a profitability measure and provides
another measure for the opportunity cost of capital.33
The expected sign on NYSE market capitalization is positive. Larger firms tend
to be more mature and have a lower opportunity cost of earnings management. It
may also be easier for large firms to move earnings. The signs on the life-cycle
variables of market-to-book and asset growth are expected to be negative. Younger
firms should have a high marginal product of capital. The expected sign of the
profitability measure is also negative. Profitable firms have a higher opportunity
cost, making EM more expensive for the manager.
Table 4 reports the regression results of this new specification with firm character32
Because the theoretical model did not have any causal predictions between EM and dividends,
the test can be reversed to regress the likelihood of being a dividend payer on absolute DA and
firm characteristics. For instance, the logit regressions run by Fama and French (2001) could be
replicated and include absolute DA on the right-hand side. This type of test shows the same
qualitative results. Larger absolute DA lowers the likelihood of being a dividend payer.
33
Variable definitions are in appendix B.
28
istics. The sign and significance of the coefficient on Dividend payer and that on
the interacted term remain. However, the coefficient on Dividend payer changes
from roughly -3 percent in the baseline to -2 percent. The expected patterns on the
year dummy coefficients remain. These coefficients suggest that DA fell 1 percent
between 2001 and 2002 and fell 1 to 2 percent between 2000 and 2001.
Three of the four coefficients on the firm characteristic variables are statistically
significant, but two of them have signs that are opposite to expectations: NYP and
Value/Lagged assets. Larger firms (NYP) should have relatively low marginal product of capital and, therefore, cheaper EM. These firms should have more evidence
of EM than smaller firms. Firms with high Value/Lagged assets, and therefore,
more expensive EM, should tend to be firms with expectations of strong growth and
less evidence of EM. These results suggest the opposite. Possible reasons for these
results are that larger firms may be more closely scrutinized, making EM more difficult. Firms with high Value/Lagged assets are likely to be younger firms and likely
to have more intangible assets. Both of these characteristics are related to having
more hidden information, making earnings announcements affect stock prices more
and, therefore, creating stronger incentives for the manager.
While the firm characteristics help control for firms that are likely to pay dividends,
it is likely that other firm characteristics affect both dividend payment and DA. To
help control for unobservables, the next set of regressions use firm fixed effects and
are shown in table 5.
The signs on Dividend payer and the interacted term continue to match expectations.
Only the interacted term coefficient is statistically significant, and it remains around
1 percent. By using fixed effects, these coefficients are identified by firms that switch
dividend policy. Yet, the point estimate on Dividend payer suggests that firms
switching to pay a dividend have lower DA. While each year there are firms that
switch to become dividend payers, many firms became dividend payers following the
dividend tax reform in 2003.34
34
Prior to the 2003 law change, dividends were taxed at the personal income rate of the investor
29
The expected patterns on the year dummy coefficients also remain. These coefficients suggest that DA fell 1 percent both between 2001 and 2002 and between 2000
and 2001. The signs of the coefficients on the firm characteristics generally follow expectations when they are statistically significant with the exception of Value/Lagged
assets.
As suggested by the graphs in figure 3 and the regression tests, absolute DA began
falling in 2001. This early change in behavior may have been a result of a change in
the corporate environment. Managers were using less aggressive accounting because
of such things as the Arthur AndersenEnron scandal. As a robustness check, I
change the SOX cutoff from 2002 to 2001 and rerun the tests. For space reasons,
the regression results are not reported in this paper, but the general results still
hold. The signs for the primary variables of Dividend payer and the interacted term
match expectations. Both coefficients are statistically significant in the baseline
regression and a regression with firm characteristic variables. The Dividend payer
coefficient is not statistically different from zero when fixed effects are used. The
expected patterns on the year dummy coefficients also remain; the dummies for 1999
and 2000 are statistically different and higher than 2001 and later.
There is still a concern about the unobserved characteristics of firms. The dividend
literature suggests that dividend payers are different types of firms than nondividend
payers. Perhaps the same characteristics related to dividend payment are related
to DA but are not tied to the mechanisms in my model, or the same characteristics
related to dividend payment are related to predictable accruals. As a robustness
check, I try to identify nondividend paying firms that most look like dividend payers
and compare the DA of those firms to those of dividend paying firms. To identify
nondividend paying firms that look like dividend payers, I use the logit model used
by Fama and French (2001).
(a high of 35 percent), and the top capital gains tax rate was 20 percent. The 2003 dividend tax
reform created a top dividend tax rate of 15 percent and lowered the top capital gains tax rate to
match (Auerbach and Hassett, 2005).
30
logit(Dividend payer = 1) = + 1 NYSE market capitalizationi,t

Valuei,t
+2
+ 3 Asset growthi,t
Assetsi,t1
Earningsi,t
+ industry dummies
+4
Assetsi,t1
(11)
I run equation 11 separately for each year. The estimated coefficients are then
used to generate probabilities of being a dividend payer. Those firm observations
that have an estimated probability greater than 0.5 are then used for DA regression
testing. Table 6 summarizes the results of these logit regressions. Notice that around
200 nondividend paying firms are included each year.
The DA regression results are shown in table 7. For three of the four models, the
primary conclusions still hold: dividend paying firms use less EM, evidence of EM
fell after SOX, and nondividend paying firms changed their behavior more than
dividend paying firms.35
Concluding Remarks
The challenge of optimizing manager behavior for shareholder value has two primary
parts. First, the manager has different incentives than shareholdersagency problems. Second, the manager knows much more about the financial viability of the
firm than shareholdersinformation asymmetry. Both of these elements need to be
incorporated into dividend models to understand the dynamics of payout selection.
Conflicting incentives explain manager behavior given compensation packages and
ownership structure, but incentives alone do not explain payout dynamics following
tougher reporting standards or explain how shareholders might learn more about
the extent of agency problems.
35
The primary conclusions still hold if the probability cut off is changed from 0.5 to 0.75. How-
ever, this change shrinks the number of nondividend paying firms to approximately 50 each year.
31
The model presented in this paper was designed to explore the relationships among
dividend policy decisions, information asymmetry, and managerial incentives. The
model shows and the tests confirm that EM behavior is different depending on payout policy. According to the DA tests, dividend payers did not appear to change
their reporting behavior as much as nondividend payers after the passage of SOX.
Furthermore, dividend payers consistently have lower absolute DA. This lower absolute DA is evidence that dividend payers inflate and deflate earnings less than
nondividend payers.
The model presented here posits that dividends help limit the discretion of management, leading to more-truthful earnings reports. The dividend commitment is
possible through a board that is perfectly aligned with shareholders. Further work
is needed to evaluate how board composition relates to monitoring levels and payout policy.36 Overall, the findings presented here suggest that dividend policies are
effective at limiting information asymmetries.
References
Aboody, D., Kasznik, R., 2000. Ceo stock option awards and the timing of corporate
voluntary disclosures. Journal of Accounting & Economics 29, 73100.
Auerbach, A.J., 1979. Wealth maximization and the cost of capital. The Quarterly
Journal of Economics 93, 433446.
Auerbach, A.J., Hassett, K.A., 2005. The 2003 dividend tax cuts and the value of
the firm: An event study. Working Paper, NBER, no. 11449.
Bettis, J.C., Coles, J.L., Lemmon, M.L., 2000. Corporate policies restricting trading
by insiders. Journal of Financial Economics 57, 191220.
36
Kay and Vojtech (2011) examine this relationship by examining the relationship between board
composition and other monitoring devices such as dividends, CEO ownership, incentive pay, and
leverage. They find some evidence that dividends, CEO ownership, and leverage are substitutes
for the monitoring provided by independent board members.
32
Bhattacharya, S., 1979. Imperfect information, dividend policy, and the bird in
the hand fallacy. The Bell Journal of Economics 10, 259270.
Bradford, D.F., 1981. The incidence and allocation effects of a tax on corporate
distributions. Journal of Public Economics 15, 122.
Brav, A., Graham, J.R., Harvey, C.R., Michaely, R., 2005. Payout policy in the 21st
century. Journal of Financial Economics 77, 483527.
Brown, S., Lo, K., Lys, T., 1999. Use of r2 in accounting research: Measuring
changes in value relevance over the last four decades. Journal of Accounting &
Economics 28, 83115.
Chaney, P.K., Lewis, C.M., 1995. Earnings management and firm valuation under
asymmetric information. Journal of Corporate Finance 1, 319345.
Chetty, R., Saez, E., 2005. Dividend taxes and corporate behavior: Evidence from
the 2003 dividend tax cut. The Quarterly Journal of Economics 120, 791833.
Chetty, R., Saez, E., 2007. An agency theory of dividend taxation. Working Paper,
NBER, no. 13538.
Cho, I.K., Kreps, D.M., 1987. Signaling games and stable equilibria. The Quarterly
Journal of Economics 102, 179222.
Coates IV, J.C., 2007. The goals and promise of the SarbanesOxley Act. Journal
of Economic Perspectives 21, 91116.
Cohen, D.A., Dey, A., Lys, T.Z., 2008. Real and accrual-based earnings management
in the pre- and post-SarbanesOxley periods. The Accounting Review 83, 757
787.
Daniel, N.D., Denis, D.J., Naveen, L., 2008. Do firms manage earnings to meet
dividend thresholds? Journal of Accounting & Economics 45, 226.
DeAngelo, H., DeAngelo, L., Skinner, D.J., 1996. Reversal of fortune: Dividend
signaling and the disappearance of sustained earnings growth. Journal of Financial
Economics 40, 341371.
33
DeAngelo, H., DeAngelo, L., Stulz, R.M., 2006.
Dividend policy and the
earned/contributed capital mix: A test of the life-cycle theory. Journal of Financial Economics 81, 227254.
Dechow, P.M., Dichev, I.D., 2002. The quality of accruals and earnings: The role
of accrual estimation errors. The Accounting Review 77, 3559.
Dechow, P.M., Schrand, C.M., 2004. Earnings Quality. The Research Foundation
of CFA Institute, Charlottesville, VA.
Dechow, P.M., Sloan, R.G., Sweeney, A.P., 1995. Detecting earnings management.
The Accounting Review 70, 193225.
Easterbrook, F.H., 1984. Two agency-cost explanations of dividends. American
Economic Review 74, 650659.
Erickson, M., Hanlon, M., Maydew, E.L., 2004. How much will firms pay for earnings
that do not exist? Evidence of taxes paid on allegedly fraudulent earnings. The
Accounting Review 79, 387408.
Fama, E.F., French, K.R., 2001. Disappearing dividends: Changing firm characteristics or lower propensity to pay? Journal of Financial Economics 60, 343.
Gordon, R., Dietz, M., 2006. Dividends and taxes. Working Paper, NBER, no.
12292.
Healy, P.M., 1985. The effect of bonus schemes on accounting decisions. Journal of
Accounting & Economics 7, 85107.
Healy, P.M., Palepu, K.G., 2001. Information asymmetry, corporate disclosure, and
the capital markets: A review of the empirical disclosure literature. Journal of
Accounting & Economics 31, 405440.
Jensen, M.C., 1986. Agency costs of free cash flow, corporate finance, and takeovers.
American Economic Review 76, 323329.
34
John, K., Knyazeva, A., 2006. Payout policy, agency conflicts, and corporate governance. Working Paper, New York University, New York, NY.
Jones, J.J., 1991. Earnings management during import relief investigations. Journal
of Accounting Research 29, 193228.
Kasanen, E., Kinnunen, J., Niskanen, J., 1996. Dividend-based earnings management: Empirical evidence from finland. Journal of Accounting & Economics 22,
283312.
Kay, B., Vojtech, C.M., 2011. How do firms switch among tools used to monitor
agency problems? Working Paper, University of CaliforniaSan Diego, La Jolla,
CA.
Kothari, S.P., Leone, A.J., Wasley, C.E., 2005. Performance matched discretionary
accrual measures. Journal of Accounting & Economics 39, 163197.
Lobo, G.J., Zhou, J., 2006. Did conservatism in financial reporting increase after
the SarbanesOxley Act? Initial evidence. Accounting Horizons 20, 5773.
Lucas, D.J., McDonald, R.L., 1990. Equity issues and stock price dynamics. Journal
of Finance 45, 10191043.
Miller, M.H., Rock, K., 1985. Dividend policy under asymmetric information. Journal of Finance 40, 10311051.
Murphy, K.J., 1999. Executive Compensation, in: Ashenfelter,O., Card,D. (Eds.).
Elsevier, Amsterdam. volume 3 of Handbook of Labor Economics. pp. 24852563.
Perry, S.E., Williams, T.H., 1994. Earnings management preceding management
buyout offers. Journal of Accounting & Economics 18, 157179.
Savov, S., 2006. Earnings management, investment, and dividend payments. Working Paper, University of Mannheim, Mannheim, Germany.
Schipper, K., 1989. Commentary on earnings management. Accounting Horizons 3,
91102.
35
Skinner, D.J., Soltes, E., 2011. What do dividends tell us about earnings quality?
Review of Accounting Studies 16, 128.
Teoh, S.H., Welch, I., Wong, T.J., 1998a. Earnings management and the underperformance of seasoned equity offerings. Journal of Financial Economics 50, 6399.
Teoh, S.H., Wong, T.J., Rao, G., 1998b. Are accruals during initial public offerings
opportunistic? Review of Accounting Studies 3, 175208.
U.S. Congress, 2002. SarbanesOxley Act of 2002.
Watts, R.L., 2003a. Conservatism in accounting part I: Explanations and implications. Accounting Horizons 17, 207221.
Watts, R.L., 2003b. Conservatism in accounting part II: Evidence and research
opportunities. Accounting Horizons 17, 287301.
36
Figure 1
Order of Events for Model
37
Figure 2
Discretionary Accruals, by Model and by Payer Type [DA/Lagged assets]
38
Figure 3
Discretionary Accruals, by Model and by Payer Type [DA/Lagged assets]
39
Table 1
Summary Statistics for Primary Variables
Variable
Mean
Median
Std. dev.
Total accrual/ Lagged assets

1/ Lagged assets
Sales Chg/ Lagged assets
(Sales Chg - Rec Chg)/ Lagged assets
PPE/ Lagged assets
ROA
NYSE market capitalization
Value/ Lagged assets
Asset growth
Earnings/ Lagged assets
-0.055
0.017
0.156
0.130
0.588
-0.012
0.350
2.991
0.234
-0.011
-0.047
0.004
0.075
0.063
0.446
0.054
0.282
1.694
0.063
0.057
0.643
0.050
1.679
1.553
0.567
0.245
0.301
6.517
2.310
0.330
Statistics based on data between 1998 and 2005 from 12,334 firm-year observations. Total accrual = [Current Assets(4) Cash(1)] [Current Liabilities(5)
Current Maturities of LT Debt(34)] Depreciation and Amortization Expense(14).
COMPUSTAT data items are in parenthesis. See appendix B for more details on variable
definitions and on the specific COMPUSTAT variables used.
40
Table 2
Summary Statistics of DA Model Results, by Payer Type
Variable
Mean
Median
Std. dev.
Discretionary accruals (scaled by lagged assets)

All12,334 firm-year observations
Jones
0.0015
0.0036
Modified Jones
0.0016
0.0030
Jones with ROA
0.0009
0.0009
Performance KLW
0.0015
0.0000
abs(Jones)
0.0634
0.0379
abs(Modified Jones)
0.0643
0.0386
abs(Jones with ROA)
0.0621
0.0377
abs(Modified Jones with ROA) 0.0628
0.0381
abs(Performance KLW)
0.0933
0.0577
abs(Perform KLW Modified)
0.0946
0.0584
Nondividend payers8,991 firm-year observations
abs(Jones)
0.072
0.043
abs(Modified Jones)
0.073
0.044
abs(Jones with ROA)
0.070
0.043
0.104
0.065
Dividend payers3,343 firm-year observations
abs(Jones)
0.0402
0.0280
abs(Modified Jones)
0.0408
0.0283
abs(Jones with ROA)
0.0396
0.0280
0.0649
0.0433
Mean/
Std. dev.
0.1409
0.1422
0.1335
0.1974
0.1258
0.1269
0.1182
0.1192
0.1739
0.1753
0.011
0.011
0.007
0.007
0.504
0.507
0.525
0.527
0.537
0.540
0.144
0.145
0.135
0.198
0.500
0.504
0.522
0.525
0.0430
0.0443
0.0419
0.0721
0.933
0.921
0.947
0.900
41
Table 2 (cont)
Summary Statistics of DA Model Results, by Payer Type
Dividend paying status is based on whether a dividend is paid in year t. Discretionary accruals
from the Jones model are estimated for each industry and year using the residual from the following

TA
REVi,t (12)
P P Ei,t (7)
1
regression: Ai,t1i,t(6) = 0 + 1 Ai,t1
+ 2 Ai,t1
+ i,t ; where T Ai,t =
(6) + 1
Ai,t1 (6)
(6)
[Current Assets(4)Cash(1)][Current Liabilities(5)Current Maturities of LT Debt(34)]
Depreciation and Amortization Expense(14).
COMPUSTAT data items are in parenthesis. REVi,t is change in sales, and P P Ei,t is property, plant and equipment. DA from the Modified Jones model changes the revenue term to

2
to the first equation, where REC is accounts receivable. DA from the
Ai,t1 (6)
Jones Model with ROA are similar to the Jones model except for the inclusion of current years
ROA as an additional explanatory variable. To obtain the Performance KLW model DA for firm i
the Jones model DA of the firm with the closest ROA that is in the same industry as firm i. DA
variables are winsorized at the 1st and 99th percentiles before taking the absolute value. Statistics
based on DA between 1998 and 2005.
42
Table 3
Baseline Regression
The table reports the coefficients from the following regression: abs(DAi,t ) =
+ 1 Dividend payeri,t + 2 SOXt Dividend payeri,t + year dummies; where
abs(DAi,t ) is the absolute value of discretionary accruals estimated by various models
separately shown in each column.
(1)
VARIABLES
Exp.
sign
Dividend payer
SOX*Div. payer
Yr 99
Yr 00
Yr 01
Yr 02
Yr 03
Yr 04
Yr 05
Constant
Observations
Number of id
R-squared
Adj. R-squared
abs(Jones)
(2)
abs(Jones
Mod)
(3)
abs(Jones
w/ ROA)
(4)
abs(KLW
Perform)
-0.0307***
(0.00177)
0.0114***
(0.00214)
0.00268
(0.00200)
0.00866***
(0.00234)
-0.00934***
(0.00198)
-0.0175***
(0.00213)
-0.0169***
(0.00225)
-0.0200***
(0.00217)
-0.0183***
(0.00220)
0.0735***
(0.00174)
-0.0313***
(0.00181)
0.0114***
(0.00219)
0.00158
(0.00208)
0.00680***
(0.00240)
-0.00993***
(0.00204)
-0.0193***
(0.00219)
-0.0180***
(0.00233)
-0.0208***
(0.00223)
-0.0200***
(0.00226)
0.0760***
(0.00180)
-0.0296***
(0.00169)
0.0106***
(0.00204)
0.00182
(0.00199)
0.00676***
(0.00223)
-0.00875***
(0.00186)
-0.0169***
(0.00206)
-0.0155***
(0.00217)
-0.0159***
(0.00215)
-0.0158***
(0.00216)
0.0716***
(0.00168)
-0.0309***
(0.00224)
0.00877***
(0.00280)
-0.00347
(0.00279)
0.00183
(0.00308)
-0.0140***
(0.00275)
-0.0230***
(0.00285)
-0.0217***
(0.00300)
-0.0259***
(0.00290)
-0.0254***
(0.00295)
0.103***
(0.00229)
12,334
2,595
0.058
0.0574
12,334
2,595
0.057
0.0565
12,334
2,595
0.053
0.0523
12,334
2,595
0.039
0.0380
Robust standard errors in parentheses are clustered at the firm level.

*** p<0.01, ** p<0.05, * p<0.10
43
Table 3 (cont)
Baseline Regression

TA
REVi,t (12)
P P Ei,t (7)
1
+ 2 Ai,t1
(6) + 1
Ai,t1 (6)
(6)

2
Ai,t1 (6)
44
Table 4
Regression with Firm Characteristics
The
table
reports
the
coefficients
from
the
following
regression:
abs(DAi,t )
+ 1 Dividend payeri,t + 2 SOXt Dividend payeri,t + 3 NYSE market capitalizationi,t +

Earnings
Value
4 Assets i,t + 5 Asset growthi,t + 6 Assets i,t + year dummies; where abs(DAi,t ) is the
i,t1
i,t1
absolute value of discretionary accruals estimated by various models separately shown in each
column.
(1)
VARIABLES
Exp.
sign
Dividend payer
SOX*Div. payer
NYP
Value /
Lagged assets
Asset growth
Earnings /
Lagged assets
Yr 99
Yr 00
Yr 01
Yr 02
Yr 03
Yr 04
Yr 05
Constant
Observations
Number of id
R-squared
Adj. R-squared
abs(Jones)
(2)
abs(Jones
Mod)
(3)
abs(Jones
w/ ROA)
(4)
abs(KLW
Perform)
-0.0186***
(0.00184)
0.00994***
(0.00210)
-0.0235***
(0.00235)
0.00135***
(0.000188)
-4.96e-05
(0.00102)
-0.0212***
(0.00335)
0.00161
(0.00197)
0.00744***
(0.00231)
-0.00896***
(0.00197)
-0.0161***
(0.00213)
-0.0163***
(0.00220)
-0.0192***
(0.00212)
-0.0176***
(0.00216)
0.0743***
(0.00190)
-0.0189***
(0.00190)
0.00989***
(0.00215)
-0.0248***
(0.00243)
0.00137***
(0.000182)
-6.80e-05
(0.00104)
-0.0210***
(0.00358)
0.000523
(0.00206)
0.00560**
(0.00238)
-0.00950***
(0.00203)
-0.0179***
(0.00219)
-0.0175***
(0.00228)
-0.0200***
(0.00219)
-0.0193***
(0.00222)
0.0769***
(0.00196)
-0.0177***
(0.00176)
0.00918***
(0.00201)
-0.0233***
(0.00228)
0.00125***
(0.000171)
-9.99e-05
(0.000860)
-0.0215***
(0.00328)
0.000891
(0.00197)
0.00557**
(0.00219)
-0.00847***
(0.00184)
-0.0157***
(0.00204)
-0.0150***
(0.00213)
-0.0151***
(0.00211)
-0.0151***
(0.00211)
0.0725***
(0.00181)
-0.0174***
(0.00234)
0.00711**
(0.00277)
-0.0260***
(0.00289)
0.00149***
(0.000212)
-0.000461
(0.000627)
-0.0250***
(0.00545)
-0.00462*
(0.00277)
0.000481
(0.00305)
-0.0138***
(0.00271)
-0.0216***
(0.00285)
-0.0211***
(0.00295)
-0.0251***
(0.00286)
-0.0246***
(0.00290)
0.104***
(0.00246)
12,334
2,595
0.104
0.103
12,334
2,595
0.102
0.101
12,334
2,595
0.099
0.0981
12,334
2,595
0.072
0.0710

*** p<0.01, ** p<0.05, * p<0.10
45
Table 4 (cont)
Regression with Firm Characteristics

TA
REVi,t (12)
P P Ei,t (7)
1
+ 2 Ai,t1
(6) + 1
Ai,t1 (6)
(6)

2
Ai,t1 (6)
46
Table 5
Regression with Firm Characteristics and Fixed Effects
The
table
reports
the
coefficients
from
the
following
regression:
abs(DAi,t )
i + 1 Dividend payeri,t + 2 SOXt Dividend payeri,t + 3 NYSE market capitalizationi,t +

Earnings
Value
4 Assets i,t + 5 Asset growthi,t + 6 Assets i,t + year dummies; where abs(DAi,t ) is the
i,t1
i,t1
absolute value of discretionary accruals estimated by various models separately shown in each
column.
Fixed Effects
(1)
VARIABLES
Exp.
sign
Dividend payer
SOX*Div. payer
NYP
Value /
Lagged assets
Asset growth
Earnings /
Lagged assets
Yr 99
Yr 00
Yr 01
Yr 02
Yr 03
Yr 04
Yr 05
Constant
Observations
Number of id
R-squared
Adj. R-squared
abs(Jones)
(2)
abs(Jones
Mod)
(3)
abs(Jones
w/ ROA)
(4)
abs(KLW
Perform)
-0.00404
(0.00414)
0.0110***
(0.00260)
0.0423***
(0.00894)
0.000716***
(0.000246)
0.000483
(0.000963)
-0.0197***
(0.00527)
-0.00254
(0.00229)
0.00412
(0.00269)
-0.0107***
(0.00243)
-0.0174***
(0.00270)
-0.0197***
(0.00281)
-0.0234***
(0.00271)
-0.0227***
(0.00283)
0.0518***
(0.00356)
-0.00446
(0.00423)
0.0113***
(0.00266)
0.0456***
(0.00933)
0.000690***
(0.000233)
0.000606
(0.000978)
-0.0199***
(0.00594)
-0.00368
(0.00240)
0.00225
(0.00276)
-0.0114***
(0.00249)
-0.0193***
(0.00277)
-0.0211***
(0.00288)
-0.0244***
(0.00279)
-0.0244***
(0.00292)
0.0531***
(0.00372)
-0.00424
(0.00380)
0.0113***
(0.00247)
0.0325***
(0.00874)
0.000598***
(0.000225)
0.000518
(0.000863)
-0.0153***
(0.00443)
-0.00258
(0.00228)
0.00172
(0.00255)
-0.0107***
(0.00228)
-0.0181***
(0.00258)
-0.0194***
(0.00269)
-0.0205***
(0.00266)
-0.0213***
(0.00277)
0.0544***
(0.00347)
-0.00181
(0.00551)
0.00911***
(0.00342)
0.00267
(0.0122)
0.00107***
(0.000320)
4.93e-05
(0.000829)
-0.0169*
(0.00894)
-0.00735**
(0.00325)
-0.00366
(0.00364)
-0.0166***
(0.00341)
-0.0247***
(0.00361)
-0.0262***
(0.00374)
-0.0303***
(0.00371)
-0.0309***
(0.00380)
0.0943***
(0.00481)
12,334
2,595
0.405
0.245
12,334
2,595
0.402
0.242
12,334
2,595
0.401
0.240
12,334
2,595
0.330
0.151

*** p<0.01, ** p<0.05, * p<0.10
47
Table 5 (cont)
Regression with Firm Characteristics and Fixed Effects

TA
REVi,t (12)
P P Ei,t (7)
1
+ 2 Ai,t1
(6) + 1
Ai,t1 (6)
(6)

2
Ai,t1 (6)
48
Table 6
Estimated Probabilities and Pseudo R-squareds from Logit Regressions
The
table
reports
the
estimated
probability
from
the
following
logit
regres-
sion:
logit(Dividend payer
=
1)
=
+ 1 NYSE market capitalizationi,t +
Earningsi,t
Valuei,t
2 Assets
+ 3 Asset growthi,t + 4 Assets
+ industry dummies.
Results are
i,t1
i,t1
reported separately for each year (19982005).
Observations are excluded if they
do not have sufficient data to construct the accrual measures shown in table 7.
1998
1999
Nondiv.
payer
Div.
payer
Pseudo R-sqr.
All observations
Observations
3869
Average
0.156
If probability > 0.5
Observations
263
Average
0.676
0.362
Nondiv.
payer
2000
Div.
payer
Nondiv.
payer
0.364
2001
Div.
payer
Nondiv.
payer
0.368
Div.
payer
0.344
1394
0.562
3685
0.156
1297
0.557
3518
0.146
1124
0.542
3049
0.158
1034
0.534
812
0.776
249
0.672
746
0.769
213
0.668
613
0.766
218
0.655
561
0.751
2002
2003
Nondiv.
payer
Div.
payer
Pseudo R-sqr.
All observations
Observations
2888
Average
0.160
If probability > 0.5
Observations
196
Average
0.664
0.332
Nondiv.
payer
2004
Div.
payer
Nondiv.
payer
0.311
2005
Div.
payer
Nondiv.
payer
0.294
Div.
payer
0.309
974
0.525
2648
0.183
1035
0.531
2526
0.202
1101
0.537
2393
0.211
1154
0.563
513
0.748
227
0.651
550
0.742
242
0.654
614
0.732
259
0.663
694
0.740
49
Table 7
Regression with Firms Estimated to Have Paid a Dividend
The
table
cept
firm-year
ability
of
reports
the
coefficients
observations
greater
than
are
0.5
limited
from
(1)
VARIABLES
Exp.
sign
Dividend payer
SOX*Div. payer
NYP
Value /
Lagged assets
Asset growth
Earnings /
Lagged assets
Yr 99
Yr 00
Yr 01
Yr 02
Yr 03
Yr 04
Yr 05
Constant
Observations
Number of id
R-squared
Adj. R-squared
from
the
the
to
logit
same
those
regression
that
regression
had
results
as
in
an
estimated
shown
abs(Jones)
(2)
abs(Jones
Mod)
(3)
abs(Jones
w/ ROA)
(4)
abs(KLW
Perform)
-0.0129***
(0.00351)
0.0110***
(0.00394)
-0.00714
(0.00441)
-0.00118
(0.000898)
0.0139**
(0.00681)
0.0363**
(0.0179)
-0.00453
(0.00293)
-0.00573*
(0.00334)
-0.00795***
(0.00295)
-0.0202***
(0.00446)
-0.0249***
(0.00439)
-0.0239***
(0.00421)
-0.0236***
(0.00434)
0.0574***
(0.00494)
-0.0137***
(0.00367)
0.0123***
(0.00409)
-0.00766*
(0.00447)
-0.00122
(0.000940)
0.0144**
(0.00716)
0.0362*
(0.0199)
-0.00558*
(0.00307)
-0.00628*
(0.00348)
-0.00883***
(0.00307)
-0.0228***
(0.00464)
-0.0264***
(0.00458)
-0.0253***
(0.00441)
-0.0257***
(0.00454)
0.0599***
(0.00516)
-0.0118***
(0.00356)
0.00989**
(0.00390)
-0.00569
(0.00432)
-0.00126
(0.000942)
0.0111*
(0.00636)
0.0578***
(0.0152)
-0.00427
(0.00285)
-0.00548*
(0.00323)
-0.00711**
(0.00289)
-0.0188***
(0.00430)
-0.0215***
(0.00439)
-0.0212***
(0.00414)
-0.0231***
(0.00428)
0.0527***
(0.00470)
-0.00806
(0.00495)
0.00237
(0.00590)
-0.00189
(0.00630)
-0.000789
(0.00130)
0.0117
(0.00882)
0.0598***
(0.0222)
-0.0116**
(0.00451)
-0.0117**
(0.00512)
-0.0161***
(0.00450)
-0.0221***
(0.00661)
-0.0271***
(0.00666)
-0.0238***
(0.00657)
-0.0258***
(0.00647)
0.0765***
(0.00665)
2,581
656
0.067
0.0627
2,581
656
0.068
0.0630
2,581
656
0.082
0.0773
2,581
656
0.042
0.0370

*** p<0.01, ** p<0.05, * p<0.10
table
in
exprob-
table
6.
50
Table 7 (cont)
Regression with Firms Estimated to Have Paid a Dividend

TA
REVi,t (12)
P P Ei,t (7)
1
+ 2 Ai,t1
(6) + 1
Ai,t1 (6)
(6)

2
Ai,t1 (6)
51
Model Details
Maximizing firm value through earnings announcement policy

The first order conditions for the situation of perfect information were given above.
Now, the second order conditions are tested.
2 V2,
= (1 )f 00 [ + c0 ]2 (1 )f 0 c00
2
a1
Since f 00 < 0, f 0 > 0, and c00 > 0
2 V2,
< 0
2 a1
2 V2,
= (1 )f 00 [ + c0 ][1 + c0 ] + (1 )f 0 c00 > 0
a1 1
a1
f 00 [ + c0 ][1 + c0 ] f 0 c00
=
1
f 00 [ + c0 ]2 f 0 c00
From the first order condition
>0
(1 )f 0
1 + c0 > + c0 > 0
a1
>0
1
+ c0 =
Conditions for a signaling equilibrium for announced earnings

To check the necessary conditions for announcement strategies that maximize compensation, test the first order conditions and cross partials. The subscripts on utility
denote partial derivatives.
UV1
UV1 1
= nM > 0
= 0
V1,0
V2,
+ d(1 )
=0
a1
a1
2 V1,0
2 V2,
Ua1 1 =
+ d(1 )
a1 1
a1 1
2
2
V2,
V1,0
Because
> 0,
>0
a1 1
a1 1
Ua1 1 > 0
Ua1
52
To satisfy the single crossing condition assumption, it must be that Ua1 1 UV1 >
UV1 1 Ua1 .
Ua1 1 UV1 UV1 1 Ua1
h 2V 0
2 V2, i
1,
=
+ d(1 )
nM 0 > 0
a1 1
a1 1
The second order condition must also be tested.

2 V2,
2 V1,0
+
d(1
)
2 a1
2 a1
2 V1,0
2 V2,
Because
<
0,
<0
2 a1
2 a1
Ua1 a1 < 0
Ua1 a1
Proof that a higher dividend lowers the earnings announcement by an

exaggerating high type
To simplify notation, let

min
U1,L a1,L (D1 ), L, D1 = U1,L (L) and U1,L a1,H (D1 ), H, D1 = U1,L (H).
To check how announcements will change, differentiate the incentive constraint with
respect to a change in the dividend. Denote period one capital as KHmin if the firm
received a high shock and the manager needed to exaggerate earnings, as KLmin if
the firm received a low shock and the manager is mimicking, and as KL if the firm
53
received a low shock and the manager makes the announcement optimal for taxes.
U1,L (L) U1,L (H)
U1,L (L) a1,L (D1 ) U1,L (L)
+
a1,L (D1 ) D1
D1
Envelope rule
amin
1,H (D1 )
D1
U1,L (H) amin

U1,L (H)
1,H (D1 )
+
min
a1,H (D1 )
D1
D1
U1,L (L) a1,L (D1 )

=0
a1,L (D1 ) D1
U1,L (H)
U1,L (L)
D
D1
1
U1,L (H)
amin
1,H (D1 )
f 0 (KHmin ) f 0 (KLmin ) < f 0 (KL )

U1,L (L) U1,L (H)
>0
D1
D1
Because
f 0 > 0; f 00 < 0
U1,L (H)
<0
amin
1,H (D1 )
Because
amin
1,H (D1 )
<0
D1
Data Appendix
The sampling method used in this paper generally follows the practices of Fama
and French (2001) and DeAngelo, DeAngelo and Stulz (2006). The broadest initial COMPUSTAT sample uses firm-year data from 1979 to 2008. Firms must be
publicly traded on the NYSE, AMEX, or NASDAQ. Utilities and financial firms are
excluded (SIC codes 4900-4949 and 6000-6999).
Observations must have data for total assets (6,at);37 stock price at the end of
the year (199,prcc f); common shares outstanding (25,csho); income before extraordinary items (18,ib); interest expense (15,xint); dividends per share by ex date
(26,dvpsx f); and (a) preferred stock liquidating value (10,pstkl), (b) preferred stock
37
The parenthetical notation contains the COMPUSTAT annual number code and the WRDS
data code, respectively.
54
redemption value (56,pstkrv), or (c) preferred stock carrying value (130,pstk). Firms
must have book equity as defined below. Observations are also required to have total
assets at the beginning of the year. Observations with total assets below $500,000
or book equity below $250,000 were excluded. To ensure that firms are publicly
traded, the firms must have share codes of 10 or 11 in the CRSP database by fiscal
year-end.
Discretionary Accrual Data
Observations must also have the data needed for the discretionary accrual regressions: change in sales (12,sale); change in receivables (2,rect); plant, property, and
equipment, gross (7,ppegt); change in current assets (4,act); change in cash (1,che);
change in current liabilities (5,lct); change in current maturities of long-term debt
(34,dlc); and depreciation and amortization expense (14,dp). Ten firm-year observations for a two-digit SIC code are needed to run the discretionary accrual regression.
Due to this constraint, only firms with fiscal year-ends the same as the calendar yearend are used.
Derived Variables
Dividend Payer = 1, if the firm had a dividend by ex-date in the current year
(=0, otherwise)
Preferred stock = preferred stock liquidating value (10,pstkl) [or preferred stock
redemption value (56,pstkrv), or preferred stock par value (130,pstk)]
Book equity = stockholders equity (216,seq) [or common equity (60,ceq) + preferred equity, or total assets (6,at) total liabilities (181,lt)] preferred stock
+ balance sheet deferred taxes and investment tax credit (35,txditc) if available
Market capitalization = stock price at the end of the year (199,prcc f) common
shares outstanding (25,csho)
Market value of the firm (Value) = total assets (6,at) book equity + market
capitalization
Earnings (E) = income before extraordinary items (18,ib) + interest expense
55
(15,xint) + income statement deferred taxes (50,txdi) if available
Asset growth =
At
At1
Return on assets (ROA) =
Et
0.5(At +At1 )

The Relationship Between Information Asymmetry and Dividend PolicyPap

Uploaded by

Copyright:

Available Formats

The Relationship Between Information Asymmetry and Dividend PolicyPap

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

The Relationship Between Information Asymmetry and Dividend PolicyPap

Uploaded by

Copyright:

Available Formats

Finance and Economics Discussion Series

Divisions of Research & Statistics and Monetary Affairs

The Relationship Between Information Asymmetry and Dividend

The Relationship Between Information Asymmetry

Division of Monetary Affairs, [email protected]. I owe special thanks to my adviser

This figure is based on firms that are in the COMPUSTAT database.

This definition is based on Schipper (1989).

Stylized Facts and Background

Reported Earnings Behavior

Background on the SarbanesOxley Act of 2002

U.S. Congress (2002) Sec. 906.

Overview and Set-up

True Earnings, Earnings Announcements, and Firm Valuation

where [0, 1) is the depreciation rate of capital and 1 is a production shock

Attention focuses on stable equilibria by restricting out-of-equilibrium beliefs, which eliminate

High Production Shock Manager

The announcement by a high-shock manager then satisfies

Board Dividend Policy

on firm value. Dividends lower announcements, lowering EM costs.

U1,L (a1,L ,L)

Sudden Decrease in the Size of Information Asymmetry

Now assume there is a large decrease in the amount of information asymmetry

Testing for Earnings Management

Estimates of Discretionary Accruals

similar to those reported in this paper.

Initial Results of Discretionary Accruals Testing

Regressions with Discretionary Accruals

NYSE market capitalization is a proxy of size. It is equal to the percentage of NYSE

logit(Dividend payer = 1) = + 1 NYSE market capitalizationi,t

Dividend policy and the

Total accrual/ Lagged assets

Discretionary accruals (scaled by lagged assets)

Robust standard errors in parentheses are clustered at the firm level.

+ 1 Dividend payeri,t + 2 SOXt Dividend payeri,t + 3 NYSE market capitalizationi,t +

Robust standard errors in parentheses are clustered at the firm level.

i + 1 Dividend payeri,t + 2 SOXt Dividend payeri,t + 3 NYSE market capitalizationi,t +

Robust standard errors in parentheses are clustered at the firm level.

Robust standard errors in parentheses are clustered at the firm level.

Maximizing firm value through earnings announcement policy

Conditions for a signaling equilibrium for announced earnings

The second order condition must also be tested.

Proof that a higher dividend lowers the earnings announcement by an

U1,L (H) amin

U1,L (L) a1,L (D1 )

f 0 (KHmin ) f 0 (KLmin ) < f 0 (KL )

Return on assets (ROA) =

You might also like