Corporate Tax Avoidance and Fraud Risk_

Corporate Tax Avoidance and Fraud Risk

Chapter 1: Introduction

There are no universal definitions of tax avoidance or tax aggressiveness in the

accounting research literature (Hanlon and Heitzman 2010; Frank et al. 2009). Rego

(2003) describes tax avoidance as the application of legal methods to minimize the

amount of tax owed to the government. Frank et al. (2009) characterizes tax

aggressiveness as the manipulation of taxable income through tax avoidance strategies

that may or may not be considered tax evasion. Similar to Hanlon and Heitzman (2010), I

define corporate tax avoidance (CTA) as a continuum of tax planning strategies, from

perfectly legal activities (e.g., municipal bond investments) to more aggressive activities

(e.g., abusive tax shelter) that may fall into grey areas.

Corporate taxes are compulsory contributions collected from firms by the

government and represent a significant cost to companies and shareholders. Before the

Tax Cuts and Jobs Act of 2017 (TCJA), companies could contribute more than one-third

of pre-tax income to the government (Chen at al. 2010). Therefore, CTA may be desired

for shareholders because, based on traditional view of wealth transfer, CTA increases

cash flow to companies and provides additional value to shareholders (Cook et al. 2008;

Dhaliwal et al. 2004). However, the agency theory view of aggressive CTA suggests that

managers may use complex CTA strategies for their own benefits at the expense of

shareholders, including aggressive financial reporting and related party transactions

(Chen et al. 2010; Desai and Dharmapala 2009). The relationship between tax avoidance

and aggressive financial reporting has been well explored (Desai and Dharmapala 2009;

Erickson et al. 2004; Frank et al. 2009; Lennox et al. 2012). However, few researchers

have explored the relationship between tax avoidance and fraud risk

. Audit standard

2401 requires audit procedures should be performed to assess fraud risks relative to

aggressive financial reporting (PCAOB AS2401). In addition, prior studies find that tax

avoidance is associated with accounting fraud (Erickson et al. 2004; Lennox et al. 2012).

Prior studies use the sample of accounting fraud firms

that were caught and formally

charged by the SEC. In this study, I examine the relationship between tax avoidance and

fraud risk. Unlike accounting fraud, fraud risk can be assessed for all companies.

Accounting fraud (fraudulent financial reporting) is defined as the deliberate

manipulation of financial statements by company’s managers to build a distorted picture

of financial condition, results of operation, and cash flow to deceive creditors and

shareholders (Nicholas 2021). Managers can manipulate financial statements by

overstating revenues or understating expenses, and misrepresenting assets and liabilities.

Fraud risk is defined as the auditors’ assessment of client’s incentives, pressures, and

opportunities to commit fraud. Brazel et al. (2009) examine the relationship between

accounting fraud and fraud risk and find that companies that are committing fraud exhibit

higher levels of fraud risk.

In this study, I explore the relationship between CTA and fraud risk. Prior studies

find that aggressive tax avoidance is associated with accounting fraud (Erickson et al.

In this study, fraud refers to accounting fraud and fraud risk refers to the risk that the entity will commit

accounting fraud.

Similar to Dechow et al. (2011) and Lennox et al. (2012), I define fraud companies as those committing

accounting fraud on the AAER list.

2004; Lennox et al. 2012). I use various measures to assess fraud risk, including accrual

quality, performance variables, and non-financial measures (NFMs) (Dechow et al.

2011). Dechow et al. (2011) find that fraud companies show unusually high accruals in

fraud years. NFMs are additional proxies I use to measure fraud risk. The PCAOB (2004)

notes that analytical procedures with financial information are not sufficient in detecting

fraud due to managers’ manipulation of financial statements and suggests financial and

NFMs be combined in detecting accounting fraud. Brazel et al. (2009) find that fraud

companies show larger difference between financial and NFMs than non-fraud

companies, suggesting that NFMs can be used to assess fraud risk. Market incentives are

focused on financing from debt securities (Dechow et al. 1995). Dechow et al. (2011)

argue that fraud companies raise more capital through debt securities in fraud years than

non-fraud companies.

Prior studies also provide arguments and evidence on how tax aggressiveness

related to aggressive financial reporting (Frank et al. 2009; Lennox et al. 2012). Some

studies argue that there is a strong and positive relationship between CTA and aggressive

financial reporting (Desai 2005; Desai and Dharmapala 2006; Frank et al. 2009).

According to this view, managers may report different amounts of income to the public

(higher) and the IRS (lower) (Desai 2005; Hanlon et al. 2012). Desai (2005) and Frank et

al. (2009) suggest that areas of nonconformity between financial and tax reporting

provide more opportunities for companies to maximize book income in the financial

statements and minimize the taxable income simultaneously. However, Erickson et al.

Frank et al. (2009) define tax aggressiveness as aggressive tax avoidance, which may be legal or illegal.

(2004) find that aggressive financial reporting is negatively related to tax aggressiveness

because companies intentionally overpaid their taxes in order to validate the fraudulent

financial income. The evidence indicates that some companies overstate their tax

obligations to cover the aggressive accounting reporting. Similar to Erickson et al.

(2004), Lennox et al. (2012) argue that companies could not manipulate taxable income

and book income in the opposite directions without being noticed by external auditors

and the IRS. Lennox et al. (2012) examine the relationship between tax reporting

aggressiveness and accounting fraud and find that companies engaging in aggressive tax

reporting are less likely to commit accounting fraud. However, I expect that CTA is

positively related to fraud risk since the agency theory view of CTA suggests that

managers can exploit financial information to conceal rent extraction

by applying

various tax strategies (Desai 2005; Desai and Dharmapala 2006). According to the fraud

risk triangle framework, fraud risk is greater when managers have more opportunities to

manipulate financial statements. Therefore, I predict there is a positive relationship

between CTA and fraud risk.

Furthermore, I examine whether the relationship between aggressive CTA and

fraud risk, is different for fraud and non-fraud companies. Specifically, I explore whether

fraud companies really overpay taxes to conceal their frauds (Erickson et al. 2004).

Drawing on fraud triangle framework, I expect that both fraud and non-fraud companies

show a positive relationship between aggressive CTA and fraud risk. However, I expect

Rent extraction refers to managers’ effort to increase their own wealth without creating additional values

for firms and shareholders.

that the magnitude of the effect of aggressive CTA on assessed fraud risk is greater for

fraud companies than for non-fraud companies because managers that commit fraud have

more opportunities and incentives to manipulate financial statements, and they have more

justifications for their fraud behaviors.

This study investigates the relationship between CTA and fraud risk using a sample

spanning the 2000-2017 timeframe. Similar to Armstrong et al. 2015; Dyreng et al. 2008;

and Robinson et al. 2010, I use the GAAP effective tax rate (GAAPETR), the Cash

effective tax rate (CashETR) and permanent book tax difference (PBTD) to measure

CTA. Lower ETR and higher PBTD indicate more aggressive CTA. Fraud risk is

measured from perspectives that include accrual quality, performance variables, and

NFMs. I examine accrual quality by using changes in receivables and inventory because

these two accruals are related to revenue recognition and cost of goods sold that affect

gross profit. Financial data are collected from Compustat. Similar to Brazel et al. (2009)

and Dechow et al. (2011), I collect revenue related NFMs, such as the number of

employees and the amount of order backlogs from Compustat. I calculate the average

change in NFMs and subtract the change in NFMs from change in sales or total assets to

determine abnormal changes in NFMs. Consistent with Brazel et al. (2009), I define

abnormal changes in NFMs of more than 20% as high fraud risk. Then I use ordinary

least square (OLS) regression to test whether CTA is positively or negatively related to

fraud risk.

Generally, I find that CTA is positively related to fraud risk when I use financial and

non-financial variables to measure fraud risk. My results are consistent with fraud

triangle theory and agency theory such that complex tax planning strategies provide more

opportunities for managers to conduct rent extraction. However, I do not find a

significant difference between fraud and non-fraud companies. Due to the considerations

of potential penalties from the IRS, reputational damage from the public, and other costs

associated with CTA, fraud companies may overpay taxes to cover up their manipulation

of financial reporting.

This study contributes to the CTA and fraud risk literature in several ways. First,

while prior studies provide competing arguments regarding whether CTA is positively or

negatively associated with financial reporting aggressiveness, I conduct an empirical

analysis to explore the relationship between CTA and fraud risk. I find that CTA is

positively related to fraud risk because managers may use complex CTA activities to

conduct rent extraction. Auditors and regulators should focus on aggressive CTA since it

may indicate higher fraud risk. Second, this study contributes to the literature relating to

fraud risk. I use accrual quality, performance variables, and NFMs to measure fraud risk.

Managers, auditors, and regulators may evaluate the effect of CTA on fraud risk from

different perspectives. Third, this study contributes to the literature on agency problems

related to CTA. Little is known about whether the relationship between CTA and fraud

risk is different for fraud and non-fraud companies. Fraud companies may incur higher

agency costs than non-fraud companies. Finally, this study provides real-world tools for

business and regulators. Aggressive CTA may be considered a red flag for fraudulent

financial reporting since a higher level of CTA is positively related to higher fraud risk.

The remainder of the paper is organized as follows. In section II, I present the

conceptual framework, literature review and hypothesis development. In section III, I

discuss the research method and data collection. In section IV, I present the research

results. In section V, I discuss the contributions, limitations, and future research

opportunities.

Chapter 2. Literature Review

2.1 Conceptual Framework

2.1.1 Corporate Tax Avoidance (CTA)

According to the key terms on the Internal Revenue Service (IRS) website, tax

avoidance is defined as a perfectly legal method to reduce tax liability and increase

aftertax income. Taxpayers are allowed to use deductions and adjustments to reduce

taxable income and credits to reduce tax liability owed to the government. However, tax

research provides various definitions of corporate tax avoidance and tax aggressiveness

(Hanlon and Heitzman 2010; Frank et al. 2009). Rego (2003) defines tax avoidance as a

legal application to minimize taxes owed to the government. Frank et al. (2009) describes

tax aggressiveness as manipulation of taxable income through tax avoidance strategies

that may or may not be considered tax evasion. Consistent with Hanlon and Heitzman

(2010), I broadly define CTA as a continuum of tax planning strategies, from perfectly

legal activities (e.g., municipal bond investments) to more aggressive activities that fall

into grey areas.

CTA activities are traditionally considered tax saving tools that transfer wealth

from the state to corporations, thus increasing net cash flows and firm performance (Cook

et al. 2008; Dhaliwal et al. 2004; Wilson 2009). From this point of view, CTA provides

benefits to companies and shareholders. However, the agency theory view of CTA argues

that managers may use opaque CTA to hide rent extraction and work for their own

benefits. The total costs from CTA planning activities, compliance, and non-tax activities

(e.g., agency cost) may exceed the tax benefits from CTA. Therefore, CTA activities may

reduce firm performance.

Slemrod (2004) provides some background information for understanding the

agency theory of CTA. He states that separation of ownership and control in large

publicly held companies causes an agency problem between owners and agents. Unlike

private companies in which owners make decisions on tax reporting, large companies

assign decision making on tax reporting to their agents. Therefore, shareholders in large

companies need to create appropriate compensation packages to match shareholders’

interests with managers’ interests. To encourage managers to choose the value enhancing

tax reporting strategies, companies need to link compensation packages to after-tax

income.

Desai and Dharmapala (2006) examined how high-powered incentives affect CTA.

Their model is based on the agency theory that states that managers who make CTA

decisions also gain personal benefits through rent extraction. Managers make decisions

on CTA and rent extraction at the same time. So, CTA and rent extraction are

complements. The level of CTA may affect the cost of rent extraction for managers. For

instance, managers who make tax sheltering decisions may experience low costs of rent

extraction since complex tax sheltering activities can hide managerial rent extraction.

Desai and Daharmapala (2006) argue that the effect of high-powered incentives on CTA

is dependent on the relationship between CTA and rent extraction. Generally,

highpowered incentives that are related to after-tax profits stimulate managers to conduct

more aggressive CTA and reduce rent extraction. Higher incentive compensation is

helpful in aligning the interests of shareholders and managers and causes managers to be

more willing to create additional firm value through CTA activities. However, the

complementary relationship between CTA and rent extraction may reverse this result.

Particularly, the direct effect of high-powered incentives on CTA (e.g., higher incentive

compensation causes more aggressive CTA) could be offset by the positive feedback

effect between CTA and rent extraction (e.g., a reduction of managerial rent extraction is

accompanied by a reduction in CTA). Desai and Dharmapala (2006) also predict the role

of corporate governance in moderating the relationship between high-powered incentives

and CTA. Companies with poor corporate governance provide more opportunities for

managerial rent extraction than well-governed companies. Therefore, for poorly governed

companies, high-powered incentives are negatively related to CTA since the tendency

toward more aggressive CTA is offset by the positive feedback effect between CTA and

rent extraction (e.g., a reduction in rent extraction and a reduction in CTA exist at the

same time).

In addition, Desai and Dharmapala (2006) argue the extent to which CTA and rent

extraction are complementary may vary in different information environments.

Transparent companies may engage in less aggressive CTA activities than opaque

companies. Furthermore, they suggest CTA may have greater effect on firm value for

transparent companies than opaque companies. In this study, I examine the relationship

between CTA and fraud risk for fraud and non-fraud companies. Fraud companies may

conduct more aggressive CTA activities than non-fraud companies because managers in

fraud companies may use more complex activities to hide rent extraction. Therefore, I

predict that the effect of CTA on fraud risk is greater in magnitude for fraud companies

than for non-fraud companies.

2.1.2 Accounting Fraud and Fraud Risk

Accounting fraud is defined as the deliberate manipulation of financial statements

by a company’s managers to build a distorted financial condition to deceive creditors and

shareholders (Nicholas 2021). Managers can manipulate financial statements by

overstating revenues, understating expenses, or misrepresenting assets and liabilities. The

auditors’ assessment of fraud risk includes the assessment of a client’s incentives,

pressures and opportunities to commit fraud. A prior study examines the positive

relationship between accounting fraud and fraud risk and finds fraud companies exhibit

higher fraud risk (larger difference between financial measures and NFMs) (Brazel et al.

2009). The PCAOB (2004) suggests that auditors combine financial data and NFMs in

detecting accounting fraud. The PCAOB indicates that analytical procedures with only

financial data are not suitable to detect fraud because managers can create false financial

information to reach their objectives. Moreover, Brazel et al. (2009) find that NFMs can

be used to assess fraud risk. They define fraud risk as difference between the financial

and NFMs growth rate.

A prior study provides different methods to identify material misstatements

(Dechow et al. 2011), which are the primary indicators of fraud risk. First, Dechow et al.

(2011) investigates accrual quality related variables, such as working capital accruals and

discretionary accruals. For working capital accruals, they analyze two specific accruals

that impact firm performance (change in receivables and change in inventory) since these

two accounts are related to revenue recognition and cost of goods sold. They find that all

fraud companies have unusually high levels of abnormal accruals and have greater ability

to manipulate short-term earnings. For discretionary accruals, they examine various

models of discretionary accruals developed in prior studies (Dechow et al. 1995; DeFond

and Jiambalvo 1994; Kothari et al. 2005). Dechow et al. (2011) notes that the residuals

from the discretionary accruals have less power to detect earnings manipulation than

working capital accruals. Second, they explore firm performance variables for fraud

companies, including returns on assets and cash sales. They find that returns on assets

(ROA) are generally increasing in fraud companies, suggesting that fraud companies

attempt to increase earnings through manipulation. However, they find that cash sales are

increasing. This could be caused by the increasing capital investments and the expanding

business operations.

Third, Dechow et al. (2011) use NFMs to detect material misstatements. One NFM

is defined as the percentage change in the number of employees minus the percentage

change in total assets. A reduction in employees compared to total assets could indicate

manipulated asset balances. Also, they describe another NFM as the difference between

the percentage change of order backlog and the percentage change of revenues. They find

that the NFM of the abnormal change in the number of employees is helpful in detecting

material misstatements. Finally, they investigate stock and debt market related variables.

They notice that fraud companies more actively raise capital through debt securities than

non-fraud companies during fraud periods. However, for the same fraud companies, they

find that the extent of financing during fraud periods is not significantly different from

early years in the company’s life. Furthermore, they examine price to earnings and market

to book ratios and note that these two ratios are extremely high for fraud companies

relative to non-fraud companies.

2.1.3 Agency Theory

The economic theory of Agency was first developed by Ross (1973) to interpret

and solve problems in the relationship between principals (shareholders) and agents

(managers). Principals have employed agents to operate business on their behalf. There

are many different opinions, priorities and interests between principals and agents

because agents are delegated to make decisions that may financially affect principals.

Principal-agent problems exist when the interests of owners are not aligned with those of

managers. Based on the definition of agency theory, principals provide resources but do

not have daily input in business operations. Agents use the resources to make business

decisions and take little or at least less risk because all losses are shared by principals

(Kopp 2020). Regarding tax avoidance, risk neutral shareholders hire managers to

maximize profits through efficient tax planning tools. But opportunistic managers may

conduct aggressive tax avoidance actions for their own benefits (Desai and Dharmapala

2009) and can utilize opaque CTA activities to mask rent extraction behaviors and

unfavorable information. Rent extraction is defined as non-value maximizing activities

that managers conduct at the expense of shareholders, including aggressive financial

reporting and related-party transactions (Chen et al. 2010). Some CTA activities, such as

seeking offshore tax havens and creating related-party transactions, are complex and easy

for managers to conceal rent extraction (Desai and Dharmapala 2006). One of the

examples of related-party transactions in rent extraction is Enron’s CFO, Andrew Fastow,

creating special purpose entities (SPEs) to transfer resources from Enron to SPEs

(McLean and Elkind, 2003). To generate benefits for themselves, managers can buy

assets at higher prices than the market dictates, pay higher consulting fees, and borrow

money at higher interest rates from SPEs (Chen et al. 2010). Managers in Dynegy

Company overstated operating cash flows by three hundred million dollars by

misclassifying cash flows produced from CTA activities as operating cash flows. Enron

Company’s management inflated its earnings until 2001 by using twelve large tax shelters

to hide its poor performance from operations. Tyco International Company utilized the

complex CTA activities to hide its rent extraction behaviors, resulting in the company’s

stock price crash after rent extraction being disclosed in 2002. Therefore, opportunistic

managers may use CTA strategies to reduce companies’ cash flows and firm performance

and increase managers’ opportunities to commit fraud.

2.1.4 Fraud Risk Triangle Framework

The fraud risk triangle (Cressey 1953) is a framework used by auditors to explain

the characteristics that must be present for a fraud to take place. It includes three

elements: opportunity, incentive, and rationalization that contribute to increasing fraud

risk. Opportunity is described as conditions under which people are more able to commit

fraud. For instance, weak internal control such as lack of separation of duties gives

employees more opportunities to perpetrate fraud. Incentive is defined as employees’

motivation towards committing fraud. Incentive-based compensation and meeting

investors’ expectations may create pressure to conduct fraudulent activities.

Rationalization is represented as employees’ justification for perpetrating fraud.

Managers that feel unfairly treated may commit fraud to get payback. Some CTA

activities such as tax-free municipal bonds investment and employees’ 401(K) and

pension plans are simple and straight forward. But other CTA activities are complicated

and obscure, including contested liability acceleration strategies, cross-border dividend

capture, and intellectual property havens (Graham and Tucker, 2006). Managers can take

advantage of the obscure nature of aggressive CTA to conceal rent extraction. Therefore,

complex CTA creates more opportunities and provides incentives and rationalization for

managers to commit fraud.

2.2 Literature Review

2.2.1 Determinants of CTA

Tax research has drawn significant attention in the last five decades. A number of

studies examine what factors cause companies to engage different tax avoidance

strategies. Some studies indicate that CTA is associated with a number of firm and

executive characteristics. Zimmerman (1983) examines the effect of firm size on effective

tax rate (ETR). Since large companies are subject to more scrutiny, their CTA strategies

are less aggressive than small companies. In addition, Zimmerman (1983) argues that

political cost plays an important role in determining CTA. He finds that firm size is

positively related to effective tax rate. Large companies are less aggressive in engaging

CTA. A prior study finds that firm size is not significantly related to fraud risk (Lawrence

et al. 2011). In this study, I control for firm size when I examine the relationship between

CTA and fraud risk. Gupta and Newberry (1997) explore the relationship between ETR

and firm characteristics other than firm size, such as capital structure, asset mix, and firm

performance. Using longitudinal data covering the Tax

Reform Act of 1986, they find that ETRs are not related to firm size when examining

firms with longer histories. However, their results indicate ETRs are related to capital

structure, asset mix and firm performance. Since prior studies show the mixed results on

the relationship between firm size and CTA, I include firm size as a control variable to

see how firm size affects CTA and fraud risk. Rego (2003) examines the effect of

economies of scale on companies’ tax planning strategies. She finds that companies with

higher profits and more foreign transactions have lower ETRs. Therefore, economies of

scale have significant effect on CTA activities. In short, to illustrate different CTA

strategies adopted by companies, these studies explore research opportunities from firm

characteristics.

Some studies extend the CTA research to incentive compensation, ownership

structure, and organization structure. Phillips et al. (2003) examine the relationship

between incentive compensation (e.g., the link of after-tax profits in CEO and managers’

bonus plan) and CTA activities. Compensation plans that are linked to after-tax profits

motivate CEOs and managers to make aggressive tax avoidance decisions. Phillips et al.

(2003) find that incentive compensation is positively related to CTA activities. Higher

incentive compensation that is linked to after-tax profits causes managers to engage in

more aggressive CTA activities. However, Phillips et al. (2003) don’t find a positive

relationship between compensation plans and after-tax profits for CEOs.

In addition, Desai and Dharmapala (2006) examine how high-powered incentive,

option-based executive compensation, is related to corporate tax sheltering. They suggest

the relationship between incentive compensation and tax sheltering depends on corporate

governance. They argue that managers in poorly governed companies have more

opportunities to conduct rent extraction than those in well-governed companies. For

poorly governed companies, high-powered incentives are negatively related to tax

sheltering because the trend toward aggressive CTA is offset by the positive feedback

between CTA and rent extraction. Therefore, Desai and Dharmapala (2006) note that both

incentive compensation and corporate governance play a significant role in determining

CTA activities.

Furthermore, Armstrong et al. (2010) investigate how the incentives of tax

directors affect tax planning. By using proprietary data with executives’ compensation

information, they examine the relationship between tax directors’ incentives and proxies

for CTA. They find that tax directors’ incentives are strongly and negatively related to

GAAP ETR. However, there is little relationship between tax directors’ incentives and

either cash ETR or book-tax differences. They illustrate that tax directors are motivated to

reduce tax expenses in the financial statements, not to save cash flows.

Unlike Phillips et al. (2003), Desai and Dharmapala (2006), and Armstrong et al.

(2010), Rego and Wilson (2008) examine how tax planning strategies affect executive

compensation. They find that executives are motivated to be aggressive in tax avoidance.

They argue that the positive relationship between tax aggressiveness and executive

compensation indicates efficient contracting, instead of rent extraction. Robinson et al.

(2010) investigate how tax department structure is related to CTA strategies. Corporate

tax departments can be structured as profit centers or cost centers. Robinson et al. (2010)

find that tax departments that are structured as profit centers can more effectively reduce

GAAP ETR, but not cash ETR.

The studies cited above examine how incentive compensation affects CTA from

different perspectives. Incentive compensation may also relate to aggressive financial

reporting. Managers with incentive compensation are motivated to manipulate financial

statements. Based on the fraud triangle concept, managers with incentive compensation

have more incentives to commit fraud. In this study, I include CEOs’ incentive

compensation as one of control variables to explore how incentive compensation affects

CTA and fraud risk.

Chen et al. (2010) examines the role of ownership structure in determining CTA

activities. Particularly, they investigate how the agency issues between dominant

shareholders (family members) and minority shareholders (non-family members) affect

CTA in family-owned businesses. Using multiple proxies to measure tax aggressiveness,

they find that family-owned companies engage in less aggressive CTA activities than

non-family-owned companies. They argue that to avoid non-tax costs from CTA,

familyowned companies are willing to sacrifice the tax benefits by taking less aggressive

CTA activities. Non-tax costs from tax avoidance in family-owned businesses include

potential penalties from the IRS, reputational damage from the public, and a loss of value

from a minority stake. Their results also indicate that family-owned companies care about

nontax costs more than non-family-owned companies. Since family-owned companies

are more concerned about non-tax costs, I expect family-owned companies are more

conservative in financial reporting. Therefore, family-owned companies are less likely to

commit fraudulent financial reporting.

In addition to studies that examine CTA determinants from a company’s

perspective, a number of studies investigate the determinants of CTA from the executive,

audit committee, and board of director’s perspectives. Gaertner (2014) examines the

relationship between the CEOs’ after-tax incentives and CTA. He finds that the use of

after-tax incentives is negatively related to ETRs. After-tax incentives motivate CEOs to

engage in more aggressive CTA. He also finds that CEOs’ cash compensation is

positively related to after-tax incentives. The result suggests that CEOs are rewarded for

engaging in more aggressive CTA. Goldman et al. (2017) investigate how CEOs tenure

affects corporate tax planning. They find that GAAP ETR decreases from the early years

to the later years of CEOs’ tenure and is the lowest during the CEOs’ final year.

However, cash ETR does not change during CEOs’ tenure. Their results indicate that

CEOs are more aggressive in financial reporting of income taxes than in corporate tax

planning and are more aggressive in the final tenure year. In addition, they find that

CEOs reinvest earnings permanently and use discretion for uncertain tax benefits. Liang

(2019) examines the relationship between CEOs’ age and CTA. He finds that CEOs’ age

plays a significant role in determining CTA policies. Particularly, he finds a positive

relationship between CEOs’ age and GAAP and Cash ETRs, and a negative relationship

between CEOs’ age and permanent book-tax differences. The results indicate that older

CEOs are less aggressive in tax avoidance. Older CEOs may also be conservative in

financial reporting. In this study, I add CEOs’ tenure and age to my model to examine the

impact of CEOs’ tenure and age on fraud risk.

Olson and Stekelberg (2016) examine how CEO narcissism affects corporate tax

sheltering. Narcissism is a personality trait that is linked to a feeling of dominance.

Narcissists don’t have moral awareness and are aggressive in chasing their goals. Olson

and Stekelberg (2016) find that CEO narcissism is positively related to tax sheltering.

They also find that CEO narcissism is positively related to uncertain tax benefits and

negatively related to cash ETR. A prior study finds that CEO narcissism is related to

earnings management (Capalbo et al. 2018). Narcissistic CEOs are more likely to

manipulate financial statements by overstating earnings, thus raising fraud risk. Chen et

al. (2019) investigate how CFO’s accounting expertise affects CTA. Accounting expertise

is highly related to CTA since managers use accounting knowledge in determining

taxable income and making adjustments based on book-tax differences in preparing tax

returns. Thus, CFOs’ accounting expertise is helpful in managing income taxes and

accounting for the effect of CTA on financial statements. Chen et al. (2019) find that

CFO’s accounting expertise is negatively related to ETRs. In addition, they find that

CFOs’ abnormal variable compensation is negatively related to ETRs. The results

indicate that CFOs’ accounting expertise and compensation plan play a significant role in

determining CTA activities. CFOs’ accounting expertise is also related to aggressive

financial reporting. CFOs with accounting expertise understand better how to manipulate

financial statements, thus increasing fraud risk.

Lanis and Richardson (2011) examine how the composition of the board of

directors affects tax aggressiveness. They find that the proportion of external members on

the board is negatively related to tax aggressiveness. This result suggests that external

board members are more independent so that they are more likely to prevent tax

aggressiveness through better governance. In addition, Lanis et al. (2017) extends the

research on the composition of the board of directors to investigate the effect of gender

diversity in the board of directors on tax aggressiveness. They find that female

representation on the board is negatively related to tax aggressiveness. The result

indicates that female board members are more risk-averse and more likely to deter tax

aggressiveness through better monitoring. Since female board members are more

riskaverse and conservative than their male counterparts, they are more concerned about

aggressive financial reporting. Female board members are more likely to reduce fraud

risk. In this study, I include female board members as one of control variables.

Robinson et al. (2012) examine the role of the audit committee in advising and

monitoring tax planning strategies. More specifically, they examine the effect of

accounting expertise on the audit committee on CTA. They find that the level of

accounting expertise on the audit committee is negatively related to CTA. The result

suggests that audit committee with accounting expertise may advise and monitor firm tax

planning, thus reducing firm’s aggressive tax avoidance. One prior study, Cohen et al.

(2014), indicates that audit committee accounting expertise is very valuable in detecting

and preventing fraudulent financial reporting.

2.2.2 Consequences of CTA

There are several possible consequences of CTA, which may be direct, such as

increasing a firm’s cash flow and shareholders’ wealth (Cook et al. 2008; Dhaliwal et al.

2004), or indirect, such as affecting a firm’s capital structure (Graham and Tucker, 2006).

One of the direct consequences of CTA is that a firm’s illegal tax activities may be

detected by the IRS or other authorities. Firms and managers may face penalties and

litigation, which may negatively affect a firm’s cash flow, stock price, and reputation. The

literature explaining the consequences of CTA is primarily focused on earnings

management, stock market reaction, firm risk, and accounting and auditing issues.

Prior studies show mixed results on whether CTA increases or decreases firm

earnings. Dhaliwal et al. (2004) examines the effect of tax expenses on earnings

management. They find that firms may manipulate tax planning to reduce ETR in the last

two quarters if pre-tax accruals earnings management does not meet the target. The

results indicate that CTA can be used as a tool to increase earnings. Cook et al. (2008)

investigate how the amount of tax fees paid to auditors is related to the change of ETR in

the last two quarters of the year. Consistent with Dhaliwal et al. (2004), Cook et al.

(2008) find that firms can change tax expenses to manage earnings and that the amount of

tax fees paid to auditors is positively associated with the change in ETR from the third to

the fourth quarter. However, building on the agency theory, Desai and Dharmapala (2009)

argue that CTA may not be positively related to firm value because managers may use

complex CTA for their own benefits. Chen et al. (2010) examines the effect of ownership

structure on CTA and argue that agency costs incurred as a result of CTA may exceed the

benefits, thus decreasing firm performance.

Hanlon and Slemrod (2009) investigate how the stock market reacts to the

announcement of news about companies’ participation in tax sheltering. They find that

stock price is negatively related to the first announcement of companies’ tax sheltering

behavior. They also document that consumer-related companies show a more negative

relationship between stock price and the announcement of tax sheltering. Frischmann et

al. (2008) analyzes how the stock market responds to the implementation of FIN 48,

which is an interpretation of the rules requiring all business entities to disclose the

taxrelated risks. They find that the stock market reacted very little to the passage of the

rule.

However, companies with the first disclosures under the rule show a small positive

return. The results suggest that the stock market’s reaction to the rule depends on

investors’ expectations. If investors anticipate an increase in tax costs, the stock market

responds negatively. In contrast, if investors expect an improvement by implementing

FIN48, the market reacts positively.

Guenther et al. (2017) examine the effect of CTA on firm risk by using several

measures of CTA. They find that CTA is not related to future tax rate volatility or overall

firm risk. The result indicates that companies use consistent strategies in engaging in

CTA, which does not increase firm risk. They also find that the volatility of cash ETR is

negatively related to the volatility of stock price. Companies may use complex CTA

activities to conceal managerial rent extraction, which increases firm risk. Kim et al.

(2011) investigate the relationship between CTA and a stock price crash and find that

CTA increases the risk of stock price crashes at the firm level. The result suggests that

CTA is accompanied by managerial rent extraction. The accumulation of managerial rent

extraction for the long term may cause a stock price to crash. They also find that the

positive relationship between CTA and stock price crash is reduced when companies have

strong governance and monitoring systems.

Aggressive CTA may have accounting and auditing consequences. Managers may

use complex CTA activities to manipulate firm’s earnings, thus reducing financial

reporting quality. Frank et al. (2009) examine the relationship between aggressive tax

reporting and aggressive financial reporting. They find that managers can use the areas of

nonconformity between tax and financial reporting to conduct tax avoidance and earnings

management. Therefore, aggressive CTA is positively related to aggressive financial

reporting. Donohoe and Knechel (2014) examine whether tax aggressiveness affects audit

pricing. Aggressive tax reporting increases auditors’ efforts in tax research and in

performing additional audit procedure. Auditors are exposed to the risk of litigation,

regulation, and reputation loss. Donohoe and Knechel (2014) find a positive relationship

between tax aggressiveness and audit fees.

In sum, the literature on the consequences of CTA examines the effects of CTA on

earnings management, stock market reaction, firm risk, and accounting and auditing

issues. Most of the studies explore the consequences of CTA using financial measures.

Little research explores the consequences of CTA using non-financial measures. In this

study, I extend the literature on consequences of CTA to fraud risk. I use different proxies

to measure fraud risk, including financial, and non-financial measures. Fraud risk is the

auditor’s assessment of their clients’ risk of committing accounting fraud. The level of

fraud risk may or may not be indicative of accounting fraud. The PCAOB suggests that

auditors assess fraud risks that are related to aggressive financial reporting. Therefore,

this study bridges the gap between aggressive financial reporting and accounting fraud.

2.2.3 Using NFMs to Measure Fraud Risk

Prior studies find that financial and nonfinancial measures of firm performance are

highly correlated (Brazel et al. 2009; Dechow et al. 2011). Companies that report an

increase in NFMs will likely exhibit a similar increase in revenues and net income. Some

airline companies use NFMs (such as the number of passengers) to predict financial

numbers such as total revenues and profits (Behn et al, 1999). Along with serving as

leading indicators for future financial performance, NFMs may be useful in detecting

fraudulent financial reporting. The PCAOB (2004) states that NFMs should be used as a

powerful benchmark to evaluate financial statement reliability and to detect fraud in

financial statements and internal control reports. Brazel et al. (2009) report that the

revenue growth rate is greater than the average NFMs growth rate in those companies

that have committed fraud. Specifically, Brazel et al. (2009) define high fraud risk as a

revenue growth rate exceeding NFMs growth rate by 20%. Unlike fraud companies,

nonfraud companies usually have better consistency between financial and nonfinancial

growth rates. Brazel et al. (2019) indicate that audit committee members can reduce fraud

risk by detecting inconsistencies between financial and non-financial measures. Audit

committee members with greater tenure and financial and industrial expertise are more

likely to detect large inconsistencies (fraud risk). Brazel et al. (2019) document that audit

committee members with greater tenure have better background information for

evaluating business operations. Cohen et al. (2014) find that audit committee industry

expertise is very valuable in monitoring external auditors and management in a specific

industry. They use financial restatements and discretionary accruals as two measures for

financial reporting quality. They report that audit committees with accounting and

industry expertise are associated with higher reporting quality than those with only

industry expertise.

2.2.4 The Relationship Between Aggressive Tax Reporting and Aggressive Financial

Reporting

There is mixed evidence on whether tax aggressiveness is positively or negatively

related to financial reporting aggressiveness. Shackelford and Shevlin, (2001) document

the trade-off that companies confront when they make decisions about financial and tax

reporting. Specifically, companies attempting to raise book income in financial

statements may experience higher tax cost when reporting a higher amount of book

income. Similarly, companies attempting to lower taxable income in the tax return may

report lower income in financial statements. Therefore, there is a negative relationship

between aggressive financial reporting and aggressive tax reporting (Ericson et al. 2004;

Lennox et al. 2012). In contrast, other studies find that companies do not always trade-off

financial and taxable income (Hanlon et al. 2012; Phillips et al. 2003). Management may

report different amounts of income to investors and creditors (higher) and the IRS

(lower). Desai (2005) suggests that areas of nonconformity between financial and tax

reporting provide more opportunities for companies to maximize book income in the

financial statements and minimize the taxable income simultaneously. Thus, there is a

strong and positive association between aggressive financial reporting and tax reporting

(Frank et al. 2009). Frank et al. (2009) develops permanent book-tax differences (BTDs)

as a proxy to measure tax reporting aggressiveness. Temporary BTDs represent pre-tax

accruals earnings management (Phillips et al. 2003). Measures of tax aggressiveness with

temporary BTDs may be falsely associated with proxy for aggressive financial reporting

since the relation is driven by pre-tax accruals earnings management, instead of tax

planning. Therefore, permanent book-tax differences are a better proxy for tax

aggressiveness (Frank et al. 2009).

Erickson et al. (2004) argue that companies may intentionally overpay their taxes

in order to validate fraudulent financial income. They choose a sample of accounting

fraud companies to analyze the taxes paid on the overstated earnings. They create a proxy

for accounting fraud from the issuance of Accounting and Auditing Enforcement

Releases (AAER) by the Securities and Exchange Commission (SEC), which describes

the SEC’s actions to enforce fairness in financial statement reporting through civil

litigation and administrative proceedings. The evidence suggests that some companies

overstate their tax obligations to cover the aggressive financial reporting. Furthermore,

Lennox et al. (2012) examine the relationship between tax reporting aggressiveness and

the incidence of accounting fraud and find managers cannot manipulate book income and

taxable income simultaneously. Aligned with Erickson et al. (2004), Lennox et al. (2012)

find that aggressive tax reporting is negatively related to aggressive financial reporting.

To cover up fraudulent financial reporting, companies may purposely overpay taxes. In

addition, Lennox et al. (2012) find that not all proxies (four of five proxies for ETR and

two of three proxies for BTD) for tax reporting aggressiveness are negatively related to

accounting fraud.

2.3 Hypothesis Development

According to the traditional view of wealth transfer, CTA may be used as a tax

saving tool to increase cash flows and create additional values for the firm and its

shareholders. Based on the traditional view, CTA has a positive effect on firm

performance and is not significantly associated with fraud risk. However, the agency

theory view of CTA suggests that aggressive CTA is accompanied by managerial rent

extraction and that managers can hide rent extraction by using various CTA strategies

(Desai and Dharmapala 2006; Chen et al. 2010). In addition, the fraud triangle framework

suggests that some complex CTA strategies may be used as one type of opportunity

(which is one corner of the fraud risk triangle) to commit fraud, as would other types of

opportunities. Therefore, I predict that CTA is positively related to fraud risk. I develop

my hypothesis as the following:

H1: CTA is positively related to fraud risk.

In addition, I examine whether the relationship between CTA and fraud risk is

different for fraud and non-fraud companies. Drawing on fraud risk triangle theory, I

expect that both fraud and non-fraud companies show a positive relationship between

CTA and fraud risk. But for fraud companies, the effect of CTA on fraud risk is greater in

magnitude than for non-fraud companies because managers have more opportunities and

incentives to manipulate financial statements. My second hypothesis is developed as the

following:

H2: The effect of CTA on fraud risk is greater in magnitude for fraud companies than for

non-fraud companies

Chapter 3. Research method

1. Sample Selection

I use a sample of all companies from 2000 to 2017. Similar to Brazel et al. (2009)

and Dechow et al. (2011), I collect revenue related NFMs, such as number of employees

and the amount of order backlogs. Previous research indicates the number of employees

and order backlog are highly associated with revenue and are recorded by most

companies as common NFMs (Brazil et al. 2009; Dechow et al. 2011). The number of

employees and order backlog are collected from Compustat. Financial data such as

revenues, accounts receivable, ETRs, and PBTD are also collected from Compustat.

Auditors’ tenure and audit fees are obtained from the Audit Analytics. Audit committee’s

tenure and chair’s gender are collected from BoardEx. Accounting fraud information is

collected from the SEC website and Lexis-Nexis AAER, a resource that contains the

results of the SEC’s investigation into accounting violations. A single fraud can cause

several AAERs as the SEC challenges and investigates different individuals implicated in

the fraud. I remove non-accounting frauds as they are not related to the research question.

Multivariate Model 1

To test my hypotheses whether CTA is positively or negatively related to fraud

risk, I develop the following regression model:

FRAUDRISK=β0+β1CTA+ β2AuditTenure + β3AuditFees +

β4ChairGender+β5ChairTenure+ β6LnTA+ β7Lev+ β8Loss+ β9ICME+ β10BM +

β11Restate+ β12Big4 + β13PCHGSales+ β14PA+ Year Dummies + Industry Dummies +ε

Dependent Variables

My dependent variable for the model is fraud risk (FRAUDRISK). I measure fraud

risk by using the following three proxies: accrual quality, NFMs, and performance

variables. Percentage change in receivables (PCHGREC), percentage change in inventory

(PCHGINV), and discretionary accrual (DISCACC) are used to measure accrual quality.

PCHGREC is the difference of current year’s accounts receivable and prior year’s

accounts receivable divided by prior year’s accounts receivable. PCHGINV is the

difference of current year’s inventory and prior year’s inventory divided by prior year’s

inventory. PCHGREC and PCHGINV are the two metrics that are closely evaluated by

investors because managers misstate these two accounts to increase revenues and gross

margin. NFMs are measured by two variables: DIFFBSE and DIFFBSO. DIFFBSE is the

difference between the percentage change of revenues and the percentage change of

employees. DIFFBSO is the difference between the percentage change of revenues and

the percentage change of order backlogs. Prior studies find that financial and nonfinancial

measures of firm performance are highly correlated (Brazel et al. 2009; Dechow et al.

2011). Inconsistency between financial and NFM performance indicates fraud risk.

Performance variables are measured by the percentage change of cash sales

(PCHGCASHSALES) and change of return on assets (CHGROA). PCHGCASHSALES

is the difference between the percentage change of total sales and the percentage change

in credit sales. CHGROA is net income divided by total assets minus prior period net

income divided by prior period total assets. The reason why PCHGCASHSALES and

CHGROA are used as proxies for fraud risk is fraud companies may increase sales and

earnings in fraud years. Therefore, I predict that PCHGCASHSALES decreases and

CHGROA increases during fraud periods.

Independent Variables

In this study, I focus on the relationship between CTA and fraud risk. The

independent variable of interest in this study is CTA. My goal is to examine the

relationships between CTA and fraud risk. Similar to Armstrong et al. 2015; Dyreng et al.

2008; and Robinson et al. 2010, I use GAAP ETR, Cash ETR and PBTD as proxies for

CTA. These CTA proxies are used to measure the effects of nonconforming transactions,

which have different impacts on financial and tax reporting (Lennox et al. 2012). Hanlon

and Heitzman (2010) define CTA as a continuum of tax planning activities. Conceptually,

GAAP ETR, Cash ETR and PBTD are connected to the continuum because CTA

strategies that produce PBTD decrease GAAP ETR and Cash ETR and increase book

income.

I define GAAP ETR as the ratio of the total tax expenses to the total pretax

BTDs and ETRs are relevant since BTDs refer to the difference between financial income and tax income

and ETRs reflect the ratio of taxes to income. In other words, BTDs represent the income effects of CTA

activities whereas ETRs represent the tax effects.

income minus special items for the same periods. I compute Cash ETR as a ratio of the

total cash tax expenses to the total pretax income minus special items. Since lower ETR

indicates high lever CTA, I predict the coefficient for GAAP ETR and Cash ETR to be

negative. PBTD is another proxy for aggressive tax reporting. PBTD is defined as the

total book-tax difference minus temporary book-tax difference, divided by total assets.

Total book-tax difference is equal to pretax income minus taxable income. The temporary

book-tax difference is equal to total deferred tax expense divided by the statutory tax rate.

Since large PBTD indicates a higher level of CTA, I expect the coefficient for PBTD to

be positive.

Next, I control for the characteristics of auditors and audit committee members.

First, I control for the tenure of auditors as an independent variable (AuditTenure) and the

tenure of the audit committee as an independent (ChairTenure). I expect the coefficient

for auditors’ tenure and audit committee tenure to be negative. Second, I control for

auditors’ effort (AuditFees) as the natural logarithm of the total audit fees billed in year t

(DeFond et al. 2005). I predict a negative coefficient for auditors’ effort.

Last, I control for gender of audit committee chair (Chairgender) as an indicator variable

of 1 if the gender of the chair is male and 0 otherwise. I predict a positive coefficient for

gender of audit committee chair since a female chair may be considered more

conservative in monitoring the audit engagement.

I also control for several variables associated with financial reporting quality

(Reichelt et al. 2010). First, I control for company size by using log of market value of

total assets (LnTA) because size is an important predictor of fraud risk (Lawrence et. al.

2011). Since large companies are subject to more strict scrutiny, I expect large companies

are more conservative with financial reporting. I predict a negative coefficient for

company size. Second, I control for financial leverage (Lev), which is measured by total

debts divided by total assets. High leverage value means great financial distress,

increasing fraud risk. I predict a positive coefficient for financial leverage. Third, I

control for operating loss as a dummy variable equal to 1 if there is an operating loss and

0 otherwise. Since companies with operating losses are more aggressive with financial

reporting, I predict a positive coefficient for operating loss. Fourth, I control for internal

control material effectiveness (ICME). The Sarbanes-Oxley Act section 404 (a) requires

the management to maintain an effective internal control system over financial reporting.

An ineffective internal control provides more opportunities for managers to commit

fraudulent financial reporting. Also, an ineffective internal control system represents an

organization environment that does not emphasize the integrity of financial reporting.

Therefore, I predict a negative relationship between internal control effectiveness and

fraud risk. Fifth, I control for operating growth measured by the market value of equity

divided by the book value of equity. Abbott et al. (2004) find that there is a negative

relationship between the growth rate of a company and financial reporting quality

because a growth company is likely to have a less effective internal control system. I

predict a positive coefficient for operating growth. Next, I control for financial

restatement (Restate) as an indicator variable equal to 1 if a financial restatement has

been reported during the last three years and 0 otherwise. I predict a positive coefficient

for financial restatement. Next, I control for sales growth rate (PCHGSALES), which is

defined as the difference between current year’s sales and previous year’s sales divided

by previous year’s sales. Since companies with rapid sales growth are more aggressive in

earnings management, I predict a positive coefficient for sales growth rate. Plant assets

(PA) is calculated as the net value of plant assets divided by total assets.

Multivariate Model 2

To test my hypotheses whether the effect of CTA on fraud risk is greater in

magnitude for fraud companies than for non-fraud companies, I develop the

following regression model:

FRAUDRISK=β0+β1Fraud*CTA+ β2Fraud+ β3CTA+β4AuditTenure + β5AuditFees +

β6ChairGender+β7ChairTenure+ β8LnTA+ β9Lev+ β10Loss+ β11ICME+ β12BM +

β13Restate+ β14Big4 + β15PCHGSales+ β16PA+ Year Dummies + Industry Dummies +ε

In model 2, I use an interaction term, Fraud*CTA, to examine whether CTA is more

significantly related to fraud risk for fraud companies than non-fraud companies. I use

Fraud as an indicator variable equal to 1 if a company is reported as an accounting

violation on the AAER website by the SEC and 0 otherwise. A company is defined as

Fraud when it is disclosed by the SEC for the current year. I don’t consider the pre or post

fraud period in this research. Since fraud companies have more incentives and

opportunities to use complex tax planning strategies to manipulate financial reporting, I

predict a positive coefficient for the interaction term Fraud*CTA. All other variables are

the same as those in model 1.

Chapter 4: Empirical Results

4.1 Descriptive statistics

The sample selection procedures are described in Table 1, Panel A. I searched

the entire Compustat database from 2000 to 2017 and start with 240,683 firm-year

observations. 136,889 firm-year observations are lost because they have either negative

or missing pre-tax income. Another 30,599 firm-year observations are excluded because

they do not have enough data to compute CashETR, GAAP ETR and PBTD. I deleted

25,726 firm-year observations to exclude financial and utility companies. I lost 22,636

firm-year observations when I merge Compustat, with Audit Analytics for auditors’

tenure and fees, and with BoardEx for audit committee chair’s tenure and gender. 449

firm-year observations are lost because they are foreign companies based on the foreign

incorporation code (FIC). I deleted 703 firm-year observations because of outliers for the

key variables. I define outliers as Z-score greater than 3 or less than -3. My final sample

includes 23,681 firm-year observations.

I report the industry distribution of firm-year observations in Table 1, Panel B. I

illustrate industry membership according to the classification scheme in Dechow et al.

(2011). There are 13 industries included in the 23,681 firm-year observations. My sample

is concentrated in the industries of transportation, computers, durable manufactures,

retail, services, and pharmaceuticals, with more than 1,500 or 5% of the firm-year

observations from each industry.

I report the distribution of the key variables of interest (CashETR, GAAPETR,

PBTD, and fraud risk) by industry in Table 2. Companies in the computers and

transportation industries show the lowest CashETR (e.g., <0.23). The agriculture, retail,

and textiles and apparel industries indicate the highest CashETR (e.g., >0.30). Companies

in the computers industries show the lowest GAAPETR (e.g., <0.28). Companies in the

transportation, retail, and service industries indicate the highest GAAPETR (e.g., >0.35).

Companies in the textiles and apparel industries show the lowest PBTD (e.g., <0.025).

Companies in the refining and extractive industries indicate the highest PBTD (e.g.,

>0.08). Companies in the lumber, furniture, and printing industries show the lowest fraud

risk (e.g., PCHGREC and PCHGINV<10%, DISCACC<0.03). The mining and

construction, and pharmaceutical industries exhibit the highest fraud risk (e.g.,

PCHGREC and PCHGINV>16%, DISCACC >0.02). The results of the key variables of

interest are comparable to those in Frank et al. (2009), Chen et al. (2010), and Lennox et

al. (2012).

I report the descriptive statistics for CashETR, GAAPETR, PBTD, fraud risk, and

other control variables in Table 3. The values of mean and median for CashETR are 0.250

and 0.248, respectively. The values of mean and median for GAAP ETR are 0.319 and

0.342, respectively. Both the mean and median values of CashETR and GAAPETR are

similar to those in Chen et al. (2010) and Lennox et al. (2012). The values of mean and

median for PBTD are 0.046 and 0.032, respectively, which are comparable to those in

Frank et al. (2009) and Lennox et al. (2012). The values of mean and median for the

percentage change of accounts receivable are 0.203 and 0.085, respectively. The values of

mean and median for the percentage change of inventory are 0.117 and 0.074,

respectively. The values of mean and median for discretionary accruals are -0.032 and

0.034, respectively, which are consistent with those in Dechow et al. (2011) and Frank et

al. (2009). The mean and median values for the difference between the sales growth rate

and the number of employees growth rate are -0.027 and 0.036, respectively. The values

of mean and median for the percentage of cash sales are 0.107 and 0.070, respectively.

The values of mean and median for the return on assets are 0.085 and 0.069, respectively.

These values are consistent with those reported in Dechow et al. (2011).

Table 3 also includes descriptive statistics for other control variables. The average

annual audit fees for the sample are $2,354,010 with the natural logarithm of

13.87. The average auditor’s tenure for the sample is 23.76 years, which is higher than

auditor’s tenure of 18 years in Brazel et al. (2019). One of the explanations is that my

sample includes longer periods and more variables than those in Brazel et al. (2019). The

average audit committee chair’s tenure is 8.07 years, which is similar to that in Brazel et

al. (2019). 90.92% of audit committee chairs are male. The average annual total assets for

the sample are $5,869 million with the natural logarithm of 6.80. The average percentage

change of sales is 12.60%. The average leverage of the sample is 17.93%. The average

book to market value ratio is 0.473. In addition, around 1.05% of companies report loss

and 80.85% of companies are audited by big four CPA firms. 11.97% of companies report

restatement. The average value for plant assets lagged by total assets is 0.246.

I report the correlations of the key variables in Table 4. GAAPETR, CashETR,

and PBTD are significantly correlated at 0.001 level. More specifically, GAAPETR and

CashETR are positively correlated, and the coefficient of correlation is 0.268. GAAPETR

and CashETR are negatively correlated to PBTD with the coefficient of correlation of -

0.064 and -0.456, respectively. The negative correlations indicate that GAAPETR,

CashETR and PBTD measure CTA from different perspectives of tax avoidance

strategies, which I explain in section 3.2. In addition, the proxies of fraud risk are

significantly correlated. PCHGREC is positively correlated to PCHGINV, DISCACC,

PCHGCASHSALES, and CHGROA with the coefficients of correlation of 0.197, 0.124,

0.156, and 0.129, respectively. However, NFM for fraud risk is not significantly

correlated to other financial measures. The main reason is that most companies do not

disclose the amount of order backlog in their financial statements.

Furthermore, Table 4 shows that the percentage change of accounts receivable is

negatively correlated to CashETR, and positively correlated to GAAPETR and PBTD

with the coefficient of correlation of -0.026, 0.007, and 0.04, respectively, indicating

higher level of CTA increases fraud risk. The percentage change of inventory is also

negatively correlated to CashETR, and positively correlated to GAAPETR and PBTD

with the coefficient of correlation of -0.052, 0.004, and 0.053, respectively. Discretionary

accrual is negatively correlated to GAAPETR and PBTD, and positively correlated to

CashETR. The percentage of cash sales is positively correlated to GAAPETR and PBTD,

and negatively correlated to CashETR with the coefficient of correlation of 0.075, 0.039,

and -0.048, respectively, suggesting high percentage of cash sales decreases fraud risk.

Return on assets is negatively correlated to GAAPETR, and positively correlated to

CashETR and PBTD with the coefficient of correlation of -0.096, 0.038, and 0.196,

respectively. The difference between percentage change of sales and percentage change

of employee’s headcount is negatively correlated to CashETR, and positively correlated

to GAAPETR and PBTD with the coefficient of correlation of -0.012, 0.024, and 0.003,

respectively. These correlations suggest that the proxies of CTA are positively correlated

to the proxies of fraud risk.

4. 2 Empirical Results

Table 5 describes the estimates of the relationship between CTA and fraud risk. In

panel A, dependent variable is percentage change of accounts receivable (PCHGREC). In

models 1A to 1C, CTA proxies are NEGCASHETR, NEGGAAPETR and PBTD,

respectively. In panel B, dependent variable is percentage change of inventory

(PCHGINV). In models 1D to 1F, CTA proxies are NEGCASHETR, NEGGAAPETR

and PBTD, respectively. In panel C, dependent variable is discretionary accruals

(DISCACC). In models 1G to 1I, CTA proxies are NEGCASHETR, NEGGAAPETR and

PBTD, respectively. For the remaining tables, I explain the detailed information in the

notes of the tables.

In my regression analysis, I use seven proxies to capture fraud risk from three

perspectives: percentage change of accounts receivable (PCHGREC), percentage change

of inventory (PCHGINV), and discretionary accrual from accrual quality variables

(DISCACC), percentage of cash sales (PCHGCASHSALES) and change of return on

assets (CHGROA) from performance variables, and difference between sales growth rate

and employee growth rate from NFM (DIFFBSE). To interpret the results consistently, I

use the converted effective tax rate to capture CTA in the regression analysis.

Specifically, I multiply effective tax rates by -1 (e.g. NEGCASHETR= - CashETR and

NEGGAAPETR = - GAAPETR) so that the large values of NEGCASHETR,

NEGGAAPETR and PBTD represent higher level of tax avoidance. I control for firm

characteristics, auditors, and audit committee chair in all my models.

In all my models, I find that generally higher levels of tax avoidance are

significantly related to higher fraud risk. In model 1A, where fraud risk is measured by

percentage change in accounts receivable (PCHGREC), the coefficient of

NEGCASHETR indicates that assuming everything else remains constant, a one percent

increase in NEGCASHETR implies a 0.232 percent increase in percentage change in

accounts receivable. This result supports hypothesis 1 such that CTA is positively related

to fraud risk. According to agency theory, a higher level of CTA provides more

opportunities for managers to conduct rent extraction. The coefficient of the natural

logarithm of audit fees (Lnauditfees) describes that a one percent increase in Lnauditfees

suggests a 0.104 decrease in percentage change of accounts receivable. This is consistent

with Brazel et al. (2019) that finds an increase in auditor efforts reduce fraud risk because

auditors understand their client’s business better and can apply more appropriate audit

procedures to alleviate fraud risk. The coefficient of auditor tenure indicates that one

percent increase in auditor tenure implies a 0.001 percent increase in percentage change

of accounts receivable. This is aligned with the results from Brazel et al. (2019) that state

long auditor-client relationships help auditors understand the nature of client’s business

and industry that may affect the risk of business operation and the risk of fraudulent

financial reporting. Auditors may use the knowledge of these risks to determine the

appropriate audit procedures. Therefore, long auditor tenure increases audit quality and

reduces fraud risk. Similar to the coefficient of auditor tenure, the coefficient (-0.007) of

audit committee chair’s tenure suggests that longer audit committee chair’s tenure

decreases fraud risk because audit committee chair with long tenure can better understand

client’s business and internal control so that they can oversee the entire audit engagement.

The coefficient of audit committee chair’s gender indicates that male audit committee

chair increases fraud risk. This is consistent with the literature on gender diversity

because my results support that female chairs are more conservative and better at

monitoring audit engagement.

The coefficient of natural log value of total assets suggests that a one percent

increase in log value of total assets is associated with a 0.043 percent increase in

percentage change in accounts receivable. This result indicates that large companies have

more complex transactions and managers may use those transactions to conduct rent

extraction, thus increasing fraud risk. The coefficient of percentage of sales growth rate

shows that a one percent increase in percentage of sales growth rate is associated with a

0.747 percent increase in percentage change in accounts receivable. This is consistent

with the literature that high growth companies are more likely to commit accounting

fraud. The coefficient for percentage of plant assets and the coefficient for big four CPA

firms do not show significant results. The coefficient of leverage shows an insignificant

relationship between leverage and percentage change in accounts receivable.

In model 1B, the coefficient of NEGGAAPETR suggests that a one percent

increase in NEGGAAPETR is associated with 0.420 percent increase in percentage

change in accounts receivable. All other variables in model 2 have the same signs as

those in model 1 and are statistically significant. In model 1C, the coefficient (0.017) of

permanent book-tax difference (PBTD) indicates that CTA is positively and significantly

related to fraud risk. All other variables have the expected signs and are statistically

significant.

In models 1D to 1F, 1G to 1I, and 1J to 1L, I examine the relationship between

CTA and percentage change in inventory, the relationship between CTA and discretionary

accrual, and the relationship between CTA and return on assets, respectively. These

results are consistent with those in models 1 to 3. I find that CTA is positively related to

fraud risk and other control variables.

In models 1M to 1O, fraud risk is measured by percentage of cash sales. According

to Dechow et al. (2011), percentage of cash sales is negatively related to fraud risk

because managers may use accruals management for accrual-based sales such as credit

sales. The coefficient of NEGCASHETR suggests that a one percent increase in

NEGCASHETR is associated with 0.080 percent increase in percentage of cash sales,

indicating a negative relationship between CTA and fraud risk. One of the explanations is

that some companies front-load earnings and make unusual transitions later, thus

increasing cash sales. Another explanation is that managers may overpay taxes to cover

their fraudulent financial reporting. Therefore, CashETR is negatively related to

percentage change in cash sales.

In models 1P to 1R, I use the difference between sales growth rate and employee

growth rate to measure fraud risk (DIFFBSE). Consistent with Brazel et al. (2009), I

define LargeRisk as an indicator variable equal to 1 if DIFFBSE is greater than 20% and

0 otherwise. In model 16, the coefficient (1.507) of CTA indicates that a higher level of

CTA is more likely to relate to large fraud risk. This result is significant with Z value of

3.85. Specifically, a one percent increase in CTA is associated with 1.507 increase in

log(P/1-P). If log(P/1-P) increases by 1.507. That means that P/(1-P) increases by exp

(1.507) =4.51. This is a 351% increase in the odds of increasing fraud risk (assuming all

other variables remain constant). In models 17 and 18, the coefficients of CTA show a

significant relationship between CTA and LargeRisk as well.

In models 2A to 2F, I examine whether the effect of CTA on fraud risk is greater in

magnitude for fraud companies than for non-fraud companies. The coefficients of the

interaction term, Fraud*CTA, in all models do not show a significantly difference

between fraud companies and non-fraud companies. Specifically, in model 2A, the

proxies for fraud risk and CTA are PCHGREC and NEGCASHETR, respectively. The

coefficient and t-stat for the interaction term are 0.149 and 0.26, respectively. In model

2B, the proxies for fraud risk and CTA are PCHGREC and NEGGAAPETR, respectively.

The coefficient and t-stat for the interaction term are 0.163 and 0.34, respectively. In

model 2C, the proxies for fraud risk and CTA are PCHGREC and PBTD, respectively.

The coefficient and t-stat for the interaction term are 0.19 and 0.51, respectively. In

models 2D to 2F, the proxy for fraud risk is PCHGINV. The coefficient and t-stat show

insignificant relationship between the interaction term and fraud risk as well. Therefore,

my results do not support hypotheses 2. One of the explanations is that some fraud

companies are not more aggressive in tax reporting when they are aggressive in financial

reporting. They may overpay taxes to cover up their fraudulent financial reporting.

Overall, my results support hypothesis 1, indicating that CTA is positively related

to fraud risk. Generally, I find CTA is positively and significantly related to fraud risk

when financial variables are used to measure fraud risk. In addition, my results show a

significant relationship between CTA and fraud risk when I use NFM proxies to measure

fraud risk. Similar to Brazel et al. (2009), I define large fraud risk (LargeRisk) as an

indicator variable equal to 1 and o otherwise if the difference between financial and

nonfinancial performance is greater than 20%. I find that all three CTA proxies are

significantly related to LargeRisk. The results from above are consistent with agency

theory in CTA such that opportunistic managers may use complex CTA strategies to

conduct managerial rent extraction, thus increasing fraud risk. However, I do not find a

significantly different effect of CTA on fraud risk between fraud and non-fraud

companies when I test hypotheses 2. According to the findings at Erickson et al. (2004),

some fraud companies may overpay taxes to cover up their fraudulent financial reporting.

They may worry about the potential penalties from the IRS, reputational damage from the

public, and some other costs associated with aggressive tax avoidance. Therefore, they

are less aggressive in tax reporting when they are more aggressive in financial reporting.

4.3 Supplemental Analysis: Using Fraud risk to Predict Fraud

From the above results, I find that higher level of CTA is related to higher fraud

risk. In this part, I analyze whether fraud risk proxies can be used to predict fraud.

According to the fraud triangle theory, managers in higher fraud risk companies have

more incentives and opportunities to commit accounting fraud. Prior studies find that

fraud companies are associated with higher fraud risk when different proxies are used to

measure fraud risk. Brazel et al. (2009) find that fraud companies have larger difference

between financial and non-financial performance. Dechow et al. (1996) note that fraud

companies are more likely to have a weak corporate governance system. Bell and

Carcello (2000) find that weak internal control system increases the fraud risk. Therefore,

I predict that fraud risk proxies in this study can be used to predict accounting fraud.

To examine whether fraud risk can predict actual fraud, I use financial variables to

measure fraud risk. I select the same number of non-fraud companies based on SIC and

size to match fraud companies. My sample includes 292 fraud companies listed on AAER

website from 2000 to 2017 and 292 non-fraud companies. I collect fraud companies from

the current year that the SEC disclose the violations. I do not consider the pre or post

fraud periods in this study. I use logistic regression to test whether fraud risk is associated

with fraud. My dependent variable is FRAUD, a dummy variable. The value of 1 is for

fraud companies and 0 otherwise. Independent variable is fraud risk. I also include some

control variables that are associated with accounting fraud. For instance, I control for

internal control material effectiveness (ICME). The Sarbanes-Oxley Act section 404 (a)

requires the management to maintain an effective internal control system over financial

reporting. An ineffective internal control provides more opportunities for managers to

commit fraudulent financial reporting. Also, an ineffective internal control system

represents an organization environment that does not emphasize the integrity of financial

reporting. Therefore, I predict a positive relationship between internal control

effectiveness and accounting fraud. Also, I control for financial restatement (Restate) as

an indicator variable equal to 1 if a financial restatement has been reported during the last

three years and 0 otherwise. I predict a positive coefficient for financial restatement.

Next, I control for sales growth rate (PCHGSALES), which is defined as the difference

between current year’s sales and previous year’s sales divided by previous year’s sales.

Since companies with rapid sales growth are more aggressive in earnings management, I

predict a positive coefficient for sales growth rate.

In model 3A, I examine the relationship between discretionary accrual and

fraud. The coefficient of percentage change in accounts receivable (PCHGREC) suggests

that a one percent increase in PCHGREC is associated with 15.90 increase in odds ratio,

indicating a very strong association between PCHGREC and accounting fraud. In model

3B, I examine the relationship between discretionary accruals (DISCACC) and fraud. The

coefficient of DISCACC indicates that a one percent increase in DISCACC is associated

with 8.82 increase in odds ratio. In model 3C, I examine the relationship between

percentage change of cash sales (PCHGCASHSALES) and fraud. The coefficient shows

that one percent of increase in PCHGCASHSALES is associated with

0.063 decrease in odds ratio, indicating a negative association between

PCHGCASHSALES and fraud. In model 3D, I examine the relationship between

percentage change of soft assets (PCHGSOFTASSET) and fraud. The coefficient shows

that one percent of increase in PCHGSOFTASSET is associated with 1.53 increase in

odds ratio. Therefore, fraud risk can successfully predict accounting fraud. The results

indicate that higher level of tax avoidance may increase regulators and external auditors’

attention since higher level of tax avoidance could be considered a red flag for accounting

fraud.

Chapter 5: Discussion

Prior studies provide mixed evidence on whether tax avoidance is positively or

negatively related to aggressive financial reporting. Some studies state that CTA is

positively related to aggressive financial reporting because managers may use the areas of

nonconformity between tax reporting and financial reporting to increase book income and

decrease tax income simultaneously. Other studies argue that managers could not control

book income and tax income in the opposite directions without being perceived by

external auditors and the IRS. In addition, some study finds that managers may

intentionally overpay taxes to cover up their aggressive financial reporting. Therefore,

there is a negative relationship between CTA and aggressive financial reporting. In this

study, I extend the research on CTA by exploring the relationship between CTA and fraud

risk because very small number of fraud companies are disclosed by the SEC on AAER’s

website and fraud risk is easier to measure for all companies. Also, I examine the

relationship between fraud risk and fraud to see if fraud risk can be used to predict fraud.

This study bridges the gap between CTA and accounting fraud through fraud risk. I

measure fraud risk from different perspectives: accrual quality, financial performance,

and NFMs.

The relationship between CTA and fraud risk

In hypothesis 1, I predict that there is a positive relationship between CTA and

fraud risk. According to the agency theory and fraud triangle concept, opportunistic

managers may use complex tax avoidance strategies to conduct rent extraction, thus

increasing fraud risk. I use different proxies to measure fraud risk. Dechow et al. (2011)

note that actual accruals are more powerful than discretionary accrual in predicting

material misstatement. Therefore, I use both the abnormal accrual and the actual accruals

to test the relationship between CTA and fraud risk. My results from discretionary

accruals are consistent with those from actual accruals. Discretionary accruals are

positively and significantly related to PBTD and negatively and significantly related to

GAAPETR and CashETR, indicating that CTA is positively related to fraud risk. In

addition, CTA is positively and significantly related to percentage change of accounts

receivable and percentage change of inventory. Furthermore, CTA is also positively and

significantly related to performance variables (e.g. percentage change of cash sales and

return on assets).

In addition, I use the difference between financial and non-financial performance

to measure fraud risk. Similar to Brazel et al. (2009), I define large fraud risk (LargeRisk)

as an indicator variable equal to 1 and 0 otherwise if the difference between financial and

non-financial performance is greater than 20%. I find that NECASHETR,

NEGGAAPETR, and PBTD are significantly related to LargeRisk. My results show a

significant relationship between CTA and fraud risk when I use NFM proxies to measure

fraud risk.

Fraud and non-fraud companies

In hypothesis 2, I predict that for both fraud and non-fraud companies, CTA is

positively related to fraud risk. However, I predict that the effect of CTA on fraud risk for

fraud companies will be greater in magnitude than for non-fraud companies because

according to agency theory and fraud triangle theory, fraud companies have more

incentives and opportunities to manipulate financial statements. I use FRAUD as an

indicator variable to separate fraud and non-fraud companies. I use the interaction term,

CTA*Fraud, to catch the effect of CTA on fraud risk for fraud companies and non-fraud

companies. The coefficients of the interaction term in all models do not show a

significant difference between fraud and non-fraud companies. My results do not support

hypothesis 2 such that the effect of CTA on fraud risk is greater in magnitude for fraud

companies than for non-fraud companies. According to the fraud triangle theory and

agency theory, fraud companies should have more incentives and opportunities to use

complex tax planning strategies to commit accounting fraud. However, some fraud

companies may overpay taxes to cover up their fraudulent financial reporting. The

positive relationship between CTA and fraud risk may be offset by the intentionally

overpaid tax. My findings in hypotheses 2 suggest that agency theory and intentionally

overpaid taxes both exist in corporate tax avoidance.

Implications for research

This study has several contributions to the literature exploring CTA, fraud risk,

and accounting fraud. First, while prior studies provide mixed evidence on whether CTA

is positively or negatively associated with aggressive financial reporting, this study

examines the relationship between CTA and fraud risk and the relationship between fraud

risk and accounting fraud. To my knowledge, this study is the first one to examine the

relationship between CTA and fraud risk. While few accounting frauds are disclosed by

the SEC, fraud risk can be assessed for all companies. Also, fraud risk can be used to

predict accounting fraud. Therefore, this study bridges the gap between CTA and

accounting fraud through fraud risk. Second, while prior studies use financial data to

assess aggressive financial reporting, this study evaluates fraud risk by using accrual

quality, performance variables, and NFMs, which provide different perspectives for

auditors and regulators to assess the effect of CTA on fraud risk. Third, this study

contributes to the literature on agency problems related to CTA. Little is known about

whether the relationship between CTA and fraud risk is different for fraud and non-fraud

companies. Agency theory and fraud triangle theory indicate that fraud companies have

more incentives and opportunities to commit fraud. However, managers may overpay

taxes to conceal fraudulent financial reporting. Therefore, higher level of CTA should be

considered a warning light for fraudulent financial reporting.

Implications for practice

This study is also meaningful to practitioners. My findings suggest that regulators

and external auditors should assess fraud risk from different perspectives. Higher fraud

risk can be used to predict accounting fraud. In addition, higher levels of CTA should

draw more attention from regulators and external auditors since a high level of CTA is

positively related to high fraud risk, which could be an indicator for actual accounting

fraud.

Limitations

Even though this study provides some insight into the relationships among CTA,

fraud risk, and accounting fraud to researchers and practitioners, it still has several

limitations. First, CTA and fraud risk are measured by various proxies, which may not

catch all of the features of CTA and fraud risk. Also, CTA is defined as a continuum of

tax avoidance strategies. It is difficult to separate CTA strategies into different levels.

Future study should examine the relationship between the specific tax avoidance

strategies such as tax shelter and the accounting fraud. Second, this study investigates the

relationship between CTA and fraud risk for publicly traded companies. The results may

not be generalized to private companies since private companies may have different

considerations when they conduct CTA. Third, CashETR, GAAPETR and PBTD are used

to measure the non-conformity between book and tax income. Future research should

consider different proxies to measure the conformity between book and tax income for

tax planning strategies. Third, only two NFM variables (number of employees and order

backlog) are collected from Compustat. Many companies do not disclose their NFM

information. The two NFM variables may not be able to measure fraud risk for all

companies. Different industries have different NFMs. Future studies should use more

NFM variables from each industry to measure fraud risk. Finally, the sample size for

fraud companies and non-fraud companies are imbalanced. I match non-fraud companies

to fraud companies based on SIC and size. Future studies may use some other techniques

to solve the issues of the data imbalance.

General conclusions

While prior studies provide mixed results on whether CTA is positively or

negatively related to aggressive financial reporting, I extend the literature on the

consequences of CTA by examining the relationship between CTA and fraud risk. Using

different proxies for CTA and fraud risk, I find that CTA is positively related to fraud risk.

However, I do not find a significantly different effect of CTA on fraud risk between fraud

and non-fraud companies. Fraud companies may be less aggressive in tax reporting when

they are more aggressive in financial reporting. Fraud companies may overpay taxes to

conceal their fraudulent financial reporting. Furthermore, my results indicate that fraud

risk proxies can be used to predict actual accounting fraud. Therefore, higher level of

CTA could be considered a red flag for fraudulent financial reporting.