Financial misconduct and employee mistreatment: Evidence from wage theft

Wage theft arises from actions taken by firms to deny employees their legally-mandated pay or benefits. Most wage theft represents violations of the Fair Labor Standards Act (FLSA). Common examples of such actions include forcing employees to underreport their hours worked, not paying for overtime work by misclassifying workers as exempt from overtime requirements, forcing employees to work through legally-mandated paid breaks, and paying workers less than the minimum wage. These actions represent direct financial savings to the firm, and in many cases these savings exceed the direct amount of underpaid wages. For instance, employees at large firms who work 30 or more hours per week are entitled to several employer-provided benefits (e.g., healthcare). If a firm were to force an employee to report that she worked 29 hours per week rather than 30, the firm would save both an hour’s wages and the cost of these benefits.

The Wage & Hour Division (WHD), an office within the US Department of Labor, enforces compliance with laws governing employees’ pay and benefits.⁶ WHD was formed as part of the enactment of the original Fair Labor Standards Act (FLSA) of 1938. The last major change in US federal wage and hour laws was in 2004, when the FLSA was overhauled to reclassify the types of workers subject to overtime pay requirements.⁷ Although WHD, like other Department of Labor agencies (e.g., the Occupational Safety & Health Administration), is not a Cabinet-level office, it operates with a high degree of autonomy.

As part of the enforcement process WHD conducts numerous on-site visits to firms’ establishments to conduct inspections. Many of these inspections are in response to workers’ complaints. Because not all victims of wage theft complain (or even know that they have been victimized), however, WHD also conducts a number of random audits each year. Although WHD does not disclose what share of its investigations arise from random audits versus employee complaints, in a 2011 Congressional hearing WHD indicated that the majority of its investigations arose from random audits.⁸ Visits are typically not preannounced to the firm, to minimize the likelihood that a firm targeted for inspection can proactively cover up evidence of wrongdoing before the inspector arrives. Firms’ uncertainty about whether they will be investigated also serves as a mechanism to deter wage theft (Weil 2010). This uncertainty is enabled by the relative frequency with which WHD conducts investigations (more than 20,000 per year on average). As with all forms of misconduct, I cannot observe wage theft that WHD did not catch; nonetheless, because of the higher frequency and lower costs of conducting investigations, it is unlikely that a firm that engages in a nontrivial amount of wage theft will have no detected violations.

Based on WHD’s investigations, if a firm is found to be noncompliant, it will be required to provide back pay to employees as well as possible civil penalties to WHD. These penalties vary in magnitude, depending on both the severity of the violation and the firm’s compliance history (i.e., conditional on committing the same type of violation, repeat offenders will pay more than first-time offenders). WHD publicly discloses enforcement actions on its website. Disclosure creates an indirect cost of noncompliance by enabling investors and news outlets to learn of corporate wage theft.⁹ Although difficult to precisely measure, this cost can be nontrivial. For example, Johnson (2020) finds that the decision by another Department of Labor office, the Occupational Safety and Health Administration, to publicly name-and-shame violators in certain states led to a decrease in subsequent workplace safety violations in those states. Johnson (2020) identifies the media as a key channel that drives his results; two-thirds of OSHA press releases in his sample were picked up by at least one newspaper. In addition, public disclosure of enforcement actions can provide employees with information that enables them to bring class-action lawsuits for wage theft (where this is not prohibited by arbitration clauses). These lawsuits serve as another significant potential cost of misconduct, as the average employee class action lawsuit settlement exceeds $10 million on average (Li and Raghunandan 2021). Thus, although the direct penalties for wage theft are small, indirect costs can serve as a deterrent.

Several existing studies (e.g., Kedia and Philippon 2009; Call et al. 2017) relate firms’ financial incentives and misconduct to compensation practices, finding that employees command a wage premium and are awarded more options during the misconduct period. As a result of these findings, I would expect that firms are less likely to engage in wage theft when they engage in financial misconduct. However, these studies focus on executive compensation or employee stock options, due to the availability of data on these measures; as a result, they consider only salaried employees, who are exempt from most wage and hour regulations. An important reason to overcompensate employees during fraud periods is to deter whistleblowing, because employees who hold equity know that they will suffer if, due to fraud being detected, the value of the company’s stock decreases. Firms are unlikely to have concerns about hourly-wage workers blowing the whistle on financial misconduct, however. Thus, although my study relates to existing research on the employment consequences of financial fraud, the mechanism underlying the conclusions of prior studies is unlikely to apply in the wage theft setting.

Recent research documents direct consequences, to employees, of their employers’ financial incentives. These studies focus on workplace safety. For example, employees suffer injuries attributable to unsafe working conditions more frequently when their employers are unexpectedly cash-constrained (Cohn and Wardlaw 2016) or are under pressure to meet earnings benchmarks (Caskey and Ozel 2017). While workplace injuries do not necessarily reflect outright misconduct, the underlying actions (e.g., reducing maintenance) that result in higher injury rates are similar to those that, if taken further, result in labor law violations.¹⁰ I contribute to this literature by focusing on another key labor-related consequence of firms’ financial reporting incentives: rank-and-file employees’ compensation. To do so, I focus on meet-or-beat situations as a measure of financial incentives. Based on these findings from prior literature, I hypothesize the following:

Hypothesis 1 does not directly imply a relation between wage theft and financial misconduct. An extensive body of literature documents the financial incentives that underpin financial misconduct but, depending on the way in which the costs of wage theft and financial misconduct are related, firms may either view the two as complements or substitutes. To develop predictions of this relation I draw on prior literature that links real and accrual-based decisions. This literature finds a negative and/or substitutive relation between the two (e.g., Roychowdhury 2006; Cohen et al. 2008; Zang 2012), especially in the period subsequent to the 2002 Sarbanes-Oxley Act. Given that wage theft represents a form of real earnings management gone too far while financial misconduct represents accrual earnings management gone too far, I expect the following:

There are reasons why Hypothesis 2 might not hold. Prior studies on real earnings management rely on the assumption that all firms in the same industry have the same cost structure and competitive strategy. Srivastava (2019) argues that this assumption means that proxies for real earnings management used in prior literature are misspecified and hence it is difficult to draw conclusions from measures of manipulative (but legal) behavior that rely on differences from industry peers; heterogeneity in such measures often reflects heterogeneous competitive strategies. For example, while a firm may cut discretionary costs in response to earnings targets, it could just as easily have done so as part of a longer-term cost-cutting strategy if it believed itself to be bloated relative to peers. I sidestep this limitation because wage theft is illegal and, as a result, cannot be the basis of a firm’s (legally permissible) competitive strategy. However, the illegality of wage theft also provides a reason why Hypothesis 2 might not hold. Certain costs may realize only if a firm engages in illegal activity (but not in legal manipulation). These costs include litigation risk and the potential for reputational damage. For example, Li and Raghunandan (2021) show that firms caught engaging in labor-related misconduct are more likely to subsequently face labor lawsuits.¹¹ If such costs are nontrivial, then the overall costs of the two types of misconduct will be positively correlated and, as a result, the two types of misconduct may instead be positively associated.

I begin my empirical analyses by first testing whether wage theft is financially motivated. To do so I follow prior literature on the operational consequences of financial incentives and study the effect of meet-or-beat incentives on the likelihood that firms engage in wage theft. Because my underlying data primarily consists of large firms, I focus on meet-or-beat behavior with respect to analyst forecasts. Moreover, Gilliam et al. (2015) show that the zero-earnings discontinuity and the discontinuity around prior-year earnings have largely disappeared since the 2002 Sarbanes-Oxley Act. Following Caskey and Ozel (2017), I define an indicator, SUSPECT_it, that equals one for firm-years that just meet or beat the consensus analyst forecast by between zero and two cents per share. I then estimate the following linear probability model of wage theft:

Because violations may systematically vary as a function of cross-sectional characteristics, I include firm fixed effects, γ_i. I also include year fixed effects, 𝜃_t, to control for time-variant changes in either the incentives for wage theft or the strength of enforcement. In this and all subsequent estimations, I cluster standard errors by firm.

In Eq. 1, WageTheft_it is an indicator for whether wage theft occurred in year t (and was ultimately detected, whether in year t or in a later year). I use the indicator as my main wage theft measure because Weil (2010) highlights that while WHD compliance action data provide a relatively accurate depiction of which firms are engaging in wage theft as well as the ability to assess firms relative to one another, the inability to inspect every establishment makes it difficult to quantify the exact amount of wage theft that a given firm is engaging in.

One limitation of the indicator-based approach to constructing WageTheft_it is that an indicator may not properly account for one-off incidents that may not reflect the firm’s behavior more generally. I therefore construct three alternative measures of WageTheft_it to account for the severity and frequency of wage theft. These are (i) the natural logarithm of the dollar value of penalties assessed for wage theft that occurred in year t; (ii) the natural logarithm of the number of distinct sites at which violations occurred during year t, irrespective of the severity in each location; and (iii) the natural logarithm of the per-affected-employee dollar value of penalties. Underlying proxies (i) and (ii) is the assumption that more severe forms of wage theft are less likely to be random.¹² Proxy (iii) accounts for the potential concern that proxies (i) and (ii) may systematically vary with firm size. While this concern is mitigated by the fact that my sample comprises only large public firms and by my use of firm fixed effects, I nonetheless measure wage theft using this alternative measure to ensure that my results are not driven by other firm characteristics. For each of these continuous variables, I use the natural logarithm of the underlying construct to mitigate the potential effects of large outliers.

While I observe the start and end dates of wage theft, I do not observe a year-by-year breakdown of the violation amount. The mean (median) wage violation in my sample lasts 817 (729) days, so in constructing proxies (i) and (iii) above, it is necessary to allocate the penalty amount over time to gauge the severity of the wage theft the firm committed in year t. For example, a firm that pays $50,000 for a single violation that lasts two years has likely engaged in less overall wage theft than a firm that pays $30,000 twice (i.e., $60,000 total) for two separate one-year violations. In addition, the underlying WHD data are at the violation level. If a firm engages in multiple instances of wage theft in a given year, I must aggregate these instances to the firm-year level to conduct my analyses. However, violation dates might not perfectly overlap. For example, a firm might engage in wage theft in one location in 2007 and 2008, and in a second location in 2008 alone. To account for these scenarios, I first attribute the total penalty for each individual violation evenly across the years in which that violation was committed, then aggregate to the firm-year level. For example, if a firm engages in wage theft in 2007 and 2008 and ultimately has to pay $60,000 in back pay and civil penalties, I allocate \(\frac {\$60,000}{2} = \$30,000\) to each of 2007 and 2008. If that firm also engaged in a separate instance of wage theft in 2008, for which it ultimately had to pay $20,000, then the total penalty allocated to 2007 would be $30,000 and the total penalty allocated to 2008 would be $50,000.

I obtain wage theft data from Good Jobs First’s Violation Tracker database.¹³ Although Violation Tracker does not contain exact violation start and end dates, I am able to obtain these dates directly from WHD’s WHISARD database. I also obtain, from WHISARD, the number of employees affected by each violation. Good Jobs First provides parent-subsidiary matching for roughly the largest 3,000 firms in the United States by size, although some of these are private companies or subsidiaries of parent companies traded abroad. To ensure that I do not erroneously mislabel unmatched violation firms as non-violation firms, I limit the sample to firms with at least one observation in either Violation Tracker (regardless of whether the firm has any incidences of wage theft specifically) or Good Jobs First’s other main firm-level database, Subsidy Tracker.¹⁴ Approximately 27% of the sample reflects firms that do not have any observations in Violation Tracker (whether for wage theft or for one of the numerous other types of violations covered in Violation Tracker) but are covered in Subsidy Tracker. Because Good Jobs First has performed parent-subsidiary matching, these firms are likely “true zeros” in the sense that their exclusion from Violation Tracker means that they genuinely do not have any violations for any agencies covered by Violation Tracker from 2000 onward. All results in the paper are robust to excluding these firms. Because the combined Violation Tracker - Subsidy Tracker universe primarily comprises large firms, imposing this sample restriction may limit external validity to the extent that the relation between financial reporting incentives and wage theft may differ in small firms. However, the inclusion of firms that appear in Subsidy Tracker mitigates potential biases in coefficients on variables of interest within the set of large firms, thereby improving internal validity with respect to understanding the behavior of large public firms.

I include two proxies for the expected costs of wage theft in Eq. 1. The first is the percentage of employees in a firm’s 2-digit NAICS industry that are unionized, obtained from UnionStats.¹⁵ The second is the industry-year violation rate, which I measure as the fraction of firms in the same 2-digit NAICS industry that received sanctions for wage theft in year t (exclusive of the firm itself). This variable picks up changes in both industry practices and enforcement intensity over time.

Other firm-specific controls include the sales per employee ratio, as a proxy for how labor-intensive the firm’s production process is, as well as several financial variables based on prior literature (Cohn and Wardlaw 2016; Caskey and Ozel 2017). These include the natural logarithm of the firm’s total number of employees, leverage, return on assets (ROA), change in ROA, and sales growth rate. I measure ROA as the ratio of net income to lagged assets. I do not include total assets as a control variable because this would introduce multicollinearity: the correlation between the log of assets and the log of the number of employees is 0.77.

Similar to Eq. 1 above, I estimate a linear probability model of fraud: where WageTheft_it represents one of the four wage theft measures outlined in Section 4.1. Following prior literature (e.g., Donelson et al. 2020), the indicator variable FinMisconduct_it equals one if firm i received an SEC Accounting and Auditing Enforcement Release (AAER) or paid a settlement in an investor lawsuit pertaining to actions taken in fiscal year t or t + 1, regardless of when the detection/settlement occurs.¹⁶ Note that one lawsuit or AAER can cover multiple fiscal years and multiple financial reporting violations. Lawsuit data come from the Stanford Securities Class Action Clearinghouse (SSCAC); I follow Karpoff et al. (2017) and exclude lawsuits that do not pertain to financial reporting or violations of SEC Rule 10(b)-5.¹⁷ I construct the dependent variable using both year t and year t + 1 to avoid potential measurement issues arising from uncertainty over whether wage theft occurs earlier in the year than financial misconduct.¹⁸

I begin my sample in 2004 for two reasons: (i) to avoid potential confounding effects of including both pre- and immediately post- Sarbanes-Oxley observations on the dependent variable, and (ii) to begin my sample after the overhaul of the Fair Labor Standards Act discussed in Section 2. I end the sample in 2015 because financial misconduct can take several years to detect and, subsequent to initial detection, class-action lawsuits can take multiple years to resolve. As with Eq. 1, I include firm and year fixed effects.

Although it is the limitations of the wage theft data that restrict my sample to larger companies, my focus on larger companies has an additional benefit for the financial misconduct sample: law firms that file class-action lawsuits prefer to sue larger companies, to attain more “bang for the buck.” Dyck et al. (2010) argue that the eagerness of law firms to file class-action lawsuits also mitigates the likelihood of undetected frauds perpetrated by large companies. Thus, by focusing on large firms, FinMisconduct_it is a reasonable representation of the set of actual financial misconduct that occurs. In addition, I omit financial services firms from the sample due to the different structure of their financial statements and regulations relative to other industries.

I select control variables based on prior fraud research (e.g., Dechow et al. 2011). These include firm size (measured as the log of firm assets), the percentage of soft assets, firm R&D, leverage, previous-year ROA as a measure of firm performance, year-over-year changes in inventories plus receivables, and the level of PP&E (scaled by assets). When R&D expenditure is missing in Compustat, I code this as zero and include an additional indicator to correct for the issues highlighted in Koh and Reeb (2015). Following Brazel et al. (2009), I include abnormal changes in the number of employees (measured as the percentage change in the number of employees minus the percentage change in the level of sales). I also include a proxy for the firm’s external financing dependence, as a measure of the firm’s incentive to raise additional capital (Rajan and Zingales 1998). This is an indicator variable that takes the value of one if \(\frac {\text {CFO}-\text {CAPX}}{\text {CAPX}} < -0.5\), where CAPX represents capital expenditures and CFO represents cash flow from operations. The greater a firm’s dependence on external financing, the higher its incentive to convince capital markets of its financial health. I provide details on the construction of the final estimation sample in Table 1.

Firm-year average levels of wage theft by 2-digit NAICS industry and year are presented in Panels A and B of Table 2. In both panels, column (1) considers the proportion of sample firm-years in which wage theft occurred (regardless of whether or not a violation was detected by WHD in those years). Columns (2), (3), and (4) provide descriptive statistics for the other three wage theft proxies used in my analyses. The incidence of (detected) wage theft is relatively stable over time, although it exhibits a decline after 2013. In untabulated analyses, I verify that my results are robust to excluding firm-year observations from 2014 and 2015 from the sample. Panel C presents descriptive statistics for the three non-indicator measures of WageTheft_it for the conditional sample of 1,573 firm-years where wage theft occurred. Panel C highlights the limitations of federal laws that set low maximum penalties; the average violation, scaled by the number of associated violation-years, results in $57,640.19 in penalties.

Summary statistics for all regression variables are presented in Table 3. Continuous variables are winsorized at the 1% and 99% levels. On average 11.6% of employees in firms’ industries are unionized, although this ranges from 2.0% to 51.3% within individual industry-years. The sample consists of large firms with average leverage ratios of 25.2%, which is low relative to the Compustat universe. The majority of firms (82%) are profitable in that they have positive ROA, reflecting the fact that my sample comprises larger firms. Approximately 2.9% of firm-years correspond to financial misconduct; because FinMisconduct_it = 1 when either the current or subsequent year is a misconduct year, this gives a slightly higher sample mean value of FinMisconduct_it of 4.0%.

Results from estimating (1) are in Table 4. I present results separately for each of the four wage theft measures outlined in Section 4.1. In all four cases, I find a positive and significant coefficient on SUSPECT_it, meaning that firms are more likely to engage in wage theft when they have incentives to meet or beat analyst forecasts. These effects are economically significant; the coefficient in column (1) suggests that the mean firm is 13.7% more likely to engage in wage theft relative to the case where it does not have meet-or-beat incentives and faces penalties that are 9.8% − 10.7% higher (based on the coefficients in columns (2) and (4)). These results are consistent with Hypothesis 1. To ensure that the results in Table 3 are unlikely to be driven by an omitted correlated variable, I follow Larcker and Rusticus (2010) and calculate the impact threshold of a confounding variable (ITCV). In untabulated tests I find that a potential omitted correlated variable would have to have an impact at least 7.6, 2.7, 4.3, and 9.4 times that of the most impactful control variable in columns (1)-(4), respectively, in order to invalidate my results. The results in Table 4 reflect meet-or-beat incentives driving the incidence, as well as the severity, of corporate wage theft: in untabulated analyses, I find that, within the set of wage theft firm-years observed in my sample, firms with meet-or-beat incentives face higher penalties per affected employee.

With respect to control variables, I find that larger firms (based on the number of employees) are more likely to engage in wage theft. This result may simply reflect the fact that firms with more opportunities to engage in wage theft are more likely to do so. Interestingly, I do not find statistically significant results for union coverage or the industry-year violation rate. These results are likely driven by the fixed effects design and minimal within-firm, between-year variation in these measures; in untabulated analyses that use 2-digit NAICS industry fixed effects in lieu of firm fixed effects, I find a significant negative relation between union coverage and the likelihood of wage theft. In these analyses I also find that ROA is negatively associated with the likelihood of wage theft; i.e., better-performing firms have less of a need to engage in actions that harm their employees.

Meet-or-beat behavior represents a response to firm-level financial incentives. However, managers’ personal incentives are also frequently cited as a cause of firms’ decisions to engage in misconduct (e.g., Armstrong et al. 2010). I therefore test whether managerial incentives play a role in firms’ decisions to engage in wage theft. I consider both positive and negative managerial incentives for wage theft. Following Coles et al. (2006) and Armstrong et al. (2013), I measure positive managerial incentives using CEO delta and vega. Delta measures the the sensitivity of executive compensation with respect to stock prices, while vega measures the sensitivity of executive compensation with respect to risk-related incentives. I obtain data on delta and vega from the authors of Coles et al. (2006) and include the natural logarithms of these quantities as additional variables in a modified version of Eq. 1.

To measure managerial disincentives I exploit five circuit court cases in the United States that shifted managerial liability for wage theft for subsets of my sample. Under the FLSA, certain employees can be held liable for wage theft. However, in order to be held personally liable for wage theft, an employee must have “significant control” over the establishment.²⁰ Courts across the United States have interpreted “significant control” differently and, as a result, the set of employees who are considered personally liable for wage theft has varied in a staggered fashion across geography and time. During my sample period, two notable circuit court cases had the effect of decreasing individual executives’ liability for firms headquartered in those circuits, and three other circuit court cases had the effect of increasing individual executives’ liability for firms headquartered in those circuits.

In 2008, the Eleventh Circuit Court of Appeals ruled in Alvarez Perez v. Sanford-Orlando Kennel Club that executives who were not regularly on site at a specific workplace did not have significant control and thus could not be held liable for wage theft at that location. Under similar logic, in 2012 the Fifth Circuit Court of Appeals handed down a similar ruling in Gray v. Powers. In contrast, in 2009 the Ninth Circuit Court of Appeals ruled in Boucher v. Shaw that executives exercised significant control over all company locations and could, as such, be held personally liable for wage theft. Rulings similar to Boucher v. Shaw were handed down in 2013 by both the First and Second Circuits, in Manning v. Boston Medical Center Corp. and Irizarry v. Catsimatidis, respectively. Finally, in 2014 the Supreme Court resolved the uncertainty created by these five circuit court rulings by affirming the decision in Irizarry v. Catsimatidis; in doing so, the Supreme Court established that individuals with “general control over corporate affairs” may be held personally liable for wage theft under the FLSA.

Using the court cases above, I conduct staggered difference-in-differences tests to examine whether these shifts in executives’ liability affect firms’ levels of wage theft. I construct two variables for use in these tests. The first, LiabDecrease_it, equals 1 for firms headquartered in the Eleventh Circuit from 2008-2014 or the Fifth Circuit from 2012-2014. The second, LiabIncrease_it, equals 1 for firms headquartered in the Ninth Circuit from 2009-2014 or the First or Second Circuits from 2013-2014. These two variables represent the product of the “treatment” and “post” variables in the differences-in-differences specifications. The main effect of the “post” variable is subsumed by year fixed effects, while, to account for the main effect of the “treatment” variable, I include headquarters state fixed effects.²¹ I obtain headquarters information data from the Loughran and McDonald Augmented 10-X Header file.

I present results from tests of managerial incentives in Table 6. For brevity, I only tabulate results using the wage theft indicator. Results using the other three wage theft proxies outlined in Section 4.1 are qualitatively similar in terms of both directional effect and statistical significance. In column (1) I consider CEO compensation incentives alone; in columns (2) and (3) I consider the effects of liability-decreasing court cases and liability-increasing court cases, respectively; and in column (4) I consider all three measures. The sample is significantly smaller in columns (1) and (4) because of limited data coverage in ExecuComp that underlies the calculations of vega and delta. My findings in column (1) are consistent with 2020) results for labor violations more broadly: I find that CEO vega is positively associated with wage theft although, unlike in their study, I do not find that CEO delta has an effect on wage theft. This result implies that executives with greater incentives for risk-taking are more likely to engage in wage theft. The effect of the circuit court cases is also consistent with my expectation that wage theft is higher (lower) after court cases that decrease (increase) executives’ personal liability for wage theft.

Having documented that wage theft arises from similar financial incentives to other forms of real activities management, I next turn to the relation between wage theft and financial misconduct. Table 7 presents results from estimating (2) for each of the four wage theft proxies (indicator, total penalties assessed, number of sites at which wage theft occurred, and per capita penalties). In all cases, the dependent variable is FinMisconduct_it, an indicator for whether firm-year t or t + 1 resulted in either an AAER or a securities lawsuit.

Control variable results are relatively standard with respect to prior literature. For example, firms that see year-over-year increases in ROA are less likely to be caught engaging in fraud. Higher R&D, which may signal more opaque financial statements, is also associated with higher levels of financial misconduct. Firms with a greater need for external financing are more likely to engage in financial misconduct, possibly due to the heightened incentive to obtain favorable financing terms.

Across all specifications in Table 7, the coefficient on the wage theft indicator is not statistically significant. However, this indicator does not differentiate undetected wage theft from detected wage theft and, accordingly, the results above could reflect countervailing forces. When wage theft is ongoing and undetected, firms may have less of a need to engage in financial misconduct. Conversely, after wage theft is caught, the costs of further wage theft increase due to WHD’s dynamic enforcement model, in which the severity of the penalty and future WHD scrutiny depend on both the severity of the underlying violation and the firm’s compliance history. Repeat violators (i) are more likely to be investigated by WHD in subsequent years and (ii) receive higher penalties than first-time offenders for the same types of violations. In addition, repeat violations may incur significant incremental litigation or arbitration risk. If the firm deems the increase in the costs of future wage theft to be material, it may shift from wage theft to other forms of misconduct subsequent to the detection of wage theft.

I am able to directly test this possibility because wage theft typically lasts for multiple years and, unlike in the case of financial misconduct, WHD typically catches wage theft while it is still ongoing and imposes sanctions immediately.²² As a result, for many of the violations in my sample, there are two distinct periods: (i) a period in which the violation was occurring and undetected, and (ii) a period in which the violation was occurring and detected. Because I can identify these periods, I can separate the effect on financial misconduct of firms’ decisions to engage in wage theft from the effect of firms’ wage theft being detected.

To implement this empirically, I create a variable, WageTheftCaught_it, measured analogously to the main wage theft measures. Specifically, for each of the four measures of WageTheft_it, the corresponding measure of WageTheftCaught_it considers only the portion of WageTheft_it that is allocable to detection years. For example, if a firm engages in wage theft in 2008 and 2009, and is caught in 2009, then the indicator form of WageTheft_it equals 1 for both 2008 and 2009 while the indicator form of WageTheftCaught_it is equal to 0 in 2008 but 1 in 2009. I then estimate the following modified version of Eq. 2:

Results from estimating (3) are in Table 8. Each of columns (1)-(4) corresponds to one of the four wage theft measures. In all four cases, the coefficient on WageTheftCaught_it is positive and significant. Conversely, the coefficient on WageTheft_it is negative, and statistically significant in all but column (3).²³ These results suggest that while wage theft is undetected, firms have less incentive to engage in financial misconduct; however, once firms have been caught engaging in wage theft, they shift toward engaging in financial misconduct. The coefficient of -0.018 on WageTheft_it suggests that firms are nearly 45% less likely to engage in financial misconduct while engaging in undetected wage theft, relative to non-wage-theft firm-years. However, once these firms are caught, the coefficient of 0.021 on WageTheftCaught_it implies that their likelihood of engaging in financial misconduct nearly doubles (relative to the firms concurrently engaging in undetected wage theft). Collectively, and perhaps most relevant, the total effect of the two coefficients suggests that firms caught engaging in wage theft are 7.5% more likely to subsequently engage in financial misconduct. Because FinMisconduct_it reflects the incidence of financial misconduct in year t or t + 1, it is unlikely that my results merely reflect companies managing earnings in the year of detection in response to the imposition of financial penalties. In sum, my findings suggest that wage theft precedes financial misconduct. This result is consistent with Choi and Gipper (2019), who find that lower-paid employees’ wages decline significantly in the years immediately preceding financial fraud; the results in Table 8 suggest one mechanism (wage theft) by which this decline occurs.

A potential alternate explanation for the result in Table 8 is that it reflects complementarities in the detection of misconduct rather than in the incidence of misconduct, i.e., that when one agency detects misconduct at a firm, other agencies also investigate that firm. However, this is unlikely with respect to financial and labor-related misconduct. The SEC’s enforcement manual²⁴ references collaboration with and referrals from agencies such as Treasury, the DOJ, and state securities regulators, but does not mention wage theft or the Department of Labor. Further, while securities lawsuits can arise from a variety of trigger events, I manually investigate all securities lawsuits in my sample, and none mention wage-related enforcement actions or lawsuits.

In this section I test whether the documented results in Sections 5.2 and 5.3 interact; that is, do increased personal incentives (or reduced disincentives) increase the likelihood that firms engage in wage theft in response to meet-or-beat incentives? To address this question I estimate modified versions of Eq. 1 that include interaction terms between the meet-or-beat indicator and the two incentive-increasing measures outlined above (CEO vega and liability-decreasing court cases). I also explore the role of firm-level incentives and opportunities in facilitating the relation between wage theft and meet-or-beat incentives. To do this, I consider habitual meet-or-beat firms (as a proxy for firms with heightened incentives for meet-or-beat behavior) and non-R&D firms (as a proxy for firms with greater reliance on hourly, rather than salaried, employees). I tabulate results in this section using only the indicator variable version of WageTheft_it; my findings are qualitatively unchanged if I instead use any of the other three proxies.

Prior literature (e.g., McVay et al. 2006) documents a link between managers’ personal incentives and earnings management behavior around analyst forecast benchmarks. If the results presented in Table 4 reflect wage theft as a form of real earnings management, I should therefore observe that stronger managerial incentive corresponds to a stronger relation between meet-or-beat incentives and wage theft.

The findings in columns (1) and (2) of Table 9 are consistent with this assertion. I find in column (1) that the interaction between SUSPECT_it and LiabDecrease_it is positive and significant. This result suggests that a decrease in managers’ personal disincentives for wage theft makes wage theft a more attractive form of real earnings management. In column (2), I interact SUSPECT_it with CEO vega and find a positive and significant coefficient on the interaction term. Although CEO vega does not correspond to wage theft incentives specifically, higher risk-reward sensitivity creates a higher overall incentive to manage earnings to meet benchmarks. Consistent with this prediction, I find that the relation between wage theft and meet-or-beat incentives is stronger when CEO vega is higher (and, hence, when there may be more top-down pressure to cut wage costs).

I next examine the role of firm-level incentives and opportunities in facilitating the relation between meet-or-beat incentives and wage theft. Chu et al. (2019) argue that meet-or-beat incentives are more important to firms with a reputation for beating earnings expectations. To test whether this finding applies to the wage theft setting, I consider whether firms with a reputation for beating earnings expectations are more likely to engage in wage theft when faced with meet-or-beat incentives. I construct a measure of reputation similar to that of Chu et al. (2019) and define a firm as a habitual meet-or-beat firm if its actual earnings exceeded the analyst consensus (regardless of the margin) in each of the previous three years. Results from this specification are presented in column (3) of Table 9. Consistent with the argument that habitual meet-or-beat firms have stronger incentives to engage in real earnings management to continue meeting earnings targets, I find that habitual meet-or-beat firms are more likely to engage in wage theft.

In addition to firm-level incentives, I consider heterogeneity in the opportunities for wage theft that may arise as a result of a firm’s business model. Because it is not possible to directly observe the proportion of a firm’s employees that are salaried, I measure opportunities for wage theft using an indicator variable for whether firms do not disclose R&D expenditures. Firms that rely upon R&D are likely to be less reliant on hourly (rather than salaried) labor. Column (4) of Table 9 presents results using this interaction. Surprisingly, I find that non-R&D firms are less likely than R&D firms to engage in wage theft in response to meet-or-beat incentives. This result may reflect differences in the signalling value of observed wage theft as a form of real earnings management. Specifically, when firms’ business models are naturally prone to wage violations, such violations may be less likely to indicate deliberate actions. Conversely, when firms’ business models are less prone to wage violations, the incidence of such violations is more likely to indicate a deliberate strategy to cut costs, even when it may not be feasible to do so legally.

Several firms have multiple instances of detected wage theft during my sample period. If firms that get caught continue to engage in wage theft, then my results may not actually reflect substitution away from wage theft (i.e., additional monitoring by WHD may have no effect). On the other hand, this may simply reflect cross-sectional characteristics. For example, suppose Firm A is systematically more likely to engage in wage theft than Firm B (because of, e.g., the industry Firm A operates in). If Firm A gets caught engaging in wage theft, it may reduce its likelihood of future wage theft. This does not mean that it will never engage in wage theft again, just that it is less likely to do so. If this is the case, then observing repeat instances of wage theft is still consistent with my interpretation of the results in Section 5.4. To distinguish between these alternative explanations, I estimate the following model: where FutureWageTheft_it is an indicator variable that equals 1 if firm i engaged in wage theft in at least one of years t + 1, t + 2, or t + 3. A negative value of β₁ in Eq. 4 implies that after being caught engaging in wage theft in year t, firm i is less likely to engage in it over the next three years. Results from estimating (4) are in Table 10. The significant negative coefficient on WageTheftCaught_it in column (1) suggests that firms are less likely to engage in future wage theft subsequent to having been caught.

As an additional mitigating factor to the concern of repeat violations, I consider the geography of violations within firms over time. Raghunandan and Ruchti (2021) show that after firms face sanctions from the Occupational Safety and Health Administration (OSHA), they are less likely to continue to engage in violations in the state they were sanctioned in. This may be because OSHA is more likely to flag a future violation as repeat if it occurs in the same state as the initial violation; as a result, the increase in the costs of future misconduct subsequent to detection may not be uniform across locations in which the firm operates. Building on these findings, I re-estimate (4) accounting for geographic dispersion in violations. I construct four new variables: FutureWageTheftHQ_it and FutureWageTheftNonHQ_it, which are analogous to FutureWageTheft_it but for a firm’s headquarters and non-headquarters states, respectively; and WageTheftCaughtHQ_it and WageTheftCaughtNonHQ_it, which are analogous to WageTheftCaught_it but for a firm’s headquarters and non-headquarters states, respectively.

I present results from this modified specification in columns (2) and (3) of Table 10. The negative and significant coefficients on WageTheftCaughtHQ_it in column (2) and WageTheftCaughtNonHQ_it in column (3) are consistent with Raghunandan and Ruchti (2021) finding that the deterrence effect of violations is strongest within rather than between geographic jurisdictions. In conjunction with the results in Section 5.4, Table 10 suggests that after getting caught engaging in wage theft, firms are subsequently (i) less likely to engage in future wage theft and (ii) more likely to engage in financial misconduct.

Recent literature finds that firms with worse corporate culture are more likely to engage in financial misconduct, suggesting that operational culture extends to financial reporting practices. For example, Kim et al. (2012) find that more “socially responsible” firms engage in earnings management less frequently and are less likely to receive SEC enforcement actions. Building on this literature, in this section I examine the role of corporate culture in driving the relation between firms’ financial reporting incentives and wage theft.

To test whether either the paper’s meet-or-beat results or the results on financial misconduct are driven by corporate culture, I construct two measures of corporate culture based on recent studies by Altamuro et al. (2017, hereafter AGZ) and Kedia et al. (2019, hereafter KLR). AGZ measure culture as firms’ internal control environments and show that, within the pharmaceutical industry, firms with internal control weaknesses (ICWs) are more likely to both restate financial statements and have Food & Drug Administration inspection failures. KLR use firms’ compliance history across five federal agencies to measure recent compliance culture, and document that past compliance behavior predicts the likelihood of receiving an AAER from the SEC.

I construct the internal control-based measure by first building a probit model of the likelihood of an internal control weakness. I then follow AGZ and label a firm as having a weak compliance culture (WeakICW_it = 1) if the firm reports ineffective internal controls despite its predicted probability of an ICW being below the median of the sample predicted ICW probability. To construct the compliance history-based measure, I draw upon the remainder of the Violation Tracker database, which contains compliance information for over 50 federal agencies. I define a firm as having a weak compliance culture (WeakV iol_it = 1) if the firm had at least one non-wage theft violation in fiscal years t, t − 1, and t − 2.²⁵

Using these measures of corporate culture, I estimate modified versions of Eqs. 1 and 3 that include one of the corporate culture proxies as well as an interaction term between the culture proxy and the main independent variable of interest. I present results in Table 11. For brevity, I do not tabulate the first-stage ICW estimation model or control variable coefficients. I also only tabulate results using the indicator form of WageTheft_it; results using the other three measures are similar. The sample size is smaller in the ICW-based tests because of data unavailability in constructing WeakICW_it. In columns (1) and (3) I consider WeakICW_it for the meet-or-beat and financial misconduct tests, respectively, but find that the coefficient on the interaction term is insignificant in both cases. In columns (2) and (4) I consider WeakV iol_it. The positive and statistically significant interaction term in column (2) suggests that firms with a weaker compliance culture are more likely to respond to meet-or-beat incentives by engaging in wage theft. However, I find no such relation between compliance culture and financial misconduct in column (4), suggesting that inasmuch as compliance culture may have an effect on financial misconduct, any such effect flows through the decision to first engage in nonfinancial misconduct.

The results in Section 5.4 suggest a potential broader implication, beyond the wage theft setting, for researchers and practitioners interested in financial misconduct. The fact that financial misconduct becomes more likely after the detection of wage theft rather than after the decision to initially engage in it suggests that using data on violation detection dates alone could be useful in predicting firms’ financial reporting quality. However, this possibility may not hold; as discussed in the introduction, a unique feature of wage theft is that it is not generally observable to outsiders until detected, while other forms of operational misconduct are.

To directly test whether the detection of other forms of non-financial misconduct predicts financial misconduct, I draw upon the entirety of the Violation Tracker database. I classify violations into one of four types: labor-related (other than wage theft), environmental, consumer protection-related, and other.²⁶ I then re-estimate a modified version of Eq. 3 incorporating indicators for whether firm i was sanctioned during fiscal year t for each of these types of violations.²⁷ For brevity, I do not tabulate this specification. While my results on wage theft continue to hold, I do not find a statistically significant relation between financial misconduct and the detection of non-wage theft labor misconduct (e.g., workplace safety violations), consumer protection violations, or environmental violations. These results suggest that the detection of misconduct may be most informative in settings where it is more difficult to externally observe ongoing misconduct.