Incentivizing Financial Regulators^*

Abstract

Enforcement is a key component of financial regulation. Every year the U.S. Securities and Exchange Commission, the primary regulator of securities markets and the subject of this paper, files hundreds of enforcement actions that impose significant costs on market participants and deter future misconduct. The SEC’s enforcement staff consists of nearly one thousand attorneys, and yet little is known on how enforcement is affected by the incentives of these individuals. In particular, how the internal organization of the enforcement staff can incentivize and stimulate effort is unknown. A common perception is that a “faceless bureaucracy” enforces the law and how the SEC chooses to organize itself is irrelevant to how it carries out its duties.¹ Building on an original data set, I reach a different conclusion. I find that tournament incentives (Lazear and Rosen 1981), as reflected in hierarchical pay gaps and promotion opportunities inside the SEC, could enhance the agency’s enforcement activity.

A key contribution of this paper is original data on all SEC enforcement attorneys from 2002 to 2017. I submitted multiple Freedom of Information Act requests to various federal bureaus, and manually collected information from more than 3,000 original court documents filed by the SEC over the years. The result is a unique data set that links each attorney’s enforcement activity to employment information, such as salary, hierarchy grade, and tenure. This granular data set allows me to overcome a major challenge for any empirical study of incentives and financial regulation: data shortage.

I start by showing that attorneys with a robust enforcement caseload are winning the tournament: they are promoted faster and in higher rates. This leads to the main hypothesis: enforcement should increase with the value of promotion and in particular with its monetary value (the expected pay raise or the “prize”). Attorneys who expect a higher “prize” should put more effort into their enforcement activities in order to win the promotion. The evidence is consistent with that prediction, and the following example illustrates the main results. Suppose the “prizes” in the Boston office are more valuable than in Chicago, that is, hierarchy grades in Boston feature large salary gaps between them. I find that the Boston office files more enforcement actions, especially against severe financial misconduct in the Boston area. Inside the Boston office, more cases are filed by grades that offer their attorneys higher “prizes,” that is, grades that are far below the next grade. Finally, within each Boston grade, lower-paid attorneys who will benefit even more from a promotion to the next grade are filing more actions.

The evidence is consistent with the central tournament hypothesis: offices, levels, and attorneys with higher promotion value exert greater effort and file more enforcement actions. A tight fixed effects specification rules out plausible alternative stories. The incentive effect is not explained by seniority or rank or by national or agencywide trends. The effect holds within office and so is not explained by various region-specific conditions, such as managerial style, the extent of financial misconduct, and career opportunities in the local labor market. In the tightest specification, I include year-office-grade indicators, which forces a comparison between attorneys who compete in the same tournament at the same time. This specification holds the “prize” fixed, which might be controlling away some of the very effect I strive to document, and yet it shows the highly significant positive relation between incentives and enforcement.

To support the tournament interpretation, I exploit a plausibly exogenous shock to the incentives of a subsample of attorneys. In 2002, the SEC transferred its workforce to a new pay system. The conversion included an arbitrary pay raise, as each salary was rounded up to the closest pay step in the new system. That pay raise had a large impact on the tournament incentive: the higher the pay raise, the lower the incentive became. It was exogenous to the employee’s characteristics, determined exclusively by the distance from the closest pay step. I estimate a 2-SLS model with the exogenous pay raise as an instrument, combined with the granular fixed effects, and document a significant positive relation between enforcement and tournament incentives.

The remainder of the paper addresses specific concerns, and I begin with the heterogeneity across enforcement actions. In the baseline specification, I count all actions equally, but in reality not all actions are born equal. Filing an enforcement action of type X could require more effort than filing an action of type Y, and ignoring that heterogeneity could lead to a measurement error in the outcome variable. Moreover, powerful incentives may possibly lead agents to allocate their effort toward less important or less desirable enforcement actions (as in Holmstrom and Milgrom 1991). To draw the line between complex and important cases to simple and routine ones, I consider several proxies from the legal literature: the factual complexity of the case (as reflected in the number of defendants) and the severity of the allegations (as reflected in parallel criminal proceedings or by alleging violations of antifraud provisions). Across all criteria, I find that tournament incentives increase the probability of filing complex enforcement cases alleging severe financial misconduct.

I consider a set of alternative incentive structures to account for different features of the SEC organizational design. Those alterations include different promotion patterns, different treatment of missing hierarchy grades, and different expectations with regards to future pay raises. None of these permutations changes the main results.

A potential omitted variable is the unobserved attorney quality. The main concern is that strong incentives were assigned to “good” attorneys, and the unobserved quality would drive enforcement activity upward and bias the estimated incentive effect. That concern appears inconsistent with the SEC’s institutional setting, based on numerous discussions I have had with current and former SEC employees. For instance, experienced attorneys who are more likely to file enforcement cases are assigned higher salaries and thus have weaker incentives. For a large subsample of attorneys, I collect additional data on bonuses and legal education, which are observable proxies for the unobserved attorney quality. I add a set of controls based on the bonus and education data and the results remain highly significant.

Overall, promotion incentives appear to motivate attorneys to engage in enforcement. But that does not necessarily affect the SEC’s overall enforcement activity, because perhaps highly incentivized attorneys convince their supervisors to be assigned to cases which would have been filed anyway. This interpretation, although possible, seems less consistent with additional evidence. I find that promotion incentives are associated with more enforcement not only across attorneys but also across grades and in the time series. I also find that stronger incentives are associated with more aggressive settlement terms, such as monetary penalties and industry bars. Taken together, the evidence suggests that stronger incentives not only shape the internal case allocation but also lead to more robust enforcement.

The average incentive for SEC enforcement staff is |$23%$|⁠. The incentive is clearly low powered compared to the |$349%$| average incentive among CEOs (Coles, Li, and Wang 2017). Nevertheless, the impact it seems to have on SEC enforcement is nontrivial. For instance, moving from the |$10$|th to the |$90$|th percentiles of the incentive distribution increases enforcement probability by |$11$| percentage points, which is |$16%$| of the mean. Similarly, recent studies of wage differences among peers report material responses to quasi-random pay gaps as small as |$5%$| (Breza, Kaur, and Shamdasani 2016) and |$10$| cents (Dube, Giuliano, and Leonard 2019).

The paper makes the following three contributions. First, I uncover a novel driver of financial regulation: tournaments among SEC staff. Studies of financial regulation in general, and SEC enforcement in particular, tend to focus on the costs and benefits of the rules for various market participants (Kedia and Philippon 2007; Karpoff, Lee, and Martin 2008a, 2008b; Yu and Yu 2011; Giannetti and Wang 2016). A growing literature studies how the organizational design of financial regulators affects regulation: the way the agencies fund their operations (Kisin and Manela 2018), delegate authority to local offices (Gopalan, Kalda, and Manela 2017), and allocate supervisory hours across regulated entities (Eisenbach, Lucca, and Townsend 2016; Hirtle, Kovner, and Plosser 2020). However, the incentives of the individual regulators who implement the rules are often overlooked. With that gap in mind, I conduct the first study on the incentives generated by the SEC’s internal organization. I document the existence of those incentives and explore their impact on the agency’s enforcement program.

Second, I illustrate how promotion opportunities affect state bureaucracy. The common perceptions are that state employees prefer a “quiet life” and mostly respond to intrinsic motivation (Weisbrod 1983; Besley and Ghatak 2005; Bénabou and Tirole 2006; Bryson, Forth, and Stokes 2017). An emerging literature undermines those assertions and shows that motivation is affected by higher starting salaries (Dal Bó, Finan, and Rossi 2013), special rewards (Ashraf, Bandiera, and Lee 2014), and outside career opportunities (Bond and Glode 2014; Lucca, Seru, and Trebbi 2014; Agarwal, Lucca, Seru, and Trebbi 2014; Tabakovic and Wollmann 2017). I contribute to this literature by focusing on internal promotion incentives, which are an indirect consequence of how the bureaucracy chose to organize its workforce. My paper is the first one to document how those incentives affect the operations of regulatory agencies.

Third, I contribute to the tournament literature by studying employee-level incentives and output. Individual productivity is typically observed only among low-skilled workers (Breza, Kaur, and Shamdasani 2016), and thus prior studies have resorted to firm-level measures of productivity (Main, O’Reilly III, and Wade 1993; Kale, Reis, and Venkateswaran 2009; Kini and Williams 2012; Coles, Li, and Wang 2017; Hibbs and Locking 2000; Ouimet, and Simintzi 2017a, 2017b). In contrast, the unique data set I assemble allows me to study tournament effects at a granular level. Moreover, I am able to distinguish between various types of litigation activities and to show how agents allocate effort toward more complicated and consequential enforcement actions (Holmstrom and Milgrom 1991).

1. Institutional Setting and Data

1.1 Institutional setting

Enforcement - This paper is centered on enforcement actions, which are legal proceedings initiated by the SEC for violations of federal securities laws (see Figure 1). Examples of such violations are insider trading, accounting fraud, and inadequate disclosure. Career staff at the SEC conduct examination and investigation, and present their findings to the Commission, a five-member body that oversees the SEC’s operations and can authorize the SEC staff to commence an enforcement action. The SEC can file a civil complaint in a U.S. District Court or institute a proceeding in front of an administrative law judge. A successful lawsuit could result in various injunctions, industry bars and suspensions, monetary penalties, and return of illegal profits (disgorgement). The SEC might also refer the case to the Department of Justice for parallel criminal prosecution.³

Figure 1

Example of a civil enforcement action

Signature page of an enforcement action filed on April 2004 in the U.S. District Court for the District of Columbia against Barry Richard Kusatzky, the former controller of California Amplifier. On February 2006, the Court entered a settled final judgment against Kusatzky for falsifying the company’s financial statements and insider trading. The final judgment imposed a permanent officer and director bar and ordered payment of partial disgorgement of |${\$}$|25,000.

Open in new tabDownload slide

Organization - The SEC consists of 12 offices: the headquarters in Washington DC and 11 regional offices. As explained below, this paper is limited to attorneys from the Enforcement Division (which operates from DC) and the 11 regional offices. Each office is organized in a 9-grades hierarchical system: from SK-12 through SK-17 and from SO-1 through SO-3. Staff attorneys are in SK-12 through SK-14, mid-level attorneys in SK-15 and SK-16, and top managers who typically carry titles such as “Assistant Director” occupy SK-17 or higher (see Figure 2).⁴

Figure 2

The SEC’S hierarchical structure

The figure presents the hierarchical structure at the SEC. Nine grades are ordered from SK-12 (lowest) to SO-3 (highest).

Open in new tabDownload slide

Salaries and Promotions - The SEC attorney’s salary consists of three main components, similar to many other federal agencies. The base pay is determined by the attorney’s pay grade, which is capped from above and is identical across all SEC staff. The base pay is supplemented by locality pay, a fixed percentage determined by the employee’s duty location. The grade caps and locality rates slowly increase over time, and, as of 2018, the locality rate is between 15.36% (Salt Lake City) and 39.3% (San Francisco). Lastly, the attorney may be awarded a cash bonus, distributed annually at the discretion of the attorney’s supervisors to reflect high performance.⁵ Promotions at the SEC are one grade at a time within the same office.⁶ In the absence of promotion, within-grade pay raises are distributed among all peers and so the contemporary salary is a function of the attorney’s starting salary and tenure. The variation in the starting salary is largely a function of the starting year and grade, education level, past work experience, and employee-specific characteristics (negotiation skills, etc.).

1.2 Enforcement data

I collect from the SEC’s website all the complaints filed by the agency in U.S. District Courts. The original court documents are the primary source of information for this paper. I focus on civil actions because the SEC rarely discloses the names of the individual attorneys who participate in administrative proceedings (Choi, Gulati, and Pritchard 2018). I obtain the majority of the complaints from the Litigation Releases section, where all announcements pertaining to civil actions are posted. I also scan the entire Accounting and Auditing Enforcement Releases section (“AAER”), which includes some enforcement proceedings that involve accountants or auditors, and the Press Releases section, which includes various official announcements. Finally, I perform a generalized web search to locate any additional litigation documents stored elsewhere on the SEC’s website. I discard actions filed before 2002 or after 2017, to match the availability of the workforce data set. This approach yields enforcement sample with |$3,178$| actions filed between January 1, 2002, and December 31, 2017.

For each enforcement action, I collect the date of the filing, the alleged violations, the names of the SEC attorneys, and the number of defendants. I read each of the releases to flag actions that are accompanied by criminal proceedings and actions that include an asset freeze or other form of temporary orders. I also identify actions that arise from the same misconduct and are therefore inherently connected. I classify actions as contested, settled, or partially settled, and, for settled actions, I code the information on the settlement terms: monetary penalties and other sanctions imposed on the defendants.

By construction, the sample excludes actions that are not stored anywhere on the SEC’s website. To better understand the selection magnitude and criteria, I obtain the appendixes of the SEC’s annual reports, where the Enforcement Division lists all enforcement actions initiated during the fiscal year and the total number of defendants. According to the SEC, between 2002 and 2017 it filed 3,863 civil actions, and, during fiscal years 2003-2017, it brought charges against 11,335 defendants. My enforcement sample in the corresponding periods includes 2,770 civil actions and 9,485 defendants, bringing the coverage ratio to |$82.3%$| and |$83.7%$|⁠, respectively. The SEC classifies actions into 11 primary categories, based on the nature of the underlying offense. The plurality of missing actions are classified as civil contempt (coverage |$4%$|⁠), essentially “secondary” cases triggered when a defendant fails to comply with court orders issued in a previously resolved enforcement action. Year-by-year analysis shows the coverage ratio improving over time and stabilizing at about |$95%$| since 2012. Note that Velikonja (2015) argues that the SEC’s published statistics may overrepresent the true scope of the SEC’s enforcement filings. By her calculations, the total number of enforcement actions in the fiscal years 2002-2014 are almost |$30%$| less than the official numbers published by the SEC. If true, it implies that the “real” coverage ratio I achieve is in fact higher.

Prior studies have extensively relied on the AAER section of the SEC’s website (Dechow, Ge, Larson, and Sloan 2011; Armstrong, Larcker, Ormazabal, and Taylor 2013; Ali and Hirshleifer 2017). As explained above, AAER designation is assigned to some of the enforcement proceedings which involve accountants or auditors, and not to all of them. Indeed, only |$18.5%$| of the actions in my sample have an AAER designation, illustrating how the AAER seems to capture only a fraction of the civil enforcement activity of the SEC. Other studies have used Lexis-Nexis and the newly released Securities Enforcement Empirical Database (“SEED”) to zero in on some of the proceedings against public companies (Licht, Poliquin, Siegel, and Li 2018; Choi, Gulati, and Pritchard 2018). Finally, Karpoff, Lee, and Martin (2008b) collected all enforcement actions for financial misrepresentation initiated by the SEC between 1977 and 2006. This data set includes 1,130 civil actions against 4,080 defendants and has been subsequently used in other studies (Karpoff, Lee, and Martin 2008a; Kedia and Rajgopal 2011; Hazarika, Karpoff, and Nahata 2012). It partially overlaps with my workforce data set (which starts from 2002) and is limited to violations of three specific provisions of the Securities Exchange Act of 1934.⁷

1.3 Workforce data

I assemble a comprehensive workforce data set on all individuals who worked at the SEC at any point between 2002 and 2017. I obtained the data through multiple Freedom of Information Act requests submitted to the SEC and to other Federal agencies, and I had numerous conversations with current and former SEC employees in order to better understand the organizational feature of the SEC’s workforce. The data set includes the employee’s full name, occupation, and year of accession and separation (if applicable). It also provides annual information on location, salary, pay grade, job title, transition across offices, tenure, overtime payments, bonus, and promotions. To the best of my knowledge, the data set is free of selection bias and includes the universe of SEC employees from that period.⁸ I augment the workforce data with additional biographic information from LinkedIn and Martindale-Hubbell, a commonly used directory for U.S. lawyers (more on that below). To the best of my knowledge, this is the most comprehensive employee-level data set on financial regulators, and the only one to tap the regulator’s official records as the primary source of workforce information.

Only attorneys sign the court documents, and I hence limit the workforce sample to attorneys. I manually match the names from the court documents to the names of the attorneys from the workforce sample. I was able to identify 99.8% of the litigating attorneys, virtually all except for six individuals. The matching confirmed that 96.9% of the litigating attorneys work at the Enforcement Division or in one of the SEC’s 11 regional offices, which is consistent with my understanding of the SEC’s functional organization. I therefore limit the study to attorneys who work at the Enforcement Division in Washington DC or in any of the 11 regional offices. The final workforce sample consists of 1,914 attorneys and 14,940 attorney-year observations.

2. Empirical Strategy and Descriptive Statistics

2.1 Measuring enforcement and incentives

The counting-based enforcement measures provide a simple, intuitive way to compare litigation portfolios across attorneys, ranks and offices. The number of enforcement actions is a closely watched metric, computed and presented annually by the SEC to Congress and the general public. Moreover, Section 2.4 demonstrates how this “simplistic” representation of the litigation portfolio is an important predictor of internal rewards at the SEC, namely, promotions. In other words, the SEC and those who oversee and study its activities seem to value the quantity of enforcement actions. At the same time, counting-based measures do not factor in potential heterogeneity across enforcement actions and across attorneys. This could introduce a measurement error in the outcome variable, and I address this and related concerns in Section 3.3 with a set of alternative outcome variables based on a deeper analysis of enforcement actions.

2.2 Empirical strategy

Proper identification of the tournament effect is difficult. At the SEC, neither the salaries nor the incentives are randomly assigned. Numerous assignment mechanisms could generate similar incentive-enforcement correlations and bias the estimated tournament effect. The identification strategy proceeds as follows. First, I estimate ordinary least squares (OLS) regressions with a set of fixed effects to rule out common alternative explanations. Next, I exploit a plausibly exogenous shock to the SEC’s pay structure and use it as an instrument for the attorneys’ promotions incentives. Finally, I develop additional tests combined with a deeper discussion of the SEC’s institutional setting to address other specific concerns and solidify the tournament interpretation of the results.

2.2.1 OLS fixed effects model

The outcome is one of the attorney-level enforcement measures and |$incentive$| is the individual incentive. |$\overrightarrow{\lambda}$| is a flexible set of fixed effects. I start with the attorney’s grade, which is necessary given the SEC’s hierarchical structure that assigns more cases to more senior attorneys (Figure 4). This specification studies deviations from the average caseload. I add year dummies which control for agencywide initiatives and other possible macroeconomic conditions, such as the financial crisis, all of which clearly shape the enforcement activity of the agency. In the third specification, I replace year dummies with year-office dummies to absorb all cross-office variation. Specifically, it absorbs any source of variation coming from local conditions: outside career opportunities, pool of financial misconduct, differences in workforce characteristics, and managerial style of the office’s top brass. Those could potentially affect the enforcement productivity of the office, and the year-office dummies force a comparison between attorneys who work at the same office (the SEC has 12 offices).

Finally, the fourth specification includes year-office-grade fixed effects. In addition to controlling for macroeconomic forces and local conditions, this specification directly compares attorneys who compete in the same tournament at the same time. This is arguably the tightest specification. Note, however, that including year-office-grade fixed effects may be “overcontrolling,” that is, it may be “controlling away” some of the very effect I strive to document. By forcing the comparison to be made between attorneys within the same “working unit,” I essentially give away the variation in the target salary (the numerator), and focus on the variation in the individual salary (the denominator): the lower the salary, the higher the incentive.

Newly promoted attorneys might be more enthusiastic or wish to prove themselves in their newly acquired position. Either way, a recent promotion is likely to increase their effort and, consequently, boost their enforcement activity. To address that issue, I include in all specifications an indicator that equals one if the attorney was recently promoted to the current grade. The results reported below use 1-year window to define “recently promoted” attorneys, but the results remain qualitatively similar if I include promotions within the last 2 or 3 years or use a continuous variable for the number of years spent at the current grade. In all specifications the explanatory variables are lagged, to rule out reverse causality, and standard errors are clustered at the attorney level.

2.2.2 Instrumental variable

A subsample of SEC enforcement attorneys participated in a quasi-random experiment, where a component of their salary was close to randomly assigned. In 2002, the SEC implemented a new compensation system. The new system consisted of 20 grades, each with 21-30 pay steps. All employees were transferred from the old system to the new one based on a fixed procedure. First, each employee was converted to the appropriate grade. Next, a fixed across-the-board percentage was added to the employee’s salary, determined only by the employee’s office and designation. Lastly, that augmented salary was rounded up to the closest pay step within the new grade. Figure 8 illustrates the mechanism for an attorney with |${\$}$|114,679 base pay in the old system. His fixed increase was 6%, based on his geographic location and occupation, which yields |${\$}$|121,560. In grade 16 in the new pay system, the closest step above |${\$}$|121,560 is |${\$}$|122,399, which became his new base pay. It follows that the employee received an individual pay raise of |$\$122,399-\$121,560=\$839$|⁠, on top of the fixed across-the-board pay raise.

The pay conversion in 2002 introduced an exogenous shock to the SEC’s pay structure. That shock had a differential impact on attorneys, depending on the magnitude of the roundup component: the salary raise up to the closest pay step in the new pay structure. I use that shock as an instrumental variable for tournament incentives. Of |$1,913$| attorneys in the full sample, |$1,177$| were not exposed to the treatment: they joined the SEC during the transition period or afterward, or left the SEC during the transition. Of the remaining |$736$| attorneys, I exclude |$266$| attorneys who received a merit-based promotion during the transition since it is impossible to separate the endogenous pay raise from the exogenous rule-based pay raise. The final sample for the IV estimation includes |$470$| attorneys who worked at the SEC before and after the transition and did not receive a separate merit-based pay raise during the process.

The instrument is defined as the log difference between the attorney’s post-transition salary to the attorney’s pretransition salary. That includes the across-the-board pay raise plus the rounding up component.¹⁰ The average pay raise was |$\$18,893$| (adjusted for inflation), or |$13.3%$| of the attorney’s pretransition salary. In the Internet Appendix, I use a more restrictive definition, where the instrument includes only the rounding up component (in logs). The average roundup component was |${\$}$|1,014 (adjusted for inflation) or |$0.7%$| of the attorney’s pretransition salary. Consequently, in that version the instrument has less power, but the results remain qualitatively similar and significant.

The instrument relies on two identifying assumptions: relevance and exclusion. For relevance, the 2002 pay raise became a permanent component of the employee’s salary. A higher raise leads to a persistently higher salary and lower incentive, which justifies the relevance assumption. I test this formally in the first stage of the 2-SLS model, and panel B of Table 5 summarizes the results. There is a significant negative correlation between the exogenous shock and the attorney’s tournament incentive, with a high |$F$|-statistic. The Internet Appendix further shows that the incentive-shock relation is weaker, albeit still significant, once the attorney is promoted to the next grade. Put differently, the exogenous shock is particularly meaningful as long as the attorney remains in the same grade.

For exclusion, the assumption is that the rule-based 2002 salary raise affects enforcement solely through the incentive, and not through any other channel. The institutional background seems consistent with this assumption. The 2002 pay raise was exogenous to the employee’s characteristics, determined by the distance from the closest pay step within the pay level. It was a unique instance in which pay raises were awarded regardless of the attorney’s tenure, skills, and accomplishments, based on a fixed formula which was equally applied to all those present. If the salary raise has a direct causal effect on enforcement, then the instrument would violate the exclusion restriction. The conceptual challenge here is that the incentive is not explicitly mentioned in the attorney’s contract. It is reconstructed based on the individual salary and a target salary. It follows that even a perfectly exogenous shock to one of those components could have direct causal effects on enforcement, on top of potential effects through the incentive. While I cannot rule out this option, an effect, if any, should go in the other direction; that is, an exogenous pay raise would increase enforcement activity (as in Dal Bó, Finan, and Rossi (2013)). While I argue that the instrument affects enforcement mainly through its effect on the incentive, I acknowledge the difficulties in identifying proper instruments that satisfy exclusion.¹¹

2.3 Summary statistics

Table 1 and Table 2 and the accompanying Figure 3 through Figure 7 summarize the key statistics. All dollar values are in |${\$}$|US (2017).

Figure 3

Compensation structure

The figure describes the compensation structure at the SEC by grade. Salaries (mean and |$95%$| confidence intervals) are in US|${\$}$|(2017). The lowest grade is SK-12, and the highest is SO-3 (see Figure 2).

Open in new tabDownload slide

Figure 4

Distribution of enforcement by hierarchy

The figure presents the distribution of enforcement over grades. panel A shows total enforcement activity by grade (total actions divided by number of attorneys), and panel B shows engagement by grade (share of attorneys filing any enforcement action).

Open in new tabDownload slide

Figure 5

Distribution of tournament incentives

The figure presents the distribution of tournament incentives. In panel A, |$Incentive$| is the ratio between current salary and target salary; the target is either the top or the median salary in the next grade. In panel B, |$Incentive$| is the median incentive within the office-grade.

Open in new tabDownload slide

Figure 6

Classification of enforcement actions

The figure describes the distribution of enforcement actions by category.

Open in new tabDownload slide

Figure 7

Geographic distribution of attorneys

The figure describes the distribution of attorney-year observations over offices. The SEC’s headquarters is in Washington, DC (DC), and its regional offices are in New York (NY), Chicago (CH), Los Angeles (LA), Boston, Miami, Denver, Fort Worth, San Francisco, Philadelphia, Atlanta, and Salt Lake City.

Open in new tabDownload slide

Figure 8

Illustration of the 2002 pay transition

Illustration of the pay raise mechanism during the 2002 transition (see Section 2.2.2). The example is based on a level-16 employee with |${\$}$|114,679 base pay in the old system. Adding 6% to |${\$}$|114,679 yields |${\$}$|121,560. In level 16 in the new pay system, the closest step above |${\$}$|121,560 is |${\$}$|122,399, which will be the employee’s new base pay. It follows that the employee received an individual pay raise of |$\$122,399-\$121,560=\$839$|⁠, on top of the 6% across-the-board pay raise.

Open in new tabDownload slide

The 3,178 enforcement actions in the sample were filed between 2002 and 2017. On average, the SEC published 186 actions during a year. The average case involves |$3.3$| defendants, including relief defendants who are not accused of wrongdoing but have received property originally obtained illegally. The vast majority of the actions (⁠|$90.4%$|⁠) are standalone, which means no follow-up action based on the same misconduct was filed during the sample period. One in three cases is settled, and the remainder (⁠|$60%$|⁠) are filed as contested actions. One in five cases is related to criminal charges brought simultaneously by the U.S. Department of Justice. One in six cases required an emergency relief in the form of asset freeze. Nearly two-thirds of the cases allege antifraud behavior (more on that in Section 3.3). Less than |$20%$| of the cases are labeled AAER and involve accountants or auditors. The average litigation team consisted of |$3.9$| attorneys, with 10 years of experience at the SEC and a combined annual income less than |$\$800,000$|⁠. One in five cases was solely litigated by DC attorneys, |$70%$| solely by regional attorneys, and the remaining by a joint team of regional and DC attorneys.

The average enforcement attorney has 9 years of experience at the SEC and earns |${\$}$|186,000 annually. The attorney’s incentive is 1.23 (1.16), relative to the top (median) salary in the next grade. The compensation structure at the SEC is concave, such that the incentives for the most senior grades are smaller on average. This seems to fall short of the optimal tournament design, which requires pay gaps between ranks to increase with the rank (Rosen 1985). Each year, one in three attorneys files at least one enforcement action (unconditional probability |$34.6%$|⁠). The probability of bringing fraud charges is slightly lower (⁠|$27.2%$|⁠), and only 1 in 10 attorneys is involved in criminally related enforcement actions. Conditional on filing enforcement actions, the average attorney brings |$2.3$| actions against |$7.8$| defendants. Every year, 1 in 10 attorneys is promoted (unconditional probability |$10.1%$|⁠). The majority of enforcement attorneys are male (⁠|$62%$|⁠), and the attrition rate among enforcement attorneys is |$5.1%$|⁠.

2.4 Enforcement and promotions

A key assumption in the paper is that enforcement positively affects promotion decisions. However, if untrue, then any incentives-enforcement correlation is spurious: even if the value of promotion increases, attorneys have no reason to invest effort in enforcement since doing so would not improve their promotion chances.

The assumption seems appropriate for the sample I study, namely, enforcement attorneys, whose promotion chances are reasonably linked to their enforcement portfolio. I test this formally in Table 3, which estimates the probability of promotion as a function of enforcement activity. All specifications include year-office-grade fixed effects, comparing attorneys who compete in the same tournament, that is, all SK-13 attorneys in Chicago in 2010. There is a significant positive relation between enforcement and winning the tournament: participating in at least one enforcement action in time |$t-1$| predicts 1.6%-2.1% increase in promotion probability at time |$t$|⁠. That relation is not explained by tenure and past promotions and it holds within employee, that is, is not affected by skills, motivation, and other employee-specific characteristics.¹²

The Internet Appendix also shows that an attorney has a 54.4% chance to be promoted at least once during his or her entire career, but this probability varies significantly between attorneys who filed any enforcement action so far and those who did not: the formers have a 58.1% chance of being promoted at least once during their career, compared to 46.4% chance for the latter, and the difference is statistically significant at the 1% level. Finally, the Internet Appendix shows how “enforcing” attorneys are promoted faster than “nonenforcing” attorneys.

To summarize, enforcement is internally rewarded: attorneys with expanded litigation portfolio are more likely to win the tournament and earn a promotion. This empirical fact leads to the main hypothesis of the paper: when the promotion value increases, attorneys should invest more effort in building their litigation portfolio. Note that I do not argue that promotions are necessarily a direct reward for enforcement. Enforcement actions could be a way for attorneys to “stand out” from the crowd in a bureaucracy and to signal their skills and determination to their supervisors. Either explanation would suffice for the purpose of this paper.

3. Results

3.1 Main result

To obtain a visual impression, Figure 9 plots the tournament incentive against the probability of participating in an enforcement action. The positive relationship between the two, at the aggregate office-grade level and also at the attorney level, is consistent with a tournament interpretation: stronger incentives lead to greater effort, which results in higher probability of enforcement.

Figure 9

Tournament incentives and enforcement activity at the SEC

The figure shows the nonparametric relation between enforcement and lagged tournament incentives, controlling for grade. |$Enforcement$| is an indicator that equals one for any enforcement.

Open in new tabDownload slide

Table 4 confirms this visual impression using regression analysis. The outcome variable is indicator which equals one for any enforcement action, or a counter for the number of enforcement actions. I start by examining individual incentives in the OLS specification (Equation (2)). I compare the litigation portfolios of attorneys throughout the SEC (year dummies) and within the same office (year-office). Finally, I compare attorneys who work at the same office, same year and same grade (year-office-grade), and compete in the same tournament for the same prize. All specifications control for recent promotion. Throughout all specification I find a significant positive relation between the attorney’s tournament incentives and his or her enforcement activities. Attorneys exhibit different propensity to participate in enforcement, depending on their tournament incentive.

Finally, I focus on the subsample of attorneys and estimate IV regressions using the |$2002$| exogenous shock as an instrument. That shock affected |$470$| attorneys, and shifted a significant portion of their salaries based on a fixed formula regardless of tenure, skills and accomplishments. I exploit this exogenous source of variation and add the same set of fixed effects as in Table 4. The results, reported in panel A in Table 5, reveal a significant positive impact of incentives on enforcement probability and total enforcement activity.

The Internet Appendix contains a set of robustness tests. For all specifications, I control for lagged enforcement and for past promotions over last 1, 2, or 3 years, and alternatively control for the length of time the attorney spent in the current grade. All the results remain qualitatively similar. For the IV regressions, I define the instrument as the roundup component instead of the full pay raise. The instrument is not as powerful and yet the results hold. I limit the sample to attorneys who were exposed to the 2002 transition and were not subsequently promoted. I exclude the attorneys from this restricted sample once they are promoted to the next grade. As explained above, as long as the attorney remains in the same grade, it is straightforward to see why the impact of the exogenous shock on the incentive persists. Indeed, even though the sample used for this test is nearly |$30%$| smaller, the main result is robust and remains highly significant. For the OLS specification, I estimate Probit model instead of linear probability model and the results remain qualitatively similar. I cluster standard errors at the grade, year, or office, as well as double-cluster at the attorney and grade, attorney and year, or attorney and office. All the results remain significant at the |$1%$| level, except for office and attorney-office, which are significant at the |$10%$| level.

3.2 Discussion

Overall, the main results support the tournament hypothesis: when pay gaps relative to the next rank increase, enforcement activity increases as well. The result holds when the pay gaps are instrumented with a plausibly exogenous shock to the SEC’s pay structure, which randomly assigned a permanent component to the attorneys’ salaries. Consider the following simplified scenario: two level-1 Boston attorneys, A and B, where A has lower salary and hence higher pay gap from the “target” salary than B. Individual salaries should commensurate with marginal productivity and therefore a priori we expect A to be less productive in terms of enforcement activity. Moreover, the low salary itself should adversely affect productivity (Dal Bó, Finan, and Rossi 2013). Finally, A’s lower salary is plausibly driven by shorter tenure and lower entry salary, and the latter is likely explained by differences in work experience, education, and negotiation skills (see Section 1.1). Because of all that we expect A to be less engaged in enforcement, but I find the opposite: attorney A is more likely to engage in enforcement. This is consistent with a tournament effect: attorney A has stronger tournament incentives, which result in higher volume of enforcement.

The economic magnitude of the effect is significant. Using the OLS point estimates, moving from the |$10$|th to the |$90$|th percentiles of the incentive distribution increases enforcement probability by |$5.5$| percentage points, or |$15.9%$| of the mean (⁠|$(1.332-1.042)*0.189/.346$|⁠). In the IV specification, which relies on a subsample of attorneys, I find that one-standard-deviation increase in the incentive leads to |$16.7$|- to |$18.6$|-percentage-point increase in probability of enforcement, which is |$46.9%-52.3%$| increase over the sample mean.

While the 2-SLS results are qualitatively similar to those of OLS, their magnitude is significantly larger. One potential explanation could have been the weakness of the instrumental variable, which may have very little linearly independent predictive power for the endogenous variable. However, this explanation seems less likely since the |$F-$|statistic for the instrument in the first stage is well above the Stock-Yogo critical values. Another possibility is that the instrument addresses an attenuation bias in the OLS specification. That bias could occur due to a potential measurement error in the endogenous variable, |$Incentive$|⁠.¹³ A downward bias in the OLS specification would also occur if an omitted variable is negatively correlated with the endogenous variable (⁠|$Incentive$|⁠) but positively correlated with the outcome [|$I(Enforcement)$|]. I will discuss this possibility in Section 3.5. Indeed, the wage-setting mechanism at the SEC suggests that “good” attorneys receive weaker incentives, which would imply a downward bias in the estimated tournament effect. Finally, as in all instrument-based estimation, the effect is identified off the compliers: attorneys whose compensation trajectory has responded to the exogenous shock (Angrist, Imbens, and Rubin 1996). Suppose that “complying” attorneys are more sensitive to promotion incentives. In that case, the coefficient in the IV specification (local average treatment effect) would be larger than the coefficient in the OLS specification (average treatment effect).

The average incentive for sample attorneys is |$23%$|⁠. To put things in perspective, Coles, Li, and Wang (2017) study within-industries tournament among CEOs and report an average incentive exceeding |$349%$|⁠. It follows that the SEC provides low-powered tournament incentives. Nevertheless, whether we consider the average treatment effect or the LATE, it appears that the impact of promotion incentives on SEC enforcement patterns is nontrivial. This is related to similar findings on the impact of wage differences among peers. Recent studies report material responses, in terms of productivity and quitting rates, to quasi-random pay gaps as small as |$5%$| (Breza, Kaur, and Shamdasani 2016) and |$10$| cents (Dube, Giuliano, and Leonard 2019). Along the same lines, the results in this paper suggest that low-powered promotion incentives could shift effort among SEC attorneys and lead to nontrivial changes in their enforcement activities.

The remainder of the paper addresses specific challenges. First, a measurement error in the outcome, for example because of heterogeneity across enforcement actions. Second, the incentive variable is not a proper measure of ex ante promotion value. Third, strong incentives are assigned to “good” attorneys who pursue enforcement due to their unobserved quality. For brevity, the results reported below use the IV specification. The Internet Appendix reports a full set of parallel results in the OLS model, which are all significant and qualitatively similar to the IV ones.

3.3 Alternative enforcement measures

In the main specification, I abstract from the nature of the enforcement cases and count them all equally toward the outcome variable. But not all enforcement actions are born equal. Some actions involve more complex legal issues, higher legal stakes and potentially larger penalties. Those distinctions are important for two complimentary reasons. First, tournament theory predicts that stronger incentives increase effort. The observed enforcement is a proxy for effort with a potential measurement error, and tightening the definition of enforcement could help reduce the measurement error (see a formal treatment in Section 2 in the Internet Appendix). Second, in the classic multitask model (Holmstrom and Milgrom 1991) an agent with multiple tasks could work harder on the more observable dimensions and neglect others. In the current setting, it is possible that powerful tournament incentives would lead enforcement attorneys to allocate effort toward less desirable enforcement actions. Empirically, no consensus has been reached on how to weight enforcement actions based on their difficulty or quality. In this section, I offer a set of measures that could capture some of those dimensions.

First, I consider the factual complexity of the case as reflected by the number of defendants named in the initial complaint. Two -thirds (⁠|$65%$|⁠) of the complaints involve two or more defendants. Multidefendant cases require greater amount of effort and also reasonably have more impact. For instance, in November 9, 2012, the SEC charged a ring of seven high school friends with insider trading in health care stocks. Tracing the long chain of tippers and “tippees” presumably required more efforts than, say, bringing a charge against a single defendant. Based on this intuition, I estimate the efficacy of tournament incentives with an outcome variable that counts the number of defendants per case (zero if the attorney filed no complaint). The results remain unchanged if I count the total number of defendants or omit relief defendants from the calculations.

Second, I consider the gravity of the allegations. One in five (⁠|$21.9%$|⁠) complaints is connected to criminal investigations, where the same nexus of misconduct led to criminal charges filed by the Department of Justice and occasionally by foreign or state-level prosecutors. Notable examples include the Ponzi scheme orchestrated by Bernard Madoff and the post-crisis settlement with UBS for enabling U.S. clients to maintain undisclosed accounts in Switzerland. Those criminally tainted cases are presumably those with the most egregious behavior, and can therefore be classified as more important than others. Practically speaking, those actions tend to involve a more complicated set of facts. In addition, I rely on the legal literature and focus on antifraud cases. Some of the violations the SEC can allege are premised on negligence or strict liability, but allegations of fraud are more challenging since they require proof of scienter. Choi, Gulati, and Pritchard (2018), based on discussions with prior Enforcement Division attorneys, conjecture that scienter cases are more complex and difficult to prove and often result in contested litigation. I follow Choi, Gulati, and Pritchard (2018) and classify cases as antifraud if they allege a provision explicitly mentioned in the Rule 506 (Rule 10b-5 and § 15(c)(1) of the Exchange Act, § 17(a)(1) of the Securities Act, § 206(1) of the Investment Advisers Act), or allege bribery (Foreign Corrupt Practices Act or (§ 13(b)(5) of the Exchange Act). For example, Rule 10b-5 is titled “employment of manipulative and deceptive devices,” and prohibits various forms of fraudulent and deceitful behavior pursuant to section 10(b) of the Securities Exchange Act of 1934. Actions pursued under the Foreign Corrupt Practices Act typically entail multiple regulators over many domestic and foreign jurisdictions that consume considerable time and resources.

Third, I consider the procedural complexity of the case and distinguish between initial and follow-up enforcement actions. A common nexus of misconduct could lead to multiple complaints filed over the course of several years. For instance, between October 2009 and January 2011 the SEC brought three insider trading actions against 14 defendants related to Galleon, a large New York-based hedge fund complex. Filing the first action in the sequence could possibly take more effort, while follow-up actions require less effort.

Lastly, I take into account different roles within the litigation team. In the tournament literature the focus is on the top executives: their effort drives the performance of the entire company, and therefore their incentives should be carefully designed. To carry on with this analogy, I study the incentives of the “CEO” of the enforcement action: the lead attorney who is in charge of the litigation effort. I utilize the order of appearance on the document and count the enforcement action only for signatory attorneys who presumably played the most dominant role. I do not count the action for attorneys who are designated as “local counsel,” “pro hac vice” (for this occasion), or “of counsel.”¹⁴

The results, relying on the exogenous shock as an instrument for |$Incentive$|⁠, are summarized in Table 6. I find that attorneys with stronger tournament incentives file complaints against more defendants to court and pursue more important enforcement cases, defined as criminally related or allegations of fraud. Omitting follow-up cases, which presumably require less effort, does not change the main results. Lastly, the incentive effect is driven by attorneys who lead the litigation effort and sign the complaints, not by those who play incendiary roles within the litigation team. In sum, attorneys with stronger tournament incentives are more likely to lead the SEC’s enforcement activities and file complex cases with grave consequences. That result seems consistent with an interpretation of incentives and effort.

3.4 Alternative incentive measures

The tournament incentive is a forward-looking measure that is not explicitly stated in the attorney’s contract. It reflects typical promotion patterns for SEC enforcement attorneys as shown in the data. For robustness, I consider a set of alternative incentive measures and obtain similar results to those reported above.

First, in the Internet Appendix I winsorize the incentive at the |$1%$|⁠, |$2%$|⁠, |$5%$|⁠, and |$10%$| levels; the results remain significant at the |$1%$| level and the coefficients in fact increase in magnitude, assuring that the results are not driven by outliers. I remove observations with effectively zero incentive, that is, when the target salary is lower than their current salary. This test essentially excludes attorneys who are at the very top of their respective grades, and the coefficients increase by a small amount.

Second, in Table 7 I consider 12 alternative incentive structures which account for different features of the SEC organizational design. First, occasionally the hierarchical structure in the office is incomplete, such that if grade |$x+1$| is unoccupied, then the incentive for an attorney from grade |$x$| is unclear. For instance, no SK-13 attorney worked at the Atlanta office during 2008. In the baseline specification, I omit those observations, but now I calculate their incentive relative to grade |$x+2$|⁠. A second complication regards SK-16, which is a unique grade at the SEC: an SK-14 attorney could ex ante be promoted to SK-15, a managerial position, or to SK-16, which is typically a nonmanagerial position albeit with better salary. For attorney in SK-15 or SK-16, the next available promotion is SK-17. In the main specification I treat SK-14 attorneys as if they compete for SK-16, which is hierarchically “closer,” and SK-15 and SK-16 attorneys as if they compete for SK-17. I now allow SK-14 attorney to aspire to SK-15 position or to be indifferent between SK-15 and SK-16, and I let SK-16 attorneys to view SK-15 as their target. Lastly, the attorney could have more moderate expectations: instead of aspiring to the top salary in the next grade, she views the median salary as more realistic. Those considerations yield various combinations, which I report in Table 7. Again, the estimation relies on the exogenous shock as instrument for |$Incentive$| and is restricted to the subsample of attorneys who were exposed to that shock. The estimated tournament effect remain similar and significant at the |$1%$| level.

Overall, the evidence in this section is consistent with tournament predictions as various measures of ex ante tournament incentives lead to similar conclusions.

3.5 Unobserved quality

A potential omitted variable is the attorney’s unobserved quality. This concern can be framed in numerous ways, but the most plausible one is that strong “incentives” are assigned to “good” attorneys, who pursue enforcement actions due to their unobserved qualifications. If this is true, then the estimated tournament effect is biased upward.

Such a concern appears inconsistent with the SEC’s institutional setting. The target salary, that is, the numerator, reflects the pay caps for each grade. Those caps are closely related to the government-wide pay schedule (GS), and I have no evidence that the SEC is pushing the target salary upward for “good” attorneys.¹⁵ The current salary, that is, the denominator, commensurates with experience which leads to the more likely conclusion that “good” attorneys have higher salaries. In sum, if anything, it appears that “good” attorneys receive weaker incentives, which implies a downward bias in the estimated tournament effect.

In the remainder of this section, I propose two observable variables which likely capture top performance: bonus and education quality. To be clear, both variables are highly endogenous. Legal education is a mix of training and selection by university admission offices. Bonus is a function of past performance as well as a separate source of motivation for future performance. I do not lean too heavily on the causal effect of bonus and education, only on their ability to proxy for the unobserved quality of SEC attorneys. Again, the estimation relies on the exogenous shock as instrument for |$Incentive$|⁠, and I restrict the analysis to the subsample of attorneys who were exposed to that shock.

I collect educational background from LinkedIn; Martindale-Hubbell, a commonly used directory for U.S. lawyers; and a generalized web search. These data are available for 81.4% of the sample. I match the education data to the rank published by U.S. News in 2018, which ranks 203 law schools from 1 (Yale) to 144. Less than one-third of attorneys (⁠|$30%$|⁠) graduated from a top-10 law school, nearly half (48%) graduated from a top-20 school, and the mean rank is 37 (of 144). Table 8 summarizes the results. Consistent with the initial intuition, attorneys with better education are more likely to file enforcement action relative to their peers. More importantly, the power of tournament incentives is not diminished by this additional control. The results remain unchanged if I replace a top-10 indicator with a top-20 indicator or a continuous variable. Note that in the latter specification (panel C) the sign flips, as higher ranked school corresponds to a lower quality.

Finally, I consider bonus recipients: employees who were judged by their superiors to perform above and beyond normal job requirements.¹⁶ The observed bonus is, by construction, correlated with the unobserved quality. Bonus distribution has virtually stopped in 2010 as part of a government-wide pay freeze, so the sample in this subsection is limited to the years 2002-2009 (including). Nearly two-thirds (61.8%) of enforcement attorneys earned a bonus during that time period. The average bonus, if earned, was US|${\$}$|(2017) 3,153, or 1.6% of the attorney’s base compensation. Table 9 summarizes the results. I include various controls based on bonus data: indicator for any bonus award; bonus as share of the base salary; and the dollar value of the bonus, conditional on receiving any, which restricts the sample even further. Regardless of the exact measurement, I find that the tournament effect remains significant at the |$1%$| level and in fact increases in magnitude.

3.6 Aggregate sample

In this section I aggregate the individual attorneys into an office-grade sample. The unit of observation in this sample is office-grade, for example, SK-12 in Boston. The aggregated sample is of course significantly smaller, but I conduct this exercise for two reasons. First, as opposed to the attorney sample, the office-grade sample abstracts from the individual salary which is driven to a large extent by the attorney’s unique characteristics. Second, conceptually, the tournament is likely designed at an aggregate level with regulatory caps and promotion patterns across pay grades. By conducting the analysis in two different levels I exploit different sources of variation, which lends more credibility to the tournament interpretation.

In this sample, the incentive is the median individual incentive. Essentially, I replace the numerator with the median salary (⁠|$incentive_{o,g,t} = \frac{RegCap_{o,g+1,t}}{MedianSalary_{o,g,t}}$|⁠). The outcome is based on the combined activity of the attorneys within the office-grade, the intensive and extensive margin (⁠|$I(enf_{o,g,t})$| and |$enf_{o,g,t}$|⁠; the latter using log plus one transformation). I also estimate |$Engagement_{o,g,t}$|⁠, the share of attorneys who participate in enforcement (⁠|$\frac{\sum_{i\in (o,g)}^{ } I(enf_{i,t})}{N}$|⁠, where |$N$| is the number of attorneys in the office-grade).¹⁷

The Internet Appendix includes descriptive statistics for the office-grade sample. The average grade includes 11 attorneys with 10 years experience at the SEC and an annual salary of |$\$196,000$|⁠. They face tournament incentives of |$1.19$| (⁠|$1.14$|⁠) relative to the top (median) salary at the next grade. The majority of grades (⁠|$73.8%$|⁠) are involved in enforcement, and half of the grades bring criminal-related actions. Conditional on filing, the average grades brought 13 actions, including 9 actions with fraud charges, 3 related to criminal charges, and 2 requiring freezing of assets. Conditional on filing, half of the attorneys were involved in enforcement.

The outcome is one of the aggregated office-grade enforcement measures, and |$incentive$| is the median incentive within the office-grade. I include office-grade controls: the number of attorneys, and the share of attorneys who were recently promoted or the median tenure. I add office-grade fixed effects, taking into account the different case assignment across grades based on relative seniority. Similar to Equation (2), I alternately add year dummies and year-office dummies. Explanatory variables are lagged and standard errors are clustered at the office-grade level.

Table 10 summarizes the results. Essentially, the table analyzes the “tournament between tournaments”: how enforcement output differs across grades as a function of the promotion-based incentives. In one set of regressions, I include office-grade indicators and year indicators. The tournament effect is then identified within the office-grade over time. In a second set I include year-office indicators, effectively comparing grades within the same office. This specifications absorbs all variation coming from local conditions, such as financial misconduct and labor market opportunities. All specifications control for the number of attorneys and the share of recently promoted attorneys, and the results show that incentives positively and significantly affect enforcement. Incentives explain the likelihood and quantity of enforcement actions, as well as the share of attorneys who participate in enforcement activity.