Prevalence of the Zero-Earnings Discontinuity in Quarterly Reported Earnings

Introduction

In their seminal paper, Burgstahler and Dichev (1997) documented the so-called zero-earnings discontinuity: a “hole” in the frequency of reported earnings just below zero followed by a “bump” at and just above zero. Following their paper, the discontinuity captured the interest of accounting scholars, leading to a stream of literature investigating its existence and implications.

The discontinuity is controversial and sparks debate as to what drives its existence. Some argue that it reflects companies’ incentives to meet and beat the zero-earnings benchmark (e.g., Burgstahler and Dichev (1997), Dichev et al. (2013), Gilliam et al. (2015), Burgstahler and Chuk (2017), and de la Rosa and Lambertsen (2022)), while others suggest that it is merely an artifact in the data (e.g., Durtschi & Easton, 2005, 2009). Scholars of the former view have extensively studied whether exogenous shocks, such as regulatory interventions, impact its presence (Enomoto & Yamaguchi, 2017; Gilliam et al., 2015).

In this paper, we contribute to this body of literature by investigating the zero-earnings discontinuity in interim financial statements surrounding the adoption of the Sarbanes-Oxley Act (hereafter SOX) in 2002. In particular, we investigate whether the discontinuity in quarterly financial statements persists after the adoption. Additionally, we explore whether the persistence is driven by the decline in the quarterly zero-earnings discontinuity being less pronounced compared to the annual zero-earnings discontinuity.

The adoption of SOX marked a major regulatory intervention that led to stricter financial reporting. It involved the creation of an oversight board to ensure sufficient audit quality, as well as accurate and reliable reporting (SOX, 2002, Section 101), limiting audit firms’ supply of nonaudit services to their audited clients to enhance auditor independence (SOX, 2002, Section 208), and disclosure on firms’ internal control measures (SOX, 2002, Section 404). As a result, the adoption of SOX likely led to lower managerial flexibility and, hence, reduced earnings management. Consistent with this, the prior literature finds that the zero-earning discontinuity in annual earnings disappeared after the adoption, and attributes this to increased costs of performing earnings management (e.g., Gilliam et al. (2015)). Similar results are found in Japan following the adoption of the Financial Instruments and Exchange Act of 2006 (hereafter J-SOX), although the discontinuity did not disappear completely (Enomoto & Yamaguchi, 2017). Since companies are required to comply with SOX irrespective of whether they issue annual or quarterly financial statements, the zero-earnings discontinuity in quarterly earnings may also disappear after the adoption.

On the contrary, quarterly financial statements, unlike their annual counterparts, are not subject to mandatory audits (17 CFR, 2021, 210.10–01.a.1). During an audit, auditors investigate whether the financial statements contain material misstatements and whether the company complies with the regulation in force. The absence of an auditor when preparing quarterly financial statements allows managers to engage in earnings management that otherwise is likely to be identified and corrected by auditors in annual financial statements. Consequently, quarterly financial statements are less likely to be compliant with SOX compared to annual financial statements, making SOX less effective in restricting earnings management in the former. Moreover, quarterly financial statements are generally more likely to contain material misstatements compared to annual financial statements, since it is challenging to determine “the results of operations on a meaningful basis for intervals of less than a full year” (GAAP, 2022) [270-10-05-2] and, “in view of the limited time available to develop complete information, many costs and expenses are estimated in interim periods” (GAAP, 2022) [270-10-05-2]. Hence, management has more leeway to manipulate earnings in quarterly financial statements compared to annual financial statements even in the absence of SOX, which is likely to cause the differential impact of the adoption to be economically significant. In contrast, prior studies (e.g., Jacob and Jorgensen (2007)) find that earnings management may not be common in quarterly earnings, as management’s compensation is often tied to annual earnings at the fiscal year-end. As a result, management may face little incentive to manage earnings on an interim basis. Conversely, as the adoption of SOX likely restricts the ability of management to manage annual earnings, managing quarterly earnings may become a viable alternative. This may result in greater incentive to manage earnings on an interim basis. Altogether, the zero-earnings discontinuity in quarterly earnings may persist after the adoption of SOX.

Discontinuities around zero have been documented in various contexts. Prior studies (e.g., Burgstahler and Dichev (1997), and (Gilliam et al., 2015)) show that the discontinuity in zero changes in earnings tends to mirror the discontinuity in zero earnings. Other studies (e.g., Doyle et al. (2013), and Huang et al. (2017), and Kim et al. (2021)) explore discontinuities in the market’s expectations of firms’ performance using financial analysts’ earnings forecast errors, and find a discontinuity which they attribute to firms’ effort to meet and beat analysts’ earnings forecasts (e.g., Chang et al. (2023), Gilliam et al. (2015), and Bird et al. (2019)). We follow the literature and extend our analysis by examining the discontinuity around zero in changes in earnings as well as in analysts’ earnings forecast errors, using both annual and quarterly data, around the adoption of SOX.

To investigate our hypotheses, we collect financial data on annual and quarterly financial statements filed by listed companies in the USA using Compustat from 1980 to 2020, and analysts’ earnings forecasts on listed firms in USA using I/B/E/S from 1983 to 2020. ¹ After truncation, our sample consists of 222,611 (829,664), 204,685 (751,274), and 138,160 (339,500) observations of annual (quarterly) earnings, changes in annual (quarterly) earnings, and analysts’ annual (quarterly) earnings forecast errors, respectively. We employ the standardized-difference test by Burgstahler and Dichev (1997) as our main test, and the manipulation test by Cattaneo et al. (2020) ² as a robustness test. The manipulation test offers statistical advantages over the standardized-difference test, including its use of a data-driven bandwidth. To this end, we address the concern commonly raised in the literature (e.g., Lahr (2014, p. 560)) that the zero-earnings discontinuity may be driven by the researchers’ choice of interval size, while simultaneously complementing recent studies on how to measure the zero-earnings discontinuity (e.g., Byzalov and Basu (2019) and Bird et al. (2019)) by adopting a test that is more robust compared to those commonly used in prior studies (e.g., Burgstahler and Dichev (1997)). We capture the zero-earnings discontinuity using (changes in) earnings scaled by lagged total assets and verify that our findings are robust toward scaling by lagged market capitalization. Moreover, we capture the discontinuity in the market’s expectations using analysts’ earnings forecast errors measured as the consensus earnings forecast per share minus the firms’ actual earnings per share as per prior studies (e.g., Das et al. (2024)).

We find that the zero-earnings discontinuity persists in quarterly earnings following the adoption of SOX. This is consistent with our prediction that the lack of mandatory audits in quarterly financial statements reduces the effectiveness of SOX in restricting earnings management. To further examine the persistence, we investigate the relative magnitude of the change in the discontinuity around the adoption. On the one hand, the zero-earnings discontinuity in quarterly earnings may persist due to SOX having a weaker impact on the quarterly discontinuity compared to the annual discontinuity. On the other hand, due to the different auditing requirements, managers are likely to have more discretion in making reporting choices for quarterly financial statements than for annual financial statements, both before and after SOX. This greater flexibility is likely to cause the discontinuity to be larger in magnitude for quarterly earnings compared to annual earnings prior to the adoption. We verify this and find that the standardized-difference test statistics are roughly twice as large in quarterly earnings compared to annual earnings. As such, the zero-earnings discontinuity in quarterly earnings could persist following the adoption even if the impact of the adoption is greater, simply due to the discontinuity being substantially larger in magnitude prior to the adoption. To investigate this, we adopt a difference-in-difference design, comparing the change in the zero-earnings discontinuity around the adoption between quarterly and annual financial statements. As a robustness test, we allow for continuous changes over time in the magnitude of the zero-earnings discontinuity to ensure that our findings are not driven by extreme-year observations (see Enomoto and Yamaguchi (2017) for more on continuous changes over time).

We find that the decrease in the zero-earnings discontinuity is smaller in magnitude for quarterly earnings compared to annual earnings. This is consistent with our prediction that the persistence is driven by the weaker impact of SOX on the quarterly zero-earnings discontinuity compared to the annual zero-earnings discontinuity. Although drawing causal inference is challenging as per previous studies examining the adoption of SOX (e.g., Gilliam et al. (2015)), our findings align with the explanation that the adoption of SOX caused the disappearance of the zero-earnings discontinuity in annual earnings, while the zero-earnings discontinuity in quarterly earnings persists because SOX is less effective in restricting earnings management in interim financial statements.

Next, we examine the discontinuity for other distributions commonly studied in the literature. We find similar results for changes in earnings, suggesting that firms are more likely to meet and beat their previous quarterly earnings compared to their previously reported annual earnings, following the adoption of SOX. For analysts’ earnings forecast errors, we find that the discontinuity for small positive forecast errors persists in both annual and quarterly earnings following the adoption. The discontinuity is larger in magnitude for quarterly forecast errors compared to annual forecasts errors, consistent with our findings using earnings and changes in earnings scaled by lagged assets. The distribution of earnings forecast errors is dissimilar to that of earnings, with both smaller negative and smaller positive forecast errors than expected under continuity. We find no discontinuity in small negative forecast errors for annual earnings, both before and after the adoption.

Lastly, we conduct a series of robustness tests. Prior studies (e.g., Durtschi and Easton (2005; 2009)) speculate that the zero-earnings discontinuity may be driven by sample selection, model specifications, and tax items. We find that our results remain robust to these effects. To further address the concern that our findings could be driven by methodological choices, we extend the robustness tests from prior studies as well as conduct several model specification tests. Our results remain robust in all of these tests, including different scaling factors, measurements, and estimation specifications.

The paper contributes to the literature in two ways. First, it adds to the vast line of research investigating the consequences of SOX (Coates & Srinivasan, 2014). Prior studies find differences in earnings management behavior (Cohen et al., 2008; Krishnan et al., 2011), altered markets for auditors (DeFond & Lennox, 2011), and changed corporate risk-willingness (Bargeron et al., 2010) following the adoption. This paper is the first to demonstrate that the adoption is also associated with differential changes in financial reporting quality for different types of financial statements. It hereby extends prior studies showing the existence of the quarterly zero-earnings discontinuity prior to the adoption (Brown & Caylor, 2005), and complements studies showing that the annual zero-earnings discontinuity disappears after the adoption (Gilliam et al., 2015).

Second, the paper adds to the literature on the adoption of SOX-like regulations (Chardonnens & Wallmeier, 2024; Enomoto & Yamaguchi, 2017). Prior studies argue that J-SOX is more lenient than SOX regarding internal controls (Enomoto & Yamaguchi, 2017). This aligns with our findings that SOX is more lenient toward quarterly financial statements compared to annual financial statements. By extending studies on earnings management (e.g., Durtschi and Easton (2005), Durtschi and Easton (2009), Burgstahler and Chuk (2015), and Burgstahler and Chuk (2017)), this paper shows that the zero-earnings discontinuity remains relevant in the current regulatory environment.

The remainder of the paper is organized as follows. Section 2 presents the hypothesis development. Section 3 outlines the research design and sample construction. Section 4 reports the findings. Section 5 depicts the robustness tests and, lastly, section 6 concludes.

Hypothesis Development

The Sarbanes-Oxley Act was enacted in 2002 in response to a series of corporate scandals in the late 1990s and early 2000s. Its primary objective was to restore trust in financial reporting “by improving the accuracy and reliability of corporate disclosures” (SOX, 2002, amendment). It represented a major regulatory intervention, leading to long-lasting changes in the financial reporting environment. Among others, SOX required the establishment of an oversight board to issue and monitor compliance with auditing standards (SOX, 2002, Section 101.c.2–101.c.6), restricted audit firms from offering nonaudit services to their clients (SOX, 2002, Section 208.b), and enhanced internal control transparency by mandating an internal control report to be disclosed (SOX, 2002, Section 404). Prior studies generally agree that the adoption was followed by increased financial disclosure and tightened auditor requirements.

As a result, substantial attention has been devoted to investigating whether the adoption of SOX leads to an increase in financial reporting quality. The literature employs various measures of financial reporting quality with earnings management proxies comprising an integral part. One type of earnings management that has received particular attention is the zero-earnings discontinuity: the “hole” in the frequency of earnings below zero followed by a “bump” at and above zero. Its existence is commonly interpreted as firms’ incentive to manage earnings from a small loss to a small profit (Burgstahler & Dichev, 1997). However, despite its widespread use in the literature to capture earnings management (e.g., Jacob and Jorgensen (2007), Li (2014), and Burgstahler and Chuk (2015)), it is surrounded by great controversy. Prior studies argue that the discontinuity may merely be an artifact of income tax items (Beaver et al., 2007), scaling (Dechow et al., 2003), and sample selection (Durtschi & Easton, 2009). Yet, these arguments are often followed by additional studies ruling out such explanations (e.g., Burgstahler and Chuk (2013), and Burgstahler and Chuk (2015)).

Nonetheless, considerable attention has been devoted to examining the existence of the zero-earnings discontinuity and how changes in the discontinuity coincide with exogenous regulatory interventions (e.g., Enomoto and Yamaguchi (2017)). Prior studies (e.g., Gilliam et al. (2015)) find that the zero-earnings discontinuity in annual financial statements disappears following the adoption of SOX. This is commonly interpreted as the consequence of the stricter reporting environment that makes accrual earnings management more challenging to perform. However, little is known about the zero-earnings discontinuity, as well as other discontinuities, in quarterly financial statements and whether they also disappear following the adoption. This is economically important, as such statements are channels through which management discloses information (e.g., Lougee and Marquardt (2004)). Prior studies (e.g., Brown and Caylor (2005)) document that the zero-earnings discontinuity exists in quarterly earnings prior to the adoption. However, to our knowledge, no studies explore the existence of the zero-earnings discontinuity in quarterly financial statements after the adoption of SOX. Our study aims to fill this void by providing insight into the behavior of discontinuities in quarterly financial statements.

It is, ex ante, unclear how the adoption of SOX coincides with changes in the zero-earnings discontinuity. On the one hand, the improved accuracy and reliability in financial reporting following the adoption are likely to reduce management’s ability to manage earnings in annual financial statements. Additionally, violation of SOX increases the risk of personal prosecution for management, which could decrease their incentive for intentionally misstating financial information. In line with this, prior studies (e.g., Gilliam et al. (2015)) conjecture and find that the zero-earnings discontinuity in annual earnings disappears after adoption. While little is known about the zero-earnings discontinuity in quarterly financial statements, a similar disappearance is likely to be present following the adoption, as compliance with SOX is also required in quarterly financial statements.

On the other hand, the existence of mandatory audits is likely to play an important role in supporting the effectiveness of SOX in reducing earnings management. During an audit, an auditor investigates whether firms’ financial statements contain material misstatements and assesses their level of compliance with the regulation in force. In contrast to annual financial statements, quarterly financial statements “need only be provided as to the registrant and its subsidiaries consolidated and may be unaudited” (17 CFR, 2021, 210.10–01.a.1). This creates an opportunity for management to remain noncompliant with SOX on a quarterly basis while still being compliant on an annual basis. As a result, annual earnings may be vastly different from the sum of quarterly earnings, since management can perform misstatements in quarterly earnings that remain undiscovered until the annual audit. ³ However, despite this opportunity, management may be disincentivized to engage in such practices, as deviations between the sum of quarterly earnings and annual earnings will become apparent when the annual financial statements are published. Unless management can convince capital providers that such deviations are due to estimation errors, their earnings management behavior is likely to be revealed through the annual earnings. Due to the potential manipulation in quarterly earnings, the zero-earnings discontinuity in quarterly financial statements may not disappear after the adoption of SOX.

Therefore, we present our first directional hypothesis.

H1

The zero-earnings discontinuity in quarterly earnings persists after the adoption of SOX.

The persistence of the zero-earnings discontinuity in quarterly earnings may be attributable to different effects. On the one hand, the discontinuity may persist if SOX is less effective in restricting earnings management in quarterly financial statements compared to annual financial statements.

On the other hand, the discontinuity may also persist if SOX is equally, or even more, effective in restricting earnings management in quarterly financial statements compared to annual financial statements. This is because earnings management is more likely to occur in quarterly financial statements than in annual financial statements prior to the adoption, due to the absence of auditors and greater potential estimation error in many line items in the statement (GAAP, 2022) [270-10-05-2]. As such, the zero-earnings discontinuity in quarterly earnings may persist following the adoption, even if the impact of SOX is greater, simply due to the discontinuity being substantially larger in magnitude prior to the adoption.

To test whether the adoption of SOX has a weaker impact, in magnitude, on the zero-earnings discontinuity in quarterly compared to annual earnings, we present our second directional hypothesis.

H2

The zero-earnings discontinuity in quarterly earnings declines less after the adoption of SOX compared to the zero-earnings discontinuity in annual earnings.

Research Design and Sample Selection

Research Design

To investigate Hypothesis 1, we follow prior studies (e.g., Brown and Caylor (2005), Beaver et al. (2007) and Gilliam et al. (2015)) and adopt the standardized-differences test by Burgstahler and Dichev (1997) (hereafter BD) as our main test. ⁴ By grouping the earnings distribution into intervals, the test identifies a discontinuity if the number of observations in a given interval differs significantly from the expected, which is defined as the average of the adjacent number of observations. Following Gilliam et al. (2015), we incorporate the variance correction proposed by Beaver et al. (2007) to obtain valid inference, causing our findings to be conservative compared to those by Burgstahler and Dichev (1997). Throughout our paper, we perform the BD test on both small profit and small loss observations in an effort to test both sides of the discontinuity. We find the BD test attractive due to its simplicity, the degree to which it is accepted in the literature, as well as its robustness towards skewness in the earnings distribution identified in prior studies (e.g., Burgstahler and Dichev (1997) and Gilliam et al. (2015)), since it relies only on the number of observations in the two adjacent intervals rather than the full sample. The test requires the subjective but important choice of interval size (hereafter bin size) to group the earnings distribution. We follow the prior literature (e.g., Burgstahler & Dichev, 1997 and Gilliam et al. (2015)) and use a bin size of 0.005 for annual earnings and one-fourth of that, 0.00125, for quarterly earnings.

To address the concern that our findings may be driven by the choice of bin size, we adopt the manipulation test by Cattaneo et al. (2020) (hereafter MT) as a robustness check. Intuitively, the MT test investigates self-selection in a variable of interest (the so-called “running” variable). In our case, we interpret self-selection as earnings management and adopt reported earnings as the running variable. The test is nonparametric and based on local polynomial techniques, is boundary adaptive and bins the data in a Means Squared Error (MSE ⁵ ) optimal fashion. This approach circumvents one of the main disadvantages of the BD test and follows prior literature (e.g., Degiannakis et al. (2022)) by applying nonparametric tests in investigating the zero-earnings discontinuity. We find the MT test attractive due to its nonparametric nature, allowing us to use more observations than when applying the BD test and to obtain estimates that are robust to skewness. Unless otherwise noted, we apply a first order local polynomial for the point estimate, a second order local polynomial for the bias correction, and a triangular kernel following recommendations in prior studies (e.g., Cattaneo et al. (2019a)). ⁶

To investigate Hypothesis 2, we adopt a difference-in-difference research design. Our approach is similar to that of prior studies (e.g., Enomoto and Yamaguchi (2017) and Chardonnens and Wallmeier (2024)), but instead of only considering the implementation of new regulation, we compare the difference in impact on annual and quarterly earnings. We adopt two measures capturing the magnitude of the zero-earnings discontinuity to ensure our findings are not driven by our choice of measurement: In our first measure, we use the difference between the actual and expected numbers of zero-earnings observations, which is equivalent to the numerator in the BD test (hereafter the unstandardized discontinuity measure). This enables us to isolate the size of the discontinuity from its corresponding standard error. In doing so, we address the concern that a discontinuity may appear economically large because the corresponding standard error is small. In our second measure, we standardize our first measure by its corresponding standard error, which is equivalent to the BD test statistic (hereafter the standardized discontinuity measure). This enables us to capture the strength of the discontinuity, hereby taking into account the uncertainty associated with the point estimates. To ensure consistency with our prior analysis, we investigate the small loss and small profit discontinuities individually using both measures. Furthermore, we scale both measures to ensure that they are comparable. Since quarterly earnings are four times as frequent as annual earnings, we divide the unstandardized discontinuity measure by four. Since the numerator in the BD test for quarterly earnings discontinuity is four times larger than for the annual earnings discontinuity, and the denominator is two times larger, we divide the standardized discontinuity measure by two.

Using these measures, we estimate the following two linear regression models using Ordinary Least Squares:

small loss = β 0 + β 1 Q + β 2 post - SOX + β 3 Q × post - SOX + ε ,

(1)

and

small profit = β 0 + β 1 Q + β 2 post - SOX + β 3 Q × post - SOX + ε .

(2)

“Small loss” (“Small profit”) measures either the unstandardized or standardized small loss (small profit) discontinuity, “Q” is an indicator variable for cases with quarterly discontinuities, and “post-SOX” is an indicator variable for cases where the discontinuity is observed in the post-SOX period. β₀ captures the discontinuity for annual earnings in the pre-SOX period. Based on prior studies (e.g., Burgstahler and Dichev (1997) and Gilliam et al. (2015)), we expect β₀ to be significantly negative (positive) for “small loss” (“small profit”) discontinuities. β₁ captures the difference in the discontinuity between annual and quarterly earnings in the pre-SOX period. On the one hand, quarterly earnings may be subject to greater earnings management compared to annual earnings, leading to a larger discontinuity. On the other hand, the incentive to perform earnings management on a quarterly basis may differ from that on an annual basis. Therefore, we do not predict the direction of the coefficient. β₂ captures the difference in the annual earnings discontinuity between the pre- and post-SOX periods. Following previous studies (e.g., Burgstahler and Dichev (1997) and Gilliam et al. (2015)), we expect β₂ to be significantly positive (negative) for “small loss” (“small profit”) discontinuities reflecting that the discontinuity becomes smaller after the adoption of SOX. Lastly, β₃ captures the incremental discontinuity in quarterly earnings relative to annual earnings in the post-SOX period. Given that the impact on the zero-earnings discontinuity is weaker in magnitude for quarterly earnings, we expect that β₃ to be significantly negative (positive) for “small loss” (“small profit”) discontinuities. We apply heteroskedasticity robust standard errors to control for nonconstant variances across time.

Discontinuities around zero are observed in various financial data. One example is changes in earnings (e.g., Burgstahler and Dichev (1997) and Gilliam et al. (2015)), which seek to capture firms’ incentives to meet and beat their previously reported earnings. Another example is analysts’ earnings forecast errors (e.g., Burgstahler and Dichev (1997), Gilliam et al. (2015), and Bird et al. (2019)). Financial analysts act as intermediaries between firms and capital markets, providing earnings forecasts, among others, to institutional investors to guide their investment decisions (Lang & Lundholm, 1996). In making earnings forecasts, analysts incorporate public and private information (e.g., Nichols (1989), Abarbanell (1991), Schipper (1991), Bowen et al. (2002), Weiss (2010), Nekrasov et al. (2022), and Campbell et al. (2023)). Prior studies (e.g., Degeorge et al. (1999)) find that companies manage earnings to meet and beat analysts’ expectations, which implies that a discontinuity may exist in analysts’ earnings forecast errors. We follow prior studies (e.g., Gilliam et al. (2015)) and extend our analysis by investigating discontinuities in both changes in earnings and analysts’ earnings forecast errors in quarterly financial statements after the adoption of SOX.

Descriptive Statistics

Table 1 reports descriptive statistics on a year-by-year basis for earnings scaled by lagged assets, change in earnings scaled by lagged assets, and analysts’ earnings forecast errors. Panel A reports the variables from annual financial statements and panel B reports those from quarterly financial statements. In both panels, the number of observations, the means, and the medians are tabulated. Throughout most fiscal years, the means of earnings and changes in earnings are negative, while the respective medians are positive, indicating a left-skewed distribution. Due to our choice of tests, we do not expect skewness to bias our findings. The means and medians of earnings and changes in earnings decrease across time, while the means and medians of analysts’ earnings forecast errors progressively increase. The number of observations tends to increase for all variables in the dataset from the beginning of the sample period until the year 2000, from which the number of observations decreases until the year 2020. Throughout the remainder of the paper, we follow Gilliam et al. (2015) and treat year 2002 as a transition year and refer to the period from fiscal year 1980 to 2001 as the pre-SOX period and the period from fiscal year 2003 to 2020 as the post-SOX period.

OPEN IN VIEWER

Table 1.

Descriptive Statistics.

Panel A: Annual financial statements
	Earnings scaled by assets			Δ earnings scaled by assets			Analysts’ forecast errors
Year	Obs	Mean	Median	Obs	Mean	Median	Obs	Mean	Median
1980	2,049	0.0692	0.0678	2,014	0.0054	0.0062	0	N/A	N/A
1981	3,641	0.0328	0.0588	3,384	−0.0051	0.0075	0	N/A	N/A
1982	4,606	−0.0077	0.0372	4,279	−0.0366	−0.0076	0	N/A	N/A
1983	4,892	−0.0235	0.0435	4,348	0.0006	0.0116	242	−0.0899	−0.0091
1984	4,970	−0.0336	0.0446	4,475	−0.0274	0.0112	1,125	−0.0256	0.0052
1985	4,973	−0.0449	0.0322	4,547	−0.0271	−0.0025	1,291	−0.0592	−0.0070
1986	5,300	−0.0943	0.0255	4,556	−0.0140	0.0040	1,482	−0.0333	−0.0157
1987	5,526	−0.0944	0.0275	4,913	−0.0145	0.0129	1,654	0.0286	0.0000
1988	5,441	−0.0735	0.0292	5,022	−0.0238	0.0117	1,896	0.0262	0.0126
1989	5,291	−0.0668	0.0250	4,954	−0.0089	0.0036	2,201	−0.0253	0.0141
1990	5,304	−0.0810	0.0200	4,902	−0.0248	0.0017	2,564	−0.0215	0.0000
1991	5,377	−0.0858	0.0159	4,959	−0.0265	0.0002	2,723	−0.0079	−0.0031
1992	5,564	−0.1042	0.0219	5,084	−0.0410	0.0071	2,766	−0.0045	0.0009
1993	5,965	−0.1029	0.0270	5,287	−0.0459	0.0088	3,116	−0.0037	0.0071
1994	6,233	−0.0938	0.0372	5,600	−0.0292	0.0150	3,491	0.0274	0.0152
1995	6,487	−0.1066	0.0344	5,797	−0.0192	0.0100	3,737	0.0190	0.0157
1996	7,295	−0.1733	0.0338	6,071	−0.0516	0.0098	4,060	0.0023	0.0138
1997	7,380	−0.1483	0.0291	6,780	−0.0765	0.0094	4,459	0.0515	0.0204
1998	7,103	−0.1931	0.0192	6,736	−0.0690	0.0027	4,646	0.0086	0.0205
1999	7,281	−0.4150	0.0135	6,404	−0.0709	0.0081	4,431	0.0281	0.0125
2000	7,117	−0.4329	0.0038	6,489	−0.2753	−0.0018	4,235	0.0237	0.0209
2001	6,801	−0.4108	−0.0192	6,353	−0.1315	−0.0130	3,880	0.0105	0.0101
2002	6,391	−0.4061	−0.0131	6,134	−0.0226	0.0102	3,520	0.0059	0.0087
2003	6,130	−0.4282	0.0094	5,819	−0.0333	0.0185	3,543	0.0550	0.0182
2004	5,956	−0.4606	0.0246	5,656	−0.1210	0.0165	3,563	0.0414	0.0250
2005	5,760	−0.4350	0.0241	5,459	−0.0997	0.0101	3,841	0.0359	0.0241
2006	5,554	−0.3887	0.0293	5,253	−0.1143	0.0086	4,041	0.0323	0.0254
2007	5,280	−0.3668	0.0230	5,041	−0.0861	0.0044	4,179	0.0348	0.0205
2008	5,064	−0.3402	0.0024	4,795	−0.0932	−0.0104	4,249	0.0294	0.0172
2009	4,987	−0.3691	0.0059	4,663	−0.0180	0.0041	3,978	0.0720	0.0088
2010	4,856	−0.3696	0.0257	4,562	−0.0124	0.0200	3,883	0.1177	0.0672
2011	4,791	−0.4394	0.0260	4,428	−0.0695	0.0084	3,749	0.0928	0.0440
2012	4,823	−0.4986	0.0183	4,395	−0.1034	0.0009	3,702	0.0210	0.0207
2013	5,094	−0.6615	0.0107	4,504	−0.1101	0.0031	3,723	0.0078	0.0159
2014	5,123	−0.6712	0.0089	4,696	−0.1575	0.0006	3,978	0.0358	0.0167
2015	4,882	−0.5791	0.0005	4,626	−0.1430	−0.0026	4,322	0.0431	0.0151
2016	4,715	−0.4788	0.0025	4,451	−0.0860	0.0025	4,315	0.0339	0.0154
2017	4,676	−0.5619	0.0068	4,353	−0.0879	0.0064	4,171	0.0363	0.0192
2018	4,609	−0.5177	0.0078	4,319	−0.1279	0.0044	4,089	0.0661	0.0262
2019	4,598	−0.4848	−0.0066	4,275	−0.1140	−0.0040	4,090	0.0223	0.0173
2020	4,726	−0.5224	−0.0308	4,302	−0.1153	−0.0093	4,176	0.0413	0.0179

Panel B: Quarterly financial statements
	Earnings scaled by assets			Δ earnings scaled by assets			Analysts’ forecast errors
Year	Obs	Mean	Median	Obs	Mean	Median	Obs	Mean	Median
1980	8,246	0.0183	0.0177	7,888	0.0007	0.0014	0	N/A	N/A
1981	8,043	0.0173	0.0166	7,849	0.0016	0.0022	0	N/A	N/A
1982	10,370	0.0031	0.0111	7,627	−0.0052	−0.0014	0	N/A	N/A
1983	16,267	−0.0043	0.0108	9,765	0.0009	0.0031	1	0.0095	0.0095
1984	19,080	−0.0083	0.0110	15,012	−0.0038	0.0029	1,686	−0.0591	0.0115
1985	19,486	−0.0103	0.0079	17,312	−0.0057	0.0000	2,649	−0.1254	−0.0207
1986	19,972	−0.0196	0.0064	17,799	−0.0020	0.0010	3,260	−0.1038	0.0000
1987	20,718	−0.0200	0.0074	18,192	0.0006	0.0031	3,407	−0.0625	0.0196
1988	21,337	−0.0185	0.0078	18,829	−0.0023	0.0029	4,123	−0.0928	0.0238
1989	20,964	−0.0182	0.0064	19,333	−0.0024	0.0010	5,497	−0.1425	0.0000
1990	20,742	−0.0207	0.0052	19,130	−0.0052	0.0005	5,905	−0.1732	0.0000
1991	20,274	−0.0217	0.0040	18,810	−0.0038	0.0001	6,143	−0.1397	0.0000
1992	20,471	−0.0257	0.0056	18,887	−0.0069	0.0020	7,005	−0.1419	0.0000
1993	21,674	−0.0236	0.0071	19,313	−0.0073	0.0022	7,084	−0.1293	0.0000
1994	23,008	−0.0221	0.0093	20,403	−0.0038	0.0033	8,912	−0.0991	0.0156
1995	24,045	−0.0227	0.0091	21,396	−0.0046	0.0025	9,400	−0.0957	0.0161
1996	25,146	−0.0336	0.0094	22,240	−0.0076	0.0025	9,925	−0.0663	0.0208
1997	26,437	−0.0351	0.0089	23,332	−0.0115	0.0026	10,811	−0.0461	0.0250
1998	26,700	−0.0411	0.0063	24,043	−0.0099	0.0014	11,041	−0.0748	0.0167
1999	26,564	−0.0651	0.0052	23,954	−0.0093	0.0018	10,535	−0.0365	0.0255
2000	26,717	−0.1031	0.0026	23,782	−0.0351	0.0006	9,908	−0.0570	0.0256
2001	26,900	−0.0956	−0.0035	23,777	−0.0253	−0.0016	8,759	−0.0552	0.0093
2002	25,495	−0.0993	−0.0013	24,131	0.0004	0.0030	9,106	−0.0026	0.0312
2003	24,081	−0.0976	0.0021	22,964	−0.0004	0.0032	8,831	−0.0156	0.0345
2004	22,926	−0.1182	0.0060	22,074	−0.0201	0.0038	9,570	0.0001	0.0455
2005	22,261	−0.1059	0.0062	21,020	−0.0208	0.0024	10,324	−0.0299	0.0379
2006	21,435	−0.1011	0.0073	20,319	−0.0193	0.0020	10,903	−0.0012	0.0410
2007	20,545	−0.0921	0.0059	19,388	−0.0229	0.0010	11,038	−0.0191	0.0380
2008	20,142	−0.0915	0.0023	18,716	−0.0193	−0.0011	10,985	0.0060	0.0484
2009	19,421	−0.0813	0.0015	18,292	−0.0010	−0.0002	10,488	0.0902	0.1111
2010	18,703	−0.0944	0.0063	17,843	−0.0003	0.0041	10,669	0.0695	0.0854
2011	18,434	−0.1078	0.0065	17,079	−0.0126	0.0022	10,284	0.0128	0.0526
2012	18,379	−0.1233	0.0049	16,878	−0.0198	0.0004	10,149	−0.0046	0.0400
2013	18,910	−0.1537	0.0033	16,973	−0.0152	0.0007	10,193	0.0091	0.0417
2014	19,130	−0.1663	0.0031	17,483	−0.0265	0.0006	10,981	0.0098	0.0361
2015	18,949	−0.1410	0.0008	17,411	−0.0257	−0.0002	11,421	0.0027	0.0401
2016	18,238	−0.1203	0.0007	17,086	−0.0137	0.0006	11,264	0.0091	0.0435
2017	17,678	−0.1272	0.0018	16,648	−0.0113	0.0012	10,904	0.0048	0.0426
2018	17,348	−0.1295	0.0024	16,272	−0.0181	0.0018	11,000	0.0204	0.0538
2019	17,276	−0.1208	0.0006	16,022	−0.0205	−0.0005	11,107	0.0030	0.0417
2020	17,152	−0.1149	−0.0038	16,002	−0.0218	−0.0016	11,281	0.0995	0.1273

Note. The table presents the number of observations (Obs), means (Mean), and medians (Median) of the variables used in the study. Panel A is based on annual financial statements and panel B is based on quarterly financial statements. Columns (2)–(4) report earnings scaled by lagged assets, columns (5)–(7) report changes in earnings scaled by lagged assets, and columns (8)–(10) report analyst consensus forecast errors, each for the years 1980 to 2020. “N/A” reflects that the value is not available due to the limited coverage in I/B/E/S.

Empirical Findings

The Zero-Earnings Discontinuity

Pre- and Post-SOX Periods

We investigate Hypothesis 1 by first examining the zero-earnings discontinuity graphically. As such, Figure 1 presents annual and quarterly earnings scaled by lagged assets for the post SOX period. The figure reveals a noticeable drop in the frequency of quarterly earnings just below zero compared to just above zero. A similar pattern is observed for annual earnings, although the effect is less pronounced. Consistent with our descriptive statistics, we find that the distributions are approximately normal though slightly left skewed. Based on the graphical inspection, we find that the zero-earnings discontinuity likely persists in quarterly earnings after the adoption of SOX. However, due to the smaller effect on annual earnings, we refrain from drawing inference about the persistence of the discontinuity in annual earnings.OPEN IN VIEWER

Figure 1.

Annual and quarterly earnings post-SOX. Note: The figure depicts the histograms of annual and quarterly earnings scaled by lagged assets around zero-earnings for the period 2003 to 2020. Observations in the range of [−0.15; 0.15] are grouped into intervals of size 0.0025 leading to 120 bins. The histogram colored in red displays annual earnings, while the histogram colored in blue displays quarterly earnings. The left hand axis shows the frequency of annual earnings, whereas the right hand axis shows the frequency of quarterly earnings. Quarterly earnings are multiplied by 4 to ensure comparability.

Figure 2 presents smoothed density functions around the zero-earnings discontinuity for earnings scaled by lagged assets, along with the corresponding confidence intervals. Panels A and B show the estimated density functions for annual earnings, while Panels C and D display the functions for quarterly earnings. Panels A and C present the estimated density functions using the pre-SOX period, and panels B and D show the post-SOX period. The zero-earnings discontinuity is present in both annual and quarterly earnings prior to the adoption of SOX, as shown by the nonoverlapping confidence intervals, which is consistent with prior studies (e.g., Burgstahler and Dichev (1997) and Brown and Caylor (2005)). Following the adoption, the zero-earnings discontinuity disappears in annual earnings, which is consistent with Gilliam et al. (2015). The discontinuity persists in quarterly earnings, which supports our prediction and is consistent with the graphical inspection in Figure 1. This is consistent with the explanation that companies continue to manage earnings around the zero-earnings threshold in quarterly financial statements following the adoption of SOX.OPEN IN VIEWER

Figure 2.

Estimated densities for annual and quarterly earnings pre- and post-SOX. (a) Panel A: Annual earnings pre-SOX. (b) Panel B: Annual earnings post-SOX. (c) Panel C: Quarterly earnings pre-SOX. (d) Panel D: Quarterly earnings post-SOX. Note: The figure visualizes the manipulation test developed by Cattaneo et al. (2020) using the companion rddensity Stata package (Cattaneo et al., 2018). A first order local polynomial is applied for point estimation, and a second order polynomial is used for bias correction, along with a triangular kernel and data-driven MSE-optimal bandwidth. The shaded areas around the lines depict the robust bias corrected 95% confidence intervals. The histogram colored in red displays negative earnings, while the histogram colored in blue displays positive earnings. The left hand axis shows the scaled frequency. Panels A and B display the density plot for annual earnings scaled by lagged assets in the pre- and post-SOX period, respectively. Panels C and D display the density plot for quarterly earnings scaled by lagged assets in the pre- and post-SOX period, respectively. In panel A (B), MSE-optimal bandwidths of 0.007 (0.007) and 0.006 (0.007) are applied to the left and right of zero, respectively, while in panel C (D), bandwidths of 0.007 (0.002) and 0.007 (0.001) are adopted.

Next, we conduct the BD and MT tests. We expect the BD test statistic for small loss (small profit) observations to be negative (positive), and the MT test statistics to be positive, documenting a lack (excess) of observations just below (above) zero relative to what is expected under continuity. Table 2 reports the findings from the pre- and post-SOX periods for annual and quarterly earnings, respectively. For annual earnings, the BD z-statistics for the zero-earnings discontinuity in the pre- and post-SOX periods are −15.71 and −2.46 for small loss observations, respectively, and 12.68 and 2.01 for small profit observations, respectively. This suggests that the zero-earnings discontinuity is significant in both periods for annual earnings with a p-value less than 1% and 5%, respectively, despite the adoption of SOX coinciding with a substantial decline in the discontinuity. These findings are in contrast to those by Gilliam et al. (2015). In additional tests (untabulated), we find that the inconsistent results are driven by our longer sampling period, which aligns with the findings of Degiannakis et al. (2022). When we restrict our sample to the period from 1980 to 2012, our findings are similar to Gilliam et al. (2015). We further explore the implications of this inconsistency in the following section. Since our findings are qualitatively similar to Gilliam et al. (2015) when using the same sampling period, we conclude that the zero-earnings discontinuity in annual earnings effectively disappeared after the adoption of SOX. We obtain consistent results using the MT test, with a significant z-statistic of 13.40 in the pre-SOX period and an insignificant z-statistic of 1.65 in the post-SOX period.

OPEN IN VIEWER

Table 2.

The BD and MT Tests for Annual and Quarterly Earnings Pre- and Post-SOX.

Period	Annual earnings			Quarterly earnings
	BD		MT	BD		MT
	Small loss	Small profit		Small loss	Small profit
Pre-SOX	−15.71***	12.68***	13.40***	−32.88***	28.44***	24.57***
Post-SOX	−2.46**	2.01**	1.65	−9.73***	11.09***	8.56***

Note. The table reports the z-statistics of the BD test and the t-statistic of the MT test using annual and quarterly earnings scaled by lagged assets. In the BD test, a bin width of 0.005 is applied for annual earnings and 0.00125 for quarterly earnings. For the MT test, a first order local polynomial is used for the point estimation and a second order one for bias correction, along with a triangular kernel, and data-driven MSE-optimal bandwidths. The asterisks denote significance ^*p < 0.10, ^**p < 0.05, ^***p < 0.01 using a two-tailed test.

For quarterly earnings, the BD z-statistics for the zero-earnings discontinuity in the pre- and post-SOX periods are −32.88 and −9.73 for small loss observations, respectively, and 28.44 and 11.09 for small profit observations, respectively. In all cases, the test statistics are significant with a p-value less than 1%, suggesting that the zero-earnings discontinuity exists in both the pre- and post-SOX periods, although being substantially smaller in the latter. Similar results are found when using the MT test with significant z-statistic of 24.57 and 8.56 in the pre- and post-SOX periods, respectively. The test statistics for the quarterly zero-earnings discontinuity are generally at least twice as large as those for the annual zero-earnings discontinuity both before and after the adoption of SOX. This suggests, as argued in Hypothesis 2, that management likely has greater reporting flexibility in quarterly earnings compared to annual earnings even before SOX was adopted.

The graphical insights and empirical tests provide evidence that the zero-earnings discontinuity persists in quarterly earnings following the adoption of SOX. This is consistent with our explanation that SOX had a weaker impact on earnings management in quarterly financial statements compared to annual statements.

Analyses by Year

To obtain additional insights into the persistence of the zero-earnings discontinuity and to shed further light on our inconsistent findings with Gilliam et al. (2015), we examine how the zero-earnings discontinuity evolves over time by employing both the BD and MT tests independently for each year. For quarterly earnings, we pool the four quarters together for each year to ensure comparability. ⁹

Table 3 reports our findings. The zero-earnings discontinuity in annual earnings generally disappears after the adoption of SOX, while it persists in quarterly earnings. The findings are robust towards both the BD and MT tests, with only a few years showing inconsistent results. Notably, a significant annual zero-earnings discontinuity is found in fiscal year 2016 (2017) using the BD (MT) test, with the remaining years being insignificant. This coincides with regulations on employee benefit plan reporting (FASB, 2015) and employee share-based payment accounting (FASB, 2016) coming into force. Although we cannot rule out that these drive the significant discontinuity in 2016 and 2017 in annual earnings, we find it unlikely that such regulations affects firms’ incentives and ability to meet and beat the zero-earnings benchmark. Furthermore, the discontinuity did not persist following 2016 and 2017, which, altogether, seems to suggest that the significant discontinuity in 2016 and 2017 is outliers. We expect that these two years, along with an aggregate effect of multiple years, explain the significant discontinuity in the annual earnings of Table 2. Interestingly, the zero-earnings discontinuity in annual earnings gradually disappears from fiscal year 1999, which can be explained by either the adoption being partially anticipated or that the adoption is not the only driving factor behind the disappearance.

OPEN IN VIEWER

Table 3.

The BD and MT Tests for Annual and Quarterly Earnings by Year.

Year	Annual earnings			Quarterly earnings
	BD		MT	BD		MT
	Small loss	Small profit		Small loss	Small profit
2020	0.32	0.98	0.42	0.06	1.75*	1.35
2019	−1.38	1.43	0.09	−1.09	0.73	0.50
2018	−1.58	1.07	0.48	0.15	1.320	−1.14
2017	−0.68	0.95	2.22**	−1.74*	2.24**	1.24
2016	−2.35**	2.10**	0.84	−1.02	2.92***	2.26**
2015	−1.51	0.89	0.67	−1.45	0.86	0.52
2014	0.25	−0.63	0.22	−0.59	2.11**	0.55
2013	−0.66	−0.19	0.51	−1.71*	2.35**	2.41**
2012	−0.83	1.40	1.32	−2.62***	2.37**	2.82***
2011	0.51	−0.23	−0.45	−2.85***	3.20***	2.37**
2010	−1.34	0.77	0.95	−3.17***	2.11**	1.65*
2009	−0.23	0.43	0.52	−3.38***	4.52***	4.44***
2008	0.80	−0.24	1.71*	−2.77***	1.92*	2.17**
2007	0.22	−0.39	−0.79	−3.50***	4.14***	4.51***
2006	0.58	−0.40	0.07	−2.92***	2.85***	3.31***
2005	−0.93	1.06	0.71	−2.73***	2.86***	2.54**
2004	1.09	−1.63	1.22	−4.74***	4.53***	4.96***
2003	−2.51**	1.06	1.07	−4.97***	3.64***	4.77***
2002	−2.00**	0.59	2.04**	−5.23***	4.04***	5.13***
2001	−2.06**	1.57	3.14***	−6.79***	5.70***	7.10***
2000	−1.28	0.37	1.89*	−5.69***	5.39***	7.05***
1999	−2.42**	1.25	2.14**	−6.84***	5.69***	6.08***
1998	−2.76***	3.41***	3.69***	−8.71***	6.54***	8.74***
1997	−3.60***	3.14***	3.23***	−8.86***	8.20***	9.42***
1996	−3.26***	2.77***	3.11***	−6.91***	6.30***	7.08***
1995	−4.25***	3.40***	2.79***	−7.85***	7.73***	9.54***
1994	−4.54***	3.92***	4.42***	−8.00***	6.54***	8.87***
1993	−2.67***	2.37**	3.26***	−7.09***	6.24***	6.95***
1992	−3.02***	3.11***	5.23***	−8.25***	8.08***	9.77***
1991	−5.61***	4.23***	5.79***	−8.18***	6.82***	7.70***
1990	−3.59***	3.26***	3.90***	−8.65***	7.32***	8.68***
1989	−4.08***	3.50***	4.19***	−6.79***	6.17***	7.26***
1988	−2.78***	1.56	2.00**	−7.79***	6.71***	7.61***
1987	−4.04***	2.80***	4.35***	−6.72***	5.33***	7.21***
1986	−5.26***	3.98***	5.20***	−7.90***	5.70***	6.96***
1985	−3.80***	3.34***	3.87***	−6.44***	6.52***	7.01***
1984	−2.74***	2.56**	3.64***	−6.13***	4.86***	6.24***
1983	−2.64***	1.47	2.45**	−6.52***	5.53***	6.83***
1982	−3.89***	2.76***	3.49***	−4.49***	3.59***	5.39***
1981	−4.30***	3.49***	4.58***	−4.49***	3.75***	5.66***
1980	−2.13**	0.570	1.20	−3.22***	2.27**	2.35**

Note. The table reports the z-statistics of the BD test as well as the t-statistic of the MT using annual and quarterly earnings scaled by lagged assets for each year in the sample period. A bin width of 0.005 for annual and 0.00125 for quarterly scaled earnings for the BD test, and a first order local polynomial for point estimation and a second order one for bias correction, a triangular kernel, and the data-driven MSE-optimal bandwidth is applied for the MT test. The asterisks denote significance ^*p < 0.10, ^**p < 0.05, ^***p < 0.01 using a two-tailed test.

The zero-earnings discontinuity in quarterly earnings persists until 2013, after which it disappears. To explore this further, we recreate Figure 2 in Figure 3 by comparing quarterly earnings from 2003 to 2013 and from 2014 to 2020. Confirming the findings of Table 3, we find that the zero-earnings discontinuity in quarterly earnings largely disappears after 2013. Several factors may contribute to this disappearance: It aligns with the explanation that SOX exhibited a delayed effect on the zero-earnings discontinuity in quarterly earnings (e.g., Chardonnens et al. (2022)). Alternatively, it may also be due to investors shifting their attention away from the zero-earnings threshold (e.g., Herrmann et al. (2011)) towards other thresholds (e.g., Bird et al. (2019) and de la Rosa and Lambertsen (2022)). Moreover, it could also be caused by the imposed regulation and increased scrutiny of financial reporting in the aftermath of the financial crisis. We do not conjecture what caused this decline due to the difficulties in attributing changes in the zero-earnings discontinuity to events (e.g., Gilliam et al. (2015)). ¹⁰ OPEN IN VIEWER

Figure 3.

Estimated densities for annual and quarterly earnings pre- and post-2013. (a) Panel A: Quarterly Earnings from 2003 to 2013. (b) Panel B: Quarterly Earnings from 2014 to 2020. Note: The figure visualizes the manipulation test developed by Cattaneo et al. (2020) using the companion rddensity Stata package. A first order local polynomial is applied for point estimation, and a second order polynomial is used for bias correction, along with a triangular kernel and data-driven MSE-optimal bandwidth. The shaded areas around the lines depict the robust bias corrected 95% confidence intervals. The histogram colored in red displays negative earnings, while the histogram colored in blue displays positive earnings. The left hand axis shows the scaled frequency. Panels A and B present the density plot for quarterly earnings scaled by lagged assets in the period from 2003 to 2013 and from 2014 to 2020, respectively, with bandwidth of 0.002 on either size of zero.

The year-by-year tests provide evidence consistent with our prediction that the adoption of SOX is associated with a weaker decline in earnings management in quarterly financial statements compared to annual financial statements. Our findings align with those of Enomoto and Yamaguchi (2017), who find that J-SOX weakened but did not eliminate the zero-earnings discontinuity in Japan, and Trimble (2018), who find the adoption of IFRS weakened but did not eliminate the zero-earnings discontinuity in both EU and non-EU countries. While our findings coincide with the adoption of SOX, establishing causal inference remains challenging as per prior studies (Gilliam et al., 2015). As a result, we caution the reader towards attributing the changes in the zero-earnings discontinuity solely to the adoption of SOX.

Relative Change in the Zero-Earnings Discontinuity

Table 4 reports the difference-in-difference estimates for Hypothesis 2. There are 15.88 (14.10) fewer (more) observations than expected under continuity for small losses (profits) in the post-SOX period for quarterly earnings relative to annual earnings. Similarly, the zero-earnings discontinuity is stronger in quarterly earnings compared to annual earnings in the post-SOX period by a test statistic of 1.33 (1.07) for small losses (profits). All coefficients are significant at less than the 1% level except the number of small profit observations, which is significant at the 5% level. This suggests that the zero-earnings discontinuity declined significantly less in magnitude for quarterly earnings compared to annual earnings. This is consistent with our prediction that SOX is likely less effective in restricting earnings management in quarterly earnings compared to annual earnings. The coefficients on the remaining variables align with our expectation. To improve comparability with Enomoto and Yamaguchi (2017) and to address the concern of potential extreme year observations, we extend our analysis and create, and interact, a time variable, measured as the fiscal year minus 2002, with the quarterly earnings indicator. Our findings (untabulated) remain unchanged further supporting the explanation that the adoption is associated with a weaker reduction in the quarterly zero-earnings discontinuity compared to the annual zero-earnings discontinuity. ¹¹

OPEN IN VIEWER

Table 4.

Relative Magnitude of the Discontinuity for Annual Compared to Quarterly Earnings.

	Unstandardized differences		Standardized differences
	(1)	(2)	(3)	(4)
	Small loss	Small profit	Small loss	Small profit
Q × post-SOX	−15.88*** (−2.63)	14.10** (2.16)	−1.33*** (−2.87)	1.07*** (2.48)
Post-SOX	35.89*** (6.95)	−34.62*** (−6.46)	3.07*** (7.37)	−2.40*** (−6.39)
Q	−0.93 (−0.24)	4.90 (1.03)	−0.09 (−0.32)	0.35 (1.23)
Constant	−39.39*** (-13.73)	37.22*** (9.95)	−3.34*** (-14.64)	2.58*** (10.64)
Observations	66	66	66	66
Adj. R2	0.568	0.491	0.642	0.542

Note. The table reports the regression estimates of the relative difference of the changes in the zero-earnings discontinuity in annual and quarterly earnings from 1980 to 2020. Q is an indicator for cases where the discontinuity is observed in quarterly earnings, and post-SOX is an indicator for cases where discontinuity is observed after the adoption of SOX. Columns (1) and (2) report the regression estimates using unstandardized differences as the dependent variable, whereas columns (3) and (4) report the estimates using standardized differences as the dependent variable. ^***, ^**, and ^* denote the two-tailed significance levels at the 1%, 5%, and 10%, respectively. We report heteroskedasticity robust t-statistics in parenthesis to control for nonconstant variances.

The validity of the difference-in-difference estimates relies on the assumption of parallel trends. To assess potential violations of this assumption, Figure 4 presents both discontinuity measures across time. Visual inspection reveals no meaningful pre-trends in either plot, although the discontinuity becomes larger in magnitude between 1980 and 1982. Since this is isolated to fiscal years with relatively few observations as evident from Table 1, we consider these years as outliers. To test the assumption, we interact the quarterly earnings discontinuity indicator, “Q,” with the continuous time variable, Time, and restrict the sample to the pre-SOX period. If a pre-trend exists prior to the adoption, the interaction term will likely be significant. Table 5 reports the findings. There is no significant pre-trend in the zero-earnings discontinuity prior to the adoption, except for a weak pre-trend for small standardized losses (p-value: 0.09) suggesting our findings are unlikely to be driven by existing trends in the discontinuities. ¹² OPEN IN VIEWER

Figure 4.

The magnitude of the discontinuity for annual and quarterly earnings across time. (a) Panel A: Small profit discontinuities, (b) Panel B: Small loss discontinuities. Note: The figure depicts the zero-earnings discontinuity in annual and quarterly earnings from 1980 to 2020. Panels A and B display the discontinuity for small profits and small losses, respectively. Each dot represents point estimates of the discontinuity across time, calculated as the difference between the number of small loss (profit) observations and the expected number of observations under continuity (i.e., the numerator in the BD test statistic). The vertical bar shows the corresponding confidence intervals for each point estimate, calculated using the standard error of each point estimate (i.e., the denominator in the BD test statistic). The red dots represent annual earnings, while the blue dots represent quarterly earnings.

OPEN IN VIEWER

Table 5.

Pre-Trends in the Discontinuity for Annual and Quarterly Earnings.

	Unstandardized differences		Standardized differences
	(1)	(2)	(3)	(4)
	Small loss	Small profit	Small loss	Small profit
Q × time	−0.98 (−1.57)	1.05 (1.22)	−0.09* (−1.75)	0.07 (1.26)
Q	−12.89* (−1.88)	17.27* (1.73)	−1.17** (−2.15)	1.14* (1.80)
Time	0.46 (0.84)	−0.45 (−0.58)	0.074 (1.67)	−0.05 (−1.00)
Constant	−36.86*** (−6.11)	35.75*** (3.90)	−2.65*** (−5.69)	2.23*** (3.84)
Observations	38	38	38	38
Adj. R2	0.093	0.114	0.157	0.140

Note. The table reports the regression estimates of the difference in trends between annual and quarterly earnings discontinuities from 1983 to 2001. Observations for fiscal year 1980 to 1982 are excluded as they are suspected outliers. Q is an indicator for cases where the discontinuity is observed in quarterly earnings and Time measures fiscal year minus 2002. Columns (1) and (2) report the regression estimates using unstandardized differences as the dependent variable, whereas columns (3) and (4) report the estimates using standardized differences as the dependent variable. ^***, ^**, and ^* denote the two-tailed significance levels at the 1%, 5%, and 10%, respectively. We report heteroskedasticity robust t-statistics in parenthesis to control for nonconstant variances.

Changes in Earnings

We next investigate the discontinuity in the distribution of changes in earnings around zero changes. Figure 5 presents changes in annual and quarterly earnings scaled by lagged assets in the post-SOX period. The figure reveals that the noticeable drop in the frequency of changes in quarterly earnings persists, while it largely disappears for changes in annual earnings. Consistent with prior studies (e.g., Burgstahler and Dichev (1997)), the magnitude of the discontinuity appears smaller compared to that of earnings scaled by lagged assets. Based on the graphical inspection, the zero-earnings discontinuity also persists in quarterly earnings following the adoption of SOX.OPEN IN VIEWER

Figure 5.

Changes in annual and quarterly earnings post-SOX. Note: The figure depicts the histograms of changes in annual and quarterly earnings scaled by lagged assets around zero changes in earnings from 2003 to 2020. Observations in the range of [−0.15; 0.15] are presented and grouped into intervals of size 0.0025 leading to 120 bins. The histogram colored in red displays changes in annual earnings, while the histogram colored in blue displays changes in quarterly earnings. The left hand side axis represents the frequency of changes in annual earnings, whereas the right hand side axis represents the frequency of changes in quarterly earnings. Quarterly earnings are multiplied by 4 to ensure comparability.

Next, we conduct the BD and MT tests on changes in earnings around the adoption of SOX. Table 6 reports the findings from the pre- and post-SOX periods. For changes in annual earnings, the BD z-statistics for the discontinuity around zero in the pre- and post-SOX periods are −5.85 and −1.16 for small negative changes, respectively, and 7.80 and 3.34 for small positive changes, respectively. These results suggest, similarly to our findings using earnings scaled by lagged assets, that the discontinuity for changes in earnings disappears in annual financial statements following the adoption. These results are largely consistent with the MT test, which yields a test statistic of 7.80 for changes in annual earnings prior to the adoption. We find weak evidence that the discontinuity for changes in annual earnings persists in the post-SOX period with a test statistic of 2.11 (p-value: 0.04). In additional analysis (untabulated), we find that the persistence is largely driven by the large number of small positive changes as well as the fiscal years 2011 and 2012. Excluding these fiscal years results in the test statistic becoming borderline significant (p-value: 0.05) at conventional levels.

OPEN IN VIEWER

Table 6.

The BD and MT Tests for Changes in Annual and Quarterly Earnings Pre- and Post-SOX.

Period	Annual earnings			Quarterly earnings
	BD		MT	BD		MT
	Small negative Δ	Small positive Δ		Small negative Δ	Small positive Δ
Pre-SOX	−5.85***	7.80***	7.80***	−13.83***	19.96***	13.83***
Post-SOX	−1.16	3.34***	2.11**	−2.78**	7.27***	4.36***

Note. The table reports the z-statistics of the BD test and the t-statistic of the MT test using changes in annual and quarterly earnings scaled by lagged assets. In the BD test, a bin width of 0.005 is applied for annual earnings and 0.00125 for quarterly earnings. For the MT test, a first order local polynomial is used for the point estimation and a second order one for bias correction, along with a triangular kernel, and data-driven MSE-optimal bandwidths. The asterisks denote significance ^*p < 0.10, ^**p < 0.05, ^***p < 0.01 using a two-tailed test.

For changes in quarterly earnings, the BD z-statistics for the discontinuity around zero in the pre- and post-SOX periods are −13.83 and −2.78 for small negative changes, respectively, and 19.96 and 7.27 for small positive changes, respectively. These findings align with those from the MT test, which yields z-statistics in the pre- and post-SOX periods of 13.83 and 4.36, respectively. This suggests, similarly to our findings using earnings scaled by lagged assets, that the discontinuity for changes in quarterly earnings persists after the adoption of SOX, whereas it generally disappears for changes in annual earnings. The discontinuities for changes in earnings are generally smaller in magnitude compared to those in earnings scaled by lagged assets, which is consistent with prior research (e.g., Burgstahler and Dichev (1997)). This is likely due to changes in earnings generally being smaller and more volatile than earnings in levels.

To obtain additional insights, we lastly examine how the discontinuity around zero changes in earnings evolves over time. We employ the BD and MT tests independently for each year and pool the four quarters together to ensure comparability. Due to the discontinuity being smaller in magnitude compared to annual earnings scaled by lagged assets (Burgstahler & Dichev, 1997), we expect the findings to be weaker although qualitatively similar compared to when we use earnings scaled by lagged assets. We find (untabulated) that the discontinuity for changes in annual earnings generally exists prior to the adoption of SOX, after which it disappears. For changes in quarterly earnings, the discontinuity persists after the adoption for small positive changes. This aligns with our prior findings, suggesting that firms are likely to manage quarterly earnings to meet and beat the previously reported quarterly earnings after the adoption of SOX. Moreover, we find that the discontinuity for changes in quarterly earnings largely disappears around 2013, consistent with our findings using earnings scaled by lagged assets. Altogether, the findings using changes in earnings are consistent with our results using earnings scaled by lagged assets and support our explanation that the discontinuity persists in quarterly earnings likely due to the absence of an auditor in ensuring compliance with SOX.

Analysts’ Earnings Forecast Errors

We next investigate the discontinuity in analysts’ earnings forecast errors. Figure 6 presents the annual and quarterly earnings forecast errors in the post-SOX period. Similarly to earnings and changes in earnings, the number of small negative (positive) forecast errors is lower (higher) than what would be expected under continuity. Hence, the discontinuity around zero appears to persist, similar to the findings for earnings and changes in earnings.OPEN IN VIEWER

Figure 6.

Analysts’ annual and quarterly earnings forecasts errors post-SOX. Note: The figure depicts the histograms of analysts’ annual and quarterly earnings forecast errors around zero from 2003 to 2020. Observations in the range of [−0.15; 0.15] are presented and grouped into intervals of size 0.0025 leading to 120 bins. The histogram colored in red displays annual earnings forecast errors, while the histogram colored in blue displays quarterly earnings forecast errors. The left hand side axis represents the frequency of annual earnings forecast errors, whereas the right hand side axis represents the frequency of quarterly earnings forecast errors. Quarterly earnings forecast errors are divided by 4 to ensure comparability.

We next conduct BD and MT tests on analysts’ earnings forecast errors. Table 7 reports the findings from the pre- and post-SOX periods. For annual earnings forecast errors, the BD z-statistics for the discontinuity around zero in the pre- and post-SOX periods are −0.47 and 0.07 for small negative forecast errors, respectively, and 7.89 and 6.00 for small positive forecast errors, respectively. Similarly, the MT t-statistic is 5.64 and 4.53 in the pre- and post-SOX periods, respectively. These results suggest that, in contrast to the discontinuity in earnings and changes in earnings, the discontinuity for annual earnings forecast errors persists for small positive errors following the adoption. For quarterly earnings forecast errors, the BD z-statistics for the discontinuity around zero in the pre- and post-SOX periods are 25.66 and 16.29 for small negative forecast errors, respectively, and 37.42 and 34.69 for small positive forecast errors, respectively. Likewise, the MT t-statistic is 13.27 and 15.86 in the pre- and post-SOX periods, respectively. These results suggest, similarly to the discontinuity in earnings and changes in earnings, that the discontinuity in quarterly earnings forecast errors persists for both small negative errors and small positive errors. Consistently with prior findings, the coefficients on quarterly forecast errors are substantially larger than those on annual forecast errors, supporting the explanation that firms possess greater flexibility in managing quarterly earnings to meet and beat analysts’ earnings forecasts. ¹³ Altogether, the findings using analysts’ quarterly earnings forecast errors are consistent with our prior analysis, while the findings using analysts’ annual earnings forecasts are not. This seems to suggest that the disappearance of the zero-earnings discontinuity has little impact on the incentive and ability by firms to meet and beat analysts’ annual earnings forecasts.

OPEN IN VIEWER

Table 7.

The BD and MT Tests for Analysts’ Annual and Quarterly Earnings Forecast Errors Pre- and Post-SOX.

Period	Annual earnings			Quarterly earnings
	BD		MT	BD		MT
	Small negative	Small positive		Small negative	Small positive
	Forecast errors	Forecast errors		Forecast errors	Forecast errors
Pre-SOX	−0.47	7.89***	5.64***	25.66***	37.42***	13.27***
Post-SOX	0.07	6.00***	4.53***	16.29***	34.69***	15.86***

Note. The table reports the z-statistics of the BD test as well as the t-statistic of the MT test using annual and quarterly earnings per share (EPS) forecast errors in the pre-SOX and post-SOX periods. In the BD test, bin width of 0.005 for annual and 0.00125 for quarterly earnings are applied. For the MT test, a first order local polynomial for point estimation and a second order one for bias correction, a triangular kernel, and the data-driven MSE-optimal bandwidth are applied. The asterisks denote significance ^*p < 0.10, ^**p < 0.05, ^***p < 0.01 using a two-tailed test.

Robustness Tests

Cross-Sectional Tests and Measurement

To address the concerns that our results might be influenced by methodological choices, we conduct a battery of sensitivity tests.

Prior studies raise concerns that firm size may drive the zero-earnings discontinuity, as large firms generally have large followings of analysts (Dempsey, 1989). This may influence investors’ attention devoted to the given company and, in turn, impact the firm’s incentive to meet and beat the zero-earnings benchmark as well as analysts’ earnings forecasts. To address this concern, we partition our dataset into two subsamples comprising firms with lagged assets below and above the median, and rerun the discontinuity tests for the period from 1980 to 2001 and the period from 2003 to 2013 for each subsample individually, respectively. To investigate the disappearance of the quarterly zero-earnings discontinuity around 2013, we extend the analysis into the period from 2014 to 2020. Table 8 reports the findings. There is virtually no difference in the discontinuity in the period from 1980 to 2001 between small and large firms for both annual and quarterly earnings. In the period from 2003 to 2012, the discontinuity is weakly significant for small firms in annual earnings, while the discontinuity in quarterly earnings remains highly significant for both large and small firms. Moreover, large firms appear to experience a greater decline in the quarterly zero-earnings discontinuity compared to small firms. These results suggest that the disappearance of the zero-earnings discontinuity in annual earnings is driven more by large firms than by small firms. This is likely to be the case if large firms experience a higher probability of being subject to investigations following the adoption of SOX, and hence, may be relatively more restricted in performing earnings management compared to small companies. For the period from 2013 to 2020, the quarterly zero-earnings discontinuity disappears for large firms. This aligns with the explanation that the disappearance around 2013 may be attributed to regulatory interventions, such as increased scrutiny following the financial crisis, as large firms are more likely to face such scrutiny. We refrain from drawing conclusions due to the challenges associated with causal inference.

OPEN IN VIEWER

Table 8.

The BD and MT Tests for Annual and Quarterly Earnings for Small and Large Firms Pre- and Post-SOX.

Period	Annual normalized earnings			Quarterly normalized earnings
	BD		MT	BD		MT
	Small loss	Small profit		Small loss	Small profit
1980–2001
Small firms	−11.10***	9.61***	11.66***	−5.96***	6.54***	20.62***
Large firms	−11.25***	8.52***	10.16***	−32.51***	27.71***	16.29***
2003–2013
Small firms	−1.83*	1.51	2.34**	−2.44**	3.85***	10.58***
Large firms	0.05	−0.48	−0.16	−10.46***	9.80***	5.51***
2014–2020
Small firms	−1.76*	1.95*	0.43	−3.06***	5.90***	2.35**
Large firms	−1.94*	1.82*	2.31**	−0.49	1.50	0.94

Note. The table reports the z-statistics of the BD test and the t-statistic of the MT test using earnings scaled by lagged assets for subsamples of small and large firms. Small firms possess lagged total assets below the median in the sample, while large firms possess lagged total assets above the median in the sample. In the BD test, a bin width of 0.005 is applied for annual earnings and 0.00125 for quarterly earnings. For the MT test, a first order local polynomial is used for the point estimation and a second order one for bias correction, along with a triangular kernel, and data-driven MSE-optimal bandwidths. The asterisks denote significance ^*p < 0.10, ^**p < 0.05, ^***p < 0.01 using a two-tailed test.

To further address the concern that firm size influences our findings, we, following the recommendation by Burgstahler and Chuk (2013), investigate the zero-earnings discontinuity in unscaled earnings. We partition firms in the pre- and post-SOX period, respectively, into four quantile subsamples based on their total assets, with firms exhibiting the smallest (largest) total assets into the first (fourth) subsample. We restrict the sample to firm-year observations between 1991 and 2013 to eliminate the effects of the disappearance of the zero-earnings discontinuity after 2013 and to ensure the number of fiscal years pre- and post-SOX is identical following the literature (e.g., Gilliam et al. (2015)). Consistent with their recommendation, we adopt progressively larger interval widths for firms in larger subsamples. Following Gilliam et al. (2015), we adopt an earnings interval width of $80,000 ($120,000), $350,000 ($650,000), and $2,000,000 ($3,000,000) for annual unscaled earnings and one fourth of that for quarterly earnings in the pre- (post-)SOX period. We find (untabulated), consistent with the prior findings, that the zero-earnings discontinuity disappears in annual earnings after the adoption of SOX, whereas it persists in quarterly earnings throughout all subsamples.

The choice of scaling has likewise been debated, with lagged market capitalization as an alternative to lagged assets. To address this, we rerun our tests using earnings scaled by lagged market capitalization. Table 9 reports the findings. We find virtually no difference in the test statistics when using lagged market capitalization as a scaling factor, although the coefficient on small profit in quarterly earnings becomes insignificant in the post-SOX period. Further analysis (untabulated) reveals that this is driven by fiscal year 2013. Interestingly, the coefficient on the test statistic of small losses and small profits changes in the period from 2014 to 2020. Additional tests (untabulated) reveal that this is driven by fiscal year 2018. Similarly, Table 10 reports the regression estimates of the difference-in-difference design using annual and quarterly earnings scaled by lagged market capitalization. Consistently, we also find no differences when adopting market capitalization as the scaling factor, although there is only a weak decline in the discontinuity for small losses.

OPEN IN VIEWER

Table 9.

The BD and MT Tests for Annual and Quarterly Earnings Scaled by Lagged Market Capitalization Pre- and Post-SOX.

Panel A: Earnings
Period	Annual normalized earnings			Quarterly normalized earnings
	BD		MT	BD		MT
	Small loss	Small profit		Small loss	Small profit
Pre-SOX	−11.27***	9.85***	9.62***	−21.32***	17.84***	16.60***
Post-SOX	−1.29	1.17	0.99	−2.54***	5.06***	3.49***

Panel B: Changes in earnings
Period	Annual normalized earnings			Quarterly normalized earnings
	BD		MT	BD		MT
	Small loss	Small profit		Small loss	Small profit
Pre-SOX	5.64***	5.64***	5.65***	−14.03***	15.11***	12.74***
Post-SOX	−1.37	3.06***	0.46	−1.86*	7.72***	2.55**

Note. The table reports the z-statistics of the BD test as well as the t-statistic of the MT test using annual and quarterly earnings scaled by lagged market capitalization. Panel A shows the test results using earnings scaled by lagged market capitalization and panel B shows the test results using changes in earnings scaled by lagged market capitalization. In the BD test, bin width of 0.005 for annual and 0.00125 for quarterly earnings are applied. For the MT test, a first order local polynomial for point estimation and a second order one for bias correction, a triangular kernel, and the data-driven MSE-optimal bandwidth are applied. The asterisks denote significance ^*p < 0.10, ^**p < 0.05, ^***p < 0.01 using a two-tailed test.

OPEN IN VIEWER

Table 10.

Relative Magnitude of the Discontinuity for Annual Compared to Quarterly Earnings Scaled by Lagged Market Capitalization.

	Unstandardized differences		Standardized differences
	(1)	(2)	(3)	(4)
	Small loss	Small profit	Small loss	Small profit
Q × post-SOX	−5.92 (−1.38)	10.91** (2.34)	−0.41 (−0.94)	0.83** (2.08)
Post-SOX	17.33*** (4.96)	−19.22*** (−5.50)	1.95*** (5.67)	−1.76*** (−5.97)
Q	5.53* (1.92)	−5.58* (−1.79)	0.31 (1.04)	−0.27 (−1.09)
Constant	−22.78*** (−9.23)	22.72*** (8.62)	−2.46*** (−9.84)	2.03*** (9.54)
Observations	66	66	66	66
Adj. R2	0.388	0.345	0.448	0.404

Note. The table reports the regression estimates of the relative difference of the changes in the zero-earnings discontinuity in annual and quarterly earnings from 1980 to 2020. Q is an indicator for cases where the discontinuity is observed in quarterly earnings and post-SOX is an indicator for cases where discontinuity is observed after the adoption of SOX. Columns (1) and (2) report the regression estimates using unstandardized differences as the dependent variable, whereas columns (3) and (4) report the estimates using standardized differences as the dependent variable. Both the unstandardized and standardized differences are estimated using earnings scaled by lagged market capitalization. ^***, ^**, and ^* denote the two-tailed significance levels at the 1%, 5%, and 10%, respectively. We report heteroskedasticity robust t-statistics in parenthesis to control for nonconstant variances.

Conclusion

We investigate the impact of the Sarbanes-Oxley Act in 2002 on the zero-earnings discontinuity in annual and quarterly earnings. Prior literature finds that the zero-earnings discontinuity in annual earnings disappears following the adoption. We hypothesize that the disappearance may not generalize to quarterly earnings, as quarterly financial statements are unaudited, and hence, lack an auditor to verify whether the firms comply with SOX.

Consistent with this, we find that the zero-earnings discontinuity in quarterly earnings persists after the adoption of the Act. The persistence is driven by SOX coinciding with a reduction in the zero-earnings discontinuity that is smaller in magnitude for quarterly earnings compared to that for annual earnings. This fits with our explanation that SOX is less able to restrict earnings management in quarterly earnings compared to annual earnings. The findings are robust to a battery of tests including commonly raised concerns in the literature as to whether the discontinuity is artificially driven.

Our findings are consistent with the explanation that the adoption of SOX is less able to restrict earnings management in quarterly financial statements compared to annual financial statements due to the absence of an auditor to ensure compliance. However, we do not claim causal inference and caution the readers to attribute our findings solely to SOX, as many factors are likely to impact the discontinuity around the adoption of SOX (Gilliam et al., 2015).