## _wp15271 - References

## Source details

**Canonical URL:** [_wp15271 - References](https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2015/_wp15271.pdf)

## Other formats

- [Markdown version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2015/_wp15271.pdf.md)
- [Structured JSON version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2015/_wp15271.pdf.json)

---

### I. Introduction: context, questions, approach
- Context and purpose:
  - After the 2008–09 financial crisis policymakers institutionalized “stress tests.”
  - Paper examines capital market implications of stress tests in the United States since the onset of the global financial crisis.
- Key research questions:
  - Do and how do capital markets react to information revealed by stress tests?
  - How does public disclosure of stress tests affect information generation and processing in capital markets?
- Empirical approach:
  - Analyze equity and bond price changes and jumps, equity and credit bid-ask spreads, implied volatilities, and CDS spreads.
  - Use difference-in-difference event-study setup around stress test announcements and results disclosure.
- Principal summary findings:
  - Important information in stress tests, especially at times of stress.
  - Markets react to stress test announcements; direction depends on nature of news and pass/fail outcomes.
  - Higher moments of return distributions and trading activity affected; spillovers to untested banks suggest revelation about systemic risk.
  - Reactions have weakened as stress tests became more established and announcement dates more predictable, but scenarios and supervisory assessments still convey year-to-year information.
  - Some indication information asymmetry increases at announcement and declines after results release.
  - Information uncertainty not significantly affected overall; markets continue producing private information.
  - Public disclosure helps reduce informational asymmetries and uncertainties in periods of distress and does not appear to have reduced private incentives to generate information.
- Policy implication highlighted:
  - Macro stress tests are more effective as crisis management and resolution tools than as early warning devices.

### II. Background: U.S. supervisory stress testing framework
- SCAP (2009):
  - Began 2009; SCAP announced February 10, 2009; detailed design and methodology made available on April 24, 2009.
  - Two macroeconomic scenarios: baseline (consensus in February 2009) and an adverse scenario (longer and more severe recession).
  - 19 bank holding companies subject to tests; assets of at least $100 billion as of end-2008; constituted two-thirds of the system by assets and more than one-half by loans.
  - Results publicly released May 7, 2009 (at 5 pm Eastern Standard Time).
- CCAR (Comprehensive Capital Assessment Review):
  - Initiated late 2010; CCAR 2011 covered same institutions as SCAP.
  - By 2014 all 30 bank holding companies with total consolidated assets of $50 billion or more included.
  - CCAR includes quantitative capital assessment and qualitative assessment of internal capital planning.
  - Banks submit stress test results under Fed-specified and internal scenarios plus a capital plan covering a nine-quarter horizon.
  - Fed may object to capital plans; objections limit capital distributions.
- DFAST (Dodd-Frank Act Stress Tests):
  - Dodd-Frank Act (2010) requires annual supervisory stress tests for bank holding companies with assets greater than $50 billion under three scenarios: baseline, adverse, and severely adverse; requires public disclosure of results.
  - Federally-regulated financial companies with $10 billion or more in total consolidated assets must annually conduct and publicly disclose internal stress tests (DFAST).
  - Supervisory DFAST uses confidential regulatory report data and standardized capital action assumptions.
  - Since 2013, DFAST and CCAR run in parallel; they share macro scenarios but differ in capital distribution assumptions (regulatory generic assumptions in DFAST vs. bank-submitted capital plans in CCAR).
- Administrative/timing notes:
  - Company-specific CCAR 2011 results were not disclosed; later years company-specific results publicly available, with release dates falling in March.
  - Releases often at or after market closing (exception: CCAR 2011 results released at 11 am Eastern Standard Time).
  - DFAST results typically released about a week before parallel CCAR results.
  - Design and scenario information released in the fall preceding the exercise; some supervisor model information disclosed but exact specifications and parameter estimates not revealed.
- Criticisms:
  - Partial disclosure criticized; concerns about “model convergence” and risk shifting.
  - Public disclosure of qualitative assessments may shift focus away from quantitative models given model uncertainty.
  - Arguments that bank-by-bank disclosure may reduce incentives for private information production and market informativeness.
  - Concerns about accounting gimmicks and distortions to avoid failing tests.

### III. Literature and theoretical trade-offs
- Empirical studies summarized:
  - Morgan, Peristiani, and Savino (2014): SCAP 2009 — abnormal returns lower for banks with bigger surprise capital shortfalls.
  - Glasserman and Tangirala (2015): Loss projections highly correlated across DFAST 2013 and 2014; projections little related to abnormal returns at disclosure.
  - Petrella and Resti (2013), Ellahie (2013), Candelon and Sy (2015), Flannery et al. (2015) — heterogeneous cross-jurisdictional and time-varying effects; evidence that stress tests can reduce opaqueness and affect returns, spreads, and trading activity.
- Theoretical considerations:
  - Disclosure trade-off: prevent market breakdown vs. destroy risk-sharing opportunities (Goldstein and Leitner, 2013).
  - Mandatory disclosure can increase welfare when contagion risk high (Alvarez and Barlevy, 2014).
  - Disclosure may interfere with risk sharing, price efficiency, and private information production (Goldstein and Sapra, 2014).
  - Regulator disclosure strategy may depend on recapitalization capacity (Spargoli, 2012).
  - Importance of credibility, governance, and backstops emphasized (Ong and Pazarbasioglu, 2013).

### IV. Empirical predictions, indicators, and heterogeneity
- Empirical focus:
  - Provide comprehensive overview of market reactions to stress tests and how disclosure affects private information production and market functioning.
- Indicators used:
  - Price reactions, higher moments of return distributions, equity/bond price changes and jumps, bid-ask spreads, implied volatilities, CDS spreads, trading activity.
- Hypotheses:
  - Public disclosure may reduce information asymmetry and uncertainty, particularly during stressed periods.
  - Disclosure could diminish private incentives to produce information, reducing price informativeness and risk sharing.
  - Optimal disclosure may vary with economic conditions; partial disclosure possibly preferable during bad times.
- Heterogeneity inquiry:
  - Are information gains greater for banks with characteristics: large, leveraged, opaque, complex, lower-quality disclosure?

### V. Data, sample, and measurement
- Data sources:
  - Equity prices: CRSP and Datastream.
  - Bond data: Bloomberg (publicly traded, nonconvertible, noncallable, on-the-run).
  - Implied volatility and CDS: Datastream.
  - Financial statements: Call Reports (via SNL).
  - Realized variance and jump measures: provided by Asger Lunde.
- Sample:
  - Largest 100 bank holding companies in the United States as of the last quarter of 2014, based on assets.
- Event window and robustness:
  - Event window: seven days total (3 days before and 3 days after event date).
  - Robust to a shorter 3-day window and to propensity-matched control groups.
- Variable definitions (selected):
  - Tier 1 Capital / RWA: percent (Basel III if available, Basel I otherwise).
  - RWA / Total assets (percent); ROAA (percent).
  - Opacity: book value of bank premises and investments in unconsolidated subsidiaries, intangible assets, and “other assets” divided by total assets.
  - Complexity: count of subsidiaries (log number reported in summary statistics).
  - Earnings management: difference between discretionary realized security gains and losses and discretionary loan loss provisions as a percent of total assets.
  - Realized volatility: daily volatility computed by bipower variation at 5-minute frequency.
  - Jump component: realized variance minus bipower variation.
  - Implied volatility: at-the-money equity call option implied volatility in standard deviations.
  - CDS spread: ratio of one-year to five-year CDS spread.
  - Spread (equity/bond bid-ask): difference between ask and bid divided by half of the difference between ask and bid.
  - Lending growth: change in total loans net of earned income (BHCK2122).

### VI. Summary statistics (Table 3, All banks)
- Assets (billion dollars): Mean 187.04; St. Dev. 54.06
- Market cap (billion dollars): Mean 16.83; St. Dev. 7.45
- Tier 1 Capital / RWA (percent): Mean 12.93; St. Dev. 3.21
- RWA / Total assets (percent): Mean 72.41; St. Dev. 15.07
- ROAA (percent): Mean 0.71; St. Dev. 1.80
- Opacity (percent): Mean 7.14; St. Dev. 4.38
- Complexity (log number): Mean 2.41; St. Dev. 1.94
- Earnings management (percent): Mean -0.01; St. Dev. 0.50

### VII. Event-study results — Announcement (selected figures, Table 4)
- All events:
  - CAR — Tested -0.14; Untested -0.08 (Difference not significant).
  - Absolute value of CAR — Tested 3.30; Untested 2.54 (Difference: Yes).
  - Abnormal volume — Tested 0.89*; Untested -0.07 (Difference: Yes).
- SCAP 2009:
  - CAR — Tested 3.95; Untested 0.53 (No).
  - Absolute value of CAR — Tested 9.47; Untested 3.66 (Yes).
  - Abnormal volume — Tested 3.80*; Untested 1.07* (Yes).
- CCAR 2011:
  - CAR — Tested -4.11*; Untested -1.36* (Yes).
  - Absolute value of CAR — Tested 4.50; Untested 2.08 (Yes).
  - Abnormal volume — Tested 0.13; Untested -1.37* (Yes).
- Note: Abnormal activities computed over the 30 trading days prior to the event. * denotes statistical significance at conventional levels (10 percent or lower).

### VIII. Event-study results — Results release (selected figures, Table 5)
- All events:
  - CAR — Tested 1.56; Untested 1.51 (No).
  - Absolute value of CAR — Tested 3.77; Untested 3.28 (Yes).
  - Abnormal volume — Tested 0.34*; Untested 0.09 (Yes).
- CCAR 2012:
  - CAR — Tested 4.35*; Untested 4.92* (No).
  - Absolute value of CAR — Tested 4.46; Untested 5.02 (No).
  - Abnormal volume — Tested 1.46*; Untested 0.65* (Yes).
- DFAST 2015:
  - CAR — Tested 4.84*; Untested 4.23* (Yes).
  - Absolute value of CAR — Tested 4.84; Untested 4.40 (No).

### IX. Information indicators — Announcement and Results Release (Tables 6 and 7, selected estimates)
- Around Announcement (selected coefficients, difference-in-difference):
  - All events: Realized volatility 0.001890 (not starred).
  - All events: Jump component 0.142***.
  - Event & Tested coefficients negative for jump component: Event & Tested -0.325***; Event & Tested -0.154***.
  - SCAP 2009: realized volatility coefficients include -0.0738** and 0.223.
- Around Results Release (selected coefficients):
  - Realized volatility: All events 1.770***; Tested 0.857; Event & Tested 3.530***.
  - Bid-ask spread, equity: All events 0.545***; Tested -0.285**; Event & Tested -0.372*.
  - CDS spread, 1-year / 5-year: All events 0.000863; Tested -0.0270; Event & Tested -0.00683.
  - SCAP 2009 pronounced: Realized volatility 7.683***, Jump component 25.96***.
- Significance notation: ***, **, and * denote 1, 5, and 10 percent levels, respectively.

### X. Passed versus Failed (Table 8, selected figures)
- All events:
  - Absolute value of CAR — Passed 3.26; Failed 7.42 (Difference significant: Yes).
  - Abnormal volume — Passed 0.19; Failed 1.47* (Yes).
- SCAP 2009:
  - Absolute value of CAR — Passed 8.82; Failed 15.74 (Yes).
  - Abnormal volume — Passed 0.74; Failed 2.11* (No).
- Regression lower-panel reports significant coefficients (e.g., All events regression: 1.852*** and 1.31115.09*** as reported).

### XI. Cross-sectional heterogeneity and market-reaction regressions (Tables 9a–9b)
- Absolute CAR and Abnormal volume regressions (selected coefficients):
  - Stress tested dummy coefficients reported across columns: 0.641*; 1.992; -0.0560; -0.125; 1.289**; 0.923.
  - Tier 1 Capital / RWA coefficients: 0.0405; 0.0884.
  - Market cap / Total assets interactions: -7.503; 5.345; Stress tested 1.441; -6.112*.
  - Complexity interaction: -0.0977; 0.157*.
  - Auditor (Big 4) interactions reported.
- Spread, equity and Implied volatility regressions (selected coefficients):
  - Stress tested coefficients: -0.222; 0.444; -0.284; -1.795**; -0.414.
  - Tier 1 Capital / RWA interactions: 0.0152.
  - RWA / Total assets: -0.0172**; 0.000242.
- Regression notes:
  - Sample = largest 100 bank holding companies as of 2014Q4.
  - Regressions include year fixed effects.
  - Robust standard errors clustered at bank holding company level.
  - ***, **, and * denote significance at 1, 5, and 10 percent levels.

### XII. Post-stress-test behavior (Table 10, selected OLS results)
- Outcomes analyzed: quarterly changes and cumulative changes two quarters following a stress test; in-between stress test news.
- Selected coefficients (OLS, year fixed effects, control for size measured by log assets):
  - Tested coefficients across outcomes: -0.0139; -0.0103; -0.00824; -0.0173; -0.00431; -0.00280.
  - Failed coefficients: -0.0201*; 0.0444**; -0.00852.
  - Additional tested and failed coefficients for other outcomes include -0.0446, -0.0450.
- Empirical patterns:
  - Failed banks tend to become less complex (number of subsidiaries decreases).
  - Failed banks show changes in earnings management: reduce earnings management cumulatively in the two quarters following the stress test but a positive and statistically significant cumulative change over the full period between stress tests.
  - Failed banks exhibit slower lending growth in the two quarters following the stress test; longer-horizon lending results not statistically significant.
  - The tested dummy not significant for longer-term behavior, suggesting post-failure adjustments plausibly reflect supervisory remediation rather than testing per se.

### XIII. Interpretation, caveats, and conclusions
- Interpretation:
  - Stress tests contain new information for markets, particularly during market turmoil (SCAP 2009 most notable).
  - Public disclosure tends to increase information asymmetry at announcement (investors acquire private information) and reduce it upon results release if information is useful and credible.
  - Disclosure reduces information uncertainty following results release in distressed periods.
  - No strong evidence that disclosure reduced private incentives to produce information.
- Potential distortions and concerns:
  - Banks might adopt complex structures or manipulate accounting to pass tests.
  - Banks may cut lending to meet test requirements; potential growth implications.
- Caveats:
  - Treatment was not random; tests covered the largest banks; sample selection could bias comparisons.
  - Sample period does not span a full credit cycle; early tests occurred in high distress, later tests in calmer environments.
  - Results are suggestive correlations rather than definitive causal links.
- Conclusions and suggested future research:
  - There is important information in stress tests, with stronger market reactions for tested banks and directional dependence on pass/fail outcomes.
  - Stress tests associated with increased trading volumes, changes in volatility measures, and reductions in information asymmetry and uncertainty in stressed environments.
  - Some post-test behavioral changes occur for failed banks (reduced complexity, altered earnings management, possible short-term lending slowdown), plausibly reflecting supervisory remediation.
  - Future research should examine stress test disclosures across a full credit cycle, especially during the upward phase, to assess effectiveness near credit peaks.

*Content based exclusively on the PDF chapter/section titled "References" (file: _wp15271 - References).*

### References .............................................................................................................

### _wp15271 - References .............................................................................................................

### I. INTRODUCTION
- Context and purpose:
  - After the 2008–09 financial crisis, policymakers and regulators sought to enhance supervisory frameworks to prevent recurrence, including institutionalizing “stress tests.”
  - The paper examines capital market implications of stress tests in the United States since the onset of the global financial crisis.
- Key research questions and approach:
  - Do and how do capital markets react to information revealed by stress tests?
  - How does public disclosure of stress tests affect information generation and processing in capital markets?
  - Empirical approach: analyze equity and bond price changes and jumps, equity and credit bid-ask spreads, implied volatilities, and CDS spreads in a difference-in-difference event-study setup around stress test announcements and results disclosure.
- Principal summary findings:
  - There is important information in stress tests, especially at times of stress.
  - Markets react to stress test announcements; the direction depends on the nature of the news (e.g., whether scenarios depict more or less stress than markets expect, or whether a bank failed or passed quantitative thresholds).
  - Higher moments of return distributions and trading activity are affected; reactions spill over to banks not subject to tests, suggesting revelation about systemic risk.
  - Reactions have weakened as stress tests became more established and announcement dates became more predictable, but scenarios and supervisory assessments still convey information year-to-year.
  - Some indication that information asymmetry increases with announcements but declines after release of results.
  - Information uncertainty is not significantly affected; markets appear to continue producing private information rather than relying solely on public supervisory disclosures.
  - Public disclosure helps reduce informational asymmetries and uncertainties, especially during market distress, and does not appear to have reduced private incentives to generate information.
- Policy implication highlighted:
  - Findings align with view that macro stress tests are more effective as crisis management and resolution tools than as early warning devices.

### II. BACKGROUND
- Scope:
  - Description of supervisory stress tests in the United States with emphasis on public disclosure of design and results.

#### A. A Brief History of Stress Testing in the United States
- Supervisory Capital Assessment Program (SCAP):
  - Beginnings of current framework in 2009 with SCAP.
  - SCAP aimed to quantify impact on capital of further deterioration in financial markets and the economy (Bernanke, 2009).
  - 19 bank holding companies subject to tests had assets of at least $100 billion as of end-2008 and constituted two-thirds of the system by assets and more than one-half by loans.
  - SCAP announced on February 10, 2009; detailed design and methodology made available on April 24, 2009.
  - Two macroeconomic scenarios: (a) a baseline reflecting the consensus in February 2009; and (b) an adverse scenario characterizing a recession longer and more severe than consensus expectation.
  - Banks projected credit losses and revenues for 2009 and 2010 under these scenarios; supervisory teams evaluated submissions and senior supervisory officials determined necessary capital buffers.
  - Results publicly released on May 7, 2009 (at 5 pm Eastern Standard Time).
- Comprehensive Capital Assessment Review (CCAR):
  - Initiated in late 2010 with first round of results in 2011; CCAR 2011 covered same institutions as SCAP.
  - By 2014, all 30 bank holding companies with total consolidated assets of $50 billion or more became part of CCAR.
  - CCAR comprises quantitative assessment of capital positions and qualitative assessment of internal capital planning processes.
  - Each bank submits stress test results under Federal Reserve-specified scenarios and internal scenarios, plus a capital plan covering a nine-quarter horizon.
  - Federal Reserve assesses plans and may object or not object to planned capital distributions; objection limits capital distributions.
- Dodd-Frank Act Stress Tests (DFAST):
  - Dodd-Frank Act passed in 2010 required annual supervisory stress tests for bank holding companies with assets greater than $50 billion (and certain SIFIs) under three scenarios: baseline, adverse, and severely adverse; requires public disclosure of results.
  - Federally-regulated financial companies with $10 billion or more in total consolidated assets must annually conduct and publicly disclose internal stress tests (DFAST).
  - Supervisory DFAST uses confidential regulatory report data as inputs into Federal Reserve models to project pre-tax net income under specified scenarios; equity capital and regulatory capital calculated using standardized capital action assumptions.
  - Since 2013, DFAST and CCAR run in parallel; DFAST projected losses used as inputs to CCAR capital planning. They share macro scenarios but differ in capital distribution assumptions (regulatory generic assumptions in DFAST vs. bank-submitted capital plans in CCAR).

- Administrative and timing notes:
  - Table references (not reproduced here) provide timelines and lists of participating bank holding companies.
  - Company-specific CCAR 2011 results were not disclosed, though some firms independently disclosed whether the Federal Reserve objected to their capital plans.
  - In later years, company-specific results publicly available, with release dates falling in March.
  - Releases are often at or after market closing (exception: CCAR 2011 results released at 11 am Eastern Standard Time).
  - DFAST results typically released about a week before parallel CCAR results.
  - Information on design and scenarios released in the fall preceding the exercise.
  - Some information on supervisor models disclosed but exact specifications and parameter estimates are not revealed.

- Criticisms and concerns:
  - Partial disclosure of model information criticized; concerns about “model convergence” and risk shifting have prevented full disclosure.
  - Public disclosure of qualitative assessments may shift focus away from quantitative models over time given model uncertainty.
  - Arguments that bank-by-bank disclosure may reduce incentives for private information production and market informativeness.
  - Concerns about accounting gimmicks and distortions in bank activities to avoid “failing” stress tests.

#### B. Literature Review
- Overview of empirical findings:
  - Morgan, Peristiani, and Savino (2014): For SCAP 2009 banks, abnormal returns were lower for banks with a bigger surprise capital shortfall (discrepancy between stress test-calculated capital gap and ex ante expectations).
  - Glasserman and Tangirala (2015): Loss projections are highly correlated across DFAST 2013 and 2014; current year’s projections highly correlated with previous year’s; projected loss rates have little relation to abnormal returns at disclosure, suggesting markets may correctly predict results.
  - Petrella and Resti (2013) (EBA 2011): Stress-tested banks with less damage in stress scenarios had cumulative abnormal returns higher than untested banks and other tested banks.
  - Ellahie (2013) (EBA 2011): Using bond, equity, and CDS spreads, concluded EBA 2011 reduced information asymmetry and revealed directional capital-strength information at expense of increasing information uncertainty.
  - Candelon and Sy (2015): Show cumulative abnormal returns vary through time and across jurisdictions; U.S. impact declined over time (large positive impact for 2009 SCAP, smaller later), whereas 2011 EBA stress test had negative significant impact on both tested and untested banks; interpretation emphasizes role of exercise design and governance.
  - Flannery et al. (2015): Stress-tested BHCs experience significant price and activity reactions, more prominent for more levered and riskier firms; found no evidence of reduced information production (proxied by analyst coverage and forecast accuracy), no decline in risk sharing (proxied by interbank borrowing), nor increased vulnerability to runs (proxied by differential response for less liquid firms).
- Theoretical literature and trade-offs:
  - Goldstein and Leitner (2013): Regulator trade-off between preventing market breakdown via disclosure and destroying risk-sharing opportunities; recommend no disclosure in normal times, partial disclosure in bad times.
  - Alvarez and Barlevy (2014): Mandatory disclosure can increase welfare when contagion risk is high.
  - Goldstein and Sapra (2014): Stress tests uncover unique information but disclosure may interfere with risk sharing, price efficiency, market discipline, and private information production, with implications for optimal bank behavior and system vulnerability.
  - Spargoli (2012) model noted: regulator may fully reveal banks’ capital shortfalls during crises if able to recapitalize, but withhold information if unable to recapitalize.
  - Ong and Pazarbasioglu (2013): Emphasize credibility through independent governance, technical expertise, and clear plans for backstops.

#### C. EMPIRICAL PREDICTIONS
- Empirical focus and indicators:
  - The paper seeks to provide a comprehensive overview of market reactions to stress tests and how disclosure affects private information production and market functioning.
  - Indicators include price reactions, higher moments of return distributions, and a range of market functioning indicators (examples earlier include equity/bond price changes and jumps, bid-ask spreads, implied volatilities, CDS spreads, and trading activity).
- Hypotheses (context from theoretical literature):
  - Public supervisory disclosure may reduce information asymmetry and uncertainty, particularly during stressed periods.
  - Disclosure could diminish private incentives to produce information, possibly reducing price informativeness and risk sharing.
  - Optimal disclosure policy may vary with economic conditions, with partial disclosure potentially preferable during bad times.

*Italicized source attribution line.*

### 1. Is there new information in stress tests? What is the reaction in price, volatility, and

### _wp15271 - 1. Is there new information in stress tests? What is the reaction in price, volatility, and

### Main research questions
- 1. Is there new information in stress tests? What is the reaction in price, volatility, and trading when the market learns about the scenarios and results?
- 1. Does this differ for “winners” and “losers”?
- 1. Is this limited to stress-tested companies?
- 1. How are information asymmetry and uncertainty in the market affected by public disclosure?

### Follow-up inquiry on heterogeneity
- 2. Are information gains greater for companies with certain characteristics (e.g., large, leveraged, opaque, complex, lower-quality disclosure)?

*Source: _wp15271 - 1. Is there new information in stress tests? What is the reaction in price, volatility, and*

### 3.   Is there a distortion of incentives? Do companies engage in more risk taking or more

### 3.   Is there a distortion of incentives? Do companies engage in more risk taking or more earnings management?

### Information content of stress tests and market reaction
- Stress tests condition on tail-event quantiles, which conveys information about the distribution and hence about the mean of future cash flows.
- A failed stress test can lead to supervisory objections to capital redistribution (stock buybacks, dividend increases), producing negative price effects.
- Examples cited:
  - Following CCAR 2012, JP Morgan Chase announced a plan to buy back as much as $15 billion of its stock and to raise its quarterly dividend from 25 to 30 cents per share.
  - CCAR 2011 reactions included a 20-cent increase in dividend per share for JP Morgan; Citigroup announced waiting until the next year for any dividend increase.
- Announcements of supervisory scenarios can also affect market behavior and capital redistribution plans (example: Bloomberg, October 23, 2014, reporting that scenarios “will limit the ability of banks to get aggressive in returning capital to shareholders”).
- Information about tested firms can transmit to untested firms via counterparty links or overall system stress.

### Empirical strategy and measures
- Event-study analysis distinguishing tested vs. untested banks and pass vs. fail at results release; examine raw cumulative abnormal returns (CAR) and absolute CAR.
- Use of high-frequency measures:
  - Daily volatility and jump components computed from 5-minute returns.
  - Daily volatility estimated by bipower variation; jump component estimated as realized variance minus bipower variation.
- Abnormal volume measure: actual trading volume minus predicted volume (based on relationship between market volume and the bank stock’s volume over past three months).
- Information-uncertainty measures: equity option implied volatilities and the ratio of one-year to five-year CDS spreads.
- Cross-sectional heterogeneity examined using bank characteristics:
  - Leverage: Tier 1 capital divided by risk-weighted assets; market capitalization divided by total assets.
  - Riskiness: risk-weighted assets divided by total assets.
  - Asset opacity: book value of bank premises and investments in unconsolidated subsidiaries, intangible assets, and “other assets” divided by total assets.
  - Complexity: count of bank and nonbank subsidiaries.
  - Earnings management: difference between discretionary realized security gains and losses and discretionary loan loss provisions as a percent of total assets.
  - Auditor identity (Big 4 vs. others).

### Data and sample
- Equity prices: CRSP and Datastream.
- Bond data: Bloomberg (publicly traded, nonconvertible, noncallable, on-the-run).
- Implied volatility and CDS: Datastream.
- Financial statements: Call Reports (via SNL).
- Realized measures for variance and jumps: provided by Asger Lunde.
- Sample: largest 100 bank holding companies in the United States as of the last quarter of 2014, based on assets.
- Event window: total of seven days (3 days before and 3 days after the event date).
- Robustness: results robust to a shorter 3-day window and to propensity-matched control groups.

### Main empirical findings
- Price effects
  - Conventional event studies show small, often positive reactions in CAR around announcements and results releases when pooling events, but pooling masks heterogeneity.
  - Directional response at results release:
    - Banks that pass the stress test: large and positive returns.
    - Banks that fail the stress test: large and negative returns.
  - Absolute CAR analysis indicates valuable information in stress test announcements.
  - Strongest price reaction observed for SCAP 2009; reactions smaller in later exercises but not monotonically diminishing.
  - Significant price reactions also observed for untested banks, suggesting spillovers.
- Trading and liquidity
  - Trading volumes increase around announcements and release dates (consistent with new information digestion).
  - Equity bid-ask spreads and implied volatility tend to increase at announcements and decrease at results release for tested banks, effects stronger in earlier tests.
- Volatility and jumps
  - Realized volatility increases significantly for tested firms around announcements and results releases (computed using realized kernel on 5-minute data).
  - Jump component generally not statistically significant except for SCAP 2009.
- Credit market indicators
  - Bond bid-ask spreads and CDS spread ratio coefficients are not consistently signed and often not statistically significant; limited data availability is a caveat.
- Cross-sectional heterogeneity
  - Disclosure appears more informative for riskier, more opaque, and more complex banks, though robust conclusions are limited.
  - Auditor identity matters: absolute CAR larger for firms audited by three of the Big 4; information asymmetry increases for tested firms audited by one of the Big 4 after results release.
- Behavior after stress tests
  - Failed banks tend to:
    - Become less complex (number of subsidiaries decreases).
    - Engage in more earnings management in the longer term; evidence shows failed banks reduce earnings management cumulatively in the two quarters following the stress test but positive and statistically significant cumulative change over the full period between stress tests.
    - Exhibit slower lending growth in the two quarters following the stress test, though longer-horizon results are not statistically significant.
  - The tested dummy is not significant for longer-term behavior, suggesting observed post-failure adjustments may reflect supervisory actions to restore a failed bank rather than stress testing per se.

### Interpretation and implications
- Stress tests contain new information for markets, especially during periods of heightened distress (SCAP 2009 most notable).
- Public disclosure tends to:
  - Increase information asymmetry at announcement (investors acquire private information), then reduce it upon results release if information is useful and credible.
  - Reduce information uncertainty following results release in distressed periods.
- Stress tests do not appear to have reduced private incentives to produce information (no evidence that disclosure eliminated private information production).
- Potential distortions and concerns:
  - Opponents: banks might adopt complex structures or manipulate accounting to pass tests.
  - Proponents: worry that banks may cut lending to meet test requirements, with potential growth implications.
- Cautionary notes:
  - Treatment was not random; tests covered the largest banks and sample selection could bias comparisons.
  - Sample period does not span a full credit cycle; earlier tests occurred in high distress, later tests in calmer environments.
  - Results should be interpreted as suggestive correlations rather than definitive causal links.

### Conclusions
- There is important information in stress tests, particularly during market turmoil.
- Market reactions are stronger for tested banks relative to untested banks, with direction depending on pass/fail outcomes.
- Stress tests are associated with increased trading volumes, changes in volatility measures, and reductions in information asymmetry and uncertainty in stressed environments.
- Some post-test behavioral changes occur for failed banks (reduced complexity, altered earnings management, possible short-term lending slowdown), plausibly reflecting supervisory remediation rather than the testing mechanism itself.
- Future research should examine stress test disclosures across a full credit cycle, especially during the upward phase, to assess effectiveness near credit peaks.

*Source: IMF working paper section titled "3.   Is there a distortion of incentives? Do companies engage in more risk taking or more earnings management?"*

### References

### _wp15271 - References

### Key bibliographic sources
- Alvarez, F. and G. Barlevy, 2014, “Mandatory disclosure and financial contagion,” Federal Reserve Bank of Chicago WP 2014–04.
- Barndorff-Nielsen, O. and N. Shephard, 2004, “Power and bipower variation in stochastic volatility and jumps,” Journal of Financial Econometrics, Volume 2 (1), pp. 1–37.
- Bernanke, B. S., 2009, “Statement regarding the Supervisory Capital Assessment Program,” May 7.
- Bookstaber, R. J., Cetina, G., Feldberg, M., Flood, and P. Glaesserman, 2013, “Stress tests to promote financial stability: Assessing progress and looking to the future,” Office of Financial Research Working Paper 0010.
- Borio, C., M. Drehmann, and K. Tsatsaronis, 2013, “Stress testing macro stress testing: Does it live up to expectations?” Journal of Financial Stability, Volume 12 (June), pp. 2–15.
- Candelon, B. and A. N. R. Sy, 2015, “How did markets react to stress tests?” IMF Working Paper WP/15/75.
- Cornett, M. M., J. J. McNutt, and H. Tehranian, 2009, “Corporate governance and earnings management at large U.S. bank holding companies,” Journal of Corporate Finance, Volume 15, pp. 412–430.
- Corrado, C., 1989, “A non parametric test for abnormal security-price performance in event studies,” Journal of Financial Economics, Volume 23, pp. 385–395.
- Dowd, K., 2015, “Central bank stress tests: Mad, bad, and dangerous,” Cato Journal, Volume 35 (3), pp. 507–524.
- Ellahie, A., 2012, “Capital market consequences of EU bank stress tests,” manuscript, London School of Economics.
- Flannery, M. J., S. H. Kwan, and M. Nimalendran, 2004, “Market evidence on the opaqueness of banking firms’ assets,” Journal of Financial Economics, Volume 71(3), pp. 419–460.
- Flannery, M. J., B. Hirtle, and A. Kovner, 2015, “Evaluating the information in the Federal Reserve stress tests,” Federal Reserve Bank of New York Staff Reports No. 744.
- Glasserman, P. and G. Tangirala, 2015, “Are the Federal Reserve’s stress test results predictable?” Office of Financial Research Working Paper 15–02.
- Goldstein, I. and Y. Leitner, 2013, “Stress tests and information disclosure,” Federal Reserve Bank of Philadelphia Working Paper 13–26.
- Goldstein, I. and H. Sapra, 2014, “Should banks' stress test results be disclosed? An analysis of the costs and benefits,” manuscript, The Wharton School.
- Hirtle, B. and A. Lehnert, 2014, “Supervisory stress tests,” Federal Reserve Bank of New York Staff Reports, no. 696.
- Morgan, D. P., S. Peristiani, and V. Savino, 2014, “The information value of the stress test,” Journal of Money, Credit, and Banking, Volume 46, pp. 1479–1500.
- Ong, L. and C. Pazarbasioglu, 2013, “Credibility and crisis stress testing,” IMF Working Paper WP/13/178.
- Petrella, G. and A. Resti, 2013, “Supervisors as information producers: Do stress tests reduce bank opaqueness?” Journal of Banking and Finance, Volume 37, pp. 5406–5420.
- Quijano, M., 2014, “Information asymmetry in U.S. banks and the 2009 bank stress test,” Economics Letters, Volume 123 (2), pp. 203–205.
- Spargoli, F., 2012, “Bank recapitalization and the information value of a stress test in a crisis,” manuscript, Universitat Pompeu Fabra.

### Timeline of Stress Tests (selected entries)
- SCAP 2009: Announcement February 10, 2009; Results release May 7, 2009. A total of 19 banks assessed; 10 had a capital gap: Bank of America, Citigroup, Wells Fargo, Morgan Stanley, PNC, GMAC, SunTrust, Regions Financial, Fifth Third Bank, and KeyCorp.
- CCAR 2011: Announcement November 17, 2010; Results release March 18, 2011. No bank-specific results released; Bank of America revealed on March 23 that the Fed had rejected its dividend distribution plan.
- CCAR 2012: Announcement November 22, 2011; Results release March 13, 2012; Revision/Correction March 16, 2012. Four banks failed: Citigroup, Ally, SunTrust, and MetLife.
- DFAST 2013: Announcement November 15, 2012; Results release March 7, 2013. Only Ally failed.
- CCAR 2013: Announcement November 9, 2012; Results release March 14, 2013. Capital plans by Ally and BB&T rejected; Goldman Sachs and JP Morgan conditionally approved.
- DFAST 2014 / CCAR 2014: Announcement November 1, 2013; Results release March 20, 2014 and March 26, 2014. Only Zions failed (DFAST); capital plans by Citigroup, RBS, HSBC, and Santander rejected (CCAR).
- DFAST 2015: Announcement October 23, 2014; Results release March 5, 2015. All banks pass.
- CCAR 2015: Announcement October 23, 2014; Results release March 11, 2015. Plans of Deutsche Bank and Santander rejected; Bank of America received a conditional non-objection.

### Stress Test Participants (selection and flags)
- ALLY FINANCIAL INC.: 111111 (USA)
- AMERICAN EXPRESS COMPANY: 111111 (USA)
- BANK OF AMERICA CORPORATION: 111111 (USA)
- BBVA COMPASS BANCSHARES, INC.: 000011 (ESP)
- BMO FINANCIAL CORP.: 000011 (CAN)
- CITIGROUP INC.: 111111 (USA)
- DEUTSCHE BANK TRUST CORPORATION: 000001 (DEU)
- HSBC NORTH AMERICA HOLDINGS INC.: 000011 (GBR)
- JPMORGAN CHASE & CO.: 111111 (USA)
- METLIFE, INC.: 1/111000 (USA) — note: Metlife, Inc. dropped out after selling its commercial bank and de-registering as a bank holding company in 2012.
- RBS CITIZENS FINANCIAL GROUP, INC.: 2/000011 (GBR/USA) — note: RBS sold about 25 percent of its stake in Citizens Financial Group, Inc. in fall 2014, with plans to shed the rest by end-2016.
- UNIONBANCAL CORPORATION: 3/000011 (JPN) — note: Effective July 1, 2014, UnionBanCal Corporation changed its name to MUFG Americas Holding Corporation.
- WELLS FARGO & COMPANY: 111111 (USA)
- ZIONS BANCORPORATION: 000011 (USA)

### Summary statistics (Table 3)
- Assets (billion dollars): Mean 187.04; St. Dev. 54.06
- Market cap (billion dollars): Mean 16.83; St. Dev. 7.45
- Tier 1 Capital / RWA (percent): Mean 12.93; St. Dev. 3.21
- RWA / Total assets (percent): Mean 72.41; St. Dev. 15.07
- ROAA (percent): Mean 0.71; St. Dev. 1.80
- Opacity (percent): Mean 7.14; St. Dev. 4.38
- Complexity (log number): Mean 2.41; St. Dev. 1.94
- Earnings management (percent): Mean -0.01; St. Dev. 0.50
- (Columns report statistics for All / Tested / Untested groups; table header indicates All, Tested, Untested but numeric values above correspond to All)

Definitions included:
- RWA: Risk-weighted assets.
- ROAA: Return on average assets.
- Opacity: Book value of bank premises and investments in unconsolidated subsidiaries, intangible assets, and “other assets” divided by total assets.
- Complexity: Count of subsidiaries.
- Earnings management: Difference between discretionary realized security gains and losses and discretionary loan loss provisions as a percent of total assets.

### Event-study results — Reaction around Announcement (Table 4, selected figures)
- All events: Cumulative abnormal return (CAR) — Tested -0.14; Untested -0.08 (Difference not significant). Absolute value of CAR — Tested 3.30; Untested 2.54 (Difference Yes). Abnormal volume — Tested 0.89*; Untested -0.07 (Difference Yes).
- SCAP 2009: CAR — Tested 3.95; Untested 0.53 (No). Absolute value of CAR — Tested 9.47; Untested 3.66 (Yes). Abnormal volume — Tested 3.80*; Untested 1.07* (Yes).
- CCAR 2011: CAR — Tested -4.11*; Untested -1.36* (Yes). Absolute value of CAR — Tested 4.50; Untested 2.08 (Yes). Abnormal volume — Tested 0.13; Untested -1.37* (Yes).
Notes: Sample = largest 100 bank holding companies as of 2014Q4. Abnormal activities computed over the 30 trading days prior to the event. * denotes statistical significance at conventional levels (10 percent or lower).

### Event-study results — Reaction around Results Release (Table 5, selected figures)
- All events: CAR — Tested 1.56; Untested 1.51 (No). Absolute value of CAR — Tested 3.77; Untested 3.28 (Yes). Abnormal volume — Tested 0.34*; Untested 0.09 (Yes).
- CCAR 2012: CAR — Tested 4.35*; Untested 4.92* (No). Absolute value of CAR — Tested 4.46; Untested 5.02 (No). Abnormal volume — Tested 1.46*; Untested 0.65* (Yes).
- DFAST 2015: CAR — Tested 4.84*; Untested 4.23* (Yes). Absolute value of CAR — Tested 4.84; Untested 4.40 (No).
Notes: Same sample and estimation window as Table 4. * denotes statistical significance at conventional levels (10 percent or lower).

### Information indicators — Announcement and Results Release (Tables 6 and 7, selected estimates)
- Table 6 (around Announcement) — Difference-in-difference estimates (sample = largest 100 bank holding companies as of 2014Q4). Examples:
  - Realized volatility: All events coefficient 0.001890 (not starred), Event & Tested 0.0341, Event & Tested -0.002030. SCAP 2009 realized volatility coefficients include -0.0738** and 0.223.
  - Jump component: All events coefficient 0.142***; Event & Tested -0.325***; Event & Tested -0.154***.
  - Significance notation: ***, **, and * denote 1, 5, and 10 percent levels, respectively.
- Table 7 (around Results Release) — Difference-in-difference estimates (sample = largest 100 bank holding companies as of 2014Q4). Examples:
  - Realized volatility: All events 1.770***; Tested 0.857; Event & Tested 3.530***.
  - Bid-ask spread, equity: All events 0.545***; Tested -0.285**; Event & Tested -0.372*.
  - CDS spread, 1-year / 5-year: All events 0.000863; Tested -0.0270; Event & Tested -0.00683.
  - SCAP 2009 shows pronounced coefficients (e.g., Realized volatility 7.683***, Jump component 25.96***).

### Passed versus Failed (Table 8, selected figures)
- All events: Absolute value of CAR — Passed 3.26; Failed 7.42 (Difference significant: Yes). Abnormal volume — Passed 0.19; Failed 1.47* (Yes).
- SCAP 2009: Absolute value of CAR — Passed 8.82; Failed 15.74 (Yes). Abnormal volume — Passed 0.74; Failed 2.11* (No).
- Lower-panel regressions report significant coefficients, e.g., All events regression: 1.852*** and 1.31115.09*** (significance at 1 percent).

### Market reaction regressions and bank characteristics (Tables 9a–9b, selected coefficients)
- Table 9a (Cumulative abnormal return (absolute value) and Abnormal volume): 
  - Stress tested (dummy = 1) coefficients reported across multiple columns, e.g., 0.641*; 1.992; -0.0560; -0.125; 1.289**; 0.923.
  - Tier 1 Capital / RWA coefficients: 0.0405; 0.0884.
  - Market cap / Total assets interaction with Stress tested: -7.503; 5.345; * Stress tested 1.441; -6.112*.
  - Opacity and Complexity interactions reported, e.g., Complexity: -0.0977; 0.157*.
  - Auditor indicators reported for Big 4 with interactions.
- Table 9b (Spread, equity and Implied volatility regressions):
  - Stress tested coefficients include -0.222; 0.444; -0.284; -1.795**; -0.414.
  - Tier 1 Capital / RWA and interactions reported (e.g., 0.0152).
  - Market cap / Total assets and interactions: -0.0183; -0.133; * Stress tested 0.4200; 0.159.
  - RWA / Total assets: -0.0172**; 0.000242.
- Notes: Sample = largest 100 bank holding companies as of 2014Q4. Regressions include year fixed effects. Robust standard errors clustered at bank holding company level. ***, **, and * denote significance at 1, 5, and 10 percent levels.

### Post-stress-test behavior (Table 10, selected OLS regression results)
- Variables analyzed: Quarterly changes and cumulative changes two quarters following a stress test; in-between stress test news.
- Selected coefficients (OLS, year fixed effects, control for size measured by log assets):
  - Tested coefficients across outcomes: -0.0139; -0.0103; -0.00824; -0.0173; -0.00431; -0.00280.
  - Failed coefficients: -0.0201*; 0.0444**; -0.00852.
  - Additional tested and failed coefficients reported for other outcomes (e.g., -0.0446, -0.0450).
- Significance notation: ***, **, and * denote 1, 5, and 10 percent levels.

### Figures and visual summaries
- Figure 1. Cumulative Abnormal Return around Results Release: Density plots of CAR for Passed versus Failed around results release (axes and density scales as in source figure).

### Appendix: Description of variables (variable definitions and data sources)
- Assets: Total assets (BHCK2170 in Call Report).
- Market cap: Market value of equity (price times shares outstanding).
- Tier 1 Capital / RWA: Already expressed in percent of risk-weighted assets. Basel III definition if available, Basel I definition otherwise (BHCA8274 and BHCK8274 in Call Report, respectively).
- RWA: Risk-weighted assets (as reported in SNL).
- ROAA: Return on average assets (as reported in SNL).
- Spread: Relative equity/bond spread calculated as the difference between ask and bid divided by half of the difference between ask and bid.
- Realized volatility: Daily volatility computed by means of the bipower variation at 5-minute frequency.
- Implied volatility: At-the-money equity call option implied volatility in standard deviations (as reported in Datastream).
- CDS spread: Ratio of the one-year CDS spread to the five-year CDS spread.
- Opacity: Book value of bank premises (BHCK2145) and investments in unconsolidated subsidiaries (BHCK2130), intangible assets (BHCK5507), and “other assets” (BHCK2160) divided by total assets (BHCK2170).
- Complexity: Count of subsidiaries including nonbank and thrift subsidiaries (RSSD9146, BHCP2794, BHCP2796).
- Earnings management: Difference between discretionary realized security gains and losses and discretionary loan loss provisions as a percent of total assets. Discretionary components obtained from fixed-effects OLS regressions following Cornett et al (2009) (Call Report items BHCK3196 plus BHCK3521 and BHCK4230).
- Auditor: Indicator for each of the Big 4 auditing companies (Deloitte, Ernst & Young, KPMG, PWC).
- Lending growth: Change in total loans net of earned income (BHCK2122 in Call Report).

*Content based exclusively on the PDF chapter/section titled "References" (file: _wp15271 - References).*

---


_Source: https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2015/_wp15271.pdf_