The Relationship Between Real Earnings Management
and the Cost of Debt:
Differences Between Local firms and Cross-Listed
Firms in the American Market
MSc Thesis
Business Administration – Organizational & Management Control - 2018-2019
University of Groningen, Faculty of Economics and Business
Supervisor: S. Wang
Word count: 12,449 words
Hidde van Lent – S2483033
Hereweg 118
9725 AK Groningen
+31623422666
h.h.l.van.lent@student.rug.nl
Abstract:
1. Introduction
In August 2018, U.S. president Donald Trump started a debate on the financial reporting frequency of U.S. companies (Bloomberg, 2018). In a tweet, Trump announced he asked the Securities and Exchange Commission to revise the current obligation for U.S. companies to report financials quarterly and move to a semi-annual system. If the U.S. would shift to a semi-semi-annually reporting system it would follow Europe. The current difference between reporting frequency is just one of many differences between European and U.S. stock exchanges. Because of these differences, it could be beneficial for companies to be traded on two different stock exchanges: cross-listing. Whereas Warren Buffet and Jamie Dimon support this idea as it focuses on executing a long-term strategy that maximizes shareholder value (Reuters, 2018), Robert Pozen mentions that reducing the reporting frequency reduces transparency towards the investors (Bloomberg, 2018).
Companies often have to make the trade-off between short-term earnings and long-term strategy. One of the tools that managers use to adapt financial reports to their needs is earnings management (EM). EM is “the use of managerial discretion (within GAAP) over accounting choices, earnings reporting choices, and real economic decisions to influence how underlying economic events are reflected in one or more measures of earnings.” (Walker, 2013, p. 446). Although earnings management via accounting choices was most used in the twentieth century, the passage of Sarbanes-Oxley (SOX) in 2002 lead to a shift towards managing earnings via real economic decisions (Cohen, Dey & Lys, 2008).
According to Roychowdhury (2006), real earnings management (REM) occurs via offering more lenient terms or price discounts to customers, overproduction to reduce the current COGS, and a reduction in discretionary expenses. Whereas these actions benefit a company’s earnings on the short term, they will negatively affect a firm’s cash flows in the future (Cohen et al., 2008; Graham, Harvey & Rajgopal, 2005). Offering more lenient terms or price discounts to customers, sales manipulation, will negatively affect future cash flows, which will harm the future debt repayment of the company. Overproduction will lead to increased future costs and a reduction in discretionary expenses, via a reduction in research and development (R&D) and selling, general and administrative (SG&A) expenses, will reduce the innovation ability of a company in the future. Because of these consequences, credit rating agencies and lenders will perceive earnings management as value-destroying. Therefore, theoretically, the relationship between REM and cost of debt should be positive. However, previous research did not always underpin this. Moreover, if this was the case, there would most likely be much less REM used than there currently is.
The relationship between accruals-based earnings management (AEM) and cost of debt is clearly defined as positive: increased AEM increases the cost of debt. However, REM is more difficult to detect than AEM (Cohen et al., 2008), which could be the reason why this relationship in practice differs from what theory predicts. On this relationship has little research been conducted, and the results have been contradicting. Ge & Kim (2014) showed a positive relationship between REM and cost of debt in the American market, as credit rating agencies perceive REM as managerial opportunism, whereas Kwak & Park (2015) showed a contrary relationship in the Korean market.
especially since one of their conclusions is that country-specific factors influence this relationship, which is likely to be the reason why their outcomes are similar to Ge & Kim (2014). Mellado-Cid, Jory & Ngo (2017) also conducted research on the U.S. market, yet they found a negative relationship between REM and cost of debt. They assigned this opposing view to a difference in data sample and sources. The conflicting results of these researches lead to the first research question of this paper:
What is the effect of real earnings management on the cost of debt in the U.S. market?
The conflicting results might be a country-related issue, as country-specific factors seems to be influential. Moreover, the relationship in the U.S. market for cross-listed firms has not yet been investigated. Firms cross-list on the U.S. exchange to, amongst others, attract cheaper external finance (Errunza & Miller, 1999). Cross-listing improves a firm’s access to lower-cost external capital (e.g. Khurana, Martin & Periera, 2007; Pagano, Röell & Zechner, 2002). Errunza and Millar (2000) find a substantial decrease in a firm’s cost of capital following an American depository receipt (ADR). According to Citigroup (2011), the establishment of an ADR program and the increased scrutiny by market forces and regulators that follow lead to a decrease in the cost of debt. If firms indeed cross-list to reduce the cost of capital, incentives to manage earnings will be higher. Moreover, cross-listed firms are less well-known to U.S. investors, thus to obtain cheap external finance there is a greater need to provide better financial statements, furtherly increasing incentives to manage earnings.
However, the increased scrutiny also reduces possibilities to manage earnings. Indeed, if firms cross-list following the “bonding hypothesis” of Coffee (2002), namely to subject themselves to the U.S. high disclosure standards and greater threat of enforcement by the SEC, the possibilities to engage in REM might decrease. In line with this, Berger, Li & Wang (2005) find that the unfamiliarity of U.S. investors might lead to stricter monitoring of cross-listed firms, which will furtherly reduce possibilities to manage earnings. However, the absence of analyst coverage might provide more opportunities to conduct REM (Enomoto, Kimara & Yamaguchi, 2015).
Kim, Son & Usmen (2010) find that, post-SOX, U.S. firms became less extreme in their accounting choices, whereas opportunistic accounting discretion of cross-listed firms remained unchanged, as cross-listed firms have been virtually free from the strict regulatory enforcement and hostile legal exposure of SOX. This lead to decreased AEM for American firms compared to their cross-listed counterparts, and therefore might also affected the REM and consequences of REM for both types of companies. The incentives of cross-listed firms to cross-listed and the different incentives and possibilities for American and cross-listed companies to engage in REM can lead to differences in the relationship between REM and the cost of debt, leading to the second research question of this paper:
How does the relationship between real earnings management and cost of debt in the American market differ between local firms and cross-listed firms?
cross-listed firms and to investigate differences in the relationship across local firms and cross-listed firms. Hereby, the paper can provide evidence whether country-specific factors influence the level of REM and how it affects the relationship that REM has with other variables, such as the cost of debt. Thirdly, the paper will mainly take into account the post-SOX period, which is the most important period for REM due to the passage of SOX (Cohen et al., 2008). The outcomes are highly valuable for managers of both U.S. and American firms, as it will provide insights into the benefits and costs of REM, but also for lenders and credit rating agencies, as it provides them with a clear view whether their current cost schedules incorporate all financial risks.
According to Roychowdhury (2006), REM is the sum of sales manipulation, reduction in discretionary expenses and overproduction. However, more recent REM literature states that a reduction in discretionary expenses and overproduction already affect the cash flow, hence including sales manipulation (calculated as abnormal cash flow) will take into account this effect twice. Therefore, in line with more recent literature, i.e. Zang (2011), this paper uses a comprehensive measure of REM via overproduction and discretionary expenses.
As the effect of the comprehensive measure may dilute different implications of the separate measures of earnings (Cohen & Zarowin, 2010), the three measures of REM will also be used individually. The cost of debt, as of year t+1, is measured as interest expense over average debt, in line with Francis, Olsson & Schipper (2008), and as credit rating, consistent with Ge & Kim (2014) and Jiang (2008). All regressions are controlled for firm size, growth opportunities, profitability, leverage, auditor, lossability, industry and year. The paper uses a sample of 65,804 firm-year observations, equally split amongst cross-listed firm-year observations and on size, industry and year matched American firm-year observations. The main regressions are run on a final sample of 14,081 firm-year observations: 8,619 American firm-year observations and 5,461 cross-listed firm-year observations.
The expected increased incentives and probabilities of cross-listed companies to engage in REM are confirmed: the level of REM for cross-listed companies is, after including control variables, almost double the size of their American counterparts. Whereas the comprehensive REM measure is insignificant on the relative interest expense, the separate measures lead to higher (via sales manipulation and overproduction) or lower (via a reduction in discretionary expenses) relative interest expense. Moreover, REM leads to a worse credit rating. The relationship between REM and cost of debt is in line with expectations, and confirms the positive relationship found by Ge & Kim (2014) and Kim et al. (2018).
To test for differences in this relationship across both samples, the regressions are run with a full factorial variable (including a dummy variable that takes value of one if the company is cross-listed), and the regressions are run separately for both samples after which the coefficients are compared via a suest test. These tests find that there are differences in the relationship across both samples. Whereas sales manipulation increases the relative interest expenses for American companies, it does not affect the relative interest expenses for cross-listed companies. Moreover, the comprehensive measure and a reduction in discretionary expenses reduce the relative interest expense for cross-listed companies, whereas American companies cannot reduce their interest expenses this way. This difference is less profound in the relationship between REM and credit rating. Although sales manipulation and overproduction have a more favourable effect (greater upgrade or lower downgrade) on credit rating for cross-listed firms than American companies, these outcomes are not robust.
stated that SOX lead to increased monitoring of U.S. firms, whereas the monitoring of cross-listed firms remained unchanged. The outcomes also indicate that country-specific features influence the relationship between REM and cost of debt, thereby providing an explanation for the contrary findings of Kwak & Park (2015) in the Korean market, and Ge & Kim (2014) and Kim et al. (2018) in the American market. Moreover, cross-listed firms have increased incentives and possibilities to manage earnings, and being cross-listed provides them with more favourable possibilities for the use of REM. However, the outcomes also show that the measure of cost of debt does matter, thereby driving the need for future research to furtherly conduct research on this topic and determine which measure of cost of debt best incorporates the effect of REM.
In the remainder of this paper, section II will elaborate on the theoretical background and section III will provide the methodology. Thereafter, sector IV will provide the results, followed by the discussion in sector V. Lastly, sector VI will provide the conclusion of the paper.
2. Literature review
2.1 Real earnings management
There are two types of EM: accruals-based earnings management (AEM) and real earnings management (REM). The former implies the management of earnings through discretionary accrual choices, allowed under the commonly accepted accounting standards. The latter implies the management of earnings by distorting real activities (Kim & Sohn, 2013). In 2002, a wave of corporate governance failures eroded trust in financial reports, leading to the passage of the Sarbanes-Oxley Act (SOX). As a result, AEM was more easily detected, and many companies shifted to REM (Cohen et al., 2008; Roychowdhury, 2006).
Because REM influences the real economic activities, it distorts the fundamentals of the business and can therefore have adverse effect on the long-term firm performance (Brown & Higgins, 2001; Cohen & Zarowin, 2010, Zang, 2011). Although REM is costlier than AEM, Francis, LaFond, Olsson & Schipper (2005) state that the overall costs of REM are lower, as it is less likely to be detected. Zang (2011) finds that managers prefer to use REM and, only if necessary, use AEM afterwards. According to Roychowdhury (2006), whose framework is widely used in research to detect REM (e.g. Cohen et al., 2008), there are three manners to detect REM: (1) sales manipulation, (2) overproduction, and (3) a reduction in discretionary expenses.
Sales manipulation is a managers’ attempt to temporarily increase sales during the year by offering price discounts or more lenient credit terms to customers (Roychowdhury, 2006). Sales manipulation will boost sales, but offering more lenient credit terms increases the exposure to uncollectible accounts (Ge & Kim, 2014). It can also backlash on future sales, as it might bring orders forward.
Whereas Roychowdhury (2006) calculates REM as the sum of overproduction, sales manipulation and discretionary expenses, recent literature became sceptical about this measurement. Overproduction and a reduction in discretionary expenses already affect the cash flow. Including sales manipulation, calculated as abnormal cash flow, in the comprehensive measure will take into account the effect of overproduction and a reduction in discretionary expenses twice. Therefore, following i.e. Zang (2011), this paper will calculate the comprehensive measure as REM as the sum of REM via overproduction and REM via a reduction in discretionary expenses. As a robustness check and to capture the effect of this assumption, the main regressions will also be run on a comprehensive measure of REM including sales manipulation.
2.2 Cross-listing
According to Merton (1987), cross-listing increases the number of investors aware of a firm as investors do not have equal information. This investor recognition reduces the “shadow cost” of not knowing about a security, and thus leads to an increase of the market value of the firm’s shares. Furthermore, cross-listing enhances liquidity (Foerster & Karolyi, 1999) and mitigates the costs of market segmentation (Lang et al., 2003a). Non-U.S. firms issue corporate bonds more frequently and at lower offering yields following an equity cross-listing on a U.S. exchange (Ball, Hail & Vasvari, 2018). They show that having cross-listing shares on the U.S. markets generates benefits for bondholders. Moreover, Citigroup (2011) showed that cross-listing can reduce an issuer’s cost of debt. This is in line with a study of the European Corporate Governance Institute, which concludes that cross-listing in the U.S. market has a positive impact on the cost of debt (Ball et al., 2018). If this reduction in debt is one of the reasons that firms cross-list, cross-listed firms have increased incentives to lower their cost of debt.
Cross-listing in the U.S. increases firm value as it limits the availability of controlling shareholders to extract private benefits of control (Doidge, Karolyi & Stulz, 2004). Coffee (1998) came up with the so-called “bonding hypothesis”. Issuers cross-list in the U.S. to subject themselves to the high U.S. disclosure standards and greater threat of enforcement (by both public and private enforcers). Hereby, cross-listing serves as a bonding mechanism that positively affects a firm’s corporate governance. The SEC supervision leads to more transparent disclosure, investor protection and effective monitoring (King & Segal, 2004), signalling good governance and reducing the cost of capital.
2.3 Differences between local firms and cross-listed firms
Furthermore, Enomoto, Kimura & Yamaguchi (2015) and Yu (2008) find that analyst coverage is negatively related with earnings management: firms followed by more analysts manage their earnings less. Due to the home bias, it is expected that American analysts follow local - American - firms more than cross-listed firms (Bae, Stulz & Tan, 2008). Therefore, cross-listed firms have increased opportunities to engage in earnings management. Furthermore, analysts have less knowledge of cross-listed firms and therefore might ask higher premiums. To obtain cheap financing, there is a greater need for cross-listed firms to provide better financial statements, thereby furtherly increasing the incentives to engage in REM. Therefore, I hypothesize;
H1: The relative level of real earnings management for cross-listed firms in the U.S. market is higher than for U.S. firms in the same market.
2.4 Earnings management and the cost of debt
Literature on the influence of REM on cost of debt is currently contradicting. Ge & Kim (2014) developed two hypotheses: (1) if REM is perceived as managerial opportunism the relationship will be positive, and (2) if bondholders do not incorporate managerial opportunism, REM is perceived as a desirable activity and the relationship will be negative. In their analysis, credit rating agencies and bondholders view REM as a credit risk-increasing factor, leading to a positive relationship. Higher credit ratings, formed by credit rating agencies on a firm’s credit worthiness, can lead to accelerated debt repayments (Kisgen, 2006).
Kim et al. (2018) link the positive relationship in the U.S. market to the development of institutional environments. The lack of this development in the Korean market explains, according to the authors, the contrary beliefs of Kwak and Park (2015). Whereas Ge & Kim (2014) argue that the adverse consequences of REM on a company’s earnings and cash flow increase the cost of debt, Mellado-Cid et al. (2017) argue that bond issuers use REM to signal superior earnings quality to potential investors. The authors assign the negative relationship they find between REM and cost of debt to the investors’ mispricing of REM due to successful signalling of increased earnings.
According to Kim & Sohn (2013), the equity market sees through REM and according to Shen & Huang (2013), REM is perceived as managerial opportunism, leading to increased costs of capital. Bharath, Sunder & Sunder (2008) found that poor accounting quality is related to higher yield spreads on bonds, and Francis et al. (2005) found that REM decreases earnings quality. Looking at the future effects of REM, Graham et al. (2005) and Roychowdhury (2006) underline the negative effects in the long-run and Kim & Sohn (2013) found a negative relationship between REM and future cash flows. Because bank lenders are able to engage in more detailed monitoring than public lenders (Goss & Roberts, 2011), they are more likely to see through REM. Moreover, SOX lead to increased monitoring of a company’s earnings. Therefore, I hypothesize:
H2: The use of real earnings management has a positive relationship with the cost of debt.
increased monitoring of U.S. firms, whereas the monitoring of cross-listed firms remained unchanged. This increases the opportunities for cross-listed firms to manage their earnings.
If this would assumption would hold, it would also imply that cross-listed firms can manage earnings without getting caught. Hence, the negative relationship between REM and cost of debt would be attenuated for cross-listed companies. In line with this, Bae et al. (2008) find that local firms have increased analyst coverage, and Yu (2008) finds that analyst coverage is negatively related with EM. This lack of analyst coverage increases the possibilities to manage earnings and reduced analyst coverage might also weaken the relationship between REM and cost of debt. Because of increased incentives and possibilities for cross-listed companies to engage in REM, less coverage of cross-listed firms, and increased monitoring of US companies after SOX, I hypothesize:
H3: The positive relationship between real earnings management and the cost of debt is attenuated for cross-listed firms in the American market compared to local firms in the same market.
3. Methodology
3.1 Data selection and matching
I sample all firms in the COMPUSTAT North America database from 1997 till 2017. The real activities of EM became more important after SOX in 2002, hence the main sample period is post-SOX. Five years before SOX are included in order to see whether there are differences between the pre- and post-SOX period. All publicly listed companies that are traded on the New York Stock Exchange, American Stock Exchange, and NASDAQ are included in the sample, leading to a total of 267,391 firm-year observations. After excluding duplicates, firms without assets, firms were accounting standards were not standard, and financial institutions, 172,657 firm-year observations remained. Then, firms are designated as either American or cross-listed.
Companies with a Foreign Incorporation Code USA are designated American. Cross-listed firms in this paper are non-U.S. firms that have common shares listed on an American Stock Exchange, next to an earlier listing on at least one other stock exchange. Firms with primary issues on other stock exchanges, with ADR or an ADR equivalent in their name, and/or an ADR ratio are assigned as cross-listed. Firms that were assigned to both or none of the groups, due to a flaw in the data, are excluded, dropping another 1,930 firm-year observations. This lead to a total of 170,727 firm-year observations: 15,017 American firms with 127,508 firm-year observations and 5,115 cross-listed firms with a total of 43,219 observations.
However, cross-listed firms are self-selected, which can lead to a self-selection bias. To control for both performance and earnings management manipulation, I adopt a matched sample technique. Every cross-listed firm is matched with an American counterpart based on industry, year and size, following other papers on this topic, such as Lang et al. (2003b) Lang et al. (2006) and Litvak (2006). A single propensity score-matching on the full sample will either keep significant asset size variations, or will lead to a significant loss of observations due to a highly small caliper. Therefore, the companies are matched on industry and size for each separate year. This method also overcomes the problem of matching a cross-listed company to a controlling firm in another year, thereby putting value on the time value of firm size due to inflation across time.
of their cross-listed counterparts. Whereas the optimal caliper according to Austin (2011) is 0.2, the caliper is set to 0.1, as this is more in line with previous studies on this topic, such as Chan, Chen, Chen & Yu (2014). An additional condition is no replacement: American companies can only be matched once.
A total of 104,923 firm-year observations is deleted because they were unmatched. This led to a sample of sample of 65,804 firm-year observations, equally split across American and cross-listed firm-year observations. Due to the lack of variables included in the empirical model (8), the main regressions are run on a final sample of 14,081 firm-year observations: 8,619 American firm-year observations and 5,461 cross-listed firm-year observations.
3.2 Variable measurement
3.2.1 Real earnings management
To measure earnings management, three proxies will be used: (1) the abnormal level of sales manipulation, (2) overproduction, and (3) the abnormal level of discretionary expenses. As indicated earlier, sales manipulation is not regarded as an input of the comprehensive measure of REM, following Zang (2011). However, REM via sales manipulation will still be calculated as a separate measure used to run separate regressions with.
The normal levels of CFO, discretionary expenses and production costs are calculated via the model of Dechow, Kothari and Watts (1998), which is also used by Roychowdhury (2006). The normal level of CFO (CFO) is calculated as a linear function of sales and change in sales, ran as a cross-sectional regression for each industry and year: 𝐶𝐹𝑂𝑖,𝑡 𝐴𝑠𝑠𝑒𝑡𝑠𝑖,𝑡−1= 𝑘1𝑡 1 𝐴𝑠𝑠𝑒𝑡𝑠𝑖,𝑡−1+ 𝑘2 𝑆𝑎𝑙𝑒𝑠𝑖,𝑡 𝐴𝑠𝑠𝑒𝑡𝑠𝑖,𝑡−1+ 𝑘3 𝑆𝑎𝑙𝑒𝑠𝑖,𝑡 𝐴𝑠𝑠𝑒𝑡𝑠𝑖,𝑡−1+ 𝑖𝑡 (1)
Abnormal CFO (REM_CFO) is the actual CFO minus the normal level of CFO as calculated using the estimated coefficients in (1).
Production costs (Prod) are calculated as the sum of COGS and change in inventory during the year. COGS (COGS) are modelled as a linear function of contemporaneous sales:
𝐶𝑂𝐺𝑆𝑖,𝑡 𝐴𝑠𝑠𝑒𝑡𝑠𝑖,𝑡−1= 𝑘1𝑡 1 𝐴𝑠𝑠𝑒𝑡𝑠𝑖,𝑡−1+ 𝑘2 𝑆𝑎𝑙𝑒𝑠𝑖,𝑡 𝐴𝑠𝑠𝑒𝑡𝑠𝑖,𝑡−1+ 𝑖𝑡 (2)
Inventory growth (Inv) is modelled as a linear function of contemporaneous change in sales and the lagged change in sales:
𝐼𝑁𝑉𝑖,𝑡 𝐴𝑠𝑠𝑒𝑡𝑠𝑖,𝑡−1= 𝑘1𝑡= 1 𝐴𝑠𝑠𝑒𝑡𝑠𝑖,𝑡−1+ 𝑘2 𝑆𝑎𝑙𝑒𝑠𝑖,𝑡 𝐴𝑠𝑠𝑒𝑡𝑠𝑖,𝑡−1+ 𝑘3 𝑆𝑎𝑙𝑒𝑠𝑖,𝑡−1 𝐴𝑠𝑠𝑒𝑡𝑠𝑖,𝑡−1+ 𝑖𝑡 (3)
By using (2) and (3), the normal level of production costs (Prod) is estimated as:
Overproduction (REM_Prod) is the actual production minus the normal level of production as calculated using the estimated coefficients in (4).
Modelling the level of discretionary expenses as a function of current sales can lead to a mechanical problem. Firms managing sales upwards to increase reported earnings in a given year will lead to significantly lower residuals than estimated in the formula. To overcome this problem, following Cohen & Zarowin (2010), discretionary expenses (DiscExp) is modelled as a function of lagged sales to derive at the normal level of discretionary expenses: 𝐷𝑖𝑠𝑐𝐸𝑥𝑝𝑖𝑡 𝐴𝑠𝑠𝑒𝑡𝑠𝑖,𝑡−1= 𝑘1 1 𝐴𝑠𝑠𝑒𝑡𝑠𝑖,𝑡−1+ 𝑘2 𝑆𝑎𝑙𝑒𝑠𝑖,𝑡−1 𝐴𝑠𝑠𝑒𝑡𝑠𝑖,𝑡−1+ 𝑖,𝑡 (5)
Abnormal discretionary expenses (REM_DiscExp) are the actual discretionary expenses minus the normal level of discretionary expenses as calculated using the estimated coefficients in (5).
In the above equations, CFOis the cash flow from operations (annual Compustat data item OANCF) in period t, Prod is the production costs in period t, calculated as the sum of costs of goods sold (annual Compustat data COGS) and the change in inventory (annual Compustat data item INVT) in period t, and DiscExpis the sum of research and development costs (annual Compustat data item XRD) and selling, general and administrative expenses (annual Compustat data item XSGA) in period t.
For the separate measures of REM, a more positive value of REM_Prod indicates a higher likelihood that a firm is engaging in REM, whereas for REM_CFO and REM_DiscExp, a more negative value indicates a higher likelihood that a firm is engaging in REM. To simplify this, REM_CFO and REM_DiscExp are multiplied by negative one, so that positive levels of REM_CFO and REM_DiscExp also indicate a greater likelihood that the firm is engaging in earnings. The comprehensive measure of REM (REM) will be the sum of the abnormal levels of discretionary expenses and production costs:
𝑅𝐸𝑀𝑖,𝑡= 𝑅𝐸𝑀_𝑃𝑟𝑜𝑑𝑖,𝑡+ 𝑅𝐸𝑀_𝐷𝑖𝑠𝑐𝐸𝑥𝑝𝑖,𝑡 (6)
A high (low) score on either measure of REM indicates a high (low) likelihood that a firm engaged in earnings management. As the effect of the comprehensive measure may dilute different implications of the separate measures for earnings (Cohen & Zarowin, 2010), the three individual measures and the comprehensive measure will be used to test whether there is a relationship between real earnings management and the cost of debt.
3.2.2 Cost of debt and credit rating
Int_Expi,t+1= 𝐷𝑇𝑖,𝑡+𝐷𝑇𝑖,𝑡+1 𝑋𝐼𝑁𝑇𝑖,𝑡+1
2
+ 𝑖𝑡 (7)
In the above equation, Int_Expt+1 will be calculated as the amount of total interest and related expense (annual Compustat data item XINT) over the total amount of outstanding debt (annual Compustat data item DT).
Consistent with Jiang (2008) and Ge & Kim (2014), I will also use the S&P credit ratings (Credit_Ratingt+1) as a second proxy for the cost of debt. The S&P credit ratings rank from AAA, the highest grade “Prime”, to D, the lowest grade “Default”. A rating below BBB- is recommended not to invest in (Standard & Poor’s, 2018). The highest rating (AAA) will be given a score of 1, and the lowest rating (D) a score of 23.1
Thus, a higher score on the credit rating implies a worse credit rating (downgrade), a lower score on the credit rating implies a better credit rating (upgrade). For example, a positive (negative) relationship between REM and
Credit_Ratingt+1 implies that an increased level of REM will lead to a worse (better) credit rating. The credit rating is not linear: a downgrade in investment grade from BBB to BBB- implies a much higher increased probability of default than a downgrade from AAA to AA+. Therefore, as a robustness check, I will run both a normal regression and an ordered logistic regression on Credit_Ratingt+1.
3.3 Control variables
To increase the reliability of the outcomes, I control for variables that influence the level of earnings management and the cost of debt. The level of earnings management can be influenced by both incentives to manage earnings and the opportunities that arise for managing earnings (Teoh, Welch & Wong, 1998), relative interest expenses and credit ratings are influenced by firm characteristics (Wald, 1999).
Following Roychowdhury (2006), I control for growth opportunities and size to control for systematic variations in the earnings management proxies. Moreover, firms with more growth opportunities are penalized more by the stock market when they miss earnings thresholds (Skinner & Sloan, 2002), giving them more incentives to manage earnings. As measurement errors can have a correlation with firm performance (Dechow et al, 1995), I also control for firm performance via net income. Moreover, the level of net income influences the urge for managers to manage earnings. These three factors also influence the cost of debt. Net income is scaled by total lagged assets; hence it equals return-on-assets (ROA). Because the return on assets in the previous year determine the level of incentives to manage earnings in the following year, and because earnings management is used to beat last year’s earnings benchmark (Jiang, 2008) ROA is taken from the lagged period (ROAt-1). Growth opportunities is measured via the market-to-book ratio (MTBRatio), which equals the ratio of the market value of equity to the book value of equity calculated as assets minus liabilities. Firm size (Size) is the natural logarithm of the market value of assets at the beginning of the year.
Capital structure also affects the opportunities of and incentives for REM. On the one hand, leverage limits REM activities (Jelinek, 2007; Zamri, Rahman & Isa, 2013) because distressed firms want or have to limit risks. On the other hand, managers use EM to avoid violations of debt covenants (Kim, Lisic & Pevzner, 2010),
1 The entire spectrum of credit ratings is as follows: AAA = 1, AA+ = 2, AA = 3, AA- = 4, A+ = 5, A = 6, A- = 7,
hence firms with a higher leverage ratio would increase their level of EM. Leverage is calculated as the ratio of total debt to total equity. As leverage in the previous year affects the level of REM in the current period, leverage is taken from the lagged period (Leveraget-1). Furthermore, large audit firms are more likely to provide higher quality audits (DeAngelo, 1981), reducing the possibilities to engage in REM, and retaining a Big Six auditor reduces debt-related monitoring costs and therefore leads to lower cost of debt (Pittman & Fortin, 2004). To control for auditor tenure, a dummy variable is included that indicates the presence of a Big Five (1997 – 2002) or Big Four (2003 – 2017, following the collapse of Arthur Anderson) auditor (BigN).
As firms avoid reporting losses, managers have increased incentives to manage earnings following a loss-giving year. Loss-making companies also have increased cost of debt and credit ratings. Firm’s beating the zero earnings benchmarks have a higher probability of rating upgrades (Jiang, 2008). Therefore, the dummy variable lossability (Losst-1) is included, which takes a value of 1 if the firm reported negative incomes in the lagged period, and 0 of the firm reported positive incomes in the lagged period. Lastly, as the possibilities and incentives for managers to manage earnings may be industry- or year-specific, and because the cost of debt and credit rating are highly influenced by industry and year, I control for industry and year by including an indicated dummy for the 2-digit SIC code as industry dummy (SIC_2) and an indicated year dummy (Fyear).
Data on the influence of stock-based compensation on incentives to manage earnings are contradicting. Ashbaugh-Skaife, Hollis, Collins & LaFond (2006) find that stock-based compensation increases the incentives for the board to manage earnings. Cohen, Dey & Lys (2008) find that the percentage of bonus compensation is not correlated with earnings management for their entire period. The relationship they find is mainly focused on the post-SOX-period and they find only a relationship between stock-based compensation and AEM, but not REM. Because of this contradicting literature, combined with the fact that there is very little data on stock-based compensation for cross-listed companies (see table 1), I do not include stock-based compensation as a control variable.
All continuous control variables in the empirical model are winsorized at their 2nd and 98th percentiles,
to rule out the influence of potential outliers, except for firm size, where the influence of outliers is ruled out by taking the natural logarithm. The cost of debt is winsorized by its 5th and 95th percentile. All winsorizing is done
before the matching process, to improve the quality of the matching. The credit rating is not winsorized because the lower the rating, the higher the probability of default; hence, it is not linear and cannot be winsorized.
3.4 Empirical model
To test my hypotheses, I run several regressions of the comprehensive and separate measures of REM on the cost of debt. This will result in the following empirical model:
𝐶𝑜𝑠𝑡 𝑜𝑓 𝐷𝑒𝑏𝑡𝑖,𝑡+1= 0+ 1∗ 𝑅𝐸𝑀𝑖,𝑡+2∗ 𝑆𝑖𝑧𝑒𝑖,𝑡+3∗ 𝑀𝑇𝐵𝑅𝑎𝑡𝑖𝑜𝑖,𝑡+ 4∗ 𝑅𝑂𝐴𝑖,𝑡−1+ … … … +
5∗ 𝐿𝑒𝑣𝑒𝑟𝑎𝑔𝑒𝑖,𝑡−1+ 6∗ 𝐵𝑖𝑔𝑁𝑖,𝑡+ 7∗ 𝐿𝑜𝑠𝑠𝑖,𝑡−1+8∗ 𝑖. 𝑖𝑛𝑑𝑢𝑠𝑡𝑟𝑦 +9∗ 𝑖. 𝑦𝑒𝑎𝑟 + 𝑖𝑡 (8)
where:
Cost of Debtt+1 = the two different measures of the cost of debt:
(ii) Credit_Ratingt+1 = the S&P credit rating in period t+1
REM = various REM measures, namely the separate measures:
(i) REM_CFO= abnormal levels of cash flow from operations; (ii) REM_Prod = abnormal levels of production costs;
(iii) REM_DiscExp = abnormal levels of discretionary expenses, and;
(iv) the comprehensive measure REM = the sum of abnormal levels of production costs and abnormal levels of discretionary expenses.
Size = the natural logarithm of total assets; MTBRatio= the market-to-book ratio;
ROAt-1 = net income scaled by lagged total assets;
Leveraget-1 = the leverage of the firm in the lagged period, calculated as total liabilities deflated by total assets;
BigN= dummy variable that equals one if the firm is audited by the Big Five (1997-2002) or the Big Four (2003 – 2017), zero otherwise;
Losst-1 = dummy variable that equals one if the firm had a negative income in the lagged period, zero otherwise;
i.industry = indicated dummy variable that controls for fixed industry effects; i.year = indicated dummy variable that controls for fixed industry effects.
Additionally, a dummy variable will be included to indicates whether a firm is cross-listed or American:
Dummy_Crosslisted = dummy variable that takes a value of one if the company is cross-listed, zero otherwise.
For the third hypothesis, a full factorial variable is added to capture the relationship of being cross-listed on REM. Hence, the empirical model will look like:
𝐶𝑜𝑠𝑡 𝑜𝑓 𝐷𝑒𝑏𝑡𝑖,𝑡+1 = 0+1∗ 𝑅𝐸𝑀𝑥𝐶𝑟𝑜𝑠𝑠𝑙𝑖𝑠𝑡𝑒𝑑𝑖,𝑡+ 2∗ 𝑅𝐸𝑀𝑖,𝑡+3∗ 𝐷𝑢𝑚𝑚𝑦𝐶𝑟𝑜𝑠𝑠𝑙𝑖𝑠𝑡𝑒𝑑+⋯ … … +
4∗ 𝑆𝑖𝑧𝑒𝑖,𝑡+5∗ 𝑀𝑇𝐵𝑅𝑎𝑡𝑖𝑜𝑖,𝑡+ 6∗ 𝑅𝑂𝐴𝑖,𝑡−1+ 7∗ 𝐿𝑒𝑣𝑒𝑟𝑎𝑔𝑒𝑖,𝑡−1+ 8∗ 𝐵𝑖𝑔𝑁𝑖,𝑡+⋯ … … … +
9∗ 𝐿𝑜𝑠𝑠𝑖,𝑡−1+10∗ 𝑖. 𝑖𝑛𝑑𝑢𝑠𝑡𝑟𝑦 + 11∗ 𝑖. 𝑦𝑒𝑎𝑟 + 𝑖𝑡 (9)
4. Results
4.1 Descriptive statistics
The descriptive statistics of all variables of the final sample are presented in table 1. The mean REM_CFO,
REM_Prod & REM_DiscExp are -0.079, -0.009 and 0.271, respectively, comparable to those reported by Cohen
et al. (2008), Gunny (2010) and Roychowdhury (2008). They indicate that, on average, firms mainly manage their earnings upwards by reducing discretionary expenses. Combining the latter two measures into the comprehensive REM measure gives a mean REM of 0.266. This indicates that, on average, firms manage their earnings upwards through the use of real earnings management. Surprisingly, the old measure of REM that includes REM_CFO,
REM_Old, is lower (0.173) than the new measure, due to the negative mean of REM_CFO.
The mean REM_CFO of the cross-listed company (-0.112) is almost double the (negative) size of the mean of the American companies (-0.049), but the mean REM_DiscExp of the cross-listed companies (0.372) also is double the (positive) size of their American counterparts (0.183). The mean REM_Prod for American companies is negative (-0.018), whereas this is slightly positive for the cross-listed companies (0.001). This leads to a mean REM of 0.369 for cross-listed companies, which is more than double the size of their American counterparts (0.179). This is in line with the research of Lang et al. (2006).
The relative interest expense of both origins is almost similar, but the foreign companies have much better mean credit ratings (BBB) than their American counterparts (BB+). Surprisingly, the mean credit rating of American companies is a non-investment grade. The asset size of both samples is almost similar due to the matching on asset size.
As can be seen from the descriptive statistics, only 200 cross-listed firm-year observations have data on stock-based compensation. Of these 200 firm-year observations, only 36 have both REM and Credit_Ratingt+1. Therefore, stock-based compensation is excluded as a control variable.
Table 1: Descriptive statistics of the full sample, the American sample and the cross-listed sample.
Full sample American companies Cross-listed companies Variable n Mean Median n Mean Median n Mean Median
REM_CFO 52,450 - 0.079 - 0.053 27,004 - 0.049 - 0.410 25,446 - 0.112 0.068 REM_Prod 42,200 - 0.009 - 0.022 21,930 - 0.018 - 0.026 20,270 0.001 - 0.019 REM_DiscExp 42,503 0.271 0.115 22,612 0.183 0.087 19,891 0.372 - 0.152 REM 33,849 0.266 0.133 18,420 0.180 0.104 15,429 0.369 0.172 REM_Old 33,658 0.173 0.088 18,339 0.114 0.067 15,319 0.245 0.116 Int_Expt+1 41,544 0.116 0.077 21,129 0.116 0.077 20,415 0.117 0.077 Credit_Ratingt+1 14,909 10.072 10 9,479 10.587 10 5,430 9.173 9 MTBRatiot-1 47,198 2.431 1.566 26,359 2.567 1.705 20,839 2.259 1.413 ROAt-1 48,628 - 0.203 0.019 24,925 - 0.246 0.022 23,703 - 0.158 0.167 Stock_comp 11,142 0.799 0 10,942 0.798 0 200 0.840 0 Leveraget-1 25,186 0.470 0.110 12,934 0.522 0.180 12,252 0.415 0.038 Bign 65,804 0.720 1 32,902 0.680 1 32,902 0.761 1 Losst-1 58,186 0.401 0 29,707 0.385 0 28,479 0.418 0 Dummy_Crosslisted 65,804 0.5 0.5 32,902 0 0 32,902 1 1 Dummy_USA 65,804 0.5 0.5 32,902 1 1 32,902 0 0
Table 1: The sample size, mean and median for all variables for (1) the full sample, (2) the American companies and (3) the cross-listed companies.
4.2 Univariate analysis
Table 2 provides the Pearson correlations of the variables included in the empirical model. First, looking at the REM components, there is a significant relationship between all components: a positive relationship between REM_CFO and REM_Prod(Pearson correlation of 0.217), a negative relationship between REM_CFO and REM_DiscExp (Pearson correlation of -0.468), and a negative relationship between REM_Prod and
REM_DiscExp (Pearson correlation of -0.118). Thus, REM_CFO and REM_Prod are used complementary,
In line with expectation, there is a significant positive relationship between Int_Expt+1 and
Credit_Ratingt+1 (Pearson correlation of 0.104), as a worse credit rating (higher Credit_Ratingt+1) leads to higher costs of debt (higher Int_Expt+1). The relationships between REM and Int_Expt+1 and Credit_Ratingt+1 are both significantly and negative (Pearson correlations of -0.033 and -0.043, respectively). This implies that a higher level of real earnings management (higher REM) decreases the cost of debt (lower Int_Expt+1) and leads to a better credit rating (lower Credit_Ratingt+1). This is surprising, as this is contrary to the first hypothesis. Moreover, the relationship for REM is significant with both measures of cost of debt, whereas these relationships are insignificant for REM_Old.
For the separate measures of REM on Int_Expt+1 and Credit_Ratingt+1, there are significant positive relationships with REM_CFO (Pearson correlation of 0.088 and 0.077, respectively), significant positive relationships with REM_Prod (Pearson correlation of 0.044 and 0.101, respectively) and significant negative relationship with REM_DiscExp (Pearson correlation of -0.046 and -0.060, respectively). This is contradicting, as it suggests that REM_CFO and REM_Prod increase Int_Expt+1 and worsen Credit_Ratingt+1, whereas
REM_DiscExp lowers Int_Expt+1 and improves Credit_Ratingt+1. Of the separate measures, REM_DiscExp has the strongest relationship with REM (0.935 against -0.418 and 0.143). The high influence of REM_DiscExp on
REM, combined with the surprising relationship between REM_DiscExpon Int_Expt+1 and Credit_Ratingt+1, provides an explanation why the relationship between REM and Int_Expt+1 and Credit_Ratingt+1 are different than hypothesized.
Dummy_Crosslisted has a significant and positive relationship with REM(Pearson correlation of 0.041): cross-listed firms engage in more REM than American companies, in line with the second hypothesis.
Dummy_Crosslisted has an insignificant relationship with Int_Expt+1, but a significant and negative relationship with Credit_Ratingt+1 (Pearson correlation of -0.178). Thus, foreign firms have better credit ratings than their American counterparts. Moreover, Dummy_Crosslisted has significant relationship with all REM components: a negative relationship with REM_CFO (Pearson correlation of -0.045) and a positive relationship with REM_Prod and REM_DiscExp (Pearson correlation of 0.019 and 0.044, respectively). This means that foreign firms engage in less REM_CFO, but in more REM_Prod and REM_DiscExp.
Other factors that influence the level magnitude of REM are growth opportunities (MTBRatio), firm performance (ROAt-1) and lossability (Loss t-1), whereas, surprisingly, firm size (Size). capital structure (Leverage t-1) and auditor (BigN) do not have a significant relationship with the comprehensive measure of real earnings management (REM). However, firm size is significant to all separate measures and auditor to REM_CFO and
REM_Prod
4.3 Multivariate analysis
Table 2: Pearson correlation table.
REM_CFO REM_Prod REM_DiscExp REM REM_Old Int_Expt+1 Credit_Ratingt+1 Size
REM_CFO 1.000 REM_Prod 0.217 1.000 REM_DiscExp -0.468 -0.118 1.000 REM -0.418 0.142 0.935 1.000 REM_Old -0.093 0.206 0.824 0.905 1.000 Int_Expt+1 0.088 0.044 -0.046 -0.033 -0.009 1.000 Credit_Ratingt+1 0.077 0.101 -0.061 -0.043 -0.008 0.104 1.000 Size -0.087 -0.031 -0.013 -0.000 -0.032 -0.241 -0.609 1.000 MTBRatioi -0.030 -0.030 0.011 0.014 -0.004 -0.020 -0.209 0.017 ROAt-1 -0.190 -0.068 0.067 0.074 0.010 -0.194 -0.251 0.410 Leveraget-1 -0.007 0.007 -0.009 -0.005 -0.006 -0.102 0.133 0.246 BigN -0.068 -0.026 -0.004 0.002 -0.016 -0.172 -0.070 0.508 Losst-1 0.075 0.057 0.009 0.015 0.042 0.148 0.417 -0.471 Dummy_CL -0.045 0.019 0.044 0.041 0.031 0.006 -0.178 -0.017 Dummy_USA 0.045 -0.019 -0.044 -0.041 -0.031 -0.006 0.178 0.017
MTBRatio ROAt-1 Leveraget-1 BigN Losst-1 Dummy_CL Dummy_USA
MTBRatio 1.000 ROAt-1 0.034 1.000 Leveraget-1 0.073 0.104 1.000 BigN 0.010 0.260 0.130 1.000 Losst-1 0.004 -0.387 -0.078 -0.253 1.000 Dummy_CL -0.028 0.049 -0.040 0.090 0.034 1.000 Dummy_USA 0.028 -0.049 0.040 -0.090 -0.034 -1.000 1.000
To see how this difference is constituted, I run the same regression on the three separate measures of REM. Firstly, Dummy_Crosslisted has a marginally significant (p = 0.063) and negative (-0.025) relationship with
REM_CFO. Thus, cross-listed firms engage in less REM_CFO than American firms. The relationship between Dummy_Crosslisted and REM_Prod and REM_DiscExp are also significant on the 5%-significance level (p =
0.000 and p = 0.004, respectively) and are both positive (0.0500 and 0.118, respectively). It can be concluded that the relative level of earnings management is higher for cross-listed firms in the U.S. market than for U.S. firms in the same market, in line with the second hypothesis. This difference is driven by increased levels of overproduction and a reduction in discretionary expenses, whilst it is slightly attenuated by the level of sales manipulation.
TABLE 3
The regression models of the four measures of REM on the cross-listed dummy variable.
Variables REM REM_CFO REM_Prod REM_DiscExp
Dummy_Crosslisted 0.199*** -0.025* 0.050*** 0.118*** (4.16) (-1.86) (5.01) (2.88) Size 0.008 -0.016*** 0.001 -0.012 (0.60) (-4.19) (0.51) (-1.00) ROAt-1 0.307*** -0.175*** -0.033*** 0.268*** (6.20) (-12.77) (-3.53) (6.36) MTBRatio -0.007 -0.003* -0.003*** -0.003 (-1.24) (-1.86) (-2.69) (-0.63) Leveraget-1 0.013 0.003 0.005* 0.007 (1.02) (0.89) (1.76) (0.65) BigN -0.055 -0.002 -0.017 -0.022 (-1.00) (-0.13) (-1.46) (-0.47) Losst-1 0.179*** 0.005 0.059*** 0.086** (3.91) (0.35) (6.07) (2.19) Fyear 0.016 -0.002 0.002 0.020** (1.64) (-1.02) (0.64) (2.51) SIC_2 0.009*** -0.004*** -0.000 0.008*** (3.50) (-4.03) (-1.23) (3.41)
The table shows the regressions of the four measures of REM on the cross-listed dummy variable and associated control variables for the complete sample. All continuous variables are winsorized at their respective 2nd and 98th percentile, except for Size, of which the natural logarithm is taken, to rule out the influence of potential outliers, and year and industry fixed effects are included.
T-statistics in parentheses. Coefficients with *, ** or *** are statistically significant at the 0.1, 0.05 or 0.01 levels, resp ectively.
To test the second hypotheses, the empirical model (8) of REM on cost of debt, discussed in section 3.4, is conducted. The results of all regressions are shown in table 4. First, I regress the comprehensive measure of REM on Int_Expt+1 (panel A) and Credit_Ratingt+1 (Panel B: normal regression. Panel C: ordered logistic regression). Whereas the univariate relationship between REM and Int_Expt+1 is significant and negative, the multivariate analysis provides an insignificant relationship, as the p-value is 0.240. Thus, REM does not significantly affect the relative interest expenses, contrary to the first hypothesis. However, although the univariate relationship between REM and Credit_Ratingt+1 is negative, the regression of REM on Credit_Ratingt+1 provides a significant (p = 0.000) and positive relationship (0.087): an increase in the level of REM worsens the credit rating. This is in line with the first hypothesis. The ordered logistic regression also provides a significant (p = 0.000) and positive relationship (0.084), making the outcome robust.
To see what drives these relationships, table 4 also provides the relationships of the separate REM measures and Int_Expt+1 and Credit_Ratingt+1. Both REM_CFO and REM_Prod have significant (p = 0.015 and
p = 0.019, respectively) and positive (0.005 and 0.008, respectively) relationships with Int_Expt+1: sales manipulation and overproduction increase the relative interest expense. However, the relationship between
TABLE 4
The regression and ordered logistic regression models of the two measures of cost of debt on the four measures of REM. Panel A: The regression models of the relative interest expense on the four measures of REM.
Variables Int_Expt+1 Int_Expt+1 Int_Expt+1 Int_Expt+1
REM measure -0.001 0.005** 0.008** -0.002** (-1.17) (2.44) (2.35) (-2.08) Size -0.009*** -0.009*** -0.009*** -0.010*** (-13.04) (-14.07) (-13.05) (-14.30) ROAt-1 -0.012*** -0.012*** -0.013*** -0.011*** (-3.79) (-4.39) (-4.50) (-3.76) MTBRatio -0.001 -0.000 -0.000 -0.000 (-1.61) (-1.06) (-1.31) (-1.44) Leveraget-1 -0.003*** -0.003*** -0.003*** -0.003*** (-4.63) (-4.94) (-4.62) (-4.92) BigN -0.015*** -0.012*** -0.013*** -0.013*** (-3.68) (-3.43) (-3.57) (-3.53) Losst-1 0.023*** 0.021*** 0.021*** 0.022*** (7.86) (7.63) (7.59) (7.61) Fyear -0.001 -0.001 -0.001 -0.001 (-0.82) (-1.35) (-0.78) (-1.23) SIC_2 -0.000 -0.000 -0.000 -0.000 (-0.54) (-0.92) (-0.64) (-0.59)
Panel B: The regression models of the credit rating on the four measures of REM.
Variables Credit_Ratingt+1 Credit_Ratingt+1 Credit_Ratingt+1 Credit_Ratingt+1
REM measure 0.087*** -0.103 0.608*** 0.067*** (3.73) (-1.61) (5.97) (2.90) Size -1.435*** -1.381*** -1.383*** -1.419*** (-42.33) (-44.66) (-42.67) (-43.62) ROAt-1 -5.927*** -5.420*** -5.760*** -5.861*** (-8.84) (-7.81) (-8.72) (-9.25) MTBRatio -0.104*** -0.106*** -0.097*** -0.107*** (-8.34) (-9.00) (-8.17) (-8.86) Leveraget-1 0.275*** 0.265*** 0.262*** 0.271*** (7.75) (8.14) (7.70) (8.06) BigN -0.076 -0.115 -0.110 -0.096 (-0.43) (-0.68) (-0.61) (-0.58) Losst-1 0.987*** 1.152*** 1.074*** 1.023*** (6.89) (8.13) (7.64) (7.38) Fyear 0.172*** 0.153*** 0.164*** 0.171*** (4.82) (3.99) (4.24) (4.90) SIC_2 -0.010 -0.003 0.003 -0.016 (-0.47) (-0.25) (0.20) (-0.94)
Panel C: The ordered logistic regression models of the credit rating on the four measures of REM.
Variables Credit_Ratingt+1 Credit_Ratingt+1 Credit_Ratingt+1 Credit_Ratingt+1
REM measure 0.084*** -0.092* 0.483*** 0.066*** (4.21) (-1.74) (5.59) (3.38) Size -1.293*** -1.220*** -1.242*** -1.264*** (-35.04) (-37.51) (-35.61) (-36.55) ROAt-1 -5.344*** -4.989*** -5.227*** -5.213*** (-8.41) (-8.05) (-8.38) (-8.69) MTBRatio -0.097*** -0.097*** -0.090*** -0.100*** (-7.22) (-7.52) (-7.01) (-7.55) Leveraget-1 0.268*** 0.258*** 0.259*** 0.262*** (7.01) (7.51) (7.13) (7.27) BigN -0.236 -0.254* -0.260* -0.247* (-1.63) (-1.91) (-1.80) (-1.86) Losst-1 0.939*** 1.006*** 0.975*** 0.954*** (7.24) (8.24) (7.74) (7.77) Fyear 0.147*** 0.135*** 0.142*** 0.147*** (5.49) (5.16) (5.40) (5.57) SIC_2 -0.021*** -0.012 -0.007 -0.022*** (-3.48) (-1.43) (-0.62) (-4.85)
The table shows the regressions and ordered logistic regression of the four measures of REM on the measures of cost of debt and associated control variables for the complete sample. Int_Expt+1 is winsorized at its respective 5th and 95th percentile, all other
continuous variables are winsorized at their respective 2nd and 98th percentile, except for Size, of which the natural logarithm is taken, to rule out the influence of potential outliers, and year and industry fixed effects are included.
REM via a reduction in discretionary expenses lowers the relative interest expense. The insignificance of the comprehensive measure but the significance of all separate measures provides a good example of the earlier discussed dilution effect Cohen & Zarowin (2010). This underlines the reasoning to also discuss all REM measures separately.
Panel B shows that REM_CFO has an insignificant relationship for the normal regression (p = 0.107) with Credit_Ratingt+1. However, the ordered logistic regression gives a p-value of 0.082 and is thus marginally significant. The coefficient is negative (-0.092), thus an increase in REM_CFO, surprisingly, improves the credit rating (lower Credit_Ratingt+1). The regressions of REM_Prod and REM_DiscExp on Credit_Ratingt+1 are both significant (p = 0.000 and p = 0.004, respectively) and positive (0.608 and 0.067, respectively) and are furtherly confirmed in the ordered logistic regression: increased REM via overproduction and a reduction in discretionary expenses downgrades the credit rating. Striking are the high coefficients of REM_Prod. Contrary, sales manipulation leads to an upgrade of the credit rating, but this relationship is only marginally significant and not robust.
To test the third hypothesis, capturing any effect of being cross-listed on the relationship between REM and cost of debt, the empirical model of (9) is used. The results can be found in table 5. REMxCrosslistedhas a significant (p = 0.050) and negative (-0.003) relationship with Int_Expt+1 (panel A). Thus, the relationship between
REM and Int_Expt+1 is more negative for cross-listed companies. However, this relationship is insignificant (p = 0.240).
REM_CFOxCrosslisted is highly significant (p = 0.001) and has a negative (-0.014) relationship with Int_Expt+1: This relationship is significant (p = 0.015) and has a coefficient of 0.005: an increase in the level of sales manipulation (higher REM_CFO) increases the relative interest expense (higher Int_Expt+1). This effect is attenuated for cross-listed firms: they face a smaller increase in relative interest expenses than their American counterparts following sales manipulation.
The relationship between REM_ProdxCrosslisted and Int_Expt+1 is insignificant (p = 0.219), thus being cross-listed does not affect the relationship between REM_Prod and Int_Expt+1. The regression of
REM_DiscExpxCrosslisted on Int_Expt+1 is marginally significant (p = 0.083) and negative (-0.003). The relationship between REM_DiscExp and Int_Expt+1 is significant (p = 0.038) and negative (-0.002): increased reductions in discretionary expenses reduce the relative interest expense. This effect is more negative, and hence strengthened, for cross-listed companies.
Panel B (regression) and panel C (ordered logistic regression) show the results of the full factorial variables between the four REM measures and Dummy_Crosslisted on Credit_Ratingt+1. The full factorial variable including the comprehensive measure, REMxCrosslisted has an insignificant (p = 0.313) relationship with
Credit_Ratingt+1 in both the normal regression (p = 0.313) and the ordered logistic regression (p = 0.481). Looking at the separate measures, the regression of REM_CFOxCrosslisted has a negative coefficient (-0.242) with Credit_Ratingt+1. The relationship is only marginally significant (p = 0.080), but is confirmed in the ordered logistic regression, which is significant at the 5%-level (p = 0.037). The relationship between REM_CFO and Credit_Ratingt+1 is negative (-0.092): increased sales manipulation improves the credit rating. This effect is stronger for cross-listed companies. However, it should be noted that the relationship between REM_CFO and
TABLE 5
The regression and ordered logistic regression models of the two measures of cost of debt on the full factorial variable of the cross-listed dummy variable and the four measures of REM.
Panel A: The regression models of the relative interest expense on the full factorial variables of the cross-listed dummy variable and the four measures of REM.
Variables Int_Exp t+1 Int_Exp t+1 Int_Exp t+1 Int_Exp t+1
REM measure * -0.003** -0.014*** -0.009 -0.003* Dummy_Crosslisted (-1.96) (-3.42) (-1.23) (-1.73) REM measure 0.000 0.011*** 0.010** -0.001 (0.09) (3.78) (2.44) (-0.75) Dummy_Crosslisted 0.016*** 0.014*** 0.013*** 0.017*** (5.09) (5.10) (4.73) (5.43) Size -0.009*** -0.009*** -0.008*** -0.009*** (-11.51) (-12.43) (-11.65) (-12.66) ROAt-1 -0.013*** -0.013*** -0.014*** -0.012*** (-4.14) (-4.59) (-4.76) (-4.10) MTBRatio -0.001 -0.000 -0.000 -0.000 (-1.49) (-0.72) (-1.14) (-1.29) Leveraget-1 -0.003*** -0.003*** -0.003*** -0.003*** (-4.52) (-4.99) (-4.63) (-4.79) BigN -0.019*** -0.017*** -0.018*** -0.018*** (-4.73) (-4.67) (-4.63) (-4.66) Losst-1 0.023*** 0.020*** 0.021*** 0.021*** (7.70) (7.48) (7.48) (7.39) Fyear -0.001 -0.001 -0.001 -0.001 (-0.88) (-1.38) (-0.82) (-1.25) SIC_2 -0.000 -0.000 -0.000 -0.000 (-0.59) (-0.87) (-0.66) (-0.65)
Panel B: The regression models of the credit rating on the full factorial variables of the cross-listed dummy variable and the four measures of REM.
Variables Credit_Ratingt+1 Credit_Ratingt+1 Credit_Ratingt+1 Credit_Ratingt+1
REM measure * 0.053 -0.242* -0.539** 0.072 Dummy_Crosslisted (1.01) (-1.75) (-2.17) (1.34) REM measure 0.076*** -0.044 0.724*** 0.053** (2.96) (-0.59) (6.13) (2.09) Dummy_Crosslisted 0.128 -0.053 -0.054 0.114 (1.04) (-0.49) (-0.48) (0.95) Size -1.432*** -1.381*** -1.386*** -1.417*** (-42.30) (-44.69) (-42.77) (-43.58) ROAt-1 -5.920*** -5.406*** -5.689*** -5.851*** (-8.84) (-7.80) (-8.57) (-9.25) MTBRatio -0.103*** -0.106*** -0.097*** -0.106*** (-8.29) (-8.98) (-8.16) (-8.81) Leveraget-1 0.275*** 0.265*** 0.263*** 0.271*** (7.74) (8.14) (7.72) (8.05) BigN -0.086 -0.109 -0.112 -0.106 (-0.49) (-0.65) (-0.62) (-0.65) Losst-1 0.980*** 1.154*** 1.084*** 1.017*** (6.86) (8.15) (7.71) (7.36) Fyear 0.172*** 0.153*** 0.163*** 0.172*** (4.81) (3.98) (4.20) (4.90) SIC_2 -0.010 -0.004 0.004 -0.016 (-0.48) (-0.31) (0.24) (-0.95)
Panel C: The ordered logistic regression models of the credit rating on the full factorial variables of the cross-listed dummy variable and the four measures of REM.
Variables Credit_Ratingt+1 Credit_Ratingt+1 Credit_Ratingt+1 Credit_Ratingt+1
(5.49) (5.12) (5.32) (5.57)
SIC_2 -0.021*** -0.012 -0.006 -0.022***
(-3.45) (-1.55) (-0.56) (-4.83)
The table shows the regressions and ordered logistic regression of the measures of cost of debt on the full factorial variables of
Dummy_Crosslisted and the four measures of REM and associated control variables for the complete sample. Int_Expt+1 is winsorized
at its respective 5th and 95th percentile, all other continuous variables are winsorized at their respective 2nd and 98th percentile, except for Size, of which the natural logarithm is taken, to rule out the influence of potential outliers, and year and industry fixed effects are included.
The four measures of REM are REM (1st column), REM_CFO (2nd column), REM_Prod (3rd column) and REM_DiscExp (4th column). T-statistics in parentheses. Coefficients with *, ** or *** are statistically significant at the 0.1, 0.05 or 0.01 levels, resp ectively.
The regression of REM_ProdxCrosslisted on Credit_Ratingt+1 shows a significant (p = 0.030) and negative (-0.539) relationship. The relationship between REM_Prod and Credit_Ratingt+1 is significant (p = 0.000) positive (0.483), thus an increase in overproduction leads to a downgrade of the credit rating. This affected is attenuated for cross-listed companies; if they engage in overproduction, their credit rating is downgraded less than for American companies engaging in the same level of overproduction. However, the ordered logistic regression of REM_ProdxCrosslistedon Credit_Ratingt+1 is insignificant (p = 0.113), making the outcome not robust. Lastly, the relationship between REM_DiscExpxCrosslistedand Credit_Ratingt+1 is insignificant for both the normal regression (p = 0.179) and the ordered logistic regression (p = 0.341). The relationship between
REM_DiscExp and Credit_Ratingt+1 is not affected by being cross-listed. The outcomes show partial confirmation of the third hypothesis.
4.4 Additional analyses
As an additional test for the third hypothesis, the regressions of REM on cost of debt are run separately for the American and cross-listed sample (table 6A and 6B), and a seemingly unrelated estimation (suest) test is run to capture differences in the coefficients of REM (table 7).
Whereas American companies have an insignificant relationship (p = 0.665) between REM and
Int_Expt+1, this relationship is significant (p = 0.039) and negative (-0.003) for cross-listed companies. Thus, REM will lower the relative interest expense for cross-listed companies, which does not hold for American companies. This difference is marginally significant (p = 0.059) (table 7), in line with the third hypothesis. Looking at the credit rating, for cross-listed companies, the use of REM does not affect Credit_Ratingt+1 (p = 0.214 and p = 0.218), whereas it is significant (p = 0.001 and p = 0.000, respectively) and positive (0.085 and 0.090, respectively) for American companies. Thus, REM will downgrade the credit rating for American companies, but not for cross-listed companies. This relationship is also more favourable for cross-listed companies and in line with the third hypothesis. However, the difference is not significantly different (p = 0.719). Both outcomes confirm the outcomes of table 5, thereby making the outcomes robust.
For American companies, REM_CFO and REM_Prod are both significant (p = 0.001 and p = 0.019, respectively) and positive (0.010 and 0.010, respectively) related to Int_Expt+1. For cross-listed companies,
the coefficient of the comprehensive measure equals the significance of REM_DiscExp, showing the big impact of REM_DiscExpon the comprehensive measure.
The suest test partially confirms the differences in relationship between REM and Int_Expt+1 across American and cross-listed companies. Although the coefficient for REM_Prod is not significantly different (p =
0.498), the coefficients for REM_CFO and REM_DiscExp are statistically significant (p = 0.004 and p = 0.087,
respectively), thereby confirming that the relationship between REM_Prod and REM_DiscExp and Int_Expt+1 are significantly different for American and cross-listed companies.
TABLE 6A
The regression and ordered logistic regression models of the two measures of cost of debt on the four measures of REM for the American sample.
Panel A: The regression models of the relative interest expense on the four measures of REM for the American sample. Variables Int_Exp t+1 Int_Exp t+1 Int_Exp t+1 Int_Exp t+1
REM measure 0.000 0.010*** 0.010** -0.000
(0.43) (3.35) (2.34) (-0.45)
Panel B: The regression models of the credit rating on the four measures of REM for the American sample. Variables Credit_Ratingt+1 Credit_Ratingt+1 Credit_Ratingt+1 Credit_Ratingt+1
REM measure 0.085*** -0.069 0.705*** 0.060**
(3.33) (-0.92) (5.95) (2.37)
Panel C: The ordered logistic regression models of the credit rating on the four measures of REM for the American sample. Variables Credit_Ratingt+1 Credit_Ratingt+1 Credit_Ratingt+1 Credit_Ratingt+1
REM measure 0.090*** -0.063 0.552*** 0.068***
(4.01) (-0.98) (5.04) (3.11)
TABLE 6B
The regression and ordered logistic regression models of the two measures of cost of debt on the four measures of REM for the cross-listed sample.
Panel A: The regression models of the relative interest expense on the four measures of REM for the cross-listed sample. Variables Int_Exp t+1 Int_Exp t+1 Int_Exp t+1 Int_Exp t+1
REM measure -0.003** -0.003 0.005 -0.003**
(-2.07) (-0.80) (0.85) (-2.53)
Panel B: The regression models of the credit rating on the four measures of REM for the cross-listed sample. Variables Credit_Ratingt+1 Credit_Ratingt+1 Credit_Ratingt+1 Credit_Ratingt+1
REM measure 0.065 -0.201* 0.243 0.078
(1.25) (-1.66) (1.16) (1.43)
Panel C: The ordered logistic regression models of the credit rating on the four measures of REM for the cross-listed sample. Variables Credit_Ratingt+1 Credit_Ratingt+1 Credit_Ratingt+1 Credit_Ratingt+1
REM measure 0.061 -0.228* 0.235 0.071
(1.23) (-1.95) (1.20) (1.39)
The tables show the regressions and ordered logistic regression of the four measures of REM on the measures of cost of debt and associated control variables for the American and cross-listed sample separately. Int_Expt+1 is winsorized at its respective 5th and 95th
percentile, all other continuous variables are winsorized at their respective 2nd and 98th percentile, except for Size, of which the natural logarithm is taken, to rule out the influence of potential outliers, and year and industry fixed effects are included. The associated control variables are included in the regression and ordered logistic regression models but excluded in the table.
The four measures of REM are REM (1st column), REM_CFO (2nd column), REM_Prod (3rd column) and REM_DiscExp (4th column). T-statistics in parentheses. Coefficients with *, ** or *** are statistically significant at the 0.1, 0.05 or 0.01 levels, resp ectively.
Panel B (regression) and panel C (ordered logistic regression) show the separate measures of REM on
Credit_Ratingt+1. The significant and positive relationship between REM and Credit_Ratingt+1 for American companies is driven by REM_Prod and REM_DiscExp, which both show significant positive relationships on both the regression and the ordered logistic regression. Whereas REM_CFO is insignificant on Credit_Ratingt+1 for American companies, it is the only REM measure that is (marginally) significant related to Credit_Ratingt+1 for cross-listed companies(-0.201 and -0.228, respectively).
