How Common Are Intentional GAAP Violations? Estimates from a Dynamic Model

• Date: 01 March 2018
• Author: ANASTASIA A. ZAKOLYUKINA
• Published date: 01 March 2018
• DOI: http://doi.org/10.1111/1475-679X.12190

DOI: 10.1111/1475-679X.12190

Journal of Accounting Research

Vol. 56 No. 1 March 2018

Printed in U.S.A.

How Common Are Intentional

GAAP Violations? Estimates from

a Dynamic Model

ANASTASIA A. ZAKOLYUKINA

∗

Received 30 March 2013; accepted 10 July 2017

ABSTRACT

This paper uses data on detected misstatements—earnings restatements—

and a dynamic model to estimate the extent of undetected misstatements

that violate GAAP. The model features a CEO who can manipulate his firm’s

stock price by misstating earnings. I find the CEO’s expected cost of mislead-

ing investors is low. The probability of detection over a five-year horizon is

13.91%, and the average misstatement, if detected, results in an 8.53% loss in

the CEO’s retirement wealth. The low expected cost implies a high fraction

of CEOs who misstate earnings at least once at 60%, with 2%–22% of CEOs

∗The University of Chicago Booth School of Business.

Accepted by Thomas Hemmer. I am grateful for helpful comments from Doug Skin-

ner (the initial editor), two anonymous referees, my dissertation committee at Stanford

Graduate School of Business—Anne Beyer, David Larcker, Maureen McNichols, Joseph Pi-

otroski, and Peter Reiss—my colleagues at the University of Chicago—especially Ray Ball,

Phil Berger, John Cochrane, Jean-Pierre Dub´

e, Alex Frankel, Christian Leuz, Canice Pren-

dergast, and Chad Syverson—and seminar participants at Stanford, Wharton, Columbia,

Booth School of Business, Yale School of Management, NYU Stern, London Business School,

2013 FARS Midyear Meeting, Carnegie Mellon University Accounting Mini Conference,

and Kellogg School of Management. I also thank Robin Weiss, Hongcen Wei, Clark Hyde,

Ravi Pillai, Jisoo Lee, Rudyard Richter, Yuriy Olshanskiy, and Hossein Pourreza for out-

standing research support, and Dennis Tanona and Olga Usvyatsky from Audit Analytics

Inc. for their help with data. This research was funded in part by the Accounting Re-

search Center at the University of Chicago Booth School of Business. I acknowledge the

Neubauer Family Foundation, Harry W. Kirchheimer Faculty Research Fund, IBM Cor-

poration Faculty Research Fund, and the University of Chicago Booth School of Busi-

ness for financial support, as well as the University of Chicago Research Computing Cen-

ter for support of this study. An online appendix to this paper can be downloaded at

http://research.chicagobooth.edu/arc/journal-of-accounting-research/online-supplements.

5

Copyright C, University of Chicago on behalf of the Accounting Research Center,2017

6A.A.ZAKOLYUKINA

starting to misstate earnings in each year 2003–2010, inflation in stock prices

across CEOs who misstate earnings at 2.02%, and inflation in stock prices

across all CEOs at 0.77%. Wealthier CEOs manipulate less, and the average

misstatement is larger in smaller firms.

JEL codes: G34; G38; K22; K42; M41

Keywords: earnings manipulation; executive compensation; earnings re-

statements

1. Introduction

On average, 4% of the U.S. companies listed on the NYSE and NAS-

DAQ restate their earnings each year (e.g., Whalen, Usvyatsky, and Tanona

[2015]). Thus, concluding GAAP violations are rare and should not con-

cern investors, boards of directors, or policy makers is tempting. However,

researchers have long recognized the ability of outside parties to detect

misstatements is imperfect.1But, except for Wang [2013] and Dyck, Morse,

and Zingales [2014], little research has tried to assess the extent of unde-

tected fraud. Making such an assessment is difficult because the incidence

of violations is not observable and the probability of detection is unknown.

For example, the 4% restatement rate can correspond to all companies mis-

stating earnings and only 4% of them being detected, or only 4% of them

misstating and all of them being detected. Any intermediate case that pro-

duces the 4% rate is also plausible, which raises the question of what the

actual rates of manipulation are.

This paper quantifies the incidence and magnitude of undetected mis-

statements by estimating a dynamic model of manipulation. This model

can accommodate both the endogeneity of manipulation decisions, which

generates an endogenous pattern of restatements, and the unobservable

features, such as the probability of detection and the penalty for manipula-

tion. When fitting this structural model, I solve for CEOs’ optimal manip-

ulation decisions, which match the pattern of restatements, and estimate

the parameters that determine these decisions. Armed with the estimated

parameters and the data on stock prices, the inference about undetected

misstatements becomes feasible.

This model features a CEO whose compensation and career path depend

on the stock price, thus inducing him not only to work hard but also to mis-

state earnings to manipulate the stock price.2,3The CEO’s manipulation

1See, for example, Feroz, Park, and Pastena [1991], Dechow, Ge, and Schrand [2010],

Dyck, Morse, and Zingales [2010], and Correia [2014].

2Margiotta and Miller [2000], Gayle and Miller [2009], and Gayle and Miller [2015] quan-

tify the extent of losses to shareholders from ignoring the moral hazard problem.

3See, for example, Goldman and Slezak [2006] and Beyer, Guttman, and Marinovic

[2014b] for equilibrium models of compensation in the presence of manipulation, and Arm-

strong, Jagolinzer, and Larcker [2010] for a review of empirical literature on manipulation

and stock-based compensation.

HOW COMMON ARE INTENTIONAL GAAP VIOLATIONS?7

decision is dynamic: old misstatements unwind, and new ones are chosen

each period. The resulting price distortion lasts until the next earnings re-

port, and benefits the CEO if he leaves at the inflated stock price. The CEO

incurs no direct cost of misstating earnings; instead, he can be penalized

when his manipulation is detected and the firm restates its financial state-

ments. The penalty allows for the possibility that few CEOs manipulate.

This study extends prior work on agency theory and performance manipu-

lation, which considers static one-period models with a manager incurring

a direct cost of misstating earnings.4

The multiperiod nature of manipulation also suggests a shift in focus

from the bias in earnings considered in prior studies to the bias in book

value. Although conditional on the prior-period bias in book value, under

clean surplus accounting, choosing the bias in book value is equivalent to

choosing the bias in earnings; this shift in focus reinforces the idea that

the prior-period bias in book value serves as an earnings management con-

straint (Barton and Simko [2002]). This constraint, however, is enforced

by the penalty for manipulation depending on the bias in book value at

detection rather than by a specific rate of accrual reversal.

I estimate the model using the simulated method of moments (SMM)

and data on executive compensation, earnings restatements, and stock

prices for a large sample of U.S. firms from 2003 to 2010. A focus on

the post–Sarbanes-Oxley Act (SOX) period takes advantage of SOX’s goal

of improving the auditing of U.S. public companies and the detection of

accounting fraud that likely changed the expected cost of manipulation

(Coates and Srinivasan [2014]). Although restatements capture detected

misstatements, not all of them are intentional GAAP violations (Dechow,

Ge, and Schrand [2010]). To address this issue, I assume misstatements cor-

rected by restatements disclosing more serious problems, such as revenue

recognition errors, are more likely to be intentional.

Estimates indicate CEOs’ expected costs of misstating earnings are low.

For the revenue recognition misstatements, once the CEO manipulates,

he faces a 2.95% probability of detection each year. This number corre-

sponds to the probability of detection of 13.91% over a five-year horizon.5

The longer the horizon, the higher the probability of detection. Once ter-

minated after being caught, which happens with a 15% chance, the CEO

incurs a penalty that reduces his retirement wealth on average by 8.53%.

The relatively low expected cost corresponds to the high fraction of ma-

nipulating CEOs. The model-implied fraction of CEOs who violate GAAP

by misstating earnings at least once is 60%, and 2%–22% of CEOs starting

4See, for example, Feltham and Xie [1994], Crocker and Slemrod [2007], and Goldman

and Slezak [2006].

5Note the CEO who manipulates in year 1 faces a 2.95% chance of being caught and,

if he is not caught, another 2.95% in year 2, and so on. If the CEO stays with the same

firm for five years, the probability of detection is g+(1 −g)g+(1 −g)2g+(1 −g)3g+

(1 −g)4g|g=2.95% =13.91%.