Effects of Real Earnings Management on credit ratings when the
investment-speculative borderline is approached.
Master Thesis Laura Klaver
Abstract
This thesis discussed whether entities with credit ratings approaching the investment-speculative borderline, that are entities with BBB and BB credit ratings, more aggressively use real earnings management after passing the Dodd–Frank Act in 2010. The dataset consists of 1.202 firm-years observation of 143 unique entities in the United States between January 1, 2002 and December 31, 2016. By using Standard & Poor’s credit ratings, this study finds no significant results that entities with credit ratings close to the borderline more aggressively use earnings management pre- or post-Dodd-Frank.
Name: Laura Klaver
Student number: 10597352
Thesis supervisor: Dr. Georgois Georgakopoulos Date: 25 June 2018
Word count: 12.331
MSc Accountancy & Control, specialization Control
2
Statement of Originality
This document is written by Student Laura Klaver 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 content
1. Introduction ... 4
2. Literature review... 6
2.1 Credit rating ... 6
2.2 General critic on credit rating agencies ... 6
2.3 Dodd-Frank Act ... 7
2.4 Earnings management ... 8
2.5 Incentives for earnings management ... 9
2.6 Earnings management and investment-speculative credit rating ... 10
3. Hypothesis ... 12
4. Methodology ... 13
4.1 Models for determination of abnormal real activities ... 13
4.2.1 Dependent and independent variables ... 14
4.2.2 Controlled Variables ... 15
4.3 Data ... 16
5. Results ... 18
5.1 Descriptive summary ... 18
5.2 Pearson correlation ... 21
5.3 Ordinary Least Square for the overall period ... 22
5.4 Ordinary Least Square for the three periods ... 24
6. Conclusion ... 29
7. Limitations and future research ... 31
8. References ... 33
4
1. Introduction
According to Gupta, Mittal and Bhalla (2010), the financial crisis of 2007-2008 was the most severe crisis since the Great Depression and resulted in a global financial meltdown. One of the many studies that examines the origin of this crisis is that of Dimitrov, Palia and Tang (2015) and EBC (2009). Studies like these states the failures of Credit Rating Agencies (CRA) as an important factor causing the crisis. CRAs determine credit ratings where these ratings reflect the creditworthiness of an issuer or obligor whereas the aim of credit ratings is to decrease information asymmetry between the issuer and investors (Securities and Exchange Commission, 2013). Prior to the financial crisis (mortgage) securities became more complex and therefore investors relied more on the credit rates provided by CRAs (Gupta et al., 2010). However, these CRAs still used the same credit rating scale to rank the default risk for all long-term, including the mortgage securities, leading to an underestimation of the systematic risk. During the 2007-2008 financial crisis, this was especially harmful for investors of and homeowners with doubtful creditworthiness mortgage securities (Gupta et al., 2010).
In response, the Congress of the United States passed the Dodd-Frank Wall Street Reform and Consumer Protection Act (Dodd-Frank Act) in 2010. The Dodd-Frank Act obtains series of broad reform to the CRA market by giving the SEC and other federal agencies the responsibility for developing specific rules and mandating internal control. As of today, the amount of corporate debt issuance is growing and together with the increasing complexity of the capital market it makes the demand for accurate ratings is still increasing (Brown, Chen & Kim, 2015). Research conducted on this act find mixed evidence concerning whether the new legislation stimulates the CRAs to provide more accurate credit ratings (Goel & Thakor, 2014). The still failing reliance on entities’ assigned credit ratings might be a problem as it can harm issuers, creditors, and investors.
CRAs use a system of letter grades to indicate ratings: a high credit rate indicates the security is associated with a low probability of default, a low credit rating indicates a higher probability of default. For example, Standard and Poor’s (S&P) uses a credit rating scale ranging from AAA (excellent) to D (poor). According to Brown et al. (2015) the investment-speculation grade is considered as one of the most important factors of decision making in managing earning. The investment-speculative rating dichotomy was introduced in 1930 and gained acceptance by financial organizations, investment communities and regulators (Standard & Poor’s, 2008). The investment-grade rating has a credit rate of BBB or above, the
speculative-5
grade rating has a credit rate of BB or below. The difference between these grades is the likelihood of default. Entities with Speculative-grades have a higher likelihood to default than those with grades. Especially when the credit rate is close to the investment-speculation borderline, managers have the incentive to manipulate earnings to inflate the entities credit rating (White, 2009). The research of Brown et al. (2015) shows that BBB and BB credit rated entities more aggressively use earnings management than entities with higher or lower credit rates during the period from 1989 to 2009. Since this period is prior to the introduction of the Dodd-Frank act in 2010, this study tests whether the use of earnings management differs prior and post 2010. The research question is stated as follows:
‘’Do entities use more aggressively real earnings management as their credit ratings approach the investment-speculative borderline in the post-Dodd-Frank period?”
This thesis contributes to existing literature in the fields of earnings management because real activities earnings management is used as a tool for earnings management where many studies use accrual-based earnings management (Zhao, 2017). Also, the effects of the Dodd-Frank act of 2010 on earnings management are studied. Examination is done whether the Dodd-Frank reduces the aggressive use of real earnings management when entities’ credit rates are near the investment-speculation borderline. Brown et al. (2015) examined the relation between real earnings management and the investment-speculation borderline in the pre-Dodd-Frank. However, no research is performed in the post-Dodd-Frank period. As a third topic, a contribution to existing literature about CRA’s is made. The Dodd-Frank Act is implemented because of the inability or unwillingness of the CRAs to deal with real earnings management. Results of this thesis show whether CRAs are able to undo this management within their predicting models after the implementation of the Act.
This thesis is structured as follows: the second chapter discusses the existing literature research on the subject. Motivated on the gap in existing literature, chapter three introduces the hypotheses. Chapter four will present the data and methodology used to empirically examine the research question, while the fifth chapter describes and explains the results. In the chapter six the conclusion of the empirical research conducted will be presented. In the last chapter, the limitations and recommendations for future research.
6
2. Literature review
The literature review provides an overview of previous conducted research and starts with the explanation of credit ratings together with the discussion that has to do with the general criticism on CRAs. This is followed up by a description of earnings management and explanation why management has an incentive to use earnings management. Also, this chapter will explain the reason to choose for a more aggressive use of earnings management when the credit rate is close to the investment-speculative borderline. It concludes with a description of the Dodd-Frank Act of 2010 and how it should improve the quality of the credit rating.
2.1 Credit rating
Credit ratings are broadly used by regulators since the adaption of the Nationally Recognized Statistical Rating Organization (NRSRO) by the SEC in 1975 (Jorion, Shi & Zhang, 2009). The credit rating can be defined as an assessment of a prospective debtors’ creditworthiness and is an implicit forecast of the likelihood of defaulting (Securities and Exchange Commission, 2013). The credit rates are determined and published by CRAs. Hereby, CRAs try to maintain credit rates which provide a stable signal about creditworthiness on the long-run and reduce information asymmetry between the borrowers and lenders (Jorion et al., 2009). The ‘’big three’’ CRAs are Standard & Poor’s (S&P), Moody’s and Fitch. These CRAs use their own system of letters and numbers for ranking the default risk (see Appendix A).
Furthermore, a distinction is made between investment- (BBB and above) and speculative-grade (BB and below) securities. When the credit rate is on the borderline of these grades, there is a possibility that the credit rate is downgraded from investment-grade to the speculative ‘’junk’’ status and vice-versa. Entities experience some disadvantages when their rate is downgraded from the investment-grade to a speculative-grade. One of the disadvantage for entities is that financial and other larger institutions place restrictions on investments in speculative-grade securities because of higher default risks (Cantor & Packer. 1997). According to Brown et al. (2015) downgrading reduces the equity prices and the overall value of the entity. Also, speculative graded securities experience a lower equity return volatility than investment graded securities which means these entities are less capable to achieve high returns.
2.2 General critic on credit rating agencies
CRAs have incentives to provide high-quality ratings, because inaccurate ratings can negatively influence the reputation of the CRAs (Zhao, 2017). Damage to their reputation has a negative
7
effect on the CRAs capital allocation. For example, lawsuits originated from low-quality ratings cause their stock price to decrease. Although CRAs have incentives to provide high-quality ratings, there is still some criticism about the questionable quality of these ratings (Bolton, Freixas & Shapiro, 2012). Criticism state CRAs use forecasting models that lack precision because of incentive issues plus economic and environment uncertainty.
Furthermore, the issuer-pay model, where CRAs are hired by the issuer, is associated with lower quality of the credit rating than when the investors-pay model is in place (Jiang, Stanford & Xie, 2012). The issuer-pay model leads to a potential conflict of interest, because CRAs can receive higher fees when the entities’ credit rate is inflated. Thereby, the credit rate will not be released when the issuer is not satisfied. This gives the issuer the possibility to “shop for ratings’’ (Zhao, 2017). Criticism also points about that CRAs have immunity for misstatements made in registration of the credit rate under Section 11 of the Security Act of 1933 (Partnoy, 2012).
Another major critic on CRAs is that they played a significant role in the 2007-2008 financial crisis (Jiang et al., 2012). The CRAs underestimated the credit risk of structured credit products and failed to adjust the credit ratings quickly enough to the changing market. CRAs were held responsible for methodological errors within their models and for the fact that they were unable to solve the appeared conflicts of interests. In response to criticism and the role of CRAs in the financial crisis, the Congress passed the Dodd-Frank Act.
2.3 Dodd-Frank Act
The Dodd-Frank Act was introduced in 2010 and fundamentally changed the environment of the CRA market. The act outlines series of broad reform to the CRA market by giving the SEC and other federal agencies the responsibility for developing specific rules and mandating internal control. Hereby, the Dodd-Frank Act increased the SECs oversight of CRAs, where the SEC grants power to impose sanctions, penalties or suspend CRAs if credit ratings are inaccurate. Furthermore, the Act improves monitoring of the rating methodologies and the internal governance. Thereby, there is an increase in potential liability for CRAs when providing inaccurate credit ratings since they are liable as ‘’expert’’. Summed up, the Dodd-Frank Act should increase the CRAs regulation oversight, improve CRAs litigation risk and improve accurate credit rates.
Dimitrov et al. (2015) states the Dodd-Frank Act has two effects on the quality of credit ratings. First, increasing legal and regulatory penalties can lead to an improvement in the quality of credit ratings. Due to these penalties, CRAs have incentives to invest in diligence and
8
improve their rating methodology. Second, the penalties can also have an adverse effect on the quality of credit rates, because of the fact that penalties are asymmetric. Hereby, CRA lower their credit rating to protect their reputation and this is, for example, the case for investment-grade rating for issuers with a high likelihood of defaulting. This means CRAs are penalized for inflated biased rating, but not for deflated biased ratings.
Furthermore, the research of Dimitrov et al. (2015) shows the credit ratings are lower in the post-Dodd-Frank period. However, they find no evidence that these credit ratings are more accurate. They also find more false warnings defined as speculative-grades issues that not default in the next-year. These results are in line with the reputation model of Morris (2001). According to this model, CRAs lower the provided credit rates to protect their reputation. In contrast, deHaan (2016) shows the credit ratings improved after 2010: they are more accurate and stable. However, the market participants reliance on the credit rating declined in the Post-Dodd-Frank period, because of the damage incurred during the 2007-2008 financial crisis to the CRAs reputation.
2.4 Earnings management
Healy and Wahlen (1999) define earnings management as “judgments and decisions of managers in financial reporting, with the goal to adjust the financial reports to either mislead some stakeholders about the economic performance of the entity or to influence contractual outcomes”. This definition is commonly used in the research of the relationship between earnings management and credit ratings.
By using earnings management, it is possible to shift the reported income between the current and future periods. Earnings management can be applied by the use of current accrual, that is for example, by reducing the provisions for bad expenses, recognition of revenues and/or expenses. Long-term accruals can also be used to manage earnings, such as delaying the recognition of the asset writing down, decreasing the amount of deferred taxes and recognition decelerating of deprecation expenses. The General Accepted Account Principle (GAAP) gives managers flexibility about the manner of reporting earning management and therefore it is not the same as accounting fraud. However, the agency theory explains that managers have incentives to only use this flexibility to obtain their personal benefits (Dimitrov et al., 2015).
Many studies examine whether entities obtain favorable credit ratings by using management earnings. Most of them use accruals-based earnings management, because managers can exercise discretion over accrual choice which are allowed under GAAP (Ge & Kim, 2014). Fewer studies use real activities earning management where managers can manage
9
earnings by altering real operating activities and their timing. In the research of Zhao (2017), both accrual-based and real activities earning management are used. There is no significant difference between both types of earnings management.
Graham, Harvey and Rajgopal (2005) states that managers prefer the use of real operating activities, like sales and production activities. Even if it deviates from sub-optimal decision, using real earnings management should draw less attention to auditors. It is harder for auditors to detect real earnings management techniques. However, the disadvantage is that it is costlier to use than accrual-based earnings management because it can negatively affect future cashflows and reduce firms value (Roychowdhury, 2006).
According to Roychowdhury (2006), the focus is on three types of real operating activities, namely (1) increasing production, (2) decreasing discretionary expenses and (3) increasing cashflow from operating activities. By increasing production, entities report a lower Cost of Goods Sold (COGS). Most of the time overproduction is used to increase the earnings. By using overproduction, the fixed overhead cost spreads over more units, which results in lower cost per unit. According to Ge and Kim (2014), using overproduction allows the managers to report a lower COGS given the already estimate sales levels. By reducing discretionary expenses, such as R&D, employee training, advertising, administration and other costs, entities can also increase their earnings. Managers have the power and can easily decrease discretionary expenses. When the discretionary expenses decrease, the Earnings Before Interest and Tax (EBIT) will increase (Roychowdhury, 2006). The intuition behind increasing cashflow from operations is that managers offer significant price cuts and lenient credit terms in an attempt to increase sale. This results in a boost of the current earnings but will lower cashflow per sale which is hurting the future earnings.
2.5 Incentives for earnings management
There are several reasons for managers to use earnings management. Using earnings management can give the appearance of a better performance, which could help to reach some pre-establish benchmarks (Xie, Davidson III & DaDalt, 2003). When a benchmark is reached, managers are compensated directly (salary and bonus) and indirectly (promotion and job security). A potential agency problem can occur when managers use earnings management to obtain their pre-established benchmark, this may not be in the best interest of shareholders. In contrast, according to Healy and Wahlen (1999), managers can also use earnings management for acting in the best interest of the shareholders. Managers are able to use earnings management to provide ‘’signals’’ to shareholders and investors (Ahmed, Takeda & Thomas, 1999). These
10
signals give shareholders and investors private information about the performance of the company.
However, using earnings management to produce a signal may harm the relationship with investors when the information turns out to be incorrect. This can decrease the ability of the shareholders to make informed decisions (Xie et al., 2003). Market efficiency is based on the information flow through the capital markets. When this information is incorrect, it is not possible for the market to determine the correct security price.
Prior research has examined the extent to which earnings management occurs around specific corporate events. Erickson and Wang (1999) find evidence of increasing earnings management in the period before both takeover and mergers. The degree of earning management is also increasing around the Initial Public Offer (IPO) and Secondary Equity Offering (SEO) (Teoh, Welch & Wong, 1998). Furthermore, according to Brown et al. (2015) earnings management is used before repurchasing stocks.
In addition, Dimitrov et al. (2015) states that managers have an incentive to manage earnings around initial credit ratings. Credit ratings influence the value of the entity because they effect the cost of future borrowing and immediate the valuation of the security (Zhao, 2017). Furthermore, entities with a good degree of creditworthiness have a high likelihood to obtain a favorable credit rating. The significant response of the capital market gives entities, included managers and shareholders, a strong incentive to improve or maintain their credit rate.
2.6 Earnings management and investment-speculative credit rating
Prior research shows earnings management is used to remain or move up credit ratings (Jiang et al, 2012). In addition, the research of Brown et al. (2015) shows the same results, but also makes a distinction between different credit rates statuses. They show the credit status is an important motivation for managers to use earnings management and that entities close to the investment-speculation grade borderline manage their earnings more aggressively compared to entities with other credit rating statuses.
There are several motivations why managers use more earnings management when the investment-speculation grade borderline is close. One is that there is a lower demand for speculative-grade securities compared to investment-grade securities. Another reason is that several institutions place restrictions on investments in speculative-grade securities, because of the higher default risks (Cantor and Packer. 1997). Already since 1931, regulations constrain banks and other financial institutes to hold debt that is below the investment-grade (Brown et al, 2015). A third reason is that downward grading from an investment-grade to a
speculative-11
grade can trigger moral hazard and may lead to liquidity problems (Dimitrov et al., 2015). For example, in the case of moral hazard a supplier could ask for the requirement of additional collateral or cash margin. Also, a bank could block the access to credit. Furthermore, severe costs are much higher for securities with a speculative-grade than for securities with an investment-grade. Increasing severe costs by downgrading also gives managers an incentive for earnings management (Brown et al., 2015).
12
3. Hypothesis
In this thesis, the aim is to examine the relation between the use of real earnings management close to the investment-speculative borderline of credit ratings. The research of Brown et al. (2015) shows entities more aggressively use real earnings management as their credit ratings approach the investment-speculative borderline in the pre-Dodd-Frank period. However, no research is conducted for the post-Dodd-Frank period. This study states hypotheses for examining these effects over multiple periods around the enforcement of the Dodd-Frank act. Hypothesis one (H1) examines whether the same results appear as the research of Brown et al. (2015) in the period post-Dodd-Frank period. Hypothesis two (H2) examines if entities close to the investment-speculative borderline more aggressively manage their earning, comparing to entities with other credit rating statuses during the post-Dodd-Frank period. Hypotheses one and two are stated as follows (in alternative form):
H1: Real earnings management is not increased as their level of credit rate of entities’ approach the investment/speculative borderline before the introduction of the Dodd-Frank act.
H2: Real earnings management is not increased as their level of credit rate of entities’ approach the investment/speculative borderline after the introduction of the Dodd-Frank act.
Prior research gives several reasons for management to increase earnings management if the investment-speculation borderline is reached (Cantor & Parker, 1997, Brown et al., 2015, and Dimitrov et al., 2015). Therefore, it is expected that the use of real earnings management will be significantly higher when the borderline is approached during both periods, while the level of significance will be lower in the post-Dodd-Frank period. The level of significance will be lower as a result of more accurate credit ratings in the post-Dodd-Frank period caused by improved monitoring of the rating methodologies and increased oversight of the SEC (Securities & Exchange Commission, 2013).
13
4. Methodology
In order to explain the relation between the dependent and independent variables, the Ordinary Least Squares (OLS) model is used for all the equations. The Roychowdhury (2006) model is used to determine the real earnings management. For estimations of this model there are two conditions. The first condition is to eliminate firms in regulated industries (SIC-code between 4400 and 5000) and bank and financial institutions (SIC-code between 6000 and 6500). The remaining companies are manufacturing entities. The Roychowdhury model is estimated by every year and industry. The second condition meets the requirement for a minimum of 15 observations for each industry-year grouping.
4.1 Models for determination of abnormal real activities
Roychowdhury (2006) uses three different measures to determine earnings management through real activities. These three measures are (1) normal levels of production costs, (2) discretionary expenses and (3) operating cashflows. The following regression equations (1-3) are used to estimate the real activities:
𝑃𝑅𝑂𝐷𝑡 𝐴𝑡−1 = 𝛽0+ 𝛽1 1 𝐴𝑡−1+ 𝛽2 𝑆𝑡 𝐴𝑡−1 + 𝛽3 ∆𝑆𝑡 𝐴𝑡−1+ 𝛽4 ∆𝑆𝑡−1 𝐴𝑡−1 + Ɛ𝑡 (1) 𝐷𝐸𝑋𝑃𝑡 𝐴𝑡−1 = 𝛽0+ 𝛽1 1 𝐴𝑡−1+ 𝛽2 𝑆𝑡 𝐴𝑡−1+ Ɛ𝑡 (2) 𝐶𝐹𝑂𝑡 𝐴𝑡−1= 𝛽0+ 𝛽1 1 𝐴𝑡−1+ 𝛽2 𝑆𝑡 𝐴𝑡−1 + 𝛽3 ∆𝑆𝑡 𝐴𝑡−1+ Ɛ𝑡 (3) Where,
PRODt = production costs in year t and is defined as the sum of the cost of goods sold and the change in
inventories divided by total assetst-1
DEXPt = discretionary expenses in year t and is defined as the sum of advertising expenses, R&D
expenses, and SG&A expenses divided by total assetst-1
CFOt = cashflow from operations in year t and is defined as operating activities from net cashflow
divided by tot assetst-1
At-1 = total assets in year t-1
St = total sales in year t
∆St = St - St-1, or sales in period t minus sales in period t – 1
14
For the estimation of the abnormal real activities, the abnormal proportions are estimated from PROD, DEXP and CFO, also see the following equations (4-6). In addition, these are the estimated residuals from equation 1 to 3.
𝐴𝐵𝑃𝑅𝑂𝐷𝑡 = 𝑃𝑅𝑂𝐷𝑡 𝐴𝑡−1 − 𝐵0− 𝐵1 1 𝐴𝑡−1− 𝐵2 𝑆𝑡 𝐴𝑡−1− 𝐵3 ∆𝑆𝑡 𝐴𝑡−1− 𝐵4 ∆𝑆𝑡−1 𝐴𝑡−1 − Ɛ𝑡 (4) 𝐴𝐵𝐷𝐸𝑋𝑃𝑡 = 𝐷𝐸𝑋𝑃𝑡 𝐴𝑡−1 − 𝛽0− 𝛽1 1 𝐴𝑡−1− 𝛽2 𝑆𝑡 𝐴𝑡−1− Ɛ𝑡 (5) 𝐴𝐵𝐶𝐹𝑂𝑡 = 𝐶𝐹𝑂𝑡 𝐴𝑡−1− 𝛽0− 𝛽1 1 𝐴𝑡−1− 𝛽2 𝑆𝑡 𝐴𝑡−1− 𝛽3 ∆𝑆𝑡 𝐴𝑡−1− Ɛ𝑡 (6)
4.2 Models for testing hypotheses
Equation 7 below is used for testing the hypotheses. Real earnings management is the dependent variable and is determined by abnormal production costs, abnormal discretionary expenses and abnormal cashflow from operation. The independent variables are the dummy variables of the S&P dataset with the credit rating statuses from AA to CCC. Ratings below CCC are not included because they are too far away from investment-speculative borderline. The rating indicator variable AAA is buried in the intercept. In total there are six independent variables included in the model:
𝐷𝐸𝑃𝑡 = 𝛽0 + 𝛽1𝐴𝐴𝑡 + 𝛽2𝐴𝑡 + 𝛽3𝐵𝐵𝐵𝑡 + 𝛽4𝐵𝐵𝑡 + 𝛽5𝐵𝑡 + 𝛽6𝐶𝐶𝐶𝑡+ 𝛽7𝑅𝑂𝐴𝑡 + 𝛽8𝐿𝐸𝑉𝑡 + + 𝛽9𝑆𝐼𝑍𝐸𝑡 + 𝛽10𝐵𝑇𝑀𝑡 + 𝛽11𝑀𝑎𝑟𝑘𝐶𝑎𝑝𝑡 + 𝛽15𝑌𝐸𝐴𝑅𝑘 + Ɛ𝑡
Where,
DEPt = ABPROD, ABDEXP, or ABCFO
AAA – CCC = An indicator variable that equals 1 if the entity has the credit rate status, otherwise 0 ROAt = Return on assets defined as income before extraordinary items divided by total assetst-1
LEVt = Leverage defined as long-term debt divided by total assetst.
SIZEt = Size defined as the natural logarithm of total assetst
BTMt = Book-to-market ratio defined as book value of equity divided by market value of equity
MarkCapt = Market capitalization defined as the natural logarithm of market value of equity
YEARk = Year and is coded 1 (2, 3) if the fiscal year falls in the pre-Dodd- Frank (financial crisis,
post-Dodd-Frank) period
4.2.1 Dependent and independent variables
Real earnings management is determined by abnormal production costs, abnormal discretionary expenses and abnormal operating cashflow. In general, the research of Brown et al. (2015) shows entities with a BBB or BB credit rating use more aggressively real earnings management.
15
Expectation for real earnings management is that in the post-Dodd-Frank period the use of it is significant lower than in the period prior to the Dodd-Frank act, which is in line with the research of Dimitrov et al. (2015). Again, the Dodd-Frank Act should increase the CRAs regulation oversight and improve accurate credit rates.
The independent variables are the credit ratings assigned by S&P from statuses AAA to CCC. According to Dimitrov et al. (2015), the regression coefficients of all rating indicator variables are significant lower in the post-Dodd-Frank period compared to the pre-Dodd-Frank period. However, independent of the period there are two expectations for the regression coefficients of the rating indicator variables. First, the regression coefficients estimated on the rating indicator variables BBB and BB are significant higher than those of other rating indicator variables in the abnormal production costs model and the abnormal cashflow model. This means entities with credit rating close to investment-speculative borderline increase their production activities and cashflows from operating as a goal to manage their earning (Kim, Kim & Song, 2013). Although the research of Brown et al. (2015) shows the coefficients in the abnormal cashflow model are not significant, expectation is a significant positive relation between cashflow from operating and the different credit rating categories, which is in line with the research of Kim et al. (2013).
Second expectation is the significant lower regression coefficients of the rating indicator variables BBB and BB compared to other rating indicator variables using the abnormal discretionary expenses model (Brown et al., 2015 and Kim et al., 2013). This means entities decrease their discretionary spending to increase earnings when their credit ratings are close to the borderline.
4.2.2 Controlled Variables
The different variables used to control for real earnings management are return on asset, amount of leverage in the company’s capital structure, book-to-market ratio, size of the company, and market capitalization. The variables size and market capitalizations are determined by the natural logarithm.
Based on the research of Brown et al. (2015), prognoses is when credit rates approach the borderline, the regression coefficient on: (1) return on assets is negative, (2) leverage is positive, (3), size is positive and (4) book-to-market ratio is positive. According to Brown et al. (2015), a negative coefficient on return on assets means an entity with a higher return on assets engage in less real earnings management. In addition, positive coefficients on leverage and book-to-market ratio means an entity with lower leverage and higher growth uses less real
16
earnings management. This is in line with the research of Roychowdhury (2006) and Cohen and Zarowin (2010).
Furthermore, Brown et al. (2015) does not relate the result of size with one of their hypotheses. They state this variable can be influenced by specific characteristics, like political costs, economies of scale and information environment. However, Blume, Lime and Mackinlay (1998) and Cohen and Zarowin (2010) state bigger entities use greater abnormal real activities as a tool for earnings management. For this reason, the variable SIZE is included in the model and a positive coefficient is expected. Market capitalization equals the total market value of an entities outstanding shares. Francis, Hasan and Li (2013) states entities depending more on the stock market engage in more real earnings management. Therefore, the expectation of variable market capitalization is a positive coefficient. Finally, the variable year is included and contains three different time periods: pre-Dodd-Frank period, financial crisis period and post-Dodd-Frank period.
4.3 Data
A dataset of S&P credit ratings and financial statement information of United States companies is obtained from the COMPUSTAT database. The data extract consists of a sample of companies which are listed in the United States and offers various information about credit ratings and real earnings management. Furthermore, the dataset has a timeframe between January 1, 2002 and December 31, 2016. For testing hypothesis one and two, a distinction between the pre- and post-Dodd-Frank period would be sufficient. However, this study divides the timeframe in three periods, namely (1) pre-Dodd Frank period before the financial crisis, (2) pre-Dodd-Frank period during the financial crisis and (3) post-Dodd-Frank period. The first period starts from January 1, 2002 until December 31, 2006 and the second period is between January 1, 2007 and December 31, 2010. Reason for creating a seperate period for the 2007-2008 financial crisis is that data during times of crisis may lead to biased results, in this case in may influence the outcomes during de pre-Dodd-Frank period. The third period starts from January 1, 2011 and ends December 31, 2016. DeHaan (2017) states this period is characterized by a period of global economic recovery until mid-2016
Raw data includes 182193 firm-year observations. In order to perform the analysis stated in this study, data exclusion is made based on non-manufacturing (SIC≤1999 and SIC≥4000) entities. The sample exclusively includes manufacturing firms because these companies can use all three real activity tools to manage earnings (Brown et al, 2015 and Zhao, 2017). Variables with missing values are also excluded from the sample which may lead to
17
certain level of selection bias. The sample size after this reduction consist of 1.202 firm-year observations with 143 unique entities. Outliers for all continuous variables are removed through winsorizing between two percent or 98 percent. The research of Cohen and Zarowin (2010), Brown et al. (2015) and Zhoa (2017) winsorized at one percent and 99 percent. However, Verardi and Croux (2008) suggest that high quality data is winsorized above one percent and lower than 99 percent.
18
5. Results
This section presents a summary of the descriptive statistics and shows the main results of the different OLS regressions of equation 7 and will answer to the hypotheses. The Pearson correlation is performed to measure multicollinearity and three OLS regressions are used to measure real earnings management for the entire timeframe. Furthermore, real earnings management is measured prior to the Dodd-Frank period, during the 2007-2008 financial crisis and in the post-Dodd-Frank period. In total, twelve different OLS regressions are performed to measure the difference in real earning management within these periods.
5.1 Descriptive summary
Table I shows the summary of statistics and the distribution for the variables of equation 7 for each credit rating group. The total average mean of abnormal production costs is -0,183. The mean of abnormal production costs is monotonically increasing when the rating indicator variables decrease from AAA to CCC, with average means from -0,604 to 0,163. This outcome is similar to that of the research of Roychowdhury (2006) and Dimitrov et al. (2015). Furthermore, the summary statistics show a total average mean of -0,247 for the abnormal discretionary expenses and a total average mean of 0,317 for abnormal cashflow of operations. The abnormal discretionary expenses are increasing and the abnormal cashflows are decreasing when the credit rates become riskier. These results are in contrast with the theory, which states real earnings management occurs with decreasing abnormal discretionary expenses and increasing abnormal cashflows of operating Roychowdhury (2006). However, the research of Brown et al. (2015) shows the same results.
The return on asset has a total average of 6 percent and decreases when the credit rating becomes riskier. The leverage ratio has a total average of 22,2 percent and monotonically increases when the credit rating decreases from AAA to CCC. According to Cohen and Zarowin (2010) entities tend to increase their leverage when the credit rate become riskier and therefore engage more in real earnings management. The maximum amount of leverage ratio is 56,4 percent and the lowest is null, meaning there are entities within the dataset that have no leverage or debt. All credit rating categories include companies without leverage, except for the AAA rating category. The book-to-market ratio has a total average of 38,4 percent and is generally positive, except for the CCC rating. The book-to-market ratio increases monotonically with credit risk, this is probably due to the low market values of the low rated stocks. Size has a total
19
average mean of $9,182 million. The size of the companies is negatively related to the rating indicator variables, which means entities with riskier credit ratings are smaller. This result is similar to Brown et al. (2015). The total average mean of market capitalization is $27,2 billion and decreases monotonically with risk. The highest amount of market capitalization is $184,5 billion where the lowest is $0,2 billion.
Table II presents the distribution of the observations sort by rating indicator variables. The rating indicator variable A contains the highest percentages of observations, namely 34,5
Table I: Summary Statistic – All Variables
Mean and Standard Deviation of variables used in equation 7, by credit ratings Rating variable Observations AAA 22 AA 109 A 415 BBB 315 BB 171 B 123 CCC 11 Total 1.202 ABCFO Mean 0,904 0,606 0,349 0,264 0,182 0,218 -0,040 0,317 S.D. 0,481 0,577 0,531 0,502 0,459 0,481 0,281 0,528 Min. 0,291 -0,781 -0,781 -0,781 -0,781 -0,781 -0,781 -0,781 Max. 1,964 2,047 2,047 2,047 2,047 2,047 0,314 2,047 ABEXP Mean -0,647 -0,429 -0,224 -0,240 -0,210 -0,170 -0,154 -0,247 S.D. 1,016 0,891 0,618 0,542 0,528 0,575 0,236 0,623 Min. -2,919 -2,919 -2,919 -2,919 -2,919 -2,919 -0,702 -2,919 Max. 0,296 0,955 0,955 0,955 0,955 0,919 0,270 0,955 ABPROD Mean -0,604 -0,412 -0,201 -0,127 -0,115 -0,130 0,163 -0,183 S.D. 0,731 0,714 0,486 0,527 0,577 0,511 0,154 0,550 Min. -3,103 -3,103 -3,103 -3,103 -3,103 -3,103 -0,002 -3,103 Max. -0,005 0,604 0,778 0,778 0,778 0,778 0,420 0,778 ROA Mean 0,128 0,100 0,086 0,058 0,031 -0,015 -0,069 0,060 S.D. 0,047 0,064 0,056 0,047 0,075 0,092 0,049 0,072 Min 0,035 -0,051 -0,146 -0,146 -0,146 -0,146 -0,146 -0,146 Max 0,207 0,207 0,207 0,207 0,207 0,207 -0,014 0,207 LEV Mean 0,076 0,151 0,184 0,250 0,242 0,315 0,394 0,222 S.D. 0,031 0,121 0,110 0,110 0,157 0,175 0,159 0,136 Min 0,029 0,000 0,000 0,000 0,000 0,000 0,000 0,000 Max. 0,159 0,538 0,564 0,564 0,564 0,564 0,564 0,564 BTM Mean 0,274 0,424 0,375 0,392 0,484 0,331 -0,648 0,384 S.D. 0,094 0,348 0,281 0,230 0,423 0,466 0,165 0,337 Min. 0,106 -0,005 -0,065 -0,157 -0,698 -0,698 -0,698 -0,698 Max. 0,481 1,233 1,233 1,233 1,233 1,233 -0,149 1,233 SIZE Mean 11,427 10,468 9,610 9,142 8,166 7,816 8,126 9,182 S.D. 0,391 1,008 1,173 1,192 1,237 1,586 1,745 1,461 Min. 10,610 8,737 7,434 6,657 6,066 6,066 6,221 6,066 Max. 11,858 12,144 11,930 12,144 12,144 12,144 12,144 12,144 MarkCap Mean 12,066 10,878 9,924 9,086 7,715 7,133 6,624 9,175 S.D. 0,104 0,955 1,063 0,981 1,168 1,250 1,330 1,587 Min. 11,691 8,533 6,915 5,952 5,304 5,304 5,304 5,304 Max. 12,125 12,125 12 ,125 11,573 11,024 11,059 8,610 12,125
The summary includes 1.202 firm-year observations with 143 unique companies firm-year from 2002 to 2016. The data is extracted from COMPUSTAT database. Non-manufacturing firms (SIC ≤1999 and SIC≥2000) and missing values are excluded. The Roychowdhury (2006) model is used to estimate the ABPROD, ABNEXP and ABCFO, which are the abnormal production costs, abnormal discretionary expenses and abnormal cashflow from operation, which are the residuals from a regression of production costs, discretionary expenses and operating cashflows. The credit rate status by Standard & Poor range from AAA to CCC and are binary dummies that equals one (zero otherwise) if it contains the status. ROA is return on assets and is defined as income before extraordinary items divided by beginning of period total assets. LEV is amount of leverage in the company’s capital structure, calculated by total long-term debt divided by total assets. BTM is the book-to-market ratio, where the book value of equity is divided by market value of equity. Size of the companies is defined as natural logarithm of total assets. MarkCap is the natural logarithm of market value of equity and is estimated as the closing price at fiscal year-end times the number of shares outstanding at fiscal year-end. Each of the dependent and continuous variables are winsorized at 2% and 98%.
20
percent and is followed by credit rate BBB with 29,2 percent. The research of Brown et al. (2015) shows the highest percentage of observations for the rating indicator variable BBB, followed by AAA. Furthermore, the rating indicator variables A and BBB are followed from high to low by rating indicator variables BB (14,2 percent), B (10,2 percent), AA (9,1 percent), AAA (1,8 percent) and CCC (0,9 percent). This distribution is similar to the research of Brown et al. (2015). According to S&P (2016) this distribution is approximately normal.
When the distribution of the observations is divided into the three different periods, the distribution is approximately similar to the overall distribution of the sample. However, during the period of the 2007-2008 financial crisis the rating indicator variables AAA, AA, BBB and BB decrease while A, B and CCC increase. The overall downwards trend could be explained by the economic downturn while the increase of almost 6 percent of rating indicator variable A could be explained by the investor-pay model (Jiang et al.,2012). BBB and BB rated entities effected by the crisis could pay the CRAs higher fees to increase the credit rating.
The comparison between the pre-Dodd-Frank period and post-Dodd-Frank period shows an overall decrease in the higher rating indicator variables (AAA and AA), while lower rating indicator variables tend to increase (A, BBB, BB and CCC), except for credit rating B. This is in line with the research of Dimitrov et al. (2015) and underwrites the prudence of CRA’s by offering credit ratings since the introduction of de Dodd-Frank act.
Table II: Distribution of Sample Firm-Years by Credit Rating Categories
Total AAA AA A BBB BB B CCC Number 1.202 22 109 415 351 171 123 11 Percentages 100% 1,8% 9,1% 34,5% 29,2% 14,2% 10,2% 0,9% Year = 1 Number 431 10 46 133 132 61 47 2 Percentages 100% 2,3% 10,7% 30,9% 30,6% 14,2% 10,9% 0,5% Year = 2 Number 318 6 26 116 78 43 45 4 Percentages 100% 1,9% 8,2% 36,5% 24,5% 13,5% 14,2% 1,3% Year = 3 Number 453 6 37 166 141 67 31 5 Percentages 100% 1,3% 8,2% 36,6% 31,1% 14,8% 6,8% 1,1%
The summary includes 1,202 firm-year observations with 143 unique companies firm-year from 2002 to 2016. The data is extracted from COMPUSTAT database. Non-manufacturing firms (SIC code below 2000 and above 3999) are excluded. Year has a scale from one until three, where one stands for the pre-Dodd-Frank period between 2002 to 2006, two stands for pre-Dodd-Frank period during the 2007-2008 financial crisis between 2007 and 2010, and three stands for the Post-Dodd-Frank period between 2011 and 2016.
21
5.2 Pearson correlation
Table III shows the Pearson correlation matrix. The correlation coefficient presents the strength and direction of two variables move together and in which direction. The coefficient is ranged between minus and plus one. There is no correlation when the coefficient is zero, a perfect correlation occurs when the coefficient is (minus) one.
Farrar and Glauber (1967) states the correlation between the variables should be smaller than (minus) 0,8. When the correlation coefficient is 0,8 or above, multicollinearity occurs, meaning inter-association could occur among independent variables. The correlation matrix shows the highest occurring correlation is -0,435, which is between the two variables credit rate category B and market capitalization. No correlation is above the threshold of (minus) 0,8 and therefore no problem for multicollinearity.
Table III: Pearson correlation matrix
ABPROD ABEXP ABCFO AAA AA A BBB BB
ABPROD 1,000 ABEXP -0,032 1,000 ABCFO -0,475*** -0,465*** 1,000 AAA -0,105*** -0,088*** 0,152*** 1,000 AA -0,132*** -0,093*** 0,173*** -0,043 1,000 A -0,024 0,027 0,044 -0,099*** -0,229*** 1,000 BBB 0,066** 0,007 -0,064** -0,088*** -0,203*** -0,466*** 1,000 BB 0,050* 0,024 -0,105*** -0,056 -0,129*** -0,296*** -0,262*** 1,000 B 0,033 0,042 -0,063** -0,046 -0,107*** -0,245*** -0,217*** -0,138*** CCC 0,061** 0,014 -0,065** -0,013 -0,030 -0,070** -0,062** -0,039 ROA -0,174*** -0,049* 0,179*** 0,129*** 0,174*** 0,261*** -0,017 -0,167*** LEV -0,043 -0,065** 0,005 -0,146*** -0,165*** -0,201*** 0,132*** 0,061* BT 0,122*** 0,036 -0,098*** -0,045 0,037 -0,020 0,015 0,121*** SIZE -0,014 -0,048* 0,114*** 0,210*** 0,278*** 0,213*** -0,017 -0,283*** MarkCap -0,132*** -0,094*** 0,225*** 0,249*** 0,338*** 0,343*** -0,036 -0,375***
B CCC ROA LEV BTM SIZE MarkCap
B 1,000 CCC -0,032 1,000 ROA -0,357*** -0,174*** 1,000 LEV 0,230*** 0,121*** -0,154*** 1,000 BT -0,054* -0,295*** -0,246*** -0,321*** 1,000 SIZE -0,316*** -0,070** 0,084*** -0,214*** 0,116*** 1,000 MarkCap -0,435*** -0,155*** 0,421*** -0,265*** -0,071** 0,850*** 1,000
The correlation table includes 1.202 firm-year observations with 143 unique companies firm-year from 2002 to 2016. The data is extracted from COMPUSTAT database. Non-manufacturing firms (SIC ≤1999 and SIC≥2000) and missing values are excluded. The Roychowdhury (2006) model is used to estimate the ABPROD, ABNEXP and ABCFO, which are the abnormal production costs, abnormal discretionary expenses and abnormal cashflow from operation. They are the residuals from a regression of production costs, discretionary expenses and operating cashflows. The credit rate status by Standard & Poor’s range from AAA to CCC and are binary dummies that equals one (zero otherwise) if it contains the status. ROA is return on assets and is defined as income before extraordinary items divided by beginning of period total assets. LEV is amount of leverage in the company’s capital structure, calculated by total long-term debt divided by total assets. BTM is the book-to-market ratio, where the book value of equity is divided by market value of equity. Size of the companies is defined as natural logarithm of total assets. MarkCap is the natural logarithm of market value of equity and is estimated as the closing price at fiscal year-end times the number of shares outstanding at fiscal year-end. Each of the dependent and continuous variables are winsorized at 2% and 98%.
22
5.3 Ordinary Least Square for the overall period
Table IV shows the results of the OLS regressions of equation 7 for the overall period between January 1, 2002 and December 31, 2016. Hereby, the AAA rating indicator variable is buried in the intercept. In total three different regression analyses are performed. The dependent variable of the first OLS regression is abnormal production costs, for the second it is abnormal discretionary expenses and the third uses abnormal cashflows. Together these three dependent variables determine real earnings management. Panel B shows the t-test results comparing the regression coefficients on pairs of rating indicator variables.
The abnormal production costs model shows a positive significant relation with the rating indicator variables from A until CCC, including the intercept. This means abnormal production costs is used for managing earnings. The monotonic relation between credit ratings and the abnormal production costs do not disappear when other incentives for earnings management are controlled. Results further show the regression coefficients for the rating indicator variables BBB and BB are not the highest. This could mean that entities with a credit rating on the investment-speculative borderline do not increase their production costs more aggressively than entities with other ratings.
The results of the abnormal discretionary expenses model show the coefficients of the rating indicator variables are positive and significant at a five percent level, except for rating indicator variables AA and CCC. Ratings BBB and BB both have an increasing effect on discretionary expenses meaning there is no increase in managing earnings. Expectations were rating indicators closer to the border line should aggressively decrease the discretionary expenses for managing their earnings and therefore the coefficients must be negative.
Furthermore, all rating indicator variables of the abnormal cashflow model are significant at a one percent level. The regression coefficients of rating indicator variables AA, A, BBB and CCC are negative. BB and B are positive. Close to the borderline, BB rated entities seem to manage their earnings by cashflows. Prior research also states entities with lower rating indicator variables should increase their cashflow to manage their earnings, especially companies with ratings at the investment-speculative borderline (Brown et al., 2015)
The assumption of higher leverage increases real earnings management is tested in the third regression analysis. For the control variables results show leverage has a negative coefficient on abnormal production costs and abnormal discretionary expenses while the coefficient on abnormal cashflow from operation is positive. Overall, results state companies with higher leverage engage more in real earnings management. This is in line with Roychowdhury (2006) and Brown et al. (2015).
23
Table IV: OLS Regression Analyses all periods
Estimation results of the OLS regression of abnormal real operating activities on credit rating status over all the periods
ABPROD ABEXP ABCFO
Panel A: OLS-regression results of equation 7
AA 0,178 0,239 -0,260*** A 0,390*** 0,440*** -0,419*** BBB 0,441*** 0,429*** -0,539*** BB 0,391*** 0,413*** 0,536*** B 0,377** 0,475*** 0,488*** CCC 0,662*** 0,429 -0,713*** ROA -0,252 0,370 0,060 LEV -0,044*** -0,573*** 0,363*** BTM 0,064 -0,056 -0,029 SIZE 0,107*** 0,075** -0,087*** MarkCap -0,109*** -0,095*** 0,119*** Intercept -0,458** -0,352 0,434*** N 1,202 1,202 1,202 R2 0,074 0,028 0,098 F 9,71*** 4,10*** 12,81*
T-Test: ABPROD ABEXP ABCFO
Panel B: Test of difference in the coefficient estimates (p-values are presented)
BBB = AA 0,000 0,012 0,000 BBB = A 0,223 0,813 0,232 BBB = BB 0,352 0,801 0,955 BBB = B 0,337 0,548 0,419 BBB = CCC 0,233 0,999 0,312 BB = AA 0,009 0,066 0,000 BB= A 0,987 0,695 0,433 BB = B 0,838 0,418 0,444 BB = CCC 0,140 0,941 0,308
The regression table includes 1.202 firm-year observations with 143 unique companies firm-year from 2002 to 2016. The data is extracted from COMPUSTAT database. Non-manufacturing firms (SIC ≤1999 and SIC≥2000) and missing values are excluded. The Roychowdhury (2006) model is used to estimate the ABPROD, ABNEXP and ABCFO, which are the abnormal production costs, abnormal discretionary expenses and abnormal cashflow from operation. They are the residuals from a regression of production costs, discretionary expenses and operating cashflows. The credit rate status by Standard & Poor’s range from AAA to CCC and are binary dummies that equals one (zero otherwise) if it contains the status. ROA is return on assets and is defined as income before extraordinary items divided by beginning of period total assets. LEV is amount of leverage in the company’s capital structure, calculated by total long-term debt divided by total assets. BTM is the book-to-market ratio, where the book value of equity is divided by market value of equity. Size of the companies is defined as natural logarithm of total assets. MarkCap is the natural logarithm of market value of equity and is estimated as the closing price at fiscal year-end times the number of shares outstanding at fiscal year-end. Each of the dependent and continuous variables are winsorized at 2% and 98%.
The variable size is significant positive in the abnormal production costs model and abnormal discretionary expenses model. Size is significant negative on abnormal cashflow from operation. Overall the dependent variable size shows mixed results. Prior research is also mixed about the results. The negative significant of size on abnormal cashflows is in line with the research of Gu, Lee and Rosett (2005). Gu et al. (2005) states larger companies have a greater benefit of economy of scale and scope, while they bear more costs when discovered in managing earning because they are politically sensitive. While other studies can state bigger entities use greater abnormal real activities as tool for earnings management (Blume et al., 1998 and Cohen & Zarowin, 2010).
Finally, results show there is a negative significant relation between market capitalization and both dependent variables abnormal production costs and abnormal
24
discretionary expenses, while there is a positive significant effect on cashflow from operation. Entities that depend more on the stock market and have a higher market value of equity engage more in real earnings management. Existing literature is in line with the results of market capitalization on the abnormal discretionary expenses. Francis et al. (2013) states firms that depend more on the stock market, and therefore have a higher market value of equity, use more real earnings management.
The table further shows the adjusted R-squared (R2) of 7,4 percent for the abnormal production costs model, R2 of 2,8 percent for the abnormal discretionary expenses model and R2 of 9,8 percent for the abnormal cashflow model. The R2 is a corrected goodness-of-fit measure for linear regression models and is modified from the R-squared adjusted for the number of variables. This makes that the R2 only increases if newly added variables improve the model. The F-statistic for the three regressions are significant at a level of one percent.
In Panel B, the difference in the coefficients of the credit rating pairs estimated is tested. Results show the majority of the pairs differ from each other, because they are not significant. Hereby, there is no indication to conclude BBB and BB companies engage more aggressive real earnings management than the other credit rating pairs.
Based on the above results, the variables close to the investment-speculative borderline have a significant effect on earnings management by increasing the production. However, discretionary expenses also significantly increase while a decrease is expected. Both ratings have a significant effect on managing earnings through cashflow from operating. Rating BBB has a decreasing effect while rating BB shows an increasing effect. Next to these borderline results, it does not seem other credit ratings act different. Also, the credit ratings pairs estimated are tested. Because there is no significance, the pairs do not differ from each other. Therefore, the statement: ratings close to the borderline more aggressively use earnings management cannot be made.
5.4 Ordinary Least Square for the three periods
Table V shows the OLS regression results for the abnormal production costs model, abnormal discretionary expenses model and abnormal cashflow model divided into the three different periods. The first period tested is the pre-Dodd-Frank period which starts from January 2002 and ends in December 2006. This period consists of 431 firm-year observations.
During this period, the abnormal costs model shows the coefficients for return on assets and leverage are negative significant at a five percent level. The abnormal discretionary model shows the coefficients of the rating indicator variables A, BBB, BB and B are positive at a level
25
of significance of one percent. Rating indicator variable AA is positive significant at ten percent. Control variables leverage (size) is negative (positive) significant at the 10 percent level.
Furthermore, for abnormal cashflow model, rating indicator variable BBB is negative significant at the one percent level; A, BB, B at five percent and CCC at ten percent. Variable leverage and return on assets are positive significant at respectively the one and five percent level. Results indicate rating category has no effect on managing earnings through abnormal costs, higher return on assets and leverage have a decreasing effect, meaning less earnings management. Furthermore, all rating categories except for category C do not show an increasing effect of earnings management by cutting expenses. It shows the opposite decreasing effect. Higher amount of leverage seems to increase this type of earnings management while the size of entities has a decreasing effect. For managing earnings trough cash flows, all rating categories, except that of rating AA have a decreasing effect. Higher return on assets and leverage do have an increasing effect.
The second pre-Dodd-Frank period starts from January 2007 and ends in December 2010 and includes a period of financial crisis. This period consists of 318 firm-year specific observations. Regarding the abnormal production costs model, the BBB credit rating is the only rating category which is significant. It is positive at a 10 percent level. The coefficients of the control variables leverage and market capitalization are negative significant while size is positive significant on the one percent level. The discretionary expenses model shows the coefficients of return on asset and leverage are negative and significant at a five percent level, size is negative significant at a one percent level while market capitalization is positive significant at a one percent level. No rating indicator variable is significant. The abnormal cashflow model for this period shows a negative significant coefficient relation between abnormal cashflow and the rating indicator variables AA, A, BBB and B, where A and BBB are significant at the one percent level, B at the five percent and AA at the ten percent. The control variables leverage and market capitalization (size) are positive (negative) at a one percent level.
Results show during times of crisis, entities with a rating at the investment borderline (BBB) seem to manage earnings through abnormal costs to ensure the investment grade remains, which is in line with the expectation. All other ratings do not differ significant from the pre-crisis period. More leverage and higher market capitalization tend to decrease this type of earnings management while the entities’ size show the opposite result. Also, in contact to the pre-crisis period, no rating category has a significant effect on the decrease of expenses. Higher
26
return on assets, leverage and size seem to increase this type of earnings management while a higher book-to-market value ratio of entities has a decreasing effect. For managing earnings through cash flows, rating category BB has no significant effect. Higher leverage and market capitalization do have an increasing effect while a bigger size decreases the managing of cash flow earnings.
The third period contains the post-Dodd-Frank period starting from January 2011 and ends in December 2016. This period consists of 453 firm-year observations. The rating indicator variables A, BBB, BB and CCC are positive significant at the one percent level, B is positive significant at a five percent level and AA is significant positive at the ten percent level for the abnormal production costs. Furthermore, the book-to-market ratio is positive significant at a level of significant of five percent. Regarding the abnormal discretionary model, the rating indicator variables A and BBB are positive significant at respectively the five and ten percent level. Control variables leverage and market capitalization are negative significant at the level of five and one percent. Size is positive significant at one percent level. For the abnormal cashflow model ratings A and BBB are negative significant at a five percent level, BB is negative significant at the ten percent level. Leverage, book to market value and market capitalization are positive significant at a one percent level of significance while size is negative significant at a one percent level.
Results of the post-Dodd-Frank period show there is a positive significant effect of entities with all sort of rating categories on the abnormal production costs. This means entities with any tested rating increase their production. However, ratings close to the borderline of investment grades do not differ from each other. It is therefore more likely that the economic state makes more companies invest in production than that they are managing their earnings. For the control variables, entities with a higher book to market value tend to engage more in abnormal production. For managing earnings by decreasing the expenses, ratings A and BBB show positive significant coefficient. Other rating coefficients are also positive while the control variables are negative (leverage, size and market capitalization are significant). It therefore seems that the rating category does not have a decreasing effect on expenses while most control variables have. For managing earnings through cash flows, both rating category near the borderline as well as category A have a significant negative effect. Most control variables have a significant positive effect. Evidence shows earnings management through cashflows is negative for ratings, the control variables do have an effect.
27
Table V: OLS Regression Analyses in different periods
Estimation results of the OLS regression of abnormal real operating activities on credit rating status for three different periods, namely (1) the pre-Dodd-Frank period, (2) the period of the 2007-2008 financial crisis and (3) the post-Dodd-Frank period.
ABPROD ABEXP ABCFO
Year 1 2 3 1 2 3 1 2 3
Panel A: OLS-regression results of equation 7
AA 0,352 0,102 0,218* 0,700* 0,154 0,228 -0,341 -0,128* -0,371 A 0,625 0,152 0,494*** 1,037*** 0,158 0,574** -0,476** -0,209*** -0,597** BBB 0,827 0,229* 0,573*** 1,122*** 0,205 0,514* -0,616*** -0,246*** -0,548** BB 0,764 0,139 0,515*** 1,339*** 0,127 0,351 -0,564** -0,122 -0,531* B 0,643 0,149 0,526** 1,447*** 0,231 0,103 -0,541** -0,197** 0,065 CCC 1,187 0,296 0,863*** 1,256 0,077 0,435 -0,893* 0,008 -0,351 ROA -3,41** -0,367 0,438 0,219 -0,681** 1,096 1,186** 0,053 -1,037 LEV -1,763** -0,353*** -0,206 -0,757* -0,378** -0,618** 0,651*** 0,225*** 0,976*** BTM -0,052 -0,029 0,223** -0,219 0,024 -0,136 -0,025 0,042 0,49*** SIZE 0,125 0,156*** 0,048 0,13* -0,097*** 0,215*** -0,007 -0,103*** -0,339*** MarkCap -0,093 -0,152*** -0,06 -0,029 0,101*** -0,333*** 0,025 0,126*** 0,499*** Intercept -0,887 -0,206 -0,596* -2,093*** -0,151 0,437 0,620* 0,104 -1,115** N 431 318 453 431 318 453 431 318 453 R2 0,053 0,219 0,107 0,020 0,046 0,095 0,073 0,269 0,174 F 2,130** 9,100*** 5,920*** 1,810** 2,390** 5,330*** 4,090*** 11,59*** 9,630***
ABPROD ABEXP ABCFO
Year 1 2 3 1 2 3 1 2 3
Panel B: Test of difference in the coefficient estimates (p-values are presented)
BBB=AA 0,167 0,066 0,000 0,0,42 0,515 0,038 0,016 0,006 0,185 BBB=A 0,399 0,072 0,100 0,555 0,340 0,446 0,080 0,153 0,521 BBB=BB 0,836 0,091 0,371 0,240 0,196 0,125 0,612 0,000 0,873 BBB=B 0,604 0,192 0,653 0,131 0,700 0,016 0,527 0,189 0,000 BBB=CCC 0,793 0,685 0,177 0,872 0,498 0,823 0,545 0,013 0,566 BB=AA 0,358 0,649 0,005 0,018 0,768 0,483 0,134 0,900 0,344 BB=A 0,689 0,818 0,771 0,149 0,623 0,063 0,445 0,015 0,573 BB=B 0,734 0,871 0,907 0,618 0,138 0,120 0,845 0,053 0,001 BB=CCC 0,754 0,337 0,101 0,919 0,792 0,809 0,465 0,198 0,593
The regression table includes 1,202 firm-year observations with 143 unique companies firm-year from 2002 to 2016. The data is extracted from COMPUSTAT database. Non-manufacturing firms (SIC code below 2000 and above 3999) are excluded. The Roychowdhury (2006) model is used to estimate the ABPROD, ABNEXP and ABCFO, which are the abnormal production costs, abnormal discretionary expenses and abnormal cashflow from operation, which are the residuals from a regression of production costs, discretionary expenses and operating cashflows. The credit rate status by Standard & Poor’s range from AAA to CCC and are binary dummies that equals one (zero otherwise) if it contains the status. ROA is return on assets and is defined as income before extraordinary items divided by beginning of period total assets. LEV is amount of leverage in the company’s capital structure, calculated by total long-term debt divided by total assets. BTM is the book-to-market ratio, where the book value of equity is divided by market value of equity. Size of the companies is defined as natural logarithm of total assets. MarkCap is the natural logarithm of market value of equity and is estimated as the closing price at fiscal year-end times the number of shares outstanding at fiscal year-end. Each of the dependent and continuous variables are winsorized at 2% and 98%.
During the 2007-2008 financial crisis, the abnormal production model has a R2 of 0.219 meaning that abnormal cashflow model is explain approximately 21,9 percent of the variation in abnormal production costs. Furthermore, an R2 of 0.269 indicates that the abnormal cashflow model is explained approximately 26,9 percent of the variation in abnormal cashflow. The other amounts of R2 are below the 0,2 and are considered to be low (Brown et al., 2015). The F-statistics for all nine models is significant.
In panel B, the majority of the t-test results comparing the coefficients on the rating indicator variables are not significant. This means that the coefficients of rating indicator variables pairs differ from each other. In contrast to the research of Brown et al (2015), the panel does not indicate that income-increasing real earnings management is the strongest for
28
BBB and BB companies. There is also no difference between the different periods in the abnormal production costs model, abnormal discretionary model and abnormal cashflow model. Comparing the different periods between the three models indicates also no differences. In sum, the results of the regression in the different period show not more aggressive real earnings management in the vicinity of the investment-speculative.
Results for the ratings close to the investment-speculative borderline show a significant effect on earnings management by increasing the production costs during the post-Dodd-Frank period. Discretionary expenses significantly increase during the non-financial crisis pre-Dodd-Frank period, while a decrease is expected. Like expected, both ratings have a significant negative effect on managing earnings through cashflows during all periods, expect for credit rating category BB during times of crisis (pre-Dodd-Frank). Like the results of the overall period, entities with other credit rating categories do not seem to act different. Also, the credit ratings pairs estimated again show no significance, the pairs do not differ from each other. Despite some effects were seen as expected, the alternative form of hypothesis one and two cannot be reject. The statements: ratings close to the borderline more aggressively use earnings management pre- or post-Dodd frank act cannot be made.
