NBER WORKING PAPER SERIES

BANK STRESS TESTING, HUMAN CAPITAL INVESTMENT AND RISK MANAGEMENT

Thomas Schneider
Philip Strahan
Jun Yang

Working Paper 30867

http://www.nber.org/papers/w30867

NATIONAL BUREAU OF ECONOMIC RESEARCH

1050 Massachusetts Avenue
Cambridge, MA 02138
January 2023

Authors received no research funding from sources other than their home institutions. The views
expressed herein are those of the authors and do not necessarily reflect the views of the National
Bureau of Economic Research.

NBER working papers are circulated for discussion and comment purposes. They have not been
peer-reviewed or been subject to the review by the NBER Board of Directors that accompanies
official NBER publications.

© 2023 by Thomas Schneider, Philip Strahan, and Jun Yang. All rights reserved. Short sections
of text, not to exceed two paragraphs, may be quoted without explicit permission provided that
full credit, including © notice, is given to the source.
Bank Stress Testing, Human Capital Investment and Risk Management
Thomas Schneider, Philip Strahan, and Jun Yang
NBER Working Paper No. 30867
January 2023
JEL No. G20

ABSTRACT

This paper studies banks’ investment in risk management practices following the Global
Financial Crisis and the advent of stress testing. Banks that experienced greater losses during the
Crisis exhibit stronger demand for risk management talents. Banks increase their demand for
highly skilled stress test labor in anticipation of a test and following poor performance on a test.
Following this higher demand, banks exhibit lower systematic risk and lower profitability. While
stress testing has modernized banks’ internal risk management by spurring the acquisition of
highly skilled risk management talent, recent changes to the tests could erode its efficacy.

Thomas Schneider Jun Yang

Old Dominion University Notre Dame
Norfolk, VA 23529 South Bend, IN 46556
[email protected] [email protected]

Philip Strahan
Carroll School of Management
324B Fulton Hall
Boston College
Chestnut Hill, MA 02467
and NBER
[email protected]
1. INTRODUCTION
The 2007-2009 Global Financial Crisis (GFC) represents a massive risk management

failure on the part of both private-sector financial firms and their regulators. The financial system

itself became over-leveraged due to the growth of products like credit default swaps in the early

2000s, and lending practices became lax due to the growth of securitization in areas such as

subprime mortgage lending (Mian and Sufi, 2009). Many of these changes occurred, or were

accelerated, by gaps in regulation. 1 In the aftermath of the GFC, regulators have attempted to fill

these gaps with new programs designed to foster resiliency, and bank stress testing is among the

most important of these regulatory innovations. Although a burgeoning literature studies the

efficacy of stress testing, little is known about how stress testing has changed the way banks

approach risk management.

This paper analyzes banks’ investment in human capital aimed at managing risk following

the GFC and the advent of bank stress testing. 2 We first document a rising demand for risk

management expertise over the subsequent decade, and we show that this demand is greater among

banks that suffered heavier losses during the GFC. 3 Next, we examine how stress testing has

shaped banks’ demand for risk managers and find that banks seek to hire highly skilled stress test

labor in anticipation of a test and following poor performance on a test. Finally, we find that this

higher demand translates to lower systematic risk and lower profitability. Overall, our results

1
OTC derivatives markets were loosely regulated and offered a means to increase economic leverage for firms like
AIG. Novel securitization techniques enabled banks to maintain ostensibly low levels of risk and hold artificially low
levels of regulatory capital. In one form of such regulatory arbitrage, Acharya et al. (2013) show that asset-backed
commercial paper vehicles used to support loan securitization were designed specifically to avoid bank capital
regulations. These structures collapsed spectacularly starting in the summer of 2007 and were the harbinger of the
full-on crisis in late 2008.
2
Our analysis is done at the level of the bank holding company (“BHC”). To avoid these clumsy terms, we instead
generally refer to ‘banks’ throughout the text.
3
Kho et al. (2000) focuses on loss exposure to emerging market crises of the 1990s, but they do not analyze subsequent
risk management hiring, as we do.

1
suggest that stress testing has catalyzed advancement in risk management through the acquisition

of highly skilled talent.

Our data come from Burning Glass Technologies (BGT), which reports comprehensive

information for online advertisements by firms looking to hire at all levels across the organization.

These data, we argue, capture labor demand, as they reflect the kinds of skills banks are seeking

to employ. As such, they help us understand what banks are trying to achieve in their human

resources policies, as opposed to whom they succeed in hiring (which would capture both labor

supply and demand effects). 4

Using BGT, we document that total demand for risk management jobs climbed nearly six-

fold from 12,000 job posts by banks in 2010 to over 70,000 posts by 2019, before falling during

the coronavirus pandemic (Figure 1A). The data clearly show that risk management has been a

persistent and increasingly important factor at banks in the years following the GFC.

To understand this pattern, we consider two effects of the GFC on banks’ labor demand.

First, we look for, and find, evidence that banks learned about their risk exposure from the trial by

fire during the Crisis and responded by demanding more risk management skills. Specifically, we

focus on banks’ exposure to losses during the 2008-2009 period, and we find that banks with higher

losses during the GFC subsequently increase their hiring of people with risk management skills

relative to less-affected banks in the cross-section. This effect is large economically, and it is not

subsumed by bank characteristics such as size, real estate exposure, and whether the bank is subject

to stress testing.

4
We do not have access to data on actual bank hiring by skill domain. As far as we know, no such systematic data
exists for U.S. banks.

2
 Next, we shift our focus to stress testing given its importance in the post-crisis regulatory

regime. Stress tested banks dominate the demand for risk management jobs, accounting for over

80% of these job posts in any given year (Figure 1A). Their demand for risk managers is higher

not only because they are bigger, but also as a fraction of their hiring. Risk management

represented about 4% of total job posts across all banks in 2010, but this share tripled to over 12%

of job posts at stress-tested banks by the end of the decade, outpacing the increase to 8% at other

banks (Figure 1B).

Empirically, the setting of stress testing allows us to exploit within-bank variation in panel

models since the set of banks subject to stress tests changes over time, and the effect of the test on

banks also varies temporally. 5 We show that stress testing spurred bank investment in highly

skilled workers, and that this led to lower risk. Specifically, banks that expect to be stress tested in

the next cycle increase their demand for stress-test specific labor. This effect is greatest at banks

who perform worse in or fail the stress tests. Following higher demand for stress-test specific labor,

banks exhibit lower systematic risk and lower profitability. This is strongest when banks demand

jobs requiring higher education or advanced quantitative skills. Overall, stress testing has helped

advance internal risk management practices at banks. By extension, the general tightening of

regulation itself helps explain the marked upward trend in banks’ demand for risk-management

skills.

The effects we document surrounding stress testing likely understate the overall impact of

regulatory tightening after the crisis on demand for risk management. Many dimensions of

5
We are using the term ‘stress testing’ a bit loosely here and later in the paper. Most banks have been subjected to
some level of these tests, as dictated in the Dodd-Frank Act. However, only banks with assets over $50 billion faced
the full effect of these tests via the supervisory stress tests known as the Comprehensive Capital Analysis and Review
(CCAR). These tests were used to tie stress test results to bank capital and were, until 2019, used to limit bank dividend
payouts in some cases. Hence, CCAR had much more ‘bite’ as a regulatory tool than the stress testing applied to
smaller banks.

3
regulatory change happened to all U.S. banks, making it hard to pin down their causal effects

empirically. We remove both time-invariant bank characteristics as well as time trends in our

analysis of stress tests to address identification concerns, but this approach may also remove

important common effects of regulatory change on skill acquisition. Even within the panel setting,

we argue that the measured effect of stress test outcomes on skill demand is likely to understate

the true effect because banks with a greater level of risk management expertise ought to perform

better on their stress tests; in other words, any reverse causation from a bank’s hiring of risk

management talent to stress test outcomes ought to attenuate our core finding.

In addition to strengthening internal risk management by talent acquisition, banks also

invest heavily in hiring external consultants. Globally, banks’ expenditure on consultants has

increased from $16 billion in 2007 to $29 billion in 2015, much of that for stress tests. 6 Since

individual banks’ expenditures on consulting services is not easily observable, the labor demand

response that we estimate is likely a lower bound on the aggregate investments that banks make in

their risk management practices. That is, our results likely understate the overall resources that

banks dedicate towards risk management and its subsequent effect on risk outcomes.

In studying post-GFC regulatory changes like stress tests, many researchers have focused

on the impact of stress testing on credit supply (e.g., Acharya, Berger and Roman (2018); Bassett

and Berrospide (2018); Cortes et al. (2020); and Doerr (2021)). There is also debate about the

efficacy of stress tests as a means of regulating capital and how results should be disclosed. Hirtle

et al. (2009) argue that stress tests have become a key tool in banking supervision, although Frame

et al. (2015) report that pre-GFC stress testing was ineffective for the GSEs. Goldstein and Leitner

(2018) model the tradeoffs faced by regulators regarding disclosure of stress test results. They

6
https://www.wsj.com/articles/stress-test-inc-billions-of-dollars-consultants-to-manage-other-consultants-
1467139620

4
focus on how disclosure of individual bank risks can affect inter-bank markets and risk sharing.

But little is known about how (and whether) stress testing reshapes banks’ internal risk

management practices, and whether such changes have affected actual risk outcomes for banks.

Our paper fills this gap by showing that banks respond to stress testing by changing their

investment in human capital, especially in highly skilled expertise, to improve their risk

management.

Because we are the first to analyze hiring patterns, our paper provides a fuller

understanding of how banks reacted to the 2008 Crisis. Many existing papers have emphasized the

risk-management failures of both banks and their regulators. The only research on the role of

bankers themselves that we know of focus on bank senior managers (e.g., Stulz and Fahlenbrach,

2011) or board members and structures (e.g., Ellul and Yerramilli, 2013; Minton et al., 2014).

While management and governance are clearly important – top executives and board members set

the priorities and incentives within an organization – these few individuals cannot singlehandedly

govern bank risk without the support of low- and mid-level employees.

Our results indicate that banks have become more highly skilled and quantitatively

sophisticated as a consequence of the trauma of the GFC, and that much of these changes were

spurred by regulatory pressure from stress tests. Recent changes in the stress testing regime,

however, have reduced the severity of these tests, as well as the level of public disclosure. Starting

in 2019 the Federal Reserve removed the qualitative assessment from stress testing, which had

formerly been a core component of the regime. Prior to 2019 the results from these qualitative

assessments – which depend on supervisory oversight of internal risk management policies,

practices, and governance – had been publicly disclosed (and are the source of our key measure),

along with the quantitative results. Our analysis suggests that this component of the stress testing

5
regime had the greatest impact on skill acquisition. Hence, these recent changes may indicate a

gradual decline of bank investment in risk management talent going forward.

2. STRESS TESTING

Stress testing began in the wake of the GFC in 2009, with the Supervisory Capital

Assessment Program (SCAP). This program succeeded in allaying concern about bank health and

led to rapid bank recapitalization. Following the success of SCAP, the Fed decided to implement

annual supervisory stress tests for banks with assets above $50 billion, rebranded as the

Comprehensive Capital Analysis and Review (CCAR), from 2011 onwards. We focus on the

CCAR banks in our analysis.

The CCAR process, until 2020, explicitly tied bank dividend and capital distributions to

the stress test results. CCAR captures each bank’s quantitative exposure to its adverse economic

scenarios (nine-quarter ahead potential paths for risk drivers) by constructing hypothetical nine-

quarter ahead paths for several regulatory capital ratios (e.g., the Tier 1 capital ratio, the Total risk-

based capital ratio and the Tier 1 leverage ratio). Until 2020, the forecasted paths of the capital

ratios embed each bank’s planned dividend increases and share repurchases. If one of the capital

ratios falls below the regulatory minimum during the forecast horizon, however, the bank was

required to reduce its planned capital payouts. In addition, CCAR provides a qualitative assessment

of each bank’s overall risk management practices. 7

In addition to CCAR, the Dodd-Frank Act requires the Fed to conduct a parallel set of

annual stress tests, known as the Dodd-Frank Act Stress Tests (DFAST). These began in 2013.

DFAST requires banks to run (and disclose) their quantitative stress tests results using their own

7
For a fuller description of the stress test regime, see Schneider, et al. (2022).

6
internally developed models. Under DFAST, each bank is required to simulate its capital ratios

using both the Fed’s scenario (common to all) and also using its own scenario. We focus on

measures using the Fed’s scenario.

During each year in our sample, the Fed first releases the results of DFAST in a document

containing the results from the Fed’s proprietary model. About one week later, they release a

second document with results from CCAR, again containing quantitative results based on the Fed’s

model, along with a discussion of the qualitative results of the test. Banks also release their results

around the same time, using a common set of scenario assumptions dictated by the Fed.

We build three stress testing measures from the publicly disclosed data for CCAR and

DFAST. First, we utilize all testing cycles from 2011 to 2021 to construct indicators for whether

a bank is tested in the current year (Testedi,t) or is expected to be tested in the following year

(Testedi,t+1).

Our second set of measure focuses on failures in CCAR. Failure occurs if the bank’s

stressed capital ratios fall below a regulatory minimum threshold, or if the Fed finds deficiencies

in a bank’s internal risk management policies or practices. As the Fed says, “The Federal Reserve’s

qualitative assessment of the capital plans focused on the robustness of a BHC’s internal capital

adequacy processes, including each BHC’s stress test under its own internally designed stress

scenario.” In some cases, a bank’s capital plans under CCAR are outright rejected; in other less

severe cases banks are required to make internal changes but may maintain their stated capital plan

(“conditional non-objections”). In these cases, banks commit to making improvements to their risk

management practice but receive approval of their planned capital distributions. These outcomes

were disclosed starting from 2013, upon which we build a failure metric which varies from 0 to 2.

7
A value of zero indicates no objection by the Fed; a value of one represents a conditional non-

objection, and a value of two represents an outright objection (failure).

Starting in 2020, the impact of the CCAR on bank capital planning was substantially

reduced. First, banks were no longer required to pre-commit to a capital plan, which until 2019

was used as a key component of the quantitative portion of CCAR. Second, the qualitative review

was no longer disclosed publicly. So, while we do extend our analysis into the last two years, we

code all banks as having passed the qualitative portion of CCAR in 2020 and 2021.

Our third set of measures equal the difference between the stressed value for a given capital

ratio (under the severely adverse scenario) from its value at the start of the test, based on each

bank’s internal model (i.e., using DFAST data). These variables are available from 2013 to 2021

and measure the exposure of a bank’s portfolio to the test. We focus on the results from bank’s

model under the Fed’s scenarios, instead of Fed’s model, because they represent bank’s own

assessment of their exposure to the scenarios before the results under the Fed’s model are disclosed.

3. HIRING DATA

To assess banks’ labor demand, we analyze granular job posting data provided by Burning

Glass Technologies (BGT). BGT collects information from online job advertisements via data

scraping techniques. These data cover the near universe of online job postings continuously from

2010 through 2021.8 Each data entry is a unique job posting and contains information such as the

name of the employer, location of the job, industry, required education/experience, and occupation

classification. One distinctive feature of the BGT database is that it provides a detailed description

of required skills listed in each job ad. We hand-match large banks from FR-Y9C data with job

8
See Hershbein and Kahn (2018) for an excellent overview of the BGT database.

8
posting data from BGT through bank/employer name and location. We consider a bank for our

sample if it ranks among the top 300 banks by total assets in any year during our sample period,

2010-2021.

We extract risk management job posts from BGT by filtering on the skill clusters and job

titles provided in the data. We define risk management jobs as those requiring at least one skill

that falls within the “Financial Risk Management” or “Financial Regulations” skill clusters and

contains one of the following key strings in the job title: analy, audit, ccar, compliance, credit,

econom, model, ppnr, quant, regulat, risk, or stress. We require this secondary filter as many bank

jobs involve extraneous forms of risk management, such as tellers managing cash drawers and

guards overseeing physical premises.

After hand-matching to BGT, we end up with 412 bank holding companies. This number

drops to 337 after we require these banks to have pre-Crisis Y9C data for our cross-sectional

analysis. Some banks drop out because previously they were not filed as bank holding companies,

such as Goldman Sachs. 9 In some tests we require stock market returns, which reduces the sample

further to 197, because some banks are private or foreign.

[Insert Figure 1 Here]

Figure 1 shows the increasing demand for risk management expertise throughout the past

decade. Figure 1.A. plots the raw number of risk management job posts and shows that it has

increased dramatically from approximately 12,000 in 2010 to over 70,000 in 2019 before the

coronavirus pandemic. This upward trend is dominated by stress-tested banks: the share of risk

management jobs posted by stress-tested banks is over 80% in any given year. Risk management

talents also become a more significant portion of labor force at banks, especially at stress-tested

9
All stress tested banks, including Goldman Sachs, are present in our panel analysis.

9
banks. Figure 1.B. demonstrates that the share of risk management job posts of all job posts is

about 4% for all banks in 2010, but this fraction climbs to over 12% for stress-tested banks by the

end of the decade, surpassing the increase to 8% at other banks.

[Insert Figure 2 Here]

Banks’ rising demand for risk management skills is not limited to the risk management

department; it can also be seen across various functions within the banking organization. We select

the largest four occupation categories: branch managers (ONET = “11-3031.02”), personal and

relationship bankers (ONET = “41-3031.02”), loan officers (ONET = “13-2072.00” or “43-

4131.00”), and analysts (ONET name contains the word “analyst”). 10 The plots in Figure 2

represent the number (left axis) and fraction (right axis) of job posts that require any risk

management skills for each major occupation. As Figure 2 demonstrates, the fraction of job posts

that require risk management skill has increased from below 10% in 2010 to over 40% in 2021 for

branch managers and personal bankers; and doubled for loan officers and analysts over the same

time period.

4. RESULTS

4.1 CROSS-SECTIONAL ANALYSIS: GFC AND LABOR DEMAND

Do banks scarred by losses during the Global Financial Crisis respond by strengthening

their risk management practice and refreshing their talent pool? To answer this question, we link

losses experienced in the Crisis with subsequent demand for risk-management skills. We merge

banks’ financial information from FR-Y9C reports and stock returns from CRSP with the BGT

data. Risk management job postings are aggregated at the bank level over 2010-2021, scaled by

10
The Occupational Information Network (O*NET) is a Standard Occupational Classification (SOC) based system.
BGT provides the O*NET occupation code using its proprietary coding rules for each job.

10
total job posts during the same period. (Summing over time removes temporal variation, which we

explore below.) We capture banks’ exposure to losses in the GFC in two ways. First, we use net

income summed across the worst quarters of the Crisis (from Q2 2008 through Q2 2009), scaled

by pre-Crisis total assets at the end of 2006. Second, we measure stock returns from the peak (June

2007) to trough (March 2009) of the Crisis. We control for banks’ pre-Crisis characteristics by

including log of assets, real estate loans/assets, and capital ratio at the end of 2006.

[Insert Table 1 Here]

Table 1 presents summary statistics. For an average bank in the sample, 5.2% of total job

posts during 2010-2021 are risk management positions. Cumulative net income over the Crisis

quarters is near zero for the average bank, and there is substantial variation in net income across

banks with a standard deviation of 2.2% of assets. Stock prices fell on average by more than 50%

from peak to trough, with a cross-sectional standard deviation of 27%.

To test the effect of GFC losses on hiring, we estimate regressions of the following form,

𝑅𝑖𝑠𝑘 𝑀𝑎𝑛𝑎𝑔𝑒𝑚𝑒𝑛𝑡 𝐽𝑜𝑏𝑠𝑖 = 𝛽 𝐿𝑜𝑠𝑠𝑒𝑠𝑖 + 𝜆𝑋𝑖 + 𝜖𝑖 , (1)

where 𝑖 indexes bank, Risk Management Jobs are measured over the twelve-year period from 2010

to 2021, Losses refers to either Net Income or Stock Return during the GFC as described above,

and 𝑋 represents a vector of pre-Crisis bank-level controls.

[Insert Table 2 Here]

The results in Table 2 support the idea that banks which experienced higher losses during

the GFC subsequently post more risk management positions. For example, Column 3 indicates

that a one standard deviation decrease in banks’ net income during the Crisis is associated with

11
0.6% (=0.27*2.23) increase in labor demand for risk management talent, corresponding to 12%

(=0.6/5.235) of the sample mean. Similarly, columns 4-6 suggests that stock returns correlate

negatively with risk management job posts, with similar economic magnitude. 11

Column 2 of Table 2 suggests that stress-tested banks demand more risk management

skills, with a very large magnitude equal to about 45% of the mean (=2.287/5.235). But this effect

becomes statistically insignificant with bank size included. The stress-tested indictor equals one

for banks ever tested during our sample period, so it is mechanically related to size. Larger banks

clearly demand more risk management talent than smaller ones, as columns 3 and 6 show, but this

effect has some ambiguity in interpretation. Large and small banks differ not only in their exposure

to post-Crisis regulatory changes such as stress testing, but also across other dimensions. 12 The

panel approach below, however, allows us to separate size from regulatory effects.

[Insert Table 3 Here]

Next, we test whether banks’ responses to losses from the GFC are distinct from banks’

responses to performance in other periods. We divide sample years into two non-overlapping

periods of 2010-2015 and 2016-2021 and calculate the fraction of risk management job postings

for each bank. In Panel A of Table 3, we regress risk management jobs from the first half of the

sample on losses from the GFC. In Panel B, we regress job posts from the second half of the sample

on losses from a placebo period from 2013 to 2015. To keep the same window length as in Panel

A, losses are measured from Q2 2014 to Q2 2015. Consistently, controls are measured at the end

11
We have also tested whether banks with a board-level risk-management committee before the 2008 crisis
increased their subsequent stress-test hiring. These results suggest little evidence that this governance variable
matters, although its inclusion does not change our main findings on GFC loss exposure.
12
For robustness, we estimate Equation (1) using only the sample of non-stress tested banks and report the results in
Appendix 3. The results are similar to the full sample, meaning that non-stress tested banks respond to GFC losses
by hiring risk managers as well.

12
of 2012. We find significant results only when losses are measured from the GFC, and we conclude

that the scarring that occurred during the crisis was indeed unique in shaping risk management

practices going forward.

4.2 PANEL ANALYSIS: STRESS TESTS AND LABOR DEMAND

Stress testing is not only the most important change in bank regulation emerging from the

GFC, but its implementation has been staggered, and its effects have varied by bank over time.

For example, failures and objections occur in some years but not others (Schneider et al., 2022).

Hence, we can bring panel analysis to bear in answering how stress testing has affected banks’

demand for relevant skills. To do so, we select jobs within the risk-management category that

require “stress testing” or “CCAR regulatory rules” as skills and categorize them as Stress Test

Jobs.13 We classify the remaining risk management jobs that do not specifically mention stress

testing skills as Other Risk Management Jobs.

[Insert Figure 3 Here]

Figure 3 plots the prevalence of stress test jobs and other risk management jobs at stress-

tested banks in the top and bottom panels, respectively. Demand for stress testing jobs increased

in the years following the implementation of the tests, accelerating in 2014 as the Fed expanded

its scope to include smaller banks (see Appendix 2), and retreating after 2017 as the Fed’s

leadership on banking supervision changed. 14 In subsequent years, changes to stress testing have

reduced its effects on large banks, and smaller banks have moved to a biannual testing cycle.

13
Although the Dodd-Frank Act contains a stress testing component (DFAST), the Act is broad and introduced a
substantial amount of risk management regulation that is not specific to stress tests. Job posts that require Dodd-Frank
regulatory skills are only classified as Stress Test Jobs if they also require stress testing skills or CCAR regulatory
skills.
14
President Donald Trump appointed Randy Quarles as the Vice Chair for Bank Supervision at the Federal Reserve
in 2017, replacing Daniel Tarullo. The stress-testing regime was developed during Tarullo’s tenure as head of
supervision at the Fed.

13
 [Insert Figure 4 Here]

In Figure 4, we plot stress test jobs posted by non-bank employers and the Federal Reserve.

Employers in the top figure include consulting firms that specialize in bank stress testing (e.g.

Accenture, Deloitte, and KPMG) and other firms that provide support services for stress testing

such as IBM. These firms expanded their labor demand for stress test jobs in the early years when

stress tests were first implemented on a large and consistent scale. This demand declined since

2017, consistent with the pattern observed for stress tested banks (Figure 3). The bottom panel of

Figure 4 depicts the Federal Reserve’s job posts for stress test positions. The rise and fall of the

Fed’s demand leads the trends in demand by stress tested banks and non-bank employers by about

one year. Even though many consulting firms have added stress testing services to their lines of

business, banks’ demand for stress test jobs dwarfs that of the consulting industry.

We further partition jobs into high- and low-skilled bins, based on whether they require

higher education or advanced quantitative skills. Since bank jobs typically require at least some

college, our measure considers whether the job post lists a preference for a master’s or doctorate

degree. To capture quantitative requirements, we flag jobs containing the most frequently listed

quantitative or programming skills such as C++, MATLAB, physics, and SAS.15 We define a job

to be High Skilled if it requires either an advanced degree or at least one quant skill and Low Skilled

otherwise.

[Insert Figure 5 Here]

[Insert Figure 6 Here]

Figure 5 shows that stress test jobs require considerably more high skilled workers than

other bank jobs. Out of all posts, only 14% are High Skilled. This fraction rises to 29% for all risk

15
We report the full list of advanced quantitative and programming skills in Appendix 1.

14
management jobs and to 53% for stress test jobs. Figure 6 plots the proportion of High Skilled

stress test jobs compared to other risk management jobs by year. The stark difference between the

panels shows that stress test jobs consistently require a higher skilled workforce than other types

of risk management. In most years, over half of stress test jobs require an advanced degree or quant

skills. In our regressions, we expect high skilled stress test jobs to produce the strongest effects,

and we find that they do.

[Insert Table 4 Here]

Panel A of Table 4 presents summary statistics for the job measures used in our bank-year

panel regressions (outlined below) for the sample of stress-tested banks. Our sample begins with

the 38 banks which are ever stress tested under CCAR between 2010 and 2021, producing 456

(=38×12) bank-years. Each job measure observation represents the number of job posts within that

category during a [t-3, t+3], where t denotes the month when that year’s stress test results were

publicly disclosed, as a fraction of the bank’s total job posts over the same period. For example,

the 2020 stress test results were released in June, so our measure considers job posts from March

through September of that year. We include the months leading up to the disclosure because banks

have private information about their performance on the test prior to the public release. We exclude

months further from the tests to avoid potential simultaneous effects between labor demand and

our variables of interest, although our results are quantitatively similar if we consider the whole

year. The mean of Risk Management Jobs is 9.15, indicating that risk management job posts

comprise 9.15% of all job posts within the sample window in a typical bank-year. This share is

1.20% for Stress Test Jobs and 7.88% for Other Risk Management Jobs. Further partitioning jobs

into High Skilled and Low Skilled confirms that stress test jobs require higher skilled workers than

15
other risk management jobs. For stress test jobs, 50% (= 0.60÷1.20) require an advanced degree

or quant skills, while only 27% (= 2.15÷7.88) of other risk management jobs do.

In our first set of panel regressions, we examine the impact of stress tests on banks’ labor

demand, including all banks which ever undergo the CCAR test. We hypothesize that banks

scheduled to be tested in the following cycle expand their stress testing work force in anticipation

of the test. Specifically, we estimate the following regression:

𝑆𝑡𝑟𝑒𝑠𝑠 𝑇𝑒𝑠𝑡 𝐽𝑜𝑏𝑠𝑖,𝑡 = 𝛽 𝑇𝑒𝑠𝑡𝑒𝑑𝑖,𝑡+1 + 𝜆𝑋𝑖,𝑡 + 𝛼𝑖 + 𝛼𝑡 + 𝜖𝑖,𝑡 (2)

where Testedi,t+1 indicates that bank i is tested in year t+1, and Stress Test Jobs is measured as a

fraction of the bank’s total job posts. The largest banks were tested across all cycles, while others

were added and dropped, as shown in Appendix Table 2. This heterogeneity provides within-bank

variation in our variable of interest (Testedi,t+1). Our model assumes that banks know at least one

year in advance whether or not they will face a test in the subsequent cycle (i.e. banks know at

time t whether they are tested at time t+1); hence, it makes sense that their hiring would respond

to their anticipation of an upcoming test. The vector X represents time-varying bank characteristics:

whether it is tested in year t, bank size (log of total assets), loan portfolios (loans and unused

commitments as a fraction of total assets), profitability (net income/assets), and capital (tier 1

capital ratio). We absorb time-invariant bank characteristics and macro time trends by including

bank fixed effects and time fixed effects, and we cluster standard errors at the bank-year level.

[Insert Table 5 Here]

We report estimates of Equation (2) in Table 5. Columns 1 to 4 of Panel A confirm that

banks increase their demand for stress test labor prior to facing a test, and that the effect is not

16
sensitive to whether or not we control for the contemporaneous test. In Column 4, our most

restrictive model, the coefficient of 0.290 on Testedi,t+1 implies that stress test job posts make up

0.29% more of a bank’s total job posts in the year prior to a test. This number is economically

large, representing an increase of over 25% (=0.29÷1.201) of the sample mean and 20%

(=0.29÷1.526) of the sample standard deviation. Columns 5 to 8 of Panel A present alternative

specifications where the dependent variable is Other Risk Management Jobs. In contrast to stress

test jobs, we find no significant relationship between facing a test and demand for other risk

management labor.

We estimate Equation (2) separately for High Skilled and Low Skilled stress test jobs in

Panel B. The results reveal that banks demand primarily high skilled stress test labor prior to a test:

the coefficient on Testedi,t+1 is statistically positive at the 1% level of significance in columns 1 to

4 for high skilled labor, while it is weak or insignificant in columns 5 to 8 for low skilled labor.

Table 5 shows that bank hiring responds to anticipated future tests. We next proceed to

investigate how banks’ labor demands respond to past stress-test performance, focusing on bank-

years for which stress test performance data are disclosed. This limits our sample to 215 bank-

years from 2013 to 2021 (see Appendix 2).16 We hypothesize that banks that fail tests subsequently

increase their demand for stress test related labor. As described above, we build a categorical

variable, Failure Score, that takes values of 0 (no objection or capital adjustments), 1 (conditional

non-objection), and 2 (objection) depending on the Fed’s response to the bank’s capital plan under

CCAR to capture the severity of a failure (Schneider, et al., 2022). Additionally, we consider the

three capital ratios consistently disclosed in all testing cycles: the Tier 1 Capital Ratio, the Total

Risk-based Capital Ratio (TRBC), and the Tier 1 Leverage Ratio. We define a bank’s Exposure

16
Due to the coronavirus pandemic, the Fed made disclosure voluntary in 2020. Our results are similar if we only
consider the 199 bank-years tested and disclosed from 2013 to 2019 (unreported).

17
for each ratio as the difference between the ratio’s actual value at the beginning of the test and the

ratio’s projected minimum value over the forecast horizon under DFAST (Cortes, et al., 2020).

Thus, Exposure measures how much a bank’s capital ratio is projected to decline during the test,

with higher values corresponding to greater losses.

[Insert Figure 7 Here]

Figure 7 visualizes the relationship between stress test performance and labor demand. The

first three pairs compare the average demand for stress test jobs following bank-years with above

vs. below median exposure using the three capital ratios. The last pair compares demand for stress

test jobs in the period following bank-years that pass vs. fail the stress test. In each pair, banks

with worse performance on the test demand more stress test related labor. We formalize this

analysis by regressing current period hiring on past stress test performance:

𝑆𝑡𝑟𝑒𝑠𝑠 𝑇𝑒𝑠𝑡 𝐽𝑜𝑏𝑠𝑖,𝑡 = 𝛽𝐹𝑎𝑖𝑙𝑢𝑟𝑒 𝑆𝑐𝑜𝑟𝑒𝑖,𝑡−1 + 𝜆 𝐸𝑥𝑝𝑜𝑠𝑢𝑟𝑒𝑖,𝑡−1 + 𝛼𝑖 + 𝛼𝑡 + 𝜖𝑖,𝑡 . (3)

Controlling for Exposure enables us to separate the effect of failing the test from the overall

riskiness of the bank.

[Insert Table 6 Here]

Table 6 presents results from estimating Equation (3). Columns 1 to 4 of Panel A show that

banks increase their demand significantly for stress test jobs following a stress test failure. The

effect of failure on hiring for other types of risk management jobs reported in columns 5 to 8 is

positive but not statistically significant. In Panel B, we find statistically positive effects of failure

on hiring for both High Skilled and Low Skilled jobs, and the magnitude of the relationship is

strongest for higher skilled labor. Taken together with the results in Table 5, we conclude that

18
stress testing has had meaningful effects on banks’ internal labor demand, particularly among high

skilled workers. The regression results also suggest that the main impact of stress testing works

through our failure score measure, which depends on the disclosed outcomes from the qualitative

portion of CCAR (recall section 2). As noted, this component of the public disclosure was ended

after 2019, suggesting that the beneficial impact of stress testing on banks’ human capital

investments may be waning.

We would like to draw a causal inference from the results of both Tables 5 and 6. That is,

we argue that stress tests causally shape banks’ human capital investment, not the reverse. In the

first set of panel models (Eq. 2), where we focus on just the effect of the stress test cycle (or, a

bank’s exposure to future stress tests), this interpretation seems easy to justify because the stress

test cycle depends only on actions taken by the Federal Reserve, not the bank. In our second set of

models, we focus on stress test outcomes. Here, reverse causality could bias the effects we observe

down because hiring fewer skilled employees for stress test compliance would likely raise the

probability of failing the test. In fact, we find a positive effect on stress test hiring on the odds of

failure. Hence, the true causal effect is, if anything, likely to be more positive than our estimates

would suggest.

4.3 HIRING AND BANK RISK

Having established the relationship between stress tests and labor demand, we next ask:

Does this new labor change bank’s risk characteristics? We hypothesize that skilled stress test

labor improves risk management by reducing systematic risk exposure. We focus on three risk

metrics: Expected Shortfall, Beta, and Volatility. 17 Expected Shortfall is the model-implied

17
The source for all three measures is the Volatility Laboratory of the NYU Stern Volatility and Risk Institute
(https://vlab.stern.nyu.edu).

19
expected fractional loss of the bank’s equity in a crisis when the aggregate market declines

significantly, following Acharya, Pederson, Phillipon and Richardson (2017); Beta is the dynamic

conditional beta with respect to the MSCI World Index, as in Engle (2016); and Volatility is the

annualized standard deviation of stock returns estimated via the Glosten, Jagannathan, and Runkle

(1993) GARCH model. The first two metrics capture systematic risk and the latter measures total

risk. Since stress testing is principally about mitigating systematic risk, we expect to observe the

strongest effects for Expected Shortfall and Beta. We regress year-end risk measures on current

year labor demand while controlling for whether the bank is tested current or upcoming cycle,

𝑌𝑖,𝑡 = 𝛽 𝑆𝑡𝑟𝑒𝑠𝑠 𝑇𝑒𝑠𝑡 𝐽𝑜𝑏𝑠𝑖,𝑡 + 𝜆 𝑇𝑒𝑠𝑡𝑒𝑑𝑖,𝑡+1 + 𝛾 𝑇𝑒𝑠𝑡𝑒𝑑𝑖,𝑡 + 𝛼𝑖 + 𝛼𝑡 + 𝜖𝑖,𝑡 (4)

and alternatively, controlling for the bank’s performance on the test:

𝑌𝑖,𝑡 = 𝛽 𝑆𝑡𝑟𝑒𝑠𝑠 𝑇𝑒𝑠𝑡 𝐽𝑜𝑏𝑠𝑖,𝑡 + 𝜆 𝐹𝑎𝑖𝑙𝑢𝑟𝑒 𝑆𝑐𝑜𝑟𝑒𝑖,𝑡 + 𝛾 𝐸𝑥𝑝𝑜𝑠𝑢𝑟𝑒𝑖,𝑡 + 𝛼𝑖 + 𝛼𝑡 + 𝜖𝑖,𝑡 . (5)

Although risk outcomes (𝑌𝑖,𝑡 ) and the variable of interest (𝛽 𝑆𝑡𝑟𝑒𝑠𝑠 𝑇𝑒𝑠𝑡 𝐽𝑜𝑏𝑠𝑖,𝑡 ) share the same

time subscript, we measure risk as of December in each year while measuring labor demand in a

window around that year’s stress test disclosures in either March or June. Thus, the relationships

in Equations (4) and (5) are not contemporaneous, and we interpret them as causal. As before,

reverse causality would predict high-risk banks to demand more stress test labor, which is the

opposite of our findings.

[Insert Table 7 Here]

20
 Equation (4) uses the full sample of banks which are ever tested, focusing on the tested vs.

non-tested years. In contrast, Equation (5) considers only bank-years with publicly disclosed test

results and studies the effects of the test outcome on risk. We report estimations of Equations (4)

and (5) in Panels A and B of Table 7, respectively. The coefficient on Stress Test Jobs is negative

in all columns, and it is statistically negative in estimations involving the two measures of

systematic risk. The coefficient of interest in column 2 of -0.487 implies that a one-standard

deviation increase in Stress Test Jobs relates to a reduction in Expected Shortfall of 74 basis points

(=0.487×1.526). 18 This effect is economically large, representing nearly 10% of Expected

Shortfall’s sample standard deviation of 7.63%. The reduction in systematic risk as measured by

Beta is of similar economic significance. The coefficient of interest in column 4 of Panel A is -

0.022, meaning that a one-sigma increase in High Skilled stress test jobs implies a reduction in

beta by 0.033 (=0.022×1.526), or 12% (=0.033÷0.280) of one sample standard deviation. The

absence of statistical significance in columns 5 and 6 for Volatility is consistent with the notion

that stress testing helps banks understand and reduce their systematic risk but not necessarily their

idiosyncratic risk.

The effects of stress test jobs are economically larger in Panel B, where we control for

stress-test outcomes. Stress test outcomes are forward looking risk metrics, and the third measure

of exposure (T1 Leverage) is consistently a good predictor of the Expected Shortfall and Beta, as

one would expect if the tests achieve their purported objectives. The increased magnitude on stress-

test jobs, after controlling for test results, supports our argument that that reverse causality – the

possibility that high-risk banks seek to hire more risk management experts – attenuates our results.

[Insert Table 8 Here]

18
Since Expected Shortfall is measured in absolute terms, a reduction in expected shortfall corresponds to a decline
in systematic risk.

21
 Next, we re-estimate these equations while considering High Skilled and Low Skilled stress

testing jobs separately and report the results in Table 8. The coefficient on High Skilled stress test

jobs is strongly negative in columns 1 to 4, confirming that it is those workers with advanced

degrees and quant skills driving the reduction in systematic risk. Low-Skilled jobs, in contrast, have

no effect. The economic effects of High Skilled jobs are also larger. Using the parameter estimate

of -1.396 in column 2 of Panel A, a one-standard deviation increase for high skilled stress test jobs

corresponds to a 100 basis point smaller Expected Shortfall (=1.396×0.713). For Beta, the

coefficient of -0.055 in column 4 of Panel A implies that a one-sigma increase in High Skilled

stress test jobs predicts a reduction of 0.039 (=0.055×0.713), representing 14% (=0.039÷0.280) of

Beta’s sample standard deviation. As before, we find no significant effects of hiring on Volatility.

And, we find larger magnitudes when we control for the stress-test outcomes (compare Panels A

and B).

The results suggest that the hiring of stress-test related labor lowers risk but, controlling

for that factor, we do not find lower risk at banks in anticipation of a test. In particular, Testedi,t+1

increases hiring (recall Tables 5 & 6), which in turn lowers risk (Tables 7 & 8); but Testedi,t+1 itself

is not correlated with risk outcomes. Hence, banks respond to the prospect of an upcoming test by

demanding skilled workers, and the amount of demand for these workers does correlate negatively

with risk outcomes.

Again, our aim is to draw a causal inference from these effects. As we argue above, if

anything reverse causality would attenuate the negative effect of skilled hiring on risk. Banks with

higher risk would tend to want more, not fewer, employees with strong risk management skills;

such banks would also face pressure from their regulator to hire such people. Both effects would

22
tend to bias our measured effect down (i.e., toward zero), and in fact our estimates become more

negative when we capture risk by controlling for stress-test outcomes.

4.4 HIRING AND BANK PROFITS

In our last set of tests, we examine the relationship between demand for stress test jobs and

bank profitability. We hypothesize that high skilled stress test jobs result in lower profitability, in

part due to reduced risk systematic exposures as evidenced above and in part because these kinds

of workers are expensive. Since high skilled jobs require advanced degrees and in-demand quant

skills, these employees may command a salary premium, thereby increasing the bank’s salary

expense. We estimate equations of the form,

𝜋𝑖,𝑡+1 = 𝛽 𝑆𝑡𝑟𝑒𝑠𝑠 𝑇𝑒𝑠𝑡 𝐽𝑜𝑏𝑠𝑖,𝑡 + 𝜆𝑋𝑖,𝑡 + 𝛼𝑖 + 𝛼𝑡 + 𝜖𝑖,𝑡 (6)

where 𝜋 is return-on-assets or average employee salary.

[Insert Table 9 Here]

We report estimates of Equation (6) in Table 9. In column 1, we find that demand for high

skilled stress test jobs corresponds to reduced bank profitability. The parameter estimate of -0.326

implies that a one standard deviation increase in High Skilled stress test labor leads to a 20 basis

point (=0.326×0.713) reduction in return-on-assets, or 22% of the sample mean of 0.91%. In

column 2, we find that Low Skilled stress test jobs also have a negative effect on profitability,

although the effect of High Skilled stress test jobs is stronger. Columns 3 and 4 show that High

Skilled stress test jobs raise the average employee salary. The coefficient on High Skilled stress

test jobs in column 3 of 5.356 implies that a one sigma increase in high skilled stress test jobs

23
increases the average employee salary by nearly $4,000 (=5.356×0.713×1000). Low Skilled stress

test jobs enters negatively in column 4, implying that these jobs receive lower pay, although it is

not statistically significant. In all specifications, Other Risk Management Jobs exhibit no

significant effect on bank profitability.

5. CONCLUSION

This paper provides new evidence on bank internal risk management practices since the

Global Financial Crisis by studying the jobs and skills demanded for risk management positions.

Banks’ overall demand for risk management jobs increased dramatically in the decade following

the Crisis, although this pattern has reversed in recent years. Banks which suffered the largest

losses in the Crisis responded by increasing their demand for risk-management talent the most. In

addition, the stress test regime led banks to increasing their demand for a skilled and educated risk

management workforce. These skilled professionals help banks reduce systematic risk at the cost

of reduced profitability. Our results suggest that stress testing has succeeded in modernizing banks’

internal risk management practices by spurring the acquisition of highly skilled risk management

talent. Recent changes, however, indicate that these tests may have lower, or perhaps no, impact

of risk management hiring in the future.

24
 REFERENCES

Acharya, V. V., Berger, A. N., and Roman, R. A. (2018) Lending implications of US bank stress
tests: Costs or benefits? Journal of Financial Intermediation 34, 58–90.

Acharya, V. V., Pedersen, L. H., Philippon, T., & Richardson, M. (2017). Measuring Systemic
Risk. Review of Financial Studies, 30(1), 2–47.

Acharya, V. V., P. Schnabl and G. Suarez. (2013). Securitization without risk transfer. Journal of
Financial Economics, 107:515–36.

Candelon, B., and Sy, M. A. N. (2015) How did markets react to stress tests?, International
Monetary Fund.

Cornett, M. M., Minnick, K., Schorno, P. J., and Tehranian, H. (2020) An examination of bank
behavior around Federal Reserve stress tests, Journal of Financial Intermediation 41, 100789.

Cortés, K. R., Demyanyk, Y., Li, L., Loutskina, E., and Strahan, P. E. (2020) Stress tests and
small business lending, Journal of Financial Economics 136(1), 260–279.

Covas, F. (2017) Capital requirements in supervisory stress tests and their adverse impact on
small business lending, The Clearing House Staff Working Paper 2017-2.

Doerr, S. (2021) Stress tests, entrepreneurship, and innovation, Review of Finance 25(5), 1609–
1637.

Ellul, A., & Yerramilli, V. (2013). Stronger risk controls, lower risk: Evidence from US bank
holding companies. The Journal of Finance, 68(5), 1757-1803.

Engle, R. F. (2016). Dynamic Conditional Beta. Journal of Financial Econometrics, 14(4), 643–
667.

Fahlenbrach, R., & Stulz, R. M. (2011). Bank CEO incentives and the credit crisis. Journal of
Financial Economics, 99(1), 11-26.

Fernandes, M., Igan D., and Pinheiro, M. (2015) March madness in wall street: (What) does the
market learn from stress tests?, IMF Working Paper 15/271.

Flannery, M., Hirtle B., and Kovner A. (2017) Evaluating the information in the Federal Reserve
stress tests, Journal of Financial Intermediation 29, 1–18.

Frame, S., Gerardi, K., and Willen P. (2015) The failure of supervisory stress testing: Fannie
Mae, Freddie Mac, and OFHEO, Federal Reserve Bank of Atlanta Working Paper 2015–3.

Gallardo, G.G., Schuermann, T. and Duane, M. (2016) Stress testing convergence, Journal of
Risk Management in Financial Institutions 9(1), 32–45.

25
Glosten, L. R., R. Jagannathan, and D. E. Runkle. (1993). On The Relation between The
Expected Value and The Volatility of Nominal Excess Return on stocks. Journal of Finance, 48:
1779-1801.

Goldstein, I., and Leitner, Y. (2018) Stress tests and information disclosure, Journal of Economic
Theory 177, 34–69.

Hirtle, B., Schuermann T., and Stiroh K. (2009) Macroprudential supervision of financial
institutions: lessons from the SCAP, Federal Reserve Bank of New York Staff Report 409.

Hershbein, B., and Kahn, L. B. (2018). “Do Recessions Accelerate Routine-Biased

Technological Change? Evidence from Vacancy Postings.” American Economic Review, 108 (7):
1737–72.

Kho, B., L. Dong, and R. Stulz (2000). “U.S. Banks, Crises, and Bailouts: From Mexico to
LTCM,” American Economic Review 90(2), 28-31.

Kohn, D., and Liang, N. (2019) Understanding the effects of the US stress tests. In Federal
Reserve System Conference: Stress testing: A discussion and review.

Minton, Bernadette, Taillard, Jerome and Rohan Williamson, 2014, “Financial Expertise of the
Board, Risk Taking, and Performance: Evidence from Bank Holding Companies,” Journal of
Financial and Quantitative Analysis 49(2), 351-380.

Mian, Atif and Amir Sufi, 2009. “The Consequences of Mortgage Credit Expansion: Evidence
from the U.S. Mortgage Default Crisis,” Quarterly Journal of Economics, November 2009,
124(4), 1449-1496.

Schneider, Thomas, Philip E. Strahan and Jun Yang, 2022. “Bank Stress Testing: Public Interest
or Regulatory Capture?” Review of Finance, forthcoming.

26
 Appendix 1: Data Definitions
Variable Name Variable Description
Job Measures*
Risk Management Jobs The number of job postings that (1) require at least one skill that can be categorized as “Financial
Risk Management” or “Financial Regulations”; and (2) the job title contains at least one of the
following key strings: “risk”, “audit”, “credit”, “analy”, “compliance”, “quant”, “model”, “regulat”,
“stress”, “ccar”, “ppnr”, “econom”.
Stress Test Jobs The subset of Risk Management Jobs that require “Stress Testing” or “CCAR Regulatory Rules”.
Other Risk Management Jobs The subset of Risk Management Jobs that are not Stress Test Jobs.
High Skilled Jobs requiring higher education (master’s or doctorate degrees) or advanced quantitative or
programming skills defined as the top-30 quant skills by frequency: Apache Hadoop, Apache Hive,
Apache Webserver, C++, Git, Java, JavaScript, Linux, MATLAB, Microsoft C#, Oracle, Oracle
PL/SQL, PERL Scripting Language, Physics, Python, R, SAP, SAS, Shell Scripting, SPSS, SQL,
SQL Server, Swift, Teradata, Teradata DBA, UNIX, UNIX Shell, VBA, Visual Basic, and .NET.
Low Skilled Jobs that are not classified as High Skilled.
*Note: All job measures are scaled by total job posts in regressions.
Stress Test Measures
Tested A binary measure equal to one if a bank-year was subjected to the CCAR stress test.
Failure A binary measure equal to one if a bank receives a conditional non-objection or an objection on the
CCAR stress test, and zero otherwise.
Failure Score A categorical measure equal to one or two if the bank received a conditional non-objection or
objection, respectively, from the Fed in response to its planned capital actions under CCAR, and
zero otherwise.
T1 Capital Tier 1 Risk-Based Capital Ratio: Tier 1 Capital divided by Risk Weighted Assets.
TRBC Total Risk-Based Capital Ratio: Total Risk-Based Capital divided by Risk Weighted Assets.
T1 Leverage Tier 1 Leverage Ratio: Tier 1 Capital divided by Total Assets.
Exposure of Ratio The difference between a ratio’s initial value and its minimum projected value during the 9-
quarter-ahead horizon under the CCAR’s severely adverse stress test scenario, minus projected
dividends.
Risk Measures†
Expected Shortfall The expected fractional loss of firm equity in a crisis when the aggregate market declines
significantly in a six-month period, following Acharya, Pederson, Phillipon and Richardson (2017).
Beta The Beta of the bank with respect to the MSCI World Index, using the Engle (2016) Dynamic
Conditional Beta model.
Stock Volatility The annualized volatility estimated via the Glosten, Jagannathan, and Runkle (1993) GARCH
model.
†
Note: Source: The Volatility Laboratory of the NYU Stern Volatility and Risk Institute
(https://vlab.stern.nyu.edu)
Financial Measures
ROA Net income divided by lagged total assets.
Salaries Total salary expense divided by the number of employees.
log Assets Natural log of total assets.
Loans / Assets Total loans and unused loan commitments to total assets.

27
 Appendix 2: Stress Tested Bank-Years
This appendix lists our sample of all stress-tested banks from 2011 to 2021. An ‘X’ indicates that the bank was stress-tested
in corresponding year. Our sample consists of 38 banks and 291 tested bank-years. Data from banks’ CCAR disclosures is
available from 2013 to 2020.
Bank 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021 Tests
Ally Financial Inc. X X X X X X X X X 9
American Express Co. X X X X X X X X X 9
Bank of America Corp. X X X X X X X X X X X 11
Bank of New York Mellon Corp. X X X X X X X X X X X 11
Barclays US LLC X X X X 4
BB&T Corp. (Truist Financial Corp.) X X X X X X X X X X 10
BVBA Compass Bancshares Inc. X X X X X 5
BMO Financial Corp. X X X X X X X 7
BNP Paribas USA Inc. X X X X 4
Capital One Financial Corp. X X X X X X X X X X X 11
CIT Group Inc. X 1
Citigroup Inc. X X X X X X X X X X X 11
Citizens Financial Group/RBS X X X X X X 6
Comerica Inc. X X X X 4
Credit Suisse Holdings USA X X X X 4
Deutsche Bank USA Corp. X X X X X X X 7
Discover Financial Services X X X X X X 6
Fifth Third Bancorp X X X X X X X X X 9
Goldman Sachs Group Inc. X X X X X X X X X X X 11
HSBC North America Holdings Inc. X X X X X X X X 8
Huntington Bancshares Inc. X X X X X X 6
JPMorgan Chase & Co. X X X X X X X X X X X 11
KeyCorp X X X X X X X X X 9
M&T Bank Corp. X X X X X X 6
Morgan Stanley X X X X X X X X X X X 11
MUFG Americas Holdings Corp. X X X X X X X 7
Northern Trust Corp. X X X X X X X X 8
PNC Financial Services Group Inc. X X X X X X X X X X X 11
RBC USA Holdco Corp X X X 3
Regions Financial Corp. X X X X X X X X X X 10
Santander Holdings USA Inc. X X X X X X 6
State Street Corp. X X X X X X X X X X X 11
SunTrust Banks Inc. X X X X X X X X 8
TD Group US Holdings LLC X X X X X X 6
UBS Americas Holding LLC X X X X X X X X X X X 11
US Bancorp X X X X 4
Wells Fargo & Co. X X X X X X X X X X X 11
Zions Bancorp X X X X 4
Number of Tested Bank-Years 18 18 18 30 31 33 34 35 18 33 23 291

28
 Appendix 3. Global Financial Crisis and Labor Demand at Non-Stress-tested Banks
This table provides robustness for Table 2 and shows that non-stress-tested banks respond to GFC losses with demand for
risk management positions. Risk Management Jobs measures aggregate risk management job posts during 2010-2021,
scaled by total job posts during the same time period. Losses are measured by Net Income from 2008Q2 through 2009Q2,
scaled by pre-crisis total assets in columns 1-3, and peak-to-through Stock Return during June 2007- March 2009 in
columns 4-5. All variables are fully defined in Appendix 1. Estimation is cross-sectional linear regression with
heteroskedasticity-robust standard errors (in parenthesis).
Risk Management Jobs 2010-2021
(1) (2) (3) (4) (5) (6)
-0.302** -0.302** -0.294**
NetIncome2008Q2-2009Q2
(0.125) (0.125) (0.137)

-0.032*** -0.032*** -0.026**

Stock Return
(0.011) (0.011) (0.013)

Log(Asset)2006 0.495** 0.451*

(0.218) (0.250)

Real Estate -0.013 0.037

Loans/Assets2006 (0.022) (0.023)

Tier 1 Cap Ratio2006 0.004 0.080

(0.045) (0.054)

Constant 5.084*** 5.084*** -1.561 3.324*** 3.324*** -5.948

(0.233) (0.233) (3.589) (0.491) (0.491) (4.196)
Observations 314 314 314 179 179 179
Adjusted R2 0.023 0.023 0.034 0.043 0.043 0.045

29
 Figure 1. Aggregate Trend in Risk Management Jobs
This figure shows that banks’ demand for risk management jobs has risen significantly over the past decade. This
trend is driven predominately by banks subject to stress testing by the Federal Reserve. The top panel reports the raw
number of banks’ risk management job posts, and the bottom panel plots risk management jobs as a fraction of total
job posts. Please see Appendix 1 for variable definitions.

Figure 1.A.

Figure 1.B.

30
 Figure 2. Banks’ Attention to Risk Management
This plot depicts the upward trend of banks’ general attention to risk management beyond the department of risk
management. The four figures demonstrate the numbers and fractions of job posts that require risk management skills
for the four major occupation categories: branch managers (upper left), personal and relationship bankers (upper right),
loan officers (lower left), and analysts (lower right).

31
 Figure 3. Stress Test Jobs at Stress-Tested Banks
This figure decomposes Risk Management Jobs into Stress Test Jobs and Other Risk Management Jobs at stress-tested
banks. Stress Test Jobs are those requiring Stress Testing or CCAR Regulatory Rules skills in BGT. The top panel
shows that banks subject to stress testing dramatically increased their demand for Stress Test Jobs during the first half
of the decade. The second panel shows these banks’ demand for Other Risk Management Jobs steadily rose throughout
the decade. Figures plot the number of job posts (left axis) and the fraction of job posts out of total job posts (right
axis).

Figure 3.A.

Figure 3.B.

32
Figure 4. Stress Test Jobs Posted by Other Employers
This figure illustrates the demand for stress test jobs by non-bank employers. Stress Test Jobs are defined in Appendix
1. The top panel shows such demand by consulting firms and others, whereas the bottom panel demonstrates the
demand by the Federal Reserve. Figures plot the number of job posts (left axis) and the fraction of job posts out of
total job posts (right axis).

Figure 4.A. Consulting Firms and Others

Figure 4.B. The Federal Reserve

33
 Figure 5. Proportion of High Skilled Jobs
This figure shows that Stress Test Jobs require higher skilled workers than other types of jobs. Stress Test Jobs is
defined as the subset of risk management jobs that require skills for “Stress Testing” or “CCAR Regulatory Rules” as
a fraction of the bank’s total job postings. High Skilled jobs are those requiring higher education (master’s or doctorate
degrees) or advanced quantitative skills as listed with variable definitions in Appendix 1.

34
 Figure 6. Demand for High-Skill Jobs
This figure shows that Stress Test Jobs require relatively high amounts of skill compared to other risk management
jobs. High Skilled jobs are those requiring higher education (master’s or doctorate degrees) or advanced quantitative
skills as listed in Appendix 1.

35
 Figure 7. Exposure to Stress Tests and Labor Demand
This figure shows that banks’ demand for stress-test specific talent increases following poor stress test performance.
Each plot represents the relative demand for stress-test specific talents at banks with high (above median) vs low
(below median) exposure to the stress tests. The red bars represent the average demand for stress test jobs for bank-
years with exposure to a certain capital ratio above the sample median for exposures (first three pairs) and stress test
failures (last pair). The green bars represent the demand for stress test jobs for bank-years with exposure to a certain
capital ratio below the sample median and stress test passes.

36
 Table 1. Cross-sectional Summary Statistics
This table presents cross-sectional summary statistics at the bank level. The cross-sectional sample consists of all
banks that rank among the top 300 banks by total assets in any year from 2010 to 2021 for which pre-Crisis Y9C
data is available. Risk Management Jobs are defined as risk management job posts scaled by total job posts by a
bank over the period of 2010 to 2021. Net Income is measured from 2008Q2 to 2009Q2, divided by total assets at
the end of 2006. Stock Return is measured at the bank level over the period of June 2007 through March 2009. All
continuous variables are winsorized at the 1st and 99th percentiles. All variables are fully defined in Appendix 1.
Obs. Mean Std. Dev. p25 p50 p75
Risk Management Jobs2010-2021 % 337 5.235 4.199 2.405 4.056 7.021
Net Income2008Q2-2009Q2 % 337 0.013 2.227 -0.260 0.611 1.280
Stock Return2007Jun-2009Mar % 197 -52.132 26.935 -73.543 -54.515 -32.860
Log(Asset)2006 337 14.887 1.450 13.923 14.518 15.510
Real Estate Loans/Asset2006 % 337 48.780 14.948 40.670 50.108 59.518
Cap Ratio2006 % 337 11.766 3.807 9.570 10.910 12.640

37
 Table 2. The Global Financial Crisis and Labor Demand for Risk Management Talents
This table shows that banks that experienced more losses during the GFC exhibit a higher demand for risk
management positions in the following decade. Risk Management Jobs measures aggregate risk management job
posts during 2010-2021, scaled by total job posts during the same time period. Losses are measured by Net Income
from 2008Q2 through 2009Q2, scaled by pre-crisis total assets in columns 1-3, and peak-to-through Stock Return
during June 2007- March 2009 in columns 4-5. All variables are fully defined in Appendix 1. Estimation is cross-
sectional linear regression with heteroskedasticity-robust standard errors (in parenthesis).
Risk Management Jobs2010-2021
(1) (2) (3) (4) (5) (6)
Net Income2008Q2-2009Q2 -0.294** -0.270 **
-0.273 **

(0.119) (0.120) (0.130)

Stock Return2007Jun-2009Mar -0.033*** -0.030*** -0.026**
(0.011) (0.011) (0.013)
**
Stress-tested Bank 2.287 -0.326 1.454 0.186
(0.968) (1.231) (1.071) (1.205)
Log(Asset)2006 0.582*** 0.492*
(0.217) (0.252)
Real Estate Loans/Assets2006 -0.018 0.019
(0.021) (0.023)
Tier 1 Cap Ratio2006 0.006 0.080
(0.044) (0.055)
Constant 5.238*** 5.082*** -2.585 3.382*** 3.441*** -5.613
(0.227) (0.233) (3.522) (0.488) (0.483) (4.126)
Observations 337 337 337 197 197 197
Adjusted R2 0.021 0.037 0.058 0.047 0.053 0.054

38
 Table 3. Global Financial Crisis, Labor Demand, and Time Split
This table shows that banks’ responses to losses from the GFC are distinct from banks’ responses to performance in other
periods. Panel A shows that banks respond to GFC losses with higher demand for risk management positions. Panel B shows
that banks exhibit limited response to performance in the latter half of the sample. Risk Management Jobs are defined as risk
management job posts scaled by total job posts by a bank over the sample period. Losses are measured by Net Income and
Stock Return. All variables are fully defined in Appendix 1. Estimation is cross-sectional linear regression with
heteroskedasticity-robust standard errors (in parenthesis).
Risk Management Jobs2010-2015
Panel A: 2010 – 2015 (1) (2) (3) (4) (5) (6)
Net Income2008Q2-2009Q2 -0.406*** -0.395** -0.367 **

(0.155) (0.156) (0.181)

Stock Return2007Jun-2009Mar -0.019 -0.017 -0.013
(0.012) (0.012) (0.013)
Stress-tested Bank 1.174 0.295 0.789 1.300
(0.884) (1.393) (0.967) (1.419)
Log(Asset)2006 0.369 0.189
(0.253) (0.303)
Real Estate Loans/Assets2006 0.015 0.040
(0.032) (0.026)
Tier 1 Cap Ratio2006 0.117 0.168
(0.099) (0.117)
Constant 4.862*** 4.779 ***
-2.788 3.601*** 3.634*** -3.039
(0.314) (0.333) (4.394) (0.730) (0.729) (5.050)
Observations 323 323 323 192 192 192
Adjusted R2 0.023 0.022 0.021 0.005 0.002 0.003
Risk Management Jobs2016-2021
Panel B: 2016 - 2021 (1) (2) (3) (4) (5) (6)
Net Income2014Q2-2015Q2 -0.280 -0.235 -0.173
(0.306) (0.294) (0.311)
Stock Return2013Jun-2015Mar -0.008 -0.002 -0.004
(0.020) (0.020) (0.021)
Stress-tested Bank 4.365*** 0.322 3.634*** 0.039
(0.976) (1.415) (1.023) (1.386)
***
Log(Asset)2012 0.860 0.902***
(0.247) (0.315)
Real Estate Loans/Assets2012 -0.022 -0.004
(0.020) (0.024)
Tier 1 Cap Ratio2012 0.007 0.050
(0.052) (0.050)
Constant 5.934*** 5.475 ***
-6.600 5.713*** 5.104*** -9.188*
(0.559) (0.537) (4.150) (0.678) (0.692) (5.369)
Observations 344 344 342 198 198 196
Adjusted R2 -0.000 0.068 0.106 -0.004 0.062 0.100

39
 Table 4. Panel Summary Statistics
This table reports summary statistics at the bank-year panel level. The panel sample includes the 38 banks which are
ever stress tested under the CCAR program, as listed in Appendix 2, over 12 years from 2010 to 2021. All continuous
variables are winsorized at the 1st and 99th percentiles. All variables are fully defined in Appendix 1.
Panel A: Labor Demand Obs Mean Std. Dev. p25 p50 p75
Risk Management Jobsi,t 456 9.146 6.151 4.419 7.891 12.695
– High Skilledi,t 456 2.780 2.298 1.121 2.061 3.705
– Low Skilledi,t 456 6.350 4.335 3.144 5.448 8.386
Stress Test Jobsi,t 456 1.201 1.526 0.247 0.624 1.549
– High Skilledi,t 456 0.607 0.713 0.126 0.372 0.803
– Low Skilledi,t 456 0.600 0.980 0.059 0.212 0.717
Other Risk Management Jobsi,t 456 7.884 4.909 4.038 6.794 11.009
– High Skilledi,t 456 2.153 1.809 0.876 1.542 3.024
– Low Skilledi,t 456 5.743 3.713 2.981 5.127 456
Panel B: Stress Test
Testedi,t+1 456 0.711 0.454 0.000 1.000 1.000
Testedi,t 456 0.638 0.481 0.000 1.000 1.000
T1 Cap Exposurei,t 227 2.246 2.441 0.649 1.514 3.432
TRBC Exposurei,t 227 2.246 2.557 0.538 1.529 3.469
T1 Lev Exposurei,t 227 1.146 1.177 0.441 0.991 1.943
CCAR Failurei,t 255 0.133 0.341 0.000 0.000 0.000
CCAR Failure Scorei,t 255 0.180 0.493 0.000 0.000 0.000
Panel C: Risk Measures
Expected Shortfalli,t 441 43.226 7.627 37.550 42.590 47.600
CAPM Betai,t 441 1.127 0.280 0.920 1.090 1.260
Volatilityi,t 441 26.581 9.914 19.200 24.350 31.460
Panel D: Bank Characteristics
LogAssetsi,t-1 410 19.244 1.061 18.568 18.931 19.736
Loan/Assetsi,t-1 410 90.800 53.130 66.765 88.136 103.243
Return-on-Assetsi,t-1 410 0.872 0.778 0.559 0.855 1.145
T1CapRatioi,t-1 410 14.030 4.078 11.591 12.911 15.020
Average Employee Salaryi,t 410 144.072 74.647 98.576 116.708 149.778

40
 Table 5. Stress Tests and Labor Demand for Risk Management Talents
This table shows that the expectation of being stress tested in the next cycle leads to higher demand of stress-test specific
talents, but not other risk management or financial regulation positions. Stress Test Jobs is defined as the subset of risk
management jobs that require skills for “Stress Testing” or “CCAR Regulatory Rules” as a fraction of the bank’s total
job postings. Other Risk Management Jobs measures the share of jobs that require skills for Financial Risk Management
or Financial Regulations but are not Stress Test Jobs. High (Low) Skilled jobs are those (not) requiring higher education
or advanced quant skills. Tested indicates whether the bank is subject to stress testing that year. All variables are fully
defined in Appendix 1. Estimation is linear regression with fixed effects and heteroskedasticity-robust standard errors
clustered at the bank-level (in parenthesis).
Panel A Stress Test Jobsi,t Other Risk Management Jobsi,t
(1) (2) (3) (4) (5) (6) (7) (8)
Testedi,t+1 0.675** 0.279** 0.710** 0.290** 0.005 -0.672 0.275 -0.521
(0.279) (0.133) (0.271) (0.134) (0.459) (0.614) (0.449) (0.584)
Testedi,t -0.097 -0.102 -0.742 -1.358**
(0.192) (0.201) (0.458) (0.567)
log Assetsi,t-1 -0.790** -0.772** 1.323 1.563
(0.350) (0.348) (1.488) (1.427)
Loans/Assetsi,t-1 -0.002 -0.002 -0.017 -0.017
(0.002) (0.002) (0.010) (0.010)
ROAi,t-1 0.013 0.014 -0.534* -0.531*
(0.088) (0.087) (0.271) (0.274)
T1 Cap. Ratioi,t-1 0.070*** 0.074*** 0.211 0.259
(0.018) (0.020) (0.163) (0.157)
Year FE Yes Yes Yes Yes Yes Yes Yes Yes
Bank FE Yes Yes Yes Yes Yes Yes Yes Yes
Observations 456 404 456 404 456 404 456 404
Adjusted R2 0.542 0.683 0.541 0.683 0.633 0.648 0.635 0.653
Panel B Stress Test Jobs – High Skilledi,t Stress Test Jobs – Low Skilledi,t
(1) (2) (3) (4) (5) (6) (7) (8)
*** *** ***
Testedi,t+1 0.363 0.209 0.345 0.200*** 0.273 0.064 0.341 *
0.085
(0.095) (0.072) (0.106) (0.071) (0.207) (0.100) (0.195) (0.105)
Testedi,t 0.048 0.082 -0.185 -0.183
(0.085) (0.072) (0.152) (0.161)
log Assetsi,t-1 -0.093 -0.107 -0.694** -0.661**
(0.142) (0.142) (0.294) (0.276)
Loans/Assetsi,t-1 0.000 0.000 -0.002 -0.002
(0.001) (0.001) (0.002) (0.002)
ROAi,t-1 0.046 0.045 -0.028 -0.027
(0.039) (0.040) (0.050) (0.049)
T1 Cap. Ratioi,t-1 0.022** 0.019 0.048*** 0.054***
(0.011) (0.011) (0.015) (0.016)
Year FE Yes Yes Yes Yes Yes Yes Yes Yes
Bank FE Yes Yes Yes Yes Yes Yes Yes Yes
Observations 456 404 456 404 456 404 456 404
2
Adjusted R 0.518 0.622 0.518 0.623 0.451 0.613 0.454 0.616
*
p < 0.1, ** p < 0.05, *** p < 0.01

41
 Table 6. Exposure to Stress Test and Labor Demand for Risk Management Talents
This table shows that banks increase their demand for stress test jobs in response to failing a stress test. Stress Test
Jobs is defined as the subset of risk management jobs that require skills for “Stress Testing” or “CCAR Regulatory
Rules” as a fraction of the bank’s total job postings. Other Risk Management Jobs measures the share of jobs that
require skills for Financial Risk Management or Financial Regulations but are not Stress Test Jobs. High (Low) Skilled
jobs are those (not) requiring higher education or advanced quant skills. Failure Score is a categorical measure equal
to one or two if the bank received a conditional non-objection or objection, respectively, from the Fed in response to
its planned capital actions under CCAR, and zero otherwise. Exposure is the difference between a ratio’s initial value
and its minimum projected value during the 9-quarter-ahead horizon under the DFAST’s severely adverse stress test
scenario, minus projected dividends. All variables are fully defined in Appendix 1. Estimation is linear regression
with fixed effects and heteroskedasticity-robust standard errors clustered at the bank-level (in parenthesis).
Panel A Stress Test Jobsi,t Other Risk Management Jobsi,t
(1) (2) (3) (4) (5) (6) (7) (8)
Failure Scorei,t-1 0.318** 0.308** 0.297** 0.309** 0.710 0.705 0.663 0.612
(0.128) (0.125) (0.121) (0.121) (0.485) (0.484) (0.481) (0.456)
T1 Cap. Exposurei,t-1 0.107* 0.084 0.159 -0.698
(0.053) (0.197) (0.167) (0.955)
TRBC Exposurei,t-1 0.102** 0.064 0.223 0.964
(0.044) (0.144) (0.178) (0.898)
T1 Lev. Exposurei,t-1 0.027 -0.109 -0.079 -0.321
(0.047) (0.109) (0.279) (0.326)
Year FE Yes Yes Yes Yes Yes Yes Yes Yes
Bank FE Yes Yes Yes Yes Yes Yes Yes Yes
Observations 215 215 215 215 215 215 215 215
Adjusted R2 0.742 0.742 0.729 0.744 0.721 0.723 0.720 0.727
Panel B Stress Test Jobs – High Skilledi,t Stress Test Jobs – Low Skilledi,t
(1) (2) (3) (4) (5) (6) (7) (8)
Failure Scorei,t-1 0.174** 0.171** 0.169** 0.173** 0.144* 0.137** 0.128* 0.136**
(0.083) (0.083) (0.082) (0.083) (0.071) (0.068) (0.067) (0.064)
T1 Cap. Exposurei,t-1 0.034* 0.029 0.073 0.056
(0.019) (0.086) (0.043) (0.124)
TRBC Exposurei,t-1 0.032* 0.012 0.070* 0.052
(0.016) (0.071) (0.037) (0.082)
T1 Lev. Exposurei,t-1 0.020 -0.017 0.007 -0.092
(0.036) (0.049) (0.026) (0.071)
Year FE Yes Yes Yes Yes Yes Yes Yes Yes
Bank FE Yes Yes Yes Yes Yes Yes Yes Yes
Observations 215 215 215 215 215 215 215 215
Adjusted R2 0.587 0.587 0.582 0.582 0.763 0.764 0.745 0.769
*
p < 0.1, ** p < 0.05, *** p < 0.01

42
 Table 7. Risk Management Labor Demand and Risk
This table shows that demand for stress test jobs leads to lower systematic risk. Outcome variables are measured as
of December in year t, while Jobs variables are measured earlier in the year in a window around the stress test
disclosure. Expected Shortfall is the expected fractional loss of equity in a crisis. Beta is with respect to the MSCI
World Index. Stock Volatility is annualized stock return volatility. Stress Test Jobs is defined as the subset of risk
management jobs that require skills for “Stress Testing” or “CCAR Regulatory Rules” as a fraction of the bank’s total
job postings. Other Risk Management Jobs measures the share of jobs that require skills for Financial Risk
Management or Financial Regulations but are not Stress Test Jobs. All variables are fully defined in Appendix 1.
Estimation is linear regression with fixed effects and heteroskedasticity-robust standard errors clustered at the bank-
level (in parenthesis).
Panel A Expected Shortfalli,t Betai,t Volatilityi,t
(1) (2) (3) (4) (5) (6)
Stress Test Jobsi,t -0.407 -0.487** -0.018* -0.022** -0.317 -0.367
(0.241) (0.239) (0.010) (0.010) (0.302) (0.304)
Other Risk 0.104 0.004 0.066
Management Jobsi,t (0.089) (0.004) (0.134)
Testedi,t+1 -0.255 -0.226 -0.011 -0.010 -0.120 -0.101
(0.751) (0.754) (0.029) (0.029) (0.889) (0.875)
Testedi,t 0.395 0.462 0.025 0.028 0.732 0.774
(0.606) (0.619) (0.023) (0.024) (0.723) (0.744)
Year FE Yes Yes Yes Yes Yes Yes
Bank FE Yes Yes Yes Yes Yes Yes
Observations 441 441 441 441 441 441
Adjusted R2 0.701 0.701 0.688 0.689 0.682 0.682
Panel B Expected Shortfalli,t Betai,t Volatilityi,t
(1) (2) (3) (4) (5) (6)
Stress Test Jobsi,t -0.794* -0.940** -0.031* -0.035** -0.271 -0.292
(0.432) (0.421) (0.016) (0.015) (0.490) (0.416)
Other Risk 0.112 0.003 0.016
Management Jobsi,t (0.119) (0.005) (0.148)
Failure Scorei,t 0.959 0.963 0.034 0.035* 1.561** 1.562**
(0.571) (0.572) (0.021) (0.020) (0.659) (0.663)
T1 Cap Exposurei,t -0.355 -0.268 -0.015 -0.013 -0.020 -0.007
(0.597) (0.644) (0.022) (0.024) (0.840) (0.885)
TRBC Exposurei,t -0.279 -0.366 -0.008 -0.010 -0.057 -0.070
(0.487) (0.545) (0.018) (0.020) (0.617) (0.661)
T1 Lev Exposurei,t 0.851** 0.792** 0.030** 0.029** 0.462 0.453
(0.320) (0.324) (0.012) (0.012) (0.401) (0.412)
Year FE Yes Yes Yes Yes Yes Yes
Bank FE Yes Yes Yes Yes Yes Yes
Observations 221 221 221 221 221 221
Adjusted R2 0.745 0.745 0.736 0.736 0.768 0.767
*
p < 0.1, ** p < 0.05, *** p < 0.01

43
 Table 8. Skilled Labor and Risk
This table shows that demand for high-skilled stress test jobs leads to lower systematic risk. Outcome variables are measured as of
December in year t, while Jobs variables are measured earlier in the year in a window around the stress test disclosure. Expected
Shortfall is the expected fractional loss of equity in a crisis. Beta is with respect to the MSCI World Index. Stock Volatility is
annualized stock return volatility. Stress Test Jobs is defined as the subset of risk management jobs that require skills for “Stress
Testing” or “CCAR Regulatory Rules” as a fraction of the bank’s total job postings. Other Risk Management Jobs measures the
share of jobs that require skills for Financial Risk Management or Financial Regulations but are not Stress Test Jobs. High (Low)
Skilled jobs are those (not) requiring higher education or advanced quant skills. All variables are fully defined in Appendix 1.
Estimation is linear regression with fixed effects and heteroskedasticity-robust standard errors clustered at the bank-level (in
parenthesis).
Panel A Expected Shortfalli,t Betai,t Volatilityi,t
(1) (2) (3) (4) (5) (6)
Stress Test Jobs-High Skilledi,t -1.280** -1.396** -0.053** -0.055** -0.777 -0.725
(0.544) (0.521) (0.023) (0.022) (0.669) (0.644)
Stress Test Jobs-Low Skilledi,t 0.150 0.002 -0.067
(0.279) (0.011) (0.355)
Other Risk Management Jobsi,t 0.114 0.111 0.005 0.005 0.065 0.066
(0.088) (0.088) (0.004) (0.004) (0.134) (0.135)
Testedi,t+1 -0.124 -0.134 -0.007 -0.007 -0.091 -0.086
(0.735) (0.737) (0.027) (0.028) (0.858) (0.864)
Testedi,t 0.578 0.609 0.033 0.033 0.846 0.832
(0.636) (0.623) (0.025) (0.024) (0.780) (0.763)
Year FE Yes Yes Yes Yes Yes Yes
Bank FE Yes Yes Yes Yes Yes Yes
Observations 441 441 441 441 441 441
Adjusted R2 0.704 0.703 0.691 0.690 0.682 0.681
Panel B Expected Shortfalli,t Betai,t Volatilityi,t
(1) (2) (3) (4) (5) (6)
** ** ** **
Stress Test Jobs-High Skilledi,t -2.079 -2.166 -0.074 -0.078 -0.326 -0.108
(0.825) (0.817) (0.030) (0.029) (1.157) (1.333)
Stress Test Jobs-Low Skilledi,t 0.206 0.008 -0.519
(0.529) (0.019) (0.740)
Other Risk Management Jobsi,t 0.151 0.147 0.004 0.004 0.074 0.084
(0.134) (0.136) (0.005) (0.005) (0.150) (0.151)
Failure Scorei,t 0.688 0.662 0.026 0.025 1.421** 1.486**
(0.603) (0.620) (0.021) (0.022) (0.636) (0.700)
T1 Cap Exposurei,t -0.356 -0.340 -0.013 -0.013 -0.080 -0.121
(0.698) (0.689) (0.025) (0.025) (1.045) (1.014)
TRBC Exposurei,t -0.484 -0.513 -0.017 -0.018 0.122 0.196
(0.529) (0.504) (0.018) (0.017) (0.715) (0.704)
T1 Lev Exposurei,t 1.092** 1.120** 0.042** 0.043** 0.294 0.224
(0.434) (0.481) (0.016) (0.018) (0.565) (0.548)
Year FE Yes Yes Yes Yes Yes Yes
Bank FE Yes Yes Yes Yes Yes Yes
Observations 193 193 193 193 193 193
Adjusted R2 0.741 0.739 0.721 0.719 0.774 0.773
*
p < 0.1, ** p < 0.05, *** p < 0.01

44
 Table 9. Labor Demand for Risk Management Talents and Bank Profits
This table shows that banks experience lower profitability and pay higher employee salaries following demand
for high skilled stress test jobs. ROA is net income divided by lagged total assets. Salaries is total salary expense
divided by the number of employees. Stress Test Jobs is defined as the subset of risk management jobs that
require skills for “Stress Testing” or “CCAR Regulatory Rules” as a fraction of the bank’s total job postings.
Other Risk Management Jobs measures the share of jobs that require skills for Financial Risk Management or
Financial Regulations but are not Stress Test Jobs. All variables are fully defined in Appendix 1. Estimation is
linear regression with fixed effects and heteroskedasticity-robust standard errors clustered at the bank-level (in
parenthesis).
ROAi,t+1 Salariesi,+1
(1) (2) (3) (4)
Stress Test Jobs-High Skilledi,t -0.326** -0.264** 5.356** 7.326**
(0.129) (0.115) (2.291) (3.016)
Stress Test Jobs-Low Skilledi,t -0.171** -5.393
(0.064) (3.513)
Other Risk Management Jobsi,t 0.0213 0.0247 -0.182 -0.0751
(0.019) (0.017) (0.349) (0.344)
Testedi,t+1 0.0581 0.0690 -0.948 -0.605
(0.114) (0.115) (4.262) (4.394)
Testedi,t 0.142 0.109 4.201 3.175
(0.112) (0.109) (4.008) (4.354)
log Assetsi,t-1 -0.430 -0.564** 15.17 10.95
(0.255) (0.247) (9.602) (11.447)
Loan/Assetsi,t-1 0.256 0.217 -16.45 -17.67
(0.540) (0.526) (26.019) (24.794)
T1 Cap. Ratioi,t-1 -0.0280* -0.0219 2.546*** 2.736***
(0.015) (0.016) (0.574) (0.655)
Year FE Yes Yes Yes Yes
Bank FE Yes Yes Yes Yes
Observations 331 331 331 331
Adjusted R2 0.610 0.622 0.944 0.945
*
p < 0.1, ** p < 0.05, *** p < 0.011

45