Zang (2012)

46
 Evidence on the Tradeoff between Real Manipulation and Accrual Manipulation Amy Y. Zang* I study whether managers use real and accrual manipulations as substitutes in managing earnings, and I study the order that managers make these decisions. I find that managers determine real manipulation  before accrual manipulation. Based on this result, I use an empirical model that captures the sequentiality of real and accrual manipulatio ns to test the tradeoffs betwee n the two. The results of the broad s ample tests are consistent with manag ers using real and accrual manipulations as substitutes. In a small sample test examining firms subject to securities class action lawsuits, I examine whether real and accrual manipulations change ov er time with changes in litigation ris k. Consistent with manager s using real and accrual manipulations as substitutes, I find that managers switch from accrual manipulation to real manipulation after lawsuit filings. Draft: December 2005 Comments welcome. *Fuqua School of Bus iness, Duke Univers ity, Box 90120, Durham, NC 27708. Email: yz15@du ke.edu. I am grateful for guidance from Jen nifer Francis, Qi Chen, Dhananj ay Nanda and Per Olsson. I thank Moshe Bareket, Allen Huang, Yvonne Lu, Shiva Rajgopal, and Mohan Venkatachalam for helpful suggestions. I appreciate the financial support f rom the Fuqua School of Business at Duke Univers ity and the Deloitte Foundati on. Errors and omissions a re my responsibility.

Transcript of Zang (2012)

Page 1: Zang (2012)

8/12/2019 Zang (2012)

http://slidepdf.com/reader/full/zang-2012 1/46

 

Evidence on the Tradeoff between Real Manipulation and Accrual Manipulation

Amy Y. Zang*

I study whether managers use real and accrual manipulations as substitutes in managing earnings, and I

study the order that managers make these decisions. I find that managers determine real manipulation

 before accrual manipulation. Based on this result, I use an empirical model that captures the sequentiality

of real and accrual manipulations to test the tradeoffs between the two. The results of the broad sample

tests are consistent with managers using real and accrual manipulations as substitutes. In a small sample

test examining firms subject to securities class action lawsuits, I examine whether real and accrual

manipulations change over time with changes in litigation risk. Consistent with managers using real and

accrual manipulations as substitutes, I find that managers switch from accrual manipulation to real

manipulation after lawsuit filings.

Draft: December 2005

Comments welcome.

Page 2: Zang (2012)

8/12/2019 Zang (2012)

http://slidepdf.com/reader/full/zang-2012 2/46

1.  Introduction

I study earnings management through both real activities management (RM) and discretionary

accruals management (AM) for a broad sample of firms over 1992-2003 and for a small sample of lawsuit

firms over 1995-2004. By real activities management, I mean a purposeful action to alter reported

earnings in a particular direction, which is achieved by changing the timing or structuring of an operation,

investment or financing transaction, and which has sub-optimal business consequences. The idea that

firms engage in RM is supported by Graham, Harvey, and Rajgopal’s [2005] survey evidence

documenting the widespread use of earnings management, especially as it concerns real manipulation.1 

They report that 80% of surveyed CFOs state that, in order to deliver earnings, they would decrease R&D,

advertising, and maintenance expenditures, while 55% said they would postpone a new project. Despite

survey and anecdotal evidence concerning real manipulation, most empirical earnings management

research focuses on accruals management.2 

The focus of my paper is on how managers trade off real and accrual manipulations based on their

relative costs. This question is important for two reasons. First, as mentioned by Fields, Lys and Vincent

[2001], examining only one earnings management technique at a time cannot explain the overall effect of

earnings management activities. In particular, if managers use RM and AM as substitutes, examining

either type of manipulation in isolation cannot lead to definitive conclusions. Second, by studying how

managers trade off RM and AM, my paper sheds light on the economic implications of accounting

choices; that is, whether the costs that managers bear for manipulating accruals affect their decisions

about real manipulations. As such, the question has implications about whether enhancing SEC scrutiny

or reducing accounting flexibility in GAAP might, for example, increase managers’ levels of real

Page 3: Zang (2012)

8/12/2019 Zang (2012)

http://slidepdf.com/reader/full/zang-2012 3/46

manipulation.

I start with a cost-benefit analysis that provides a structure for empirical tests of the relation between

RM and AM. This analysis produces different models depending on whether RM and AM are determined

simultaneously or sequentially. My empirical tests show that RM and AM are determined sequentially:

specifically, the RM decision precedes the AM decision. Based on this result, I further test the

substitutive relation between RM and AM using an empirical model that controls for this sequentiality. In

testing managers’ tradeoff of RM and AM, I first identify cost determinants of each. I then use a broad

sample of firms to show that the relations between RM, AM and their cost determinants are consistent

with managers using both types of manipulations as substitutes. The substitutive relation is further

supported by a small sample test examining firms experiencing securities class action lawsuits. I find that

subsequent to the filing of the lawsuit, sued firms’ AM levels drop abruptly, while their RM levels (via

cutting R&D expenditure and overproduction) increase significantly. These results are consistent with my

 prediction that managers turn to RM as a response to increased litigation risk and outside scrutiny.

My paper contributes to the earnings management literature in several ways. First, it is one of a few

studies (Hunt, Moyer and Shevlin [1996], Beatty, Chamberlain and Magliolo [1995], Gaver and Paterson

[1999], Barton [2001], Pincus and Rajgopal [2001]) to examine how managers use multiple tools to

manage earnings. It is also one of the first to specifically examine how managers trade off RM and AM.

Second, I extend the prior literature by incorporating cost factors for RM and AM in the empirical tests. I

argue that the main costs of RM are the economic consequences of deviating from optimal business

operations and therefore, jeopardizing the firm’s competitive advantage; in contrast, AM is costly

 primarily due to auditor and regulators’ scrutiny and litigation risk. Although many prior studies of

multiple accounting choices recognize the importance of the relative costs of different accounting (and/or

Page 4: Zang (2012)

8/12/2019 Zang (2012)

http://slidepdf.com/reader/full/zang-2012 4/46

substitutive relation between RM and AM.

Third, by empirically testing the simultaneity/sequentiality of real and accrual manipulations, my

 paper expands our understanding of managers’ decision processes. Most prior studies on multiple

accounting (and/or economic) choices implicitly assume that managers use multiple choices

simultaneously and treat all of the choice variables as endogenous, without testing the sequential decision

 process as an alternative hypothesis (one exception is Pincus and Rajgopal [2001]).4  In contrast, I

explicitly model and test the sequentiality versus simultaneity of the two manipulations, and show

evidence consistent with managers deciding the RM decision before the AM decision.

Finally, my paper is among the first to validate RM proxies measured by cross-sectional estimation

models. In the first validity test, I find that real transaction manipulators manipulate more in the fourth

fiscal quarter than in other fiscal quarters. This evidence is consistent with managers having, in the fourth

quarter, both more information about the total amount of earnings management needed and more

incentive to manipulate. In the second validity test, I use a performance matching procedure and show

that real transaction manipulators are associated with negative abnormal performance in subsequent years,

consistent with RM proxies capturing suboptimal business decisions.

The rest of the paper is organized as follows. In section 2, I develop a simple model to show how

managers trade off real and accrual manipulations; I then use this model to develop hypotheses. Section 3

constructs the proxies for real and accrual manipulations, and section 4 conducts validity tests of the RM

 proxies. Section 5 contains the research design, measurement of independent variables, sample selection,

and empirical results for the broad sample test. The small sample test is presented in section 6. Section 7

concludes and discusses implication of my results.

Page 5: Zang (2012)

8/12/2019 Zang (2012)

http://slidepdf.com/reader/full/zang-2012 5/46

 benefits of earnings management, the levels of RM and AM, and their cost determinants. Some stylized

simplifying assumptions are made, but I expect the basic results to hold in a more general setting where

empirical tests are conducted.5 

I assume that managers obtain benefits (denoted B) from earnings management, with the benefit

function being concave and twice continuously differentiable. A concave benefit function means that the

marginal benefit is decreasing in the amount of earnings being managed. For simplicity and without loss

of generality, I assume a quadratic function for B, B = a·K – β·K 2/2, where K is the total amount of

earnings being managed, and α and β > 0. Therefore, the marginal benefit of earnings management at K

can be expressed as MBK  = ∂B/∂K = α – β·K. I assume that managers have two techniques, AM and RM,

to achieve an expected level of K. Let the costs of implementing AM and RM be denoted as CA and CR ,

respectively. Again, for simplicity and without loss of generality, I assume a quadratic cost function for

 both CA and CR : CA = a·AM2/2 and CR  = b·RM2/2, with a > 0 and b > 0. Therefore, the marginal cost of

AM is MCA = a·AM, and the marginal cost of RM is MCR  = b·RM.

To capture the possibility that realized earnings may differ from managers’ choices of AM or RM due

to exogenous shocks, I assume that realized earnings management K is the sum of managers’ choice of

RM and AM, and their disturbance terms u and v : K = RM +u + AM + v , where u is the exogenous

shock to RM and v is the exogenous shock to AM.

Managers’ decisions with respect to the optimal levels of RM and AM depend on whether managers

can observe these exogenous shocks when they make their decisions. Prior studies investigating multiple

earnings management tools (Hunt, Moyer and Shevlin [1996], Beatty, Chamberlain and Magliolo [1995],

Gaver and Paterson [1999], Barton [2001], Pincus and Rajgopal [2001]) implicitly assume that managers

make the manipulation decisions simultaneously without observing any exogenous shocks However if

Page 6: Zang (2012)

8/12/2019 Zang (2012)

http://slidepdf.com/reader/full/zang-2012 6/46

simultaneously and, separately, under the assumption they are determined sequentially. As I will show,

the assumption of simultaneity or sequentiality leads to different empirical models for how managers

trade off RM and AM. I leave the test of this assumption to section 5.

In the first scenario, assuming that managers cannot observe any shocks before their AM aqnd RM

decisions, managers will choose their optimal levels of RM and AM simultaneously, as follows:

R K 

A K 

MC MB

MC MB

=⎧⎨

=⎩

**

**

α-β AMRM =

 b+β

α-β RMAM =

a+β

⎧   ⋅⎪⎪

⇔ ⎨⋅⎪

⎪⎩

. (I)

 Note that the marginal cost of the optimal level of AM (AM*) and the marginal cost of the optimal level

of RM (RM*) simultaneously equal the marginal benefit of earnings management (i.e., MCA = MCR  =

MBK ), since managers want to be cost efficient.

In the second scenario, I assume that managers make RM and AM decisions sequentially because it is

likely that real transactions are performed during the fiscal year but accrual manipulation is conducted

close to or after the fiscal year end. If this is the case, a shock that affects RM (u ) during the fiscal year

can affect AM* by changing the marginal benefit of earnings management (MBK ). In contrast, a shock

that affects AM ( v ) after the fiscal year end cannot change RM*, since RM is already realized. In other

words, there is no feedback from AM to RM*, since RM is the lower level of the causal chain.

Assuming managers can observe the exogenous shocks to RM (u, the realized value of u that

managers can observe when making the AM decision) before deciding the optimal level of AM (AM*),

we can obtain the following:

R K MC MB=⎧⎨

( )

( )

aRM

a b a b

α ⋅⎧ =⎪ ⋅ + + ⋅β⎪⇔ ⎨

*

(II)

Page 7: Zang (2012)

8/12/2019 Zang (2012)

http://slidepdf.com/reader/full/zang-2012 7/46

shocks to RM (u) for their AM decisions. Thus, AM* will depend on realized RM (RM* + u) and other

exogenous variables including benefit factors (α and β), and its own cost factor (a).

The above cost-benefit analysis has the following implications. First, simple comparative analysis

shows that RM and AM are negatively correlated with their own cost factors and positively correlated

with earnings management incentives. Second, the substitutive relation between RM and AM is manifest

 by a positive relation between the amount of one manipulation technique and the cost of the other because

managers trade off the two until the marginal costs are equal (i.e., MCA = MCR ). Note that, for testing the

tradeoff between RM and AM, it is important to separate the benefits of earnings management from the

costs of these manipulations. The reason is that managers trade off RM and AM according to their

marginal costs (which differ) but not their marginal benefit (which are the same, i.e., a $1 increase in RM

and $1 increase in AM both result in $1 increase in reported earnings). Third, the fact that real and

accrual manipulations are jointly determined does not necessarily imply they are simultaneous. In

 particular, the two can be joint but managers can determine RM first (based on the cost functions of both

manipulation techniques and the benefit function of an earnings increase) and then determine AM (based

on realized RM, AM’s cost and incentives). Fourth, irrespective of whether the two decisions are made

sequentially or simultaneously, AM is negatively correlated with RM as indicated by the negative sign on

RM in the AM function. These findings lead to the following hypotheses:

 H1: Both RM and AM are negatively correlated with their own cost determinants.

 H2: The level of RM is positively correlated with the cost determinants of AM.

 H3: The level of AM is negatively correlated with the level of RM, controlling for the cost determinants of

 AM and the incentives to increase earnings.

Page 8: Zang (2012)

8/12/2019 Zang (2012)

http://slidepdf.com/reader/full/zang-2012 8/46

general and administrative expenditures, overproducing inventory to reduce the cost of goods sold, and

selling fixed assets with a market value greater than book value to report a gain.

Following Berger [1993] and Gunny [2005], I estimate the normal level of R&D expenditure using

equation (1). The regression is estimated cross-sectionally for each industry-year with at least 15

observations, where industry is defined following Fama and French [1997]:6

 

t t 1 t t0 1 2 3 t 4 t

t 1 t 1 t 1 t 1

RD RD Funds CapitalExpTobinsQ

A A A A

− − − −

= α + α + α + α + α + ε   (1) where, RDt = R&D expense = Data 46;

At = Total assets = Data 6;

Fundst = Internal funds = IBEI + R&D + Depreciation = Data 18 + Data 46 + Data14;

TobinsQt = (MVE + Book value of preferred stock + Long-term debt + Short-term debt)/ Total assets = (Data 199μData 25 + Data 130 + Data 9 + Data 34)/Data 6;

CapitalExpt = Capital expenditure = Data 128.

Lag R&D expense (RDt-1) proxies for the firm’s innovation opportunity; the coefficient on this

variable is expected to be positive. Internal fund (Fundst) is included based on the argument that

expanding R&D investment is cheaper for firms with more internal funds since external funds are more

expensive for R&D projects than internal funds. Therefore, I expect a positive coefficient on this variable.

Tobin’s Q (TobinsQt) is measured as average Q, which captures the firm’s growth potential. Similarly,

capital expenditure (CapitalExpt) represents the firm’s investing activities in the current year. I expect the

coefficients on both Tobin’s Q and capital expenditure to be positive.

Following Anderson, Banker, and Janakiraman [2003], I estimate the normal level of selling, general

and administrative (excluding R&D expense) expenses using equation (2):

t t t t-1 t-11 2 3 t 4 5 t-1 t

t-1 t-1 t-1 t-2 t-2

SG&A S S S SLog =α +α Log +α Log DS +α Log +α Log DS +ε

SG&A S S S S

⎛ ⎞ ⎛ ⎞ ⎛ ⎞ ⎛ ⎞ ⎛ ⎞× ×⎜ ⎟ ⎜ ⎟ ⎜ ⎟ ⎜ ⎟ ⎜ ⎟

⎝ ⎠ ⎝ ⎠ ⎝ ⎠ ⎝ ⎠ ⎝ ⎠  (2) 

Page 9: Zang (2012)

8/12/2019 Zang (2012)

http://slidepdf.com/reader/full/zang-2012 9/46

SG&A expenditures when sales decrease. They also find that this stickiness in SG&A costs reverses in

subsequent periods. Following their work, I use the ratio form and log specification to improve the

comparability of the variables across firms and to mitigate heteroskedasticity. DSt, a dummy variable

equal to one when sales are deceasing and zero otherwise, is included to capture the asymmetric relation

 between SG&A costs and sales levels when sales increase and decrease. I expect the coefficient on

Log(St/St-1) to be positive since the change in SG&A is generally positively correlated with the change in

sales. The coefficient on Log(St/St-1)·DSt is expected to be negative because of the stickiness of SG&A

costs. The coefficient on Log(St-1/St-2), which captures the lagged adjustment to SG&A for the change in

sales, is expected to be positive. I expect the coefficient on Log(St-1/St-2)·DSt-1 to be positive, reflecting

the reversal of SG&A stickiness.

I use Roychowdhury’s [2004] model to estimate the normal level of production costs:

t t t t-112 3 4 t

t-1 t-1 t-1 t-1 t-1

Prod S   ΔS   ΔSα= +α +α +α

A A A A A+ ε   (3) 

where, Prodt  = COGSt + ΔInventoryt = Data 41 + ΔData 3;

St = Net sales = Data 12;ΔSt = St – St-1.

Roychowdhury [2004] develops equation (3) based on Dechow, Kothari, and Watts [1998], who model

COGS and change in inventory as a linear function of sales and change in sales. Again, the regression is

estimated for each industry-year with at least 15 observations. According to Dechow et al. [1998]’s

model, all of the coefficients, except that onΔSt-1/At-1, are expected to be positive.

I follow Gunny [2005]’s model to estimate the normal level of gains on asset sales:

t 0 t t t1 2 3 t

t 1 t 1 t 1 t 1 t 1

GLA PPESales ISales S

A A A A A− − − − −

α Δ= + α + α + α + ε   (4) 

Page 10: Zang (2012)

8/12/2019 Zang (2012)

http://slidepdf.com/reader/full/zang-2012 10/46

investments (GLAt) is modeled as a linear function of sales of PPE and investment, both of which are

expected to have positive coefficients. Change in sales is included to control for firm growth, because

growth firms tend to be younger firms who are less likely to recognize gains from selling assets.

Table 1, Panel A reports the estimation results for equations (1) to (4). All of the equations are

estimated cross-sectionally for each industry-year over 1988-2004. The mean coefficients, number of

observations, and adjusted R 2 across all the industry-years are reported. The t-statistics are calculated

 based on the standard errors of the mean coefficients. All of the mean coefficients (except for the

coefficient on Log(St-1/St-2)·DSt-1) are significant and have the expected signs. The mean adjusted R 2 

ranges from 26% for the gain on asset sales model to 94% for the production cost model, indicating

reasonable to substantial explanatory power for all models.

I measure the abnormal level of each transaction (R&D, SG&A, Prod, and GLA) as the residual from

the relevant estimation model.8  I multiply the residuals from the estimation models of R&D, SG&A, and

GLA by negative one, such that higher values (denoted as Ab_RD, Ab_SGA, and Ab_GLA) indicate a

higher probability that firms cut R&D, SG&A expenses and incur abnormally lower gains from asset

sales to increase reported earnings. Additionally, the higher the abnormal production cost (Ab_Prod,

measured as the residual from the production cost model), the more likely that managers overproduce

inventories to reduce reported cost of goods sold. In measuring real manipulation with assets sales

(Ab_GLA), I impose an additional restriction that the gains from asset sales (GLA) themselves be

 positive. Therefore, firms with lower abnormal gains from asset sales (i.e. higher Ab_GLA) and positive

total gains from asset sales (GLA) are more likely to have engaged in real manipulation.9 

8 Note that the residual of equation (2) is in log form The following transformation derives the Ab SGA from the

Page 11: Zang (2012)

8/12/2019 Zang (2012)

http://slidepdf.com/reader/full/zang-2012 11/46

I use Dechow, Richardson and Tuna’s [2003] (hereafter, DRT) model to estimate the normal level of

total accruals (i.e., nondiscretionary accruals):10 

( ) j t j t j t j t j t 1 j t 101 2 3 4 j t

 j t 1 j t 1 j t 1 j t 1 j t 2 j t

TAC 1 k S REC PPE TAC S

A A A A A S

 , , , , , ,

 ,

 , , , , , ,

− +

− − − − −

+ Δ − Δ Δα= + β + β + β + β + ε   (5) 

where, TACt = Total accruals = Data 123 – Data 308;

ΔSt = Salest – Salest-1;

ΔRECt  = Change in account receivable = ΔData 2;

k = Estimated slope coefficient from a regression of ΔREC on ΔSales for eachFama-French industry-year grouping, i.e., ΔREC = a + k·ΔS + ε;

PPEt = Property, plant and equipment = Data 8.

Similar to the estimation models for the normal level of real transactions, equation (5) is estimated

cross-sectionally for all Fama-French industry-years with at least 15 observations. DRT improve the

cross-sectional modified Jones model (Dechow, Sloan and Sweeney [1995], hereafter DSS) in several

ways. First, DRT’s model does not assume that all credit sales are discretionary. Instead, it estimates k as

the slope coefficient ofΔREC on ΔSales for each industry-year; k captures the expected change in credit

sales for a given change in sales. Second, lag total accruals (TACt-1/At-2) is included to capture the

 predictable portion of total accruals. Third, the model is forward-looking since it includes next-year’s

sales growth (ΔSt+1/St) to incorporate the increase in inventory that is due to growth prospects.

Panel B, Table 1 reports the estimation results for equation (5). Results of DSS’s modified Jones

model are also provided for comparison purposes. Consistent with DRT, I find that the average adjusted

R 2 of the DRT model (of 44.77%) is higher than that of the DSS model (of 35.50%). All of the estimated

coefficients are significant with the same signs as found by DRT. I use the residuals from equation (5) to

measure the abnormal level of accruals (Ab_Accr). I assume that the higher the abnormal accrual, the

more likely managers have managed accruals.

Page 12: Zang (2012)

8/12/2019 Zang (2012)

http://slidepdf.com/reader/full/zang-2012 12/46

skewness and kurtosis of Ab_GLA are due to the requirement that raw GLA must be positive. The

Pearson correlations among the variables are shown in Panel D, Table 1. Recall that the larger the

Ab_RD, Ab_SGA, and Ab_Prod, the higher the probability of RM. Therefore, the significant positive

correlations between Ab_Accr and Ab_RD, between Ab_Accr and Ab_SGA, and between Ab_Accr and

Ab_Prod, indicating positive relations between AM and RM achieved with R&D, SG&A and

overproduction. The positive correlation between RM and AM is explained by both RM and AM being

 positively correlated with incentives to manage earnings.

4.  Validity Tests of Real Manipulation Proxies 

Before I test the main hypotheses of the paper, I provide two validity tests for the RM proxies

(Ab_Rd, Ab_SGA, Ab_Prod, Ab_GLA) developed in the previous section. Validity tests of these

measures are important both because of the conceptual difficulty in defining RM and because of the

 paucity of prior research on RM. The first validity test considers the timing of managers’ RM decisions.

 Note that RM is expected of managers during the fiscal year. When a manager is making the RM

decision, presumably two conditions should be met. The first is that the manager has strong incentives to

manipulate earnings for the current quarter; the second is that he has gathered adequate information about

 both the true earnings performance and the market’s expectation to estimate how far unmanipulated

earnings are from the earnings target – in order to determine the amount of RM needed. Given these

requirements, I argue that managers are more likely to perform RM during the fourth fiscal quarter than in

the other fiscal quarters.

Along the same lines, several prior studies use quarterly patterns in profit margins to detect potential

earnings management For example Thomas and Zhang [2002] use unusual patterns of COGS scaled by

Page 13: Zang (2012)

8/12/2019 Zang (2012)

http://slidepdf.com/reader/full/zang-2012 13/46

Taking firms in the top quintile of the RM measures (Ab_RD, Ab_SGA, Ab_Prod, and Ab_GLA) as

“suspect RM firms,” I predict they have less R&D, SG&A expenses, and gains from asset sales, and more

 production costs, in the fourth fiscal quarter than the other fiscal quarters, compared with other firms in

the same industry-year. I construct the following model to estimate the quarterly pattern of these real

transactions:4 4 12

f k f c

q,t q q t q q m m t

q=1 q=1 m=1

X =   γ D TopRM + f D + c D +ε× × × ×∑ ∑ ∑   (6) where, Xq,t  œ  {QRDq,t/At-1, QSGAq,t/At-1, QProdq,t/At-1, QGLAq,t/At-1};

QRD = Quarterly R&D expenses = Data 4;

QSGA = Quarterly SG&A expenses = Data 1 – Data 4;

QProd = Quarterly production costs = Data 30 + change in Data 3 from last quarter;QGLA = Quarterly gains from asset sales = – Data 102;

Df q  = Dummy variable equals 1 if it is fiscal quarter q (q=1 to 4), zero otherwise;

Dcm  = Dummy variable equals 1 if it is calendar month m (m = 1 to 12), zero otherwise;

TopRMk t = Dummy variable equals 1 if the firm’s Ab_RD (Ab_SGA, Ab_Prod, or Ab_GLA) is

in the top quintiles in its industry-year. For example, if the independent variable isQRDqt/At-1 from fiscal quarter 2, TopRM is a dummy variable equals one if this

firm’s annual Ab_RD is in the top quintile of Ab_RD of the same industry-year.Here, k = RD, SGA, Prod, or GLA.

Equation (6) is estimated cross-sectionally for each Fama-French industry-year that satisfies the

following requirements: (1) at least 15 observations; (2) fewer than 80% of the firms have a December

fiscal year end; (3) there are at least three different fiscal year ends among the firms.11 The dependent

variable is the quarterly amount of real transactions scaled by lagged annual assets. Dummy variables for

calendar months (Dcm) are used to control for calendar seasonality; dummy variables capturing fiscal

quarters (Df q) are used to proxy for average quarterly patterns in real transactions. TopRMk 

t equals one if

the firm’s RM proxy (Ab_RD, Ab_SGA, Ab_Prod, and Ab_GLA) is in the top quintile in its industry-year.

Page 14: Zang (2012)

8/12/2019 Zang (2012)

http://slidepdf.com/reader/full/zang-2012 14/46

these firms have abnormally low annual R&D, SG&A expenses and gains from asset sales. For

overproducing firms, I expect to find positive values of γ1 to γ4, since they have abnormally high annual

 production costs. Under the view that managers are more likely to use RM in the fourth fiscal quarter

than in the other fiscal quarters, the focus of the tests is on the difference between γ4 and the mean of γ1 ,

γ2 and γ3. If suspect RM firms cut R&D, SG&A expenses, or sell assets at abnormally lower prices

(overproduce inventories) in the fourth fiscal quarter, I expect to find that γ4 is smaller (larger) than the

mean of γ1 , γ2 and γ3. Table 2, Panel A reports the results of this test. For quarterly R&D expense

(production costs), the results are consistent with the prediction that γ1 to γ4 are negative (positive), and

that γ4 is smaller (larger) than the mean of γ1, γ2 and γ3 (significant at the 0.001 level). These results

indicate that not only do the suspect RM firms have lower (higher) R&D expense (production costs)

throughout the year than the industry average, they cut R&D expense (overproduce inventories)

significantly more in the fourth fiscal quarter than in the first three fiscal quarters. I do not, however, find

significant results for the SG&A or asset selling manipulation proxies.

My second validity test examines whether suspect RM firms are associated with adverse subsequent

 performance. Gunny [2005] also studies the consequence of RM and finds negative performance-

matched operating performance for firms she identifies as real manipulators for years subsequent to the

RM behavior. She identifies real manipulators as firms with abnormal real transactions but little accrual

manipulation flexibility. My tests are similar to hers, but our definitions of suspect RM firms differ,

 because I argue that firms that engage in RM can conduct AM at the same time, as evidenced by the

 positive correlation between RM and AM proxies in Table 1, Panel D. In particular, I identify a

 performance-matched control firm for each suspect RM firm using the following procedure: (1) the

control firm is in the same industry year as the suspect firm; (2) it is selected from the bottom three

Page 15: Zang (2012)

8/12/2019 Zang (2012)

http://slidepdf.com/reader/full/zang-2012 15/46

measured as the difference between the performance measure of the suspect firm and that of the control

firm. Two performance measures are examined, return on assets (ROA) and cash flow from operation

scaled by assets (CFO/At-1). I expect to find subsequent negative abnormal performance as the

consequences of suboptimal decisions are revealed.

Table 2, Panel B reports the results of this second validity test. Note that all of the abnormal

 performance measures in year t are around zero, indicating the success of the matching procedure.

Consistent with RM having subsequent adverse economic consequences, most of the subsequent

abnormal ROA and scaled CFO in year t+1 to t+3 for firms in the top Ab_SGA and Ab_Prod quintiles are

negative (significance levels of 0.005 or better). Also consistent with predictions, the subsequent

abnormal ROAs from year t+1 to t+3 are significantly negative for firms in the top Ab_RD quintile.

In sum, the first validity test provides support for Ab_RD and Ab_Prod as capturing RM; the second

validity test provides support for Ab_RD, Ab_SGA and Ab_Prod as capturing RM. These tests increase

confidence in the results reported in the next section which focus on the tradeoffs between RM and AM.

However, neither validity test provides support for Ab_GLA as capturing RM. The latter finding is

consistent with Graham et al. [2005] who report little evidence that managers use timing of asset sales to

manage earnings. For these reasons, I do not include Ab_GLA in subsequent empirical tests.

5.  Empirical Tests – Broad Sample Analysis

5.1. Research design

As discussed in section 2, whether managers determine RM and AM simultaneously or sequentially

leads to different empirical models for testing the tradeoffs between them. In order to identify the correct

empirical model for testing my hypotheses I first test the assumption of the simultaneity/sequentiality of

Page 16: Zang (2012)

8/12/2019 Zang (2012)

http://slidepdf.com/reader/full/zang-2012 16/46

of simultaneity of RM and AM decisions. Since RM and AM are assumed to be endogenous, I estimate

(7) using two-stage least squares. Equation (7) can be identified because the cost determinants in the two

equations differ. I test the simultaneity/sequentiality of RM and AM with the Hausman test. As discussed

in section 2, if RM and AM are simultaneous, RM should be endogenous in the AM equation, i.e.,

correlated with the error term of the AM equation; hence, the Hausman test should reject exogeneity of

RM in the AM equation. On the other hand, if RM is determined before AM, there is no feedback from

AM to RM, which means RM should be orthogonal to the error term of the AM equation; that is, the

Hausman test should fail to reject the exogeneity of RM in the AM equation. Moreover, if managers

make AM decisions based on realized RM, the Hausman test should reject the exogeneity of AM in RM

equation since AM will be correlated with the disturbance term of RM.

If the Hausman test rejects simultaneity, I further test H1-H4 using the following recursive

simultaneous equation system which captures the sequentiality of RM and AM:

i,t 0 1,j j,i,t 2,p p,i,t 3,k k,i,t 4,l l,i,t i,t

 j p k l

RM = + Cost of RM + Cost of AM + Incentives + Controls +uλ λ λ λ λ∑ ∑ ∑ ∑ ,

i,t 0 1 i,t 2,p p,i,t 3,k k,i,t 4,l l,i,t i,t

 p k l

AM = + RM + Cost of AM + Incentives + Controls +vδ δ δ δ δ∑ ∑ ∑ .(8) 

 Note that in this recursive equation system, RM is predetermined by the costs of RM and AM, as well

as incentives. On the other hand, the AM equation in equation system (8) has RM as an independent

variable. Note that, under the assumption of sequentiality, when managers determine the level of AM,

they observe the realized RM. Hence, the RM variable in the AM equation is exogenous. Since equation

system (8) is a recursive equation system, it can be estimated consistently with OLS.

H1 predicts that in system (8), both RM and AM are negatively correlated with their own cost

determinants. H2 predicts that in the RM equation, RM variable is positively correlated with the cost

Page 17: Zang (2012)

8/12/2019 Zang (2012)

http://slidepdf.com/reader/full/zang-2012 17/46

the cost determinants of manipulations and for earnings management incentives. For RM, I identify five

such proxies. The first proxy, the total benefit of R&D expenditure (TotalBenefit_RD), is based on Lev

and Sougiannis [1996], who study the relation between R&D expenditure and future earnings. They

develop a procedure to measure the total effect of $1 invested R&D on current and future operating

income. When managers cut R&D expenditure to increase current earnings, they forgo the current and

future benefits from the R&D investment that is cut. Hence, the forgone benefit is a cost determinant for

R&D manipulation. Following Lev and Sougiannis [1996], I measure the total benefits of R&D

expenditure to current and future earnings using the following model:13 

0 1 2,k 3 i,t

i,t i,t-1 i,t-k i,t-1k 

OI TA RD AD= α +α +   α +α +e

S S S S

⎛ ⎞ ⎛ ⎞ ⎛ ⎞ ⎛ ⎞⎜ ⎟ ⎜ ⎟ ⎜ ⎟ ⎜ ⎟⎝ ⎠ ⎝ ⎠ ⎝ ⎠ ⎝ ⎠

∑ , (9) where, OI = Annual operating income, before depreciation, advertising and R&D expenses;

TA = The value of plant and equity, inventory and investment in unconsolicated subsidiariesand goodwill, in current dollars;

RD = Annual R&D expenditures, in current dollars;

AD = Annual advertising expenses, in current dollars.

To address the potentially severe collinearity problem in equation (9) due to the inclusion of lagged R&D

expenses as independent variables, an Almon lag procedure is used to impose a lag structure for the

coefficients of lagged R&D expenses. The procedure also reduces the parameters needed to be estimated.

The total benefit of current R&D to current and future earnings (TotalBenefit_RD) is measured as the sum

of significant 2,∑ k 

k α  .

The second proxy for RM cost is level of competition, proxied by the Herfindahl index and calculated

as the sum of the squared share of each company’s sales in total sales of the industry.14  This index ranges

from one in the case of pure monopoly to zero in the case of perfect competition. In order to reduce

Page 18: Zang (2012)

8/12/2019 Zang (2012)

http://slidepdf.com/reader/full/zang-2012 18/46

French industries, based on 3-digit SIC codes.

Within an industry, different firms likely face different levels of competition and therefore, different

 pressure when deviating from optimal business strategy. Management research (as reviewed by Woo

[1983]) has shown that market leaders enjoy more competitive advantages than market followers due to

leaders’ greater cumulative experience, economic of scale, bargaining power with suppliers and customers,

attention of investors, as well as influence on their competitors. Therefore, managers in market leader

firms may perceive RM as less costly since the erosion to their firms’ competitive advantage is relatively

small. I use market share (MarketShare), my third proxy for RM cost, to capture companies’ leadership in

the industry, measured as the percentage of the company’s sales to the total sales of its industry.

For a firm close to bankruptcy, the marginal cost of deviating from optimal business strategy is likely

to be high. In this case, managers might perceive RM as quite costly since their primary goal is the

survival of the firm.15  Following previous research, I use a modified version of Altman’s Z-score (Altman

[1968]) to proxy for a firm’s financial health:

 j,t j,t j,t j,t

 j,t

 j,t j,t j,t j,t

 j,t

 j,t

 NI Sales Retained Earnings Working Capital

Z-score = 3.3 +1.0 +1.4 +1.2Assets Assets Assets Assets

Stock Price Shares Outstanding  +0.6

Total Liabilities

×.

Altman [2000] examines three samples of subsequent distressed firms from 1969-1975, 1976-1995, and

1997-1999, and finds that a cutoff score of 2.675 yields prediction accuracy of 82% to 94%. Following

Altman [2000], my fourth proxy for RM cost is a dummy variable (Distressed) that equals one if the

firm’s Z-score is smaller than 2.675, and zero otherwise.

My last proxy concerns cost of overproduction. Managers can reduce the cost of goods sold through

Page 19: Zang (2012)

8/12/2019 Zang (2012)

http://slidepdf.com/reader/full/zang-2012 19/46

 Note that some of the cost determinants of RM are costs of deviating from optimal business strategy

(assumed to be the industry norm) in both directions. For example, in a highly competitive industry both

under- and over-investment in R&D may be costly for the firm. To distinguish between positive and

negative deviations, I create a dummy variable (D_Neg) that equals -1 when the independent variable is

negative, and +1 otherwise. I interact D_Neg with the three cost determinants (Herfindahl, MarketShare,

Distressed) that constrain firms to deviate from optimal level in both directions.

Unlike RM, AM does not have a cash flow effect. Instead, managers are constrained by the flexibility

within GAAP and the scrutiny from outsiders. I identify the following five cost determinants for AM.

The first, auditor reputation, is based on research showing that Big Eight audit firms constrain earnings

management through discretionary accruals (Becker, DeFond, Jiambalvo, and Subramanyam [1998];

Francis, Maydew, and Sparks [1999]).16  The reason is that Big Eight audit firms are likely to be more

experienced, invest more resources in auditing, and have more reputation at risk. To proxy for auditors’

reputation, I use a dummy variable (Big8) that equals one if the firm’s auditor is one of the Big Eight,

zero otherwise. My second AM cost proxy is auditor tenure. Beck, Frecka, and Solomon [1988] and Lys

and Watts [1994] find that auditor independence decreases with auditor tenure. However, Stice [1991]

finds the opposite relation and argues that the risk of not detecting errors due to unfamiliarity decreases

with tenure. Therefore, I consider auditor tenure as a proxy for auditor scrutiny without predicting its

sign. I measure auditor tenure (AuditorTenure) as the number of years the auditor has audited the client.

One obvious cost of manipulating accruals upward is that abnormal accruals will mechanically

reverse in the short-run, reducing earnings in the next period. Follow Hunt et al. [1996], I use firm-

specific estimation of current accruals’ first-order autocorrelation over the sample period as a proxy for

the reversal rate of accruals (ReversalRate) my third AM cost proxy 17 In some growing firms the

Page 20: Zang (2012)

8/12/2019 Zang (2012)

http://slidepdf.com/reader/full/zang-2012 20/46

  Due to the limited flexibility within GAAP and the reversal of previous accruals, managers’ ability to

manipulate accruals upward in the current period is constrained by AM activities in previous periods. To

capture the level of abnormal accruals in previous periods (my fourth proxy for AM cost), I use Barton

and Simko’s [2002] balance sheet measure of previous accounting choices (NOAt-1/Salest-1), which equals

net operating assets at the beginning of the year divided by lagged sales (net operating assetst-1 =

shareholders’ equity t-1 – cash and marketable securities t-1 + total debt t-1). The rationale for this proxy is

that because of the articulation between the income statement and the balance sheet, previous abnormal

accruals reflected in past earnings are also reflected in net assets; hence, the latter are overstated.19 

My last AM cost captures the level of scrutiny from investors and regulators. Prior research (Francis,

Philbrick, and Schipper [1994]) identifies four industries with particularly high litigation risk:

 biotechnology, computer, electronics, and retailing industry. I use a dummy variable to indicate firms in

these four industries (HighLitigation) and use it to proxy for litigation risk.20 

Prior earnings management work examines numerous managerial incentives to manage earnings.

Recent studies (reviewed by Dechow and Skinner [2000], Fields et al. [2001], Healy and Wahlen [1999])

have shown that capital market incentives dominate other incentives. The first capital market incentive I

consider is equity issuance. In particular, managers have incentives to boost stock price when they are

issuing equity, in the belief that inventors cannot “see through” earnings management at the time of

equity issuance. Consistent with this argument, Teoh, Welch, and Wong [1998] and Rangan [1998] find

that managers manage earnings at the time of seasoned equity offerings. Following Choudhary, Rajgopal,

and Venkatachalam [2005], I measure this incentive using a dummy variable (StockIssuance) that equals

one if the firm has issued equity in the last three fiscal years, and zero otherwise.

Recent research documents that analysts’ forecast consensus is an important earnings threshold that

Page 21: Zang (2012)

8/12/2019 Zang (2012)

http://slidepdf.com/reader/full/zang-2012 21/46

analysts’ forecast consensus at the end of the third fiscal quarter as the earnings target. Following Hunt et

al. [1996], I calculate earnings per share excluding the effects of manipulation as EPS before

extraordinary items reduced by the sum of the per share effect of unscaled abnormal real transactions and

abnormal accruals. Ex ante distance to earnings threshold at the end of third quarter (ExAnteDistance) is

the difference between analysts’ forecast consensus at the end of the third quarter and EPS excluding the

effects of manipulations. If managers have incentives to meet analyst forecast with manipulation, they

need to exert more effort if ExAnteDistance is large.

Since the earnings target (i.e., analysts’ forecast consensus) is a per share number, one penny short in

EPS translates into more dollars of actual earnings for firms with more shares outstanding than for firms

with fewer shares outstanding. Therefore, I predict that controlling for the distance to earnings threshold

at the end of third quarter, the amounts of RM and AM are positively correlated with firm’s number of

shares outstanding (Shares).

Furthermore, Bartov, Givoly, and Hayn [2002] find that firms that meet or beat analysts’ earnings

forecasts enjoy a higher return than firms that fail to meet the target. Both Bartov et al. [2002] and

Kasznik and McNichols [2002] find that this premium is greater for “habitual beaters” than for “sporadic

 beaters.” Therefore, managers of firms who repeatedly beat earnings targets have stronger incentives to

keep beating targets. I use the percentage of beating/meeting analysts’ forecast consensus in the past four

quarters (Beater) to measure a firm’s history of beating earnings thresholds.

The next capital market incentive proxy relates to stock compensation where I argue that managers

with more stock compensation have greater incentives to increase stock price. I measure the earnings

management incentives from equity compensation in two ways. The first is the top five managers’ total

stock wealth (StkWealth) calculated as the sum of their stock option value under Black Scholes method

Page 22: Zang (2012)

8/12/2019 Zang (2012)

http://slidepdf.com/reader/full/zang-2012 22/46

total stock compensation for a 1% change in the stock price. In order to aggregate the individual-level

measure to firm-level, I average the values of Sensitivity for the top five executives.

There are two views in the literature about the analysts’ role in managers’ earnings management

decision. One view is that analysts play a monitoring role (Jensen and Meckling [1976], Healy and

Palepu [2001]), the other is that analysts pressure managers to beat forecast targets. A recent paper by Yu

[2005] finds that the monitoring role of analysts dominates the pressure they exert on managers to manage

earnings. Therefore, I include analyst following (LogAnalystFoll), measured as the total number of

analysts who issue at least one forecast for the firm’s current fiscal year, as a proxy for managerial

incentive to manage earnings, and predict a negative sign.22 

It is well recognized that Jones’ model estimates of abnormal accruals are correlated with firm

 performance (Dechow et al. [1995], Kasznik [1999], McNichols [2000]). Specifically, research shows

that estimated abnormal accruals are negative for firms with low return on assets (ROA) and positive for

firms with high ROA, even in the absence of earnings management. Therefore, I include ROA in the AM

equations and expect it to have a positive sign. Since there is no prior research about the relation between

RM proxies and firm performance, I include ROA in the RM equations to control for firm performance

without predicting its sign. Other control variables include firm size, measured as the log value of net

sales (LogSales), and firms’ growth prospect, proxied by ratio of market value to book value (MtoB).

5.3. Sample selection

I start from the population of CRSP/COMPUSTAT Merged Database from 1988 to 2003, when

statement of cash flow data are available (Collins and Hribar [2001]). I exclude financial institutions

(SIC 6000-6999) and regulated industries (SIC 4400-5000), and obtain a sample of 84,178 firm-year

observations Of this sample 54 105 firm year observations have at least one RM proxy and the AM

Page 23: Zang (2012)

8/12/2019 Zang (2012)

http://slidepdf.com/reader/full/zang-2012 23/46

24,401 firm-year observations remain. The sample size declines to 10,980 firm-year observations after

obtaining independent variables calculated from Execucomp Database (StkWealth and Sensitivity). Note

that I do not require all firms to have all dependent and independent variables. Therefore, the number of

observations varies across the regressions. Table 3 reports summary statistics for the variables.

5.4. Tests of Hypotheses

To test whether managers make RM and AM decisions simultaneously or sequentially, I conduct the

Hausman test by regressing Ab_Accr on the exogenous variables (i.e., cost determinants of AM,

incentives, and control variables), the instrument for Ab_RM (the predicted value from the first-stage

regression), and the actual Ab_RM. If RM is determined after AM, then the coefficient on the

instrumental variable of Ab_RM should equal zero. Table 4 reports the results of Hausman tests for

equation (7). Consistent with sequentiality, all of the Hausman tests fail to reject the exogeneity of

Ab_RD, Ab_SGA, and Ab_Prod, in the AM regressions (with p-values ranging from 0.1244 to 0.1758).

In contrast, all of the Hausman tests reject the exogeneity of Ab_Accr in the RM equations, which means

Ab_Accr is correlated with RM’s error term. These results indicate that RM and AM are determined

sequentially, with RM preceding AM.

Given the finding of the sequentiality of RM and AM, I use the recursive equation system (8) to test

H1-H4. Since the equation system is triangular, it can be estimated consistently with OLS; results are

reported in Table 5. The adjusted R 2’s for the RM equations of Ab_RD and Ab_Prod are 13.57% and

57.10%, respectively; for the AM equations, the adjusted R 2

s range from 23.90% to 54.32%. These

results indicate reasonable explanatory power for Ab_RD, Ab_Prod, and Ab_Accr. However, the adjusted

R 2 for the Ab_SGA equation is 2.97%, indicating that this regression has limited explanatory power.

H1 predicts that in both equations RM and AM are negatively related with their own cost

Page 24: Zang (2012)

8/12/2019 Zang (2012)

http://slidepdf.com/reader/full/zang-2012 24/46

likely to deviate from optimal levels of R&D expenditures (i.e., manipulate R&D expenditures).

In the AM equation of Panel A, I find significant negative coefficient on HighLitigation and significant

 positive coefficient on AuditorTenure, consistent with H1 which argues that AM is negatively correlated

with litigation risk and auditor independence. However, the significant positive coefficient on NOA/Sales

in the AM equation is not consistent with H1. This result may be due to the measurement error in this

 proxy (discussed by DeFond [2002]) or to the possibility that firms with greater accounting flexibility in

the past are likely to have greater accounting flexibility in the current period. Panel B reports the results

of SG&A manipulation. The significant positive (negative) coefficient on D_Neg*MarketShare

(D_Neg*Distressed) suggests that firms are reluctant to deviate from the optimal SG&A decision if they

are market followers or closer to bankruptcy, consistent with H1. In the AM equation of Panel B,

abnormal accruals are negatively related with cost determinants, litigation risk (HighLigtigation) and

auditor reputation (Big8), consistent with H1. However, the significant negative coefficient of

 NOA/Sales is inconsistent with predictions, potentially due to the reasons discussed above. Finally, Panel

C also shows evidence consistent with H1. Here, results for the Ab_Prod equation show that distressed

firms are less willing to overproduce inventories, and that firms with higher fixed cost (PPE/Sales) are

more likely to overproduce inventories. Results for the AM equation show that firms with higher auditor

independence (i.e. smaller AuditorTenure), higher auditor reputation (Big8), higher accrual reversal rate

(ReversalRate), and higher litigation risk (HighLitigation) are less likely to use AM.

If managers trade off RM with AM, the level of RM should increase with the cost of AM (H2). Tests

of H2 are indicated by the coefficient estimates for the cost determinants of AM (i.e., AuditorTenure, Big8,

 NOA/Sales, ReversalRate, and HighLitigation) in the RM equations. In Panel A of Table 5, the cost

determinants of AM ReversalRate and HighLitigation are significantly positively correlated with Ab RD

Page 25: Zang (2012)

8/12/2019 Zang (2012)

http://slidepdf.com/reader/full/zang-2012 25/46

incentives to cut SG&A when accruals reverse faster and when auditors’ scrutiny is lower. In Panel C of

Table 5, H2 is supported by the significant positive coefficients on ReversalRate and HighLitigation in

Ab_Prod equation, and by the significant negative coefficient on AuditorTenure, suggesting that firms

with higher accrual reversal rate, higher litigation risk, and higher auditor independence are more likely to

use overproduction to increase reported income.

H3 predicts a negative relation between AM and RM in the AM equation. Consistent with H3, all

three panels of Table 5 show that RM levels are significantly negatively correlated with AM (p-values <

0.05), providing strong evidence for the substitutive relation between RM and AM.

A positive relation between earnings management incentives and manipulation level is predicted by

H4. In Panel A of Table 5, the significant coefficients on Beater, LogAnalystFoll, Shares, and

ExAnteDistance in both Ab_RD and Ab_Accr equations suggest that firms which consistently beat

earnings, have less analyst monitoring, more shares outstanding, and greater ex ante distance to analysts’

forecast consensus, are more likely to manipulate both R&D expenditures and discretionary accruals,

consistent with H4. Panel B of Table 5 shows that firms that consistently beat earnings and have more

shares outstanding are more likely to manipulate SG&A expenditures, consistent with H4. In the AM

equation, Panel B shows that the relation between AM (Ab_Accr and most of the earnings management

 proxies (except for StkWealth) are significant and consistent with the H4. Panel C of Table 5 reports a

significant positive relation between incentives captured by Beater, LogAnalystFoll, EquityIssuance,

Sensitivity, and ExAnteDistance and both RM and AM. Overall, the results reported in Table 5 support

H4, which predicts that RM and AM are positively correlated with earnings management incentives.

Taken together, the broad sample tests generally show results consistent with my models’ predictions.

That is both RM and AM are negatively related with their own cost determinants (H1); RM is negatively

Page 26: Zang (2012)

8/12/2019 Zang (2012)

http://slidepdf.com/reader/full/zang-2012 26/46

sample tests. There are two distinguishing features of the small sample test. First, I focus on the over-

time effects of the change in cost determinants (firm-specific litigation risk) on RM and AM behaviors.

Prior research (Dechow et al. [1996]; Lu [2004]) shows that firms subject to SEC enforcement or

securities class actions have distinct patterns of abnormal accruals: abnormal accruals increase over the

alleged years of earnings management, and decrease precipitously after the alleged years of earnings

management. Based on H1 and the results of the broad sample tests, I assume that the documented

decrease in abnormal accruals after the class period is due to the increase in litigation risk and the

negative relation between AM and its cost determinants. Assuming a substitutive relation between RM

and AM, I expect that sued firms engage in more RM after security lawsuits are filed, as implied by H2.

Therefore, lawsuit firms offer an opportunity for comparing over-time behaviors in RM and AM due to

the change in cost determinants before and after the lawsuit filing.23 

Second, I adopt a conservative matching procedure to control for the time-series properties of

abnormal real transactions and abnormal accruals. Such a procedure is necessary because abnormal

levels of real transactions and accruals are likely to be mean-reverting. Without controlling for mean-

reversion, the decrease in abnormal accruals after SEC enforcement or securities class actions

documented by the prior research cannot be solely attributable to the change in cost determinants of AM.

I use the following procedure to match lawsuit firms with control firms:24 (1) the control firm is in the

same industry-year as the lawsuit firm; (2) it does not experience any securities class action lawsuits in

the sample period; (3) it has the closest abnormal real transactions or accruals (depending on the type of

manipulations being examined) to that of the lawsuit firm in the year prior to the lawsuit filling; (4) a firm

cannot serve as control firm more than once. I calculate the difference between the abnormal levels of

real transactions and accruals of the lawsuit firm and its matched control firm and study the change in

Page 27: Zang (2012)

8/12/2019 Zang (2012)

http://slidepdf.com/reader/full/zang-2012 27/46

maintained by Stanford Law School in cooperation with Cornerstone Research. The Clearinghouse

maintains an index of filings of 1,880 issuers that have been named in federal class action securities

lawsuits since 1995. For each lawsuit, I code information about the name of the sued company, the

company’s ticker, the starting and ending dates of the class period, and the date of the class action filing.

I merge the lawsuit sample into CRSP/ COMPUSTAT Merged Database by ticker and company name.

After removing financial institutions, firms in regulated industries, and firms with multiple lawsuits

during the sample period, I obtain a sample size of 823 lawsuit firms over 1995-2004 which have both the

AM proxy and at least on of the RM proxies.

Table 6 reports descriptive statistics for the 823 lawsuit firms. Panel A shows that lawsuit firms are

larger than their control group in terms of market value of equity and total assets (both p-values < 0.01).

However, there are no significant differences in market to book ratios or ROA between lawsuit firms and

control firms. Panel B of Table 6 lists the number of lawsuit firms by industry. Note that the lawsuit

sample is concentrated in the computer industry (38.3%). Panel C reports the lawsuit filings by year.

Panel D lists the number of lawsuit firms by the length of their class periods.

Table 7 compares the performance-matched abnormal real transactions and abnormal accruals for the

lawsuit sample before and after the lawsuit filing year. To better illustrate the over-time changes in

 performance-matched RM and AM measures, I plot the mean values in Figure 1. As illustrated there, the

mean performance-matched RM and AM measures in year -1 (one year prior to lawsuit filing) are close to

zero, indicating the success of the matching procedure. Consistent with H1 and prior research, the

 performance-matched abnormal accruals drop abruptly in the year of lawsuit filing (year 0) and in the

following two years (years 1 and 2).25  Note that this for the lawsuit firms is not due to potential reversal

of abnormal accruals because the matching procedure ensures that the matched firms have the same

Page 28: Zang (2012)

8/12/2019 Zang (2012)

http://slidepdf.com/reader/full/zang-2012 28/46

  If managers trade off the relative costs of RM and AM when making RM decisions (as posited by H2)

then the increase in litigation risk and outside scrutiny due to the lawsuit filing will lead to higher levels

of RM following the lawsuit. Consistent with H2, Figure 1 shows that the performance-matched Ab_RD

increases substantially in year 1 and in year 3. Similarly, the performance-matched abnormal production

costs increase dramatically in the year of lawsuit filing. However, I do not find evidence of managers

increasing RM by cutting SG&A expenditure after the filing of lawsuit. Overall, I believe the small

sample results support the broad sample evidence concerning H2. Specifically, both indicate a

substitutive relation between RM and AM activities, consistent with H2.

7.  Conclusion

My paper provides evidence on how managers trade off real manipulation and accrual manipulation.

The real manipulations I examine include cutting R&D and SG&A expenditures, overproducing inventory,

and timing of asset sales. I use cross-sectional models to estimate abnormal levels of real transactions as

 proxies for real manipulation, and I provide two validity tests of these proxies. The first validity test

shows that suspect manipulator firms conduct more real manipulation in the fourth fiscal quarter than in

the other fiscal quarters, consistent with managers engaging in more manipulation in periods when they

have both more information about the total amount of earnings management needed and greater

incentives to manage earnings to achieve annual targets. The second validity test shows that suspect

manipulator firms have negative abnormal performance in years after real manipulation is detected,

consistent with my real manipulation proxies capturing suboptimal business decisions.

My main (broad sample) tests begin by establishing the simultaneity or sequentiality of managers’

real manipulation and accrual manipulation decisions Hausman tests reject simultaneity showing instead

Page 29: Zang (2012)

8/12/2019 Zang (2012)

http://slidepdf.com/reader/full/zang-2012 29/46

with earnings management incentives. The hypothesized substitutive relation between real manipulation

and accrual manipulation is further supported by the results of a small sample test examining over-time

 patterns of real manipulation and accrual manipulation for firms subject to securities class action lawsuits

during 1995-2004. I find that accrual manipulation decreases and real manipulation increases after

lawsuit filings, consistent with managers changing their earnings management strategies in response to

the increase in litigation risk and outside scrutiny.

My finding that managers treating real manipulation and accrual manipulation as substitutes has

implications for both researchers and regulators. For researchers, this substitutive relation suggests that

focusing on accrual manipulation exclusively may not fully explain earnings management activities. For

regulators, this result implies that increasing scrutiny or constraints over accounting discretion may not

eliminate earnings management activities, but only change managers’ priority of earnings management

strategies, some of which (real manipulation for example) may be more costly to investors.

Page 30: Zang (2012)

8/12/2019 Zang (2012)

http://slidepdf.com/reader/full/zang-2012 30/46

References

Altman, E. 1968. Financial ratios, discriminate analysis and the prediction of corporate bankruptcy.

 Journal of Finance 23(4): 589-609.

Altman, E. 2000. Predicting financial distress of companies: Revisiting the Z-Score and ZETA models.

Working paper, New York University.

Anderson, M., Banker, R. and S. Janakiraman. 2003. Are selling, general, and administrative costs

“sticky”?  Journal of Accounting Research. 41(1): 47-63.

Barber, W., P. Fairfield and J. Haggard. 1991. The effect of concern about reported income on

discretionary spending decisions: The case of research and development. The Accounting Review 66(4):

818-829.

Barton, J. 2001. Does the use of financial derivatives affect earnings management decisions? The

 Accounting Review 76(1): 1-26.

Barton, J. and P. Simko. 2002. The balance sheet as an earnings management constraint. The Accounting

 Review 77, Supplement: 1-27.

Bartov, E. 1993. The timing of assets sales and earnings manipulation. The Accounting Review 68(4):

840-855.

Bartov, E., D. Givoly and C. Hayn. 2002. The rewards to meeting or beating earnings expectations.

 Journal of Accounting and Economics 33: 173-204.

Beatty, A., S. Chamberlain and J. Magliolo. 1995. Managing financial reports of commercial banks: the

influence of taxes regulatory capital, and earnings. Journal of Accounting Research 33(2): 231-261.

Beck, P., T. Frecka and I. Solomon. 1988. An empirical analysis of the relationship between MAS

i l d di li i f di i d d

Page 31: Zang (2012)

8/12/2019 Zang (2012)

http://slidepdf.com/reader/full/zang-2012 31/46

Burgstahler, D., 1997. Incentives to manage earnings to avoid earnings decreases and losses: evidencefrom quarterly earnings. Working paper. University of Washington.

Burgstahler, D. and I. Dichev. 1997. Earnings management to avoid earnings decreases and losses.

 Journal of Accounting and Economics 24: 99-126.

Bushee, B. 1998. The influence of institutional investors on myopic R&D investment behavior. The

 Accounting Review 73(3): 305-333.

Cheng, S. 2004. R & D expenditures and CEO compensation. The Accounting Review 79(2): 305-328.

Choudhary, P., S. Rajgopal and M. Venkatachalam. 2005. Accelerated vesting of employee stock options

in anticipation of FAS 123-R. Working paper, Duke University and University of Washington.

Core, J. and W. Guay. 2002. Estimating the value of employee stock option portfolios and their

sensitivities to price and volatility. Journal of Accounting Research 40(3): 613-630.

Dechow, P., S. Kothari and R. Watts. 1998. The relation between earnings and cash flows. Journal of

 Accounting and Economics 25(2): 133.

Dechow, P., S. Richardson and I. Tuna. 2003. Why are earnings kinky? An examination of the earnings

management explanation. Review of Accounting Studies 8: 355-384.

Dechow, P. and D. Skinner. 2000. Earnings management: Reconciling the views of accounting academics,

 practitioners, and regulators. Accounting Horizon 14(2): 235-250.

Dechow, P. and R. Sloan. 1991. Executive incentives and the horizon problem, An empirical investigation.

 Journal of Accounting and Economic 14: 51-89.

Dechow, P., R. Sloan and A. Sweeney. 1995. Detecting earnings management. The Accounting Review 70:

3-42.

d i i f h b l h i i

Page 32: Zang (2012)

8/12/2019 Zang (2012)

http://slidepdf.com/reader/full/zang-2012 32/46

Francis, J., E. Maydew and H. Sparks. 1999. The role of Big Six auditors in the credible reporting ofaccruals. Auditing: A Journal of Practice and Theory 18: 17-35.

Francis, J., D. Philbrick and K. Schipper. 1994. Shareholder litigation and corporate disclosures. Journal

of Accounting Research 32(2): 137-164.

Gaver, J. and J. Paterson. 1999. Managing insurance company financial statements to meet regulatory and

tax reporting goals. Contemporary Accounting Research 16(2): 207-241.

Graham, J., C. Harvey and S. Rajgopal. 2005. The economic implications of corporate financial reporting.

Working paper, Duke University, National Bureau of Economic Research and University of Washington.

Gunny, K. 2005. What are the consequences of real earnings management? Working paper, University of

California, Berkeley.

Hand, J. 1989. Did firms undertake debt-equity swaps for an accounting paper profit or true financial gain.

The Accounting Review 64(4): 587-623.

Hausman, J.A. 1978. Specification tests in econometrics.  Econometrica 46 (6): 1251-1271.

Healy, P. and K. Palepu. 2001. Information asymmetry, corporate disclosure, and the capital markets: A

review of the empirical disclosure literature. Journal of Accounting and Economics 31(1-3): 405-440.

Healy, P. and J. Wahlen. 1999. A review of the earnings management literature and its implications for

standard setting. Accounting Horizons 13(4): 365-383.

Herrmann, D., T. Inoue and W. Thomas. 2003. The sale of assets to manage earnings in Japan. Journal of

 Accounting Research 41(1): 89-108.

Hribar, P. and D. Collins. 2002. Errors in estimating accruals: Implication for empirical research. Journal

of   Accounting Research 40(1): 105-134.

ib ki d h S k h i d i

Page 33: Zang (2012)

8/12/2019 Zang (2012)

http://slidepdf.com/reader/full/zang-2012 33/46

Jensen, M. and W. Meckling. 1976. Theory of the firm: Managerial behavior, agency costs and ownershipstructure. Journal of Financial Economics 3(4): 305-360.

Kasznik, R., 1999. On the association between voluntary disclosure and earnings management. Journal of

 Accounting Research 37: 57-81.

Kasznik, R. and M. McNichols. 2002. Does meeting earnings expectations matter? Evidence from analyst

forecast revisions and share prices. Journal of Accounting Research 40(3): 727-759.

Lev, B. 2003. Corporate earnings: Fact and fiction, Journal of Economic Perspectives 17: 27-50.

Lev, B. and T. Sougiannis. 1996. The capitalization, amortization and value-relevance of R&D. Journal of

 Accounting and Economics 21: 107-138.

Lu, Yvonne., 2004. Earnings management and securities litigation. Working paper, University of Southern

California.

Lys, T. and R. Watts. 1994. Lawsuits against auditors. Journal of Accounting Research 32 Supplement:

65-93.

McNichols, M. 2000. Research design issues in earnings management studies. Journal of Accounting and

Public Policy 19: 313-345.

Pincus, M. and S. Rajgopal. 2002. The interaction between accrual management and hedging: evidence

from oil and gas firms. The Accounting Review 77(1): 127-160.

Oyer, P. 1998. Fiscal year ends and nonlinear incentive contracts: the effect on business seasonality. The

Quarterly Journal of Economics 113(1): 149-185.

Rangan, S. 1998. Earnings management and the performance of seasoned equity offerings. Journal of

Financial Economics 50(1): 101-122.

h dh S f i h h h i l i f l i i i h ff

Page 34: Zang (2012)

8/12/2019 Zang (2012)

http://slidepdf.com/reader/full/zang-2012 34/46

Teoh, S., J. Welch and T. Wong. 1998. Earnings management and the underperformance of seasonedequity offerings. Journal of Financial Economics 50: 63-99.

Thomas, J. and H. Zhang. 2002. Inventory changes and future returns. Review of Accounting Studies 7:

163-187.

Woo, C. 1983. Evaluation of the strategies and performance of low ROI market share leaders. Strategic

 Management Journal 4: 123-135.

Yu, F. 2005. Analyst coverage and earnings management. Working paper, University of Chicago.

Zang, A.Y., 2005, Evidence on the tradeoff between real manipulation and accrual manipulation.

Unpublished dissertation, Duke University.

Page 35: Zang (2012)

8/12/2019 Zang (2012)

http://slidepdf.com/reader/full/zang-2012 35/46

Appendix I: Variable DefinitionsVariable Measurement

Variable used in the estimation models for normal levels of real transactions and accruals

RDt R&D expense = Data 46

At Total assets = Data 6

Fundst Internal funds = IBEI + R&D + Depreciation = Data 18 + Data 46 + Data 14

TobinsQt (MVE + Book value of preferred stock + Long-term debt + Short-term debt) / Total assets =(Data 199 * Data 25 + Data 130 + Data 9 + Data 34)/Data 6

CapitalExpt Capital expenditure = Data 128

SG&At SG&A expenses, excluding R&D expense = Data 189 – Data 46St  Net sales = Data 12

DSt Dummy for decreasing sales that equals 1 if St < St-1, 0 otherwise

Prodt COGSt + ΔInventoryt = Data 41 + ΔData 3

ΔSt St – St-1 

GLAt Gain or loss from sale of PPE and investment = – Data 123. A negative sign is needed sinceCOMPUSTAT presents losses as positive numbers for Gain or loss from sale of assets.

PPESalest Sale of PPE = Data 107

ISalest Sale of investment = Data 109

TACt Total accruals = Data 123 – Data 308ΔRECt  Change in account receivables = ΔData 2

k Estimated slope coefficient from a regression of ΔREC on ΔSales for each Fama-Frenchindustry-year grouping, i.e.,ΔREC = a + k ΔS + ε.

PPEt Property, plant and equipment = Data 8

Proxies for real and accrual manipulations

Ab_RD Residuals from the following regression multiplied by -1, estimated cross-sectionally for Fama-French industry years with at least 15 observations:

RD RD Funds CapitalExpt t 1 t tTobinsQ

0 1 2 3 t 4 tA A A At 1 t 1 t 1 t 1

−= α + α + α + α + α + ε

− − − −.

Ab_SGA Ab_SGA = -1*{Exp[Log(SGAt/SGAt-1)] – Exp[Log(SGAt/SGAt-1) – residual ofLog(SGAt/SGAt-1)]}*SGAt-1, where Log(SGAt/SGAt-1)] is the residuals from the followingmodel estimated cross-sectionally for Fama-French industry years with at least 15 observations:

SG&A S S S St t t t-1 t-1Log =α +α Log +α Log *DS +α Log +α Log *DS +ε1 2 3 t 4 5 t-1 tSG&A S S S St-1 t-1 t-1 t-2 t-2

⎛ ⎞ ⎛ ⎞ ⎛ ⎞ ⎛ ⎞ ⎛ ⎞⎜ ⎟ ⎜ ⎟ ⎜ ⎟ ⎜ ⎟ ⎜ ⎟⎜ ⎟ ⎜ ⎟ ⎜ ⎟ ⎜ ⎟ ⎜ ⎟⎝ ⎠ ⎝ ⎠ ⎝ ⎠ ⎝ ⎠ ⎝ ⎠

.

Ab_Prod Residuals from the following regression, estimated cross-sectionally for Fama-French industryyears with at least 15 observations:Prod S   ΔS   ΔSαt t t t-11

= +α +α +α2 3 4 tA A A A At-1 t-1 t-1 t-1 t-1 +ε .

Ab_GLA If the firm has positive GLA, Ab_GLA is measured as residuals from the following regressionmultiplied by -1, estimated cross-sectionally for Fama-French industry years with at least 15observations:GLA PPESales ISales St 0 t t t

1 2 3 tA A A A At 1 t 1 t 1 t 1 t 1

α Δ= +α +α +α +ε .

Page 36: Zang (2012)

8/12/2019 Zang (2012)

http://slidepdf.com/reader/full/zang-2012 36/46

TopRMk t  Dummy variable equals 1 if the firm’s abnormal transaction is in the top quintiles of RM in its

industry-year. For example, if the independent variable is QRDt/At-1 from fiscal quarter 2,TopRM is a dummy variable equals 1 if this firm’s annual Ab_RDt is in the top quintile ofAb_RDt of the same industry-year. Here, k = RD, SGA, Prod, or GLA.

 Independent variables in the main test

OI Annual operating income, before depreciation, advertising and R&D expenses.

TA The value of equipment and equity, inventory and investment in unconsolidated subsidiariesand goodwill, in current dollars.

AD Annual advertising expenses.

IRD Industry average R&D expense using four-digit SIC industry grouping.TotalBenefit_RD The total benefit of current R&D to current and future earnings (TotalBenefit_RD) is measured

as the sum of significant  2,k α  , estimated with the following model, which is run cross-

sectionally for each industry-year using Almon lag procedure:

0 1 2, 3, , 1 , , 1

OI TA RD ADe

k it S S S S  k i t i t   i t k i t  

α α α α  

⎛ ⎞ ⎛ ⎞   ⎛ ⎞ ⎛ ⎞∑= + + + +⎜ ⎟ ⎜ ⎟   ⎜ ⎟ ⎜ ⎟⎝ ⎠ ⎝ ⎠   ⎝ ⎠ ⎝ ⎠−   − −, where RD/S is replaced by its

instrumental variable estimated with the following regression: ,, ,

 RD IRDa b e

i t S S i t i t  

⎛ ⎞ ⎛ ⎞= + +⎜ ⎟ ⎜ ⎟

⎝ ⎠ ⎝ ⎠.

Herfindahl The sum of the squared share of each company in total sales of the industry.MarketShare The percentage of the company’s sales to the total sales of its industry

Distressed Dummy variable that equals 1 if the firm’s Z-score is less than 2.675, and 0 otherwise, where NI Sales Retained Earnings Working Capital Stock Price SharesOutstanding

 jt jt jt jt jtZ-score = 3.3 +1.0 +1.4 +1.2 +0.6 jt Assets Assets Assets Assets Total Liabilities

 jt jt jt jt jt

×.

PPE/Sales Data 8/Data 12

D_Neg Dummy variable equals -1 if the independent variable is negative, 0 otherwise.

Big8 Dummy variable equals 1 if the firm’s auditor is a Big Eight, 0 otherwise.

AuditorTenure Number of years the auditor has worked for the client.

ReversalRate Firm-specific estimation of current accruals’ first-order autocorrelation over the sample period.I require the time series of firm-specific current accruals to be at least 7 years. I set theaccruals’ autocorrelation coefficient to 0 if the estimated value is positive.

 NOAt-1/Salest-1  (Shareholders’ equity – cash and marketable securities + total debt)t-1/Salest-1

HighLitigation Dummy variable equals 1 if the firm belongs to one of the following industries: biotechnology(SIC 2833-2836), computer (SIC 3570-3577, 7370-7374), electronics (SIC 3600-3674), andretailing industry (SIC 5200-5961).

StockIssuance Dummy variable equals 1 if the firm has issued equity in the last three fiscal years and 0otherwise.

Beater Percentage of beating/meeting analysts’ forecast consensus in the past 4 quarters.StkWealth The sum of the top five executives’ stock option, valued under Black-Scholes method, and thevalue of their other stock holdings. The Black-Scholes value of top executives’ stock option iscalculated with “one-year approximation” procedure developed by Core and Guay [2002].

Sensitivity The average sensitivity of stock compensation to stock price for top five executives. Thesensitivity of stock compensation to stock price is calculated as the sum of the changes inBlack Scholes value of the stock options and in other stock holding value for a 1% change in

Page 37: Zang (2012)

8/12/2019 Zang (2012)

http://slidepdf.com/reader/full/zang-2012 37/46

Table 1: Estimation of the Normal Levels of R&D Transactions and Accruals Panel A: Estimation of normal level of real transactions

The following regressions are estimated cross-sectionally within each industry-year over 1988-2004. Fama-French

48 industry grouping is used. The model is estimated for an industry-year with at least 15 observations. The

reported coefficients are the mean value of the coefficients across the industry-years. T-statistics are calculated

using the standard error of the mean coefficients across the industry-years. The adjusted R 2 (number of observations)

is the mean adjusted R 2 (number of observations) across the industry-years.

t t 1 t t0 1 2 3 t 4 t

t 1 t 1 t 1 t 1

RD RD Funds CapitalExpTobinsQA A A A

− − − −

= α + α + α + α + α + ε   (1) t t t t-1 t-1

1 2 3 t 4 5 t-1 t

t-1 t-1 t-1 t-2 t-2

SG&A S S S SLog =α +α Log +α Log *DS +α Log +α Log *DS +ε

SG&A S S S S

⎛ ⎞ ⎛ ⎞ ⎛ ⎞ ⎛ ⎞ ⎛ ⎞⎜ ⎟ ⎜ ⎟ ⎜ ⎟ ⎜ ⎟ ⎜ ⎟⎝ ⎠ ⎝ ⎠ ⎝ ⎠ ⎝ ⎠ ⎝ ⎠

  (2) t t t t-11

2 3 4 t

t-1 t-1 t-1 t-1 t-1

Prod S   ΔS   ΔSα= +α +α +α

A A A A A+ ε   (3) 

t 0 t t t

1 2 3 t

t 1 t 1 t 1 t 1 t 1

GLA PPESales ISales S

A A A A A− − − − −

α Δ

= + α + α + α + ε   (4) 

t

t 1

RD

A −

  t

t-1

SG&ALog

SG&A

⎛ ⎞⎜ ⎟⎝ ⎠

  t

t-1

Prod

A  t

t 1

GLA

A −

 

Intercept0.0011(0.91)

Intercept0.0179(7.49)

t 1

1

A −

  0.0446(0.51)

t 1

1

A −

  0.0059(0.88)

t 1

t 1

RD

A

  0.9263(47.47)

t

t-1

SlogS

⎛ ⎞⎜ ⎟⎝ ⎠

  0.6220(39.90)

t

t-1

S

A  0.7067

(167.45)t

t 1

PPESales

A −

  0.2679(7.94)

t

t 1

Funds

A −

  0.0008(0.26)

tt

t-1

SLog *DS

S

⎛ ⎞⎜ ⎟⎝ ⎠

 -0.2396(-1.97)

t

t-1

ΔS

A  0.0638

(6.79)t

t 1

ISales

A −

  0.0591(2.52)

TobinsQt 0.0027(6.54)

t-1

t-2

SLog

S

⎛ ⎞⎜ ⎟⎝ ⎠

 0.0716(6.24)

t-1

t-1

ΔS

A  -0.0174

(-2.26)t

t 1

S

A −

Δ  -0.0174

(-1.11)

t

t 1

CapitalExp

A −  0.1089(8.06)

t-1 t-1

t-2

SLog *DSS

⎛ ⎞⎜ ⎟⎝ ⎠

  0.1993(1.77)

Adj. R 2(%) 82.08 49.58 94.08 26.00# of obs. 102.37 100.13 111.00 98.97# of ind-yrs 448 603 620 577

Page 38: Zang (2012)

8/12/2019 Zang (2012)

http://slidepdf.com/reader/full/zang-2012 38/46

Panel B: Estimation of normal level of total accrualsThe following regressions are estimated cross-sectionally within each industry-year over 1988-2004. Fama-French

48 industry grouping is used. The model is estimated for an industry-year with at least 15 observations. The

reported coefficients are the mean value of the coefficients across the industry-years. T-statistics are calculated

using the standard error of the mean coefficients across the industry-years. The adjusted R 2 (number of observations)

is the mean adjusted R 2 (number of observations) across the industry-years.

DRT [2003] model: ( ) j t j t j t j t j t 1 j t 101 2 3 4 j t

 j t 1 j t 1 j t 1 j t 1 j t 2 j t

TAC 1 k S REC PPE TAC S

A A A A A S

 , , , , , ,

 ,

 , , , , , ,

− +

− − − − −

+ Δ − Δ Δα= + β + β + β + β + ε   (5)

DSS [1995] model: j t j t j t j t0

1 2 j t

 j t 1 j t 1 j t 1 j t 1

TAC S REC PPE

A A A A

 , , , ,

 ,

 , , , ,− − − −

Δ − Δα= + β + β + ε  

DRT model DSS model

a0/A j,t-1-0.2228(-4.35)

a0/A j,t-1 -0.4354(-6.74)

( )  j t j t

 j t 1

1 k S REC

A

 , ,

 ,  −

+ Δ − Δ  0.0196

(2.99)

 j t j t

 j t 1

S REC

A

 , ,

 ,  −

Δ − Δ  0.0401

(5.90)

PPE j,t/A j,t-1-0.1172(-22.74)

PPE j,t/A j,t-1 -0.1505(-23.29)

TAC j,t-1/A j,t-20.2160(15.30)

ΔS j,t+1/S j,t 0.0243(3.86)

Adj. R 2(%) 44.77 35.50# of obs. 99.71 117.66# of ind-yrs 543 636

Panel C: Summary statistics for abnormal real and accrual levels

Ab_RD Ab_SGA Ab_Prod Ab_GLA Ab_Accr

Mean 0.0015 -0.0001 -0.0285 -0.0130 -0.0079Median -0.0008 0.0000 -0.0319 0.0020 -0.0018

Std. Dev. 0.0645 0.0216 0.2680 0.0409 0.139325% -0.0075 -0.0007 -0.1582 -0.0098 -0.059875% 0.0170 0.0007 0.0908 0.0001 0.0519Skewness -0.9698 -0.2829 0.2895 -4.5216 -0.4484Kurtosis 8.0336 16.4449 2.1356 23.8872 3.5897# of obs. 45,860 60,378 68,820 12,937 54,142

Page 39: Zang (2012)

8/12/2019 Zang (2012)

http://slidepdf.com/reader/full/zang-2012 39/46

Table 2: Validity Tests of Real Manipulation Proxies

Panel A: Timing test

The following regressions are estimated cross-sectionally for each Fama-French industry-year over 1988-2003. The

model is estimated for an industry-year that satisfies the following requirements: (1) ≥ 15 observations; (2) ≤ 80%

of the firms with a December fiscal year end; 3) ≥ three different fiscal year ends among the firms. The reported

coefficients are the mean of the coefficients across the industry-years. The significance of the mean coefficients is

 based on the standard errors across the industry-years.4 4 12

f k f c

qt q q t q q m m t

q=1 q=1 m=1

X =   γ D TopRM + f D + c D +ε× × × ×∑ ∑ ∑   (6) Xqt  QRDt/At-1  QSGAt/At-1  QProdt/At-1  QGLAt/At-1 

Pred.sign

ResultsPred.sign

ResultsPred.sign

ResultsPred.sign

Results

Df 1μTopRM (γ1) – -0.0104 – -0.0788 + 0.1681*** – 0.0000

Df 2μTopRM (γ2) – -0.0058 – -0.0280 + 0.1897*** – 0.0001

Df 3μTopRM (γ3) – -0.0023 – 0.0231 + 0.1963*** – -0.0005

Df 4μTopRM (γ4) – -0.0095** – 0.0232* + 0.2264*** – -0.0011

γ4 – mean (γ1, γ2, γ3) – -0.0034*** – 0.0511 + 0.0417*** – -0.0010

# of industry-years 383 567 1,072 348

Panel B: Subsequent firm performance

A performance-matched control firm (control firm) is identified for each firm in the top abnormal real transaction

quintiles (target firm). The abnormal performance in year t+1 to t+3 of is measured as the difference between the

 performance measures of the target firms and those of the control firm.

Performance-matched ROA Performance-matched CFO/At-1 

 N Mean Median N Mean Median

Top Ab_RD quintile:

t 4,894 0.0000 0.0006*** 2,793 0.0003 0.0009***t+1 4,894 -0.0100* 0.0012 2,793 0.0015 0.0059*t+2 4,894 -0.0187* 0.0009 2,793 -0.0051 0.0086***t+3 4,894 -0.0215*** 0.0022 2,793 -0.0145 0.0072

Top Ab_SGA quintile: t 10,128 0.0001 0.0000*** 3,416 0.0005*** 0.0007***

t+1 10,128 -0.0104*** 0.0002 3,416 -0.0221*** -0.0162***t+2 10,128 -0.0140*** -0.0035*** 3,416 -0.0072*** -0.0180***t+3 10,128 -0.0289** -0.0086*** 3,416 -0.0366 -0.0198***

Top Ab_Prod quintile: t 9 079 -0 0002 0 0002 3 957 0 0000 0 0000

Page 40: Zang (2012)

8/12/2019 Zang (2012)

http://slidepdf.com/reader/full/zang-2012 40/46

Table 3: Descriptive Information About Independent Variables

Variable N Mean Median Std. Dev. 25% 75%

Cost determinants of real manipulation:

TotalBenefit_RD 6827 10.2154 3.5329 49.8448 0.0000 17.3446Herfindahl 11129 0.1440 0.1124 0.0965 0.0777 0.1830MarketShare 11129 0.0669 0.0225 0.1083 0.0051 0.0762

Distressed 11129 0.2805 0.0000 0.4493 0.0000 1.0000PPE/Sales 11129 0.4090 0.2277 0.6930 0.1295 0.4116

Cost determinants of accrual manipulation:

Big8 11129 0.9825 1.0000 0.1312 1.0000 1.0000AuditorTenure 11129 11.3527 9.0000 8.1872 5.0000 19.0000ReversalRate 11129 -0.1747 -0.0573 0.2533 -0.2829 0.0000 NOA/Sales 11129 0.7777 0.5868 0.7397 0.3711 0.8841HighLitigation 11129 0.3600 0.0000 0.4800 0.0000 1.0000

 Incentives of earnings management:Beater 11129 0.3860 0.2500 0.3175 0.0000 0.6667LogAnalystFoll 11129 2.4073 2.4849 0.7414 1.9459 2.9444StockIssuance 11129 0.8934 1.0000 0.3086 1.0000 1.0000Sensitivity 11129 0.3398 0.0768 3.2071 0.0314 0.1960StkWealth 11129 35.0971 6.7581 346.5280 2.7121 17.7762Shares 11129 114.0876 40.4190 231.1821 21.2280 98.8000ExAnteDistance_RD 7445 0.3858 0.2453 1.7429 -0.3997 1.0263ExAnteDistance_SGA 10328 0.2659 0.1849 1.7724 -0.4574 0.9207

ExAnteDistance_Prod 10932 -0.3380 -0.6623 5.3202 -2.7909 1.4070ExAnteDistance_GLA 7372 0.2318 0.1673 1.6831 -0.4417 0.8433

Controls:

ROA 11129 0.0556 0.0617 0.1143 0.0207 0.1074LogSales 11129 6.8006 6.7266 1.5986 5.7665 7.8286MtoB 11129 3.3064 2.4350 3.2413 1.5862 3.8889

For variable definitions, please refer to Appendix I.

Page 41: Zang (2012)

8/12/2019 Zang (2012)

http://slidepdf.com/reader/full/zang-2012 41/46

Table 4: Hausman Test for Simultaneity versus Sequentiality of Real and Accrual Manipulations

The following regressions are estimated for the pooled sample over 1992-2003. Costs of RM are TotalBenefit_RD,

D_Neg*Herfindahl, D_Neg*MarketShare, D_Neg*Distressed PPE/Sales; costs of AM are AuditorTenure, Big8,

 NOA/Sales, ReversalRate, HighLitigation; incentives are Beater, LogAnalystFoll, EquityIssuance, Sensitivity,

StkWealth, Shares, ExAnteDistance; other controls are ROA, LogSales, MtoB. The Hausman test is conducted by

regressing Ab_Accr on the exogenous variables (i.e., cost determinants of AM, incentives, and control variables), the

instrument for Ab_RM (the predicted value from the first-stage regression), and the actual Ab_RM. The coefficient

on the predicted value of Ab_RM (P_Ab_RM) is tested against zero.

i,t 0 1 i,t 2,j j,i,t 3,k k,i,t 4,l l,i,t i t

 j k l

RM = γ +γ AM +   γ Cost of RM +   γ Incentives +   γ Other Controls u ,

+∑ ∑ ∑ ,

i,t 0 1 i,t 2,j' j',i,t 3,k k,i,t 4,l l,i,t i,t

 j' k l

AM = φ +φ RM +   φ Cost of AM +   φ Incentives +   φ Other Controls +v∑ ∑ ∑ ,(7) 

Panel A: Panel B:

Ab_RD equation(N = 5,104)

Ab_Accr equation(N = 5,104)

Ab_SGA equation(N = 10,276)

Ab_Accr equation(N = 10,276)

Coef. p-value Coef. p-value Coef. p-value Coef. p-value

 Endogenous

variables:

Ab_Accr   -0.0370 <.0001 0.0000 0.4333Ab_RM -0.1044 <.0001 0.1051 0.4332

P_Ab_Accr -0.2083 0.0057 -0.0138 <.0001P_Ab_RM -1.0877 0.1758 -19.3241 0.1244 Hausman test:

1st-stage adj. R 2 (%) 13.57 45.59 2.97 54.202nd-stage adj. R 2 (%) 13.69 45.73 2.79 54.18

p-value for

Hausman stat.0.0057 0.1758 <.0001 0.1244

Panel C:

Ab_Prod equation(N = 10,453)

Ab_Accr equation(N = 10,453)

Coef. p-value Coef. p-value

 Endogenous

variables:

Ab Accr -0.1749 <.0001

Page 42: Zang (2012)

8/12/2019 Zang (2012)

http://slidepdf.com/reader/full/zang-2012 42/46

Table 5: Tests of the Tradeoff between Real and Accrual Manipulations 

The following regressions are estimated for the pooled sample over 1992-2003. Costs of RM are TotalBenefit_RD,

D_Neg*Herfindahl, D_Neg*MarketShare, D_Neg*Distressed PPE/Sales; costs of AM are AuditorTenure, Big8,

 NOA/Sales, ReversalRate, HighLitigation; incentives are Beater, LogAnalystFoll, EquityIssuance, Sensitivity,

StkWealth, Shares, ExAnteDistance; other controls are ROA, LogSales, MtoB.

i,t 0 1,j j,i,t 2,p p,i,t 3,k k,i,t 4,l l,i,t i,t

 j p k l

RM = + Cost of RM + Cost of AM + Incentives + Other Controls +uλ λ λ λ λ∑ ∑ ∑ ∑ ,

i,t 0 1 i,t 2,p p,i,t 3,k k,i,t 4,l l,i,t i,t

 p k l

AM = + RM + Cost of AM + Incentives + Other Controls +vδ δ δ δ δ∑ ∑ ∑ ,(8) 

Panel A: R&D manipulation

Ab_RD equation(N = 5,104)

Ab_Accr equation(N = 5,104)

Pred. sign Coef. p-valuePred.sign

Coef. p-value

Intercept ? -0.0256 0.0002 ? 0.0155 0.0886Ab_RD – -0.1166 <.0001Cost determinants of RM:TotalBenefit_RD – -0.0001 0.0211D_Neg*Herfindahl + 0.0256 0.0001

D_Neg*MarketShare + 0.0356 <.0001D_Neg*Distressed – -0.0033 0.0022Cost determinants of AM:

AuditorTenure +/– 0.0001 0.1723 +/– 0.0002 0.0956Big8 + -0.0147 0.2600 – -0.0027 0.2785 NOA/Sales + 0.0008 0.2282 – 0.0068 <.0001ReversalRate + 0.0027 0.0472 – -0.0029 0.2475HighLitigation + 0.0055 <.0001 – -0.0116 <.0001

 Incentives:

Beater + 0.0034 0.0437 + 0.0203 <.0001LogAnalystFoll – -0.0061 <.0001 – -0.0044 0.0083EquityIssuance + 0.0005 0.4140 + 0.0037 0.1353Sensitivity + -0.0001 0.4547 + 0.0016 0.2124StkWealth + 0.0000 0.4389 + -0.0000 0.0580Shares + 0.0000 0.0178 + 0.0000 <.0001

Page 43: Zang (2012)

8/12/2019 Zang (2012)

http://slidepdf.com/reader/full/zang-2012 43/46

Panel B: SGA manipulation

Ab_SGA equation(N = 10,276)

Ab_Accr equation(N = 10,276)

Pred. sign Coef. p-value Pred. sign Coef. p-value

Intercept ? 0.0005 <.0001 ? -0.0335 <.0001Ab_SGA – -0.0266 0.0483

Cost determinants of RM:D_Neg*Herfindahl + 0.0001 0.2578

D_Neg*MarketShare + 0.0004 <.0001D_Neg*Distressed – -0.0001 <.0001

Cost determinants of AM:AuditorTenure +/– -0.0000 0.0055 +/– 0.0000 0.3297Big8 + -0.0000 0.3904 – -0.0081 0.0348 NOA/Sales + 0.0001 <.0001 – 0.0052 <.0001ReversalRate + 0.0001 0.0065 – -0.0021 0.2240HighLitigation + -0.0000 0.4857 – -0.0023 0.0519

 Incentives:

Beater + 0.0001 <.0001 + 0.0068 0.0005LogAnalystFoll – -0.0000 0.1582 – -0.0058 <.0001EquityIssuance + -0.0000 0.2267 + 0.0007 0.3729Sensitivity + 0.0000 0.4518 + 0.0020 0.0232StkWealth + -0.0000 0.4651 + -0.0000 0.0009Shares + -0.0000 0.0212 + 0.0000 0.0738ExAnteDistance + 0.0000 0.4790 + 0.0339 <.0001Control variables:ROA ? -0.0001 0.3206 + 0.3700 <.0001

LogSales ? 0.0001 <.0001 ? 0.0007 0.1315MtoB ? -0.0000 0.4282 ? -0.0030 <.0001

Adj. R 2 (%) 2.97 54.32

For variable definitions, please refer to Appendix I.

Page 44: Zang (2012)

8/12/2019 Zang (2012)

http://slidepdf.com/reader/full/zang-2012 44/46

Panel C: Production cost manipulation

Ab_Prod equation(N = 10,453)

Ab_Accr equation(N = 10,453)

Pred. sign Coef. p-value Pred. sign Coef. p-value

Intercept ? -0.1940 <.0001 ? -0.0184 0.0120Ab_Prod – -0.0517 <.0001

Cost determinants of RM:D_Neg*Herfindahl + -0.0748 <.0001

D_Neg*MarketShare + -0.0153 0.1123D_Neg*Distressed – -0.0305 <.0001PPE/Sales  + 0.0070 0.0474Cost determinants of AM:

AuditorTenure +/– -0.0003 0.0576 +/– 0.0004 <.0001Big8 + -0.0328 0.0017 – -0.0105 0.0079 NOA/Sales + 0.0208 <.0001 – 0.0130 <.0001ReversalRate + 0.0049 0.0521 – -0.0047 0.0613HighLitigation + 0.0195 0.0754 – -0.0014 0.0859

 Incentives:Beater + 0.0483 <.0001 + 0.0041 0.0589LogAnalystFoll – -0.0298 <.0001 – -0.0035 0.0055EquityIssuance + 0.0178 0.0001 + 0.0038 0.0781Sensitivity + 0.0042 0.0482 + 0.0018 0.0856StkWealth + -0.0000 0.0268 + -0.0000 0.0087Shares + -0.0001 <.0001 + 0.0000 <.0001ExAnteDistance + 0.0282 <.0001 + 0.0074 <.0001Control variables:

ROA ? -0.5044 <.0001 + 0.3180 <.0001LogSales ? 0.0299 <.0001 ? -0.0035 <.0001MtoB ? -0.0060 <.0001 ? -0.0025 <.0001

Adj. R 2 (%) 57.10 23.90

For variable definitions, please refer to Appendix I.

Page 45: Zang (2012)

8/12/2019 Zang (2012)

http://slidepdf.com/reader/full/zang-2012 45/46

Table 6:  Summary Statistics for 823 Lawsuit Firms (1995-2004) 

Panel A: Selective descriptive statistics:

Lawsuit firms(N=823)

Control firms(N=2,577)

Difference

Variable Mean Median Mean Median Mean Median

MVE ($million) 5,036.23 485.41 2,343.99 108.06 2,692.24*** 377.35***

MtoB 3.49 1.90 3.05 1.89 0.44 0.01

Assets ($million) 2,432.80 273.29 1,431.35 95.22 1,001.45*** 178.07***

ROA (%) -0.18 -0.06 -0.14 0.01 -0.04 -0.07

***, **, and * represents significance at 1%, 5%, and 10% respectively.

Panel B: Number of lawsuit firms by industry:

Industry SIC code Number of suits

1. Mining and construction 1000-1299, 1400-1999 132. Food 2000-2111 123. Textile, printing and publishing 2200-2799 264. Chemicals 2800-2824, 2840-2899 155. Pharmaceuticals 2830-2836 666. Extractive industries 2900-2999, 1300-1399 13

7. Durable manufacturers 3000-3569, 3580-3669, 3680-3999 1578. Computers 7370-7379, 3570-3579, 3670-3679 3159. Retail 5000-5999 9210. Services 7000-7369, 7380-8999 10111. Other 13

Total 823

Panel C: Number of filings by year:

1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 Total

3 45 101 108 107 88 198 105 67 1 823

Table 7: Performance Matched Analysis For Small Sample

Page 46: Zang (2012)

8/12/2019 Zang (2012)

http://slidepdf.com/reader/full/zang-2012 46/46

  45

Table 7:  Performance-Matched Analysis For Small SampleThe following procedure is used to match lawsuit sample with control firms: (1) the control firm is in the same industry-year as lawsuit firm; (2) it does notexperience any securities class action lawsuits over 1995-2004; (3) it has the closest abnormal real transactions or accruals (depending on the type ofmanipulation being examined) to that of the lawsuit firm in the year prior to the lawsuit filling year; (4) one firm cannot serve as control firm more than once.

The difference between the abnormal level of real transactions or accruals of a lawsuit firm and that of its control firm is reported.

Ab_Accr Ab_RD Ab_SGA Ab_Prod

Year Obs Mean p-value Obs Mean p-value Obs Mean p-value Obs Mean p-value

-3 322 -0.0097 0.2072 228 -0.0121 0.0567 317 -0.0014 0.0924 362 -0.0033 0.4193-2 426 0.0025 0.4031 319 -0.0121 0.0330 395 -0.0017 0.0657 466 -0.0077 0.2765-1 576 -0.0004 0.3188 564 -0.0002 0.2275 534 -0.0001 0.2086 600 -0.0045 0.00820 435 -0.0590 0.0000 459 -0.0042 0.1887 452 -0.0021 0.0088 488 0.0296 0.0092+1 310 -0.0251 0.0159 341 0.0119 0.0039 342 0.0013 0.0688 387 0.0042 0.3840

+2 202 -0.0124 0.1909 241 0.0073 0.1456 247 0.0003 0.4005 264 -0.0101 0.2831+3 131 0.0056 0.3502 152 0.0146 0.0334 165 -0.0031 0.0120 180 -0.0429 0.0301

Figure 1. Mean difference of RM/AM between litigation firms and control firms*

-0.07

-0.06

-0.05

-0.04

-0.03

-0.02

-0.01

0

0.01

0.02

0.03

0.04

-3 -2 -1 0 1 2 3

Year relative to lawsuit filing year

   A  s  s  e   t  -  s  c  a

   l  e   d   V  a   l  u  e   f  o  r   A   b_

   A  c  c  r ,   A   b_

   P  r  o   d

 ,

   A   b_

   R   D   (   %   )

-0.007

-0.006

-0.005

-0.004

-0.003

-0.002

-0.001

0

0.001

0.002

0.003

0.004

   A  s  s  e   t  -  s  c  a   l  e   d   V  a   l  u  e   f  o  r   A   b_   S

   G   A   (   %   )

Ab_Accr

Ab_RD

Ab_Prod

Ab_SGA

 For variable definitions, please refer to Appendix I.