The effect of debt financing pressure on earnings management

(1)

The effect of debt financing pressure

on earnings management

Erwin Koning

S2506777

24-06-2019

(2)

1

Abstract

The current study examines whether debt financing pressure has a positive effect on accrual-based earnings management, real earnings management and total earnings management. Debt financing pressure is measured using three proxies: changes in short-term debt, long-term debt issues, and the frequency of issuing long-term debt. This study argues that these proxies are incentives for managers to inflate earnings. Significant positive effects indicate that debt financing pressure has a positive effect on earnings management. Robustness tests support the results. Additional analysis provide evidence for the reversal effect of discretionary accruals. These results fill a niche in scientific knowledge as the effect of debt financing pressure on real earnings management and total earnings management and the effect of a lower frequency of issuing debt on earnings management are not addressed in existing literature. Thereby, the current results provide guidance for future studies.

JEL classification

G32 • M41

Keywords

Earnings management • debt capital • financing pressure • debt issue frequency

Author: E.C.Y. Koning Supervisor: dr. C.A. Huijgen

University: University of Groningen

Master: Accountancy

(3)

2

1. Introduction

McKinsey Global Institute (2018) states that many observers expected the global debt levels to decline following the financial crisis of 2008. However, growing from $37 trillion in 2007 to a total of $66 trillion in 2017, the corporate worldwide debt has almost doubled during the last decade. Furthermore, the Pacific Investment Management Company (PIMCO, 2018) states that U.S. companies are becoming more heavily indebted, which is illustrated by an increase of the net leverage ratio from 1.4 in 2007 to 2.4 in 2017. Additionally, a growing percentage of the debt is rated BBB or lower, to a percentage of 48 in 2017. This suggests that during the previous decade, debt financing has become increasingly important for companies.

Therefore, PIMCO (2018) states that companies and investors should be aware of the higher debt levels and the decreasing quality of debts. Particularly since the Federal Reserve in the U.S. started raising the interest rates four times in 2018. In line with the quality of new debts, Caton et al., (2011) and Liu et al., (2010) state that companies use higher levels of accrual-based earnings management prior to debt issues. Furthermore, Fields et al. (2018) and Fung and Goodwin (2013) argue that debt financing pressure influences the manager’s behaviour, resulting in higher discretionary accruals for instance.

Following Generally Accepted Accounting Principles (GAAP), companies can exercise some discretion over accrual choices to a desired level. In addition to these discretionary accruals, surveys have shown that real earnings management is practiced by CFOs to increase the valuation of their company (Graham et al., 2005). Zang (2011) defines real earnings management as an intentional action to change the structuring or timing of a transaction to affect earnings in a specific direction. Some suggest that the Sarbanes-Oxley Act (SOx) in 2002 has caused an increase of real earnings management and decrease of accrual-based earnings management in the years following the implementation of the SOx (Cohen et al., 2008 and Zang, 2011). In literature, knowledge on the effect of debt financing pressure on real earnings management is forthwith absent. To fill this niche, the effect of debt financing pressure on both real earnings management and accrual-based earnings management needs to be analysed. To evaluate the overall effect of earnings management incentives, the total earnings management is measured in line with the work of Franz et al. (2014). The reversal effect of discretionary accruals is assessed to create a more complete view of the use of accrual-based earnings management.

(5)

4 The modified jones-model Jones (1991) is used for measuring the discretionary accruals and the three empirical proxies (sales, production, and discretionary expenditures manipulation) by Roychowdhury (2006) is used for measuring real earnings management. Total earnings management is defined as the sum of accrual-based earnings management and real earnings management (Franz et al., 2014).

1.1 Research gap and scientific contributions

Prior research has focussed on the effects of financial events like initial public offerings and seasoned equity offerings on earnings management (Chang and Lin, 2018; Rangan, 1998; Shu and Chiang 2014; Teoh et al., 1998a, 1998b). Furthermore, the use of earnings management surrounding companies in technical default or close to their debt covenants has been studied as well (Defond and Jiambalvo, 1994; Dichev and Skinner, 2002; Franz et al., 2014; Sweeney, 1994).

Several authors (Caton et al., 2011; Fields et al., 2018; Fung and Goodwin, 2013; Liu et al., 2010) discuss the use of accrual-based earnings management prior to the issues of debts. Fields et al. (2018) focussed on debt refinancing pressure, in particular the effect of both changes in short-term debt (debt coming due within one year) and subsequent debt financing on discretionary accruals. Fung and Goodwin (2013) focussed on short-term debt and the effect on discretionary accruals, relating the effect to the financial distress theory. The current study aims to expand the work of Fields et al. (2018) and Fung and Goodwin (2013) by determining the effect of debt financing pressure on earnings management. Both researchers exclusively studied the effect on accrual-based earnings management. It seems that the question of whether debt financing pressure leads to real earnings management or total earnings management cannot be answered with existing literature. The amount of debt financing of companies differs for each company from year to year due to the refinancing of maturing debt and additional capital needs. Furthermore, it appears there is no evidence on a possible relation between long-term debt issues and real earnings management and total earnings management in the previous year. In addition, companies with a lower debt financing frequency face higher debt financing pressure when issuing debt, compared to companies issuing long-term debt every year. The effect of the frequency of issuing long-term debt on the level of earnings management remains unclear as well.

The purpose of this study is to examine whether managerial behavior is influenced by debt financing pressure. To achieve this, debt financing pressure is measured in three ways: changes in short-term debt (debt coming due within one year), long-term debt issues, and the frequency

(6)

5 of issuing long-term debt (companies not issuing debt every year). Earnings management is measured in three ways as well: accrual-based earnings management (discretionary accruals), real earnings management and total earnings management. The current study focusses on nonfinancial companies in the U.S. in the years 2005 to 2017, after the Sarbanes-Oxley Act became effective in 2002.

This study will contribute to the scientific knowledge on this topic, as it will pioneer in determining the effect of debt financing pressure on real earnings management and total earnings management. It will expand to existing studies that focused on accrual-based earnings management alone (Fields et al., 2018 and Fung and Goodwin, 2013) and provide clarity as three earnings management methods are used instead of one. Moreover, the effect of the frequency of issuing long-term debt on the earnings management level has not yet been clarified.

1.2 Research question

Combining the growing corporate debt market (McKinsey Global Institute, 2018), need for awareness of decreasing debt quality (PIMCO, 2018) and lack of knowledge on consequences of debt financing pressure on managerial behaviour (i.e. use of earnings management), a need for further research becomes clear. Therefore, the following research question is formulated: ‘What is the effect of debt financing pressure on earnings management?’

The current study proceeds as follows. In the next section, section 2, prior literature will be discussed which will support the hypotheses. In section 3, the methodology will be described. The results will be presented in section 4. In the final section, the findings will be discussed, and the conclusions of this study will be stated.

(7)

6

2. Theoretical framework

In the current section, the theoretical background of earnings management is discussed. First, the definitions of the different types of earnings management are explained. Second, agency theory and the lemons problem are discussed. Third, survey evidence is reviewed. Fourth, the consequences of potential earnings management are discussed. Fifth, the debt covenant hypothesis is related to earnings management. Thereafter, the use of earnings management prior to equity offerings and debt issues is discussed. Finally, the hypotheses are developed in this section.

2.1 Definitions earnings management and debt financing pressure

Earnings management

Hepworth (1953) was one of the first to find an income increasing and decreasing effect due to accounting policy choices. Hepworth observed this as income smoothing and suggested that companies manage earnings because investors perceived variability in earnings as an indicator of firm-risk. More recently, Healy and Wahlen (1999) defined earnings management as managerial decisions or judgments in financial reporting that adjust financial reports to influence contractual outcomes or to mislead some stakeholders about the underlying financial performance. The Securities and Exchange Commission (SEC) defines earnings management as “the practice of distorting the true financial performance of the company”. Walker (2013) uses a broader definition of earnings management: "the use of managerial discretion over (within GAAP) accounting choices, earnings reporting choices, and real economic decisions to influence how underlying economic events are reflected in one or more measures of earnings." (Walker, 2013, page. 446). As described by Walker, there are two ways to manage earnings: accrual-based earnings management and real earnings management. In line with the study of Franz et al. (2014), total earnings management is used in the current study to evaluate the overall earnings management.

The Sarbanes-Oxley Act in 2002 resulted in relatively higher real earnings management and less accrual-based earnings management in the years following the implementation (Cohen et al., 2008; Cohen and Zarowin, 2010; Zang, 2011). The findings of Cohen et al. (2008) state that the passage of the Sarbanes-Oxley Act has motivated managers to switch from accrual-based earnings management to real earnings management. Consistent with Cohen et al. (2008), Cohen and Zarowin (2010) find a change from the use of accrual-based earnings management to real earnings management and argue that the passage of Sarbanes-Oxley Act has made accrual-based earnings management more costly than real earnings management. Zang (2011) finds that

(8)

7 accrual-based earnings management and real earnings management substitute for each other. However, both methods are used to achieve manage earnings, increasing the relevance of measuring total earnings management in further research (Franz et al., 2014).

Accrual-based earnings management

Managers have discretion in reporting earnings as written in the GAAP. Managers have some discretion to shift reported income between current and future periods by temporarily boosting or reducing accruals (Bergstresser and Philippon, 2006). They define accruals as: “components of earnings that are not reflected in current cash flows, and a great deal of managerial discretion goes into their construction” (p. 512).

Real earnings management

Unlike accrual-based earnings management, real earnings management has direct consequences on current and future cash flows. Roychowdhury (2006) defines real earnings management as “management actions that deviate from normal business practices, undertaken with the primary objective of meeting certain earnings thresholds” (p. 336). Roychowdhury (2006) argues there are three main types of real earnings management. First, managers can manipulate sales by accelerating the timing of sales or generating additional unsustainable sales through more tolerant credit terms or excessive sales discounts. Second, managers can create overproduction by increasing production in order to report lower cost of goods sold. Third, managers can cut discretionary expenses in the current period like selling, administrative and general costs. Total earnings management

In line with Franz et al. (2014), total earnings management is defined as the sum of accrual-based earnings management and real earnings management.

Debt financing pressure

The proxies used by Fields et al. (2018) for debt refinancing pressure are short-term debt coming due and subsequently if new debt is obtained. Like Fields et al. (2018), Fung and Goodwin (2013) use debt coming due as a proxy for financing pressure. In line with Fields et al. (2018) and Fung and Goodwin (2013), the current study uses three proxies for debt financing pressure: (1) the amount of debt coming due within one year, (2) the amount of long-term debt issues (3) companies with a lower frequency of issuing long-term debt. The third proxy for debt financing pressure is a result of reasoning. Companies can issue long-term debt every year or once every few years. If companies do not issue long-term debt every year, they have a lower

(9)

8 debt financing frequency. This results in higher debt financing pressure when issuing long-term debt compared to the companies issuing debt more frequently.

2.2 Agency theory and the lemons problem

Agency theory is closely tied to earnings management (Palliam and Shalhoub, 2003). They argue that earnings management can be the result of information asymmetry that arises between capital providers and managers, since capital providers base their decisions on information provided by earnings announcements. Riahi-Belkaoui (1999) state that earnings management is an effort of managers to satisfy the consensus of earnings estimates and to project a smooth earnings path. Watts and Zimmerman (1990) argue that certain discretion over accounting policies by the management is accepted by capital providers. Regarding debt financing, earnings management is relevant considering it as a source of potential conflict and information asymmetry between debt capital providers and the management.

The ‘lemons problem’ of Akerlof (1970) describes that the value of an investment is due to asymmetric information between the buyer and the seller of fresh capital. This can result in an incentive for managers to manage earnings, in order to look attractive to debt capital providers. When earnings are inflated, the financial report is due to information asymmetry. For debt capital providers, it can be difficult to distinguish good quality debt issues from bad quality debt issues.

2.3 Survey evidence

From a managerial view, survey evidence shows that managers believe earnings management can increase the value of their company which motivates them to inflate earnings in current periods when they believe it is necessary (Graham et al., 2005). Managers do so in order to build credibility with the capital markets and to reveal future growth prospects. They also revealed in their survey that managers are willing to manage earnings through real actions (real earnings management). The managers admitted that they would delay maintenance or advertising. Also, positive net present value projects would be given up to meet short-term earnings benchmarks. From a debt financing perspective, Graham et al. argue that managers feel they are making the right choice when sacrificing economic value to achieve a specific earnings target, because a negative earnings surprise can be costly for the company. De Jong et al. (2014) surveyed financial analysts the same way as Graham et al. (2005) surveyed investors in order to compare the results. They argue that analysts are unable to unravel certain earnings management of CFOs which might perhaps explain the CFOs earnings management preferences. Kim and Sohn (2013) argue that investors rely on reported earnings when

(10)

9 estimating the future earnings of the company. Additionally, previous literature shows that investors underestimate the reversal effect of discretionary accruals, which results in overestimating the future performance of the company giving managers an incentive to inflate current earnings (DeFond and Park, 2001; Sloan, 1996; Xie, 2001).

2.4 Consequences of potential earnings management

Existing literature concerning the consequences of earnings management on the cost of debt argues that companies using higher levels of earnings management, both accrual-based and real earnings management, show higher cost of debt (Bharath et al., 2008; Francis et al., 2005; Gray et al., 2010; Kim et al., 2018; Prevost et al., 2008; Shen et al., 2013). They predominantly argue that earnings management results in an information risk for investors, which leads to a risk premium. These results could motivate managers to use less earnings management in general. However, they did not include the timing of the earnings management, for example, if managers used earnings management specifically for debt financing reasons.

Regarding the effect of earnings management due to debt financing pressure, Caton et al. (2011), Liu et al. (2010) and Crabtree et al. (2014) investigated the effect of earnings management prior to debt issues on the cost of debt and found contradictory results. Liu et al. (2010) find that the cost of the debt is lower when earnings are inflated prior to the debt issue. Additionally, Demirtas and Cornaggia (2006) show that companies inflate earnings prior to obtaining a credit rating on issued debt. This leads to higher credit ratings, implying a lower cost of debt. Caton et al. (2011) and Crabtree et al. (2014) find opposite results and state that the cost of debt is higher if earnings are inflated prior to debt issues. The results of Demirtas and Cornaggia (2006) and Liu et al. (2010) are inconsistent with the literature arguing that higher levels of accrual-based earnings management lead to a higher cost of debt (Bharath et al., 2008; Francis et al., 2005; Gray et al., 2010; Kim et al., 2018; Prevost et al., 2008; Shen et al., 2013). Crabtree et al. (2014) is the only research assessing the effect of real earnings management prior to debt issues on the cost of debt. They focused only on the effects of earnings management and do not measure if earnings are actually inflated prior to the debt issue. In summary, it is unclear what the consequences of earnings management are on the cost of debt due to debt financing pressure (Caton et al., 2011; Crabtree et al., 2014; Demirtas and Cornaggia, 2006; Liu et al., 2010). Noteworthy is that the results of the current research may be relevant for future research implications, regarding the consequences of the use of earnings management due to debt financing pressure.

(11)

10

2.5 Debt covenant hypothesis

The debt covenant hypothesis argues that the closer a company is to violating its debt covenants, the more likely managers will use earnings management to prevent a technical default (Dichev and Skinner, 2002; Franz. et al., 2014; Watts and Zimmerman, 1990). Dichev and Skinner (2002) find large sample evidence of the debt covenant hypothesis using accrual-based earnings management. More recently, Franz et al. (2014) also assess the debt covenant hypothesis using three ways of measuring earnings management: accrual-based earnings management, real earnings management, and total earnings management. They find support for the debt covenant hypothesis using all three measures. Additionally, Franz et al. suggest that managers manage their earnings upwards to improve their bargaining power in debt renegotiation. From a debt financing pressure perspective and in line with Franz et al., the debt covenant hypothesis suggests higher levels of earnings management when managers are under debt financing pressure.

Financial distress is related to the debt covenant hypothesis, since companies that are close to their debt covenants may be in a situation of financial distress. Fung and Goodwin (2013) argue that debt financing pressure and the effect on earnings management is consistent with the rationale of financial distress. Gorden (1971) defines financial distress as a situation where a company creates a probability that it may not be able to pay interest or repay its debt. In a situation of financial distress, the debt of the company will be sold with higher yields to maturity in comparison to other companies, because investors ask for a risk premium. Fung and Goodwin (2013) argue that accrual-based earnings management can be benchmarked against the financial distress rationale because debt financing pressure leads to discretionary accruals.

2.6 Earnings management prior to equity offerings

Prior literature reveals that companies tend to manage earnings upward prior to initial public offerings of equity (Teoh et al., 1998a), because companies aim to influence the perception of investors. Companies also manage earnings upward before seasoned equity offerings (Chang and Lin, 2018; Rangan, 1998; Shu and Chiang 2014; Teoh et al., 1998b). For companies engaging in seasoned equity offerings, earnings decline significantly in the years following (Loughran and Ritter, 1995). Chang and Lin (2018) studied both accrual-based earnings management and real earnings management and found higher levels of earnings management prior to seasoned equity offerings for both methods of earnings management. These findings reporting positive earnings management because of equity financing pressure suggest that earnings may also be managed upward prior to debt financing pressure.

(12)

11

2.7 Earnings management prior to debt offerings

The use of accrual-based earnings management prior to debt issues has been studied by several researchers (Caton et al., 2011; Crabtree et al., 2014; Demirtas and Cornaggia, 2006; Liu et al., 2010). They all focused on discretionary accruals and report that managers inflate earnings prior to a debt issue due to debt financing pressure. Caton et al. (2011) used data between 1995 and 2005. Liu et al. (2010) conducted similar research using data from 1970 to 2004. The results of Caton et al. (2011) and Liu et al. (2010) suggest that debt financing pressure leads to higher accrual-based earnings management.

The studies of Fields et al. (2018) and Fung and Goodwin (2013) are closer to the current study. Fung and Goodwin (2013) investigated the effect of maturing debt on accrual-based earnings management and found a positive effect (i.e. higher debt financing pressure leads to higher levels of discretionary accruals). They relate the debt financing pressure to the financial distress theory. Fields et al. (2018) focussed specifically on debt refinancing pressure, analysing the effect of both changes in short-term debt (debt coming due within one year), like Fung and Goodwin (2013), and subsequent debt financing (if a new debt was obtained) on accrual-based earnings management. They found that earnings are inflated due to the positive changes in short-term debt and that this effect is stronger if new debt is obtained subsequently. They state that refinancing pressure leads to higher discretionary accruals and base their research on the debt covenant hypothesis (Watts and Zimmerman, 1990) and research by Roberts and Sufi (2009) who state that companies are inclined to use earnings management when long-term debt is coming due. Roberts and Sufi (2009) argue that managers believe there could be negative consequences if they need to finance new debts and they appear to be unattractive, resulting in higher levels of earnings management.

2.8 Hypotheses

Based on the literature in the prior section, it is expected that debt financing pressure has a positive effect on earnings management. Because of different proxies for debt financing pressure and using multiple methods of measuring earnings management, the hypotheses are separated for each proxy of debt financing pressure.

Regarding accrual-based earnings management, there has been prior research concerning debt financing pressure and the effect on the level of discretionary accruals, finding a positive effect (Caton et al., 2011; Crabtree et al., 2014; Demirtas and Cornaggia, 2006; Liu et al., 2010). It seems that no previous literature has studied the effect of debt financing pressure on real earnings management and total earnings management. The literature in the current section

(13)

12 suggests a positive effect between debt financing pressure and real earnings management and total earnings management.

Change in debt in current liabilities

Fields et al. (2018) analysed the effect of changes in short-term debt (debt coming due within one year) on discretionary accruals and finds a positive effect. Based on these studies and the literature described in this section, a positive effect is expected:

H1a: A change in debt in current liabilities is positively associated with accrual-based earnings management.

H1b: A change in debt in current liabilities is positively associated with real earnings management.

H1c: A change in debt in current liabilities is positively associated with total earnings management.

Long-term debt issues

Caton et al. (2011) and Liu et al. (2010) state that managers use accrual-based earnings management prior to debt issues. Fields et al. (2018) argue that earnings are inflated due to positive changes in short-term debt and that this effect is stronger if new debt is subsequently obtained. In line with the literature, a positive effect is expected:

H2a: Long-term debt issues are positively associated with accrual-based earnings management in the prior year.

H2b: Long-term debt issues are positively associated with real earnings management in the prior year.

H2c: Long-term debt issues are positively associated with total earnings management in the prior year.

(14)

13 Frequency of issuing long-term debt

Based on the literature in the current section, it is expected that companies use higher levels of earnings management when they have a lower frequency of issuing long-term debt. Therefore, a positive effect is expected:

H3a: A lower frequency of issuing debt is positively associated with accrual-based earnings management in the prior year.

H3b: A lower frequency of issuing debt is positively associated with real earnings management in the prior year.

H3c: A lower frequency of issuing debt is positively associated with total earnings management in the prior year.

(15)

14

3. Methodology

This section provides an overview of the methodology by describing the sample and explaining the dependent and independent variables. Additionally, the control variables are identified and models for testing the hypotheses are deployed.

3.1 Sample

The initial data for the current study is obtained from Compustat North America over the years from 2003 to 2017. Data from the years 2003 and 2004 are necessary to measure earnings management in the year 2005. Non-US companies, financial industries (SIC codes 6000–6999) and regulated industries (SIC codes 4400-4499) are deleted from the sample in line with previous research (Fields, 2018; Fung and Goodwin, 2013; Roychowdhury, 2006). Companies with missing values and small companies with revenues or total assets below one million dollars are eliminated. Including the years 2003 and 2004, the sample for measuring earnings management consist of 56,533 company years (8,144 unique firms) and 53 industries, based on the two-digit industry SIC numbers. Companies with missing or zero long-term debt are deleted for the final sample because the current study focuses on debt financing pressure. The final sample over the years from 2005 to 2017 consists of 34,107 company years (5,941 unique firms) in 53 industries, based on the two-digit industry SIC numbers. The variables are winsorized at the 1 and 99 percent level.

3.2 Dependent variable: earnings management

Earnings management is measured using three methods: accrual-based earnings management, real earnings management, and total earnings management. Normal values of earning management are used in the current study, resulting in both positive and negative earnings management.

Accrual-based earnings management

The discretionary accruals are measured using the Modified Jones Model. Jones (1991) developed this model and a few years later Dechow et al. (1995) added some modifications. In

Table 1. Sample composition Company years Unique companies Number of industries Years

Sample earnings management 56,533 8,144 53 2003-2017

(16)

15 the Modified Jones Model, the discretionary accruals are calculated by measuring the non-discretionary accruals as a portion of the total accruals. To estimate the non-discretionary accruals, A cross-sectional regression model is used for every industry and year.

The Modified Jones Model:

𝑇𝐴𝐴𝐶_𝑡 𝐴_𝑡−1

= 𝛼

1 1 𝐴_𝑡−1

+ 𝛼

2

(

∆𝑅𝐸𝑉_𝑡 𝐴_𝑡−1

−

∆𝑅𝐸𝐶_𝑡 𝐴_𝑡−1

) + 𝛼

3 𝑃𝑃𝐸_𝑡 𝐴_𝑡−1

+ 𝜀

𝑡 (1) Where:

𝐴_𝑡−1= total assets (Compustat #6) in year t-1

𝛥𝑅𝐸𝑉𝑡 = change in revenue (Compustat #12) in year t

𝛥𝑅𝐸𝐶_𝑡 = change in receivables (Compustat #2) in year t

𝑃𝑃𝐸𝑡 = gross property, plant and equipment (Compustat #7) in year t

𝛼₁, 𝛼₂ and 𝛼₃ = industry-year specific parameters, following from the regression

𝜀_𝑡 = error (discretionary accrual)

The parameters for measuring the discretionary accruals are estimated for every year and industry. The current study has 53 industries (two-digit SIC numbers) and measures the discretionary accruals for the years 2005 to 2017. The total accruals (TACC) are calculated as follow:

𝑇𝐴𝐶𝐶_𝑡= 𝛥𝐶𝐴_𝑡− 𝛥𝐶𝑎𝑠ℎ_𝑡 − 𝛥𝐶𝐿_𝑡 + 𝛥𝐷𝐶𝐿_𝑡 – 𝐷𝐸𝑃_𝑡 (2)

Where:

𝑇𝐴𝐶𝐶𝑡 = Total accruals in year t

𝛥𝐶𝐴_𝑡 = Change in current assets (Compustat #4) in year t

𝛥𝐶𝑎𝑠ℎ_𝑡 = Change in cash and cash equivalents (Compustat #162) in year t

𝛥𝐶𝐿𝑡 = Change in current liabilities (Compustat #5) in year t

𝛥𝐷𝐶𝐿_𝑡 = Change in debt included in current liabilities (Compustat #34) in year t

𝐷𝐸𝑃_𝑡 = Depreciation and amortization (Compustat #14) in year t

The error is the residual in the Modified Jones Model. Consequently, the remainder of the equation must be calculated first. The non-discretionary accruals (NDACC) are measured for each industry-year group:

𝑁𝐷𝐴𝐶𝐶_𝑡 𝐴_𝑡−1

= 𝛼

1 1 𝐴_𝑡−1

+ 𝛼

2

(

∆𝑅𝐸𝑉_𝑡 𝐴_𝑡−1

−

∆𝑅𝐸𝐶_𝑡 𝐴_𝑡−1

) + 𝛼

3 𝑃𝑃𝐸_𝑡 𝐴_𝑡−1

(3)

(17)

16 The discretionary accrual (DACC) is calculated by subtracting the NDACC of the TACC.

𝐷𝐴𝐶𝐶_𝑡 𝐴_𝑡−1

=

𝑇𝐴𝐶𝐶_𝑡 𝐴_𝑡−1

−

𝑁𝐷𝐴𝐶𝐶_𝑡 𝐴_𝑡−1

(4)

Real earnings management

Three types of real earnings management are calculated using the model developed by Roychowdhury (2006): (1) offering excessive sales discounts or soft credit terms to temporarily boost sales in the current period (i.e. abnormal operating cash flows (ACFO)), (2) creating overproduction to report a lower cost of goods sold in the current period (i.e. abnormal production costs (APROD)), and (3) reducing discretionary expenditures in the current period (i.e.. abnormal discretionary expenses (ADISEXP)). To estimate each model of real earnings management, a cross-sectional regression for every industry and year is used. In line with the measure of the discretionary accruals by the Modified Jones Model, 53 industries are identified (two-digit SIC numbers). In every model, the “abnormal” part (i.e. error) of the CFO, PROD and DISEXP are measured by calculating the actual CFO, PROD and DISEXP and then subtracting the “normal” part. The models of Roychowdhury (2006) for calculating each type of real earnings management are as follow:

𝐶𝐹𝑂_𝑡 𝐴_𝑡−1

= 𝛼

1 1 𝐴_𝑡−1

+ 𝛼

2 𝑅𝐸𝑉_𝑡 𝐴_𝑡−1

+ 𝛼

3 ∆𝑅𝐸𝑉_𝑡 𝐴_𝑡−1

+ 𝜀

𝑡

(5)

𝑃𝑅𝑂𝐷_𝑡 𝐴_𝑡−1

= 𝛼

1 1 𝐴_𝑡−1

+ 𝛼

3 ∆𝑅𝐸𝑉_𝑡 𝐴_𝑡−1

+ 𝛼

4 ∆𝑅𝐸𝑉_𝑡−1 𝐴_𝑡−1

+ 𝜀

𝑡

(6)

𝐷𝐼𝑆𝐸𝑋𝑃_𝑡 𝐴_𝑡−1

= 𝛼

1 1 𝐴_𝑡−1

+ 𝛼

+ 𝜀

𝑡

(7)

Where:

𝐶𝐹𝑂_𝑡= Operating cash flows (Compustat #308) in year t

𝑃𝑅𝑂𝐷_𝑡 = Production costs in year t, production costs are the sum of the costs of goods sold (Compustat #41) and the change in inventory (Compustat #3).

𝐷𝐼𝑆𝐸𝑋𝑃_𝑡 = Discretionary expenses in year t, discretionary expenses are the sum of research

and development costs (Compustat #46), advertising expenses (Compustat #45) and selling, general and administrative expenses (Compustat #189). In line with Roychowdhury (2006), advertising expenses and research and development costs are set to zero if missing. Companies

(18)

17 are not required to report their advertising expenses and if it is not separately reported, advertising expenses are included in selling, general and administrative expenses in Compustat. 𝑅𝐸𝑉_𝑡 = Revenue in year t

Δ𝑅𝐸𝑉_𝑡 = Change in revenue in year t Δ𝑅𝐸𝑉𝑡−1= Change in revenue in year t -1

𝛼1, 𝛼2, 𝛼3 and 𝛼4 = industry-year specific parameters, following from the regression

𝜀_𝑡 = error (ACFO, APROD, and ADISEXP)

Given the “normal” sales level and cash flow from operations, the error of ACFO is negative if sales are unusually high. The error of ADISEXP also results in a negative value if a company shows unusually low discretionary expenses, in comparison with its sales level. ACFO and ADISEXP are calculated by multiplying the error by a negative value of one so that higher values indicate higher income-increasing earnings management. The error of APROD is positive if real earnings management is used because the error represents an unusually high production level.

In line with prior studies (Cohen and Zarowin, 2010; Kim, 2018; Zang 2011), three proxies are used for real earnings management. REM1 is the average of ACFO and ADISEXP, REM2 is the average of APROD and ADISEXP and REM3 is the average ACFO, APROD, and ADISEXP. ACFO and APROD are not combined because they account for similar activities, such as overproduction. REM3 can be identified as a final robustness check. In summary, the proxies for real earnings management are as follow:

REM1 = average of ACFO and ADISEXP REM2 = average of APROD and ADISEXP

REM3 = average of ACFO, APROD, and ADISEXP

Total earnings management

Total earnings management TEM is the sum of REM3 and DACC in line with the research of Franz et al. (2014). The overall impact of earnings management incentives can be evaluated by measuring TEM.

3.2 Independent variable: debt financing pressure

Three proxies are used for debt financing pressure: (1) change in short-term debt (∆STDEB), (2) long-term debt issues (LTDIS), and (3) a lower debt financing frequency (DEBFREQ).

(19)

18 ∆STDEB is measured in line with Fields et al. (2018). Short-term debt (STDEB) is calculated by the debt in current liabilities (Compustat #34) in year t is divided by total assets (Compustat #6) in year t-1. STDEB in year t is divided by STDEB in year t-1, the outcome minus 1 results in the ∆STDEB.

∆𝑆𝑇𝐷𝐸𝐵 = 𝑆𝑇𝐷𝐸𝐵𝑡

𝑆𝑇𝐷𝐸𝐵𝑡−1 − 1 (8)

LTDIS is calculated as the long-term debt issues (Compustat #111) in year t divided by the total assets in year t-1.

𝐿𝑇𝐷𝐼𝑆 = 𝐿𝑜𝑛𝑔−𝑡𝑒𝑟𝑚 𝑑𝑒𝑏𝑡 𝑖𝑠𝑠𝑢𝑎𝑛𝑐𝑒𝑡

𝐴𝑡−1 (9)

DEBFREQ is measured by using a dummy variable. The dummy variable for DEBFREQ1 is 1 for company years that comply to the following conditions: (1) issued long-term debt in year t, (2) the total amount of long-term debt increased at least 10 percent relative to year t-1, (3) the company did not issue long-term debt at year t-1. The dummy variable for DEBFREQ2 is 1 when a company year complies to a fourth condition: (4) the company did not issue long-term debt in both years t-1 and t-2.

3.3 Control variables

Prior literature argues that industry, year, size, profitability and leverage ratio influence earnings management and are relevant to use as control variables (Fields, 2018; Franz, 2014; Fung, 2013; Kim, 2018). In line with their research, industry and year are fixed effects variables. Industry (IND) is measured as the first two-digit SIC number. Year (YEAR) is defined as the fiscal year of the company.

Size (SIZE) is expected to have a negative effect on earnings management and is calculated as the Log of the total assets. Profitability is measured as the net income (Compustat #172) divided by the total assets, resulting in the return on assets (ROA). ROA is expected to have a positive effect on earnings management, because positive earnings management results in higher net income. Leverage (LEV) is defined as the total long-term debt (Compustat #9) divided by the total assets and is expected to have a positive effect on earnings management. All variables are winsorized at the 1 and 99 percent level.

(20)

19

3.4 Models

The current study focusses on debt financing pressure and the effect on earnings management. The proxies for debt financing pressure and the control variables are included in the following four models. Earnings management (EM) represents DACC, REM1, REM2, REM3, and TEM. In table 2 all abbreviations in the models are given. The models that are used for testing the hypotheses are as follows:

Hypothesis 1

𝐸𝑀_𝑡= 𝛽₀ + 𝛽₁(∆𝑆𝑇𝐷𝐸𝐵𝑡) + 𝛽2(𝐿𝐸𝑉𝑡) + 𝛽3(𝑅𝑂𝐴𝑡) + 𝛽4(𝑆𝐼𝑍𝐸𝑡) +

𝛽₅(𝐼𝑁𝐷_𝑡) + 𝛽₆(𝑌𝐸𝐴𝑅_𝑡) (10)

In equation 10 both EM and ∆𝑆𝑇𝐷𝐸𝐵 are measured in year t because when the debt in current liabilities increases in year t, it is expected that the earnings are positively inflated in the same year since within a short period of time debt must be issued.

Hypothesis 2 𝐸𝑀𝑡−1= 𝛽0 + 𝛽1(𝐿𝑇𝐷𝐼𝑆𝑡) + 𝛽2(𝐿𝐸𝑉𝑡) + 𝛽3(𝑅𝑂𝐴𝑡) + 𝛽4(𝑆𝐼𝑍𝐸𝑡) + 𝛽₅(𝐼𝑁𝐷𝑡) + 𝛽6(𝑌𝐸𝐴𝑅𝑡) (11) Hypothesis 3 𝐸𝑀_𝑡−1= 𝛽₀ + 𝛽₁(𝐷𝐸𝐵𝐹𝑅𝐸𝑄1_𝑡) + 𝛽₂(𝐿𝐸𝑉_𝑡) + 𝛽₃(𝑅𝑂𝐴_𝑡) + 𝛽₄(𝑆𝐼𝑍𝐸_𝑡) + 𝛽₅(𝐼𝑁𝐷_𝑡) + 𝛽₆(𝑌𝐸𝐴𝑅_𝑡) (12) 𝐸𝑀_𝑡−1= 𝛽₀ + 𝛽₁(𝐷𝐸𝐵𝐹𝑅𝐸𝑄2_𝑡) + 𝛽₂(𝐿𝐸𝑉_𝑡) + 𝛽₃(𝑅𝑂𝐴_𝑡) + 𝛽₄(𝑆𝐼𝑍𝐸_𝑡) + 𝛽₅(𝐼𝑁𝐷_𝑡) + 𝛽₆(𝑌𝐸𝐴𝑅_𝑡) (13)

In equation 11, 12 and 13 EM is measured in year t-1 and LTDIS, DEBFREQ1 and DEBFREQ2 are measured in year t. All measures of LTDIS, DEBFREQ1, and DEBFREQ2 represent the issue of long-term debt in year t. It is expected that earnings management is used in the year prior to debt issues.

(21)

20

Table 2.

Variables abbreviations

Abbreviation Explanation

Dependent variables

DACC accrual-based earnings managmeent: discretionary accrual

REM1 real earnings management: average of abnormal operating cash flows and abnormal discretionary expenses

REM2 real earnings management: average of abnormal production costs and abnormal discretionary expenses

REM3 real earnings management: average of abnormal operating cash flows, abnormal production costs and abnormal discretionary expenses TEM total earnings management: sum of REM3 and DACC

EM DACC, REM1, REM2, REM3 or TEM

Independent variables

STDEB short-term debt

∆STDEB change in short-term debt LTDIS long-term debt issue

DEBFREQ1 debt financing frequency: company did not issue long-term debt at year t-1 DEBFREQ2 debt financing frequency: company did not issue long-term debt at year t-1

and year t-2

Control variables

LEV leverage ratio

ROA return on assets

SIZE log total assets

IND industry (first two-digit SIC code)

(22)

21

4. Results

The results are discussed in this section. Descriptive statistics are presented for the variables used in the regression models. Additionally, a Pearson correlation matrix and the results of the regression models will be discussed. The reversal effect of the discretionary accruals is discussed in the additional tests. Finally, robustness tests are performed to confirm the results.

4.1 Descriptive statistics

In table 3 are the descriptive statistics of the variables. The total observations (N), mean, median, standard deviation (Std. dev.), minimum and maximum are shown. All the means and medians of the dependent variables (DACC, REM1, REM2, REM3, and TEM) are close to zero because earnings are managed both up- and downwards. Only companies with positive long-term debt are included. Therefore, the descriptive statistics of the independent variables may be relatively high. The mean of STDEB is 0.0568 and the median is 0.0201. LTDIS shows a mean of 0.1517 and a median of 0.0352. In other words, the companies have an average short-term debt in current liabilities of 5,68% and long-term debt issues of 15,17% relative to the total assets in year t-1. The ∆STDEB has a mean of 2.3626 and median of -0.0475, suggesting that the STDEB of the companies changes substantially each year. The highest amount of ∆STDEB is 74.2830, meaning an increase of 7,428% in STDEB relatively to the prior year. The mean of 236% of ∆STDEB is close to the mean in a similar study by Fields et al. (2018) of 216%, using the same measure of ∆STDEB. DEBFREQ1 and DEBFREQ2 are dummy variables. In the final sample, most of the observations result in a 0. The mean of the LEV is 0.2128. The ROA has an average of 0.0213 and a median of 0.0320. The mean and median of the SIZE are 2.8603 and 2.8958 respectively.

(23)

22

4.2 Correlation matrix

The Pearson correlation matrix for the dependent, independent and control variables is shown in table 4. Earnings management is measured in year t for ∆STDEB and in year t-1 for LTDIS, DEBFREQ1, and DEBFREQ2. Therefore, both year t and year t-1 are included in table 4. The correlations between the independent variables and control variables are low. This shows there is no multicollinearity in one of the models (equation 10, 11, 12, 13). All control variables have a significant effect on the dependent variables. LEV has a positive effect on DACC, REM1, REM2, REM3, and TEM. ROA also has a positive effect on all dependent variables. SIZE shows a negative effect on DACC and a positive effect on REM1, REM2, REM3, and TEM.

The correlations between ∆STDEB and DACC, REM1, REM2, REM3, and TEM are all positive. However, the correlations with REM1 and REM3 are insignificant. Further analysis including the full model will clarify the relation. A positive relation is expected following the prior

sections. 𝐷𝐴𝐶𝐶_𝑡−1, 𝑅𝐸𝑀1_𝑡−1, 𝑅𝐸𝑀2_𝑡−1, 𝑅𝐸𝑀3_𝑡−1 and, 𝑇𝐸𝑀_𝑡−1 show a positive correlation

with LTDIS. The positive correlation between LTDIS and 𝐷𝐴𝐶𝐶𝑡−1 is insignificant and will be

studied in further analysis. The correlations between the dependent variables and DEBFREQ1 or DEBFREQ2 are comparable. Both show positive and negative relations between

Table 3.

Descriptive statistics for the final sample 2005-2017

Variables N Mean Median Std. dev. Min Max

Dependent variables DACC 31,792 -.0032 -.0100 .1694 -.6665 1.0213 REM1 31,787 .0046 .0006 .1748 -1.2236 1.2158 REM2 30,221 .0103 .0048 .1310 -1.1544 1.2244 REM3 30,220 .0084 .0111 .1001 -1.0443 .9216 TEM 30,220 .0038 -.0008 .1889 -1.5302 1.9430 Independent variables STDEB 29,545 .0568 .0201 .1030 0.000 1 ∆STDEB 26,473 2.3626 -.0475 10.6074 -1.000 74.2830 LTDIS 22,243 .1517 .0352 .2473 0.000 1 DEBFREQ1 34,107 .0586 0 .2350 0 1 DEBFREQ2 34,107 .0318 0 .1755 0 1 Control variables LEV 34,107 .2128 .1806 .1760 0 .9337 ROA 34,107 .0213 .0320 .0871 -.1233 .2841 SIZE 34,107 2.8603 2.8958 .9774 .0030 5.7055

(24)

23 DEBFREQ1 or DEBFREQ2 and 𝐷𝐴𝐶𝐶_𝑡−1, 𝑅𝐸𝑀1_𝑡−1, 𝑅𝐸𝑀2_𝑡−1, 𝑅𝐸𝑀3_𝑡−1 and, 𝑇𝐸𝑀_𝑡−1.

𝐷𝐴𝐶𝐶_𝑡−1 has a positive significant correlation and 𝑅𝐸𝑀1_𝑡−1, 𝑅𝐸𝑀2_𝑡−1, 𝑅𝐸𝑀3_𝑡−1 show a

negative correlation. The negative correlation is only significant for 𝑅𝐸𝑀1𝑡−1 and DEBFREQ1

or DEBFREQ2. 𝑇𝐸𝑀_𝑡−1 has a positive correlation with DEBFREQ1 and DEBFREQ2. This

positive correlation is only significant for DEBFREQ1 indicating that the total earnings management is positive in the year prior to a debt issue in case the company did not issue debt in year t-1.

Most correlations between the discretionary accruals and the independent variables are significantly positive, only the positive correlation with LTDIS is insignificant. This suggests that debt financing pressure leads to positive accrual-based earnings management.

The measures for real earnings management show significant positive, significant negative and insignificant correlations with the debt financing pressure variables. ∆STDEB and LTDIS show positive correlations with real earnings management and the frequency of issuing debt variables show negative correlations.

Total earnings management is positively and significantly correlated with all independent variables, only DEBFREQ2 is insignificant. The results suggest that debt financing pressure leads to higher levels of total earnings management. Further statistical tests including all control and fixed-effects variables in the next paragraph will give a better understanding of the relations.

(25)

24 T ab le 4. P ea rs on c or re la tion m at ri x LEV R O A S IZ E D A C C D A C C t-1 R E M 1 R E M 1t -1 R E M 2 R E M 2t -1 R E M 3 R E M 3t -1 TEM T E M t-1 S T D E B ∆ S T D E B L T D IS D E B F R E Q 1 LEV RO A -.072 3** * S IZ E .2334** * .3105** * D A C C .0130** .0873** * -.016 9** * D A C C t-1 -.018 0** .0308** * -.024 5** * -.034 0** * R E M 1 .0435** * .1770** * .1376** * .0205** * .0658** * R E M 1t -1 .0255** * .1381** * .1398** * .0296** * .0477** * .4688** * R E M 2 .0230** .1215** * .1066** * .0380** * .0544** * .7468** * .2177** * R E M 2t -1 .0140** * .0774** * .1041** * .0078 .0431** * .2059** * .7436** * .2408** * R E M 3 .0276** * .3388** * .1878** * -.023 3** * .0528** * .7556** * .2150** * .8374** * .1941** * R E M 3t -1 .0219** * .2633** * .1894** * .0096 -.013 2** .1933** * .7513** * .1820** * .8374** * .3060** * TEM .0347** * .2621** * .0937** * .8479** * -.001 2 .4225** * .1397** * .4768** * .1129** * .5102** * .1756** * T E M t-1 .0024 .1709** * .0915** * -.021 0** * .8531** * .1504** * .4197** * .1335** * .4740** * .1991** * .5104** * .0911** * S T D E B -.108 2** * -.113 2** * -.156 9** * .1232** * .0464** * .0802** * .0167** * .0727** * .0105 -.000 9 -.039 1** * .0831** * .0128** ∆ S T D E B -.016 3** -.018 4** * .0215** * .0190** * -.022 8** * .0052 -.017 7** * .0160** -.016 6** -.000 2 -.018 3** * .0161** -.029 5** * .1341** * L T D IS .3488** * -.028 3** * .0216** * .0644** * .0062 .1157** * .0391** * .1341** * .0395** * .0896** * .0260** * .0978** * .0207** * .1207** * .0466** * D E B F R E Q 1 .0264** * .0010 .0080 .0091 .0129** .0147* -.015 9** * .0324** * -.005 2 .0173** -.002 1 .0194** * .0104* -.019 3** * .0491** * .0928** * D E B F R E Q 2 .0036 -.002 9 -.006 5 .0127** .0100* .0055 -.019 9** * .0201** * -.005 6 .0082 -.002 2 .0174** .0068 -.014 4* .0568** * .0568** * .7263** * * * * , * * , a n d * i n d ic a te p < .0 1 , .0 5 , a n d . 1 0 , re s p e c ti v e ly ( tw o -t a il e d ).

(26)

25

4.3 Regression models

The results of twenty regression models are presented in table 5. The four models in the methodology section numbered 10, 11, 12 and 13 include all five methods of measuring earnings management. The control variables and industry and year fixed-effects are included in all twenty models. The coefficients of the control variables in most models are significant in the expected direction. The hypotheses are linked to the models in table 5.

∆STDEB as independent variable

Model 1 H1a: A change in debt in current liabilities is positively associated with

accrual-based earnings management.

Models 2/3/4 H1b: A change in debt in current liabilities is positively associated with real earnings management.

Model 5 H1c: A change in debt in current liabilities is positively associated with total

earnings management.

Model 1 shows a positive relation between ∆STDEB and DACC of .0003 with a significance level of p<.01. Models 2, 3 and 4 show a positive relation between ∆STDEB and REM1, REM2 and REM3. Only model 3 (REM2) is significantly positive with a coefficient of .0002 and a significance level of p<.05. This indicates that the effect of ∆STDEB on abnormal sales levels has no positive effect because real earnings management by boosting sales (ACFO) is not included in REM2. ∆STDEB has a positive significant effect on TEM at the p<.01 level in model 5. The coefficient of .0004 is higher than the coefficient of .0003 in model 1. This indicates that both accrual-based earnings management and real earnings management (except for ACFO) are used to inflate earnings due to debt financing pressure.

H1a and H1c are accepted. H1b is rejected, considering that real earnings management by boosting sales is not used to inflate earnings due to changes in debt in current liabilities. LTDIS as independent variable

Model 6 H2a: Long-term debt issues are positively associated with accrual-based

earnings management in the prior year.

Models 7/8/9 H2b: Long-term debt issues are positively associated with real earnings management in the prior year.

Model 10 H2c: Long-term debt issues are positively associated with total earnings

(27)

26 LTDIS has a positive but insignificant effect on 𝐷𝐴𝐶𝐶_𝑡−1 with a coefficient of .0053 and a significance level of p=.212 in model 6. This result will be discussed in the next section. Models 7, 8 and 9 all show positive and significant effects with a significance level of p<.01. Long-term debt issues have a clear positive effect on real earnings management in the prior year. Additionally, LTDIS has a positive effect which is significant at the p<.01 as presented in model

10. This suggests that LTDIS should also affect 𝐷𝐴𝐶𝐶_𝑡−1, because 𝑇𝐸𝑀_𝑡−1 is the sum of

𝐷𝐴𝐶𝐶_𝑡−1 and 𝑅𝐸𝑀3_𝑡−1. However, there is no certainty about H1a because model 6 is

insignificant. H1b and H1c are accepted. H1a is rejected, although a positive effect is suggested by the results.

DEBFEQ1 and DEBFREQ2 as independent variables

Models 11/16 H3a: A lower frequency of issuing debt is positively associated

with accrual-based earnings management in the prior year.

Models 12/13/14/17/18/19 H3b: A lower frequency of issuing debt is positively associated

with real earnings management in the prior year.

Models 15/20 H3c: A lower frequency of issuing debt is positively associated

with total earnings management in the prior year.

Models 11 and 16 show positive significant effects of DEBFREQ1 and DEBFREQ2 on

𝐷𝐴𝐶𝐶_𝑡−1. DEBFREQ1 has a coefficient of .0117 and a significance level of p<.01 and

DEBFREQ2 has a coefficient of .0118 and a significance level of p<.05. This shows that companies with a lower debt financing frequency show higher discretionary accruals in the prior year when issuing long-term debt. Contrary to the expected relation, the models 12, 13, 14, 17, and 18 show a negative effect which is insignificant in four of these models. DEBFREQ1

and DEBFREQ2 both have a significant negative effect on 𝑅𝐸𝑀1_𝑡−1 with coefficients of -.0078

and -.0138 and significance levels of p<.10 and p<.5 respectively. 𝑅𝐸𝑀2_𝑡−1 and 𝑅𝐸𝑀3_𝑡−1 are

both insignificant, suggesting real earnings management is not practiced by boosting

production, because 𝑅𝐸𝑀1𝑡−1 excludes real earnings management through abnormal

production costs. Models 15 and 20 both show positive significant effects with coefficients of .0086 and .0097 and significance levels of p<.5 and p<.10 respectively. The results indicate that

both DEBFREQ1 and DEBFREQ2 have a significant positive effect on 𝑇𝐸𝑀_𝑡−1. However, the

coefficients and significance levels are lower than models 11 and 16 with 𝐷𝐴𝐶𝐶_𝑡−1. Considered

the negative effect of real earnings management and the positive effect of discretionary accruals, the results suggest that total earnings management in the prior year is positive due to

(28)

27 the high positive discretionary accruals in the prior year. H1a and H1c are accepted while H1b is rejected. Earnings may be negatively managed by cutting sales and boosting abnormal discretionary expenses.

To conclude, all independent variables, ∆STDEB, LTDIS, DEBFREQ1, and DEBFREQ2, have a positive significant effect on total earnings management. Three out of four independent variables have a significant positive effect on accrual-based earnings management. The effects of the independent variables on real earnings management are mixed.

(29)

28 T ab le 5. R es ul ts f rom r eg re ss ion m ode ls V ar iab le s P re d ic ti on M od e ls / d e p e n d e n t var iab le 1. D A C C 2. R E M 1 3. R E M 2 4. R E M 3 5. T E M Co e ff ic ie n t p -v al u e Co e ff ic ie n t p -v al u e Co e ff ic ie n t p -v al u e Co e ff ic ie n t p -v al u e Co e ff ic ie n t p -v al u e ∆ S T D E B + .0003** * .000 .0001 .372 .0002** .020 .00002 .711 .0004** * .001 LEV + .0341** * .000 .0353** * .000 .0152** * .005 .0243** * .000 .0585** * .000 R O A + .2200** * .000 .3155** * .000 .1768** * .000 .3854** * .000 .6055** * .000 S IZ E --.010 3** * .000 .0165** * .000 .0114** * .000 .0869** * .000 -.001 6 .216 V ar iab le s P re d ic ti on 6. D A C C t-1 7. R E M 1t -1 8. R E M 2t -1 9. R E M 3t -1 10. T E M t-1 Co e ff ic ie n t p -v al u e Co e ff ic ie n t p -v al u e Co e ff ic ie n t p -v al u e Co e ff ic ie n t p -v al u e Co e ff ic ie n t p -v al u e L T D IS + .0053 .212 .0367** * .000 .0252** * .000 .0101** * .000 .0150** * .003 LEV + -.018 0** * .005 -.012 3* .065 -.013 7** * .008 .0014 .710 -.016 1** .032 R O A + .0829** * .000 .2342** * .000 .0855** * .000 .2767** * .000 .3690** * .000 S IZ E --.007 1** * .000 .0236** * .000 .0146** * .000 .0136** * .000 .0081** * .000 V ar iab le s P re d ic ti on 11. D A C C t-1 12. R E M 1t -1 13. R E M 2t -1 14. R E M 3t -1 15. T E M t-1 Co e ff ic ie n t p -v al u e Co e ff ic ie n t p -v al u e Co e ff ic ie n t p -v al u e Co e ff ic ie n t p -v al u e Co e ff ic ie n t p -v al u e D E B F R E Q 1 + .0117** * .003 -.007 8* .059 -.003 1 .305 -.001 1 .622 .0086** .050 LEV + -.016 0** * .008 .0066 .297 -.000 9 .857 .0065* .067 -.009 1 .196 R O A + .0830** * .000 .2361** * .000 .0870** * .000 .2773** * .000 .3696** * .000 S IZ E --.007 2** * .000 .0229** * .000 .0142** * .000 .0135** * .000 .0079** * .000 in d u str y (f ir st tw o -d ig it SI C co d e ) V ar iab le s P re d ic ti on 16. D A C C t-1 17. R E M 1t -1 18. R E M 2t -1 19. R E M 3t -1 20. T E M t-1 Co e ff ic ie n t p -v al u e Co e ff ic ie n t p -v al u e Co e ff ic ie n t p -v al u e Co e ff ic ie n t p -v al u e Co e ff ic ie n t p -v al u e D E B F R E Q 2 + .0118** .025 -.013 8** .013 -.003 3 .414 .0004 .904 .0097* .098 LEV + -.015 6** * .010 .0064 .310 -.001 0 .837 .0065* .071 -.008 8 .212 R O A + .0831** * .000 .2361** * .000 .0870** * .000 .2773** * .000 .3697** * .000 S IZ E --.007 2** * .000 .0229** * .000 .0141** * .000 .0135** * .000 .0079** * .000 * * * , * * , a n d * i n d ic a te p < .0 1 , .0 5 , a n d . 1 0 , re s p e c ti v e ly ( tw o -t a il e d ). A ll m o d e ls i n c lu d e i n d u s tr y a n d y e a r fi xe d -e ff e c ts v a ri a b le s .

(30)

29

4.4 Additional tests

The reversal effects of the discretionary accruals give a more complete view of the use of accrual-based earnings management. Table 6 shows the reversal effect in the two years following accruals-based earnings management is expected. The models 21, 24, 27 and 30 in table 6 are the same as the models 1, 6, 11 and 16 in table 5. Models 22 and 23 for example, show the accrual-based earnings management in the next two years.

The reversal effect is demonstrated in the models 27, 28 and 29 regarding the independent variable DEBFREQ2. Positive accrual-based earnings management is practiced in the year prior to the long-term debt issue (model 27). In the same year as the debt issue (model 28), an insignificant effect is presented. The reversal effect is observable two years later (model 29), resulting in a negative coefficient of -.0068 with a significance level of p<.05. Models 30, 31 and 32 reveals the reversal effect as well. In the year t-1 (model 30) there is a significant positive effect with a coefficient of. 0118. In the year t (model 31) the significant positive effect is lower with a coefficient of .0088. Year t+1 (model 32) shows significant negative accrual-based earnings management with a coefficient of -.0114. The reversal effect is also observable in the models 21 and 22. In the year following the ∆STDEB increased (model 22), the accrual-based earnings management is negative with a coefficient of -.0002 and a significance level of p<.10. In conclusion, the inflated earnings show a reversal effect in the following years, resulting in decreasing accrual-based earnings management.

Another interesting result in table 6 is model 25. The outcome state that higher amounts of long-term debt issues lead to higher discretionary accruals in the same year. Model 31 also shows a positive significant effect, which indicates that companies with a lower frequency of issuing debt have higher discretionary accruals in the same year as they issue long-term debt. In the current study fiscal years are used. Companies may use earnings management in the first quarter and issue long-term debt in the fourth quarter of the year. Quarterly earnings management research could give some new interesting insights. This will be further discussed in the next section.

(31)

30

4.5 Robustness tests

A robustness test for ∆STDEB and LTDIS is presented in table 7 including ten models. These independent variables are changed into dummy variables. These two dummy variables, ∆STDEB20 and LTDIS20, are used in the models. ∆STDEB20 is 1 for the 20% highest increases of STDEB and LTDIS is 1 for the 20% most substantive long-term debt issues relative to the total assets in year t-1.

All ten models show positive relationships with earnings management. Only models 36 and 38 are insignificant. Model 36 shows that ∆STDEB20 does not have a significant positive effect on REM3 and model 38 states that LTDIS20 does not have a significant positive effect on

𝐷𝐴𝐶𝐶𝑡−1. These results are similar to models 4 and 6 in table 5 respectively. Contrary to table

5, ∆STDEB20 does have a positive significant effect on REM1 in model 34, with a coefficient of .0074 and a significance level of p-<.01. Additionally, model 35 shows a significance level of p-<.01, which is lower than the significance level of p-<.05 in model 3 of table 5. Hypotheses H1a and H1c are confirmed. H1b cannot be accepted with certainty and is therefore rejected. The robustness tests indicate that more research is needed to investigate the effect of changes in short-term debt on real earnings management. The current study cannot conclude there is a significant positive effect. The models 38, 39, 40, 41 and 42 regarding LTDIS20, show the same relationships in table 7 as the models 6, 7, 8, 9 and 10 regarding LTDIS in table 5. Therefore, the results in table 5 are confirmed. H2b and H2c are confirmed and H2a is rejected again.

Table 6.

Reversal effect accrual-based earnings management

Variable Prediction Models / dependent variable

21. DACC 22. DACCt+1 23. DACCt+2

Coefficient p -value Coefficient p -value Coefficient p -value

∆STDEB + .0003*** .000 -.0002* .076 -.0001 0.586

Variable Prediction 24. DACCt-1 25. DACC 26. DACCt+1

LTDIS + .0053 .212 .0408*** .000 -.0003 .940

DEBFREQ1 + .0117*** .003 .0024 .551 -.0068* .074

DEBFREQ2 + .0118** .025 .0088* .10 -.0114** .024

***, **, and * indicate p<.01, .05, and .10, respectively (two-tailed).

(32)

31 T ab le 7. R obu st ne ss t es ts V ar iab le s P re d ic ti on M od e ls / d e p e n d e n t var iab le 33. D A C C 34. R E M 1 35. R E M 2 36. R E M 3 37. T E M Co e ff ic ie n t p -v al u e Co e ff ic ie n t p -v al u e Co e ff ic ie n t p -v al u e Co e ff ic ie n t p -v al u e Co e ff ic ie n t p -v al u e ∆ S T D E B 20 + .0188** * .000 .0074** * .004 .0096** * .000 .0011 .466 .0198** * .000 V ar iab le s P re d ic ti on 38. D A C C t-1 39. R E M 1t -1 40. R E M 2t -1 41. R E M 3t -1 42. T E M t-1 Co e ff ic ie n t p -v al u e Co e ff ic ie n t p -v al u e Co e ff ic ie n t p -v al u e Co e ff ic ie n t p -v al u e Co e ff ic ie n t p -v al u e L T D IS 20 + .0018 .469 .0185** * .000 .0136** * .000 .0060** * .000 .0085** * .005 * * * , * * , a n d * i n d ic a te p < .0 1 , .0 5 , a n d . 1 0 , re s p e c ti v e ly ( tw o -t a il e d ). ∆S TD EB 20 a nd L TD IS 20 a re d um m y va ria bl es , t he h ig he st 2 0% is 1 . A ll m o d e ls i n c lu d e i n d u s tr y a n d y e a r fi xe d -e ff e c ts v a ri a b le s a n d t h e c o n tr o l v a ri a b le s L E V , R O A a n d S IZE .

The effect of debt financing pressure on earnings management