The use of classification shifting during the
financial crisis: A European study
Abstract: Prior literature has investigated three forms of earnings management: real earnings management, accrual management, and classification shifting. In this study, I examine the effects of a financial crisis on the use of classification shifting. Literature suggests that the use of real earnings management and accrual management decreases during a financial crisis. Because prior research suggests a substitution effect between classification shifting and the other earnings management tools, classification shifting could increase during a financial crisis. Thereby, the increased possibility of debt covenant violation during a financial crisis could result in an increased use of classification shifting. In my study I found evidence suggesting that classification shifting decreases during a financial crisis, but increases if firms in a financial crisis are also in financial distress. I also performed a test for high leveraged firms, but no significant results were generated, so no conclusions can be drawn.
Keywords: classification shifting, earnings management, financial crisis, special items, analyst forecasts, debt-covenant violations.
Data Availability: Data are available from public sources identified in the paper.
Author: J.B.F. Reinerink
Student number: 2546876
Supervisor: Prof. Dr. C.K. Hoi University: University of Groningen
Faculty: Economics and Business
Date: 04-08-2019
City: Tubbergen
Table of Contents
I. INTRODUCTION 3
II. BACKGROUND LITERATURE 6
III. HYPOTHESIS DEVELOPMENT 13
IV. MEASURING CLASSIFICATION SHIFTING 14
V. DATA, SAMPLE SELECTION, AND DESCRIPTIVE STATISTICS 17
VI. RESEARCH DESIGN AND RESULTS 23
VII. DISCUSSION AND CONCLUSION 31
I. INTRODUCTION
Firms can manipulate earnings to meet analyst expectations by engaging in accrual management, real-activities management, and classification shifting (Roychowdhury, 2006; Athanasakou, Strong & Walker, 2011). Whereas the extant literature has explored earnings management, with primary foci on manipulations of accruals and real activities (e.g., Dechow, Sloan & Sweeney, 1995; Jones, 1991; Roychowdhury, 2006), McVay (2006) has introduced another earnings-management tool: classification shifting. According to McVay (2006), the main purpose of classification shifting is to deliberately misclassify items within the income statement. Managers opportunistically shift core expenses (defined as cost of goods sold and selling, general and administrative expenses) to special items to inflate current core earnings, for example, thereby producing a positive relation between unexpected core earnings and income-decreasing special items.
Analysts, investors, senior executives and boards of directors consider earnings to be the most important item in financial reports (DeGeorge, Patel & Zeckerhauser, 1999). Analysts use these earnings to generate forecasts for future firm performance. In turn, a manager could use classification shifting to meet or beat forecasts (McVay, 2006). According to McVay (2006), analysts tend to exclude special items from core earnings; thus, shifting the core expenses (cost of goods sold and selling, general, and administrative expenses) to special items could result in firms meeting the analyst benchmarks when they would otherwise not have met these benchmarks. Special items, as used by McVay, are material events that arise from a firm’s ongoing, continuing activities; they are either unusual in nature or infrequent in occurrence and must be disclosed as a separate line item as part of income from continuing operations or in footnotes to the financial statements (Revsine, Collins & Johnson, 2005).
Empirical evidence (e.g., Bloomfield & Libby, 1996; Gelb & Zarowin, 2002) suggests that Generally Accepted Accounting Principles (GAAP) earnings may not provide sufficient value-relevant information for investors and may need to be supplemented by non-GAAP measures. According to Bloomfield and Libby (1996, p. 204), investors may perceive information disclosed in more available locations to be more reliable, may misinterpret information in less available locations such as footnotes, and may recognize that financial contracts are typically written based on more available recognized balances.
In addition, according to Gelb and Zarowin (2002), greater disclosure increases informativeness. Recent research has provided evidence that managers include pro forma earnings to supplement GAAP earnings (Bhattacharya, Black, Christensen, & Mergenthaler,
2004). According to Entwistle, Feltham and Mbagwu (2010), pro forma earnings are the most value relevant to investors. According to Bhattacharya et al. (2004), to provide a clearer picture of a firm’s sustainable or recurring earnings, pro forma earnings exclude special or transitory items. Thus, Fan, Thomas and Yu (2019) found evidence which suggests that managers misclassify core expenses as income-decreasing special items to increase earnings before interest, taxes, depreciation and amortization (EBITDA) and thereby endeavour to avoid debt-covenant violations.
Additionally, Poonawala and Nagar (2019) provide evidence that managers could also use classification shifting to inflate the gross profits of a firm. They provide evidence that managers sometimes shift cost of goods sold to research and development (R&D) to meet the prior year’s gross margin.
Prior research suggests that about half of the earnings quality is determined by innate factors, for example, business model, industry and macroeconomic conditions (Dichev, Graham, Harvey & Rajgopal, 2013). In this study, I focus on the effect of such a macroeconomic condition: the financial crisis. During a financial crisis, firms will notice a sharp contraction of economic activity (Waymire & Basu, 2011). Filip and Raffournier (2014) provide evidence that, during financial crisis, earnings smoothing decreases and accrual quality improves. This suggests that macroeconomic conditions like a financial crisis have a negative effect on earnings management.
In addition, Filip and Raffournier (2014) report different reasons why earnings management would decrease during a financial crisis. First, managers have fewer incentives to manage their earnings during a financial crisis in comparison with a period with higher financial performance because of a higher market tolerance for poor performance. Second, litigation risk increases during a financial crisis. In addition, Kudrna (2016) reports evidence that litigation output in the seven years after the financial crisis almost doubled in comparison with the litigation input of the seven years prior to the financial crisis. Finally, a decrease in earnings management might also be a response to a higher demand for timely earnings during more troubled periods.
However, prior research also suggests that there could also be incentives to increase earnings management during a financial crisis. Due to the sharp contraction of economic activity during a financial crisis (Waymire & Basu, 2011), firms are more likely to face debt-covenant violations. Fan et al. (2019) provide evidence that managers engage in classification shifting to increase their EBITDA and thereby avoid debt-covenant violations, because a
violation of debt covenants can be a bad sign for investors (DeFond & Jiambalvo, 1994; Dichev & Skinner, 2002).
The aim of this paper is to explore whether European firms experience higher or lower levels of classification shifting as a result of the financial crisis. To determine the impact, I will consider the use of classification shifting before and during the financial crisis.
The financial crisis began in 2007 in the United States but peaked during 2008 with the bankruptcy of several major financial institutions such as Bear Stearns, Lehman Brothers, Merrill Lynch and Citigroup (Kothari & Lester, 2012). However, as shown in Figure 1, the crisis struck Europe in 2008. Consequentially, because I focus on Europe, I consider 2006-2007 to be pre-crisis years and 2008-2009 to be crisis years. I treat the pre-crisis years, the first phase of this study, as a benchmark so I can compare the years affected by the financial crisis with the years prior to the financial crisis.
Figure 1: Development of the EuroSTOXX50 index.
The second phase of this study, as mentioned above, is comprised of the years 2008-2009. During these years, the financial crisis struck Europe, and the European stock market collapsed (Figure 1). The sample consists of European firms, because the financial crisis had a greater impact on Europe than on the rest of the world (Filip & Raffournier, 2014). According to Filip and Raffournier (2014), the GDP growth rate in Europe fell to -4.30% whilst the GDP growth rate fell to -2.05% in the rest of the world. A comparison will be made between the first and second phase to identify the change in the use of classification shifting after the start of the financial crisis.
According to Abernathy, Beyer and Rapley (2014), classification is used as a substitute for both real earnings management and accrual management. They provide evidence that
0 1000 2000 3000 4000 5000 1-1-2005 1-1-2006 1-1-2007 1-1-2008 1-1-2009 1-1-2010 1-1-2011
EuroSTOXX50
substitution between classification shifting and real earnings management is likely in poor financial times and that substitution between classification shifting and accrual management is likely when accounting flexibility decreases. Thus, prior research states that earnings management decreases during a financial crisis (Filip & Raffournier, 2014; Cimini, 2015). In addition, managers can engage in classification shifting to avoid debt-covenant violations (Fan et al., 2019). Therefore, I expect an increase in the level of classification shifting during the financial crisis.
Taken together, to examine all the information above, this study uses the following research question to examine the use of classification shifting:
What is the impact of the financial crisis on the use of classification shifting?
This paper contributes to the accounting literature and to the literature of incentives and constraints which confront managers who are willing to manipulate earnings to meet or beat certain analysts’ forecasts. Thereby, this study also illustrates that the use of classification shifting is sensitive to external economic conditions, like a financial crisis. Finally, since research into classification shifting is relatively new, this study also contributes to the literature of classification shifting. By obtaining a better understanding of the developments of the use of classification shifting, a better understanding of the behaviour of managers is acquired.
This study is comprised of seven sections. In the next section, I discuss the background literature I use to construct my hypothesis. In the third section, I develop my hypothesis. In section four, I provide a better understanding of the determination of the measurement of classification shifting and the other variables. In the fifth section, I formulate the data selection and the descriptive statistics. Section six describes the research design, the tests of hypothesis and the results of this study. Section seven offers discussion and the conclusion.
II. BACKGROUND LITERATURE
Earnings management
Prior research has extensively examined earnings management and has frequently found substantial evidence of managers engaging in earnings management.1 According to Healy and Wahlen (1999, p. 368), “Earnings management occurs when managers use judgment in
1 Examples of studies provide evidence of the use of earnings management by executives are Dichev et al. (2013), Healy and Wahlen (1999), Kasznik (1999) and McVay (2006).
financial reporting and in structuring transactions to alter financial reports to either mislead some stakeholders about the underlying economic performance of the company or to influence contractual outcomes that depend on reported accounting numbers”. More recent research places less emphasis on the deception and more on the influence. According to Walker (2013), “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.” To make it hard to identify earnings management, management uses invisible and subtle accounting choices to mislead outsiders (Dichev et al., 2013).
Managers employ several methods of earnings management (Malikov, Manson & Coakley, 2018; McVay, 2006). One method is accrual management (e.g., Healy & Wahlen, 1999). The purpose of accruals is to reflect the real performance of a firm, but managers can manipulate accruals to alter firm performance on an accounting basis (Healy & Wahlen, 1999). By doing so, managers can shift income between periods by changing the timing of the reported or actual events of a payment (Degeorge et al., 1999). Managers could, for example, use acceleration of sales, alterations of shipment schedules, and delaying of research and development and maintenance expenditures to alter their earnings (Dechow & Skinner, 2000; Fudenberg & Tirole, 1995; Healy & Wahlen, 1999).
In case of accrual management, managers could also manipulate the judgment which is required to estimate numerous future economic events. According to Healy and Wahlen (1999, p. 369), examples of such events include salvage values of long-term assets, obligations for pension benefits and other post-employment benefits, deferred taxes and losses from bad debts and asset impairments. Because these events include several judgements, management could easily influence these events by managing the accruals. Management must consider which method to use for reporting economic transactions, such as the straight-line or accelerated depreciation method or the First In First Out (FIFO), Last In First Out (LIFO), or weighted average inventory valuation methods (Healy & Wahlen, 1999).
As a consequence of accrual management, accounts can look more positive (or negative) than they actually are, so certain benchmarks (e.g., zero earnings, last year’s earnings, and analysts’ forecasted earnings) could be met, which was not the case otherwise (Xu, 2016). Nevertheless, in the following years, these positive earnings will decline as the accruals reverse (Sloan, 1996).
Another method of earnings management is real earnings management. This method does not only occur by accounting estimates and methods, but a firm could also use operational
decisions (Roychowdhury, 2006). In case of manipulation of real economic activities or real earnings management, firms engage in earnings management by affecting cash flows and earnings (Roychowdhury, 2006). Roychowdhury (2006, p. 337) defines real activities manipulation as “departures from normal operational practices, motivated by managers’ desire to mislead at least some stakeholders into believing certain financial reporting goals have been met in the normal course of operations.” McVay (2006) claims that management can engage in the manipulation of real economic activities by using price discounts to increase sales and cutting discretionary expenditures to manage earnings. By doing so, the profit will increase, but it can give no indication of the real performance of the firm. The reason managers engage in real earnings management is to avoid losses or meet or beat earnings benchmarks (Baber, Fairfield & Haggard, 1991; Herrmann, Inoue & Thomas, 2003; Roychowdhury, 2006).
In 2006, McVay introduced a third method of earnings management: classification shifting. In her paper, McVay (2006) defines classification shifting as a way to inflate the core earnings by shifting income-decreasing core expenses (such as cost of goods sold and selling, general, and administrative expenses) to special items. Managers use classification shifting to inflate core earnings to meet or beat analysts’ forecasts (McVay, 2006). In the development of their expectations and forecasts, analysts tend to fixate on the core earnings instead of on the bottom-line GAAP earnings (Fan, Barua, Cready & Thomas, 2010). The method could be preferred because special items, which increase via this method, tend to be excluded from both pro forma and analyst earnings definitions (McVay, 2006). Thus, analysts tend to value different line items in comparison to sales differently (e.g., Lipe, 1986; Fairfield, Sweeney & Yohn, 1996; Francis, Hanna & Vincent, 1996). Pro forma earnings are adjusted earnings, which exclude special or transitory items to provide a clear picture of the firm’s sustainable or recurring earnings (Entwistle et al, 2010).
According to Fan et al. (2010), the use of classification shifting is not constant during the year. Fan et al. (2010) use quarterly data as opposed to the annual data of McVay (2006). They found evidence to suggest that the use of classification shifting is higher in the last quarter of the year in comparison with the prior quarters. Their study includes the models of McVay (2006) to help identify classification shifting. In contrast to the model of McVay (2006), Fan et al. (2010) add returns to their model to control for deteriorating performance and decrease in expectations. In this study, such control is not included, as the deterioration of the returns during the financial crisis is included. When testing their hypothesis—i.e., that “Managers shift core expenses to special items more in the fourth quarter”—they compare the coefficient of the
fourth quarter with the coefficient of the previous quarters. If the coefficient of the fourth quarter is more positive (less negative), there is evidence for more classification shifting.
Additionally, Fan and Liu (2017) found evidence which suggests that managers, to achieve different profitability benchmarks, can distinguish between the misclassification of cost of goods sold and that of sales, general and administrative expenses. In their study, they state that managers misclassify cost of goods sold, instead of sales, general and administrative expenses, to inflate the gross margin ratio. In addition, managers can use both the misclassification of cost of goods sold and sales, general and administrative expenses when their firms report small positive core earnings or small positive changes in core earnings or when the earnings just meet or beat the analyst forecast. Poonawala and Nagar (2019) find evidence that managers shift cost of goods sold to R&D to manipulate gross profits upward. According to Fan et al. (2019), this manipulation of core expenses can be used to avoid EBITDA-related debt-covenant violations. They provide evidence which suggests that managers use classification shifting to increase EBITDA and thereby avoiding debt-covenant violations.
Prior research also suggests that classification shifting is more likely to occur in countries with weak investor protection (Behn, Gotti, Herrmann & Kang, 2013). According to Behn et al. (2013), it is easier to pursue in classification shifting in countries with weak investor protection because there are fewer restrictions.
Differences between the earnings-management tools
The main difference between accrual management and real earnings management is that accrual management manipulates the total accruals, and the manipulation of real economic activities manipulates the net cash flows (Healy & Wahlen, 1999; Roychowdhury, 2006).
When comparing the different earnings-management tools with classification shifting, prior research suggests that firms use classification shifting as a substitute for both accrual management and the manipulation of real economic activities (Abernathy et al., 2014). In their study, they provide evidence that managers are likely to engage in classification shifting when real earnings management is constrained by poor financial performance, high institutional ownership and low industry market share. Thus, the flexibility of lower accounting system and an increase in analyst issuance of cash flow forecasts motivate managers to substitute accrual management to classification shifting. Overall, Abernathy et al. (2014) provide evidence that managers use classification shifting as a substitute for real earnings management and accrual
management. Finally, they compare the use of the different tools directly to determine how they change with regard to each other. The test with the use of real earnings management and accrual management indicated a significantly negative relation with classification shifting.
Haw, Ho and Li (2011) claim that classification shifting and accrual management are substitutes. They also claim that firms are more likely to shift core expenses to special items than they are to manage discretionary accruals to inflate their core earnings. According to Bagnoli and Watts (2000), firms adopt the earnings-management tool which is cheaper and harder to see through. Haw et al. (2011) adopt this evidence in their study and conclude that the use of classification shifting is the less costly mechanism.
In addition, McVay (2006) identifies some other differences between classification shifting and the other earnings-management tools. At first, GAAP earnings do not change by the use of classification shifting. Only the classification of the earnings changes, but the result does not change. Second, classification shifting will not change future earnings while accrual management and the manipulation of real activities will change future earnings. Finally, classification shifting will not change GAAP net income.
The reasoning behind earnings management
According to DeGeorge et al. (1999, p. 8), executives have three main thresholds when they report their earnings:
1. to report positive profits, that is, to report earnings that are above zero; 2. to sustain recent performance, that is, make at least last year’s earnings; and
3. to meet analysts’ expectations, particularly the analysts’ consensus earnings forecast. DeGeorge et al. (1999) examine earnings management as a whole. They claim that, of these thresholds, meeting the analysts’ expectations and the analysts’ consensus earnings forecast is the least important. More recent earnings-management studies show that this preference of the thresholds has shifted. According to Dechow et al. (2003), the most important threshold is to meet analysts’ expectations. Graham et al. (2005) and Brown and Caylor (2005) support the findings of Dechow et al. (2003). They also state that the main reason for managers to meet the expectations of analysts is to build credibility with the capital market and maintain or increase stock prices.
According to prior research (e.g., McVay, 2006; Haw et al., 2011), the main incentive for the use of classification shifting is to convey a strong operating performance to meet or beat
analysts’ earnings forecasts. According to Fairfield et al. (1996), inflating core earnings by shifting expenses results in forecast improvements and potentially increases equity values.
Managers of a firm during a crisis period consider pro forma earnings more important than earnings for reporting purposes (Graham et al., 2005). Graham et al. (2005) argue that managers use pro forma earnings to make the GAAP earnings more palatable. Entwistle et al. (2010) also suggest that pro forma earnings are the most value relevant for investors in comparison with I/B/E/S earnings and GAAP earnings. In comparison to GAAP earnings, pro forma earnings exclude special or transitory items to provide a clearer picture of a firm’s sustainable or recurring earnings (Bhattacharya, Black, Christensen and Mergenthaler, 2004). According to McVay (2006, p. 504), classification shifting can be used to increase pro forma earnings and provide managers with a relatively low-cost tool with which to meet analyst forecasts.
Poonawala and Nagar (2019) state that classification shifting is not limited to the inflation of core earnings but can also inflate the gross profit of a firm. Evidence provided by their study suggests that firms are more likely to shift cost of goods sold to R&D than to sales or general and administrative expenses. They also provide evidence which suggests that the shift from cost of goods sold to R&D rather than to sales, general and administrative expenses, was made just to meet the prior year’s gross profits. Because of this inflation of gross profits, a firm could use classification shifting to avoid violations of EBITDA-related covenants (Fan et al., 2019). This evidence is consistent with evidence provided by Malikov, Coakley and Manson (2019). Their results, provided by a UK study, suggest that the interest coverage EBITDA-based covenant is the most-used covenant, and that firms use classification shifting to avoid violations of these covenants.
Classification shifting could be less desirable in some situations. If there are no other income-decreasing special items on the income-statement, the use of classification shifting, and thereby the shift from core expenses to income-decreasing special items, could be made more detectable (Fan et al., 2019). So, the presence of these income-decreasing special items on the income statement is desired to camouflage the misclassification of the core expenses.
The financial crisis
“Economic crises are periods of sharp drops in economic activity that cause widespread changes in expectations about future economic prospects and are frequently sandwiched between periods of speculative frenzy in the economic and political markets” (Waymire & Basu, 2011).
Many causes of the financial crisis have been identified, including tax regulation over mortgage lending, a growing housing bubble, the rise of derivatives instruments such as collateralized debt obligations and questionable banking practices. According to prior research (e.g., Kothari & Lester, 2012), accounting and earnings-management methods contributed to the start of financial crisis. In addition, aside from accounting methods, auditors were also blamed for the financial crisis (Sikka, 2009). According to Sikka (2009), auditors were blamed for their lack of auditing activities. They should have foreseen the bankruptcy of firms to some extent. In addition, Chen, Krishnan and Yu (2018) have examined the earnings quality and the audit quality during the financial crisis. Although auditors were under pressure from clients and audit committees to lower their audit fees after the start of the financial crisis, neither earnings quality nor audit quality suffered significant change.
Due to rising interest rates in the years prior to 2007, a large number of house owners were no longer able to pay their mortgages. According to Kothari and Lester (2012), banks did not have sufficient capital to cover the losses and therefore were struck hard when lenders could no longer pay their mortgages. This led to major losses for firms dealing with these mortgages, including Bear Stearns, Lehman Brothers, Merrill Lynch and Citigroup (Kothari & Lester, 2012). According to Ramey (2019), macroeconomic events, like economic crises, lacked in research since the 1960s. Therefore, when the financial crisis started in 2008, firms and regulators were not sure how to proceed. The financial crisis of 2008 had a huge impact on Europe in comparison to the world. Filip and Raffournier (2014) state that the GDP growth rate in Europe fell to -4.30% whilst the GDP growth rate fell to -2.05% in the rest of the world.
Filip and Raffournier (2014) have examined the effects of the financial crisis on earnings management. They did not include classification shifting in their study but focused only on accrual management. They expected that earnings management is affected by macroeconomic events (for example, an economic crisis). In their analysis, they evaluated the income smoothing and accrual quality for 8,266 observations. According to their results, the use of earnings management, and thus accrual management, will decrease during a financial crisis. They found a decrease in income smoothing and an increase accruals quality, which could imply a decrease of earnings management. In addition, Cimini (2015) also predicts a decrease in accrual management during a financial crisis. He focused on earnings quality and the monitoring activities of Big 4 auditors during a crisis. The evidence provided by Cimini (2015) also suggests an increase in earnings quality and a decrease in accrual management after 2008. Dimitras, Kyriakou and Iatridis (2015) examine financially distressed companies during the financial crisis and measure the effect of the financial crisis on accrual management in
Europe. According to their study, some individual firms increased their earnings management to avoid bankruptcy, but their evidence suggests that the overall use of earnings management decreases. In addition, Abernathy et al. (2014) found evidence which suggests that there was more substitution between real earnings management and classification shifting in a period with poor financial conditions.
III. HYPOTHESIS DEVELOPMENT
McVay (2006) was the first to identify the use of classification shifting by managers. She concluded that managers engage in classification shifting by shifting core expenses to special items. Managers (e.g., Lougee & Marquardt, 2004) frequently exclude special items in the core expenses when reporting financial reports to market participants. This could result in meeting analyst forecasts which would not have been met if the manager included the core expenses in the special items. In this section, the examined literature is constructed into hypotheses which I use to resolve my main research question.
There are two main reasons to motivate an increase in classification shifting during a financial crisis. First, a violation of debt covenants could be a bad indicator for the investors of the firm. Earnings are included in debt covenants (Dichev & Skinner, 2002), so earnings management to increase earnings could reduce the probability of violation of those covenants (DeFond & Jiambalvo, 1994; Saleh & Ahmed, 2005; Sweeney, 1994).2 Prior research (e.g., Dichev & Skinner, 2002; Franz, HassabElnaby & Lobo, 2014) on earnings management used to avoid debt-covenant violations has focused on using accrual management or real earnings management to manipulate the balance sheet (e.g., current ratio). According to Fan et al. (2019), managers also misclassify core expenses as income-decreasing special items to increase EBITDA and thereby endeavour to avoid debt-covenant violations. In a financial crisis, it is more likely that firms would violate debt covenants than in other situations. Shifting the core expenses to income-decreasing special items will indicate stronger operating performance and signal more persistent future earnings, which would in turn motivate higher equity value (Fan et al., 2019). Consequently, the use of classification shifting could increase as firms facing unexpectedly difficult financial situations attempt to reduce the likelihood of debt violations.
2 For completeness, Dichev and Skinner (2002) state that the violation of a debt covenant are not necessarily an indication of financial distress.
Second, according to prior research, classification shifting can be used as a substitute for both accrual management and real earnings management (Abernathy et al., 2014; Haw et al, 2011) and there is evidence that firms reduce accruals and real earnings management during financial crisis (e.g., Filip and Raffournier, 2014). Specifically, Abernathy et al. (2014) provide evidence suggesting that poor financial performance decreases real earnings management and would increase the use of classification shifting. Thus, evidence suggests that lower accounting system flexibility and an increase in analyst issuance of cash flow forecasts decrease the use of accrual management and would result in an increase of classification shifting. However, accrual management constraints—like Big N auditors, long tenured auditors and regulations— would not indirectly increase classification shifting. Nevertheless, the results provide support for a substitute relationship between classification shifting and both real earnings management and accrual management. In addition, Haw et al. (2011) provide more evidence that classification shifting is a substitute for accrual management. These authors provide evidence that managers are more likely to shift core expenses to special items than they are to manage discretionary accruals to inflate core earnings, as classification shifting is a relatively low-cost mechanism for earnings management.
On the other hand, Filip and Raffournier (2014) find a significant decrease in income smoothing and an increase in accruals quality during the financial crisis. They interpret these findings as indicating that managers have reduced incentives to manipulate earnings in crisis period due to either higher market tolerance for poor performance or heightened litigation risk. The change in the behaviour of companies may respond to a higher demand for more timely earnings in troubled periods. Cimini (2015) provides corroborating evidence and contends that firms want to attract investors by providing high-quality financial reporting in financial crisis. Taken together, the foregoing arguments imply that the use of classification shifting will increase during the financial crisis, leading to the following hypothesis.
H1: The use of classification shifting will increase in the second phase (2008-2009) relative to the first phase (2006-2007).
IV. MEASURING CLASSIFICATION SHIFTING
In this section, I develop a methodology to measure classification shifting. In her study, McVay (2006) develops models which she uses to determine classification shifting. In this study, I use the same models to measure classification shifting during and after the financial crisis. If
managers use classification shifting, the core earnings are overstated in the year the special item is recognized. According to the methodology of Fan et al. (2010), Model 2 is modified to examine the effects of the financial crisis.
Regarding the measurement of predicted core earnings, I first estimate the determinants of the core earnings. The following model (Model 1) estimates these determinants by using a regression with the core earnings as the dependent variable.
(1) 𝐶𝐸𝑡 = 𝛽0+ 𝛽1𝐶𝐸𝑡−1+ 𝛽2𝐴𝑇𝑂𝑡+ 𝛽3𝐴𝐶𝐶𝑅𝑈𝐴𝐿𝑆𝑡−1+ 𝛽4𝐴𝐶𝐶𝑅𝑈𝐴𝐿𝑆𝑡+ 𝛽5∆𝑆𝐴𝐿𝐸𝑆𝑡+ 𝛽6𝑁𝐸𝐺_∆𝑆𝐴𝐿𝐸𝑆𝑡+ 𝜀𝑡
The predicted core earnings are generated by using the model of McVay (2006). First, I determine the core earnings from year t-1. These core earnings are calculated by subtracting the cost of goods sold and the selling, general and administrative expenses from the sales and dividing the result by the sales (McVay, 2006). Second, the asset turnover ratio is calculated by dividing the sales by the current net operating assets and the net operating assets from year t-1 (McVay, 2006). The net operating assets are determined by subtracting the operating liabilities from the operating assets. Operating assets are calculated as total assets less cash and short-term investments. The operating liabilities are calculated as total assets less total debt, less book value of common and preferred equity, less minority interest. Third, the accruals are calculated by subtracting the cash from operations from the net income before extraordinary items and then dividing the result by the sales (McVay, 2006). Fourth, the change in sales is determined by subtracting the sales from year t-1 from the sales and dividing the result by the sales. These amounts were multiplied by -1 with the negative change in sales. If the amount was less than zero, it was set to 0 (McVay, 2006).
After determining all the variables mentioned in Model 1, I can determine the predicted core earnings. The betas are calculated with the use of a regression, with the core earnings as the dependent variable. Following McVay (2006), each industry was identified according to the industry classification of Fama and French (1997). Missing industry data was gathered manually. After this identification, the betas were calculated by performing the regression analysis for each firm-year per industry.
Calculation of the predicted core earnings was done on a firm-by-firm level. The betas, which are calculated for each firm, for each year, are multiplied by the related variable. The predicted core earnings are calculated for each firm, for each year, by adding up the necessary variables. Once the predicted core earnings are calculated, the unexpected core earnings are
calculated by subtracting the predicted core earnings from the reported core earnings. When the reported core earnings are higher than the predicted core earnings, they are considered unexpected core earnings (McVay, 2006).
Next, the relation between the unexpected core earnings and the special items as a percentage of the sales is determined by the use of Model 2 and used to identify the use of classification shifting. The special items, as used in the definition of McVay, are material events that arise from a firm’s ongoing, continuing activities but are either unusual in nature or infrequent in occurrence and so must be disclosed as a separate line item as part of income from continuing operations or in footnotes to the financial statements (Revsine, Collins & Johnson, 2005). The special items are generated from the Compustat database, and if no data was available, the value is set to zero. This is consistent with McVay (2006) and Elliott and Hanna (1996). Because I consider only income-decreasing special items, income-increasing special items are also set to zero, consistent with McVay (2006). The coefficient which indicates the relation between the unexpected core earnings and the income-decreasing special items is generated according to the following model:
In this model, 𝑈𝐸_𝐶𝐸𝑡,𝑗 refers to the unexpected core earnings in year t for firm j; %𝑆𝐼𝑡 refers to the income-decreasing special items, divided by sales in year t.
If the coefficient on the special items as a percentage of the sales (𝛼1) is positive, it indicates that a decrease in income-decreasing special items will lead to an increase in unexpected core earnings. Therefore, it can be assumed that the income-decreasing special items make a contribution to the increase of the unexpected core earnings. Prior research also states that, if (𝛼1) is positive, then managers use classification shifting (McVay, 2006; Fan et al. 2010). In conclusion, managers shift core expenses to special items so that the core earnings will be above the expectation (Fan et al. 2010). McVay (2006, p. 517) states that, when using the models above, no further control variables are necessary.
To test the level of classification shifting during the financial crisis in comparison with the level of classification shifting before the financial crisis, I first examine the test conducted by Fan et al. (2010). Fan et al. (2010) examined the level of classification shifting in the last quarter of the year in comparison with the other quarters. They modified Model 2 of McVay by adding a dummy variable which indicates the fourth quarter. They also added a variable which indicates the special items as a percentage of the sales in case it is the fourth quarter.
To test Hypothesis 1, Model 2 is modified to include the financial crisis. In Model 3, dummy variable FC (Financial Crisis) is included. This variable values to 1 during years 2008 and 2009 of the financial crisis and to 0 for 2006 and 2007. Thus, consistent with Fan et al. (2010), the special items as percentage of the sales during the financial crisis are included in the model. SI_FC is calculated by multiplying the dummy variable with the income-decreasing items as a percentage of the sales. To test Hypothesis 1, only the years 2006-2009 are considered. Once the results are generated, the SI_FC coefficient will indicate the difference between the two periods:
In this section, I developed methods to measure the use of classification shifting and to determine the effect of the financial crisis on the use of classification shifting. In section VI, I use these measurement methods of classification shifting, which are determined in this section, to estimate the use of classification shifting.
V. DATA, SAMPLE SELECTION, AND DESCRIPTIVE STATISTICS
Data, sample selection
Data for this research is obtained for the years 2005 to 2011 from the Compustat Global - Fundamentals Annual file. Although the research consists of data for the years 2006-2009, 2005 is also included in the sample because the measurement of the predicted unexpected earnings makes use of both the core earnings (𝐶𝐸𝑡−1) and the accruals (𝐴𝐶𝐶𝑅𝑈𝐴𝐿𝑆𝑡−1) from the previous year. These core earnings are calculated by subtracting the cost of goods sold and the selling, general and administrative expenses from the sales and dividing the result by the sales (McVay, 2006). Accruals are calculated by subtracting the cash from operations from the net income before extraordinary items and then dividing the result by the sales (McVay, 2006). The exact calculations with the corresponding Compustat data item numbers are stated in Table 1.
TABLE 1:
Variable definitions with corresponding Compustat data item number
Variable Definitions
𝑪𝑬𝒕 = Core Earnings, calculated as (Sales (#12) – Cost of Goods Sold (#131) –
Selling, General, and Administrative Expenses (#132) )/Sales (#12)
𝑼𝑬_𝑪𝑬𝒕 = The Unexpected Core Earnings is the difference between the reported core
earnings and the predicted Core Earnings. The predicted Core Earnings are calculated according to the following model:
𝐶𝐸𝑡 = 𝛽0+ 𝛽1𝐶𝐸𝑡−1+ 𝛽2𝐴𝑇𝑂𝑡+ 𝛽3𝐴𝐶𝐶𝑅𝑈𝐴𝐿𝑆𝑡−1+ 𝛽4𝐴𝐶𝐶𝑅𝑈𝐴𝐿𝑆𝑡
+ 𝛽5∆𝑆𝐴𝐿𝐸𝑆𝑡+ 𝛽6𝑁𝐸𝐺_∆𝑆𝐴𝐿𝐸𝑆𝑡+ 𝜀𝑡
%𝑺𝑰𝒕 = Special items (#17) as a percentage of the sales (#12). Following McVay
(2006), special items (#17) are multiplied by -1 and the values below zero are set to zero. By doing this only the income-decreasing items remain and the income-increasing items are set to zero and thus eliminated.
𝑨𝑪𝑪𝑹𝑼𝑨𝑳𝑺𝒕 = Accruals, calculated as net income before extraordinary items (#123) minus
cash from operations (#308-#124) / sales (#12).
𝑨𝑻𝑶𝒕 = Asset Turnover Ratio, calculated as 𝑆𝐴𝐿𝐸𝑆𝑡/(𝑁𝑂𝐴𝑡+𝑁𝑂𝐴𝑡−1
2 ), where net
operating assets is operating assets – operating liabilities. Operating assets are total assets (#6) minus cash and short-term investments (#1). Operating liabilities are the total assets (#6) minus total debt (#9 and #34), less book value of common and preferred equity (#60 and #130), less minority interest (#38).
∆𝑺𝑨𝑳𝑬𝑺𝒕 = Percentage change in sales, calculated as
(𝑆𝐴𝐿𝐸𝑆𝑡(#12) − 𝑆𝐴𝐿𝐸𝑆𝑡−1)/𝑆𝐴𝐿𝐸𝑆𝑡−1
𝑵𝑬𝑮∆𝑺𝑨𝑳𝑬𝑺𝒕 = ∆𝑆𝐴𝐿𝐸𝑆𝑡 if percentage change in sales is less than 0, otherwise 0.
FC = A dummy variable which indicates the financial crisis period. The variable is
1 if the observation is in the years 2008-2009 and 0 otherwise.
HL = A dummy variable which indicates a high leveraged firm. The leverage is
calculated by dividing the total debt by the total assets. The variables values one if the calculated value is above the median of HL and zero otherwise.
Distress = A dummy variable which indicates if the firm reports a negative net income.
The variable values one if the firm reported negative net income and zero otherwise.
SI×FC = Change in the use of classification shifting as a result of the financial crisis.
TABLE 1 (Continued)
SI×HL = Variable indicates the interrelated relation between the use of classification
shifting and the fact if the firm is a high leverage firm. The variable is calculated by multiplying %SI with the HL dummy variable.
SI×Distress = Variable indicates the interrelated relation between the use of classification
shifting and the fact if the firm is in financial distress. The variable is calculated by multiplying %SI with the Distress dummy variable.
SI×FC×HL = Variable indicates the interrelated relation between the use of classification
shifting, the financial crisis and the fact that the firm is a high leverage firm. The variable is calculated by multiplying %SI with FC and HL.
SI×FC×Distress = Variable indicates the interrelated relation between the use of classification
shifting, the financial crisis and the fact that the firm is in financial distress. The variable is calculated by multiplying %SI with FC and Distress.
Control variables
Size = A dummy variable which values one if the firm sales are included in the upper
50 percent of the sample.
BIG4 = A dummy variable which indicates if the firm hires a Big4 audit firm or
another firm. The variable values one if the firm hires a Big4 audit firm and zero otherwise.
ROA = Return on assets defined as the ratio of income before extraordinary items to
total assets (Christensen & Nikolaev, 2012).
R&D = R&D expense divided by total revenue. Missing R&D expense is replaced
with zero (Christensen & Nikolaev, 2012).
From the Compustat database, a sample of 179,428 firm years was generated. In this research I focus on the European market, so all 140,074 non-EU firm-year observations were excluded from the sample. Following McVay (2006), I removed 3,680 firm-year observations with less than one million in net sales to keep major outliers at a minimum and 3,099 firm-year observations with a changing fiscal-year-end to ensure the years are comparable. Because the firms are compared with prior years, there need to be at least 4 firm-years of each firm in the sample, therefore 2,702 firm-year observations were excluded from the sample.
Finally, some firms had more than one observation per year, no sales, or there was missing information regarding the calculation of the core earnings. This could lead to biases when comparing years, so these 1,426 firms-year observations were excluded as well. Because
some missing data in the net operating assets, 26 asset turnover ratios are divided by zero and therefore excluded from the sample, when the asset turnover ratio is included.
To be able to calculate all the values in 2006, 2005 was included in the sample. This was necessary to determine the 𝐴𝐶𝐶𝑅𝑈𝐴𝐿𝑆𝑡−1, 𝐶𝐸𝑡−1, and the ∆𝑆𝐴𝐿𝐸𝑆𝑡 . After the calculations of these variables for the year 2006, 2005 was excluded from the sample.
The years 2010-2011 were included in the sample because they were part of a later excluded test. Once the test was excluded, the data was excluded as well.
Items in data-set 179,428
Non-EU Members 140,074
Sales below 1 Million 3,680
Change in fiscal-year-end 3,099
Three or less firm-years in sample 2,702
Year 2005 4,450
Other 1,426
Years 2010-2011 7,698
Total 16,299
Descriptive statistics
Table 2 provides the descriptive statistics of the whole sample for the main variables. The core earnings are scaled by sales and have a mean of 0.134 and a median of 0.128. The income-decreasing items, scaled by the sales, contain a mean of 1.6 percent. Because there are no income-increasing items in this study, so in the dataset the values above 0 are set to zero. To eliminate outliers, which could influence the results, all variables are winsorized at the first and the ninety-ninth percentile (McVay, 2006). The descriptive statistics are in line with the statistics calculated by McVay (2006) in her paper.
TABLE 2:
Descriptive Statistics: 2006-2009
Variable Mean Median Standard Deviation 25% 75% 𝑺𝑨𝑳𝑬𝑺𝒕 2,183.781 144.192 7,905.418 32.603 731.985 𝑷𝑬𝑹𝑪𝑬𝑵𝑻 𝑪𝑯𝑨𝑵𝑮𝑬 𝑰𝑵 𝑺𝑨𝑳𝑬𝑺𝒕−𝟏,𝒕 14.2% 7.2% 0.435 -4.8% 22.2% 𝑵𝑬𝑮𝑨𝑻𝑰𝑽𝑬 𝑪𝑯𝑨𝑵𝑮𝑬 𝑰𝑵 𝑺𝑨𝑳𝑬𝑺𝒕−𝟏,𝒕 -0.058 0.000 0.126 -0.048 0.000 𝑪𝑶𝑹𝑬 𝑬𝑨𝑹𝑵𝑰𝑵𝑮𝑺𝒕 0.133 0.129 0.258 0.058 0.236 𝑼𝑵𝑬𝑿𝑷𝑬𝑪𝑻𝑬𝑫 𝑪𝑶𝑹𝑬 𝑬𝑨𝑹𝑵𝑰𝑵𝑮𝑺𝒕 0.000 -0.000 0.102 -0.035 0.039
INCOME-DECREASING SPECIAL ITEMSt 14.505 0.000 72.361 0.000 0.805
𝑰𝑵𝑪𝑶𝑴𝑬 − 𝑫𝑬𝑪𝑹𝑬𝑨𝑺𝑰𝑵𝑮 𝑺𝑷𝑬𝑪𝑰𝑨𝑳 𝑰𝑻𝑬𝑴𝑺 𝑨𝑺 𝑨 𝑷𝑬𝑹𝑪𝑬𝑵𝑻 𝑶𝑭 𝑺𝑨𝑳𝑬𝑺 t
1.5% 0.0% 0.056 0.0% 0.5%
𝑨𝑪𝑪𝑹𝑼𝑨𝑳𝑺𝒕𝒂 -0.071 -0.038 0.244 -0.106 0.011
𝑨𝑺𝑺𝑬𝑻 𝑻𝑼𝑹𝑵𝑶𝑽𝑬𝑹 𝑹𝑨𝑻𝑰𝑶𝒕 2.629 1.778 6.933 0.948 3.196
There is a maximum of 16,299 firm-year observations. Variables are defined in Table 1. All variables are winsorized at the 1st and 99th percentile (McVay, 2006).
As mentioned in previous sections, this study will include two phases to evaluate the changes in the use of classification shifting. Table 3 provides the descriptive statistics of these phases. The amount of income-decreasing special items as a percentage of the sales doubles after the start of the financial crisis, phase 2, from 0.9% to 2.1%. The core earnings decrease after the start of the financial crisis with a mean (median) of 0.144 (0.135) to 0.123 (0.122). The mean of the unexpected core earnings on the other hand increase during the financial crisis from -0.000 to 0.001, which indicates that the reported core earnings were higher than the predicted core earnings, compared to the first phase.
Table 4 represents the correlations between the different variables used in this research. In this table, the correlations between the asset turnover ratio and the sales and the asset turnover ratio and the unexpected core earnings are not significant. This is in accordance with McVay (2006). The correlation between the special items as a percentage of the sales and the unexpected core earnings is not significant, this could be because of the sample size, because if more years (2010-2011) are included in the sample this correlation is significant.
TABLE 3
Descriptive Statistics for each Phase Phase 1
2006-2007
Phase 2 2008-2009
Variable Mean Median Mean Median
𝑆𝐴𝐿𝐸𝑆𝑡 2,174.150 139.516 2,192.467 148.789
𝑃𝐸𝑅𝐶𝐸𝑁𝑇 𝐶𝐻𝐴𝑁𝐺𝐸 𝐼𝑁 𝑆𝐴𝐿𝐸𝑆𝑡−1,𝑡 22.7% 12.1% 6.5% 1.7%
𝑵𝑬𝑮𝑨𝑻𝑰𝑽𝑬 𝑪𝑯𝑨𝑵𝑮𝑬 𝑰𝑵 𝑺𝑨𝑳𝑬𝑺𝒕−𝟏,𝒕 -0.032 0.000 -0.081 0.000
𝐶𝑂𝑅𝐸 𝐸𝐴𝑅𝑁𝐼𝑁𝐺𝑆𝑡 0.144 0.135 0.123 0.122
𝑈𝑁𝐸𝑋𝑃𝐸𝐶𝑇𝐸𝐷 𝐶𝑂𝑅𝐸 𝐸𝐴𝑅𝑁𝐼𝑁𝐺𝑆𝑡 -0.000 -0.000 0.001 0.000
INCOME-DECREASING SPECIAL ITEMSt 7.879 0.000 20.481 0.000
𝐼𝑁𝐶𝑂𝑀𝐸 − 𝐷𝐸𝐶𝑅𝐸𝐴𝑆𝐼𝑁𝐺 𝑆𝑃𝐸𝐶𝐼𝐴𝐿 𝐼𝑇𝐸𝑀𝑆 𝐴𝑆 𝐴 𝑃𝐸𝑅𝐶𝐸𝑁𝑇 𝑂𝐹 𝑆𝐴𝐿𝐸𝑆t
0.9% 0.0% 2.1% 0.0%
𝐴𝐶𝐶𝑅𝑈𝐴𝐿𝑆𝑡𝑎 -0.034 -0.021 -0.104 -0.055
𝐴𝑆𝑆𝐸𝑇 𝑇𝑈𝑅𝑁𝑂𝑉𝐸𝑅 𝑅𝐴𝑇𝐼𝑂𝑡 2.795 1.916 2.480 1.640
There is a maximum of 16,299 firm-year observations. Variables are defined in Table 1. All variables are winsorized at the 1st and 99th percentile (McVay, 2006). In phase 1 the maximum number of firm-year observations is 7,729 and in phase 2 the maximum number of firm-year observations is 8,570.
TABLE 4:
Pearson Correlation Matrix
Variable 𝑆𝐴𝐿𝐸𝑆𝑡 ∆𝑆𝐴𝐿𝐸𝑆𝑡 𝐶𝐸𝑡 𝐶𝐸𝑡−1 𝑈𝐸_𝐶𝐸𝑡 %𝑆𝐼𝑡 𝐴𝐶𝐶𝑅𝑈𝐴𝐿𝑆𝑡 𝐴𝑇𝑂𝑡 𝑺𝑨𝑳𝑬𝑺𝒕 1.0000 ∆𝑺𝑨𝑳𝑬𝑺𝒕 -0.0345 (0.0000) 1.0000 𝑪𝑬𝒕 0.0638 (0.0000) 0.0680 (0.0000) 1.0000 𝑪𝑬𝒕−𝟏 0.0606 (0.0000) -0.1201 (0.0000) 0.8014 (0.0000) 1.0000 𝑼𝑬_𝑪𝑬𝒕 0.0172 (0.0281) -0.0305 (0.0001) 0.4677 (0.0000) 0.0340 (0.0000) 1.0000 %𝑺𝑰𝒕 -0.0272 (0.0005) -0.0736 (0.0000) -0.1430 (0.0000) -0.0780 (0.0000) 0.0047 (0.5527) 1.0000 𝑨𝑪𝑪𝑹𝑼𝑨𝑳𝑺𝒕 0.0129 (0.0985) 0.1178 (0.0000) 0.2368 (0.0000) 0.1226 (0.0000) -0.0135 (0.0843) -0.4076 (0.0000) 1.0000 𝑨𝑻𝑶𝒕 0.0001 (0.9938) 0.0409 (0.0000) -0.0164 (0.0359) -0.0303 (0.0001) -0.0007 (0.9322) -0.0385 (0.0000) 0.0610 (0.0000) 1.0000
All bold amounts are at least significant at the 0.1 level. There is a maximum of 16,299 firm-year observations. Variables are defined in Table 1. All variables are winsorized at the 1st and 99th
percentile. Amounts reported are the Pearson correlations and the significance levels in parentheses.
Coefficients regarding classification shifting
Table 5 shows the coefficients of each beta and the predicted sign, calculated according to Model 1 of McVay (2006). The core earnings of the previous year have a coefficient of 0.7910 and is significant at the level of 0.001. Like previous research (McVay, 2006) asset turnover ratio has a small impact on the core earnings with a significant coefficient of -0.0004. As predicted, the lagged accruals have a negative beta coefficient of -0.1079, because the accruals reverse in the next year and have a negative impact on the core earnings. The current accruals have the opposite effect with a beta coefficient of 0.1203. All variables, except the asset turnover ratio, are significant at a level of 0.001. The asset turnover ratio is significant at the level of 0.01. Concluding, all variables are significant regarding the core earnings.
TABLE 5:
Model of Predicted Core Earnings
Dependent Variable: 𝑪𝑬𝒕
Independent Variables
Predicted Sign Coefficient
Intercept 0.0358 (23.12)*** 𝑪𝑬𝒕−𝟏 + 0.7868 (198.33)*** 𝑨𝑻𝑶𝒕 - -0.0004 (-2.69)** 𝑨𝑪𝑪𝑹𝑼𝑨𝑳𝑺𝒕−𝟏 - -0.1032 (-22.24)*** 𝑨𝑪𝑪𝑹𝑼𝑨𝑳𝑺𝒕 + 0.1236 (28.27)*** ∆𝑺𝑨𝑳𝑬𝑺𝒕 + 0.0611 (22.02)*** 𝑵𝑬𝑮_∆𝑺𝑨𝑳𝑬𝑺𝒕 + 0.2396 (25.51)*** 𝑹𝟐 69.43%
*, **, *** Indicates significant at the 0.05, 0.01 and 0.001 levels, respectively. There is a maximum of 16,299 firm-year observations. All variables are defined in Table 1. All variables are winsorized at the 1st and 99th percentile. The reported amounts in the table are the regression coefficients (t-statistics).
VI. RESEARCH DESIGN AND RESULTS
In this section I develop my research design to be able to compare the different phases. In addition I will test the hypotheses to examine the change in the use of classification shifting during the sample period. In addition, to strengthen the results, some extra control variables will be implemented to identify the effect of these variables on classification shifting.
Identifying the classification shifting
To determine the change in the use of classification shifting during the financial crisis, I have to identify the classification shifting and compare the use of classification shifting before
the financial crisis with the use of classification shifting during the financial crisis. As stated in the previous section, classification shifting will be identified by using the following model:
In Model 3 %𝑆𝐼𝑡 represents the special items as a percentage of the sales. The special items are multiplied by -1, and the negative, income-increasing, items were set to zero. Estimated is that the coefficient is more positive (less negative) in phase 2 and therefore it could be assumed that the use of classification shifting increases (McVay, 2006; Fan et al. 2010). If the estimated coefficient is more negative (less positive), the firm performance is the more dominating effect (Fan et al. 2010). In this situation, an increase in the income-decreasing special items would result in a decrease of the unexpected core earnings. Previous research also claimed that firms incurring large write-offs or corporate restructuring charges are often poor performing firms (Elliot and Shaw, 1998; DeAngelo et al. 1994).
In Model 3, FC is a dummy variable which indicates the financial crisis. The years 2008 and 2009 the value is set to one and during the other years the value is set to zero. The next variable, SI×FC is calculated by multiplying the dummy variable with the income-decreasing items as a percentage of the sales. The result of this variable indicates the difference between the phases. I will only use Model 3 to test Hypothesis 1, so it will only consider the years 2006-2009.
To measure the impact of the financial crisis on the level of classification shifting, I examine the level and sign of the use of classification shifting, as presented in Table 6. The total sample will contain the years 2006-2009. This was the total sample which was generated from the Compustat Database. Thereby, phase one contains the years prior to the financial crisis (2006-2007), the second phase contains the years during the financial crisis (2008-2009). The years 2008 is chosen as the start of the financial crisis, because from there the financial crisis started in Europe. In Table 6, the use of classification shifting is determined according to Model 2 of McVay (2006). The modified Model 3, is presented in Table 7.
I use the Model 2 to determine the use of classification shifting during the different phases. Thereafter, the control group (phase 1) will be used to examine the developments during the start of the crisis (phase 2).
TABLE 6:
Regression of Unexpected Core Earnings on Special Items as a Percentage of Sales Dependent Variable: 𝑼𝑬_𝑪𝑬𝒕
Total Sample Phase 1 Phase 2
Intercept 0.000 (0.32) 0.000 (0.02) 0.001 (0.51) %𝑺𝑰𝒕 0.009 (0.59) -0.017 (-0.50) 0.012 (0.75) 𝑹𝟐 0.01% 0.01% 0.01% N 16,299 7,729 8,570
*, **, *** Indicates significant at the 0.05, 0.01 and 0.001 levels, respectively. In phase 1 the maximum number of firm-year observations is 7,729, in phase 2 the maximum number of firm-year observations is 8,570. All variables are defined in Table 1. The Unexpected Core Earnings (UE_CE) is the difference between the reported core earnings and the predicted Core Earnings. The predicted Core Earnings are calculated according to the Model 1, estimated by industry and fiscal year: 𝐶𝐸𝑡 = 𝛽0+ 𝛽1𝐶𝐸𝑡−1+ 𝛽2𝐴𝑇𝑂𝑡+ 𝛽3𝐴𝐶𝐶𝑅𝑈𝐴𝐿𝑆𝑡−1+ 𝛽4𝐴𝐶𝐶𝑅𝑈𝐴𝐿𝑆𝑡+ 𝛽5∆𝑆𝐴𝐿𝐸𝑆𝑡+
𝛽6𝑁𝐸𝐺_∆𝑆𝐴𝐿𝐸𝑆𝑡+ 𝜀𝑡.
Special items as a percentage of the sales (%SI). Income-decreasing items are multiplied by -1 and values below zero are set to zero to eliminate the income-increasing items.
All variables are winsorized at the 1st and 99th percentile. The reported amounts in the table are the regression coefficients (t-statistics).
Table 6 presents the outcomes of the regression of the unexpected core earnings on the special items as a percentage of the sales. These statistics were generated by using a regression with 𝑈𝐸_𝐶𝐸𝑡 as the dependent variable and %𝑆𝐼𝑡as the independent variable. In the first column the total sample is represented. Here the coefficient (t-statistic) of the %𝑆𝐼𝑡 is 0.009 (0.59). The coefficient is positive, but not significant. For this study, the following columns are more relevant because these show the coefficients of the different phases. In the period before the financial crisis (phase 1) the coefficient is -0.017, the negative sign suggests that the firm performance has a more dominating effect. After the start of the financial crisis (phase 2) a change is identified since the coefficient becomes positive (0.012), so in this phase the use of classification shifting might increase compared to the first phase. No conclusions can be drawn from this, because the coefficients are not significant.
To find direct evidence of classification shifting, one would find a positive relation between UE_CE and %SI, consistent with McVay (2006). The first and second phase have no
significant coefficient, this can be due to the small sample size. To indicate the effect of a larger sample I also performed the test including the years 2010 and 2011. This test provides a positive significant coefficient of 0.022 and is in accordance with the study of McVay (2006).
Hypothesis testing
The hypothesis stated that the use of classification shifting would increase after the start of the financial crisis in 2008. In Table 6 no significant coefficient was identified. To compare the first and second phase, I use Model 3 to identify the effect of the financial crisis and thereby be able to compare the two periods. Table 7 indicates the results of Model 3.
Table 7 tests H1 according to Model 3. The test illustrates a coefficient of 0.030 for SI×FC. This is the difference between %SI of phase 1 and phase 2. A significant increase in the coefficient would imply an increase in the use of classification shifting. Nevertheless, the increase is not significant, so no conclusions can be drawn from the results generated from Model 3.
To test this hypothesis further, I include control variables which could also influence the use of classification shifting and thereby generated Model 4.
First, I include the size of the firm to identify the influence of the size of the firm on the use of classification shifting. The size is scaled by the sales of the firm because all prior calculations were also scaled by the sales to make them comparable. Size is a dummy variable which values one if it considers a big firm and is included in the upper 50 percent of the sample and values zero otherwise. Second, consistent with Fan et al. (2019), I include the BIG4 as a control variable in Model 4. BIG4 equals 1 if the firm hired a Big4-auditor and values zero otherwise. Third, return on assets will be included as a control variable. The variable is defined as the ratio of income before extraordinary items to total assets (Christensen & Nikolaev, 2012). Fourth, consistent with Christensen and Nikolaev (2012), research and development expenses divided by the sales will also be included as a control variable. Table 7 provides the results of Model 4. Once the control variables are included in the model, results provide evidence suggesting that firms make use of classifications shifting because %SI has a positive significant coefficient. Nevertheless, the interrelated variable SI×FC does not provide a significant coefficient so no evidence can be drawn.
(4) 𝑈𝐸_𝐶𝐸𝑡,𝑗= 𝛼0+ 𝛼1%𝑆𝐼𝑡+ 𝛼2𝐹𝐶 + 𝛼3𝑆𝐼 × 𝐹𝐶 + 𝛼4𝑆𝑖𝑧𝑒 + 𝛼5𝐵𝐼𝐺4 + 𝛼6𝑅𝑂𝐴 + 𝛼7𝑅&𝐷 + 𝜀𝑡
Table 7:
Regression of Unexpected Core Earnings on Special Items as a Percentage of Sales 2006-2009 Dependent Variable: 𝑼𝑬_𝑪𝑬𝒕 Test H1: Model 3 Test H1: Model 4 Intercept 0.000 (0.02) -0.010 (-6.31)*** %𝑺𝑰𝒕 -0.017 (-0.48) 0.224 (6.13)*** FC 0.001 (0.35) 0.005 (3.22)*** SI×FC 0.030 (0.75) -0.056 (-1.44) Size -0.002 (-1.09) BIG4 0.006 (3.66)*** ROA 0.168 (25.58)*** R&D 0.028 (3.79)*** 𝑹𝟐 0.01% 4.22% Number of observations 16.299 16,299
*, **, *** Indicates significant at the 0.05, 0.01 and 0.001 levels, respectively. The maximum number of firm-year observations is 16,299. Variable definitions are specified in Table 1. All variables are winsorized at the 1st and 99th percentile. The reported amounts in the table are the regression coefficients (t-statistics).
The first test is done according Model 3: 𝑈𝐸_𝐶𝐸𝑡,𝑗 = 𝛼0+ 𝛼1%𝑆𝐼𝑡+ 𝛼2𝐹𝐶 + 𝛼3𝑆𝐼 × 𝐹𝐶 + 𝜀𝑡
The second test is done according Model 4: 𝑈𝐸_𝐶𝐸𝑡,𝑗= 𝛼0+ 𝛼1%𝑆𝐼𝑡+ 𝛼2𝐹𝐶 + 𝛼3𝑆𝐼 × 𝐹𝐶 +
𝛼4𝑆𝑖𝑧𝑒 + 𝛼5𝐵𝐼𝐺4 + 𝛼6𝑅𝑂𝐴 + 𝛼7𝑅&𝐷 + 𝜀𝑡
According to Fan et al. (2019), managers also misclassify core expenses as income-decreasing special items to increase EBITDA and thereby endeavour to avoid debt-covenant violations. If firms report a negative net income or have high leverage, it is more likely that firms would violate debt covenants than in other situations. Graham et al. (2005) claim in their study that highly leveraged and unprofitable firms are closer to debt-covenant violations. Therefore I generate 2 extra models. One including high leverage (HL) and one including negative net income (Distress).
The leverage is calculated by dividing total debt by the total assets (Haw et al., 2010). The total debt and assets were defined by McVay (2006), so I will use the same definition in this variable calculation. Firms with leverage above the median are categorized as high leverage firms and firms with leverage below the median are categorized as low leverage firms. Therefore HL values one if leverage is above the median and zero otherwise.
Distress indicates if a firm reports a negative net income and thus could be in financial distress. The variable values one if the firm reports a negative net income and zero otherwise. Table 8 provides the results of both Model 5 and Model 6.
Table 8 presents evidence for the use of classification shifting in both models. Regarding Model 5, no evidence is presented suggesting that HL affects classification shifting because SI×HL is not significant. Thereby, SI×FC×HL is not significant, indicating that the inclusion of the financial crisis has no effect neither.
In contrast with the estimation based on the papers of Dimitras et al. (2015) and Abernathy et al. (2014), financially distressed firms are less eager to engage in classification shifting. The results provide evidence that the use of classification shifting decreases for firms in financial distress. Due to the fact that Dimistras et al. (2015) found evidence that accrual management decreases during financial distress and classification shifting increases due to the substitution effect when accrual management decreases (Abernathy et al., 2014), this should be further examined in future research. When the effect of the financial crisis is included, there
(5) 𝑈𝐸_𝐶𝐸𝑡,𝑗= 𝛼0+ 𝛼1%𝑆𝐼𝑡+ 𝛼2𝐹𝐶 + 𝛼3𝐻𝐿 + 𝛼4𝑆𝐼 × 𝐹𝐶 + 𝛼5𝑆𝐼 × 𝐻𝐿
+ α6𝑆𝐼 × 𝐹𝐶 × 𝐻𝐿 + 𝛼4𝑆𝑖𝑧𝑒 + 𝛼5𝐵𝐼𝐺4 + 𝛼6𝑅𝑂𝐴 + 𝛼7𝑅&𝐷 + 𝜀𝑡 (6) 𝑈𝐸_𝐶𝐸𝑡,𝑗= 𝛼0+ 𝛼1%𝑆𝐼𝑡+ 𝛼2𝐹𝐶 + 𝛼3𝐷𝑖𝑠𝑡𝑟𝑒𝑠𝑠 + 𝛼4𝑆𝐼 × 𝐹𝐶
+ 𝛼5𝑆𝐼 × 𝐷𝑖𝑠𝑡𝑟𝑒𝑠𝑠 + α6𝑆𝐼 × 𝐹𝐶 × 𝐷𝑖𝑠𝑡𝑟𝑒𝑠𝑠 + 𝛼4𝑆𝑖𝑧𝑒 + 𝛼5𝐵𝐼𝐺4 + 𝛼6𝑅𝑂𝐴 + 𝛼7𝑅&𝐷 + 𝜀𝑡
Table 8:
Regression of Unexpected Core Earnings on Special Items as a Percentage of Sales, including multiple control variables.
2006-2009 Dependent Variable: 𝑼𝑬_𝑪𝑬𝒕 Model 5 Model 6 Intercept -0.013 (-7.69)*** -0.010 (-6.15)*** %𝑺𝑰𝒕 0.206 (4.32)*** 0.610 (8.25)*** FC 0.005 (3.13)** 0.006 (3.40)*** HL 0.008 (4.76)*** Distress -0.000 (-0.18) SI×FC -0.048 (-0.93) -0.343 (-3.99)*** SI×HL 0.047 (0.67) SI×Distress -0.500 (-6.00)*** SI×FC×HL -0.026 (-0.34) SI×FC×Distress 0.384 (4.07)*** Size -0.004 (-2.04)* -0.002 (-1.25) BIG4 0.006 (3.40)*** 0.006 (3.70)*** ROA 0.173 (26.06) 0.162 (21.11)*** R&D 0.034 (4.55)*** 0.028 (3.80)*** 𝑹𝟐 4.39% 4.47% Number of observations 16,295 16,295
