supplementary information for · discreet encounters. between married individuals. married dating...

1

Supplementary Information for Personal Infidelity and Professional Conduct in 4 Settings John M. Griffin, Samuel Kruger, Gonzalo Maturana John M. Griffin Email: [email protected] This PDF file includes:

Data availability Supplementary text Figs. S1 to S9 Tables S1 to S19 References for SI reference citations

www.pnas.org/cgi/doi/10.1073/pnas.1905329116

2

Data Availability:

The paper uses financial advisor information from the Financial Industry Regulatory Authority’s (FINRA’s) BrokerCheck data; information on defendants to Securities and Exchange Commission (SEC) litigation from the SEC’s litigation release archives; information on Chicago police officers from the Citizens Police Data Project; company and stock data from Execucomp, Compustat, the Center for Research in Security Prices (CRSP), MSCI, Institutional Shareholder Services (ISS), and Thomson Reuters; Ashley Madison data that was publicly posted after the company was hacked; and public records data from LexisNexis. These data are all in the public domain or available to researchers subject to contracts with the data providers, which are reasonably common.

Methodology:

Ashley Madison Data

Ashley Madison is an online dating service for married people, operating under the slogan “Life is short. Have an affair.” Their focus on facilitating marital infidelity could not be more explicit. Fig. S1 shows the ashleymadison.com homepage as of June 23, 2015. In addition to the affair slogan, the “o” in “Madison” is a depicted as a wedding ring, and the woman in the photograph is wearing a wedding ring. The service description at the bottom of the page reads in full:

Ashley Madison is the most famous name in infidelity and married dating. As seen on Hannity, Howard Stern, TIME, BusinessWeek, Sports Illustrated, Maxim, USA Today. Ashley Madison is the most recognized and reputable married dating company. Our Married Dating Services for Married individuals Work. Ashley Madison is the most successful website for finding an affair and cheating partners. Have an Affair today on Ashley Madison. Thousands of cheating wives and cheating husbands signup everyday looking for an affair. We are the most famous website for discreet encounters between married individuals. Married Dating has never been easier. With Our affair guarantee package we guarantee you will find the perfect affair partner. Sign up for Free today. (Emphasis and capitalization theirs.)

To validate that AM users are married, we conduct in-depth investigations of the 94 CEO and CFO AM transaction users in our sample. CEOs and CFOs are more conducive for this investigation than our other samples because they are frequently public figures with more available biographical information. Of the 94 CEOs and CFOs with Ashley Madison transactions, 87 have at least one property record on LexisNexis that is co-owned with a woman who uses the same last name. Of the remaining seven CEOs and CFOs, four have evidence of marriage in other biographical sources. For the last three, we were unable to find any evidence as to whether or not they were married. Overall, the evidence indicates that at least 97% (91 of 94) of CEO and CFO Ashley Madison users in our sample were married at least at some point. There is not enough publicly available data to determine the precise timing of these marriages.

Ashley Madison’s scope is staggering. Their website boasted 36,705,000 anonymous members as of June 23, 2015. Despite Ashley Madison’s claim, many of those members are no longer anonymous. In July and August of 2015, a group calling itself “The Impact Team” hacked into

3

Ashley Madison’s computer system and publicly posted Ashley Madison usage data, transaction data, and company documents, including records on 36 million user accounts.

To ensure that our measures are as accurate as possible, our analysis focuses on the subset of Ashley Madison users for whom we have transaction data. While signing up for Ashley Madison and creating a profile is free, users are charged for many activities, including sending and receiving messages (5 credits per message) and chatting with other users (30 credits for a 30-minute chat session). For $49, users can purchase 100 credits. For $249, they get an “affair guarantee.”

The Ashley Madison transaction data spans March 21, 2008 to June 28, 2015 and includes 9.7 million transaction records, representing 4.1 million individual transactions. Each transaction typically has multiple records, for example separate credit card authorization and settlement records. These purchases are associated with 1.4 million user accounts. We focus on the 1.0 million transaction users in the United States. The median AM purchase size is $98.

Ashley Madison usage is widespread throughout the United States. Fig. S2 maps per capita paid AM usage by core-based statistical area (CBSA). Paid usage rates represent the number of paid users in a CBSA divided by total population from the 2010 census. Paid AM usage is slightly higher in the Northeast, but similarly high usage rates occur in other regions, including the metro areas of Atlanta, Austin, Dallas, Houston, Denver, Salt Lake City, and Seattle. AM usage appears to be higher in larger CBSAs. For example, paid usage rates in the top 20 CBSAs range from 0.32% and 0.52%, with an average of 0.42%, compared to the national paid usage rate of 0.33% (Table S1). Paid AM usage by CBSA and county is available upon request for researchers interested in this as a potential measure of regional culture.

In addition to reflecting active usage, Ashley Madison transaction data has the benefit of including billing names and addresses. We use this information to merge the Ashley Madison data with financial advisors, SEC white-collar crime defendants, and Execucomp executives of U.S. public companies. We search LexisNexis to find addresses associated with each person in the sample. Then, to count as an Ashley Madison transaction match, we require a match on both name and address. We exhaustively search LexisNexis for misconduct financial advisors, SEC defendants, CEOs, and CFOs and find profiles we can confirm as correct for 85% of the individuals in our samples. Specifically, we find 95% of police officers, 88% of financial advisors, 62% of SEC defendants, 84% of CEOs, and 83% of CFOs in LexisNexis. LexisNexis profiles are confirmed based on search uniqueness, employment information, and other identifiable information. We treat our AM usage indicator as missing for individuals that we are not able to find in LexisNexis. For executives who are not CEOs or CFOs, we decrease the number of required LexisNexis searches by first matching executive names to Ashley Madison names with transaction billing addresses within 50 miles of executives’ corporate headquarters and then searching LexisNexis to check whether these potential matches represent actual executive Ashley Madison usage. Using this process, we identify paid Ashley Madison usage for 44 misconduct financial advisors (3.3% of the sample), 18 SEC defendants (4.0% of the sample), and 214 executives (1.2% of the sample), including 47 CEOs (1.8% of the sample) and 48 CFOs (1.7% of the sample).

Chicago Police Data

The police data come from the Citizens Police Data Project, which collected detailed information on Chicago police officer misconduct. In addition to complaints against Chicago police officers, the data include full work histories, names, and enough identifying information to find the police

4

officers in LexisNexis searches. The work history information includes which of Chicago’s 25 geographic districts an officer was assigned to at any point in time. Officers that were assigned to other more specialized districts are combined into a single general purpose district for purposes of our analysis. We focus on male police officers who joined the police department by 2000 and served at some point between 2010 and February of 2018 when the data end. This results in a sample of 7,198 officers.

We define misconduct officers as anyone with a sustained complaint (i.e., a complaint that resulted in disciplinary action or a reprimand) or at least 5 complaints in 2010-2018. This results in a misconduct sample of 1,343 police officers, 19% of the original sample. We then match each misconduct officer to a control officer who has never had a sustained complaint and had no more than one complaint during 2010-2018. Matched officers work in the same district at the time of the misconduct officer’s first 2010-2018 complaint and have age and experience differences of no more than five years, with matches prioritized to minimize age and experience differences. These matching criteria result in matched control officers for 1,063 of the 1,343 misconduct officers. We successfully find both officers in LexisNexis for 960 pairs, resulting in a final sample of 960 misconduct police officers matched to 960 control officers.

Financial Advisor Data

Financial advisors play an important role in shaping the financial decisions of millions of individual investors. Due to federal and state regulations, FINRA collects and publishes individual-level data on financial advisor licenses, certifications, employment, and detailed disclosures of everything from customer disputes to criminal convictions.

Existing research finds that misconduct by financial advisors is related to both an individual’s and a firm’s past professional misconduct (1). We follow the methodology and misconduct definition used by this research and build a dataset of all individuals employed as financial advisors during 2015 and 2016. The data consist of 736 thousand financial advisors, 6.7% of whom have misconduct on their records and 0.8% of whom have misconduct in 2015 to 2016. To generate a sample that is similar to Ashley Madison’s target audience, we further restrict our sample to male advisors who have been registered since at least 2000. This results in a sample of 2,376 misconduct financial advisors. Of the 736 thousand financial advisors employed in 2015 or 2016, 5,771 (0.8%) have misconduct during their 2015 or 2016 employment. 4,613 (80%) of these financial advisors are male. Restricting the sample to financial advisors who have been registered since 2000 reduces it to 2,361 financial advisors. We then match the misconduct financial advisors to a control group of financial advisor with no misconduct. The matches are by firm, county, year, an indicator for whether the person passed a Series 65 or Series 66 exam, and years of experience (with a maximum experience difference of five years and final matches selected to minimize experience differences). This methodology controls for any firm-county-year effects. We additionally match on whether a person has passed a Series 65 or 66 examination because advisors with this qualification are more likely to work with retail clients and financial advisors who hold these exams have higher rates of misconduct (1). The resulting sample has 1,574 unique pairs of misconduct and control financial advisors. We found LexisNexis contact information for 1,319 of the pairs, which represent our final matched sample.

We then use LexisNexis data on all addresses a person is associated with in public records to match financial advisors to names and billing addresses in the Ashley Madison transaction data. LexisNexis public records include property addresses and mailing addresses associated with

5

everything from real estate transactions to hunting and fishing licenses. Ashley Madison billing addresses come from credit card transaction data. In addition to verified AM transaction users, we also examine AM usage among all those who opened an account (most of whom never had a paid transaction with AM). The AM non-transaction user data does not have full address information and is usually missing names so these matches are based email address and zip code. LexisNexis profiles include email addresses associated with an individual.

SEC Defendant Data

We next turn to an even more egregious form of professional misconduct, fraud and white-collar crime alleged in SEC lawsuits. We identify defendants to all civil litigation initiated by the SEC between 2010 and 2015 by reviewing complaints filed by the SEC in federal court, which are available in the SEC’s litigation release archives. These are civil lawsuits alleging criminal activity such as insider trading, Ponzi schemes, pump and dump operations, and other financial fraud. Starting with 1,063 defendants who live in the United States, we locate LexisNexis public records information for 663 individuals. We then use this data to match SEC defendants to Ashley Madison transaction data, with matches based on name and mailing address. For benchmarking purposes, we match each SEC defendant to two control samples: financial advisors with no misconduct on their records and Execucomp CEOs and CFOs serving during the same time period. The financial advisor matches are to advisors working in the same county with the same gender and age within five years. Unique one-to-one matches are selected to minimize expected age differences based on financial advisor start dates with actual age differences verified based on public records age data. SEC defendants are also matched to a unique CEO or CFO with the same gender, age within 5 years of the SEC defendant’s age, and whose company is located within 50 miles of the SEC defendant by first minimizing age difference and then minimizing geographic distance. The resulting matched samples consist of pairs of 613 SEC Defendants matched to misconduct-free financial advisors and 569 SEC defendants matched to CEOs and CFOs.

Corporate Data

Our indicators for corporate infractions come from security class action filings and financial restatement records. These indicators are commonly used as proxies for corporate misconduct (2, 3-5). The Accounting and Auditing Enforcement Releases (AAER) database, which reports information on SEC investigations about financial misconduct, is also frequently used in this literature, but we are unable to use this data because it includes few observations after 2012, and most AM usage is concentrated in these later years.

Like previous studies, our data on class action lawsuits come from the Securities Class Action Clearinghouse (SCAC) database. In our sample, 286 firms were subject to class action lawsuits alleging misconduct during the 2008 to 2014 time period. Our securities class action indicator variable takes a value of one in all firm-years affected by the alleged misconduct. For our baseline specifications, we ignore class action lawsuits that have been dismissed in order to avoid cases where a firm is falsely accused. This results in 181 settled or ongoing lawsuits, which affect 3.17% of the firm-year observations in our sample. Our main results remain unchanged if we also include dismissed cases.

Data on financial restatements come from Audit Analytics’ Non-Reliance Restatements file, which includes corrections to financial statements with significant errors disclosed to the SEC. Our restatement indicator takes a value of one in all firm-years that were restated for non-accounting

6

reasons. Because accounting-related restatements are frequently due to new interpretations or guidance on accounting rules as opposed to firm-level actions, we follow the prior literature and drop accounting-related restatements (4). In our sample, 123 firms have non-accounting restatements, which affect 2.69% of firm-year observations. The correlation between class actions and financial restatements is 0.108. For most of our tests, we use a combined corporate infraction indicator that takes the value of one for firm-year observations that were affected by either a class action or a restatement. The mean value of this variable, Corporate Infraction, is 5.46%.

Our analysis is at the firm-year level and employs a merged panel dataset of Ashley Madison usage, executive characteristics, firm characteristics, and infraction data for 9,981 firm-year observations of 1,817 unique firms between 2008 and 2014. Panel B of Table S5 summarizes the data and compares the 213 firm-years with Ashley Madison CEOs or CFOs to the 9,768 firm-years without a CEO or CFO matched to Ashley Madison. A CEO or CFO is considered to be an Ashley Madison user if they are matched to a transaction that occurred before or during the firm year being considered. Our indicator variable for firm-years with an AM CEO or CFO (AM CEO/CFO) is missing when we do not identify the CEO or CFO in Ashley Madison and are unable to find a confirmed LexisNexis profile for either the CEO or CFO. This results in dropping approximately 24% of the firm-year observations. County-level AM usage is similar for firms with and without AM CEOs and CFOs, and AM CEOs and CFOs are similar to other CEOs and CFOs on average age and CEO tenure. Firm characteristics, including size (log of book assets), return on assets, Tobin’s Q, investment rate, acquisition rate, R&D activity, leverage, and dividend payouts are also largely similar across AM CEO/CFO and non-AM CEO/CFO firm years. By contrast, corporate infractions are significantly more common in AM CEO/CFO firm years. Of AM CEO/CFO firm years, 8.9% have a class action lawsuit, compared to a 3.0% frequency for non-AM CEO/CFO firm years. Similarly, 6.6% of AM CEO/CFO have a financial restatement compared to a 2.6% frequency for other firm-years. Overall, 11.7% of AM CEO/CFO firm years have an infraction of some kind, compared to a 5.3% frequency for other firm-years. These differences are all statistically significant with standard errors clustered by firm.

While our methodology for finding executives in the Ashley Madison data is entirely based on publicly available data, we believe we may be the first to actually do this match. In particular, we perform a news search on executive name and “Ashley Madison” for all the CEOs and CFOs in the data and do not find mentions in the press of Ashley Madison usage by these executives. This is somewhat surprising because some of the firms with a CEO or CFO in AM are quite prominent, including major financial services, pharmaceutical, defense, insurance, and retail firms. Nine firms with a CEO or CFO in AM have market capitalizations above ten billion dollars and 56 have market capitalizations above one billion dollars. Additionally, retention rates after the Ashley Madison data was posted are the same for AM and non-AM CEOs and CFOs. (Retention rates from 2015 to 2016 are 89% for AM CEOs and CFOs and 86% for non-AM CEOs and CFOs.) Nonetheless, it is possible that market participants analyzed these executives. Stock prices generally react negatively to news of scandals, including sexual misadventures, suggesting that we should expect AM firms to have negative abnormal returns if hedge funds or other investors are using this information in a large-scale manner (6). In Fig. S4, we examine stock returns around the date of the Ashley Madison leak for firms with CEOs and CFOs in the data and find no evidence of abnormal returns.

In Fig. S3, we explore whether AM CEOs and CFOs are concentrated in certain industries. The figure shows the fraction of firms with a CEO or CFO who is a paid Ashley Madison user along

7

with a 95% confidence interval for each of the Fama-French 12 industry classifications. To obtain the estimated probabilities of AM usage by industry, we regress the indicator for AM CEO/CFO on industry fixed effects. Confidence intervals are computed using heteroscedasticity-robust standard errors, clustered by firm. Levels are obtained by adding the amount of AM CEO/CFO from the omitted category (Business Equipment) to each fixed effect. Consequently, the confidence interval for Business Equipment is not reported. The highest probabilities are for Manufacturing and Healthcare, with estimated values of 5.1% and 2.7%, respectively. However, most of the differences across industries are not statistically significant. The only statistically significant differences are between Manufacturing and Chemicals, and between Manufacturing and Consumer Durables. Overall, AM CEOs and CFOs do not seem to be systematically concentrated in any particular industry.

Supplementary Results:

Police Officers

Panel A of Fig. S5 plots overall AM usage for misconduct and matched non-misconduct police officers for misconduct officers with and without a sustained complaint. Both groups have elevated Ashley Madison usage with differences that are significant at the 1% level. Panel B of Fig. S5 groups misconduct officers by the severity of the complaints made against them. These complaints are limited to personnel and traffic complaints for 139 officers. Another 139 officers have at least one complaint that falls into a more severe category such as illegal search, lockup procedures, verbal abuse, false arrest, conduct unbecoming, drug and alcohol abuse, or first amendment violations. Finally, 258 officers have at least one complaint of the highest severity, consisting of complaint categories involving use of force, criminal misconduct, domestic abuse, bribery, official corruption, racial profiling, or excessive force. AM usage differences are concentrated entirely in the most severe complaint categories. To ensure that domestic abuse and other off-duty complaints are not driving the results, we drop them in Panel C of Fig. S5 by excluding misconduct officers who would not have been classified as having misconduct without domestic abuse and other off-duty complaints. Results are unchanged.

Financial Advisors

If personal conduct is related to firm culture, Ashley Madison usage could vary across firms. The results in Table 1 control for any firm effects by matching financial advisors within firms. Nonetheless, we are interested in whether Ashley Madison usage varies across firms. Past research indicates that misconduct rates vary across firms (1). In particular, within the universe of firms with at least 1,000 financial advisors, the top ten firms have misconduct rates in excess of 13%, whereas the bottom ten firms have misconduct rates below 1.75%. The high misconduct firms are Oppenheimer & Co., First Allied Securities, Wells Fargo Advisors Financial Network, UBS Financial Services, Cetera Advisors, Securities America, National Planning Corporation, Raymond James & Associates, Stifel, Nicolaus & Company, and Janney Montgomery Scott. The low misconduct firms are Morgan Stanley, Goldman Sachs, BNP Paribus Securities, Suntrust Robinson Humphrey, Blackrock Investments, UBS Securities, Jeffries, Prudential Investment Management Services, Wells Fargo Securities, and Pershing. Analysis is based on financial advisors employed in May 2015.

We analyze the high and low misconduct firms to test whether they differ in their rates of Ashley Madison usage. For this analysis we match high misconduct firm employees to low misconduct

8

firm employees by county, years of experience (with a maximum difference of five years), and an indicator for whether the person has passed a Series 65 or Series 66 exam. After searching for financial advisors in LexisNexis, this results in a sample of 1,837 matched pairs. Table S4 reports the results. Ashley Madison usage is essentially the same at high and low misconduct firms. While there may be important cultural differences across firms, these do not translate into significant differences in personal Ashley Madison usage. This result may be surprising at first glance because misconduct is associated with Ashley Madison usage at the individual level. However, since the high misconduct firm sample has a misconduct rate of 14.9% compared to 3.3% for the low misconduct firm sample, this effect alone is only 0.4 percentage points (3.4% AM usage difference between misconduct and non-misconduct advisors times 11.6% misconduct rate difference between high and low firms). To observe significant differences in Ashley Madison usage across firms, Ashley Madison usage would have to relate to firm culture more generally.

While high and low misconduct firms do not have different rates of Ashley Madison usage, it is worth noting that Ashley Madison transaction rates for both high and low misconduct firms (1.9% and 2.3%, respectively), are elevated relative to the AM transaction rate for the general population, which is 1.2% after controlling for gender and age. We calculate the general population Ashley Madison transaction usage rate for every combination of gender and age, and then take the average across all of the gender-age combinations in our sample (which is entirely male in this case). For our general population comparisons, we focus entirely on transaction users because without knowing which non-transaction AM accounts are real, we cannot accurately estimate the non-transaction AM usage rate of the general population.

SEC Defendants

Fig. S7 plots SEC defendant and matched CEO/CFO AM usage rates by type of infraction alleged in the SEC complaint. AM usage is elevated relative to matched CEOs and CFOs for all types of fraud. However, due to the small sample sizes, the differences are independently significant only for the other fraud category, which includes general securities and accounting fraud.

Corporate Infractions

Our findings on corporate infractions are most closely related to recent papers on the relation between personal characteristics of executives and firm misconduct. This literature finds that CEOs and CFOs of companies that perpetrate corporate fraud are more likely to have legal records of traffic and criminal convictions compared to CEOs and CFOs of similar firms (7, 8). CEOs with past military experience are less likely to be involved in corporate fraud, consistent with military service instilling a strong sense of ethics (2). Public announcements of accusations of dishonesty, substance abuse, sexual adventure, and violence by executives are more likely from poorly governed firms and are associated with negative returns, managed earnings, future corporate lawsuits and investigations, and a future loss of salary for the executives (6). Relatedly, corporate infractions are more common at companies with CEOs who have high testosterone levels (9). Our analysis of infidelity spans three different professional settings and explores a new dimension of unethical personal behavior that is of widespread interest and practical importance and is intimately connected to personal trust and honesty. In contrast to most previous work, our measure of personal misconduct captures behavior that violates trust but is not illegal. Analyzing financial advisors and SEC defendants allows us to observe the connection between personal and professional conduct beyond the managerial settings previously analyzed.

9

The role of personal ethics in professional settings also relates to a relatively new literature on the importance of firm culture (10-12). Several recent papers consider geographic determinants of firm culture (4, 13-16). As a cultural proxy for regional differences in personal cheating, we examine Ashley Madison transaction usage as a percentage of the general population in each county. We find that county AM usage is somewhat predictive of corporate infractions. However, executive AM usage is uncorrelated with county AM usage, and our results are unaffected by controlling for county AM usage, indicating that firm and individual behavior can be quite distinct from regional proxies.

Previous research finds that corporate infractions are more common at firms whose CEOs have personal legal infractions based on detailed private investigator investigations of a sample of CEOs with and without corporate infractions (7). Replicating this for our full sample of all Execucomp CEOs and CFOs is likely not possible. Nonetheless, we collect similar data based on criminal records available from LexisNexis. While this data is limited by the fact that it often lacks important details, and it is not clear how complete the data are for all states based on LexisNexis documentation, it is similar to the type of legal infractions previously studied. 52% of these infractions are traffic-related, and 42% have no information about the infraction. Table S7 reports regression results controlling for personal legal infractions. Including this control variable has no impact on the Ashley Madison CEO/CFO coefficients.

To further explore the relation between personal conduct and corporate misconduct, we repeat our baseline logistic estimations using indicators for class action lawsuits and financial restatements as separate dependent variables. Results are reported in Table S8. In Panel A, the dependent variable is Class Action, a dummy that takes the value of one for the firm-years affected by a class action lawsuit. In Panel B, the dependent variable is Restatement, a dummy that takes the value of one for the firm-years affected by a financial restatement. The marginal effect of AM CEO/CFO is economically large and statistically significant for both class action lawsuits and financial restatements. In particular, in the column (4) regressions with our full set of control variables, the marginal effect of CEO/CFO AM usage is 4.8 percentage points (significant at the 1% level) for class action lawsuits and 3.7 percentage points (significant at the 5% level) for restatements. As with the combined indicator, these effects imply that having an AM CEO or CFO doubles the probability of infraction (the unconditional probability of class action and restatement are 4.1% and 3.5%, respectively).

So far, we have jointly considered firms with AM CEOs and AM CFOs. This gives us a larger sample of firms with AM users and increases the power of our tests, but it comes at the cost of hiding any differences between CEOs and CFOs and of losing firms whose CEOs or CFOs we do not find in LexisNexis. We consider the effects of AM CEOs and CFOs on corporate infractions separately in Panels A and B of Table S9, respectively. The CEO and CFO marginal effects are significant and similar to one another in the first two specifications. In specifications (3) and (4) with industry and state fixed effects, the AM CEO coefficient shrinks moderately, and the AM CFO is moderately larger. With the full set of control variables in column (4), the marginal effects are 3.8 percentage points for CEOs and 6.7 percentage points for CFOs, and the CEO effect is not statistically significant. To further understand this result, in Table S10, we repeat the estimations in Table S9 for class actions and financial restatements separately. The AM CFO marginal effect is a significant 5.8 to 5.9 percentage points for both types of infractions, whereas the AM CEO marginal effect is significant (3.9 percentage points) only for class action lawsuits. Overall, the

10

results show that CEO and CFO AM usage both predict corporate infractions and suggest that CFOs are at least as important as CEOs, particularly for financial restatements.

Interpretation:

The results in the previous section show a strong relation between corporate misconduct and the personal ethics of CEOs and CFOs, and this relation is not explained by county AM usage, executive age, CEO tenure, executive gender, firm characteristics, or year, industry, and state fixed effects. There are three leading explanations for this finding. First, unethical CEOs and CFOs cause corporations to engage in questionable corporate practices. Relatedly, CEOs and CFOs who engage in corporate misconduct may become more unethical in their private lives. These explanations are similar in that there is a personal characteristic that drives both behaviors. Second, firm culture, which may in part reflect the culture of the local geographic area, jointly influences executive personal conduct and corporate infractions. Third, some other omitted firm characteristic could drive both AM executive usage and corporate infractions. In particular, corporate boards may screen executives for ethical characteristics related to Ashley Madison usage that we do not observe. If this is the case, companies inclined to engage in unethical conduct may endogenously match with unethical executives.

We examine these potential explanations in four ways. First, we investigate the firm culture channel by analyzing AM usage of executives other than CEOs and CFOs and regional characteristics that may be related to culture. Second, we analyze the relation between CEO and CFO AM usage and firm decisions that are not typically associated with firm misconduct. Third, we use propensity score matching as an alternative approach to control for firm heterogeneity. Fourth, we consider additional characteristics, including CEO and CFO optimism, corporate governance, returns, return volatility, and accounting patterns potentially associated with misreporting. We also consider two alternative measures of AM usage and sample extensions. None of these methods is a precise causal test, but each helps to better understand the relation between AM usage and corporate misconduct.

Firm Culture

We control for geographic and industry cultural differences in our baseline regressions by including county AM usage, state fixed effects, and industry fixed effects, but culture is still likely to vary across firms even after controlling for these characteristics. If this is true, less ethical firms might hire unethical executives and exhibit more corporate misconduct even if the personal conduct of executives has nothing to do with corporate misconduct.

To assess whether unobserved firm culture is driving our results, we analyze AM usage by executives other than CEOs and CFOs. The other executives we examine are top executives included in Execucomp data, including Chief Operating Officers, Chief Marketing Officers, and Chief Technology Officers. Our sample includes 15,360 non CEO/CFO executives compared to 2,654 CEOs and 2,797 CFOs. Firm culture should affect not just CEOs and CFOs, but also other top executives. Moreover, we should obtain a more precise measure of corporate culture with other executives because our non-CEO/CFO sample is approximately three times as large as our sample of CEOs and CFOs. Column (1) of Table S13 reports results for regressions of corporate infractions on both CEO/CFO and other executive AM usage. Other executive AM usage is unrelated to corporate infractions, and including it as a control variable has no impact on the AM CEO/CFO marginal effect, which is 4.9 percentage points, compared to 5.0 percentage points in

11

the baseline (Table 3) specification. The lack of explanatory power of other executive AM usage is not due to the inclusion of AM CEO/CFO in the regression. The coefficient on AM other executive and its statistical significance remain unchanged when AM CEO/CFO is excluded. The fact that other executive AM usage is not related to corporate infractions while CEO/CFO AM usage is, suggests that either CEO and CFO AM usage is a much stronger proxy for firm culture than other executive AM usage, or CEO and CFO ethics directly affect firm corporate actions. Either way, personal and professional conduct are tightly connected, and CEOs and CFOs play an important role in corporate culture and conduct.

To investigate the relation between AM usage and local culture, we examine how AM usage varies with political corruption, religious adherence, and regional demographics. Following the literature, we measure political corruption at the federal judicial district level as federal public corruption convictions between 2004 and 2013 per one million residents (17). Because it is not comparable to other federal judicial districts, we drop Washington, DC from this analysis. Public corruption convictions have been shown to be correlated with the incidence of financial misconduct for firms headquartered in the region (13). Religious adherence represents the percent of the CBSA’s population with a religious affiliation in 2010, as reported by the Association of Religious Data Archives. This measure is similar to religiosity measures used in other papers (4, 14, 15). In Fig. S9, we plot CBSA-level overall and executive AM usage versus political corruption and religious adherence and find that they are largely unrelated. Overall AM usage is moderately negatively correlated with political corruption within the top 50 CBSAs, but this is entirely driven by New Orleans and Memphis. Executive AM usage is moderately positively correlated with political corruption, but this is also driven by two outliers. Overall and executive AM usage are both uncorrelated with religious adherence. Table S1 further describes and summarizes the data.

In columns (2) to (4) of Table S13, we examine adding regional variables potentially related to culture to our baseline regressions. In column (2), we add county religious adherence and federal judicial district political corruption. The inclusion of these cultural measures does not affect the relation between corporate infractions and CEO/CFO AM usage, which has a marginal effect of 5.2 percentage points in this specification. County AM usage also continues to significantly predict corporate infractions. Column (3) adds the log of the CBSA population and county-level college education rate and median household income (based on the location of the firm’s headquarters) to the baseline specification. The marginal effect of AM CEO/CFO again remains virtually unchanged. By contrast, the marginal effect of county AM usage decreases and loses its statistical significance. Finally, when including all the regional control variables together in column (4), the marginal effect of AM CEO/CFO remains strong and highly significant at 5.3 percentage points.

Ashley Madison Usage and Other Corporate Decisions

CEO and CFO AM usage could be related to non-fraud corporate decisions either because unethical CEOs and CFOs manage their firms differently or because AM usage is related to omitted firm characteristics. A large and growing literature shows that CEOs effect firm decisions (18). Specific examples include the effect of CEO perks and ability on firm performance (19, 20); the effect of CEO personality, overconfidence, formative experiences, and personal behavior on corporate investment, leverage, and risk (21-27); the effect of CEO military service on corporate decisions and performance (2, 28); and the effect of CEO personal tax aggressiveness on corporate tax avoidance (29).

12

To assess this possibility, we study the relation between having an AM CEO or CFO and corporate decisions that are not typically associated with corporate misconduct. Specifically, we analyze Investment (CAPEX divided by lagged total assets), Acquisitions (total value of acquisitions divided by lagged total assets), Dividend Payout (the sum of preferred and common dividends paid divided by lagged total assets), R&D (research and development expenses divided by lagged total assets), and CEO/CFO Compensation (the natural log of total CEO and CFO compensation), all of which are analyzed in other studies of the impact of executives on firm actions. Once again, the explanatory variable of interest is AM CEO/CFO. Results are presented in Table S18. All specifications are OLS regressions with executive and firm controls, as well as year, 2-digit SIC, and state fixed effects. Standard errors are clustered by firm. Having an AM CEO or CFO has no effect on acquisitions, dividend payouts, research and development expenses, or compensation. However, we find that firms with AM CEO/CFOs have lower investment rates. This is consistent with unethical CEOs and CFOs being less aggressive in their investment decisions and cuts against interpreting AM usage as reflecting risk-taking, but we caution about reading too much into this result given the lack of any significant relation to the four other variables analyzed in Table S18.

One could also ask how CEO and CFO AM usage is related to firm performance. We explore the relation between stock returns and CEO/CFO AM usage in Table S19. We do not find a significant relation between a firm’s annual stock return and AM CEO/CFO. However, the noise inherent in firm-level stock returns limits the power of this test (standard errors in these regressions are in the range of 4 to 5 percentage points).

Propensity Score Matching and Additional Firm Characteristics

Given the potential for unobserved firm characteristics to influence our results, we would ideally like to include firm fixed effects in our regressions. However, with our limited sample of CEO and CFO AM users, we lack power to estimate this specification. Concerns about heterogeneity across firms are mitigated by the similarities across firms we have already documented and by the control variables included in our regression specifications. To assess the robustness of our results and control for firm heterogeneity as well as possible, we conduct propensity score matching analysis and consider additional firm characteristic covariates. We also consider alternative measure of AM usage and expanded samples.

To control for differences in firm characteristics, we match each firm-year observation with an AM CEO or CFO to a similar firm-year observation without an AM CEO or CFO. Specifically, we match based on year, Fama-French 12 industry classification, and propensity scores estimated using a logistic regression with executive age, executive gender, CEO tenure, firm size, ROA, Tobin’s Q, investment, acquisitions, and dividend payouts. Results of the logistic propensity regression are reported in Table S11. The matching uses the nearest neighbor technique (one-to-one). We require the differences between the propensity scores of the treatment (AM CEO/CFO) and control (Non-AM CEO/CFO) groups to be at most 0.5% to be considered a match. This results in matching 87% of the firm-year observations with AM CEOs or CFOs. Table S12 shows that the matching procedure achieves the intended objective of reducing the differences in firm characteristics between the two groups of firms. Table S12 also reports corporate infraction rates and the average treatment effect (ATT) for firm-years with AM CEO/CFOs compared to matched firms. The corporate infraction rate of firms with AM CEOs or CFOs is three times as large as the infraction rate for the matched sample, and the average treatment effect of 9.2 percentage points

13

is statistically significant at the 5% level (with standard errors clustered by AM firm). In short, the matching results validate our previous findings.

We now turn to controlling for additional firm-level measures of risk and governance. These variables are not included in our baseline regressions because of concerns about potential endogeneity, missing values, and redundancy with the control variables and fixed effects already included in the baseline specifications. We first consider CEO/CFO optimism, which we calculate following the previous literature (30). Specifically, we consider a CEO or CFO as optimistic if the executive holds stock options that are more than 100% in the money in at least two years during the sample period. Once an executive is categorized as optimistic, we keep him or her under that classification for the rest of the sample period. We define the indicator variable Optimistic CEO/CFO as one when the firm has either an optimistic CEO or an optimistic CFO. As reported in the first column of Table S14, CEO/CFO optimism is unrelated to corporate infractions, and including it has no impact on the relationship between corporate infractions and CEO/CFO AM usage. Corporate infractions and AM usage could also be related to stress and pressure induced by firm performance. In column (2), we control for returns and return volatility contemporaneously and in the previous year. Past return volatility is negatively related to corporate infractions. In contrast, current return volatility is positively related to corporate infractions, and current stock returns are negatively related to infractions. Controlling for these return variables has no impact on the relation between either CEO/CFO AM usage or county-level AM usage and corporate infractions.

In column (3), we examine the importance of the E-index corporate governance measure (31). The E-index has a small and statistically insignificant coefficient, and its addition has no impact on the AM CEO/CFO coefficient. In column (3), we control for F-score, which predicts financial misstatement probability based on accounting data using coefficients from past research applied to our 2008-2014 sample of firms (32). F-score has a small positive relationship with corporate infractions, but including it has essentially no impact the relation between corporate infractions and CEO/CFO AM usage. E-index is available for 75% of the observations in our sample, and F-score is available for 71% of the observations in our sample. Where missing, we replace E-index and F-score with industry averages to avoid reducing the sample size.

In Table S15, we consider a broader set of governance-related measures that have been commonly used in the literature, including executive stock option value and firm ownership percent (33), product market competition (34), audit committee size and independence (33), and institutional ownership (35). Director data is available for 79% of the observations in our sample, and CEO/CFO ownership is available for 77% of the observations in our sample. Where missing, we replace director data and CEO/CFO ownership with industry averages. Specifically, in column (1), we consider executive stock option value and percentage of firm ownership. In column (2), we consider product market competition based on industry sales Herfindahl-Hirschman index (HHI). Because competition is calculated at the industry level, this specification does not include industry fixed effects. In column (3), we consider the size of the firm’s board of directors and the audit committee, as well as the percentage of non-independent directors in each. Finally, in column (4), we consider institutional investor firm ownership. The effect of AM CEO/CFO usage remains unaffected by the inclusion of these controls. We also assess whether stronger governance measures moderate the relation between AM usage and corporate infractions. We focus on E-index because it is our main governance variable, and we also consider audit committee size and percentage of non-independent directors in the audit committee because these two variables are

14

the most significant in Table S15. In Table S16, we interact AM CEO/CFO with indicators for above-median values of these variables. The effect of AM usage does not vary with E-index or the fraction of non-independent directors on the audit committee. We find some evidence that the effect of AM usage on corporate infractions is somewhat mitigated in firms with larger audit committees. The lack of interaction effects for E-index and non-independent audit committee members may be in part due to the limited power of these regressions. For example the standard error on the E-index interaction term is 0.038, which is over half of the baseline AM effect. Thus, we cannot rule out the governance may have important moderating effects on AM usage.

In Table S17, we consider a broader definition of AM usage based on email address matches to AM user data in addition to our more restrictive transaction data matches. We also relax the requirement that AM transaction usage occur before or during the year being considered. Specifically, we redefine our AM usage variable such that CEOs and CFOs in the AM transaction data are AM users in all years, even those prior to their AM usage. We also consider expanded samples with 2015 observations and non-Execucomp firms. Corporate infractions for 2015 are less reliable because it frequently takes multiple years for infractions to be identified. Non-Execucomp data on CEOs and CFOs comes from MSCI’s GMI data, which does not appear to be as accurate as Execucomp data. The generalizations and sample expansions result in slightly lower marginal effects, but the results remain large and statistically significant.

Interpretation Discussion

A leading interpretation of our results is that CEOs and CFOs who violate ethical norms in their personal lives are more likely than other CEOs and CFOs to violate ethical norms and legal restrictions in their professional lives, thereby making the companies they manage more prone to corporate misconduct. If personal ethics influence professional ethics, the strong effect from CEOs and CFOs is what one would expect given the large and growing literature showing that CEOs exert strong influence on the firms they manage. While our baseline result could be driven by omitted firm characteristics, its robustness across different specifications and methodologies weighs against this possibility. The lack of a relation between CEO/CFO AM usage and most other firm decisions also weighs against our results being driven by omitted firm characteristics. Additionally, the lack of any association between non-CEO/CFO executive AM usage and corporate infractions is evidence against our results being driven by omitted firm culture. Finally, our analysis of financial advisors and SEC defendants provides a window for examining personal and professional conduct in other contexts without the same concerns over omitted firm characteristics. Here, too, we find that professional misconduct is associated with higher rates of AM usage.

Reverse causality and intentional selection of CEOs and CFOs based on their ethics could also potentially explain our results. Under the reverse causality channel, CEOs and CFOs become less ethical in their personal lives as a result of their participation in firm misconduct. Under the intentional selection channel, firms that allow corporate misconduct recruit CEOs and CFOs with more personal misconduct. Broader versions of both of these stories would apply to non-CEO/CFO executives as well, which is inconsistent with our empirical results on non-CEO/CFO AM executives. While we cannot rule out that some firms intentionally hire less ethical CEOs and CFOs (but not other executives) or that an unethical work environment causes CEOs and CFOs (but not other executives) to act unethically in their personal lives, even if these channels drive part of the results, this still implies a strong relation between personal and professional conduct.

15

Importantly, culture may also play an important role in firm conduct. Though its significance depends on what other regional characteristics we control for, the relation between county-level AM usage and corporate infractions points in this direction. Other dimensions of regional and firm culture could also affect firm conduct. The fact that CEO and CFO AM usage remains important with an unchanged coefficient after controlling for a wide range of regional characteristics and AM usage by other executives suggests that CEOs and CFOs have a unique relationship with firm conduct, whether directly or through firm culture.

At the individual level, our results demonstrate an intimate connection between personal and professional conduct but do not reveal the exact nature of that connection. A leading possibility is that character traits such as honesty and faithfulness to commitments affect behavior across contexts. It is also possible that risk aversion or perceptions of the likelihood and cost of being caught affect behavior across contexts. Regardless of what trait the underlying connection comes from, personal actions are informative about professional conduct. Understanding this connection and how it changes over time is an important area for further inquiry but is beyond the scope of what we can learn in this empirical setting.

16

Fig. S1. Ashley Madison website. This figure shows the ashleymadison.com home page as of June 23, 2015, collected from the Wayback Machine (https://web.archive.org/).

17

Fig. S2. Ashley Madison usage by CBSA. This figure shows the concentration of paid Ashley Madison usage by CBSA. Paid usage rates represent the number of paid users in a particular CBSA divided by total population from the 2010 census. Ashley Madison usage data comes from Ashley Madison’s transaction records.

18

Fig. S3. Probability of AM CEO/CFO transaction usage by industry. This figure plots the estimated probability of having a CEO/CFO who is an AM transaction user by industry along with a 95% confidence interval. Estimated probabilities come from regressing an indicator for AM CEO/CFO on industry fixed effects. Levels are obtained by adding the amount of AM CEO/CFO from the omitted category (Business Equipment) to each fixed effect. Consequently, the confidence interval for the industry “Business Equipment” is not reported.

19

Fig. S4. Cumulative market-model abnormal return for AM data release. This figure shows average cumulative abnormal returns (CAR) for the 41 firms in the sample with AM CEOs or CFOs when the Ashley Madison data was released (August 18, 2015). Abnormal returns are based on a market-model is estimated using returns from January 2, 2015 to 16 days prior to the event. Dashed lines represent 95% confidence intervals for the CAR.

20

(A) Ashley Madison usage by sustained vs. unsustained complaints

Fig. S5. Ashley Madison usage by police complaint characteristics. This figure plots overall Ashley Madison usage rates for misconduct and matched non-misconduct police officers. Misconduct police officers have at least one sustained complaint or at least five total complaints in 2010-2018. Panel A splits misconduct police officers by whether or not they have a sustained complaint. Panel B splits misconduct police officers by the severity of their most serious complaint. Moderate complaints are defined as complaint categories involving illegal searches, lockup procedures, verbal abuse, false arrest, conduct unbecoming, drug and alcohol abuse, or first amendment violations. Severe complaints are defined as complaint categories involving use of force, criminal misconduct, domestic abuse, bribery, official corruption, racial profiling, or excessive force. Panel C excludes off-duty complaints by dropping matched pairs in which the misconduct officers who would not have been classified as having misconduct without domestic abuse complaints and without any off-duty complaints, including domestic abuse and any other complaints classified as off-duty. *** indicates that the differences between misconduct and matched advisors are significant at the 1% level, ** indicates 5% significance, and * indicates 10% significance, with standard errors clustered by police district.

21

(B) Ashley Madison usage by complaint severity

(C) Ashley Madison usage after dropping off-duty complaints

Fig. S5 (continued) Ashley Madison usage by police complaint characteristics.

22

(A) Ashley Madison usage by financial damage

(B) Ashley Madison usage by number of prior reports

Fig. S6. Ashley Madison usage by advisor misconduct characteristics. This figure plots overall Ashley Madison usage rates for misconduct and matched non-misconduct financial advisors. Misconduct advisors have misconduct on their FINRA records in 2015 or 2016. Panel A splits misconduct financial advisors by magnitude of financial damages (when available). Panel B splits misconduct financial advisors by the amount of misconduct on their records prior to 2015. *** indicates that the differences between misconduct and matched advisors are significant at the 1% level, ** indicates 5% significance, and * indicates 10% significance, with difference standard errors clustered by firm and county.

23

Fig. S7. Ashley Madison usage of SEC defendants by infraction type. This figure plots overall Ashley Madison usage rates for SEC defendants and matched CEOs and CFOs by type of infraction alleged in the SEC complaints between 2010 and 2015. The types included are insider trading, Ponzi schemes, pump and dump schemes, and other fraud (e.g., securities or accounting fraud). *** indicates that the differences between misconduct and matched advisors are significant at the 1% level, ** indicates 5% significance, and * indicates 10% significance, with standard errors clustered by lawsuit.

24

Fig. S8. Probability of corporate infraction by year and CEO/CFO Ashley Madison usage. This figure shows the probability of a firm being affected by a class action lawsuit or by a financial statement restatement, by year, for firms with and without CEOs or CFOs who are Ashley Madison transaction users. A firm-year is affected by an infraction if its earnings are restated or if a class action lawsuit involves content that occurred in that year.

25

Fig. S9. Ashley Madison usage, political corruption, and religious adherence. Panel A plots the amount of overall paid AM usage and political corruption by CBSA. Panel B plots the amount of overall paid AM usage and religious adherence by CBSA. Panel C plots the amount of paid executive AM usage and political corruption by CBSA. Panel D plots the amount of paid executive AM usage and religious adherence by CBSA. The top-50 CBSAs are considered. Political corruption represents public corruption convictions per one million residents between 2004 and 2013 in the federal judicial district headquartered in the primary CBSA city. Religious adherence represents the percent of the CBSA’s population with a religious affiliation in 2010.

26

Fig. S9 (continued). Ashley Madison usage, political corruption, and religious adherence.

27

Table S1. Ashley Madison data summary by CBSA

Paid AM Usage Political Religious Overall Executives Corruption Adherence Top 20 CBSAs New York 0.46% 1.02% 34.2 55.5% Los Angeles 0.39% 1.13% 20.4 51.4% Chicago 0.39% 1.18% 41.4 57.2% Dallas-Fort Worth 0.46% 0.44% 28.2 55.3% Philadelphia 0.44% 1.38% 47.6 54.7% Houston 0.45% 0.57% 33.5 55.3% Washington, DC 0.55% 0.16% 59.6 44.5% Miami 0.41% 0.00% 43.2 38.1% Atlanta 0.45% 1.18% 29.6 49.7% Boston 0.52% 0.47% 30.9 56.8% San Francisco 0.43% 0.23% 5.9 37.8% Detroit 0.38% 0.53% 20.7 44.6% Riverside, CA 0.30% 0.00% 20.4 42.3% Phoenix 0.40% 0.68% 35.1 37.5% Seattle 0.47% 0.19% 11.3 35.6% Minneapolis 0.42% 0.00% 11.1 52.2% San Diego 0.44% 0.25% 26.0 43.9% St. Louis 0.32% 0.55% 41.0 49.2% Tampa 0.36% 0.37% 23.5 34.8% Baltimore 0.40% 1.83% 50.2 42.1%

Correlations - Top 50 CBSAs Overall AM Usage 1.00 Executive AM Usage 0.01 1.00 Political Corruption -0.24* 0.29** 1.00 Religiosity -0.09 -0.05 0.27* 1.00

This table reports paid Ashley Madison usage, political corruption, and religious adherence by CBSA. Paid AM usage is based on Ashley Madison’s transaction records. Executive users of Ashley Madison are identified by matching Ashley Madison transactions to Execucomp executives based on the executives’ names and addresses, which were obtained through LexisNexis searches. The sample is restricted to the United States. Political corruption represents public corruption convictions per one million residents in a federal judicial district between 2004 and 2013. CBSA political corruption is the weighted average (by population) of federal judicial districts within the CBSA. We drop conviction data from the District of Columbia judicial district because it not comparable to other districts. As a result, political corruption in the Washington, DC CBSA is based on conviction rates in Maryland and Virginia. Religious adherence represents the percent of the CBSA’s population with a religious affiliation in 2010. ***p<0.01, **p<0.05, *p<0.1.

28

Table S2. Additional financial advisor characteristics

Misconduct Matched Brokers Brokers Difference t-statistic

Inv. Adviser Exam (65/66) 66.6% 66.6% 0 Sec. Agent St. Law (63) 93.3% 92.4% 0.9% 1.03 Gen. Sec. Rep. (7) 88.0% 84.8% 3.2% 2.73 Inv. Co. Prod. Rep. (6) 28.7% 32.8% -4.2% -3.06 Gen Sec. Principal (24) 24.0% 24.2% -0.2% -0.15 Number of other qualifications 0.8 0.8 0.0 -0.48 Registered in > 3 states 49.4% 65.4% -16.1% -6.38 N 1,319 1,319

Financial advisor data comes from FINRA BrokerCheck. Reported qualifications summarize the fraction of financial advisors who have passed a particular exam, the mean number of other exams they have passed, and the fraction who are registered in more than three states. Financial advisors with FINRA-reported misconduct in 2015 or 2016 are matched to financial advisors who have never had misconduct. Matches as based on firm, county, year, experience (within five years), and whether or not the advisor has passed a Series 65 or 66 exam. Within these criteria, matches are chosen to minimize experience differences. The sample and control groups are limited to male financial advisors who have been registered since at least 2000. Reported t-statistics are based on standard errors that are double clustered by firm and county.

29

Table S3. Financial advisor matched-pair regressions

(1) (2) AM Transaction Overall AM Usage Misconduct 0.019*** 0.033***

(2.84) (4.07)

Years of experience -0.001 0.001 (-0.18) (0.17)

Age 0.000 -0.001 (-0.35) (-1.02)

Sec. Agent St. Law (63) 0.052*** 0.107*** (3.27) (3.71)

Gen. Sec. Rep. (7) -0.004 -0.001 (-0.32) (-0.02)

Inv. Co. Prod. Rep. (6) -0.003 0.013 (-0.25) (0.77)

Gen Sec. Principal (24) 0.003 0.009 (0.24) (0.42)

Number of other qualifications 0.004 0.002 (0.84) (0.23)

Registered in > 3 states 0.001 -0.001 (0.09) (-0.07)

N 1,319 1,319 Adjusted R2 0.002 0.008 Mean of dep. Variable 0.019 0.034

This table reports results from regressions within the matched financial advisor pairs described in Tables 1 and S2. Each observation is a matched pair of financial advisors. The dependent variables are differences between the Ashley Madison transaction and overall usage indicators for the pairs. The control variables are differences between financial advisor characteristics. The intercepts of the regressions, which are reported as the Misconduct coefficients, are the average differences between the pairs that are not explained by differences in financial advisor characteristics. Regressions are OLS, and t-statistics are based on standard errors that are double clustered by firm and county.

30

Table S4. Ashley Madison usage by financial advisors, cross-firm comparison High-Misconduct Low-Misconduct Firms Firms Difference t-statistic AM transaction usage 1.9% 2.3% -0.4% -0.53 Overall AM usage 4.0% 3.6% 0.4% 0.46 Years of experience 24.4 24.0 0.4 2.81 Age 53.3 51.1 2.2 5.63 Inv. adviser exam (65/66) 38.9% 38.9% 0 N 1,292 1,292

Ashley Madison usage data comes from Ashley Madison’s transaction and usage records. Financial advisor data comes from FINRA BrokerCheck. Financial advisors employed by high-misconduct firms as of May 2015 are matched to financial advisors employed by low-misconduct firms as of May 2015. The matches are based on county, experience (within five years), and whether or not the advisor has passed a Series 65 or 66 exam. Within these criteria, matches are chosen to sample from all firms and to minimize experience differences. The sample and control groups are again limited to male financial advisors who have been registered since at least 2000. Financial advisors are matched to Ashley Madison transactions based on name and address and zip code (for transaction usage) or email address (for non-transaction usage). Addresses and email addresses for financial advisors were obtained through LexisNexis searches. Reported t-statistics are based on standard errors that are clustered by county.

31

Table S5. Corporate data summary statistics

Panel A. CEO / CFO Summary Age Tenure Gender N % of Total (mean) (mean) (% Male) CEOs Total 2,654 100 55.88 7.12 96 AM Users 68 2.56 53.95 5.63 100 AM Transaction Users 47 1.77 53.65 5.94 100 AM Transaction Users (Prior) 36 1.36 54.31 6.07 100

CFOs Total 2,797 100 50.68 90 AM Users 65 2.32 49.49 97 AM Transaction Users 48 1.72 48.77 100 AM Transaction Users (Prior) 38 1.36 48.62 100

This table reports summary statistics for CEO/CFO characteristics, firm characteristics, and corporate infraction indicators. The sample period starts in 2008 (when Ashley Madison transaction data starts) and ends in 2014. The sample consists of firms in Execucomp that are headquartered in the U.S. In Panel A, CEOs and CFOs are matched to Ashley Madison based on name and address (for transaction usage) or email address and zip code (for non-transaction usage). Addresses and email addresses for CEOs and CFOs were obtained through LexisNexis searches. In Panel B, firm-year observations are split based on whether the firm’s CEO or CFO has a paid Ashley Madison transaction in the current year or in a prior year. The infraction dummy variables take the value of one in all firm-years affected by an infraction. Reported t-statistics are based on standard errors that are clustered at the firm level.

32

Table S5 (continued). Corporate data summary statistics

Panel B. Firm summary (firm-year observations) Ashley Non-Ashley Madison CEO/CFO Madison CEO/CFO Mean Mean Difference t-statistic Firm characteristics Firm size 7.85 7.80 0.05 0.17 Return on assets 0.11 0.12 -0.010 -0.20 Tobin’s Q 1.57 1.81 -0.24 -0.72 Investment 0.03 0.05 -0.02 -1.41 Acquisitions 0.04 0.03 0.004 0.43 R&D 0.05 0.05 0.00 -0.19 Book leverage 0.43 0.46 -0.03 -0.09 Market leverage 0.28 0.28 0.003 0.09 Dividend payouts 0.01 0.02 -0.005 -0.92

Geographic characteristics AM paid usage (county) 0.46 0.44 0.02 0.66 Religious adherence (county) 53.74 52.62 1.11 0.57 Political corruption (jud. district) 32.73 30.03 2.70 0.89

Infraction variables Class action 0.089 0.030 0.059 3.13 Restatement 0.066 0.026 0.040 2.10 Class action or restatement 0.117 0.053 0.064 2.53 N 213 9,768

33

Table S6. AM CEOs/CFOs and corporate infraction likelihood (OLS estimation)

(1) (2) (3) (4) AM CEO/CFO 0.094** 0.094** 0.089* 0.086*

(2.02) (2.01) (1.94) (1.87) AM paid usage 0.054 0.043 0.103**

(1.57) (1.17) (2.27) CEO age -0.001 -0.001 -0.001 -0.001

(-0.93) (-0.92) (-1.02) (-1.23) CEO tenure 0.000 -0.000 0.000 0.000

(0.01) (-0.11) (0.09) (0.30) CEO male -0.022 -0.021 -0.019 -0.023

(-0.89) (-0.86) (-0.74) (-0.86) CFO age 0.001 0.001 0.001 0.001

(1.57) (1.53) (1.16) (1.39) CFO male 0.019* 0.019* 0.018* 0.024**

(1.87) (1.87) (1.67) (2.26) Firm size -0.004 -0.004* -0.003 -0.003

(-1.63) (-1.82) (-1.10) (-0.99) Return on assets 0.018 0.019 0.014 0.017

(0.60) (0.63) (0.51) (0.59) Tobin's Q 0.000 0.000 -0.000 0.000

(0.20) (0.14) (-0.05) (0.01) Market leverage 0.082*** 0.084*** 0.099*** 0.094***

(3.01) (3.09) (3.53) (3.31) Year FE y y y y 2-digit SIC FE n n y y State FE n n n y N 7,899 7,862 7,862 7,862 R2 0.011 0.012 0.027 0.048 Mean of dep. Variable 0.056 0.056 0.056 0.056

This table reports OLS regression results equivalent to those in Table 3. The dependent variable is a dummy variable that takes the value of one in all firm-years with restated financial statements or that were affected by conduct alleged in a class action lawsuit. The explanatory variable of interest is AM CEO/CFO, a dummy variable that takes the value of one for the firm-years where a firm has either a CEO or CFO that is a paid user of the AM website. The variable AM paid usage represents the per capita paid AM usage rate at the county of the firm’s headquarter. Executive and firm controls, county AM usage, as well as year, industry, and state fixed effects are included as indicated. Reported t-statistics in parentheses are heteroscedasticity-robust and clustered by firm. ***p<0.01, **p<0.05, *p<0.1.

34

Table S7. Additional control variable for CEO/CFO personal legal infractions

(1) (2) (3) (4) AM CEO/CFO 0.056*** 0.055*** 0.052*** 0.049**

(2.87) (2.83) (2.63) (2.44) Infraction CEO/CFO 0.000 0.001 0.001 0.004

(0.01) (0.17) (0.07) (0.43) Exec. & firm controls y y y y AM paid usage control n y y y Year FE y y y y 2-digit SIC FE n n y y State FE n n n y N 7,899 7,862 7,521 7,342 Pseudo R2 0.023 0.026 0.052 0.085 Mean of dep. Variable 0.056 0.056 0.059 0.060

This table reports marginal effects of logistic regressions. Specifications are the same as in Table 3 except that the regressions control for personal legal infractions of CEOs and CFOs. Legal infractions are based on LexisNexis criminal records. Infraction CEO/CFO is a dummy variable that takes the value of one for the firm-years where a firm has either a CEO or CFO with a legal infraction in LexisNexis criminal records. Executive and firm controls, county AM usage, as well as year, industry, and state fixed effects are included as indicated. Reported z-statistics in parentheses are heteroscedasticity-robust and clustered by firm. ***p<0.01, **p<0.05, *p<0.1.

35

Table S8. AM CEOs/CFOs and likelihood of class action lawsuit or financial restatement

Panel A. Class action lawsuit (1) (2) (3) (4) AM CEO/CFO 0.039*** 0.038*** 0.048*** 0.048***


Panel B. Financial restatement

(1) (2) (3) (4) AM CEO/CFO 0.033** 0.033** 0.032* 0.037**


This table reports marginal effects of logistic regressions. In Panel A, the dependent variable is a dummy variable that takes the value of one in all firm-years affected by a class action lawsuit. In Panel B, the dependent variable is a dummy variable that takes the value of one in all firm-years where the financial statements were restated. The explanatory variable of interest is AM CEO/CFO, a dummy variable that takes the value of one for the firm-years where a firm has either a CEO or CFO that is a paid user of the AM website. Executive and firm controls, county AM usage, as well as year, industry, and state fixed effects are included as indicated. Reported z-statistics in parentheses are heteroscedasticity-robust and clustered by firm. ***p<0.01, **p<0.05, *p<0.1.

36

Table S9. AM CEOs and CFOs and corporate infraction likelihood

Panel A. Ashley Madison CEO (1) (2) (3) (4) AM CEO 0.055** 0.055** 0.042 0.038


Panel B. Ashley Madison CFO (1) (2) (3) (4) AM CFO 0.054* 0.054* 0.063** 0.067**


This table reports marginal effects of logistic regressions. The dependent variable is a dummy variable that takes the value of one in all firm-years with restated financial statements or that were affected by conduct alleged in a class action lawsuit. The explanatory variables of interest are AM CEO or AM CFO, two dummy variables that take the value of one for the firm-years where a firm has a CEO or a CFO that is a paid user of the AM website, respectively. Executive and firm controls, county AM usage, as well as year, industry, and state fixed effects are included as indicated. Reported z-statistics in parentheses are heteroscedasticity-robust and clustered by firm. ***p<0.01, **p<0.05, *p<0.1.

37

Table S10. Separate effect of AM CEOs and CFOs on corporate infraction types

Panel A. AM CEO and class action lawsuit (1) (2) (3) (4) AM CEO 0.038** 0.037** 0.042** 0.039**


Panel B. AM CFO and class action lawsuit (1) (2) (3) (4) AM CFO 0.039** 0.039** 0.055** 0.059**


This table reports marginal effects of logistic regressions. In Panels A and B, the dependent variable is a dummy variable that takes the value of one in all firm-years affected by a class action lawsuit. In Panels C and D, the dependent variable is a dummy variable that takes the value of one in all firm-years where the financial statements were restated. The explanatory variables of interest are AM CEO and AM CFO, dummy variables that take the value of one for the firm-years where a firm has a CEO or a CFO that is a paid user of the AM website, respectively. Executive and firm controls, county AM usage, as well as year, industry, and state fixed effects are included as indicated. Reported z-statistics in parentheses are heteroscedasticity-robust and clustered by firm. ***p<0.01, **p<0.05, *p<0.1.

38

Table S10 (continued). Separate effect of AM CEOs and CFOs on corporate infraction types

Panel C. AM CEO and financial restatement (1) (2) (3) (4) AM CEO 0.016 0.016 0.007 0.013


Panel D. AM CFO and financial restatement (1) (2) (3) (4) AM CFO 0.042** 0.042** 0.051** 0.058**


39

Table S11. Logistic propensity score regression

(1) CEO age 0.00004

(0.13) CEO tenure -0.00046

(-1.06) CEO male 0.01925

(0.91) CFO age -0.00012

(-0.25) CFO male 0.01494

(1.09) Firm size -0.00052

(-0.31) Return on assets 0.00485

(0.22) Tobin’s Q -0.00253

(-1.26) Investment -0.13760**

(-2.26) Acquisitions 0.00601

(0.74) Div. Payout -0.15840

(-0.99) N 8,745 Pseudo R2 0.022

This table reports marginal effects of the logistic regression used to compute the propensity score. The dependent variable is AM CEO/CFO, a dummy variable that takes the value of one for the firm-years where a firm has either a CEO or CFO that is a paid user of the AM website. The set of explanatory variables include executive age, executive gender, CEO tenure, and firm variables. Reported t-statistics in parentheses are heteroscedasticity-robust and clustered by firm. ***p<0.01, **p<0.05, *p<0.1.

40

Table S12. Propensity score matching treatment effect and sample comparison

Ashley Non-Ashley

Madison

CEO/CFO Madison

CEO/CFO Difference t-statistic Combined infraction 0.135 0.043 0.092 2.16

Executive characteristics CEO age 55.708 55.903 -0.195 -0.23 CEO tenure 7.423 7.610 -0.187 -0.22 CEO male 0.984 0.995 -0.011 -0.63 CFO age 50.935 51.232 -0.297 -0.28 CFO male 0.951 0.930 0.022 0.79

Firm characteristics Firm size 7.760 7.422 0.338 1.31 Return on assets 0.105 0.109 -0.005 -0.48 Tobin’s Q 1.598 1.719 -0.121 -1.36 Investment 0.034 0.032 0.002 0.66 Acquisitions 0.039 0.045 -0.006 -0.61 R&D 0.053 0.058 -0.005 -0.32 Market leverage 0.272 0.261 0.020 0.88 Dividend payouts 0.009 0.008 0.0005 0.20

This table reports the average executive and firm characteristics of matched firm-year observations of firms with AM CEOs or CFOs and firms without CEOs or CFOs who are paid users of the AM website. The matching is performed based on year, Fama-French 12 industry classification, and propensity scores estimated based on a logistic regression with executive age, executive gender, CEO tenure, and firm controls (Table S10). The matching uses the nearest neighbor technique (one-to-one). A maximum difference between the propensity scores of 0.5% is required. Reported t-statistics are heteroscedasticity-robust and clustered by AM firm.

41

Table S13. Other AM executives and additional cultural variables

(1) (2) (3) (4) AM CEO/CFO 0.049** 0.052*** 0.049** 0.053***

(2.41) (2.59) (2.45) (2.66) AM other executive 0.007

(0.31) AM paid usage (county) 0.114** 0.119** 0.068 0.055

(2.49) (2.51) (0.90) (0.70) Religious adherence (county) 0.001* 0.001*

(1.68) (1.89) Political corruption (jud. district) -0.000 -0.000

(-0.47) (-0.15) Population (CBSA) 0.003 -0.002

(0.40) (-0.25) College educated (education) 0.001 0.001

(0.73) (0.93) Income (county) -0.000 -0.000

(-0.14) (-0.07) Exec. & firm controls y y y y Year FE y y y y 2-digit SIC FE y y y y State FE y y y y N 7,342 6,982 7,342 6,982 Pseudo R2 0.086 0.092 0.086 0.093 Mean of dep. Variable 0.060 0.060 0.060 0.060

This table reports marginal effects of logistic regressions in which the dependent variable is a dummy variable that takes the value of one in all firm-years with restated financial statements or that were affected by conduct alleged in a class action lawsuit. AM CEO/CFO is a dummy variable that takes the value of one for the firm-years where a firm has either a CEO or CFO that is a paid user of the AM website. AM other executive is a dummy variable that takes the value of one for the firm-years where a firm has a top executive (other than the CEO or CFO) that is a paid user of the AM website. AM paid usage is the per capita paid AM usage rate in the county of the firm’s headquarter. Political corruption is the number of public corruption convictions in the federal judicial district of the firm’s headquarters (excluding the District of Columbia) per 1 million residents between 2004 and 2013 (as reported by the U.S. Department of Justice). Religious adherence is the percent of the county’s population with a religious affiliation in 2010 (as reported by the Association of Religious Data Archives). Population is the log of CBSA population in 2010 (as reported by the U.S. Census). College educated is the percentage of the county’s residents who are college educated residents in the county of the firm’s headquarter in 2010 (as reported by the U.S. Department of Agriculture). Income is the median household income in the county of the firm’s headquarters in 2010 (as reported by the U.S. Census). Reported z-statistics in parentheses are heteroscedasticity-robust and clustered by firm. ***p<0.01, **p<0.05, *p<0.1.

42

Table S14. Additional control variables

(1) (2) (3) (4) AM CEO/CFO 0.045** 0.050** 0.051** 0.049**

(2.07) (2.34) (2.53) (2.44) AM paid usage (county) 0.098** 0.113** 0.113** 0.115**

(2.22) (2.29) (2.51) (2.57) Optimistic CEO/CFO -0.002

(-0.22) Stock return -0.019*

(-1.91) Past year's stock return 0.005

(1.09) Return volatility 0.076**

(2.42) Past year's return volatility -0.075**

(-2.19) E-index -0.005

(-1.17) F-score 0.010*

(1.65) Exec. & firm controls y y y y Year FE y y y y 2-digit SIC FE y y y y State FE y y y y N 7,188 6,339 7,342 7,342 Pseudo R2 0.086 0.091 0.086 0.089 Mean of dep. Variable 0.058 0.062 0.060 0.059

This table reports marginal effects of logistic regressions in which the dependent variable is a dummy variable for firm-years with a corporate infraction. Optimistic CEO/CFO is a dummy variable that takes the value of one if a firm has an optimistic CEO or CFO (30). E-index is a corporate governance index (31), and F-score is the financial misstatement predictor based on accounting data (32). Where missing, we replace E-index and F-score with industry averages so as not to reduce the sample size. Reported z-statistics in parentheses are heteroscedasticity-robust and clustered by firm. ***p<0.01, **p<0.05, *p<0.1.

43

Table S15. Additional governance-related control variables

(1) (2) (3) (4) AM CEO/CFO 0.050** 0.054*** 0.052*** 0.043**

(2.47) (2.76) (2.60) (2.13) Stock option CEO 0.106

(1.37) Stock option CFO -0.580

(-1.09) CEO ownership 0.000

(0.49) CFO ownership -0.008

(-0.98) Product market competition -0.077

(-1.24) Ln(board size) -0.020

(-0.87) Non-independent directors 0.057

(1.15) Ln(audit committee size) -0.032*

(-1.95) Non-independent directors (audit) -0.135*

(-1.72) Institutional ownership -0.017

(-0.94) Exec. & firm controls y y y y AM paid usage control y y y y Year FE y y y y 2-digit SIC FE y n y y State FE y y y y N 7,335 7,676 7,342 6,329 Pseudo R2 0.0866 0.0563 0.0904 0.0572 Mean of dep. Variable 0.060 0.057 0.060 0.055

Marginal effects of logistic regressions. Dependent variable is a dummy variable for firm-years with a corporate infraction. Stock option is the estimated value of in-the-money exercisable and unexercisable stock options owned. Ownership is the fraction of total shares owned. Product market competition is industry HHI based on sales. Board size and audit committee size are number of members. Non-independent directors is the fraction of members classified as non-independent by ISS. Institutional ownership is the fraction of shares owned by institutional investors. Where missing, we replace director data and CEO/CFO ownership with industry averages. Reported z-statistics in parentheses are clustered by firm. ***p<0.01, **p<0.05, *p<0.1.

44

Table S16. Interactions with governance-related control variables

(1) (2) (3) AM CEO/CFO 0.059*** 0.074*** 0.038*

(2.65) (3.69) (1.72) High e-index -0.012

(-1.54) AM CEO/CFO*High e-index -0.025

(-0.66) High ln(audit committee size) -0.011

(-1.21) AM CEO/CFO*High ln(audit committee size) -0.114**

(-2.08) High non-independent directors (audit) -0.022*

(-1.91) AM CEO/CFO*High non-independent directors (audit) 0.046

(1.20) Exec. & firm controls y y y AM paid usage control y y y Year FE y y y 2-digit SIC FE y y y State FE y y y N 7,342 7,342 7,342 Pseudo R2 0.087 0.090 0.088 Mean of dep. Variable 0.060 0.060 0.060

This table reports marginal effects of logistic regressions similar to those of Table 3, but including additional firm governance-related controls and interaction terms. The dependent variable is Corporate Infraction, a dummy variable that takes the value of one in all firm-years with restated financial statements or that were affected by conduct alleged in a class action lawsuit. High e-index is a dummy variable that the value of one for all firm-years where E-index is above median value. High ln(audit committee size) is a dummy variable that the value of one for all firm-years where Ln(audit committee size) is above median value. High non-independent directors (audit) is a dummy variable that the value of one for all firm-years where Non-independent directors (audit) is above median value. The table also includes marginal effects associated to AM CEO/CFO interacted with the previously defined variables. E-index, Ln(audit committee size), and Non-independent directors (audit) are defined in Tables S14 and S15. Reported z-statistics in parentheses are heteroscedasticity- robust and clustered by firm. ***p<0.01, **p<0.05, *p<0.1.

45

Table S17. Alternative definitions of AM CEO/CFO and extended samples

Main Main Main Extended Extended Sample Sample Sample 2015 GMI

(1) (2) (3) (4) (5) AM CEO/CFO 0.050** 0.041** 0.034*

(2.47) (2.16) (1.78) AM CEO/CFO (All) 0.041** (2.34) AM CEO/CFO (Backfilled) 0.040** (2.19) Exec. & firm controls y y y y y AM paid usage control y y y y y Year FE y y y y y 2-digit SIC FE y y y y y State FE y y y y y N 7,342 7,342 7,342 8,208 12,268 R2 0.085 0.085 0.085 0.079 0.055 Mean of dep. Variable 0.060 0.060 0.060 0.061 0.064

This table reports marginal effects of logistic regressions. The dependent variable is a dummy variable that takes the value of one in all firm-years with restated financial statements or that were affected by conduct alleged in a class action lawsuit. The explanatory variables of interest are different definitions of AM CEO/CFO usage. AM CEO/CFO, is a dummy variable that takes the value of one for the firm-years where a firm has either a CEO or CFO that is a paid user of the AM website. AM CEO/CFO (all users), is a dummy variable that takes the value of one for the firm-years where a firm has either a CEO or CFO that has an account (i.e., not necessarily paid) at the AM website. AM CEO/CFO (backfilled), considers CEOs and CFOs that are paid users of the AM website, but relaxes the requirement that AM transaction usage occur before or during the year being considered. Columns 1 to 3 use the main sample. The sample in Column 4 also includes data from 2015, and the sample in Column 5 includes firms from MSCI’s GMI data. Executive and firm controls, county AM usage, as well as year, industry, and state fixed effects are included as indicated. Reported z-statistics in parentheses are heteroscedasticity-robust and clustered by firm. ***p<0.01, **p<0.05, *p<0.1.

46

Table S18. AM CEOs/CFOs and other firm decisions

Investment Acquisitions Div. Payout R&D CEO/CFO Comp. (1) (2) (3) (4) (5)

AM CEO/CFO -0.015** -0.003 -0.004 0.003 0.029 (-2.43) (-0.29) (-1.48) (0.39) (0.55) Exec. & firm controls y y y y y AM paid usage control y y y y y Year FE y y y y y 2-digit SIC FE y y y y y State FE y y y y y N 7,859 7,478 7,853 4,756 7,852 R2 0.371 0.045 0.179 0.446 0.658 Mean of dep. Variable 0.058 0.037 0.016 0.054 8.531

This table reports the coefficients of OLS regressions. The dependent variables are Investment (Column 1), Acquisitions (Column 2), Dividend Payout (Column 3), R&D (Column 4), and CEO/CFO Compensation (Column 5). Investment is CAPEX divided by lagged total assets. Acquisitions is the total value of acquisitions during the year divided by lagged total assets. Dividend Payout is the sum of preferred and common dividends paid divided by lagged total assets. R&D is research and development expenses divided by lagged total assets and CEO/CFO Compensation is the natural log of the sum of total CEO compensation and total CFO compensation. Total compensation consists of salary, bonus, value of restricted stock granted, value of options granted, long-term incentive payouts, and other compensation. The explanatory variable of interest is AM CEO/CFO, a dummy variable that takes the value of one for the firm-years where a firm has either a CEO or CFO that is a paid user of the AM website. Executive and firm controls, county AM usage, as well as year, industry, and state fixed effects are indicated. Reported t-statistics in parentheses are heteroscedasticity-robust and clustered by firm. ***p<0.01, **p<0.05, *p<0.1.

47

Table S19. AM CEOs/CFOs and stock returns

(1) (2) (3) (4) AM CEO/CFO 0.048 0.048 0.049 0.052

(1.00) (1.01) (0.99) (1.03) Exec. & firm controls y y y y AM paid usage control n y y y Year FE y y y y 2-digit SIC FE n n y y State FE n n n y N 7,477 7,440 7,440 7,440 R2 0.209 0.209 0.228 0.231

This table reports the coefficients of OLS regressions. The dependent variable is the firm’s annual stock return. The explanatory variable of interest is AM CEO/CFO, a dummy variable that takes the value of one for the firm-years where a firm has either a CEO or CFO that is a paid user of the AM website. Executive and firm controls, county AM usage, as well as year, industry, and state fixed effects are included as indicated. Reported t-statistics in parentheses are heteroscedasticity-robust and clustered by firm. ***p<0.01, **p<0.05, *p<0.1.

48

References:

1. Egan M, Matvos G, Seru A (2017) The market for financial adviser misconduct. Journal of

Political Economy, forthcoming.

2. Benmelech E, Frydman C (2015) Military CEOs. Journal of Financial Economics, 117:42-59.

3. Dyck A, Morse A, Zingales L (2010) Who blows the whistle on corporate fraud? Journal of

Finance, 65:2213-2253.

4. McGuire ST, Omer TC, Sharp NY (2011) The impact of religion on financial reporting

irregularities. Accounting Review, 87:645-673.

5. Karpoff JM, Koester A, Lee DS, Martin GS (2017) Proxies and databases in financial

misconduct research. Accounting Review, 92:129-163.

6. Cline BN, Walkling RA, Yore AS (2018) The consequences of managerial indiscretions: Sex,

lies, and firm value. Journal of Financial Economics, 127:389-415.

7. Davidson R, Dey A, Smith A (2015) Executives’ “off-the-job” behavior, corporate culture,

and financial reporting risk. Journal of Financial Economics, 117:5-28.

8. Mironov M (2015) Should one hire a corrupt CEO in a corrupt country? Journal of Financial

Economics, 117:29-42.

9. Jia Y, van Lent L, Zeng Y (2014) Masculinity, testosterone, and financial misreporting.

Journal of Accounting Research, 52:1195-1246.

10. Grieser W, Li R, Simonov A (2018) Integrity, creativity, and corporate culture. Working paper,

Texas Christian University and Michigan State University.

11. Guiso L, Sapienza P, Zingales L (2015) The value of corporate culture. Journal of Financial

Economics, 117:60-76.

49

12. Biggerstaff L, Cicero DC, Puckett A (2015) Suspect CEOs, unethical culture, and corporate

misbehavior. Journal of Financial Economics, 117:98-121.

13. Parsons CA, Sulaeman J, Titman S (2018) The geography of financial misconduct. Journal of

Finance, 73:2087-2137.

14. Grullon G, Kanatas G, Weston J (2009) Religion and corporate (mis) behavior. Working paper,

Rice University.

15. Hilary G, Hui KW (2009) Does religion matter in corporate decision making in America?

Journal of Financial Economics, 93:455-473.

16. DeBacker J, Heim BT, Tran A (2015) Importing corruption culture from overseas: Evidence

from corporate tax evasion in the United States. Journal of Financial Economics, 117:122-

138.

17. Glaeser EL, Saks RE (2006) Corruption in America. Journal of Public Economics, 90:1053-

1072.

18. Bertrand M, Schoar A (2003) Managing with style: The effect of managers on firm policies.

Quarterly Journal of Economics, 118:1169-1208.

19. Yermack D (2006) Flights of fancy: Corporate jets, CEO perquisites, and inferior shareholder

returns. Journal of Financial Economics, 80:211-242.

20. Kaplan SN, Klebanov NM, Sorensen M (2012) Which CEO characteristics and abilities

matter? Journal of Finance, 67:973-1007.

21. Gow ID, Kaplan SN, Larcker DF, Zakolyukina AA (2016) CEO personality and firm policies.

Working paper, University of Melbourne, University of Chicago, and Stanford University.

22. Malmendier U, Tate G (2005) CEO overconfidence and corporate investment. Journal of

Finance, 60:2661-2700.

50

23. Malmendier U, Tate G, Yan J (2011) Overconfidence and early‐life experiences: The effect of

managerial traits on corporate financial policies. Journal of Finance, 66:1687-1733.

24. Cain MD, McKeon SB (2016) CEO personal risk-taking and corporate policies. Journal of

Financial and Quantitative Analysis, 51:139-164.

25. Cronqvist H, Makhija AK, Yonker SE (2012) Behavioral consistency in corporate finance:

CEO personal and corporate leverage. Journal of Financial Economics, 103:20-40.

26. Graham JR, Harvey CR, Puri M (2013) Managerial attitudes and corporate actions. Journal of

Financial Economics, 109:103-121.

27. Kallunki JP, Pyykkö E (2013) Do defaulting CEOs and directors increase the likelihood of

financial distress of the firm? Review of Accounting Studies, 18:228-260.

28. Law KK, Mills LF (2017) Military experience and corporate tax avoidance. Review of

Accounting Studies, 22:141-184.

29. Chyz JA (2013) Personally tax aggressive executives and corporate tax sheltering. Journal of

Accounting and Economics, 56:311-328.

30. Campbell TC, Gallmeyer M, Johnson SA, Rutherford J, Stanley BW (2011) CEO optimism

and forced turnover. Journal of Financial Economics, 101:695-712.

31. Bebchuk L, Cohen A, Ferrell A (2008) What matters in corporate governance? Review of

Financial Studies, 22:783-827.

32. Dechow PM, Ge W, Larson CR, Sloan RG (2011) Predicting material accounting

misstatements. Contemporary Accounting Research, 28:17-82.

33. Khanna V, Kim EH, Lu Y (2015) CEO connectedness and corporate fraud. Journal of Finance,

70:1203-1252.

51

34. Giroud X, Mueller HM (2011) Corporate governance, product market competition, and equity

prices. Journal of Finance, 66:563-600.

35. Chung KH, Zhang H (2011). Corporate governance and institutional ownership. Journal of

Financial and Quantitative Analysis, 46:247-273.

supplementary information for · discreet encounters. between married individuals. married dating...

Documents