Abstract
We investigate the cost of debt effects for firms that manage earnings per share (EPS) through abnormal share repurchases. Although prior research finds a significant cost of debt decrease for firms that meet earnings benchmarks, our results suggest that firms using the abnormal share repurchase strategy realize no cost of debt decrease associated with meeting earnings benchmarks. We find some evidence of a smaller decrease in cost of debt associated with measures of abnormal decreases in cash flows but weak evidence for measures that are cash flow increasing. We also find that the effect of using abnormal stock repurchases to meet earnings benchmarks leads to smaller reductions in the cost of debt when compared with the cost reduction when earnings benchmarks are met through accruals management. This study extends prior literature regarding the effects on the cost of debt through alternative strategies to meet earnings benchmarks and will be of interest to managers as they consider the impact of their managerial decisions.
Introduction
Prior research has uncovered benefits associated with meeting earnings benchmarks in the debt market. Specifically, Jiang (2008) finds that firms that meet earnings benchmarks enjoy a decreased cost of debt. This decrease, however, is of smaller magnitude (but not zero) for firms that meet or beat (hereafter stated as “meet”) earnings benchmarks through earnings management (Jiang, 2008, p. 407). Jiang primarily uses proxies for accruals management but fails to fully test the possibility that bond raters may respond differently when earnings benchmarks are met using strategies that may distort current or future cash flows (hereafter, “real activities manipulation”). We measure seven aspects of real activities manipulation and find that firms using real activities management tend to experience an increased cost of debt. Specifically, they are less likely to have a credit rating upgrade and/or more likely to have a credit rating downgrade.
We define real activities manipulation as the actions of management that deviate from normal business practices, undertaken with the primary objective of meeting certain earnings per share (EPS) thresholds. Examples include stock repurchases, price discounts to temporarily increase sales, reduction in discretionary expenditures, and overproduction to lower cost of goods sold. The majority of prior research on earnings management has focused on detecting abnormal accruals (hereafter stated as “accruals manipulation”). Accruals manipulation does not have any direct effect on the cash flows of the firm. Real activities manipulation, on the contrary, affects both cash flows and other accruals as well. In particular, our article is unique in analyzing the effects on the cost of debt for firms who meet earnings benchmarks using judicious stock repurchases.
We expect bondholders respond more intensively to manipulations involving cash flows than those that solely involve accruals. Bondholders are likely to react to accruals manipulation differently than real activities manipulation due to the nature of their claim in the firm. Bondholders own a fixed claim in the firm that exposes them to the firm’s downside risk, but they do not share in the firm’s upside growth. In addition, the primary concern of debtholders is cash flows, liquidity, and solvency. Therefore, accruals manipulation, which does not affect cash flows, ought not to be as important to bondholders as abnormal stock repurchases that typically transfer wealth to stockholders (Maxwell & Stephens, 2003). Stock repurchases at a favorable price are effectively dividends paid to shareholders while inflating the EPS of the company. The cash flow distributions, in turn, decrease cash and other assets available to meet firm obligations, such as debt. A significant cash flow distribution increases the probability of default on this debt, which adversely affects bondholders and bond prices. 1
Using a commonly used proxy for cost of debt—credit ratings—we determine whether the cost of debt decrease (documented by Jiang, 2008) from meeting earnings benchmarks is attenuated for those firms that manage earnings via real activities manipulation. We examine abnormal stock repurchases as well as several measures of real activities manipulation including abnormal cash flows from operations, abnormal production costs, abnormal discretionary expenditures, and two aggregate measures from prior literature. Our evidence suggests that firms that manage EPS through share repurchases realize a smaller decrease in the cost of debt than that associated with alternative choices of real activities management.
We gauge these results by also examining the effects of using accruals manipulation to meet earnings benchmarks on the cost of debt. As hypothesized, based on bondholders’ interests in cash flow effects of transactions, we find that the attenuation of the previously identified cost of debt decrease is more pronounced when earnings benchmarks are reached using stock repurchases compared with accruals manipulation. The economic magnitude of our results suggests that firms which meet earnings benchmarks using abnormal repurchases have a higher probability of receiving an adverse credit rating change. Thus, the cost of debt decrease achieved when meeting an earnings benchmark identified by Jiang (2008) appears to be completely eliminated. Although we find similar results when managers use a combination of other real activities manipulation strategies (proxied by our aggregate measures) to meet earnings benchmarks, at least some reduction in the cost of debt under these alternative strategies.
Our study contributes to the existing literature in several ways. First and foremost, our evidence suggests that the use of abnormal stock repurchases matters more to bondholders than accruals management, likely due to the transfer of resources to equity holders when stocks are repurchased. This is consistent with bondholders’ natural tendency to focus on cash flows rather than accruals. Our results also imply that bondholders may be more concerned with real activities manipulation activities that are cash flow decreasing compared with those that are cash flow increasing. 2 We also expand the findings of Jiang (2008) by examining the effect of real activities manipulation on the relationship between meeting earnings benchmarks and cost of debt. Third, Gunny (2010) finds that firms using real activities manipulation to meet earnings benchmarks experience better future performance. We add to this stream of literature by identifying a potential negative consequence of using real activities manipulation to meet earnings benchmarks.
Fourth, our study also adds to the existing credit ratings literature by providing evidence on the impact of managerial decision making on credit ratings and ratings changes. Our results point to real activities manipulation possibly influencing credit ratings, credit ratings changes, or even credit watch listing. We build on Maxwell and Stephens (2003), who find that the announcement of stock repurchase programs negatively affects bond ratings. We find that specific stock repurchase transactions used to meet earnings benchmarks are associated with a decrease in bond ratings. Finally, our study adds to the literature on bondholders’ perspectives on the real activities measures that we study.
The remainder of this article is organized as follows: The next or second section reviews the related literature in more detail and guides our hypotheses. The third section describes our sample, empirical methodology, and measurement of real activities manipulation. The fourth section reports our empirical results. The fifth and concluding section summarizes our findings and suggests future extensions to our research.
Literature Review and Hypothesis Development
Prior empirical research provides evidence that firms manage earnings to report positive profits, avoid earnings decreases, meet analysts’ earnings expectations, and maintain long earnings “strings” (Burgstahler & Dichev, 1997; Burgstahler & Eames, 2006; Degeorge, Patel, & Zeckhauser, 1999; Hayn, 1995; Myers, Myers, & Skinner, 2007). The incentives driving these actions are illustrated by additional literature which suggests that investors reward firms that report consistent earnings growth, consistently meet analysts’ earnings forecasts, and avoid earnings disappointments (Barth, Elliot, & Finn, 1999; Bartov, Givoly, & Hayn, 2002; Kasznik & McNichols, 2002; Myers et al., 2007; Skinner & Sloan, 2002).
Managers’ sensitivity to these incentives is confirmed by Graham, Harvey, and Rajgopal (2005). In a field/survey study in which more than 400 executives were surveyed or interviewed, managers say that they are motivated to meet analysts’ quarterly EPS forecasts to build credibility and preserve their reputation with capital markets, maintain or increase the firm’s stock price, and avoid the uncertainty created by missing the forecast. Graham et al. (2005) show that managers recognize the importance of four different benchmarks: (a) same quarter last year (85.1% agree or strongly agree that the metric is important), (b) analyst consensus estimate (73.5%), (c) reporting a profit (65.2%), and (d) previous quarter EPS (54.2%). Their surveys also show that managers are more inclined to manage earnings through real activities manipulation than through accounting actions (accruals manipulation). Eighty percent of the survey participants reported that they would decrease discretionary spending to meet an earnings target, while the use of accounting actions received notably little support. 3 Since much of prior research has focused on accruals manipulation, our motivation in this article is to provide more evidence about real activities earnings management, particularly given managers’ stated preferences for this form of manipulation.
Although most prior research on benchmark meeting focuses on stocks, Jiang (2008) investigates bonds. He finds that the cost of debt decreases for firms that meet earnings benchmarks, but the use of earnings management erases a part of this reduction. We expect the same effect to be evident when managers use real activities manipulation to meet these benchmarks. Meeting earnings benchmarks through abnormal accounting or abnormal activities devalues the performance of the firm and should be treated unfavorably by both equity and bond investors.
Given the nature of their financial claims, bondholders are exposed to the downside risk of the firms they lend to but are not privy to the advantages that come along with the inherent growth potential of these firms. For this reason, bondholders and stockholders typically have differing preferences regarding activities that affect the cash flow of the firm. Due to the direct cash flow effect of real activities manipulation, we conjecture that these manipulations will have a more significant effect on cost debt than will accruals manipulation. That is, not only is the performance of the firm in meeting the earnings benchmark more suspect, the division of future cash flows between stockholders and bondholders may shift in favor of stockholders.
We examine several measures of real activities manipulation. The first method we study is stock repurchases. Prior literature has shown that stock repurchases are often intended to boost EPS. A review of the mechanics of repurchases provides insight as to how this method can be achieved. In general, the formula used to calculate EPS is as follows: Net Income / Common Shares Outstanding. Therefore, if a company repurchases shares of their common stock from their investors, this decreases the denominator of the EPS formula, while the numerator of the formula remains theoretically unchanged. 4
The most common way that U.S. firms repurchase their stock is through open-market repurchase programs. 5 Open-market programs accounted for 94.3% of all repurchases announced during the 1990s and 95.2% of the total dollar value of shares repurchased (Grullon & Ikenberry, 2000). Cook, Krigman, and Leach (2003) point out that open-market repurchase plans offer considerable flexibility to managers in choosing when to buy stock or even whether to buy any stock at all. These firms are not required to provide details about the timing, price, or volume of their individual repurchase transactions. Therefore, analysts and investors typically learn about open-market repurchase transactions indirectly from quarterly or annual report disclosures about firm cash flows and shares outstanding. These flexible characteristics make stock repurchases an ideal way to manipulate EPS in a subtle manner.
Brav, Graham, Harvey, and Michaely (2005) confirm the impetus of repurchases as being earnings management related. They report that over 76% of the CEOs, CFOs, and treasurers, who responded to their survey on dividend and stock buyback policies, said that increasing EPS is an “important” or “very important” consideration in their firms’ stock repurchase decisions. A few empirical studies have provided support for the use of stock repurchases as an earnings management device. Bens, Nagar, and Wong (2002) and Bens, Nagar, Skinner, and Wong (2003) report that managers repurchase stock to avoid EPS dilution arising from employee stock exercises and employee stock grants. Bens et al. (2002) find that, on average, firms repurchase shares at an overall value of approximately 3.2% of their sales. Bens et al. (2003) find that, on average, firms repurchase 2% of their shares outstanding at the beginning of each year, which increases EPS by 2%. Hribar, Jenkins, and Johnson (2006) find a disproportionally large number of accretive stock repurchases among firms that would have missed analysts’ forecasts without the repurchase. Myers et al. (2007) find that firms use various earnings management tools, including repurchases, to maintain long “strings” of consecutive increases in EPS. We expect to uncover a substantial number of firms using repurchases to manage EPS, given the survey evidence on the importance of EPS in repurchase decisions, the timing and flexibility of open-market repurchase plans, and the proof of repurchases being used as an earnings management tool. This allows us to determine the debt market consequences of these repurchases.
Stock repurchases, as a form of real activities manipulation, ought to be of particular concern to bondholders. The cash used to repurchase shares could have been used to pay debt payments. If this decrease in cash flow is significant, it could increase the default risk of the firm, which causes a loss to bondholders. Maxwell and Stephens (2003) find that, consistent with this hypothesis, there is a negative and significant abnormal bond price reaction to the announcement of stock repurchase programs. 6 Yet to be studied by prior literature is how specific stock repurchases affect the company’s cost of debt.
We also use three real activities manipulation measures identified in Roychowdhury’s (2006) study. First, we consider abnormal cash flows from operations. We proxy for real activities manipulation with negative abnormal cash flows from operations (see more measurement details in the next section), which signals an increased likelihood of managers attempting to inflate current sales by offering price discounts, channel stuffing, and so on. (Roychowdhury, 2006). This form of real activities manipulation is cash flow decreasing and earnings increasing; thus, we expect bondholders to react more negatively than shareholders.
Second, we consider abnormal production costs. We proxy for real activities manipulation with positive abnormal production costs (see more measurement details in the next section), which signals an increased likelihood of managers producing more goods than necessary to report lower cost of goods sold and, thus, overstate earnings (Roychowdhury, 2006). This form of real activities manipulation is also cash flow decreasing and earnings increasing; thus, we expect bondholders to react more negatively than shareholders.
Third, we consider abnormal discretionary expenditures. We proxy for real activities manipulation with negative abnormal discretionary expenditures (see more measure details in the next section), which signals that managers are cutting discretionary expenditures such as research and development, advertising, and so on (Roychowdhury, 2006). We are less sure of how bondholders will react to this proxy because it is cash flow increasing and earnings increasing.
Finally, following Zang (2012) and Cohen and Zarowin (2010), we calculate two aggregate real activities manipulation measures (see the measurement details in the next section). Again, using these real activities manipulations, managers more directly affect the cash flows of a firm compared with accruals management, which is of significant importance to a firm’s bondholders.
Method
Empirical Model Development
To investigate whether firms that meet earnings benchmarks using real activities manipulation have attenuated cost of debt decreases (H1), we use firm credit ratings to proxy for a firm’s cost of debt following prior research (Ahmed, Billings, Morton, & Stanford-Harris, 2002; Jiang, 2008; Shi, 2003). Credit ratings represent rating agencies’ assessments of a firm’s creditworthiness and default risk. Prior research suggests that ratings downgrades affect both bond and stock prices (Dichev & Piotroski, 2001; Holthausen & Leftwich, 1986) and can also trigger accelerated debt repayment. Credit ratings serve as a proxy for cost of debt, given that default risk is a determinant of yield spreads that measure the risk premium charged by investors in the debt market.
The main models are specified as rating change models, similar to prior research. We cannot observe the metrics that credit ratings agencies use to determine specific credit ratings nor can we observe the weights the agencies put on each metric. Therefore, a levels analysis would suffer the possibility of correlated omitted variables driving the results. A changes analysis mitigates this concern. Also, credit ratings are often described as “sticky,” implying that error terms in a levels model could be autocorrelated. The changes specification mitigates this concern as well.
The following ordered logit equation developed in Jiang’s (2008) study is used to test our hypotheses7,8:
Beatit takes one of the three following specifications: Profitit is equal to 1 if firm i’s basic EPS before extraordinary items is greater than or equal to 0 in year t, and 0 otherwise; Incrit is equal to 1 if firm i’s EPS before extraordinary items in year t is greater than or equal to that of year t– 1, and 0 otherwise; Surpit is equal to 1 if firm i’s reported EPS in year t is greater than or equal to its most recent analyst forecast in year t, and 0 otherwise. RealEMit is equal to 1 if firm i has been determined to have been taking part in real activities manipulation (to be described in the next section), and 0 otherwise. EarningsControlit takes one of the following continuous earnings variables corresponding to each earnings benchmark: EPSit is equal to firm i’s EPS before extraordinary items in year t divided by its stock price at the end of year t– 1 and is used as a control with Profitit; ΔEPSit is equal to the change in firm i’s EPS before extraordinary items between year t and t– 1 divided by its stock price at the end of year t– 1 and is used as a control with Incrit; UE_EPSit is equal to firm i’s actual EPS minus the single most recent analyst forecast for year t divided by its stock price at the end of year t– 1 and is used as a control with Surpit. See definitions for other control variables in Appendix A.
H1, that the effect of meeting an earnings benchmark is lower when using real earnings management, is consistent with α2 being found to be positive and significant.
The control variables in Equation 1 are meant to capture base determinants of bond ratings (Ahmed et al., 2002; Campbell & Taksler, 2003; Shi, 2003). To control for the effect of performance on cost of debt, we include operating cash flow (CFOit), times interest earned ratio (Timesit), and book-to-market ratio (BMit). To control for the effect of risk on cost of debt, we include standard deviation of return on assets (StdROAit), standard deviation of daily stock returns (StdRetit), size (Sizeit), and leverage (Levit). We also control for time-varying factors related to ratings in Equation 1 by including year fixed effects following Blume, Lim, and Mackinlay (1998).
We use an approach similar to Jiang (2008), looking to improve on his attempt to capture the interaction among earnings management, benchmark meeting, and cost of debt. He includes a proxy for real activities manipulation, abnormal cash flows from operations, in his tests on the attenuating effects of earnings management on cost of debt reductions from firms meeting earnings benchmarks. Abnormal cash flows from operations (Roychowdhury, 2006), however, is used to capture the effect of firms that meet earnings targets through lower abnormal cash expenditure. This proxy could contain both negative abnormal cash flow effects, caused by price discounts and lenient credit terms offered to temporarily increase sales, and positive abnormal cash flow effects, resulting from a reduction in discretionary expenses. The net effect of these activities is ambiguous; thus, this proxy is not a direct method of isolating the directional effects of real activities manipulation.
To investigate whether using real activities manipulation to meet earnings benchmarks causes a greater attenuation in the cost of debt decrease than using accruals management to meet those benchmarks (H2), we again use the ordered logit regression (see above). This time, we include two interaction variables: one where the earnings benchmark is interacted with the accruals management proxy (EM_ACCR) and another where the earnings benchmark is interacted with a real activities manipulation proxy. We then use a chi-square test to determine whether the difference between the coefficients of these two variables is different from 0. Our H2 expectation is that the coefficient of the real activities manipulation interaction variable will be significantly greater than the coefficient of the accruals management interaction variable.
The most important part of our research design is the determination of which firms are managing earnings by real activities manipulation. This process is described in the following section.
Expectations Models
Repurchases—ACCRE
We estimate expected repurchases to determine whether unexpected repurchases were used to manage EPS. To estimate expected levels of repurchases, we use a two-stage Heckman expectations model using quarterly data 9 similar to Hribar et al. (2006) developed from factors identified in Dittmar’s (2000) study as determinants of repurchases made by firms. 10 This approach computes expected share repurchases as the product of the estimated probability that a firm will repurchase shares in a given quarter, the first stage, and the expected level of repurchases conditional upon a repurchase occurring, the second stage. Our sample includes firms that make repurchases as well as firms that do not; this approach allows us to compute a repurchase expectation for every firm-quarter in the sample. 11 The first-stage probit equation evaluates the repurchase decision as a binary variable:
We include two lags of repurchases to capture the persistence of repurchase activity. The remaining variables in the equation are determinants of share repurchases (Dittmar, 2000). See Appendix A for the variable definitions.
The second-stage regression is identical to the first-stage regression with two exceptions: (a) the binary repurchase variables are replaced by the dollar value of repurchases, and (b) the second-stage regression uses two variables computed from the first stage to incorporate information about the probability of a repurchase. The first stage predicted values are transformed into estimates of λ and Λ that represent the standard normal probability density function and cumulative density function, respectively. The variable λ is included as a separate control variable in the second stage, and Λ is multiplied by each of the independent variables in the second stage. 12 The purpose of this procedure is to compute estimates for repurchasing firms as well as those that do not make repurchases. Both the first- and second-stage descriptive statistics and the regression results are respectively shown in Table 1 and Appendix B. All variables in the first stage are significant at the 1% level in the predicted direction and the pseudo R2 is 34.8%. All the variables in the second stage except CAPEX are significant at the 1% level in the predicted direction and the adjusted R2 is 57.6%. These results are consistent with those reported in Hribar et al.’s (2006) study. 13
Descriptive Statistics.
Note. The sample period is 1988-2008. Rating is firm i’s S&P senior debt rating in year t and ΔRatingit+1 = Ratingit+1–Ratingit; Profit equals to 1 if firm i’s basic EPS before extraordinary items is greater than or equal to 0 in year t, and 0 otherwise; Incr equals to 1 if firm i’s EPS before extraordinary items in year t is greater than or equal to that of year t– 1, and 0 otherwise; Surp equals to 1 if firm i’s reported EPS in year t is greater than or equal to its most recent analyst forecast in year t, and 0 otherwise; EPS is firm i’s EPS before extraordinary items in year t divided by its stock price at the end of year t– 1; ΔEPS is change in firm i’s EPS before extraordinary items between year t and t– 1 divided by its stock price at the end of year t– 1; UE_EPS is firm i’s actual EPS minus the single most recent analyst forecast for year t divided by its stock price at the end of year t– 1; StdROA is firm i’s standard deviation of ROA calculated using 5 years of data from years t– 4 to t. ROA is calculated as net income before extraordinary items deflated by beginning of year total assets; Times is the logarithm of (1 + times interest earned ratio) where times interest earned ratio is firm i’s operating income before depreciation and interest expense divided by interest expense at year t; StdRet is the standard deviation of firm i’s daily stock returns during year t; BM is the logarithm of (book value of equity / market value of equity) for firm i measured at the end of year t; Size is the logarithm of total assets for firm i at the end of year t; Lev is firm i’s long-term debt divided by total assets at the end of year t; CFO is firm i’s operating cash flows at year t deflated by beginning of year total assets; RND is firm i’s research and development expense deflated by beginning of year total assets; RepAmt is the continuous measure of the amount of stock repurchased in a given firm-quarter (in US$M); IfRepurchase is equal to 1 in firm-quarters in which a stock repurchase was made, and 0 otherwise; Cash is the ratio of cash and cash equivalents to total assets; CAPEX is the ratio of capital expenditures over the past year to total assets; DividendYield is the dividends per share divided by beginning of quarter stock price; Debt is the ratio of current and long-term debt to total assets; Size is the logarithm of total assets for firm i at the end of year t. S&P = Standard & Poor’s Financial Services; EPS = earnings per share; ROA = return on assets.
Note. Panels B, C, and D provide descriptive information for Credit Ratings Changes (our dependent variable) in the sample. Panel B provides information for the frequency of credit ratings changes for the entire sample, as well as for high default risk and low default risk firms. Downgrade includes all firm-years whose one year ahead rating has gone down. Unchanged includes all firm-years whose one year ahead rating has not changed. Upgrade includes all firm-years whose one year ahead rating has gone up. The high default risk subsample includes observations that are rated as noninvestment grade (BB+ and below). The Low Default Risk subsample includes observations that are rated as investment grade (BBB– and above). Panel C provides information on the frequency of missing/meeting benchmarks among the three credit rating changes subsamples. Loss is all observations where net income was negative. Profit is all observations where net income was positive. Decrease is all observations where current year net income was less than that of the prior year. Increase is all observations where current year net income was greater than that of last year. Miss is all observations where EPS was less than analyst estimate. Meet is all observations where EPS was greater than analyst estimate. Panel D reports the same descriptive information as Panel C but is only reported for observations in the high default risk subsample. EPS = earnings per share.
The second-stage equation produces unconditional expectations of the dollar amount of repurchases for each firm and quarter based only on information known at the beginning of the quarter. The estimates of dollar amounts of repurchases are then divided by beginning of quarter stock price to arrive at an expected number of shares repurchased. Then, we calculate measures of “As-if” EPS which is an estimate of investors’ expectations of EPS. We then compare this estimate with reported EPS to determine whether unexpected repurchases were used to manage EPS:
where NIt is net income; Ct is foregone earnings on cash used to repurchase stock; Shares Outstandingt is the number of common shares outstanding; UE_Repurchasest is unexpected shares repurchased (actual repurchases minus expected repurchases using the procedure described above).
Essentially, we assume that investors can perfectly predict all firm values except for repurchases. The unexpected number of shares repurchased is multiplied by .125 in the spirit of calculating weighted-average shares outstanding. 14 The “As-if” EPS number is then compared with reported EPS. Those firms where “As-if” EPS is less than actual EPS are labeled as having managed earnings through share repurchases, and the indicator variable ACCRE is set equal to 1, and 0 otherwise. 15
Abnormal CFO—RealEM_CFO
Following Dechow, Kothari, and Watts (1998), we model cash flows from operations as a linear function of sales and change in sales. 16 The following equation is estimated using a cross-sectional regression for every industry-year similar to Roychowdhury (2006):
The Equation 4 regression is run by industry-year; residuals from the equation are ranked into quintiles. Firms that rank in the lowest quintile of the industry-year are classified as having managed earnings using real activities manipulation in that year, and the variable RealEM_CFO is set equal to 1. For all other firms the variable RealEM_CFO is equal to 0. See other variable definitions in Appendix A.
Abnormal production costs—RealEM_PROD.
Following Roychowdhury (2006), we model production costs as a linear function of sales, change in sales in the current year, and prior year change in sales. The following equation is estimated using a cross-sectional regression for every industry-year:
The Equation 5 regression is run by industry-year; residuals from the equation are ranked into quintiles. Firms that rank in the highest quintile of the industry-year are classified as having managed earnings using real activities manipulation in that year, and the variable RealEM_PROD is set equal to 1. For all other firms the variable RealEM_PROD is equal to 0. See other variable definitions in Appendix A.
Abnormal discretionary expenditures—RealEM_DISX.
Again, following Roychowdhury (2006), we model discretionary expenditures as a linear function of lagged sales. The following equation is estimated using a cross-sectional regression for every industry-year:
The Equation 6 regression is run by industry-year; residuals from the equation are ranked into quintiles. Firms that rank in the lowest quintile of the industry-year are classified as having managed earnings using real activities manipulation in that year, and the variable RealEM_DISX is set equal to 1. For all other firms the variable RealEM_DISX is equal to 0. See other variable definitions in Appendix A.
Abnormal accruals—EM_ACCR.
We use the cross-sectional Jones model to estimate abnormal accruals (DeFond & Jiambalvo, 1994). The following equation is estimated for every industry-year:
The Equation 7 regression is run by industry-year; residuals from the equation are ranked into quintiles. Firms that rank in the highest quintile of the industry-year are classified as having managed earnings using accrual manipulation in that year, and the variable EM_ACCR is set equal to 1. For all other firms the variable EM_ACCR is equal to 0. 17 See other variable definitions in Appendix A.
Aggregating measures—RealEM_RM1 and RealEM_RM2.
Following Cohen and Zarowin (2010) and Zang (2012), we compute aggregate measures out of three of the real activities to entertain the idea that firms may use more than one mechanism to manage earnings toward their desired income level. The first aggregates abnormal discretionary expenditures with abnormal production costs. RM1 is calculated by adding abnormal production costs (the residual of Equation 5) to the product of abnormal discretionary expenditures (the residual of Equation 6) and (–1). Therefore, the higher the aggregate measure, the more likely the firm has managed earnings through the combination of these two measures. The second aggregates abnormal cash flows from operations (the residual of Equation 4) and abnormal discretionary expenditures (the residual of Equation 6). RM2 is calculated by multiplying both measures by (–1) and adding them together, so that higher levels signal higher probability of earnings management through sales manipulation and cutting of discretionary expenditures.
Once calculated, the aggregate measures are ranked into quintiles by industry-year; those firms in the top quintile are labeled as having managed earnings (RealEM_RM1 or RealEM_RM2 set equal to 1, and 0 otherwise).
Sample
Our main dependent variable is derived from the Standard & Poor’s Financial Services (S&P) senior debt ratings listed in Compustat. All our control variables come from Compustat and the Center for Research in Security Prices (CRSP), with analyst EPS forecasts being taken from the Institutional Brokers’ Estimate System (I/B/E/S) detail file. Our sample consists of the years 1988 to 2008, as we calculate cash flows from operations using the statement of cash flows. We exclude firms in public utilities (two-digit SIC [Standard Industrial Classification] codes 40-49) and financial service firms (two-digit SIC codes 60-67) because these firms have different operating characteristics and debt financing activities than industrial firms. 18 We also require that firms be listed on the NYSE (New York Stock Exchange), AMEX (American Stock Exchange), or NASDAQ (National Association of Securities Dealers Automated Quotations). This gives us a sample of 8,884 firm-year observations which consist of 1,264 unique firms (see Appendix C for more details of our sample selection process).
For the purpose of estimating expected stock repurchases, we use a different sample drawn from the quarterly Compustat database. To be in this stock repurchase sample, we require that firms have reported a positive purchase of stock, greater than US$10,000 (Compustat PRSTKCY), on the quarterly cash flow statement during the sample period. We exclude observations where more than 20% of outstanding shares were repurchased. 19
Results
Table 1, Panel A, presents descriptive statistics for key variables. Our sample consists of profitable firms that have increasing earnings and are on average meeting their analyst estimate for EPS, as evidenced by the mean and median of Profit, Incr, and Surp being greater than 0.5. Otherwise, the sample characteristics are generally in line with the sample characteristics from Jiang’s (2008) study. For example, the mean of Profit is 0.855 (0.836 reported by Jiang), the mean of EPS is 0.038 (0.035 reported by Jiang), and the mean of ΔSize is 0.100 (0.094 reported by Jiang). Minor differences are expected due to the difference in sample periods (Jiang’s sample period is 1985-2002 while ours is 1988-2008). Quarterly data descriptive statistics are presented in Table 1, Panel A, because quarterly data are used to calculate abnormal repurchases in determining whether firms used repurchases to manage EPS. 20
In Panel B, we show the distributions of credit ratings changes in our sample. Approximately 22% of firm-years in our sample experience a credit rating change; rating downgrades are more common than upgrades (13% vs. 9%). Panels C and D show a comparison of the distributions of ratings changes across our sample when a firm-year achieves our three earnings benchmarks versus when it does not. With all our measures, achieving (not achieving) an earnings benchmark is associated with a higher proportion of firms receiving a credit rating upgrade (downgrade). These results appear reasonable given the main results of Jiang (2008): There is a positive association between credit ratings and meeting an earnings benchmark.
The results in Table 1 are generally consistent with Jiang (2008), supporting the validity of our sample and results. For example, in Panel B, we show that in our aggregate sample, 13.2% of observations experienced a credit rating downgrade, 77.5% experienced no credit rating change, and 9.3% experienced a credit rating upgrade. Jiang (2008) shows these percentages as 14%, 78%, and 8%, respectively (Table 1, Panel A, p. 388). In Panel C, we show that in our credit rating downgrade sample, 31.4% of observations reported a loss on their financial statements, 12.4% of observations reported a loss in our unchanged credit rating sample, and 9.2% reported a loss in our credit rating upgrade sample. Jiang shows these percentages as 33%, 14%, and 10%, respectively (Table 1, Panel B, p. 388). In Panel D, we show that, in our high default risk sample, 47.7% (52.3%) of observations that meet (miss) their analyst earnings forecast experience a credit rating downgrade, 63.4% (36.6%) experience no credit rating change, and 69.3% (30.7%) experience a credit rating upgrade. Jiang shows that of high default risk firms that meet (miss) their earnings benchmarks 45% (55%) experience a credit rating downgrade, 59% (41%) experience no credit rating change, and 65% (35%) experience a credit rating upgrade. (Table 1, Panel C, p. 388).
Table 2 shows the Pearson and Spearman correlations of our regression variables. We report Spearman correlations above the diagonal and Pearson correlations below. Correlations that are significant at at least the 10% level (two-tailed test) are noted in bold. Consistent with Jiang (2008), each of the three earnings benchmarks is significantly and negatively correlated with our dependent variable. Also, our seven real activities manipulation proxies are largely significantly and positively correlated with our dependent variable. Most of our real activities manipulation measures have a significant negative correlation with our three earnings benchmarks, consistent with Gunny (2010), while our accruals management measure has a significant positive correlation with the earnings benchmarks. Zang (2012) finds that managers use real activities manipulation and accruals earnings management as substitutes—for example, when a firm participates in more real activities manipulation, it also tends to participate in less accruals earnings management. Thus, these correlation results appear reasonable. Furthermore, it appears that, on average, the firms in our sample that meet their earnings benchmarks use accruals earnings management rather than real activities manipulation. ACCRE has a significant positive correlation with the earnings benchmarks. This correlation appears reasonable due to the way ACCRE is defined: The company must meet an earnings benchmark using stock repurchases for this variable to have a value of 1. Except for ΔRND, all our control variables are significantly correlated with our dependent variable.
Correlation Matrix.
Note. Rating is firm i’s S&P senior debt rating in year t and ΔRatingit+ 1 = Ratingit+ 1–Ratingit; Profit equals to 1 if firm i’s basic EPS before extraordinary items is greater than or equal to 0 in year t, and 0 otherwise; Incr equals to 1 if firm i’s EPS before extraordinary items in year t is greater than or equal to that of year t– 1, and 0 otherwise; Surp equals to 1 if firm i’s reported EPS in year t is greater than or equal to its most recent analyst forecast in year t, and 0 otherwise; RealEM_RM1 equals to 1 if included in the top quintile of RM1, and 0 otherwise; RealEM_RM2 equals to 1 if included in the top quintile of RM2, and 0 otherwise; RealEM_CFO equals to 1 if included in the bottom quintile of residuals from Equation 4, and 0 otherwise; RealEM_PROD equals to 1 if included in the top quintile of residuals from Equation 5, and 0 otherwise; RealEM_DISX equals to 1 if included in the bottom quintile of residuals from Equation 6, and 0 otherwise; ACCRE equals to 1 if “As-if” EPS < actual EPS, and 0 otherwise; EM_ACCR equals to 1 if included in the top quintile of residuals from Equation 7, and 0 otherwise; EPS is firm i’s EPS before extraordinary items in year t divided by its stock price at the end of year t– 1; ΔEPS is change in firm i’s EPS before extraordinary items between year t and t– 1 divided by its stock price at the end of year t– 1; UE_EPS is firm i’s actual EPS minus the single most recent analyst forecast for year t divided by its stock price at the end of year t– 1; StdROA is firm i’s standard deviation of ROA calculated using 5 years of data from years t– 4 to t. ROA is calculated as net income before extraordinary items deflated by beginning of year total assets; Times is the logarithm of (1 + times interest earned ratio) where times interest earned ratio is firm i’s operating income before depreciation and interest expense divided by interest expense at year t; StdRet is the standard deviation of firm i’s daily stock returns during year t; BM is the logarithm of (book value of equity / market value of equity) for firm i measured at the end of year t; Size is the logarithm of total assets for firm i at the end of year t; Lev is firm i’s long-term debt divided by total assets at the end of year t; CFO is firm i’s operating cash flows at year t deflated by beginning of year total assets; RND is firm i’s research and development expense deflated by beginning of year total assets. Those correlations noted in bold are significant at at least the 10% level (two-tailed test). The p values are listed below the correlations. Spearman correlations are listed above the diagonal; Pearson correlations are listed below the diagonal. S&P = Standard & Poor’s Financial Services; EPS = earnings per share; ROA = return on assets.
Table 3 tests H1 with the benchmark variable of zero EPS (Profit). The coefficients in all columns for the Profit variable are negative and significant at the 1% level, effectively replicating the result from Jiang (2008) that finds firms that meet the zero-earnings threshold enjoy a lower cost of debt. Columns 1 and 2 test whether the combined real activities metrics from Cohen and Zarowin (2010) are associated with a mitigated cost of debt decrease for firms that meet the profit benchmark. The coefficients on Profit × RealEM_RM1 and Profit × RealEM_RM2 are both positive and significant at the 10% level, which provides weak evidence supporting our first hypothesis that firms which meet benchmarks through real activities manipulation realize a mitigation in the cost of debt decrease associated with meeting an earnings benchmark. Our H1 is further supported by results found in columns 3, 4, and 7. In these columns, abnormal cash flow from operations (Profit × RealEM_CFO), abnormal production expenditures (Profit × RealEM_PROD), and abnormal repurchases (Profit × ACCRE) that increase EPS are all shown to mitigate the cost of debt decrease associated with meeting an earnings benchmark. The result in column 5 fails to support our H1. Cutting discretionary expenditures does not seem to mitigate the cost of debt decrease associated with meeting an earnings benchmark. These results are not surprising considering that positive abnormal discretionary expenditures, at least in the short-term, are cash flow increasing. Also, we conjecture that debt markets may have difficulty in determining whether discretionary expenditures were cut due to market pressures surrounding earnings benchmarks, economic reasons, or cost-cutting related to efficiency goals. Column 6 shows the result when using our earnings management measure (EM_ACCR), which is insignificant. Overall, the results in Table 3 generally support H1.
Ordered Logit Regressions Testing H1 With Profit Benchmark.
Note. Table 3 reports coefficients and p values from ordered logit regressions testing H1 with the Profit Benchmark. In each column from 1 through 7, a different proxy for each of the seven earnings manipulation variables is tested. Profit equals to 1 if firm i’s basic EPS before extraordinary items is greater than or equal to 0 in year t, and 0 otherwise; RealEM_RM1 equals to 1 if included in the top quintile of RM1, and 0 otherwise; RealEM_RM2 equals to 1 if included in the top quintile of RM2, and 0 otherwise; RealEM_CFO equals to 1 if included in the bottom quintile of residuals from Equation 4, and 0 otherwise; RealEM_PROD equals to 1 if included in the top quintile of residuals from Equation 5, and 0 otherwise; RealEM_DISX equals to 1 if included in the bottom quintile of residuals from Equation 6, and 0 otherwise; EM_ACCR equals to 1 if included in the top quintile of residuals from Equation 7, and 0 otherwise; ACCRE equals to 1 if “As-if” EPS < actual EPS, and 0 otherwise; EPS is firm i’s EPS before extraordinary items in year t divided by its stock price at the end of year t– 1; CFO is firm i’s operating cash flows at year t deflated by beginning of year total assets; StdROA is firm i’s standard deviation of ROA calculated using 5 years of data from years t– 4 to t. ROA is calculated as net income before extraordinary items deflated by beginning of year total assets; Times is the logarithm of (1 + times interest earned ratio) where times interest earned ratio is firm i’s operating income before depreciation and interest expense divided by interest expense at year t; StdRet is the standard deviation of firm i’s daily stock returns during year t; BM is the logarithm of (book value of equity / market value of equity) for firm i measured at the end of year t; Size is the logarithm of total assets for firm i at the end of year t; RND is firm i’s research and development expense deflated by beginning of year total assets; Lev is firm i’s long-term debt divided by total assets at the end of year t. The p values are in parentheses (two-tailed tests). In all columns, the dependent variable is the change in credit rating in year t+ 1. All standard errors are clustered by firm. EPS = earnings per share; ROA = return on assets.
Significance of coefficients is marked by asterisks, with *, **, and *** representing significance at the 10%, 5%, and 1% levels, respectively.
Table 4 is identical to Table 3, save one difference: The benchmark variable in each of the seven columns is the prior year’s EPS, Incr. The coefficients in all columns for the Incr variable are negative and significant at the 1% level. This effectively replicates the result from Jiang (2008), which finds firms that meet their prior year’s EPS enjoy a lower cost of debt. Columns 1 and 2 test whether the combined real activities metrics from Cohen and Zarowin (2010) are associated with a mitigated cost of debt decrease for firms that meet the Incr benchmark. The coefficients on Incr × RealEM_RM1 and Incr × RealEM_RM2 are both positive and significant at the 5% and 1% levels, respectively, which supports our first hypothesis. The remaining columns are similar to the results in Table 3, with the exception that the abnormal cash flow metric Incr × RealEM_CFO is positive but insignificant. Overall, we find some support for H1 in Table 4.
Ordered Logit Regressions Testing H1 With Incr Benchmark.
Note. Table 4 reports coefficients and p values from ordered logit regressions testing H1 with the Incr Benchmark. In each column from 1 through 7, a different proxy for each of the seven earnings manipulation variables is tested. Incr equals to 1 if firm i’s EPS before extraordinary items in year t is greater than or equal to that of year t– 1, and 0 otherwise; RealEM_RM1 equals to 1 if included in the top quintile of RM1, and 0 otherwise; RealEM_RM2 equals to 1 if included in the top quintile of RM2, and 0 otherwise; RealEM_CFO equals to 1 if included in the bottom quintile of residuals from Equation 4, and 0 otherwise; RealEM_PROD equals to 1 if included in the top quintile of residuals from Equation 5, and 0 otherwise; RealEM_DISX equals to 1 if included in the bottom quintile of residuals from Equation 6, and 0 otherwise; EM_ACCR equals to 1 if included in the top quintile of residuals from Equation 7, and 0 otherwise; ACCRE equals to 1 if “As-if” EPS < actual EPS, and 0 otherwise; ΔEPS is the change in firm i’s EPS before extraordinary items between year t and t– 1 divided by its stock price at the end of year t– 1; CFO is firm i’s operating cash flows at year t deflated by beginning of year total assets; StdROA is firm i’s standard deviation of ROA calculated using 5 years of data from years t– 4 to t. ROA is calculated as net income before extraordinary items deflated by beginning of year total assets; Times is the logarithm of (1 + times interest earned ratio) where times interest earned ratio is firm i’s operating income before depreciation and interest expense divided by interest expense at year t; StdRet is the standard deviation of firm i’s daily stock returns during year t; BM is the logarithm of (book value of equity / market value of equity) for firm i measured at the end of year t; Size is the logarithm of total assets for firm i at the end of year t; RND is firm i’s research and development expense deflated by beginning of year total assets; Lev is firm i’s long-term debt divided by total assets at the end of year t. the p values are in parentheses (two-tailed tests). In all columns, the dependent variable is the change in credit rating in year t+ 1. All standard errors are clustered by firm. EPS = earnings per share; ROA = return on assets.
Significance of coefficients is marked by asterisks, with *, **, and *** representing significance at the 10%, 5%, and 1% levels, respectively.
Table 5 repeats the tests in Tables 3 and 4 with the benchmark being the firm’s most recent analyst estimate of EPS, Surp. The coefficients in all columns for the Surp variable are negative and significant at the 1% level. This effectively replicates the result from Jiang (2008), which finds firms that beat their most recent analyst estimate of EPS enjoy a lower cost of debt. In this table, we find some support of our hypothesis in Column 3; we continue to find robust support of H1 when considering abnormal stock repurchases (Surp × ACCRE). Columns 1, 2, 4, 5, and 6 show positive coefficients on the variables of interest, but no significance. 21
Ordered Logit Regressions Testing H1 With Surp Benchmark.
Note. Table 5 reports coefficients and p values from ordered logit regressions testing H1 with the Surp Benchmark. In each column from 1 through 7, a different proxy for each of the seven earnings manipulation variables is tested. Surp equals to 1 if firm i’s reported EPS in year t is greater than or equal to its most recent analyst forecast in year t, and 0 otherwise; RealEM_RM1 equals to 1 if included in the top quintile of RM1, and 0 otherwise; RealEM_RM2 equals to 1 if included in the top quintile of RM2, and 0 otherwise; RealEM_CFO equals to 1 if included in the bottom quintile of residuals from Equation 4, and 0 otherwise; RealEM_PROD equals to 1 if included in the top quintile of residuals from Equation 5, and 0 otherwise; RealEM_DISX equals to 1 if included in the bottom quintile of residuals from Equation 6, and 0 otherwise; EM_ACCR equals to 1 if included in the top quintile of residuals from Equation 7, and 0 otherwise; ACCRE equals to 1 if “As-if” EPS < actual EPS, and 0 otherwise; UE_EPS is firm i’s actual EPS minus the single most recent analyst forecast for year t divided by its stock price at the end of year t– 1; CFO is firm i’s operating cash flows at year t deflated by beginning of year total assets; StdROA is firm i’s standard deviation of ROA calculated using 5 years of data from years t– 4 to t. ROA is calculated as net income before extraordinary items deflated by beginning of year total assets; Times is the logarithm of (1 + times interest earned ratio) where times interest earned ratio is firm i’s operating income before depreciation and interest expense divided by interest expense at year t; StdRet is the standard deviation of firm i’s daily stock returns during year t; BM is the logarithm of (book value of equity / market value of equity) for firm i measured at the end of year t; Size is the logarithm of total assets for firm i at the end of year t; RND is firm i’s research and development expense deflated by beginning of year total assets; Lev is firm i’s long-term debt divided by total assets at the end of year t. The p values are in parentheses (two-tailed tests). In all columns, the dependent variable is the change in credit rating in year t+ 1. All standard errors are clustered by firm. EPS = earnings per share; ROA = return on assets.
Significance of coefficients is marked by asterisks, with *, **, and *** representing significance at the 10%, 5%, and 1% levels, respectively.
Overall, our findings suggest that abnormal repurchases that increase EPS are associated with a mitigated cost of debt decrease (from meeting earnings benchmarks). 22 We find some support for H1 using the combined real activities metrics and abnormal cash flows from operations as the proxies for real activities manipulation. We find marginal support for H1 when using abnormal production costs and no support when using abnormal discretionary expenditures as a proxy for real activities manipulation. Again, these results are not surprising considering that positive abnormal discretionary expenditures, at least in the short-term, are cash flow increasing. We show more consistent support for H1 for real activities manipulation measures that are cash flow decreasing. 23 As expected, the most consistent support of H1 comes from using abnormal repurchases as our proxy for real activities manipulation. 24
An interesting implication of these results relates to Gunny (2010), who finds that firms participate in real activities to meet earnings benchmarks, and managers’ underlying motivation for these actions is to use current-period benefits to allow for better future firm performance. Our results suggest that the debt market perceives these management actions as more opportunistic in nature.
Table 6 shows the results from the ordered logit regressions and chi-square tests to test H2. Columns 1, 2, and 3 show the regressions using the Profit, Incr, and Surp benchmarks, respectively. In all three columns, we show initial support for H2 when using abnormal stock repurchases as the proxy for real activities manipulation: The coefficients for the ACCRE interaction variable are more positive and significant than the EM_ACCR interaction variable coefficients. The results of the direct test of H2 are shown at the bottom of the table. Again, for all three columns, we have significant chi-square statistics verifying that the differences in the coefficients are significantly different from 0, which supports H2. We completed these tests using our other real activities manipulation proxies (untabulated) but did not find any evidence to support H2 using those proxies. Overall, we find support for H2 using our abnormal repurchases proxy for real activities manipulation. When firms repurchase shares to increase their EPS to meet an earnings benchmark, the magnitude of the associated cost of debt decrease attenuation is significantly more than if firms use accruals management for the same reason.
Ordered Logit Regressions and Chi-Square Tests of H2.
Note. Table 6 reports coefficients and p values from ordered logit regressions testing H2. In column 1, H2 is tested using the Profit benchmark. In column 2 H2 is tested using the Incr benchmark. In column 3 H2 is tested using the Surp benchmark. Profit equals to 1 if firm i’s basic EPS before extraordinary items is greater than or equal to 0 in year t, and 0 otherwise; EM_ACCR equals to 1 if included in the top quintile of residuals from Equation 7, and 0 otherwise; ACCRE equals to 1 if “As-if” EPS < actual EPS, and 0 otherwise; Incr equals to 1 if firm i’s EPS before extraordinary items in year t is greater than or equal to that of year t– 1, and 0 otherwise; Surp equals to 1 if firm i’s reported EPS in year t is greater than or equal to its most recent analyst forecast in year t, and 0 otherwise; EarningsControl is one of the following continuous earnings variables: EPS, ΔEPS, and UE_EPS; CFO is firm i’s operating cash flows at year t deflated by beginning of year total assets; StdROA is firm i’s standard deviation of ROA calculated using 5 years of data from years t– 4 to t. ROA is calculated as net income before extraordinary items deflated by beginning of year total assets; Times is the logarithm of (1 + times interest earned ratio) where times interest earned ratio is firm i’s operating income before depreciation and interest expense divided by interest expense at year t; StdRet is the standard deviation of firm i’s daily stock returns during year t; BM is the logarithm of (book value of equity / market value of equity) for firm i measured at the end of year t; Size is the logarithm of total assets for firm i at the end of year t; RND is firm i’s research and development expense deflated by beginning of year total assets; Lev is firm i’s long-term debt divided by total assets at the end of year t. The p values are in parentheses (two-tailed tests). In all columns, the dependent variable is the change in credit rating in year t+ 1. All standard errors are clustered by firm. EPS = earnings per share; ROA = return on assets.
Significance of coefficients is marked by asterisks, with *, **, and *** representing significance at the 10%, 5%, and 1% levels, respectively.
Using this real activities manipulation proxy, our H1 and H2 findings meet our expectations due to the wealth transfer hypothesis discussed earlier (Maxwell & Stephens, 2003). Also, our abnormal repurchases measure more directly identifies a documented management action (Hribar et al., 2006): The stock repurchases made by management directly increase EPS. Thus, we believe that this proxy is more directly specified as real activities manipulation than our other proxies.
Due to the financial crisis, credit markets in 2007 and 2008 were severe. Since our dependent variable is change in credit ratings in the previously mentioned analyses, including these years may significantly affect our results. Therefore, as a robustness check, we excluded observations from 2007 and 2008 and reran the regressions in Tables 3 to 6. Our results are robust to this 2-year exclusion of data.
Table 7 reports the economic significance of our results pertaining to a credit ratings upgrade or downgrade. We construct new dichotomous variables capturing whether the change in credit rating is an upgrade, downgrade, or no change at all. We then run logit models replacing the change in credit rating with these dichotomous variables and compute the marginal effects of our variables of interest holding all other variables at their mean values. Therefore, the coefficients show the percentage change in the predicted probability of a credit ratings change, holding all other variables in the regressions at their mean values similar to Jiang’s (2008)Table 6. A firm that reaches the Profit earnings benchmark has a 5.2% lesser probability of receiving a credit rating downgrade, while reaching that benchmark does not have a significant effect on receiving a credit rating upgrade. Our interaction variable coefficients show us that if a firm reaches the Profit benchmark using a stock repurchase transaction, this benefit is offset by a 7% decreased likelihood of receiving a credit rating upgrade. Using our first (second) aggregate measure of real activities manipulation, this benefit is offset by a 3% (2.4%) increased likelihood of receiving a credit rating downgrade.
Economic Significance—Changes in Predicted Probabilities of Ratings Changes When the Benchmark and Earnings Management Indicators Change From 0 to 1, Holding All Other Variables at Their Mean Values.
Note. Table 7 reports the economic significance of the coefficients in Tables 3, 4, and 5. Profit equals to 1 if firm i’s basic EPS before extraordinary items is greater than or equal to 0 in year t, and 0 otherwise; Incr equals to 1 if firm i’s EPS before extraordinary items in year t is greater than or equal to that of year t– 1, and 0 otherwise; Surp equals to 1 if firm i’s reported EPS in year t is greater than or equal to its most recent analyst forecast in year t, and 0 otherwise; ACCRE equals to 1 if “As-if” EPS < actual EPS, and 0 otherwise; RealEM_RM1 equals to 1 if included in the top quintile of RM1, and 0 otherwise; RealEM_RM2 equals to 1 if included in the top quintile of RM2, and 0 otherwise. Ratings changes are classified as either upgrades, downgrades, or no change at all, and logit models are estimated through which the marginal effects were calculated. Therefore, the tabulated coefficients can be interpreted as percentage changes in the predicted probability of a credit rating change, holding all other variables in the regressions at their mean values. EPS = earnings per share.
Significance of change in predicted probability is marked by asterisks, with *, **, and *** representing significance at the 10%, 5%, and 1% levels, respectively.
A firm that reaches the Incr earnings benchmark has a 3.5% (2.2%) lesser (greater) probability of receiving a credit rating downgrade (upgrade). This benefit is completely offset when a firm uses real activities manipulation to reach the benchmark: If a firm reaches the Incr benchmark using a stock repurchase transaction, there is an 8.7% decreased likelihood of receiving a credit rating upgrade. Using our first (second) aggregate measure, there is a 3.9% (4.0%) increased likelihood of receiving a credit rating downgrade.
We also find similar results with the Surp earnings benchmark: A firm that reaches the Surp benchmark has a 2.7% (1.4%) lesser (greater) probability of receiving a credit rating downgrade (upgrade). If a firm, however, reaches the Surp benchmark using a stock repurchase transaction, there is a 12.5% decreased likelihood of receiving a credit rating upgrade. Using our first (second) aggregate measure, there is a 3% (2.3%) increased likelihood of receiving a credit rating downgrade.
Taking the results of the last three paragraphs together, we find a complete attenuation of the cost of debt decrease achieved by meeting an earnings benchmark in several instances of real activities manipulation. Firms that meet the Incr and Surp earnings benchmarks, using abnormal stock repurchases, have a net significant lower probability of receiving a credit rating upgrade of 6.5% and 11.1%, respectively. Also, firms that meet the Incr benchmark have a net significant higher probability of receiving a credit rating downgrade of 1.9%, 0.4%, and 0.5% for the abnormal repurchases measure and the first and second aggregate measures, respectively. Finally, firms that meet the Surp benchmark have a net significant higher probability of receiving a credit rating downgrade of 0.3% for the first aggregate measure. Thus, as expected, we find our strongest results using the stock repurchase measure of real activities manipulation.
Conclusion
We investigate the cost of debt effects of meeting earnings benchmarks using real activities manipulation. Meeting earnings benchmarks is viewed favorably by investors. However, meeting these benchmarks through abnormal accruals or real activities manipulation will reduce the favorable impact. Jiang (2008) finds that when firms meet their earnings benchmarks, their cost of debt goes down but this decrease is smaller if the target is met using accruals earnings management. We expect to find that the favorable impact of meeting an earnings benchmark will be diminished even more when earnings benchmarks are met through real activities rather than accruals management. Not only is the current performance in meeting the benchmark suspect, future cash flows may also be reduced due to the activities manipulation.
In the specific case when earnings benchmarks are met through share repurchases, there is a triple effect on the cost of debt in that abnormal stock repurchases are expected to transfer wealth to shareholders from bondholders (Maxwell & Stephens, 2003).
Supporting our H1 that firms meeting earnings benchmarks through real activities manipulation should be viewed less favorably than firms that meet their benchmarks without manipulation, we find some evidence of smaller cost-of-debt reductions for firms managing earnings. Specifically, we find consistent evidence of this effect for real activities manipulation measures that are cash flow decreasing and weak evidence for those measures that are cash flow increasing. In other words, there is little reduction in the cost of debt by meeting earnings benchmarks through real activities manipulation that reduce cash flows. The strongest effect, as hypothesized, is found when share repurchases are used to meet earnings benchmarks. In this case, the entire benefits (in terms of cost of debt) of meeting earnings benchmarks are eliminated. These results may suggest a debt market perception of these managerial actions as opportunistic in nature (Gunny, 2010).
We also find some support for our H2 that meeting earnings benchmarks through real activities manipulation that adversely affect cash flows is viewed less favorably by the market. Measuring real activities manipulation using abnormal stock repurchases, our results show that there is an adverse debt market reaction (in terms of credit rating downgrades) for firms that use share repurchases to meet their earnings benchmarks. This contrasts with the overall reduction in the cost of debt for firms that use accruals earnings management to meet those benchmarks.
Finally, we examine the economic magnitude of our regression results. For several of our real activities manipulation proxies, we find that the complete attenuation of the cost of debt decrease is achieved by meeting an earnings benchmark when real activities manipulation is used. These findings are likely of interest to managers: They more prevalently use real activities manipulation in the management of their companies, compared with their use of accruals earnings management (Graham et al., 2005). Thus, it would be of great interest to them to know how their actions affect their firms’ cost of debt. Identification of this debt market effect will also be of interest to bondholders and regulators.
Footnotes
Appendix A
Appendix B
Two-Stage Regression Used to Calculate Abnormal Repurchases.
| Predict | Second-stage regression | First-stage regression | |
|---|---|---|---|
| RepAmtt – 1 | (+) | 0.480 | |
| (<.001)*** | |||
| RepAmtt – 2 | (+) | 0.262 | |
| (<.001)*** | |||
| IfRepurchaset – 1 | (+) | 1.331 | |
| (<.001)*** | |||
| IfRepurchaset – 2 | (+) | 0.699 | |
| (<.001)*** | |||
| Casht – 1 | (+) | 13.790 | 0.097 |
| (<.001)*** | (<.001)*** | ||
| CAPEXt – 1 | (–) | −4.642 | −0.167 |
| (0.109) | (0.004)*** | ||
| DividendYieldt – 1 | (–/+) | −325.204 | 3.824 |
| (<.001)*** | (<.001)*** | ||
| Debtt – 1 | (–) | −22.705 | −0.642 |
| (<.001)*** | (<.001)*** | ||
| Sizet – 1 | (+) | 9.480 | 0.099 |
| (<.001)*** | (<.001)*** | ||
| Lambda | 9.946 | ||
| (<.001)*** | |||
| Pseudo R2 | .5755 | .3481 | |
| N | 180,008 | 180,008 | |
| Year fixed effects | Yes | Yes | |
| Industry fixed effects | Yes | Yes | |
| Quarter fixed effects | Yes | Yes |
Note. The above table lists the coefficients and p values (in parentheses) of a two-stage Heckman model used to produce unconditional expectations of stock repurchases for every firm-year in our sample following the determinants documented in Hribar, Jenkins, and Johnson’s (2006) study. Column 2 lists the estimates for the first stage where the dependent variable IfRepurchase is equal to 1 in firm-quarters in which a stock repurchase was made, and 0 otherwise. Column 1 lists the estimates for the second stage where all the variables are identical to the first stage except the binary variable IfRepurchase is replaced by a continuous measure of the amount of stock repurchased in a given firm-quarter RepAmt. Also, the second-stage regression uses two variables computed from the first stage to incorporate information about the probability of a repurchase. The first stage predicted that values are transformed into estimates of λ and Λ that represent the standard normal probability density function and cumulative density function, respectively. The variable λ is included as a separate control variable in the second stage, and Λ is multiplied by each of the independent variables in the second stage. Cash is the ratio of cash and cash equivalents to total assets; CAPEX is the ratio of capital expenditures over the past year to total assets; DividendYield is the dividends per share divided by beginning of quarter stock price; Debt is the ratio of current and long-term debt to total assets; Size is the logarithm of total assets for firm i at the end of year t.
Significance of coefficients is marked by asterisks, with *, **, and *** representing significance at the 10%, 5%, and 1% levels, respectively.
Appendix C
Sample Selection.
| Step | Description | Effect on sample | Observations |
|---|---|---|---|
| S&P Financial Services | Number of firm-years that have data for senior credit ratings in COMPUSTAT. | 43,626 | |
| CRSP | Number of firm-years that have return data for control variables. | −13,718 | 29,908 |
| I/B/E/S | Number of firm-years that have analyst EPS forecasts. | −4,282 | 25,626 |
| Industry | Delete observations in public utilities and financial services (two-digit SIC codes 40-49 and 60-67) or missing industry code. | −10,159 | 15,467 |
| Exchange listing | Delete observations of firms that are not listed on the NYSE, AMEX or NASDAQ. | −0 | 15,467 |
| COMPUSTAT | Number of firm-years that have all available data to construct control variables in COMPUSTAT. | −6,583 | 8,884 |
Note. This table presents a summary of sample selection procedures with detailed descriptions of each step and the effect on the sample for this study. S&P = Standard & Poor’s Financial Services; CRSP = Center for Research in Security Prices; I/B/E/S = Institutional Brokers’ Estimate System; EPS = earnings per share; SIC = Standard Industrial Classification; NYSE = New York Stock Exchange; AMEX = American Stock Exchange; NASDAQ = National Association of Securities Dealers Automated Quotations.
Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author(s) received no financial support for the research, authorship, and/or publication of this article.