Master Thesis
Short-term Earnings Management and Long-term Value Creation:
Looking Beyond Returns
Ralitza K. Todorova (11400870)
August 2017
Thesis Supervisor: David Veenman
Abstract: This study expands the scope of the existing literature’s findings about how managing earnings to achieve short-term financial objectives impacts firm performance. I focus on the impact on long-term value creation and define it as the spread between a firm’s Return on Net Operating Assets (RNOA) and its Weighted Average Cost of Capital (WACC). I find that, in the long-term, firms suspected of myopic earnings management have significantly lower accounting returns and a lower cost of capital in comparison to the firms that are least likely to have engaged in myopic earnings management. The effect on RNOA is stronger than the effect on WACC, so the impact on the RNOA-WACC spread is negative. Ultimately, myopic earnings management appears to harm long-term value creation, but the negative effect is moderated by the beneficial impact on WACC. Further, I find that the severity of the negative impact on long-term value creation depends on the frequency with which the firm has engaged in myopic earnings management in past years. More frequent myopic earnings management in past years is associated with a lower RNOA-WACC spread today.
Contents
1. Introduction ... 2 2. Literature overview ... 5 3. Hypotheses... 8 4. Study design ... 11 5. Results ... 23 6. Conclusion ... 32 References ... 341. Introduction
In the agency-based model of corporate leadership, public company managers are viewed as agents representing the interests of shareholders, and they have a responsibility to maximize shareholder value, as articulated by Jensen and Meckling (1976). Unfortunately, the principal-agent relationship is well-understood to create an opportunity for certain conflicts of interest between shareholders and managers. One such potential conflict is that executives can be tempted to behave myopically, making particular decisions in order to achieve short-term objectives without regard for, or sometimes knowingly in spite of, the implied costs of those decisions for longer-term firm value. In recent years, studies have documented an increase in the pressure and temptation facing managers to deliver short-term financial results, such as meeting analyst earnings forecasts1. The potential threats that
this type of behavior poses to long-term value has been receiving growing attention in the popular press as well as within the financial industry2. It has also been the subject of prominent academic
studies over the past three decades3.
The existing literature has identified the incentives that often tempt executives to engage in myopic earnings management [e.g. Stein (1989); Degeorge et al. (1999), and Graham et al. (2005)]. The literature has also identified the mechanisms by which myopic earnings management takes place as well as its identifiable symptoms [e.g. Roychowdhury (2006), Edmans et al. (2014), and Ladika and Sautner (2016)]. The literature has reached mostly consistent conclusions about the overall direction of the association between myopic earnings management and firm performance, measured as equity or accounting returns [e.g. Bhojraj et al. (2009) and Brochet et al. (2015)], but some notable exceptions have been found, such as in a study by Rahmandad et al. (2016). Adding a layer of complexity to the analysis, there seems to be little consensus as to what is the interaction between earnings smoothing (closely related to managing earnings to meet short-term benchmarks, such as analyst expectations) and a firm’s cost of capital, as evidenced by McInnis (2010) and Strobl (2013).
With this thesis, I would like to continue the inquiry into whether myopic earnings management has a negative effect on long-term firm value creation. My study aims to contribute to the academic community’s current understanding of the association between myopic earnings management and value creation by extending the scope of existing studies to examine a more wholistic and academically robust measure of value creation. The study after which I model my methodology measures returns
1 As an example, refer to the findings from a 2013 survey of global business leaders, Looking toward the long term, commissioned by McKinsey & Company and the Canada Pension Plan Investment Board. Review of findings
available at http://www.shareholderforum.com/access/Library/20131226_McKinsey.pdf 2 https://www.sec.gov/news/speech/spch370.htm
using the Return on Assets (ROA), which is defined as income before extraordinary items divided by beginning total assets. From an academic perspective, ROA is not an internally consistent metric because it does not appropriately match the numerator and denominator; income before extraordinary items is net of interest expense, but total assets are supported by debt.
Instead of ROA, my study uses the Return on Net Operating Assets (RNOA), which is a more internally consistent measure of returns. In addition to refining the returns metric, I extend previous studies’ reach beyond examining just returns. While there is some agreement in the literature that managing for the short-term can harm accounting returns, some findings suggest that smoother and more predictable earnings, which can be achieved by managing earnings to meet analyst expectations, may be associated with a lower cost of capital. In order to explore the potentially complex interaction between these two contributing factors to value creation, I define long-term value creation as the spread between a firm’s RNOA and its Weighted Average Cost of Capital (WACC).
I also examine whether long-term value creation reacts differently in firms who manage earnings for the short-term only occasionally in contrast to firms that practice this more frequently. For instance, if myopic earnings management is a rare practice within a firm, when it does occur it may not deplete investment enough to endanger long-term performance. But if it is practiced frequently, then I would expect that value suffers in the long-term as the impacts of the myopic decisions accumulate year after year. I create a variable to measure the frequency of myopic earnings management by dividing the number of years in which a firm is suspected of managing earnings to achieve short-term goals by the total number of years in the sample that have available data for that firm.4 I expect that this
earnings management frequency variable has a negative association with firm operating returns. My study’s findings support my expectations that myopic earnings management impedes long-term value creation, as measured by a firm’s RNOA-WACC spread. The long-term future RNOA-WACC spread of firms suspected of managing earnings myopically is significantly lower than the spread of those firms that are least likely to have done so. Myopic earnings management affects the two components of the spread in the same direction: it leads to lower accounting returns and also to a lower cost of capital. Even though the ultimate impact on the RNOA-WACC spread is negative, the effect is moderated by the beneficial impact on WACC. I also find that firms that have managed earnings myopically with greater frequency in past years tend to have a lower RNOA-WACC spread
4 In addition to using the design discussed in this proposal, frequency of earnings management could be examined by identifying firms with high versus low net operating assets. If accruals are managed regularly, then this accumulates on the balance sheet and the activity would be identifiable [Ho, Liu and Ouyang (2012)]. My study does not employ this method, but it can be applied in future studies or robustness checks.
today. This implies that the severity of the negative impact on long-term value creation depends on the frequency with which the firm’s earnings have been managed myopically in the past.
On the one hand, these findings might suggest that it is not always completely unreasonable to manage earnings to meet short-term objectives, such as analyst earnings estimates. The long-term benefit to a firm’s WACC may be sufficiently important for some firms so as to make the sacrifice in accounting returns worthwhile. I suspect that few firms sufficiently benefit from the reduced WACC to compensate for the lower returns, but future studies should attempt to identify such cases and understand what is unique about them. On the other hand, the economically significant harm to long-term returns should give capital providers pause. Perhaps when investors reward smoother earnings (achieved through myopic earnings management) with a lower required return, they do not take into account the significant reduction to long-term value creation that is likely to follow.
My findings clearly illustrate that managing for the short-term benefit of a firm’s stock price and its shareholders’ returns is not to be confused with managing for the long-term success of the firm itself. Many managers’ behavior is still governed by the widely-accepted top priority to generate returns to shareholders5, such as the immediate gains in equity returns that can be achieved with myopic
earnings management. This guiding principle should perhaps give way to a stronger focus on protecting the firm’s long-term value-creation potential, which has important implications for many stakeholders beyond just the shareholders. Regulators and boards may find that limiting the shareholder-friendliness of current governance models may ultimately prove beneficial to a firm’s long-term value creation potential.
5 The 1997 issue of the Statement on Corporate Governance from The Business Roundtable of CEOs stated that “the paramount duty of management and of boards of directors is to the corporation’s stockholders” and that “the principal objective of a business enterprise is to generate economic returns to its owners.” Since then, the most recent updates to the Business Roundtable’s 2016 Principles of Corporate Governance have taken a different tone, advocating for “strategies that are intended to build sustainable long-term value” and putting
2. Literature overview
2.1. Motivation for myopic earnings management
In one of the earliest studies of myopic management practices, Stein (1989) characterizes myopic management behavior as the Nash equilibrium outcome of a non-cooperative game. The outcome is that managers inflate current results to signal strong future performance to the market, and the market in turn anticipates and prices in this myopic behavior. Degeorge et al. (1999) document managers’ use of earnings management techniques in order to avoid reporting losses and missing analyst earnings expectations, with the understanding that reporting losses or earnings below the expected level would have real negative consequences for the firm.
In addition to having negative consequences for the firm, studies have found that falling short of analyst and board earnings expectations can have negative consequences for the firm managers themselves. Farrell and Whidbee (2003) show evidence that a firm’s board of directors evaluates the CEO’s performance based on how a period’s actual earnings compare to earnings projected by analysts. Their findings suggest that CEO turnover is higher when accrual earnings fall short of the expected value, highlighting the personal pressures motivating managers to adhere to analyst forecasts.
In a set of extensive surveys and interviews with executives, Graham et al. (2005) confirm prior theories about myopic earnings management in a more direct way. Among their findings are the following: managers focus more on accounting earnings than on cash flows for reporting purposes; meeting earnings benchmarks and maintaining smooth earnings are a high priority for managers; and the majority of sampled managers would smooth earnings by taking real actions that have negative consequences for long-term economic value, but that only a minority would smooth earnings using within-GAAP6 accounting adjustments.
2.2. Causes and symptoms of myopic earnings management
Roychowdhury (2006) finds evidence of manipulation in real operational activities in order to avoid reporting annual losses, as well as to meet annual analyst forecasts (but the evidence for this second finding is less robust). With further evidence of reduction in real activities, Edmans et al. (2014) find that CEO short-term concerns for the firm’s stock price is associated with lower growth in Research and Development (R&D) and net capital expenditure. They define short-term concerns for the stock price as the presence of equity and options vesting in a given quarter.
6 Within-GAAP accounting adjustments refer to accounting choices that are within the bounds of the Generally Accepted Accounting Principles and are not fraudulent.
In a natural experiment, Ladika and Sautner (2016) use an exogenous shock to show that a reduction in executives’ incentive horizons had a negative effect on corporate investment. The exogenous shock was a change in reporting requirements for options, which reduced executive incentive horizons by accelerating the vesting of executives’ existing stock options.
2.3. Consequences for value of myopic earnings management
To examine the consequences of myopic earnings management, Bhojraj et al. (2009) compare the performance of firms that just beat earnings forecasts with low quality earnings to firms that just miss earnings forecasts with high quality earnings. Earnings quality is examined through changes in discretionary spending and accruals. The combination of just beating the earnings forecast and an indicator of poor earnings quality is used as a signal of myopic earnings management; this is the methodology that my study will follow.
The results of Bhojraj et al. (2009) show that despite an increase in short-term stock price returns, firms with a low-quality beat perform worse in the longer-term than firms with a high-quality miss, suggesting long-term value erosion results from managing earnings for short-term gains. This study measures firm performance using both share price and Return on Assets (ROA), which is defined as income before extraordinary items divided by beginning total assets.
A similar conclusion is reached by Brochet et al. (2015) regarding the directional impact of short-termism on returns. The study establishes that the time horizon stressed by management in conference calls and presentations can serve as a good proxy for myopic management. They find that communication focused on a relatively short time horizon is associated with lower ROE in the next one and two years, confirming the negative association expected between short-termism and firm performance.
In one of the most recent studies of short-termism and firm performance, the results appear to be more nuanced. Rahmandad et al. (2016) conclude that the direction of the long-term consequences of earnings management depends on how intensely the firm engages in the activity. The variability in outcomes is due to the impact of earnings management on the accumulation or depletion of a firm’s “capability” level. As long as a certain threshold of “capability” depletion is not breached, they find that firms can avoid suffering long-term negative consequences from earnings management.
2.4. Earnings smoothness and cost of capital
An additional motivator to engage in earnings management, beyond those discussed in earlier sections of this paper, may be related to a firm’s cost of capital. Burgstahler and Dichev (1997) propose that earnings are managed to avoid relative decreases or absolute losses; they argue that managers
expect that avoiding these outcomes can reduce the firm’s costs of transacting with stakeholders. If I extend their reasoning to consider stakeholders such as providers of debt and equity capital, then perhaps using earnings management to achieve certain targets could be a useful tool in reducing a firm’s cost of capital. Since my study’s metric for long-term value creation is partly driven by the firm’s cost of capital, the findings in this area are of particular interest to me.
While Burgstahler and Dichev (1997) do not explicitly examine the impact on cost of capital, other studies have suggested that smoother, more predictable earnings may help lower a firm’s cost of capital. For example, Graham et al. (2005) discover through interviews and surveys that CFOs tend to associate missing earnings expectations with high earnings volatility, and that both lead to higher investor uncertainty. The CFOs interviewed believe that investors require a lower risk premium to invest in companies with predictable and smooth earnings, which explains part of the pressure on managers to avoid negative surprises.
However, a study by McInnis (2010) uses asset-pricing tests to find that no relation can be identified between earnings smoothness and average returns to equity holders (which can be used as a proxy for the cost of equity capital). He argues that any association between the implied cost of equity and earnings smoothness is a result of optimistic bias in analyst earnings forecasts for firms with higher earnings volatility. The results of Strobl (2013) are also nuanced and leave some open questions about the link between earnings management and the cost of capital. He finds that earnings manipulation affects a firm’s cost of capital only if the manipulation is significant enough to affect cash flows. The existing literature on myopic earnings management is rich, but it leaves unaddressed the question of how managing earnings for the short-term impacts long-term value creation when taking into account the interaction between operating returns and cost of capital. In this study, I propose to connect these two elements in order to understand their relative contributions to value creation.
3. Hypotheses
I expect that myopic earnings management has, on net, an adverse effect on long-term value creation, as defined by the spread between a firm’s Return on Net Operating Assets (RNOA) and its Weighted Average Cost of Capital (WACC). This net negative effect results from the interaction between the impact of myopic earnings management on returns and on the cost of capital. The hypotheses described below elaborate on this prediction and are all stated in the alternative form.
3.1. Hypothesis 1
Firms suspected of myopic earnings management (those who beat earnings expectations with low quality earnings) should, in the long term, have lower RNOA than (a) all other firms, and (b) firms with high quality earnings who miss earnings expectations. This outcome follows intuitively because, as it is defined, myopic earnings management involves accounting and operating decisions that are made only in order to meet or beat earnings expectations, and that would not be made in the normal course of business if earnings pressures were not present.
The existing literature supports my expectation that the indicators I use as a proxy for low quality earnings may compromise future firm performance. Xie (2001) finds that discretionary accruals have lower persistence than other components of earnings, and Dechow and Dichev (2002) find that extreme levels of accruals are associated with less persistent earnings. Sloan (1996) also finds that when the composition of a firm’s earnings is relatively high in accruals, then the firm’s ROA declines (reverts to the mean) more quickly than when the composition of earnings is relatively high in cash flows. Based on these findings, I expect high discretionary accruals to signal less sustainable earnings. Similarly, outsized reductions in discretionary investment today, if undertaken only to inflate the current period’s earnings, are likely to be reversed and compensated for in future periods in order to maintain the required level of investment to support the firm’s operations. The subsequent compensation in discretionary investment in future periods would have a negative impact on future earnings, resulting in lower returns. Low discretionary investment can also hurt future returns if the investments the firm forgoes have a positive net present value (NPV). By cutting value-adding investments, the firm’s future performance is likely to be lower than it would have been if the positive-NPV investments had been made.
Although RNOA is a slightly different measure of returns than other studies have used, this hypothesis is in line with the findings of other studies that use similar return measures to assess the effects of earnings management. Bhojraj et al. (2009) and Brochet et al. (2015) each find that managing for short-term objectives is associated with underperformance in stock price returns, ROA, and ROE in
the following one-to-three years. My study examines the impact farther into the future and defines the long term as the five years following the year in which earnings management is suspected. 3.2. Hypothesis 2
On the other hand, firms who beat earnings expectations with low quality earnings could, in future years, have lower WACC than (a) all other firms, and (b) firms with high quality earnings who miss earnings expectations. Intuitively, I predict that by avoiding negative earnings surprises, firms can improve investor confidence in their future performance; I expect that improving investor confidence could lower the investors’ required return, and therefore result in a lower WACC. This expectation is in line with the findings of Graham et al. (2005); they collect first-hand accounts from CFOs who believe that firms with smooth and predictable earnings are assigned a lower risk premium by investors, potentially benefiting those firms’ cost of capital.
3.3. Hypothesis 3
Firms who beat earnings expectations with low quality earnings should, in the long term, have a lower spread between their RNOA and their WACC than (a) all other firms, and (b) firms with high quality earnings who miss earnings expectations. This hypothesis extends my study beyond the scope of the existing literature by considering the effects on value creation measured not only by returns but also by the cost of a firm’s capital.
If my predictions for Hypothesis 1 and 2 hold, then the effects of myopic earnings management on value creation would be more nuanced than the effect on returns alone. If myopic earnings management is associated with a lower cost of capital, then a firm with low quality earnings may not suffer negative consequences to its long-term value creation in the special case where the favorable effect on WACC outweighs the unfavorable effect on RNOA. However, this special case seems unlikely, and I propose a more modest prediction. While I expect that earnings management is associated with a reduced capacity for value creation, as measured by the RNOA-WACC spread, I predict that this value creation metric suffers less in absolute terms than RNOA alone7, because the favorable impact on
WACC tempers the negative impact on RNOA. 3.4. Hypothesis 4
If a negative association is found between a single instance of myopic earnings management and long-term value creation, then I would also expect that firms who are subject to more myopic earnings management have more severe adverse outcomes. I predict that the magnitude of the impact of
7 All tests are performed on identical samples to allow for direct comparison between the individual impact on RNOA and WACC and the combined impact on the RNOA-WACC spread.
myopic earnings management on long-term value grows with the frequency with which myopic earnings management is exercised. In other words, the negative impact of myopic earnings management on long-term value is stronger in firms who manage earnings for short-term benefits with greater frequency.
I assume that firms may be able to rebound from the occasional year of myopic earnings management, marked by high discretionary accruals or low discretionary investment, particularly if the accounting or cash activities are relatively small and easy to reverse the following year. However, I expect that the negative consequences of repeatedly making such decisions over a prolonged period would accumulate, become difficult to reverse, and ultimately hinder the firm’s ability to create value. Conceptually, this hypothesis is similar to one of the findings in Rahmandad et al. (2016). They find that the outcomes of earnings management can vary, and the direction of the outcome may depend on the intensity of the earnings management activity.
4. Study design
4.1. Myopic earnings management, independent variable
I define the concept of myopic earnings management as operating and accounting activity that is tailored specifically to meet short-term earnings expectations, such as analyst estimates or management guidance. This paper will also use the terms earnings management or managing earnings for the short-term to refer to the same concept. The literature referenced in the previous section suggests that operating and accounting activity in a given period would be different if short-term earnings pressure was not present (such as if the firm could meet earnings estimates without any earnings management). A frequent result of this short-term pressure is that management forgoes long-term positive-NPV investments with the purpose of meeting the short-term targets; this is demonstrated in the survey results of Graham et al. (2005).
4.1.1. Identifying myopic earnings management
Empirically, I identify suspected cases of myopic earnings management using the methodology from Bhojraj et al. (2009). In their study, myopic earnings management is suspected when a firm ends a fiscal year with a small positive earnings surprise, while at the same time displaying symptoms indicative of low earnings quality for the period. Low earnings quality is suspected when a firm has relatively high discretionary accruals and relatively low discretionary investment. The logic is that if a firm marginally beats analyst earnings expectations in a year when its earnings quality is low, the firm would have missed earnings expectations if it avoided the accounting and operating decisions that lower its earnings quality. Assuming that firms prefer to have high quality earnings, all else equal, I assume that the accounting and operating activities that reduce earnings quality are undertaken in order to avoid missing earnings expectations.
In contrast to the focus group of firms suspected of myopic earnings management, I also identify observations where myopic earnings management is least likely to have taken place. The characteristics used to define this group are the opposite of those used to identify suspected cases of earnings management. I assume that the firms least likely to have engaged in myopic earnings management have ended a fiscal year with a small negative earnings surprise, while displaying symptoms indicative of high earnings quality for the same period. A discussion of the definitions and calculations of the relevant metrics for earnings surprises and earnings quality follows in the next sections.
4.1.2. Earnings expectations miss / beat
To identify how actual earnings per share compare to analyst estimates, I compare the actual reported earnings per share for a given fiscal year to the analyst consensus estimate for the same period. Data
for the actual and estimated earnings per share is collected from the I/B/E/S database. The consensus estimate used is the median of all analyst estimates as calculated by I/B/E/S. Following the logic in Bhojraj et al. (2009), I use the median estimate as of two months prior to the end of the fiscal year, or in other words, the estimate that was issued in the eleventh month of the fiscal year.
To maximize the probability of identifying earnings management activity, I focus on cases where the difference between the actual and estimated earnings per share is one penny when rounded to the nearest penny. Dechow et al. (2010) surveys a number of studies, which find a consistent association between meeting or beating analyst estimates and other established determinants of earnings management. A possible interpretation of these findings is that meeting or beating analyst estimates may indicate a case of earnings management. I limit my focus to a positive surprise of just one penny, because these are the cases where it is most likely the firm would have fallen short of earnings expectations had it not engaged in earnings management.
It is important to note that this approach is not a perfect indicator of earnings management. For one, some firms that beat earnings expectations by a penny could have done so without any earnings management activity. In other cases, firms where earnings management did take place could beat earnings expectations by more than just a penny, such as firms with high earnings per share values, where a larger surprise in terms of dollars per share would still be a small fraction of earnings per share. Additionally, findings by Bissessur and Veenman (2016) have called into question the usefulness of using a small earnings surprise to identify earnings management activity.
4.1.3. Earnings quality
The concept of earnings quality is not strictly defined by the academic literature. Dechow et al. (2010) survey the most commonly used empirical proxies for earnings quality in the literature and argue that there is no single best measure of earnings quality. Some of the proxies most frequently used include the persistence of earnings over time, the magnitude of accruals, the residuals from accrual models, the smoothness of earnings relative to cash flows, timely recognition of incurred losses, and performance relative to a benchmark, among others. My study uses two of these proxies, the magnitude of accruals and performance relative to a benchmark, in addition to a third proxy, discretionary investment, to assess the quality of earnings and identify suspected cases of earnings management.
Following the methodology in Bhojraj et al. (2009), I determine earnings quality based on a combination of accounting and operating decisions. I construct an earnings quality score for each observation in my sample using the results of the discretionary accruals and discretionary investment
calculations, which are discussed in the next two sections. Low levels of scaled discretionary accruals and high annual changes in discretionary expenditures are treated as indicative of high earnings quality. The justification for each of these indicators is discussed in detail in the following sections of this paper.
The numeric earnings quality score of each firm-year observation is calculated by awarding 1 point for each of the following:
• Low discretionary accruals • High annual R&D change
• High annual advertising expense change
A score of 2 or 3 is considered high, and a score of 0 is low. Scores of 1 are not used to determine earnings quality because they may result from missing data and are not sufficiently informative. 4.1.4. Discretionary accruals
Managers have discretion over certain accounting decisions that can affect accruals and reported net income. This is why I use discretionary accruals to identify potential instances of earnings management. Discretionary accruals are measured as the difference between total accruals and an estimated value for non-discretionary accruals. Total accruals are calculated as income before extraordinary items minus cash flow from operations. Non-discretionary accruals are estimated for each industry-year group using the modified Jones model as in Dechow et al. (1995). The model used for the estimation is as follows:
ACC𝑖,𝑡 / 𝐴𝑖,𝑡−1 = β0 + β1(1 / 𝐴𝑖,𝑡−1) + β2(Δ𝑅𝐸𝑉i,t – ΔARi,t) / Ai,t-1 + β3(𝑃𝑃𝐸𝑖,𝑡 / Ai,t-1)+ 𝜀𝑖,𝑡
Where:
ACC = Total Accruals,
A = Total Assets,
REV = Revenue,
AR = Accounts Receivable,
PPE = Gross Property Plant & Equipment,
i = the 2-digit SIC industry code, and
t = the fiscal year
All variables are scaled by total assets in the prior fiscal year. The scaled variables are winsorized at the 1st and 99th percentiles in order to limit the occurrence of extreme observations and outliers, with
the exception of total assets and gross property plant & equipment, which are only winsorized at the 99th percentile because they have a natural limit at zero. Each industry-year group must have at least
The resulting error term from the estimation represents the discretionary accruals (scaled by total assets) for each firm-year observation, based on 1) each observation’s actual scaled total accruals and 2) the scaled non-discretionary accruals estimate for the industry-year group to which the observation belongs. Once the scaled discretionary accruals have been estimated, I determine for each observation if its scaled discretionary accruals are higher or lower than the median of all firms in the relevant fiscal year. An observation is considered to have low discretionary accruals if its scaled discretionary accruals are lower than the year’s median for all firms. An observation is considered to have high discretionary accruals if its scaled discretionary accruals are equal to or higher than the year’s median for all firms.
High discretionary accruals are interpreted as a potential signal of low earnings quality. Conversely, low discretionary accruals are considered to signal high earnings quality. I make this assumption based on two related findings in the literature. First, studies have shown that earnings composed of relatively higher levels of accruals are less informative about future performance, and therefore are of lower quality. For example, Sloan (1996) finds that the accrual component of earnings is a less persistent contributor to earnings performance than the cash flow component of earnings. Second, Dechow et al. (2010) find that firms with high accruals also tend to have high discretionary accruals.
4.1.5. Discretionary investment
In addition to accounting decisions that affect accruals, managers also have discretion over certain operating investment expenses, which affect reported net income. I use annual reported R&D and advertising spending data as a proxy for discretionary investment. Reporting of this data is not required but it is widely available, so I limit my sample to firms that report at least one of the two expense items. For each firm-year observation with data available for at least one expense item, I calculate the annual change in the level of discretionary expense and scale it by the previous year’s total assets, as shown:
ΔR&Di,t = (R&Di,t – R&Di,t-1) / Ai,t-1
ΔADVi,t = (ADVi,t – ADVi,t-1) / Ai,t-1
Next, I compare each observation’s annual scaled change in discretionary expense to the median scaled change for all observations in the relevant year. For each observation, this comparison is done individually for both R&D and advertising expense. Observations whose scaled change in discretionary expense is higher than the year’s median are labeled as having a high change in discretionary expense. A high change in discretionary expense is interpreted as a sign of high earnings quality. Conversely, a relatively low change in discretionary expense is considered a sign of low earnings quality.
4.1.6. Frequency of myopic earnings management
With Hypothesis 4, I would like to test if the frequency with which a firm engages in myopic earnings management impacts the firm’s ability to create value over time. In other words, I would like to determine if occasional myopic earnings management affects value creation differently than frequent myopic earnings management. To examine this, I create an indicator of the frequency of myopic earnings management in years prior to the current year. In order to test Hypothesis 4, this myopic earnings management frequency metric is used as an independent variable in a regression with returns as the dependent variable.
For each firm-year observation, I create a variable that measures the frequency with which myopic earnings management activity is detected during the previous five years in the sample. The variable takes on a value equal to the percentage of years that the firm has beat expectations with low quality earnings. For example, for the observation of Firm A in year 2005, the myopic frequency variable equals 0.2 if Firm A has just beat earnings expectations with low quality earnings in only one year during the period 2001-2005, inclusive.
If data are available for fewer than five of the previous years, then the frequency calculation is adjusted to take into account the actual number of prior years for which data are available. For example, for the observation of Firm B in year 2005, if Firm B data is only available beginning in 2002, then the denominator for the myopic frequency calculation would be four instead of five. If Firm B just beat earnings expectations with low quality earnings in only one year over the period 2002-2005, then the myopic earnings management frequency variable for the year 2005 observation would equal 0.25.
4.2. Value creation, dependent variable
I define the concept of value creation as employing capital in such a way as to generate returns in excess of its costs8. A firm’s ability to generate excess returns is assumed to be an indicator of its future
value creation potential. Consistent with the findings of Bhojraj et al. (2009) and others, I expect excess returns to respond differently to myopic earnings management in the short- vs. long-term. In the short-term, myopic earnings management may result in higher RNOA (due to an inflated numerator) and lower WACC (if the market rewards firms for meeting earnings expectations with lower required returns). In the long-term, I expect that the accrual and real investment decisions made in order to
8 The academic literature refers to this and similar concepts as “economic profit”, “abnormal earnings”, “economic value added”, and “residual income”. Among the first to articulate this concept was Alfred Marshall in The Principles of Economics (1890).
meet or beat expectations will undermine the firm’s ability to undertake positive-NPV investments and create value.
Excess returns may also respond differently to occasional versus frequent myopic earnings management, as alluded to in Rahmandad et al. (2016). The accrual and real investment decisions made in an isolated period of myopic earnings management could be compensated for in future years, such that long-term value is not impacted if the firm has previously built a sufficient “cushion” of investment. Alternatively, the effects of frequent myopic earnings management are expected to accumulate over time and ultimately to destroy value, where the value destruction severity increases with the frequency of historical myopic earnings management.
Empirically, value creation is defined as the difference between the firm’s RNOA and its WACC (the “RNOA-WACC spread”). This metric captures firm excess returns and allows us to examine returns and cost of capital using comparable measurements. In the enterprise version of the residual income (also called abnormal earnings) valuation model, RNOA is compared to WACC in order to identify excess returns and estimate enterprise value. Research by Feltham and Ohlson (1995) shows that abnormal earnings depend on operating activities and that analyzing firm value requires forecasts of RNOA. The metrics that I use to calculate RNOA are defined as follows:
Table 1: RNOA variable definitions
Variable Definition
Return on Net Operating Assets = Net Operating Profit After Tax / Average Net Operating Assets Net Operating Profit After Tax9
= Operating Income After Depreciation & Amortization and Before Interest10 * (1 – Effective Tax Rate) CRSP Compustat field names = oiadp * (1 – Effective Tax Rate)
Effective Tax Rate = Income Taxes Total / Pre-tax Income
CRSP Compustat field names = txt / pi
Net Operating Assets
= Operating Assets – Operating Liabilities = (Total Assets – Cash & Equivalents – Investment & Advances Other) – (Total Assets – Long-term Debt – Debt in Current Liabilities – Common/Ordinary Equity – Preferred Stock – Minority Interest)11
CRSP Compustat field names = (at – che – ivao) – (at – dltt – dlc – ceq – pstk – mib)12
9 Effective tax rate is assumed to equal zero for observations with total income taxes of zero and a non-null pre-tax income. The calculated effective pre-tax rate is winsorized at the 1st and 99th percentiles to limit the occurrence of outliers and extreme values.
10 See Fairfield, P. M., S. Ramnath, and T. L. Yohn (2009) and Soliman, M. (2008) 11 See Soliman, M. (2008)
The data for the RNOA calculations is collected from the CRSP Compustat Merged database. In order to avoid extreme RNOA values resulting from a negative or small denominator, RNOA is not calculated for observations where the average calculated value for Net Operating Assets is negative or below $5 million. I also winsorize the resulting RNOA values at the 1st and 99th percentiles to further eliminate
outliers and extreme values.
I collect the WACC data from Bloomberg, starting with a data sample of monthly WACC estimates for all of the firms for which I have the required data from the CRSP Compustat Merged database and from the I/B/E/S database. Bloomberg’s WACC estimates are updated on a quarterly basis, so downloading the data on a monthly basis results in identical entries for the three months of each quarter, but makes it easier to merge the WACC data with my remaining data. Each observation from CRSP Compustat is assigned the WACC value for the same month and year as the month and year of the fiscal year-end date of the CRSP Compustat observation.
This WACC data is available from May 2000, which limits my sample to observations with a fiscal year end date no earlier than May 2000. This limitation is beneficial; it forces my study to focus on a more recent sample, improving confidence in the relevance of the findings for managers who face earnings management decisions today. Bloomberg calculates the weighted average cost of capital as follows:
𝑅𝑊𝐴𝐶𝐶 = 𝐸 𝐷 + 𝐸𝑘𝐸+ 𝐷 𝐷 + 𝐸𝑘𝐷(1 − 𝑡𝐶) Where:
E = market value of firm’s equity
D = value of firm’s debt
ke = cost of the firm’s equity, as calculated by the Capital Asset Pricing Model
kD = cost of the firm’s debt, and
tc = corporate tax rate
4.3. Sample
The sample includes observations with fiscal years-end dates from May 2000 through February 2017. The limiting factor for the fiscal years used is the availability of WACC data from Bloomberg. If WACC data availability were not a constraint, then the sample could begin in 1988; in this year, firms began to report accurate accruals data in the cash flow statement, as discussed by Hribar and Collins (2002). The study focuses on public companies listed on major US exchanges. I use the CRSP Compustat Merged data base instead of the Compustat database alone because the Compustat database also includes companies that are registered with the SEC but that are not necessarily publicly listed. To avoid including small and micro-cap stocks, I exclude observations with assets less than $10 million and price per share less than $5 as of the fiscal year-end date. I also exclude companies with unique
characteristics that may affect the interpretation of the data, or those for which required fields are typically not reported; these are financial companies and utilities.
The data sample is further limited to firms and years covered by all of the databases I use, and for which the required historical data are available. The table below summarizes the data requirements that define the ultimate sample used in the study. The order in which the requirements are listed corresponds to the order in which the data were collected, cleaned, and merged. Note that the number of observation lost due to any particular requirement would be different if the data collection, cleaning, and merging operations were executed in a different order.
Table 2: Description of data and requirements that limit sample size
Data Description by Source
Observations Lost
Observations Remaining I/B/E/S data cleaning:
Query for earnings per share (EPS) estimate and actual data for fiscal
year-end dates between Jan. 1988 and Apr. 2017 1,579,530
Missing median EPS estimate 32 1,579,498
Missing actual EPS 103,694 1,475,804
Forecast issued after earnings announcement 18,560 1,457,244
Actual EPS currency different from estimate currency 3,853 1,453,391
Keep only forecast from 11th month of firm’s fiscal year 1,328,676 124,715
Firm-year duplicates 17 124,698
Missing PERMNO identifier 6,867 117,831
CRSP Compustat Merged file cleaning:
Query for fiscal year-end dates between Jan. 1988 and Dec. 2016, with positive Total Assets, non-null SIC code, and data for at least one of R&D or Advertising expense fields
110,784
Firm-year duplicates 1,432 109,352
Unclassifiable SIC code 45 109,307
Non-US firms 11,587 97,720
Firms in Utilities industry 2,921 94,799
Firms in Financials industry 13,784 81,015
Total Assets below $10 million 8,313 72,702
Price per share at fiscal year-end below $5 18,833 53,869
Minimum Data Requirements for Regressions:
Records missing all dependent or all independent13 variables tested 2,136 51,733
Records missing any of the shared control variables14 7,809 43,924
13 Refers to all of the focus independent variables being tested, which are the dummy variables defining the nature of the current year’s earnings surprise and earnings quality, and the historical frequency of just beating earnings expectations with low earnings quality. Observations lost due to missing control variable data are detailed in the next line of the table.
14 All regressions include as control variables (1) the natural logarithm of the firm’s market capitalization, (2) the natural logarithm of the number of individual analyst estimates comprising the median earnings per share
4.4. Descriptive statistics
The descriptive statistics included in this section represent sub-sets of the sample described in the previous section. The sample used to test Hypotheses 1, 2, and 3 is limited to those observations that have all of the required data to test each of the three hypotheses. This allows a meaningful comparison among the results of each hypothesis so that I can trace the individual contribution of each factor, RNOA and WACC, to the value creation metric. To test Hypotheses 1a, 2a, and 3a, I use a sample of 9,970 firm-year observations. The descriptive statistics for that sample’s independent, dependent, and control variables are shown in Table 3.
Table 3: Descriptive Statistics for Sample Used to Test Hypotheses 1-3
Statistics av5rnoa av5wacc av5rnoa_wacc rnoa wacc rnoa_wacc mc numest N 9,970 9,970 9,970 9,970 9,970 9,970 9,970 9,970 mean 0.1997 0.0978 0.1019 0.2011 0.0975 0.1036 8,152 9.5 sd 0.1922 0.0190 0.1906 0.2624 0.0249 0.2609 26,812 7.2 min 0.0002 0.0208 (0.1708) (2.1263) 0.0274 (2.2551) 20 1.0 p25 0.0935 0.0848 (0.0001) 0.0856 0.0789 (0.0046) 531 4.0 p50 0.1448 0.0972 0.0487 0.1454 0.0954 0.0505 1,363 7.0 p75 0.2303 0.1104 0.1325 0.2410 0.1128 0.1415 4,361 13.0 max 1.4644 0.1851 1.3975 1.4644 0.2580 1.4157 504,240 47.0
Table 4 shows the mean and median of the dependent and independent variables used in Hypotheses
1-3 for each of the relevant categories of observations. Observations that just beat expectations with
low earnings quality are shown separately from all other observations; these two categories add up to the total sample in Table 3 and they are used to test H1a, H2a, and H3a. Table 4 also shows descriptive statistics for observations that just missed earnings expectations with high earnings quality; this group is compared to the first group to test H1b, H2b, and H3b.
Table 4 shows that the long-term average RNOA (labeled av5rnoa) for firms suspected of myopic earnings management (group 1) is lower than for all others (group 2) and is also lower than for firms who are least likely to have managed earnings (group 3). The group 1 median is lower than the medians for group 2 and group 3 by 155 basis points (bps) and 392bps, respectively. This suggests that future accounting returns of firms who engage in myopic earnings management may suffer economically significant consequences compared to those who do not manage earnings. It is also interesting to compare the groups’ RNOA in the event year itself. The group 1 event year RNOA is lower than that of group 2 and group 3 by 145bps and 96bps, respectively. The gap in accounting returns between myopically managed firms and other firms appears to grow over time.
The long-term average WACC (labeled av5wacc) of firms suspected of myopic earnings management is also lower than that of the two other groups, but the differences are smaller for WACC than they
are for RNOA. The median long-term WACC of group 1 is lower than that of group 2 and group 3 by 47bps and 78bps, respectively. While not negligible, the impact of myopic earnings management on cost of capital appears less economically significant than the impact on accounting returns.
Given that long-term returns suffer more than cost of capital benefits, it follows that myopically managed firms have a lower long-term RNOA-WACC spread. The RNOA-WACC spread for firms in group 1 is lower than that of groups 2 and 3 by 113bps and 297bps, respectively.
Table 4: Descriptive Statistics for Sample Used to Test Hypotheses 1-3 by Observation Category
Observation category Statistics av5rnoa av5wacc av5rnoa_wacc rnoa wacc rnoa_wacc
1. Just beat with low earnings quality
N 226 226 226 226 226 226 mean 0.1689 0.0927 0.0762 0.1592 0.0909 0.0683 median 0.1297 0.0926 0.0377 0.1313 0.0878 0.0387 2. All others who did
not just beat with low earnings quality
N 9,744 9,744 9,744 9,744 9,744 9,744 mean 0.2004 0.0979 0.1025 0.2020 0.0977 0.1044 median 0.1452 0.0973 0.0490 0.1458 0.0956 0.0509
3. Just miss with high earnings quality
N 187 187 187 187 187 187 mean 0.2311 0.1008 0.1303 0.1875 0.1004 0.0871 median 0.1689 0.1004 0.0674 0.1409 0.0986 0.0437
Table 5 shows the correlation among the variables used to test Hypotheses 1-3. The first three variables listed are the average value of the firm’s RNOA, WACC, and its RNOA-WACC spread during the five years following the event year. Each of these variables is used as the dependent variable in the regressions for H1, H2, and H3, respectively. The next three variables listed are the event year’s RNOA, WACC, and RNOA-WACC spread. Each one of these is used as a control variable for the regression containing the corresponding dependent variable. The last two variables shown in the table represent the firm’s market capitalization and the number of analyst estimates comprising the median consensus EPS estimate for the given period. The natural logarithms of both of these values are used as control variables in each of the three regressions used to test H1, H2, and H3.
Table 5: Correlation Among Variables for Hypotheses 1-3
av5rnoa av5wacc av5rnoa_wacc rnoa wacc rnoa_wacc mc numest av5rnoa 1.0000 av5wacc 0.1307 1.0000 av5rnoa_wacc 0.9951 0.0322 1.0000 rnoa 0.6123 0.0744 0.6099 1.0000 wacc 0.1225 0.5882 0.0650 0.1079 1.0000 rnoa_wacc 0.6042 0.0187 0.6072 0.9955 0.0132 1.0000 mc 0.1266 (0.1113) 0.1388 0.1080 (0.0678) 0.1151 1.0000 numest 0.2260 (0.0303) 0.2308 0.1825 0.0706 0.1768 0.4323 1.0000
Hypothesis 4 is tested using a different sub-set of the data sample. The sample used is limited to those
observations that have all of the required data to test the impact of historical myopic earnings management frequency on current RNOA, WACC, and the RNOA-WACC spread. The 13,248 firm-year observations that meet these requirements are shown in Table 6 with descriptive statistics for the relevant dependent and independent variables used in the regressions.
Table 6: Descriptive Statistics for Sample Used to Test Hypothesis 4
Statistics rnoa wacc rnoa_wacc jblfreq5 l_rnoa l_wacc l_rnoa_wacc mc numest N 13,248 13,248 13,248 13,248 13,248 13,248 13,248 13,248 13,248 mean 0.2084 0.0961 0.1123 0.0225 0.2003 0.0959 0.1044 9,130 10.0 sd 0.2244 0.0235 0.2222 0.0706 0.2469 0.0242 0.2451 29,187 7.6 min 0.0000 0.0224 (0.1783) 0.0000 (2.1263) 0.0219 (2.2493) 24 1.0 p25 0.0890 0.0794 (0.0017) 0.0000 0.0859 0.0783 (0.0038) 609 4.0 p50 0.1445 0.0940 0.0497 0.0000 0.1441 0.0937 0.0502 1,622 8.0 p75 0.2364 0.1102 0.1399 0.0000 0.2373 0.1106 0.1398 5,121 14.0 max 1.4644 0.2404 1.4157 0.6000 1.4644 0.2580 1.4157 547,815 49.0
Table 7 shows mean and median values of the three dependent variables tested for each category of historical myopic earnings management frequency (labeled jblfreq5). Because I limit the sample to firms that have at least four years of the historical data needed to identify potential myopic earnings management, the values that the myopic earnings management frequency variable can take on are limited to 0, 0.2, 0.25, 0.4, 0.5, 0.6, 0.75, 0.8, and 1. The sample contains no observations with a myopic earnings management frequency variable greater than 0.6. With some exceptions, it appears that the lower frequencies of historical myopic earnings management tend to be associated with higher RNOA, higher WACC, and a higher RNOA-WACC spread. This pattern suggests that the negative consequences of myopic earnings management for value creation may be exacerbated by more frequent earnings management activity.
Table 7: Descriptive Statistics for Sample Used to Test Hypothesis 4 by Category of Myopic Earnings Management Frequency
jblfreq5 Statistics rnoa wacc rnoa_wacc
0 N 11,904 11,904 11,904 mean 0.2120 0.0967 0.1154 median 0.1456 0.0947 0.0506 0.2 N 1,164 1,164 1,164 mean 0.1788 0.0921 0.0867 median 0.1351 0.0890 0.0453 0.25 N 68 68 68 mean 0.1637 0.0894 0.0743 median 0.0952 0.0878 0.0123 0.4 N 89 89 89 mean 0.1648 0.0878 0.0770 median 0.1282 0.0877 0.0439 0.5 N 5 5 5 mean 0.1890 0.1090 0.0800 median 0.1079 0.1005 0.0030 0.6 N 18 18 18 mean 0.1271 0.0796 0.0475 median 0.1041 0.0792 0.0364 Total N 13,248 13,248 13,248 mean 0.2084 0.0961 0.1123 median 0.1445 0.0940 0.0497
Table 8 shows the correlation among the variables used to test Hypothesis 4. The first three variables are the current year’s RNOA, WACC, RNOA-WACC spread; each one is used as the dependent variable in three separate equations. The fourth variable listed represents the frequency of suspected earnings management activity in the five years prior to the current year. This is used as the main independent variable in each of the three aforementioned regressions. The following three variables are the lagged values of RNOA, WACC, and the RNOA-WACC spread for the year prior to the current year. Each one is used as a control variable in the regression containing the corresponding current year value as the dependent variable. The last two variables in the table represent the firm’s market capitalization and the number of analyst estimates comprising the median consensus EPS estimate. The natural logarithms of these values are used as control variables in the regressions used to test H4.
Table 8: Correlation Among Variables for Hypothesis 4
rnoa wacc rnoa_wacc jblfreq5 l_rnoa l_wacc l_rnoa_wacc mc numest
rnoa 1.0000 wacc 0.1483 1.0000 rnoa_wacc 0.9945 0.0441 1.0000 jblfreq5 (0.0483) (0.0687) (0.0416) 1.0000 l_rnoa 0.7015 0.1139 0.6967 (0.0394) 1.0000 l_wacc 0.1397 0.7315 0.0639 (0.0767) 0.1213 1.0000 l_rnoa_wacc 0.6928 0.0424 0.6953 (0.0320) 0.9952 0.0233 1.0000 mc 0.1247 (0.0821) 0.1347 (0.0226) 0.1197 (0.0732) 0.1278 1.0000 numest 0.2115 0.0368 0.2097 (0.0093) 0.2035 0.0757 0.1975 0.4298 1.0000
5. Results
To test my hypotheses, I use panel data regressions with controls for year and industry fixed effects, in addition to several other control variables. Each hypothesis, the regressions used to test it, and its results, are discussed in detail in the sections that follow.
To test the long-term effects on value creation in Hypothesis 1, 2, and 3, the dependent variables are the average Return on Net Operating Assets, Weighted Average Cost of Capital, and the spread between the two, respectively, during the five years following the year in which the suspected earnings management event took place. I also test the effects in the mid-term, in which case the dependent variable is the average during the three years following the event year. The mid-term results are not shown, but they are similar to the long-term results. The focus independent variable is a dummy variable indicating whether or not the firm-year observation has just beat earnings expectations with low quality earnings.
To test the effects on value creation in Hypothesis 4, the dependent variables are the current year’s value of RNOA, WACC, and the RNOA-WACC spread. The independent variable is the historical frequency of myopic earnings management over the previous five years.
5.1. Hypothesis 1
Firms suspected of myopic earnings management (those who beat earnings expectations with low quality earnings) should, in the long term, have lower RNOA than:
(a) all other firms, and
(b) firms with high quality earnings who miss earnings expectations.
To test Hypothesis 1a, I regress the average RNOA in future years on a dummy variable indicating whether or not the firm-year observation has just beat earnings expectations with low quality earnings. I control for the firm’s market capitalization, the number of analyst estimates comprising the median EPS estimate, and the event year’s RNOA,in addition to year and industry fixed effects. I control for the natural log of a firm’s market capitalization because size may impact a firm’s accounting returns, the availability of real investment opportunities, and its cost of capital.I control for the number of analyst estimates comprising the median EPS estimate because analyst coverage quality and forecast accuracy may vary with the level of analyst coverage. Firms with greater analyst coverage may have more accurate forecasts, affecting the likelihood that they meet earnings expectations. I control for the firm’s RNOA in the event year, because it may impact the RNOA in future years. Freeman et al. (1982) have demonstrated that there is regression toward the mean in accounting returns measures such as ROE. In order to control for external market or economic
conditions that may affect real investment decisions, performance, and the cost of capital from year to year, I control for time by using fixed effects for the fiscal year. Finally, because there may be inherent differences in the returns and cost of capital among different industries, I use fixed effects on the two-digit SIC code to control for the industry group to which the firm belongs15.
The sample used for this regression is limited to observations whose average 5-year future RNOA is positive and which have available RNOA data for at least four out of five years following the event year. I also limit the sample to observations that have all of the required data needed to test
Hypothesis 2 and 3, so that the results among H1-3 can be compared.
The estimated coefficient on the dummy variable that indicates myopic earnings management is negative, as predicted, indicating that firms who exercise myopic earnings management have lower future RNOA than all other firms. However, the coefficient is not significant with a p-value of 0.117, so I cannot conclude that there is a statistically significant difference in the future returns of these two groups. The results of this regression are shown in Table 9 as Regression 1. I also test an alternative version of this regression, excluding the firm’s RNOA in the event year from the control variables. The estimated coefficient from this regression is negative, as predicted, and it is statistically significant at the 5% level with a p-value of 0.017. The disadvantage of this alternative test is that the model fit is poorer with a much lower R-squared of 0.13 compared to 0.42 in the first equation. The results of this alternative regression are shown in Table 9 as Regression 2.
To test Hypothesis 1b, I use the same regression as that used for H1a, but only a sub-sample of the data used to test H1a. In order to compare the outcome for firms who beat with low quality earnings only to firms who missed with high quality earnings, I run the regression on a sub-sample of the data that includes only observations in these two categories.
The estimated coefficient on the dummy variable that indicates myopic earnings management is negative, as predicted, and significant at the 1% level. This indicates that firms suspected of myopic earnings management have RNOA in the five years following the event year that is significantly lower than that of the firms least likely to have engaged in myopic earnings management. This result remains
15 Some additional control variables may also be appropriate to use, but I have not included them in this study. (1) Distress: The characteristics I use to identify myopic management may also be indicative of tactics used by companies in distress to try to cope with threats to the business. In these cases, the distress would be related to both the myopic management activity and to the long-term performance of the firm. One could control for distressed cases by identifying companies with low cash flows but relatively high earnings. (2) Ownership: Management’s focus on meeting short- vs. long-term objectives may be impacted by the investment horizon and sophistication of the firm’s investors, so it may be useful to control for investor type. (3) Management incentives: The prominence of stock options in Management’s compensation as well as the timing of stock
significant if I exclude the event year’s RNOA from the list of control variables. The results of these regressions are shown in Table 9 as Regressions 3 and 4.
Table 9 summarizes the output of the regressions used to test Hypothesis 1. The dependent variable in each regression is the average RNOA over the five years following the event year (labeled av5rnoa). The independent variable jbl is a dummy variable that equals 1 if the observation just beat earnings expectations with low earnings quality for the given year, and equals 0 otherwise. The control variables included are the natural logarithm of the firm’s market capitalization (lnmc), the natural logarithm of the number of individual analyst estimates that comprise the median consensus estimate (lnnumest), and the firm’s RNOA in the event year (rnoa). In addition to the independent variables shown, the regressions include fixed effects to control for the firms two-digit SIC code as well as the fiscal year of the observation; coefficients for these dummy variables are not shown. Regressions 1 and 2 are used to test H1a. Regression 1 includes the event year’s RNOA as a control variable, whereas Regression 2 does not. The negative coefficient estimated on the myopic earnings management dummy variable is significant only in Regression 2. Regressions 3 and 4 are used to test H1b. The sample size for Regressions 3 and 4 is much smaller because it is limited to observations that either just beat with low earnings quality or just missed with high earnings quality. Regression 3 includes the event year’s RNOA as a control variable, whereas Regression 4 does not. The coefficient estimated on the myopic earnings management dummy variable is negative and significant in both regressions.
Regression # (1) (2) (3) (4)
Dependent variable av5rnoa av5rnoa av5rnoa av5rnoa
jbl (0.016) (0.023) (0.054) (0.038) (0.117) (0.017) * (0.003) ** (0.015) * lnmc 0.008 0.022 0.010 0.016 (0.000) *** (0.000) *** (0.293) (0.041) * lnnumest 0.007 (0.002) (0.014) (0.020) (0.007) ** (0.403) (0.396) (0.141) rnoa 0.414 0.327 (0.000) *** (0.000) *** Intercept 0.081 0.106 0.186 0.205 (0.065) (0.008) ** (0.161) (0.120) N 9,970 14,583 413 610 R-sq 0.42 0.13 0.40 0.21 adj. R-sq 0.41 0.13 0.30 0.12
p-values in parentheses and italics * p<0.05, ** p<0.01, *** p<0.001
5.2. Hypothesis 2
Firms who beat earnings expectations with low quality earnings should, in future years, have lower WACC than:
(a) all other firms, and
(b) firms with high quality earnings who miss earnings expectations.
To test Hypothesis 2a, I regress the average WACC in future years on a dummy variable indicating whether or not the firm-year observation has just beat earnings expectations with low quality earnings. I control for the firm’s market capitalization, the number of analyst estimates comprising the median EPS estimate, and the event year’s WACC,in addition to year and industry fixed effects. The sample used for this regression is limited to observations that meet all of the requirements outlined for testing H1 and all of the required data for testing H3, so that the results among H1-3 are based on the same sample and can be compared.
The estimated coefficient on the dummy variable that indicates myopic earnings management is negative, as predicted, and significant at the 5% level. This confirms my prediction that firms who engage in myopic earnings management in order to slightly beat earnings expectations enjoy a lower cost of capital in future years than all other firms. The results of this regression are shown in Table 10 as Regression 5.
To test Hypothesis 2b, I use the same regression as that used for H2a, but the test is run on a sub-sample of that used for H2a. The data used is limited to observations that either just beat the estimate with low earnings quality or just missed the estimate with high earnings quality. This allows us to contrast the firms most likely to have engaged in myopic earnings management to those least likely to have done so.
The estimated coefficient on the dummy variable that indicates myopic earnings management is negative, as predicted, and significant at the 1% level. Firms that engage in myopic earnings management have significantly lower cost of capital in future years than those firms that missed earnings expectations and have probably not engaged in myopic earnings management. The results of this regression are shown in Table 10 as Regression 6.
Table 10 summarizes the output of the regressions used to test Hypothesis 2. The dependent variable in each regression is the average WACC over the five years following the event year (labeled av5wacc). The independent variable jbl is a dummy variable that equals 1 if the observation just beat earnings expectations with low earnings quality for the given year, and equals 0 otherwise. The control variables included are the natural logarithm of the firm’s market capitalization (lnmc), the natural logarithm of the number of individual analyst estimates that comprise the median consensus estimate (lnnumest), and the firm’s WACC in the event year (wacc). In addition to the independent variables shown, the regressions include fixed effects to control for the firms two-digit SIC code as well as the fiscal year of the observation; coefficients for these dummy variables are not shown. Regression 5 is used to test H2a. The coefficient estimated on the myopic earnings management dummy variable is negative and significant at the 5% level. Regression 6 is used to test H2b. The sample size for Regression 6 is much smaller because it is limited to observations that either just beat with low earnings quality or just missed with high earnings quality. The coefficient estimated on the myopic earnings management dummy variable is negative and significant at the 1% level.
Regression # (5) (6)
Dependent variable av5wacc av5wacc
jbl (0.002) (0.004) (0.029) * (0.003) ** lnmc (0.001) (0.001) (0.000) *** (0.131) lnnumest 0.001 (0.001) (0.008) ** (0.302) wacc 0.419 0.446 (0.000) *** (0.000) *** Intercept 0.073 0.057 (0.000) *** (0.000) *** N 9,970 413 R-sq 0.56 0.65 adj. R-sq 0.56 0.60
p-values in parentheses and italics * p<0.05, ** p<0.01, *** p<0.001
