Msc Accountancy & Control, variant Accountancy Faculty of Economics and Business, University of Amsterdam
Earnings Management around Financial
Performance Thresholds
A comparison of earnings management patterns across three different
earnings thresholds
Petru-Cătălin David (10824618) June 22, 2015
Supervisor: Dr Réka Felleg Word count: 13,253
2 Statement of Originality
This document is written by student Petru-Cătălin David who declares to take full responsibility for the contents of this document.
I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.
The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
3 Table of Contents
1. Introduction ... 5
2. Literature review and hypotheses development ... 6
2.1 Earnings Management ... 6
2.2 The enactment of the Sarbanes-Oxley Act (2002) ... 9
2.3 Earnings Benchmarks ... 11
2.4 Hypothesis development... 13
3. Research Design ... 16
3.1 Sources of data... 16
3.2 Description of the regression model used to test hypotheses ... 17
3.3 Methodology used for estimating accrual-based earnings management ... 18
3.4 Methodology used for estimating real earnings management ... 19
4. Descriptive Statistics ... 21
4.1 Firm characteristics ... 21
4.2 Earnings distribution around financial performance thresholds ... 23
5. Results ... 27
5.1 Estimation models ... 27
5.1.1 The zero-earnings threshold ... 28
5.1.2 The analysts' mean EPS estimate benchmark ... 29
5.1.3 The previous year’s reported EPS benchmark ... 31
5.2 Comparing earnings management patterns across financial performance thresholds 32 5.3 Robustness test ... 34
5.4 Supplementary analysis ... 35
6. Conclusions ... 36
Appendices ... 38
4
Abstract
The existing literature on earnings management across financial performance thresholds is fragmented in multiple studies, none of which having the main goal of comparing earnings management patterns across different thresholds. I seek to fill out this gap and based on a unitary sample of firm-year observations that extends between 1987 and 2014, I compare earnings management patterns across three financial performance thresholds: zero-earnings, the analysts' mean EPS estimate and the previous year’s reported EPS.
In spite of some statistical limitations, my study brings a contribution to financial accounting literature by delivering new insights related to the companies’ financial performance motivations for engaging in earnings management. First, I do not find any similarities in earnings management patterns across thresholds and moreover, in opposition to previous studies, I do not identify any significant shift from accrual-based earnings management to real earnings management in the Post-SOX period. Still, I find some evidence that the reduction in discretionary expenses represents a common mechanism to meet thresholds in all three instances. Second, by considering each threshold on an individual basis, I classify zero-earnings as the threshold which induces the highest extent of earnings manipulation. On the other hand, although surprising when taking into account previous research, I do not find any evidence that firms which meet or barely exceed analysts’ EPS benchmarks engage in earning management.
Keywords: accrual-based earnings management, real earnings management, financial performance thresholds, SOX, earnings manipulation
5
1. Introduction
The objective of my study is to provide empirical evidence regarding the financial performance motivations of companies for engaging in earnings management. The firm’s main capital providers, shareholders and lenders, set financial performance thresholds that are strongly related to accounting data which in many situations is at the discretion of the management. Prior research shows that the distribution of earnings is discontinued in two instances – at the zero-earnings threshold (Burgstahler and Dichev, 1997) and at the analysts’ mean EPS estimate benchmark (Burgstahler and Eames, 2006). Starting from this apparent anomaly, authors use empirical methods and demonstrate that companies manipulate earnings to avoid small losses and negative earnings surprises so consequently it can be inferred that an important reason for managing earnings is to avoid missing performance thresholds set by capital providers. In addition to these two financial performance thresholds, the survey data gathered by Graham (2005) reveals the previous period’s earnings as an equally important threshold in the view of chief financial officers (CFOs).
The significant academic evidence showing that companies manipulate earnings to meet stakeholders’ performance thresholds and the large financial reporting scandals from early 2000s have triggered a reaction from US regulators in the form of the Sarbanes-Oxley Act (2002). Expectations were that this new regulation would put an end to earnings management but as subsequent literature reveals, even though the prevalence of accrual-based earnings management decreased in the Post-SOX era, real earnings management has emerged as a substitute which may prove to have even increased negative consequences for firm value in the long term (Bartov and Cohen, 2008). Surveys show that managers acknowledge real earnings management as an important drag on firm value because it encourages short-termism and results in missing numerous long-term investment opportunities (Graham et al., 2005).
Existing theory concerning earnings management is based on separate studies, each of them emphasizing one particular earnings threshold at a time. For instance, Roychowdhurry (2006) concentrates on real earnings management around the zero-earnings threshold while Burgstahler and Dichev (1997) focus on accrual earnings management around the analysts’ mean EPS estimate benchmark. While most academics are preoccupied of studying how the mechanisms used to manipulate earnings have evolved over time in reaction to the Sarbanes-Oxley Act (SOX), a literature gap remains due to the lack of studies which compare the patterns of earnings
6
management across different financial performance thresholds. Consequently, I am motivated to fill out this gap through the current study. I emphasize the zero-earnings threshold and the analysts’ mean EPS estimate benchmark because these have been associated by academics with the performance thresholds set by capital providers. Nevertheless, I also include in my analysis the previous period’s earnings threshold because of its high importance in the view of executive management.
Based on reviewing previous literature, I expect that, after controlling for the Pre- and Post-SOX regulatory regime, firms suspected to have managed earnings display characteristics such as higher discretionary accruals, lower abnormal operating cash flow, lower abnormal discretionary expenses and higher abnormal production costs relative to non-suspect firms. Nevertheless, it is the comparison of the earnings manipulation patterns displayed by suspect firms relative to each of the three performance thresholds that represents my original contribution to the financial accounting literature on earnings management.
There are five additional sections that complete my study. The second section represents a literature review of the relevant topics and more details on how the hypotheses were developed. The third section explains the research design and methodology whereas the fourth presents a set of descriptive statistics of the sample considered for testing the hypotheses. The last two sections reveal the study’s findings and the conclusions, respectively.
2. Literature review and hypotheses development
2.1 Earnings Management
Over the last decades, a large body of academic literature has been centered on earnings management which became one of the most recurring topics in financial accounting research. According to Verbruggen et al. (2008), only between 2000 and 2006 there have been published over 150 academic articles focusing on or relating to earnings management. As the authors note, this indicates that there is a continuous interest in this field which periodically translates in new research topics such as the emergence of studies on real earnings management in the post Sarbanes-Oxley Act (SOX) era.
The concept of earnings management has been defined by Healy and Wahlen (1999) as following: “Earnings management occurs when managers use judgment in financial reporting and in structuring transactions to alter financial reports to either mislead some stakeholders about the
7
underlying economic performance of the company or to influence contractual outcomes that depend on reported accounting numbers.” Once having defined the general concept of earnings management, it is important to note that it actually encompasses two different mechanisms for managing earnings: accrual-based earnings management and real earnings management.
Accruals arise based on the fundamental accounting concept of matching expenditure or income to the accounting period in which these are incurred rather than paid. For example, if a supplier provides a service in the current period but does not send an invoice before the closing of the current period, an accrued expense is required according to the fundamental accounting principles. Due to this event net income and operating cash flow for the current period will differ. In fact, according to Van Praag (2001) net income can be seen as the adjustment of the operational cash flow for transitory components resulting in net income from operations. It is often assumed that accruals are open to more discretion than cash flows and that total accruals have two components, discretionary and non-discretionary accruals. The situation in which management exercises discretion on accruals without having any economic reason to do so, represents the traditional case of accrual-based earnings management (AEM). Studies on AEM have been abundant since the 1980s and have covered a vast area of topics such the measurement of discretionary accruals, the motives for engaging in AEM, the degree of investor understanding of AEM or the changes in AEM in the Post-SOX period.
Compared to AEM, the concept of real earnings management (REM) has only recently become a frequent topic in financial accounting research and its occurrence has often been associated with the enactment of SOX. In one of the first papers focused exclusively on REM, Roychowdhury (2006) defines real activities manipulation as departures from normal operational practices, motivated by management’s desire to mislead at least some stakeholders into believing certain financial reporting goals have been met in the normal course of operations. He notes that these departures do not necessarily contribute to firm value even though they enable managers to meet reporting goals.
Existing literature distinguishes between three main mechanisms of REM: sales manipulation, overproduction and the abnormal reduction of discretionary expenses (e.g. R&D, sales, general and administrative expenses). Sales manipulation is defined as the practice of stuffing sales at the end of the reporting period by offering additional discounts and more lenient credit terms to clients relative to other periods of the year. While this practice increases the
top-8
line of the income statement, the other two mechanisms of REM result in reduced reported expenses. By overproducing, the firm capitalizes an increased amount of fixed costs as part of inventory and by cutting discretionary expenses, managers avoid recording any income-decreasing transactions.
Compared to AEM where the multiple financial reporting options allowed by standard setters enable management to manipulate earnings in several ways making it difficult for researchers to agree on a general model that fully captures discretionary accruals, in the case of REM there seems to be less debate on how to measure the level of real activities manipulation. The original model developed by Roychowdhury (2006) has been consistently used in subsequent studies and there have been relatively few papers centered on pointing out measurement and modelling issues. In the last couple of years, papers on REM tend to focus mainly on the evolution, motives and consequences of REM so the literature gaps in this areas have closed gradually.
Healy and Wahlen (1999) review a large part of the earnings management literature and reveal three traditional motives for earnings management: capital market expectations and valuation; contracts written in terms of accounting numbers; and anti-trust or other government regulation.
Capital market motivations appear because of the widespread use of accounting information by investors and financial analysts in valuing stocks which often creates incentives for managers to manipulate earnings in an attempt to influence short-term stock price performance. Burgstahler and Eames (2006) present evidence that managers take actions to avoid negative earnings surprises, as distributions of earnings surprises contain an unusually high frequency of zero and small positive surprises and an unusually low frequency of small negative surprises. Their paper is one of the first which goes beyond analyzing earnings management through accruals by also providing evidence that managers also take real operating actions in order to achieve analyst targets. Besides periodical earnings releases, another capital market related event for which strong evidence of earnings management has been found, is represented by equity issues. Findings suggest that prior to public equity offers some managers inflate reported earnings in an attempt to increase investors’ expectations of future performance and to increase the offer price.
Earnings management for contracting reasons emerged since accounting data has been used to help monitor and regulate the contracts between the firm and its stakeholders. Two contracts which often have provisions related to accounting performance indicators are management
9
compensation contracts and lending contracts. Management compensation contracts are designed with the purpose of aligning the interests of executive management and shareholders while lending contracts contain covenants that protect creditors from actions that would benefit shareholders at their expense. Healy and Wahlen (1999) note that compensation and lending contracts induce at least some firms to manage earnings in order to increase bonus awards, improve job security and mitigate potential violation of debt covenants.
Finally, out of the regulatory motivations that have been primarily researched most relate to industry-specific, anti-trust and capital markets regulation. Regulatory monitoring that is explicitly tied to accounting data may be more or less prevalent depending on each industry, the banking sector representing probably the extreme case in terms of the importance of accounting data in setting regulatory requirements. On more than one occasions, The Basel Committee (one of the global supervisory authorities responsible for issuing guidance on banking regulation) has expressed its intention to promote convergence between the regulatory approaches to risk measurement and the international financial reporting standards. For instance, Basel III bases its capital composition on IFRS data and the definition of exposures relates back to fair value as defined by the accounting standards. Healy and Wahlen (1999) conclude that strong research evidence supports the hypothesis that accounting discretion is used to manage industry-specific regulatory constraints after citing several studies, most of which focused on the banking sector. Anti-trust regulation is also often recalled as a source of incentives to manage earnings but according to the same authors, there is limited evidence on whether this behavior is widespread or rare and also very little evidence on the effect on regulators or investors.
2.2 The enactment of the Sarbanes-Oxley Act (2002)
One of the most significant regulatory events that triggered a “regime shift” in terms of reporting requirements for all US listed companies and their international subsidiaries, was the enactment of the Sarbanes-Oxley Act (SOX) at July 30, 2002. The aim of this reform was to regain the confidence of investors in capital markets after it abruptly decreased once the large corporate financial reporting frauds such as Enron and WorldCom were exposed. SOX enforces accountability for senior management by requiring it to certify the accuracy of the reported financial statements (Section 302). Moreover, it also specifies that an independent auditor shall attest to and report on the assessment of the effectiveness of the internal control structure of the
10
firm and its procedures for financial reporting (Section 404). Violating or failing to comply with the SOX requirements imposes penalties of fines and/or up to 20 years imprisonment for altering, destroying, mutilating, concealing, falsifying records, documents or tangible objects with the intent to obstruct, impede or influence a legal investigation (Section 802).
As financial accounting literature reveals, SOX had important consequences on earnings management practices. Given the focus of SOX on improving corporate governance and deterring fraudulent financial reporting, several studies document a significant decrease in accrual-based earnings management in the Post-SOX period. Cohen et al. (2008) study the level of accrual-based earnings management relative to analysts’ targets over an extended period and reveal that AEM increased steadily from 1987 until the passage of SOX in 2002, followed by a significant decline in subsequent periods. On the other hand, they observe the opposite evolution in real earnings management, concluding that firms substituted AEM with REM after the passage of SOX and that overall the level of earnings management returned to the Pre-SOX trend line. Bartov and Cohen (2008) perform a similar study on the changes in earnings management mechanisms in the Post-SOX era and besides analyzing AEM and REM, they introduce a third mechanism used for meeting or beating analysts’ targets, named earnings expectations management. This is defined as walking down analyst earnings expectations so as to transform an otherwise negative earnings surprise into a positive one. They find that the frequency of just meeting or beating analyst earnings targets diminished in the Post-SOX period because of significant decreases in AEM and in expectations management. The authors find evidence of an increase in REM activities but this does not compensate the decrease in the two other components of the mix.
Roychowdhurry (2006) focuses on REM in his paper which became an important point of reference for a large part of subsequent literature, at least in terms of the regression models used for estimating REM. The author presents evidence that REM has been used by management in order to meet zero-earnings thresholds even before the enactment of SOX. Moreover, he develops no less than eight hypothesis on different factors that seem to be associated with the incidence of REM. According to his findings, the prevalence of leverage and elevated market-to-book ratios are associated with increased REM while high institutional ownership seems to reduce REM.
Summing up the most important aspects of existing literature on AEM and REM, it seems that managers engage in one or both of the two earnings manipulation mechanisms depending on the time period and on the earnings benchmarks they want to exceed. Overall, significant evidence
11
has been found that managers use both AEM and REM to meet either analysts’ mean EPS targets or zero-earnings thresholds. However, the incidence of AEM seems to be higher in the Pre-SOX era, whereas REM seems to be more present in the Post-SOX era.
2.3 Earnings Benchmarks
Seen from a corporate financial management perspective, the two main earnings benchmarks discussed in my paper reflect proxies for the expectations of the two providers of capital: shareholders and lenders. While shareholders expect listed companies to at least fulfill analyst earnings estimates in order to validate their market price per share, lenders expect companies to generate the positive cash flows necessary to repay the principal and interest as scheduled in their mutually agreed debt contracts.
Studies performed prior to the enactment of SOX find that firms which slightly meet earnings expectations achieve positive short-term market responses and a greater share value than those which do not, after controlling for the book value of equity and the present value of analysts’ estimates of future abnormal earnings (Bartov et al., 2002; Kasznik and McNichols, 2002). On the other hand, Skinner and Sloan (2001) reveal that even small negative earnings surprises are associated with significant stock price declines. In a study that accounts for both the Pre- and Post-SOX eras, Keung et al. (2010) reveal that investors have gradually become more aware of earnings management and consequently have gradually become skeptical about zero or small positive surprises in firm’s earnings releases. The authors analyze the earnings response coefficient (ERC) for three intervals and demonstrate that for the interval which corresponds with the Post-SOX era (2002-2006), the coefficient is significantly lower than for the other two intervals (1992 – 1996; 1997 – 2001).
According to Dechow et al. (1998), an explanation for the importance put by shareholders and lenders on earnings is that current earnings are a better predictor of future operating cash flows than current operating cash flows. This finding also implies that firms with positive earnings performance are less likely to default on their debt obligations. A common practice in debt contracts is to set financial ratio covenants where the borrower is required to maintain threshold levels of specified accounting ratios. In this way the lenders look to decrease the risk of their exposure, by setting interest penalties or obligations to repay debt in advance in case the borrower exceeds the limits specified in covenants. Nevertheless, it is not uncommon that over the duration
12
of their contractual relationship, the two parties amend covenants in order to reflect changes in economic conditions, lender risk aversion or borrower financial performance. For instance, Roychowdhurry (2006) suggests in his paper that covenants tend to become tighter when the borrower incurs losses. Beaty and Weber (2003) present evidence that borrowers closely monitor the costs of covenant violations and that they tend to make income-increasing accounting choices in order to prevent the activation of covenant provisions that stipulate increases in the interest rates in case of poor financial performance.
In a recent study on earnings quality, Graham et al. (2013) provide insights from surveys of 169 chief financial officers of public companies. The authors claim to provide, for the first time ever, point estimates of the economic magnitude of opportunistic earnings management. Based on responses from CFOs they reveal that, in any given period, roughly 20% of firms manage earnings and the typical magnitude for such firms is about 10% of reported earnings per share (EPS). Concerning the motivations for engaging in earnings management, the most common reasons recalled by CFOs are related to capital market factors, followed by debt contracting and career and compensation issues. Respondents were asked to rate the importance of a set of motivations for managers that use earnings to misrepresent economic performance. The following motivations achieved a level of approval which exceeded 80% from the total responses: to influence stock price (94.1%), to face the outside (90.6%) and inside (86.7%) pressures of hitting earnings benchmarks, to influence executive compensation (93%), to avoid the adverse career consequences caused by poor performance (83.7%) and last but not least, to avoid violation of debt covenants (89.2%). A shared opinion among survey respondents was that it is difficult for outside observers such as analysts to discover earnings management, especially when such earnings are managed using subtle unobservable choice or real actions. This finding comes as an extension to a previous study which was led by the same author (Graham et al., 2005) and which revealed that an increasing number of CFOs preferred engaging in real earnings management rather than manipulating earnings through accruals, in order to meet earnings benchmarks such as zero-earnings, previous period’s earnings and analyst forecasts. Given the increased scrutiny from regulators and auditors in the post-SOX era, CFOs seem to prefer manipulating earnings through real activities even though they are aware that such practices lead to a long-term decrease in firm value. For instance, an opportunistic cut in R&D expenditures with the sole aim of hitting earnings targets and not
13
based on economic reasons, reduces the probability of developing new products and diminishes the competitive advantage of being the “first-mover” in new market segments.
Considering the third financial performance threshold discussed in my study, the past year’s earnings, academics provide evidence that supports its importance for executive management. Degeorge et al. (1999) note the relevance of this threshold due to the attention paid by stakeholders to “sustaining recent performance, that is, to meet or surpass the most recent level of comparable earnings”. Merchant and Van der Stede (2011) describe the performance target setting process and identify historical targets as one of the most common thresholds set by boards in executive management’s compensation packages.
To sum up, managers face the pressure of meeting the earnings benchmarks set by their capital providers. In addition, management compensation contracts frequently contain bonuses dependent on exceeding historical performance. Based on the existing literature, I consider that the most common proxy for the benchmark set by equity holders in terms of earnings is represented by the analysts’ mean EPS targets. At the same time, the zero-earnings threshold seems to be the cut-off point at which lenders take defensive actions (e.g. triggering early repayment provisions) which lead to adverse consequences for borrowers. Finally, the previous year’s earnings per share serve as a proxy for managerial contingent compensation. As Degeorge et al. (1999) reveal, all these three earnings benchmarks provide increased incentives for managers to engage in earnings manipulation. The authors also infer that thresholds are hierarchically ranked from the management’s point of view, the most important being the zero-earnings threshold followed by the previous year’s earnings and by the analysts’ mean EPS estimates.
2.4 Hypothesis development
As discussed in the previous sub-sections, until now there have been several instances in which academics studied the managers’ motivations for engaging in earnings management, the different mechanisms of manipulating earnings and the variation of these mechanisms through time. Nevertheless, the existing literature is fragmented in multiple studies, each of them being focused on a particular aspect related to earnings management. For instance, Roychowdhurry (2006) concentrates on detecting real earnings management around the zero-earnings threshold on an annual basis from January 1987 to December 2001, an interval which does not capture the enactment of SOX. The study of Bartov and Cohen (2008) takes into account the regulatory
14
“regime shift” triggered by SOX and compares the prevalence of both accrual and real earnings management in the Pre- and Post-SOX eras. However, it only focuses on earnings management around quarterly analysts’ mean EPS estimates over a period which starts in January 1987 and ends in December 2006.
I consider that the absence of a study which compares the patterns of earnings management across different performance thresholds represents a literature gap that should be filled in order to derive new knowledge in this area. My study aims to analyze and compare the prevalence of earnings management around the zero-earnings threshold, the analysts’ mean EPS estimate benchmark and the previous year’s reported EPS benchmark.
Based on previous literature, I expect that, after controlling for the Pre- and Post-SOX regulatory regime, for each of the three thresholds, firms suspected to have managed earnings display characteristics such as higher discretionary accruals, lower abnormal operating cash flow, lower abnormal discretionary expenses and higher abnormal production costs relative to non-suspect firms. While higher discretionary accruals represent a proxy for accrual-based earnings management, the other three characteristics are meant to capture manipulation of real activities. Therefore, in the first instance, I examine firm-year observations from the interval 1987 – 2014 and test the following directional hypotheses:
H1: Relative to non-suspect firm-years, suspect firm-years engage in accrual-based earnings management in order to exceed financial performance thresholds.
H2: Relative to non-suspect firm-years, suspect firm-years engage in real earnings management in order to exceed financial performance thresholds.
In both hypotheses I mention the concept of suspect-firm years so it is important to define it for each of the three financial performance thresholds. Burgstahler and Dichev (1997) showed that the distribution of firm earnings displays discontinuity around zero and argued that this happens because firms that would otherwise report small losses, manage earnings in other to report small profits. Dechow et al. (2003) performed a study in order to conclude if this is indeed the case but they did not find significant evidence that firms reporting small profits manage accruals to exceed the zero threshold. Roychowdhurry (2006) also addressed this topic and provided strong evidence that firms which reported income before extraordinary items between 0% and 0.5% of total assets have previously engaged in real earnings management in order to exceed the zero-earnings
15
threshold. Therefore, I define suspect firms relative to the zero-earnings threshold as those which at the end of the reporting period report reported income before extraordinary items between 0% and 0.5% of total assets. At the same time, Burgstahler and Eames (2006) found a similar discontinuity in the distribution of earnings surprises, which has also been associated with earnings management by firms that find it important to meet or exceed analysts’ expectations. Bartov and Cohen (2008) provide evidence that firms which just meet or exceed the analysts’ mean EPS estimate benchmark engage in a mix of manipulation mechanisms out of which AEM prevails in the Pre-SOX era while REM strongly emerges after the enactment of SOX. Similar to these prior studies, I define suspect firms relative to the analysts’ mean EPS estimate benchmark as those firms which just meet or beat the mean EPS estimate by one cent or less. Finally, based on the study of Degeorge et al. (1999) which depict a discontinue distribution in the case of all three earnings thresholds, I define suspect firms relative to prior year’s EPS as those firms which just meet or exceed the lagged EPS by one cent or less.
It is important to mention that the discontinuity phenomenon around financial performance thresholds is not supported by all authors. Durtschi and Easton (2005) claim that the discontinuity is only caused by scaling earnings surprises by market capitalization and they argue that the presence or absence of earnings management is unrelated with the shape of the distribution.
In the second part of my study, I seek to determine if any significant differences in earnings management patterns can be identified between the three financial performance thresholds. Since there doesn’t seem to be any studies that might point to an expected finding, I formulate the following non-directional hypothesis:
H3: There is no significant difference in earnings management patterns across different financial performance thresholds.
Thus, according to this hypothesis I expect to see the same pattern of earnings management (for instance, a higher prevalence of AEM relative to REM in the Pre-SOX era) across all three financial performance thresholds.
16
3. Research Design
The process of testing the hypotheses described in the previous section has three stages and is summarized in the table below:
Stage one
1.1 Estimate the level of accrual-based and real earnings management for the relevant firm-year observations from the COMPUSTAT universe;
1.2 Obtain analysts’ mean EPS estimates and actual EPS data from I/B/E/S; 1.3 Merge the files into a unitary database and identify suspect-firm years.
Stage two (tests H1 and H2)
For each of the three financial performance thresholds, determine whether suspect firms display the following differences relative to non-suspect firms:
- unusually high discretionary accruals - unusually low cash flow from operations - unusually low discretionary expenses - unusually high production costs
Stage three (tests H3)
Compare the patterns of earnings management identified around each financial performance threshold and identify the significant differences.
3.1 Sources of data
All the data I use in my study is obtained via WRDS from COMPUSTAT and the I/B/E/S Summary History files. For all my tests I use annual data. As Roychowhurry (2006) notes, annual reports are viewed more seriously by stakeholders because these are audited and considered more reliable. Moreover, the annual interval is common for assessing managerial performance and it also smoothes out for seasonality bias in several industries. The annual frequency of observations is similar to Burgstahler and Eames (2006) and to Roychowdhurry (2006) but is different from Bartov and Cohen (2008) who use quarterly data in their earnings management estimation models.
I eliminate from the COMPUSTAT universe firms from the financial sector (Standard Industry Code [SIC] codes 6000 – 6700), utilities (SIC codes 4800 – 4999) and other regulated industries (SIC codes 4000 – 4499) because these present particular balance sheet structures dependent on their specific business models and regulatory factors. This results in 69,454 firm-years over the period 1987 – 2014, including 5,555 individual firms. Based on this sample I estimate the models for normal cash flow from operations, normal discretionary expenses, normal
17
production costs and non-discretionary accruals. Each model is estimated on a yearly basis for every industry based on the 3-digit SIC industry level. If less than 20 observations are available for a year and 3-digit (2-digit) industry group combination, 2-digit (1-digit) SIC codes are used. The advantage of estimating models in a way to take into account every industry-year is described by Bartov and Cohen (2008) as the capability to control for industry-wide changes in economic conditions that would affect independent variables and to allow the coefficients to vary across time. For each firm-year included in this sample I estimate the level of accrual-based and real earnings management as the difference between actual and normal (predicted) values (the process is further detailed in sub-sections 3.2 and 3.3).
I use the I/B/E/S Summary History database to obtain data on analysts’ mean earnings targets and actual EPS values for each firm-year. In order to avoid potential biases caused by “stale” forecasts, I consider as the analysts’ mean earnings estimate, the most recent mean EPS target preceding the fiscal year-end of each firm-year observation. After merging the data from COMPUSTAT with data from I/B/E/S, I drop all observations for which no information regarding estimated and actual EPS is available. This results in a sample containing a total 24,451 firm-year observations over the period 1987-2014, including 2,779 individual firms. This is the final sample on which I proceed with the second stage of the research process.
3.2 Description of the regression model used to test hypotheses
In order to compare suspect firm-years with the rest of the sample, I estimate the following regression:
Yt= ∝ + β1(SIZE)t−1+ β2(MTB)t−1+ β3(NI)t+ β4SUSPECTPRE−SOX+ + β5SUSPECTPOST−SOX+ ε (1)
The independent variable (Yt) represents, in turn, each of the proxies for real or accrual-based earnings management (described in detail in sections 3.3 and 3.4). The control variables are SIZE, MTB and NI. The SUSPECT variables, which are of interest for testing hypotheses, are dummy variables which take the value of 1 depending whether the suspect firm-year observation is in the Pre- or Post-SOX era (the data point which departs the two SOX eras is set at August 1, 2002). SIZE is the logarithm of the market value of equity at the beginning of the year whereas MTB is the ratio of market value of equity to book value of equity. NI is introduced to control for
18
measurement errors correlated with performance and it is represented by net income scaled by lagged total assets.
As a general approach for estimating regression coefficients I use the Fama-MacBeth (1973) two step procedure which encompasses the following process: in the first step, for each single time period a cross-sectional regression is performed. Then, in the second step, the final coefficient estimates are obtained as the average of the first step coefficient estimates. Academics consider that this approach is useful for avoiding auto-correlation. Nevertheless, taking into account that the section on descriptive statistics (Section 4) reveals that sub-samples of suspect- and non-suspect firms present significant differences in median for most variables, I first control for these differences via propensity score matching before applying the Fama-MacBeth (1973) procedure. For each financial performance threshold, I describe my approach in more detail in sections 5.1.1 – 5.1.3 which ultimately serve at testing hypotheses H1 and H2.
3.3 Methodology used for estimating accrual-based earnings management
In accrual-based earnings management estimation models, researchers break down total accruals in two components, namely discretionary (also referred to as abnormal accruals throughout the paper) and non-discretionary accruals, with the following relationship existing between these two components:
DAt= TAt− NDAt
Where DAt stands for discretionary accruals in year t while TAt and NDAt stand for total accruals and non-discretionary accruals in the same period.
Based on the definition stating accruals as the difference between cash flow and earnings, the total accruals are calculated as income before extraordinary items (IBEI) less operating cash flows adjusted for discontinued operations and extraordinary items (CFO):
TAt= IBEIt− CFOt
In the process of determining the degree of accruals earnings management I used the performance-adjusted modified Jones (1991) model which estimates non-discretionary accruals scaled by total lagged assets. More precisely, I followed the steps provided by Cahan and Veenman (2011) in their case study on the fraudulent reporting practices of the company Waste Management.
19 Accrualst At−1 = b0+ b1 1 At−1 + b2 (∆St− ∆ARt) At−1 + b3 PPEt At−1 + b4 IBEIt−1 At−1 + ε (2)
In this model, besides the elements already defined above, At-1 represent total lagged assets, ΔSt and ΔARt are the current period’s changes in sales and accounts receivable while PPEt represents the current gross property, plant and equipment. The variable represented by the scaled lagged income before extraordinary items (IBEIt-1) is specific to this model and according to Kothari et al. (2005) which have applied it for the first time, it reduces the probability of model misspecification by controlling for differences in firm performance.
For every firm-year, discretionary accruals are calculated as the difference between the actual total accruals and the non-discretionary accruals resulted by using estimated coefficients from the corresponding industry-year model and the firm-year’s sales and lagged assets, accounts receivable, gross PPE and lagged IBEI.
3.4 Methodology used for estimating real earnings management
As I mentioned earlier, the models introduced by Roychowhury (2006) have been adopted in subsequent studies and became a frequently used approach for estimating real earnings management. In order to detect real activities manipulation to avoid losses, academics investigate patterns in cash flow from operations (CFO), discretionary expenses (DISEXP), and production costs (PROD) for suspect firms. The purpose is to focus on three manipulation methods and their effects on the abnormal level of the three variables. First, abnormal CFO represents a proxy for manipulating sales via offering price discounts or more lenient credit terms to clients in order to boost sales volumes before the end of the reporting period. Second, reductions in discretionary expenditures (R&D, advertising, maintenance, employee training, etc.) represent a convenient way for management to avoid recording additional expenses in the current year and consequently to preserve earnings. Last, overproduction represents a method for spreading fixed overhead costs over a larger number of units, lowering fixed costs per units. Both IFRS and US GAAP require the full absorption costing method for valuing the inventory of manufactured goods at the end of the reporting period. Full absorption costing requires fixed manufacturing overhead to be allocated to the total number of units produced. If some of those units are not sold during the period, the fixed overhead costs assigned to excess units are not charged as expenses. Since in most cases, the reduction in fixed costs per unit is not offset by any increase in marginal cost per unit, total
20
cost per unit declines and consequently reported cost of goods sold (COGS) is lower resulting in increased operating profit margins. The models used for estimating abnormal CFO, abnormal discretionary expenses and abnormal production costs are detailed below.
Following Roychowdhurry (2006) and Dechow et al. (1998), normal CFO is estimated as a linear function of sales and change in sales in the current period. The model is obtained by running the following cross-sectional regression for every industry and year:
CFOt At−1 = b0+ b1 1 At−1 + b2 St At−1 + b3 ∆St At−1 + ε (3)
For every firm-year, abnormal CFO is calculated as the difference between the actual CFO and the normal CFO calculated using estimated coefficients from the corresponding industry-year model and the firm-year’s sales and lagged assets.
Roychowdhurry (2006) defines production costs as the sum of COGS and change in inventory (ΔINV) during the period. He demonstrates that examining production costs instead of COGS has two advantages. On one hand, accrual manipulation to lower reported COGS through the inventory account, does not affect production costs because the sum of COGS and inventory change is unaffected. The author shows that the practice of delaying write-offs of obsolete inventory reduces COGS but generates correspondingly higher ending inventory so ultimately production costs are not affected. On the other hand, the choice of LIFO or FIFO accounting modifies reported COGS, but not production costs, due to offsetting effects on COGS and inventory change. To estimate the model, I run the following cross-sectional regression for every industry and year:
PRODt At−1 = b0+ b1 1 At−1 + b2 St At−1 + b3 ∆St At−1 + b4 ∆St−1 At−1 + ε (4)
For every firm-year, abnormal production costs are calculated as the difference between the actual production costs and the normal production costs calculated using estimated coefficients from the corresponding industry-year model and the firm-year’s sales and lagged assets.
The model used by the author for estimating normal discretionary expenses is relatively simple as it requires only two independent variables: inverse lagged assets and lagged sales:
DISEXPt At−1 = b0+ b1 1 At−1 + b2 St−1 At−1 + ε (5)
21
For every firm-year, abnormal discretionary expenses are calculated as the difference between the actual production discretionary expenses and the normal discretionary expenses calculated using estimated coefficients from the corresponding industry-year model and the firm-year’s sales and lagged assets.
Concluding on this part, it is important to note that the models for estimating AEM and REM have been estimated based on the entire COMPUSTAT sample of relevant industries in order to capture a large amount of firm-year observations for each industry and from Pre- and Post-SOX eras. Nevertheless, after merging COMPUSTAT with I/B/E/S, in order to have a unitary sample for testing my hypotheses, I retained only those observations for which available information existed concerning both analysts’ EPS targets and actual EPS. Therefore, I test my hypotheses based on a sample of 24,451 firm-year observations with 5,570 suspect firm-years based on just meeting or slightly exceeding analysts’ mean EPS estimate benchmark, 364 suspect firm-years based on just meeting or slightly exceeding the zero-earnings threshold and 332 suspect firm-years based on just meeting or slightly exceeding the previous year’s EPS.
4. Descriptive Statistics
4.1 Firm characteristics
Table 1 presents a set of descriptive statistics for all suspect- and non-suspect firm-years included in the sample, relative to each financial performance threshold considered in my study. Compared to the rest of the sample, in each of the three cases, characteristics of suspect firms do not present a consistent pattern and in several instances do not seem to point to any of the real or accrual-based management mechanisms which were documented in previous studies. Table 1 displays the median of each variable for each group (suspect or non-suspect) and the z-statistics from Wilcoxon tests for differences in medians. Taking into consideration that the samples of suspect firms are relatively thin in the case of the last two thresholds (364 and 332 suspect firm-years compared to over 24,000 non-suspect firms in each case), I use the median in my analysis in order to minimize any potential bias caused by outliers.
In general, suspect firms tend to have a lower market capitalization compared with the rest of the sample. However, it is interesting to note that for two out of the three financial performance thresholds, suspect firms have a similar level of total assets. Taking this two aspects into account, it can be inferred that investors require discounts for holding suspect firms.
22
Table 1
Median comparison for full sample - 24,451 firm-years, period 1987 - 2014
Zero-earnings threshold
Analysts' mean EPS estimate benchmark Previous year’s reported EPS benchmark Suspect firm-years Rest of the sample Difference (z-stat) Suspect firm-years Rest of the sample Difference (z-stat) Suspect firm-years Rest of the sample Difference (z-stat) MVE ($ mln) 631.12 823.98 -192.86 756.57 1056.26 -299.69 453.40 829.04 -375.64 (-3.72)** (-10.62)** (-5.33)** Total Assets ($ mln) 776.51 673.98 102.53 665.86 678.73 -12.87 317.77 679.82 -362.05 (1.92)* (-1.71)* (-6.12)** Sales ($ mln) 675.18 650.34 24.84 624.67 655.63 -30.96 287.67 657.48 -369.81 (1.22) (-0.02) (-6.29)** IBEI ($ mln) 1.53 27.63 -26.10 24.04 36.01 -11.97 12.59 27 -14.41 (-10.76)** (-13.02)** (-4.11)** CFO ($ mln) 40.36 60.85 -20.49 68.05 58.58 9.47 31.49 61.28 -29.79 (-2.22)** (6.24)** (-4.45)** Accruals ($ mln) -39.02 -26.82 -12.20 -24.01 -27.72 3.71 -15.52 -27.16 11.64 (-2.64)** (3.08)** (3.57)** Sales/At-1 (%) 0.92 1.03 -0.11 1.09 1.01 0.08 0.99 1.03 -0.04 (-3.29)** (9.69)** (-1.31) IBEI/At-1 (%) 0.25% 5.66% -5.41% 7.76% 4.92% 2.84% 5.08% 5.56% -0.48% (-16.53)** (25.46)** (-1.45) CFO/At-1 (%) 6.13% 10.37% -4.24% 12.54% 9.64% 2.90% 9.78% 10.31% -0.53% (-9.49)** (20.76)** (-0.90) Accruals/ At-1 (%) -5.82% -5.37% -0.45% -4.99% -5.50% 0.51% -5.82% -5.37% -0.45% (-0.58) (5.59)** (-0.58) PROD/At-1 (%) 60.33% 61.87% -1.54% 62.51% 60.01% 2.50% 57.65% 61.94% -4.29% (-0.71) (1.68)* (-2.38)** DISEXP/ At-1 (%) 21.42% 30.54% -9.12% 34.90% 28.99% 5.91% 32.75% 30.33% 2.42% (-6.42)** (14.69)** (2.12)** Inventory turnover 4.87 4.91 -0.04 4.70 4.99 -0.29 4.72 4.92 -0.2 (-0.117) (-5.84)** (-1.15) No. of obs. 5,570 18,881 362 24,087 332 24,119
23
The proxies for real-earnings management present inconsistencies across the three financial performance thresholds. The ratios of CFO and DISEXP to lagged assets are significantly lower for suspect firms in the case of the zero-earnings threshold. This is similar to the descriptive statistics presented by Roychowdhurry (2006) in his study on real earnings management.
In the case of analysts' mean EPS estimate benchmark suspect firms appear to be more profitable that non-suspect firms. It seems that companies which meet or just exceed analyst thresholds are on average the firms which already report higher positive EPS whereas firms from the rest of the sample present a weaker performance (an operating profit ratio of 4.92% for non-suspect firms compared to 7.76% for non-suspect firms). When testing H1 and H2 I use propensity score matching in order to control for these inconsistencies between the two groups.
Relative to previous year’s reported EPS benchmark, the table reveals mixed signals. On one hand the lower CFO and inventory turnover ratios indicate potential real earnings management for suspect firms but on the other hand their discretionary expenses and production cost ratios are significantly higher than those of non-suspect firms.
Tables 1A and 1B are attached as appendices and present the same set of variables for the Pre- and Post-SOX era. In this way I seek to observe if the preliminary conclusions from Table 1 are different once I control for the regulatory regime. However, it remains difficult to identify a clear pattern for suspect firms relative to any of the three financial performance thresholds. I expect that once I perform the regression analysis and control for a series of risk factors, more insights will be available.
4.2 Earnings distribution around financial performance thresholds
In this sub-section I proceed with illustrating the distribution of earnings for each financial performance threshold. Prior research reveals that the distribution of earnings scaled by market value is discontinued in two instances – at the zero-earnings threshold and at the analysts’ mean EPS estimate benchmark (Burgstahler and Dichev, 1997, Burgstahler and Eames, 2006). Roychowdhurry (2006) scales earnings by total lagged assets and also identifies a discontinuous distribution but less evident than those documented in previous studies. Durtschi and Easton (2005) claim that the discontinuity is only caused by scaling earnings surprises by market capitalization and they argue that the presence or absence of earnings management is unrelated with the shape of the distribution.
24
I present in the figures below, the distribution of earnings (or earnings surprises depending on which threshold is considered) both in absolute terms and also scaled by market value (or total assets in the case of the zero-earnings threshold). Starting with the zero-earnings threshold, I group firm-years into intervals based on IBEI scaled by lagged assets. Figure 1 presents 80 earnings intervals over the range -5% to 5%, thus each interval is of width 0.125%. Compared to prior studies, the discontinuity identified at the left of zero is still present but not as evident in this case.
Figure 1 – Number of firm years by earnings interval relative to the zero-earnings threshold
In the case of earnings surprises relative to the analysts' mean EPS estimate benchmark (Figures 2 and 2A), the discontinuity is noticeable when plotted based on two cent intervals of earnings surprises (thus, as absolute values) ranging from –10 cents to +10 cents. When scaled by market value at the beginning of the year, the distribution presents no discontinuity at the left of zero but still presents a large shift at the right of zero.
0 50 100 150 200 No . o f fi rm y e a rs -.05 0 .05 Earnings interval
25
Figure 2 – Number of firm years by earnings surprises intervals (cents)
Figure 2A – Number of firm years by earnings surprises intervals (scaled by market value)
Finally, relating to the previous year’s EPS benchmark, Figure 3 highlights the strongest discontinuity in absolute terms, with a very steep drop in companies which report a small decrease in EPS relative to the prior year. The histogram contains one cent intervals of year-to-year earnings variation ranging from -5 cents to +5 cents.
0 2000 4000 6000 No . o f fi rm y e a rs -.1 -.05 0 .05 .1
Earnings surprises (cents)
0 1000 2000 3000 F re q u e n c y -.00001 -0.000005 0 0.000005 .00001 Earnings surprise scaled by market value
26
Figure 3 – Number of firm years by year-to-year earnings variation intervals (cents)
Figure 3A displays a totally different picture and it reveals the absence of any discontinuity left to zero and even more, no significant shift just right to zero.
Figure 3A – Number of firm years by year-to-year earnings variation intervals (scaled by market value)
Overall, this sub-section reveals that earnings distributions present large variations depending on whether variables are plotted as absolute values or as scaled values. Moreover, the size of the intervals represents another factor that influences the form of the distribution. These
0 100 200 300 400 No . o f fi rm s -.05 0 .05
Year-to-year earnings variation (cents)
0 500 1000 1500 F re q u e n c y -.0001 -.00005 0 .00005 .0001
27
findings seems to correspond with the arguments of Durtschi and Easton (2005) who suggest that shape of the earnings surprise distribution is highly sensitive and that it is unrelated to the presence or absence of earnings management.
5. Results
5.1 Estimation models
The first stage of my research requires the computation of abnormal levels for accruals, CFO, DISEXP and PROD. Using each regression equation discussed in Section 3, I predict the normal levels of these variables. The differences between actual and predicted levels are labeled as abnormal and are investigated in the following sub-sections. Given the fact that all dependent and independent variables are Winsorized at the 1% level, their standard deviation decreases and consequently, the models have an increased degree of robustness. Table 2 displays the mean coefficients across all industry-years and t-statistics calculated using the standard error of the mean across industry-years. The standard error is consistently low across each model and consequently, all coefficients are significant at the 5% level. The explanatory power of the models is high, ranging from 49% in the case of CFO to 82% in the case of PROD.
Table 2
Model parameters
CFOt/At-1 DISEXPt/At-1 PRODt/At-1 Accrualst/At-1
Intercept -0.045** 0.2163** -0.0817** -0.0294** (-290) (855) (-720) (270) 1/At-1 -0.9496** 2.5834** -0.4425** -0.1044** (-300) (499) (-87) (-64) St/At-1 0.1072** 0.1102** 0.7165** (721) (505) (280) ΔSt/At-1 -0.0545** 0.0172** 0.0171** (-18.8) (608) (62) ΔSt-1/At-1 0.0284** (111) PPEt-1/At-1 -0.0685** (-310) IBEIt-1/At-1 0.2649** (732)
28
Adjusted R2 0.49 0.52 0.82 0.54
* Significant at the 10% level. **Significant at the 5% level.
5.1.1 The zero-earnings threshold
In order to detect potential earnings management practices in the case of the zero-earnings threshold I follow Roychowdhurry’s (2006) approach and apply directly the regression model presented in section 5.1. Since the profitability itself is the factor that differentiates between suspect and non-suspect firms, it is not necessary to apply propensity score matching for performance before estimating regression (1), as it is the case for the other two earnings benchmarks discussed in sub-sections 5.1.2 and 5.1.3.
Table 3
Comparison of firm-years that barely exceed zero-earnings with the rest of the sample
Abnormal Accruals Abnormal CFO Abnormal DISEXP Abnormal PROD Intercept 0.0614 -0.0283 -0.0180 0.0163 (5.74)** (-3.61)** (-1.22) (2.39)** SIZE -0.0096 0.0096 -0.0127 0.0004 (-5.26)** (9.58)** (-6.98)** (0.69) MTB 0.0003 0.0057 0.0158 -0.0139 (0.36) (6.82)** (7.1)** (-9.94)** NI 0.1209 0.3929 -0.2717 -0.3792 (4.03)** (11.81)** (-3.76)** (-16.93)** SUSPECTPRE-SOX 0.0096 -0.0025 -0.0521 0.0185 (1.55) (-0.56) (-2.78)** (2.02)** SUSPECTPOST-SOX 0.0091 -0.0131 -0.0043 0.0148 (2.61)** (-2.08)** (-0.5) (1.68)* Adjusted R2 4.05% 18.10% 7.52% 12.92%
* Significant at the 10% level. ** Significant at the 5% level.
In order to test H1, I estimate regression (1) with abnormal accruals as an independent variable. The coefficient of determination is very low, so the findings must be treated with caution. By studying the coefficients of SUSPECTPRE-SOX and SUSPECTPOST-SOX it can be observed that in both the Pre- and Post-SOX eras suspect firms present abnormal accruals that are higher with 0.91% and 0.96%, respectively, than those of non-suspect firms. However, the coefficient of SUSPECTPRE-SOX is only significant at the level of 13.3%. Overall, even though the evidence is thin, it can be claimed that H1 fails to be rejected and therefore, concerning the zero-earnings threshold, suspect firms seem to engage in accrual-based earnings management.
29
In relation to the real earnings manipulation mechanisms considered for testing H2, after estimating regression (1) for each of them, the coefficient of determination ranges from 7.52% to 18.10%. Starting with the abnormal CFO which also has the highest R2, it can be observed that suspect firm-years from the Post-SOX era display a negative difference of 1.31% in abnormal CFO relative to the rest of the sample. This finding is in line with previous studies. Continuing with abnormal DISEXP, the table reveals an interesting situation – suspect firms seem to display lower abnormal discretionary expenses in the Pre-SOX era relative to the Post-SOX era, when compared to non-suspect firms. This is in contrast with prior studies that found the opposite, with suspect firms using manipulation through discretionary expenses preponderantly in the Post-SOX era. Finally, coefficients attached to abnormal PROD are positive in both periods and suggest that suspect firms consistently manipulate earnings through overproduction. It is important to note that the intensity of this mechanism seems to be higher in the Pre-SOX era which, as in the case of abnormal DISEXP, is in contradiction with previous studies. As a whole, the findings derived from table 3 indicate that H2 fails to be rejected and thus, in the case of the zero-earnings threshold, suspect firms seem to engage in real earnings management.
5.1.2 The analysts' mean EPS estimate benchmark
The section on descriptive statistics revealed significant differences in terms of size and profitability between suspect- and non-suspect firms relative to the analysts' mean EPS estimate benchmark. More precisely, suspect firms tend to be larger and more profitable than the rest of the sample so applying regression model (1) without controlling for these differences might lead to biased findings. I look to control for any possible errors by matching each suspect firm with its equivalent non-suspect firm in terms of total assets, earnings per share and operating profit margin. Therefore, by applying propensity score matching, I obtain a sample of 11,140 firm years. This serves as the basis for regression model (1) which results in the following coefficients:
Table 4
Comparison of firm-years that just meet analyst forecasts with the rest of the sample Abnormal Accruals Abnormal CFO Abnormal DISEXP Abnormal PROD Intercept 0.0638 -0.0337 -0.0163 0.0381 (5.68)** (-3.53)** (-1.29) (3.09)** SIZE -0.0099 0.0093 -0.0151 -0.0002 (-5.22)** (6.31)** (-6.81)** (-0.14)
30 MTB 0.0006 0.0062 0.0170 -0.0149 (0.65) (5.3)** (6.14)** (-9.52)** NI 0.0663 0.3808 -0.1177 -0.4399 (2.03)* (10.72)** (-1.28) (-9.82)** SUSPECTPRE-SOX -0.0026 0.0128 0.0041 -0.0149 (-1.13) (3.08)** 0.92 (-2.86)** SUSPECTPOST-SOX 0.0027 0.0078 -0.0117 -0.0033 (1.30) (3.64)** (-2.64)** (-1.24) Adjusted R2 5.02% 18.44% 7.66% 16.59%
* Significant at the 10% level. ** Significant at the 5% level.
Starting with abnormal accruals, the regression analysis reveals that both in the Pre- and Post-SOX eras suspect firms do not present significant differences from the rest of the sample. This finding is not consistent with Burgstahler and Eames (2006) who found that suspect firms present significant positive abnormal accruals relative to non-suspect firms. Thus, I reject H1 and conclude that suspect firms do not seem to engage in accrual-based earnings management for exceeding the analysts' mean EPS estimate benchmark.
Going on with the proxies for real earnings management, it is difficult to directly accept or reject H2 because the regression analysis reveals a mixed picture. First, concerning discretionary expenses I find significant evidence that in the Post-SOX era suspect firms present abnormal DISEXP that are 1.17% lower than those of non-suspect firms. This is not the case in the Pre-SOX era where there is no significant difference between the two groups of firms. Second, in the case of abnormal CFO, the coefficients of suspect firms in both eras are positive. This is in opposition with theory on real earnings management and suggests not only that suspect firms do not engage in manipulation of real activities but also that on average their abnormal CFO exceeds that of non-suspect firms by 1.28% in the Pre-SOX era and 0.78% in the Post-SOX era. Lastly, abnormal PROD does not bring any clear evidence that would lead to not rejecting H2. Whereas in the Pre-SOX era suspect firms display lower abnormal production costs, in the Post-Pre-SOX era they do not present any significant differences relative to the rest of the sample. Taking into account that two out of the three real management earnings proxies do not reveal any manipulation practice, I reject H2 and conclude that suspect firms do not seem to engage in real earnings management for exceeding the analysts' mean EPS estimate benchmark.
31 5.1.3 The previous year’s reported EPS benchmark
I adopt an approach similar with the one described in section 5.1.2 so before estimating regression (1), I first control for differences in total assets, earnings per share and operating profit margin between suspect- and non-suspect firms. By applying propensity score matching, I obtain a sample of 656 firm- years. This serves as the basis for regression model (1) which results in the following outputs:
Table 5
Comparison of firm-years that barely exceed past year's earnings with the rest of the sample
Abnormal Accruals Abnormal CFO Abnormal DISEXP
Abnormal PROD Intercept 0.0696 -0.0215 0.0598 -0.0940 (2.08)** (-0.56) (1.10) (-1.70)* SIZE -0.0089 0.0077 -0.0341 0.0224 (-1.75)* (1.30) (-4.48)** (3.49)** MTB 0.0022 0.0102 0.0255 -0.0304 (0.65) (2.69)** (2.44)** (-3.72)** NI 0.0391 0.3509 0.2368 -0.4436 (0.30) (3.24)** (1.45) (-3.13)** SUSPECTPRE-SOX 0.0143 -0.0097 0.0205 -0.0206 (1.84)* (-1.16) (1.07) (-0.94) SUSPECTPOST-SOX -0.0030 0.0015 -0.0264 0.0222 (-0.35) (1.05) (-1.74)* (1.30) Adjusted R2 34.78% 35.37% 25.29% 37.56%
* Significant at the 10% level. ** Significant at the 5% level.
It is important to note that coefficients of determination range from 25.29% to 37.56%, considerably higher than those displayed in previous instances. On the other hand, it can also be observed that coefficients of SUSPECTPOST-SOX andSUSPECTPRE-SOX are only in a few instances significant at the 10% level. Therefore, even though models seem to explain more of the phenomenon, the findings should still be treated with caution due to the reduced significance in the variables I am interested in.
First, concerning abnormal accruals, the table reveals that in the Pre-SOX era suspect firms display a positive difference of 1.43% compared with the rest of the sample. This is not also the case in the Post-SOX era where it can be assumed that there is no significant difference between suspect and non-suspect firms. This finding is similar with those of previous studies in the sense that accrual-based earnings management seems to have been less prevalent after the enactment of
32
SOX (Bartov and Cohen, 2008). Putting all together it might be difficult to draw a clear conclusion on whether to reject H1 or not, but given that the general pattern revealed by the regression model is similar to existing literature it can be claimed that H1 fails to be rejected and therefore concerning the previous year’s EPS benchmark, suspect firms seem to engage in accrual-based earnings management.
Second, considering the manipulation of real activities, the general picture corresponds with the theoretical insights provided by previous studies. It seems that compared to the Pre-SOX era, after the SOX enactment suspect firms engage in real earnings management by reducing discretionary expenses and by overproducing goods. Abnormal DISEXP and Abnormal PROD for suspect firms in the Post-SOX era are 2.64% lower and 2.22% higher, respectively, than non-suspect firms. Abnormal CFO does not indicate any significant difference between firms in neither the Pre- or Post-SOX eras. Summing up the findings for all three real earnings management proxies, in spite of the low level of significance for the estimated regression coefficients, it can be claimed that H2 fails to be rejected and therefore concerning the previous year’s EPS benchmark, suspect firms seem to engage in real earnings management.
5.2 Comparing earnings management patterns across financial performance thresholds Up to this point I analyzed whether, relative to each financial performance threshold, suspect firms seem to engage in earnings management. Table 6 summarizes the findings from previous sub-sections and also allows drawing a conclusion regarding H3.
Table 6: Summary of findings (Pre-, Post-SOX) Manipulation
mechanisms: The zero-earnings threshold
The analysts' mean EPS
estimate benchmark
The previous year’s reported EPS benchmark
Persistence of positive abnormal accruals
Insignificant evidence PRESOX
Significant evidence POSTSOX
Absent evidence PRESOX
Insignificant evidence POSTSOX
Significant evidence PRESOX
Absent evidence POSTSOX
Persistence of negative abnormal CFO
Insignificant evidence PRESOX
Significant evidence POSTSOX
Absent evidence PRESOX
Absent evidence POSTSOX
Insignificant evidence PRESOX
Absent evidence POSTSOX
Persistence of negative abnormal DISEXP
Significant evidence PRESOX
Insignificant evidence POSTSOX
Absent evidence PRESOX
Significant evidence POSTSOX
Absent evidence PRESOX
Significant evidence POSTSOX
Persistence of positive abnormal PROD
Significant evidence PRESOX
Significant evidence POSTSOX
Absent evidence PRESOX
Absent evidence POSTSOX
Absent evidence PRESOX
