The Geography of Financial Misconduct

• Published date: 01 October 2018
• Author: CHRISTOPHER A. PARSONS,SHERIDAN TITMAN,JOHAN SULAEMAN
• DOI: http://doi.org/10.1111/jofi.12704
• Date: 01 October 2018

THE JOURNAL OF FINANCE •VOL. LXXIII, NO. 5 •OCTOBER 2018

The Geography of Financial Misconduct

CHRISTOPHER A. PARSONS, JOHAN SULAEMAN, and SHERIDAN TITMAN∗

ABSTRACT

Financial misconduct (FM) rates differ widely between major U.S. cities, up to a fac-

tor of 3. Although spatial differences in enforcement and firm characteristics do not

account for these patterns, city-level norms appear to be very important. For exam-

ple, FM rates are strongly related to other unethical behavior, involving politicians,

doctors, and (potentially unfaithful) spouses, in the city.

Traditional models of crime begin with Becker (1968), who frames the choice to

engage in misbehavior like any other economic decision involving cost-benefit

tradeoffs. Though somewhat successful when taken to the data, the theory

fails to account for the enormous variation in crime rates observed across

both time and space. Indeed, as Glaeser, Sacerdote, and Scheinkman (1996)

argue, regional variation in demographics, enforcement, and other observables

is simply not large enough to explain why,for example, two seemingly identical

neighborhoods in the same city have drastically different crime rates.1Their

conclusion is that social factors—culture—play an important role in a person’s

decision to break rules, or to respect them.

In this paper, we attempt to quantify the importance of social norms in white

collar crimes, exploiting geography as the main source of variation. Using

∗Christopher A. Parsons is with the University of Southern California. Johan Sulaeman is with

National University of Singapore. Sheridan Titman is with the University of Texas at Austin.

The paper uses a number of hand-collected or proprietary data sources, and would not have been

possible without the cooperation of other researchers. We are particularly grateful to Jonathan

Karpoff, Allison Koester, Scott Lee, and Gerald Martin for making their data on financial miscon-

duct available. We also acknowledge Joey Engelberg for providing the prescription/payment data

set, Stephen Dimmock, Will Gerken, and Nathaniel Graham for data on fraud by financial advi-

sors, and Gennaro Bernille and Gregg Jarrell for data on options backdating. We are also thankful

to two anonymous referees, an Associate Editor, and the Editor (Michael Roberts), whose insights

have markedly improved the paper. Feedback from Serdar Dinc (discussant), Ray Fisman (discus-

sant), Francisco Gallego (discussant), Rajkamal Iyer (discussant), and seminar participants at the

American Finance Association Meetings (2015), University of Michigan, University of Oregon,

Southern Methodist University, University of Washington, 2014 SFS Cavalcade, NBER Summer

Institute Corporate Finance (2014), UT Dallas Spring Finance Conference (2017), and 7th In-

ternational Conference at Finance UC (2016) was also very helpful in revising the paper. Johan

Sulaeman acknowledges research support from NUS Start-Up Research Grant WBS R-315-000-

113-133. We have read the Journal of Finance’s disclosure policy and have no conflicts of interest

to disclose. All errors are ours.

1For a theoretical justification of this idea, see Sah (1991). Other empirical studies of peer

effects in crime include Kling, Ludwig, and Katz (2005) and Ludwig and Kling (2007).

DOI: 10.1111/jofi.12704

2087

2088 The Journal of Finance R

Karpoff, Koester, Lee, and Martin’s (KKLM, 2017) hand-collected data set of

financial misconduct (FM) by public firms over the four decades beginning in

1970, we find that

1. FM tends to disproportionately cluster in certain cities, with rates differ-

ing by up to a factor of 3.

2. Geographic variation in social norms—informal understandings that gov-

ern a wide range of (mis)behaviors—accounts for a large part of these

patterns, whereas regional differences in enforcement or firm character-

istics do not.2

3. A city’s social norms, as measured by other (non-FM) types of misbehav-

ior, such as spousal infidelity or political corruption, strongly explain the

geographic cross-section of FM.

After providing a stylized model (Section I) and description of the data

(Section II), we begin our empirical analysis with a measurement exercise

in Section III. Here, we measure regional differences in FM, provide a feel for

the magnitudes, and ask whether FM events cluster geographically more than

would be predicted by chance. In other words, although simple averages indi-

cate that FM is three times more common in some places (e.g., Miami, Dallas)

than others (e.g., Seattle, Minneapolis), are these differences statistically

significant?

We follow the empirical approach of Ellison and Swanson (2016), who use

negative binomial regressions to examine whether the incidence of exception-

ally high scores on a difficult high school mathematics exam cluster at the

school level. The variable of interest in our study, FM, is also a rare event, and

like the number of high test scores, the number of firms committing financial

misconduct takes discrete values.

In the negative binomial regression we estimate, the number of FM events

occurring in each city over our four-decade sample period is regressed on the

population at risk (total number of firm-years). We estimate a highly significant

(p<0.001) excess variance parameter, which indicates that FM clusters more

in certain cities than we would expect by chance. The same picture emerges if

we expand the unit of analysis to the city-year level, resulting in 20 cities ×

40 years =800 city-year observations, and estimate city fixed effects. In Poisson

regressions, city fixed effects are highly significant ( p<0.001) determinants of

FM incidence.3

With these basic patterns established, in the remainder of the paper we

attempt to understand the factor(s) underlying this geographic heterogeneity

2The Oxford Dictionary of Sociology defines social norms as “common standards within a social

group regarding socially acceptable or appropriate behaviour in particular social situations, the

breach of which has social consequences. The strength of these norms varies from loose expectations

to unwritten rules.”

3Alternatively,ordinary least square (OLS) regressions with city-level FM rates ( #ofFM events

#offirms )as

the dependent variable (rather than counts of FM incidences) also give highly significant estimates

for city fixed effects.

The Geography of Financial Misconduct 2089

in FM. We consider three possibilities. First, the risk of detection or enforce-

ment may differ across cities, in which case differences in observed misconduct

may arise even if actual FM rates are identical. Second, the potential benefit

of “cooking the books” may vary across cities, for example, with regional dif-

ferences in financial distress or incentive-based compensation. Finally, social

attitudes toward right and wrong may differ across cities: a given behavior

may be frowned upon in Minneapolis, questionable in Phoenix, but acceptable

in Houston.

While we cannot definitively rule out the first two explanations, city-level

social norms appear to be a key factor determining the prevalence of FM. In

Section IV, we assemble city-level data on a wide variety of activities intended

to capture a city’s prevailing ethical norms: (1) political corruption, (2) fraud

by financial advisors, (3) backdating of executive stock options, (4) spousal in-

fidelity, and (5) financial relationships between doctors and drug companies.

Importantly, these behaviors span the range from being explicitly illegal (#1

and most cases of #2) to violating social contracts and/or generalized expec-

tations for proper behavior (cases #3 to #5); in the latter three cases, formal

enforcement plays little to no role.

In cross-sectional analyses we find that the prevalence of these questionable

behaviors is strongly related to rates of FM committed by local executives.

Despite the low statistical power of regressions estimated using only 20 ob-

servations (one for each city), the individually estimated coefficients on the

incidence of each activity are all positive, with four significant at the 5% level,

and the fifth (financial advisor fraud) exceeding the 10% threshold. When we

combine these activities into a single city-level index based on their average

ranks, the t-statistic is over four. The strength of this relation (ρ=0.60) is

readily apparent in Figure 2, which plots each city’s average FM rate against

its ranking of other misbehaviors.

Remarkably, this index provides a near-complete account of the (significant)

cross-city variation in FM. When we omit the misbehavior index as a covariate,

the random effect parameter in a negative binomial regression of FM has a

p-value of less than 0.001, indicating that FM clusters too frequently in some

cities to be explained by chance. However, when we include the misbehavior

index, the p-value increases to 0.5, implying that any residual heterogeneity

across cities can be attributed to sampling variation.

While we interpret these results as strong evidence that city-level culture

affects a wide range of misbehaviors, including financial misconduct, our anal-

ysis does not identify the source(s) of these norms. Following Manski’s (1993)

canonical taxonomy, city-level norms could arise from exogenous (e.g., genetic)

differences between people in certain areas,4correlated/environmental influ-

ences (e.g., church or educational institutions), or endogenous peer effects. Any

of these, or their combination, can contribute to city-wide misbehavior.

4A twin study by Loewen et al. (2013), for example, concludes that a substantial fraction of

variability in the acceptability of dishonesty can be traced to genetics.