The Information Content of Stress Test Announcements

Finance and Economics Discussion Series Divisions of Research & Statistics and Monetary Affairs Federal Reserve Board, Washington, D.C. The Information Content of Stress Test Announcements Luca Guerrieri and Michele Modugno 2021-012 Please cite this paper as: Guerrieri, Luca, and Michele Modugno (2021). “The Information Content of Stress Test Announcements,” Finance and Economics Discussion Series 2021-012. Washington: Board of Governors of the Federal Reserve System, https://doi.org/10.17016/FEDS.2021.012. NOTE: Staff working papers in the Finance and Economics Discussion Series (FEDS) are preliminary materials circulated to stimulate discussion and critical comment. The analysis and conclusions set forth are those of the authors and do not indicate concurrence by other members of the research staff or the Board of Governors. References in publications to the Finance and Economics Discussion Series (other than acknowledgement) should be cleared with the author(s) to protect the tentative character of these papers.

The Information Content of Stress Test Announcements Luca Guerrieri∗, Michele Modugno†. ‡ February 17, 2021 Abstract We exploit institutional features of the U.S. banking stress tests to disentangle different types of information garnered by market participants when the stress test results are released. By examining the reaction of different asset prices, we find evidence that market participants value the stress test announcements not only for the information on possible future capital distributions but also for the signals about bank resilience. These results back the use of stress tests by central banks to inform the broader public about the soundness of the banking system. JEL Codes: E58, G21 Keywords: Stress tests, event study, banks, overnight stock returns, CDS spreads ∗E-mail: luca.guerrieri@frb.gov †E-mail: michele.modugno@frb.gov ‡TheauthorsareeconomistsattheBoardofGovernorsoftheFederalReserveSystem. WethankWilliamBassett, JoseBerrospide,JasonBrown,SebastianInfante,AnnaKovner,KleopatraNikolaou,DinoPalazzoandSkanderVan denHeuvelforhelpfulcomments. ThematerialinthispaperdoesnotrepresenttheviewsoftheBoardofGovernors of the Federal Reserve System or any other person associated with the Federal Reserve System. 1

1 Introduction In the wake of the Global Financial Crisis, it has become standard practice around the world to rely on stress tests to gauge the soundness of banks. In the United States, the annual release of the stress test results is subject to intense public scrutiny accompanied by voluminous media coverage.1 Not surprisingly, financial market participants have also shown a keen interest in the results of the stress tests. Flannery, Hirtle and Kovner (2017) and Fernandes, Igan and Pinheiro (2020) show that stress test disclosures are associated with significantly higher returns and trading volume for the stress-tested banks. While it is already known that financial markets react to stress test announcements, it is not clear whether this reaction is driven by the immediate impact on capital distribution plans to investors, whose approval by the Federal Reserve is linked to the stress test results, or whether it is driven by the stress test results conveying news about the ability of the largest banks to withstand harsh economic conditions. The former hypothesis implies that only a restricted set of financial actors benefits from the stress tests, while the latter implies that the stress test results convey information on the soundness of the banking system of importance for the broader public. A corollary of this latter hypothesis is that the stress test scenarios are severe but plausible and the related stress test results are credible. We find evidence that market participants value the stress tests not only for the information on possible future capital distributions but also for the signals about bank resilience. A focus on the U.S. stress tests allows us to gauge the reaction to different types of information as, until 2019, the annual results used to be announced in two phases. For the first phase, which carried no supervisory consequences, we find that when results pointed to a more sizable capital cushion, CDS spreads declined and stock prices rose systematically. This configuration of reactions of stock returns and CDS spreads point to financial market participants acting on information about the underlying probability of default of each firm, as opposed to only reacting to changes in expected payouts. In connection with a second phase of the stress tests that had supervisory consequences, we also find evidence that when regulators prevented dividends or share buybacks from expanding, stock prices declined, but we do not detect a systematic response of CDS spreads. For this second phase, the consequences for payouts dominate. Nonetheless, taken together, the results for the two phases point to stress tests being informative about the resilience of each firm even abstracting from supervisory consequences. Our analysis is based on an array of event studies. A typical requirement of these studies is to isolate the surprise component for the event by controlling for market participants’ expectations. 1Supervisory objections to the capital plans submitted as part of the stress tests have also been associated with staff changes at the very top of the largest U.S. bank holding companies. Perhaps most prominently, the objections to the capital plans of Citigroup are often cited in press coverage as explanation for the resignation of the CEO in 2012. See for instance, https://www.wsj.com/articles/SB10000872396390443854204578060280201488530 2

We exploit the peculiar structure of the U.S. stress tests that used to differentiate them from the stress tests conducted in other countries. Through 2019, results for two distinct phases of the stress testswereannouncedoneweekapart,withtheresultsoftheDoddFrankActStressTests(DFAST) comingbeforetheresultsfortheComprehensiveCapitalAnalysisandReview(CCAR).Theresults were based on the same supervisory scenarios but differed by their assumption regarding capital plans for dividends and share buybacks and by the supervisory consequences. The DFAST results used a dummy capital plan based on the distributions for the previous stress test cycle, while the CCAR results incorporated the new capital plans. Furthermore, the DFAST results bore no direct supervisory consequences for the firms, while the Federal Reserve could curb capital distributions based on the subsequent CCAR results. This particular structure allows us to take the results for DFAST and CCAR within the same cycle to size the surprise change in capital plans, whereas we can take the difference in results for DFAST and CCAR across cycles to size the surprise change in the capital cushion. Furthermore, the fact that new capital plans are only approved when the CCAR results are released helps us isolate information about bank resilience at the DFAST phase.2 We first show that, following the release of the DFAST results, increases in loss absorption capacity, as implied by a higher minimum value of tier 1 capital relative to the minimum in the previous stress test cycle lead to declines in CDS spreads and increases in stock prices. Although a higher tier 1 capital minimum under DFAST does not guarantee that bank will be able to expand its payouts, we cannot exclude that market participants take such a signal from the results. However,theassociatedtighteningofCDSspreadsclearlypointstomarketparticipantsinterpreting the DFAST results as indicative of greater resilience, and may explain part of the stock price reaction through the risk-premium channel. Turning to the CCAR results, the reaction of stock prices was systematically related to actions undertaken by the Federal Reserve that curbed capital distributions. We show that when capital plans were objected to or not approved, stock prices declined. This result seems exclusively driven by the limitations on capital distributions, as we do not find evidence of a significant response of CDS spreads. In sum, we find clear evidence that financial markets scrutinize the stress test results to understand whether participating firms can withstand harsh economic conditions. If we had found that only stock prices reacted systematically, it would have pointed to benefits from the stress tests accruingtoalimitednumberofindividual. Bycontrast,ourfindingthatCDSspreadsrespondwith an opposite sign relative to the response of stock returns highlights the importance of information from the stress tests for the broader public. 2See Lehnert and Hirtle (2015) for a detailed description of the institutional details for the stress tests. 3

2 Literature Review The academic literature on macroeconomic stress testing is expanding rapidly building on early contributionstoresearchonstressteststhatdrewheavilyontheexperienceofpractitioners. Henry et al. (2018) provide a comprehensive review. Blaschke et al. (2001) and Kohn and Liang (2019) provideageneraloverviewoftheimplementation,benefits,andconsequencesofstresstestingbanks. Perhaps, Flannery, Hirtle and Kovner (2017) and Fernandes, Igan and Pinheiro (2020) are the closest papers to ours. They find that markets do react to stress test announcements. Building on their results, we use firm-specific data to investigate whether the market reaction to the stress test announcements is linked to the ability of banks to withstand harsh economic conditions or is connected to the expected cash flow. Our paper is also related to Philippon, Pessarossi and Camara (2017) and to Georgescu et al. (2017), whose focus is on the stress tests conducted by the European Banking Authority. Philippon, Pessarossi and Camara (2017) use the scenarios for macroeconomicfactors-—GDPgrowth,inflation,andunemployment—toestimatethesensitivityof individualbankstomacroeconomicshocks. Theyfindtheseestimatedsensitivitiespredictrelatively well the realized losses of banks in subsequent years, backing the value of the stress test results. Georgescuetal.(2017)alsofocusonstockreturnsandCDSspreadsfindingasystematicrelationship with the stress test results. In addition, they look at CDS spreads on sovereign debt for the headquarter countries of banks in the stress tests, finding a systematic impact there, too. We focus on the U.S. stress test announcements to discriminate between the risk premium versus the cash flow information content of those announcements. A separate strand of the literature has focused on the value and desirability of disclosing supervisory information. For instance, Goldstein and Sapra (2014), Alvarez and Barlevy (2014), and Goldstein and Leitner (2015) theoretically show that disclosing supervisory information may be welfare improving and may promote financial stability. Our empirical results contribute to this literature showing that the information content of stress test announcement can indeed affect market participants’ perception of the riskiness of banks. Finally, our work is also related to the vast literature that explores the reaction of asset prices to macroeconomic announcements or to corporate announcements, such as dividends. On the macroeconomic side, early papers, such as Fama (1981), focused on the connection between stock returns and inflation measures or monetary aggregates. Apart from news for inflation, Flannery and Protopapadakis (2002) shows a systematic reaction of stock returns to the balance of trade, employment, and housing starts. Boyd, Hu and Jagannathan (2005) investigate the information content of news about unemployment. On the corporate side, Brav and Heaton (2015) provide a helpful review of evidence from event studies.3 Unlike ours, most event studies are either focused 3See, in particular, Appendix 4 of Brav and Heaton (2015). 4

on broad measures of stock returns or exclude financial firms. Unlike ours, most event studies are either focused on broad measures of stock returns or exclude financial firms. An exception is Kelly, Lustig and Nieuwerburgh (2016), which is squarely focused on banks. 3 Sizing Surprises and Market Reactions Forouranalysis,weexploitthepeculiarstructurefortheannouncementsoftheU.S.stresstests. Through 2019, the U.S. stress tests consisted of two separate tests with results published about a week apart. The first results to be released were for DFAST, followed by those for CCAR. Both tests evaluated capital adequacy assuming exactly the same supervisory macroeconomic scenario. However, while the DFAST analysis was conducted assuming a dummy capital distribution plan, basedonthepreviousyear’sdistributions, thecapitalcalculationsfortheCCARresultswerebased on the proposed capital distributions for the current stress test cycle. We rely on the difference between the CCAR and the DFAST results to isolate the unexpected component of the stress test announcements. A further important difference between CCAR and DFAST is that they entail different supervisory consequences: while the Federal Reserve did not take supervisory actions following the DFAST results, it could object to the capital distribution plans and put limits on payouts following the CCAR results.4 Our analysis, based on an array of event studies, faces two challenges. First, we need to select a summary measure of the stress test results; and second, we need to control for market participants’ expectations of this summary measure in order to isolate the surprise component. Given that the principalaimofthestresstestsistoassessthecapitaladequacyandcapitalplanningpracticesofthe participating firms, we focus on capital. In particular, our summary measure for each participating firm is the minimum value of the tier 1 capital ratio over the nine-quarter assessment period used by the U.S. stress tests.5 Inordertoidentifythesurprisecomponentofthesemeasures,weadaptourapproachdepending onwhetherweareanalyzingtheCCARortheDFASTannouncements. Fortheformer,oursurprise component is the difference between the minimum value for the tier 1 capital ratio attained under thecurrentyear’sDFASTminustheoneattainedunderthecurrentyear’sCCAR.Whilethecapital pathunderDFASTassumesadummyplanbasedonlastyear’scapitaldistributions, CCARresults are based on capital plans for the current year. Accordingly, when this difference is positive, it 4In some years the Federal Reserve issued objections to the capital plans, in others it issued “non-approvals.” Bothcarriedtherequirementtosubmitrevisedcapitalplansandcurbpayoutstoshareholderspriortotheapproval of the new plans. 5The Federal Reserve publishes stress test scenarios that cover 13 quarters. However, the last four quarters are used to compute provisions. Accordingly, the stress test results report the minimum tier 1 capital ratio over a nine-quarter assessment period. 5

points to an expansion in approved capital distributions relative to last year’s. Given the relatively short time lapse between the release of the DFAST and the CCAR results, this measure is a good proxy of the surprise component of the CCAR results. To identify the surprise component of the DFAST results, we take the difference between the minimum value for the tier 1 capital ratio in the current year’s DFAST and the minimum value for tier 1 capital ratio in the previous year’s CCAR.6 Given that this year DFAST results are based on the assumption that the BHC will implement the same capital distributions approved by the Federal Reserve in the previous year CCAR, a higher minimum capital requirement may point to greater resilience and signal that the BHC can proceed with a larger capital distribution compared to last year, if allowed by the following week’s CCAR results. 3.1 Abnormal Overnight Trading Volumes and Returns For the event studies, we focus on the banks headquartered in the United States.7 Figure 1 shows overnight returns around the CCAR announcements. The data are from CRSP.8 The figure expresses these returns as percentiles of the distribution of overnight returns for each bank for the six months preceding each release. Overnight returns following CCAR announcement are mainly abnormal, coming from the tails of the distributions for the six months preceding the announcements. The figure also highlights important variation year by year.9 As we only report the overnight returns for the banks included in our study, Figure 1 also gives a complete account of the firms in our dataset. Notice that Ally Financial and Citizens Financial Group became public companies after the release of the 2014 results, which explains missing values in the figure for those two banks for the 2013 and 2014 cycles. Missing values for other banks in certain cycles indicate that the Federal Reserve did not test those banks in those particular cycles. We next explore whether abnormal returns could be connected with markets seizing up, but find little evidence of malfunctions. Figure 2 reports analogous statistics to those in Figure 1 for 6Among other factors, we also control for changes in starting capital. Recent papers have focused on the predictability of stress test results. Glasserman and Tangirala (2016), for example, make note of the high correlation across the scenario results across years and across banks. 7We exclude foreign intermediate holding companies that are also subject to the stress tests since the market reaction, only available for the parent company, would not be comparable to the market reaction for the domestic bank holding companies. 8Center for Research in Security Prices, CRSP 1925 US Stock Database, Wharton Research Data Services, https://wrds-web.wharton.upenn.edu/wrds/. 9The results for 2017 stand out even among the outliers. A blog post by David Dowd published on the website of the Cato Institute shortly after the release of the CCAR 2017 results noted:“This year, the news is particularly good. As usual, thekey capital metrics across the system are betterthan ever. And whereas in previousyears there werealways[banksthat]failed,thelatestsetofstresstestsarethefirstinwhichallthebankspassedandthisyear’s class laggard, Capital One, got only the mildest of slaps on the wrist.” 6

tradingvolumes. ThedatasourceisagainCRSP.10 ThefigureshowsthatthedaysfollowingCCAR releases are also characterized by abnormally elevated trading volumes, mainly in the upper tails of the distributions, which we interpret as indicating that markets were functioning smoothly. Overnight returns on the day of DFAST results are similarly abnormal. The middle panel of Figure 3 shows data analogous to those in Figure 1 but the returns pertain to DFAST releases and are shown as a bar chart. The top and bottom panels of that figure allow a comparison with overnight returns for the day before the release of the stress test results and for the day after. This comparison highlights that overnight returns straddling the release of stress test results are indeed special: Abnormal returns, those in the upper percentiles of the distribution for the six months prior to each release, are much more prevalent.11 By contrast, on the day preceding and the day following the release of the results, the distribution of overnight returns is more uniform across the percentile bins shown. 3.2 Other Data For the event studies, apart from data on stock returns, we also use data on CDS spreads relative to Treasuries. The data are from Markit. We select the most liquid contracts quoted in dollars with a maturity of 5 years. We focus on contracts for the senior debt tranche since these contracts are more widely traded than contracts on other tranches. CDS contracts are traded over the counter. The Markit data report quotes for 9:00 PM, the close of the trading day. While the CDS spread movements on the day surrounding stress test announcements do not tend to be as extreme (relative to their distribution) as the reaction of stock returns and traded volumes, they can still be sizable. Furthermore, there is significant heterogeneity in the results across banks.12 Moving to additional controls, in some of our regressions we use: the starting level of the tier 1 capital ratio, as reported in the CCAR/DFAST results; the difference between the starting level of the tier 1 capital ratio across cycles, when we compare two different stress test cycles; the forced decrease in payouts — the difference between the tier 1 capital minimum in the final capital plan submission and the original submission; and an indicator for whether the capital plans were objected to or not approved.13 The consequences of an objection were the inability of making any 10Markit North America, Inc. Credit Default Swaps (CDS), Wharton Research Data Services, https://wrdsweb.wharton.upenn.edu/wrds/. 11An online appendix shows an analogous pattern of overnight returns for CCAR result releases. 12The online appendix includes a figure for CDS spreads analogous to Figure 1 and another analogous to Figure 3. 13Before the public disclosure of the CCAR results, each participating BHC was shown its own results and given an opportunity to reduce dividends and stock repurchases. In practice only those firms whose original capital distributions would have pushed them below statutory minima took advantage of this opportunity, which is why we are calling this revision a forced reduction in payouts. 7

capital distributions, unless expressly permitted by the Federal Reserve.14 Non-approvals generally carried less-dire consequences, with firms having to limit their capital distributions to the levels of prior years. 4 Event Study The stress tests had a constant format from 2013 through 2019, yielding seven events for our study. However, the day following the 2016 DFAST announcement coincided with the release of the results for the Brexit referendum, which dramatically affected the prices of bank stocks as well as other asset prices. Accordingly, we exclude the 2016 results when analysing the reaction to the DFAST releases. To compensate for the relatively short temporal dimension of the data, we exploit a larger cross-section. There are as many as 25 BHCs headquartered in the United States that have participated in the stress tests over the 2013-2019 period. Accordingly, we rely on panel regressions for our analysis. We use the following panel regression model: y = α+βs +Φ +Ω +Ψ +u (1) i,t+v i,t t i,t i i,t+v wheret isthedayoftheCCARorDFASTannouncementandwheretheleft-hand-sideterm, y , i,t+v is in turn: • the overnight percentage change in the stock price of BHC i through t+1 (which implies v = 1). In other words, this is the opening price on the day t+1, following the announcement, minus the closing price on the day t of the announcement, divided by the closing price of day t, times 100. • the daily change of the CDS spread for BHC i for day t (which implies v = 0). In other words, this is the CDS spread surveyed at the end of day t minus the CDS spread surveyed at the end of day t-1. Giventhatthestresstestresultsareusuallyreleasedat4:30p.m.,aftertheclosingofthetrading day on the New York Stock Exchange (at 4:00 p.m.), the overnight stock price changes identify the impact of the new information in the announcements. In contrast with the stock prices, the CDS spreads are not available at a frequency higher than daily. However, given that CDS contracts are traded over the counter and that trading stops at 9:00 p.m., we can rely on market surveys for the 14Afirmwhosecapitalplanwasobjectedtocouldresubmitanewcapitalplanaheadofthefollowingyear’sstress tests, but was not required to do so. 8

end of the day when the stress test results were released. We difference the closing CDS spreads relative to the previous day’s values to gauge the market reaction to the stress test results. The terms on the right-hand side of Equation 1 differ depending on whether we are analyzing CCAR or DFAST results. Starting with the regression equation for CCAR, the term s is the i,t difference between the minimum value for the tier 1 capital ratio attained under the current year’s DFAST minus the one attained under the current year’s CCAR. The term Ω includes a dummy i,t that captures supervisory actions of the Federal Reserve in the context of CCAR, non-approvals or objections; the starting capital; and forced decreases in payouts. The terms Φ and Ψ include t i time fixed effects and firm-specific fixed effects, respectively. Turning to regression equation for DFAST, the term s is defined as the difference between the i,t minimum value for the tier 1 capital ratio in the current year’s DFAST and the minimum value for the same ratio in the previous year’s CCAR. The term Ω includes three types of firm-specific i,t controls: 1)thedifferenceinthestartingtier1capitalratiobetweenDFASTandthepreviousyear’s CCAR; 2) the starting capital; and 3) a dummy that captures decisions of the Federal Reserve for the previous CCAR cycle, non-approvals or objections. The terms Φ and Ψ include time fixed t i effects and firm-specific fixed effects, respectively. 5 Capital Distributions vs. Resilience The following two sections describe the regression results starting with DFAST and moving on to CCAR. 5.1 DFAST Results Offer Information on Resilience Table 1 shows the results of panel regression models where the dependent variables are the percentagechangesinstockpricesandthechangesinCDSspreadsaroundDFASTannouncements, expressed as basis points. At a broad brush, when the DFAST minimum tier 1 capital ratio is higher than the previous year’s CCAR minimum, stock prices systematically increase, as we can see from the estimated coefficients in columns (1) and (2). According to those coefficients, when a bank’s minimum of the tier 1 capital ratio is one percentage point higher, the overnight stock return increases about 0.22 percentage point, all else equal.15 This increase may happen for two non-mutually exclusive reasons: 1) market participants interpret increases in tier 1 capital ratio minima as a signal of 15This increase is statistically significant at conventional levels based on standard errors that are robust to heteroscedasticity. This increase remains significant when considering standard errors that are clustered at the firm level, as documented in the online appendix. 9

greater resilience to adverse conditions, reducing the risk of holding the stocks of those banks; 2) market participants interpret the increase in tier 1 capital minima as a signal that BHCs may have more capital to distribute compared to the previous year, although there is no guarantee that the Federal Reserve will approve greater payouts until the following week’s CCAR results are released. The reaction of CDS spreads can help corroborate or exclude the first reason. Columns (3) and (4), point to a systematic decrease of CDS spreads when the DFAST minimum tier 1 capital ratio is higher than the previous year’s CCAR minimum. This decrease is even more significant, in statistical terms, than the increase in stock prices. This result points to financial market participants reading a higher stressed capital ratio as an indication of greater resilience of banks in the face of harsh economic conditions. A corollary of this finding is that financial market participants view the stress test scenario as relevant and the results as credible. 5.2 CCAR Results Point to Changes in Capital Distributions Table 2 shows the results of panel regression models where the dependent variables are the overnight stock price returns and CDS spread changes around CCAR announcements. In this case, and in contrast with the results for DFAST releases, the difference between the minimum value of the tier 1 capital ratio under DFAST and under CCAR has an insignificant coefficient for both stock returns and CDS spreads. However, from column (2) we can see that when the Federal Reserve issued a non-approval or an outright objection to the proposed capital plans, stock returns systematically decreased. This decrease is sizable for overnight returns. On average, it is sized at about 2.2 percentage points, all else equal. However, the insignificance of the coefficient on the same dummy in the regression of CDS spreads in column (4) indicates that the main concern of market participants at this phase of the stress tests is the impact of supervisory actionsoncapitaldistributions.16 AfterallthepreviouslyreleasedDFASTresultsalreadyprovided informationonbankresilienceinthefaceofadverseconditionsthat,asweshowed,issystematically related to CDS spreads. 6 Sensitivity Analysis An online appendix documents robustness of our regression results to numerous specification changes. Werefertheinterestedreadertothatappendixforthoseadditionalresultsandonlysketch herethemostsalientchangesweconsidered. Wefocusontwotypesofsensitivityanalysis. Thefirst type considers alternative specifications keeping the change in tier 1 minimum capital across stress 16Theexpectationthatafirmcouldraiseadditionalcapitalthroughequityissuanceinconnectionwithanobjection would also be consistent with our regression results. 10

test cycles as the surprise measure for the event studies. The second type of sensitivity analysis considers alternative surprise measures. In sum, the baseline results are strikingly robust. 6.1 Robustness to Specification Changes Under CCAR, firms had a chance to reduce their proposed capital distributions to avoid a stressed capital minimum that fell short of the statutory ratios. The change in tier 1 capital minimum from DFAST to CCAR in our baseline specification is based on the CCAR minimum under the original plans. However, the regression includes a term for the (forced) decrease in payouts when firms resubmitted capital plans with lower distributions when the original plans put them below a statutory ratio.17 This term is flagged in the results released by the Federal Reserve. As an alternative, we can compute the change in tier 1 capital minimum across DFAST and CCAR with the CCAR minimum based on the revised plans and exclude the term that captures the forced reduction in capital distributions. When we make these changes the regression results vary little quantitatively and are unchanged qualitatively. The samples for the CCAR regressions and the DFAST regressions have different number of observations, complicating the comparison of results. The CCAR regressions have more observations because we do not need to compute the change in tier 1 minimum capital across stress test cycles. Moreover, wedonotneedtodrop2016fortheCCARregressions—rememberthatthereleaseofthe DFAST results coincided with the announcement of the results for the Brexit referendum. When we drop the additional observations from the sample for the CCAR regressions, the coefficient on the objection dummy is even more statistically significant. There are no other changes of note in the regression results. 6.2 Robustness to Alternative Surprise Measures The measure summarizing the information provided by the release of stress test results in our baseline regressions is based on tier 1 capital. However, the stress test results report the starting value and minimum conditional on the stress test scenarios for other statutory ratios: the tier 1 leverage ratio, the total risk-based capital ratio, and the common equity tier 1 ratio. We also consider surprise measures based on each of these alternative ratios and find that our results are extremely robust. 17ThebanksparticipatingintheU.S.stresstestsweregivenalimitedsetofoptionsforrevisingtheircapitalplan submissions. See the section entitled “Limited Adjustments to Planned Capital Actions” in Board of Governors of the Federal Reserve (2019). 11

7 Conclusion We have found evidence that market participants extract different information from the two components of the U.S. stress test. The DFAST results, released first, point to market participants taking signal for the resilience of banks from changes in capital positions under stressed conditions. Banks with a higher stressed capital minimum relative to the previous stress test cycle systematically experience an increase in their stock prices and a decrease in their CDS spreads. Our analysis of the market reaction to the release of the CCAR results points to a systematic response of stock returns to restrictions on payouts. Theseresultsbackthewidespreaduseofstressteststoinformmarketparticipantsonthesoundness of banks. Our analysis shows that market participants value the stress test announcements not only to gauge future capital distributions, which would simply benefit a limited set of investors, but also as indicators of bank resilience, with importance for the broader public. In the United States, stress tests have supervisory consequences for bank capital, but our results point to the usefulness of the information provided by stress tests even when the results are not tied to capital actions by the regulator, as is the case in other countries. 12

References Alvarez, Fernando and Gadi Barlevy. 2014. Mandatory Disclosure and Financial Contagion. Working Paper Series WP-2014-4 Federal Reserve Bank of Chicago. Blaschke, Winfrid, Matthew T Jones, Giovanni Majnoni and Soledad M Peria. 2001. “Stress Testing of Financial Systems: An Overview of Issues, Methodologies, and FSAP Experiences.” IMF Working Paper . Board of Governors of the Federal Reserve. 2019. Comprehensive Capital Analysis and Review 2019, Summary Instructions. Technical report. Boyd, John H., Jian Hu and Ravi Jagannathan. 2005. “The Stock Market’s Reaction to Unemployment News: Why Bad News Is Usually Good for Stocks.” Journal of Finance 60(2):649–672. Brav, Alon and J.B. Heaton. 2015. “Competing Theories of Financial Anomalies.” The Review of Financial Studies 15(2):575–606. Fama, Eugene F. 1981. “Stock Returns, Real Activity, Inflation, and Money.” American Economic Review 71(4):545–565. Fernandes, Marcelo, Deniz Igan and Marcelo Pinheiro. 2020. “March madness in Wall Street: (What) does the market learn from stress tests?” Journal of Banking & Finance 112(C). Flannery, Mark, Beverly Hirtle and Anna Kovner. 2017. “Evaluating the information in the federal reserve stress tests.” Journal of Financial Intermediation 29(C):1–18. Flannery, Mark J. and Aris A. Protopapadakis. 2002. “Macroeconomic Factors Do Influence Aggregate Stock Returns.” Review of Financial Studies 15(3):751–782. Georgescu, Oana-Maria, Marco Gross, Daniel Kapp and Christoffer Kok. 2017. Do stress tests matter? Evidence from the 2014 and 2016 stress tests. Working Paper Series 2054 European Central Bank. Glasserman, Paul and Gowtham Tangirala. 2016. “Are the Federal Reserve’s stress test results predictable?” The Journal of Alternative Investments 18(4):82–97. Goldstein,ItayandHareshSapra.2014. “Shouldbanks’stresstestresultsbedisclosed? Ananalysis of the costs and benefits.” Foundations and Trends in Finance 8(1):1–54. Goldstein, Itay and Yaron Leitner. 2015. Stress tests and information disclosure. Working Papers 15-10 Federal Reserve Bank of Philadelphia. Henry, Jerome, Patrizia Baudino, Roland Goetschmann, Ken Taniguchi and Weisha Zhu. 2018. Stress-testing banks - a comparative analysis. Fsi insights on policy implementation no 12 Bank for International Settlements. Kelly, Bryan, Hanno Lustig and Stijn Van Nieuwerburgh. 2016. “Too-Systemic-to-Fail: What OptionMarketsImplyaboutSector-WideGovernmentGuarantees.”American Economic Review 106(6):1278–1319. 13

Kohn, Donald and Nellie Liang. 2019. Understanding the Effects of the U.S. Stress Tests. Report Brookings. Lehnert,AndreasandBeverlyHirtle.2015. “SupervisoryStressTests.”Annual Review of Financial Economics 7(1):339–355. Philippon, Thomas, Pierre Pessarossi and Boubacar Camara. 2017. Backtesting European Stress Tests. NBER Working Papers 23083 National Bureau of Economic Research, Inc. 14

Table 1: Market Reaction to DFAST Announcements (1) (2) (3) (4) Stock returns Stock returns ∆CDS spreads ∆CDS spreads DFAST-CCAR 0.216∗ 0.217∗ -0.383∗∗ -0.407∗∗ minimum (0.017) (0.017) (0.005) (0.004) DFAST-CCAR(-1) 0.0710 0.0776 -0.212 -0.162 start (0.393) (0.334) (0.174) (0.310) Starting capital -0.236∗ -0.236∗ 0.340+ 0.342+ (0.015) (0.015) (0.059) (0.060) Objection or -0.0660 -0.446 non-approval, lagged (0.749) (0.434) r2 0.634 0.634 0.422 0.424 N 102 102 93 93 p-values in parentheses + p<0.1, ∗ p<0.05, ∗∗ p<0.01 Note: The dependent variables in the panel regressions in columns (1) and (2) are overnight returns surrounding DFAST announcements; columns (3) and (4) are for changes in CDS spreads at the end of the day of the DFAST announcements relative to the end of the day prior. DFAST-CCAR min. is the difference between the current year DFAST and the previous year CCAR minimum value of the tier 1 capital ratio over the nine-quarter assessment periodusedintheU.S.stresstests. DFAST-CCAR(-1) start isthedifferencebetweenthestartinglevelofthetier1 capitalratioacrosscycles. Startingcapital isthestartinglevelofthetier1capitalratio. Objectionsornon-approvals, lagged isadummythatassumesvalueoneifthecapitalplanswereobjectedtoornotapprovedinthepreviousyear CCAR.Alltheregressionsincludebanksandyearfixedeffects. Inparentheseswereportthep-valueswhere+p<0.1, *p<0.05, **p<0.01. These values are based on standard errors that are robust to heteroscedasticity. 15

Table 2: Market Reaction to CCAR Announcements (1) (2) (3) (4) Stock returns Stock returns ∆ CDS spreads ∆ CDS spreads DFAST-CCAR 0.0629 0.0401 -0.266 -0.269 minimum (0.620) (0.724) (0.463) (0.459) Forced decrease 0.311 0.324 -0.301 -0.368 in payouts (0.438) (0.347) (0.570) (0.486) Starting capital 0.305∗ 0.170+ 0.114 0.144 (0.015) (0.083) (0.718) (0.659) Objections or -2.145∗∗ 0.532 non-approvals (0.000) (0.450) r2 0.590 0.717 0.524 0.527 N 150 150 111 111 Note: Thedependentvariablesinthepanelregressionsincolumns(1)and(2)areovernightstockreturnssurrounding CCAR announcements; columns (3) and (4) are for changes in CDS spreads at the end of the day of the CCAR announcements relative to end of the day prior. DFAST-CCAR min. is the difference between the current year DFAST and CCAR minimum value of the tier 1 capital ratio over the nine-quarter assessment period used in the U.S. stress tests. Forced decrease in payouts is the difference between the tier 1 capital minimum in the final capital plansubmissionandtheoriginalsubmission. Startingcapital isthestartinglevelofthetier1capitalratio. Objections or non-approvals is a dummy that assumes value one if the capital plans were objected to or not approved. All the regressions include banks and year fixed effects. In parentheses, we report the p-values where +p < 0.1, *p < 0.05, **p<0.01.These values are based on standard errors that are robust to heteroscedasticity. 16

Figure 1: Overnight Returns Following CCAR Announcements (percentiles of the distribution for the preceding six months) Bank 2013 2014 2015 2016 2017 2018 2019 Ally Financial Inc. 100 100 99 100 American Express Company 74 53 87 74 92 93 BB&T Corporation 1 23 93 91 98 92 Bank of America Corporation 98 77 26 91 98 93 98 Capital One Financial Corporation 67 85 89 85 7 86 88 Citigroup Inc. 54 1 97 89 98 92 92 Comerica Incorporated 31 93 90 99 Discover Financial Services 28 97 80 96 92 Fifth Third Bancorp 40 17 90 89 98 92 Huntington Bancshares Incorporated 68 89 81 98 84 JPMorgan Chase & Co. 4 54 81 81 98 95 96 KeyCorp 40 78 94 96 97 95 M&T Bank Corporation 32 84 97 93 94 Morgan Stanley 55 34 99 69 97 91 93 Northern Trust Corporation 47 93 10 97 87 94 RBS Citizens Financial Group, Inc. 49 94 99 87 Regions Financial Corporation 65 19 99 87 99 95 State Street Corporation 44 38 92 86 97 22 89 SunTrust Banks, Inc. 11 63 91 89 99 98 The Bank of New York Mellon Corporation 24 59 98 91 96 90 94 The Goldman Sachs Group, Inc. 4 54 89 86 98 74 97 The PNC Financial Services Group, Inc. 23 52 95 90 99 84 93 U.S. Bancorp 15 34 89 87 97 93 88 Wells Fargo & Company 91 98 95 81 98 99 92 Zions Bancorporation 84 94 93 99 Note: Overnight returns following the release of the CCAR results expressed as percentile of the distribution of overnight stock returns for each bank for the six months preceding the release of the results. The percentiles reported are based on calculations by the authors on data from CRSP. 17

Figure 2: Trading Volumes Following CCAR Announcements (shown as percentiles of the distribution for the preceding six months) Bank 2013 2014 2015 2016 2017 2018 2019 Ally Financial Inc. 88 99 65 80 American Express Company 88 34 86 83 58 53 BB&T Corporation 97 91 77 86 82 74 Bank of America Corporation 96 88 80 80 84 73 73 Capital One Financial Corporation 92 70 76 86 85 45 59 Citigroup Inc. 81 87 81 76 79 67 60 Comerica Incorporated 79 72 87 73 Discover Financial Services 28 73 54 73 36 Fifth Third Bancorp 92 88 91 90 80 74 Huntington Bancshares Incorporated 75 63 96 87 91 JPMorgan Chase & Co. 96 77 67 72 80 68 73 KeyCorp 94 81 89 96 80 81 M&T Bank Corporation 83 79 90 90 92 Morgan Stanley 91 72 83 85 78 72 78 Northern Trust Corporation 60 47 90 67 41 94 RBS Citizens Financial Group, Inc. 11 97 73 85 Regions Financial Corporation 93 81 95 92 95 81 State Street Corporation 94 85 79 94 46 57 97 SunTrust Banks, Inc. 91 64 64 74 80 79 The Bank of New York Mellon Corporation 92 83 79 80 61 82 82 The Goldman Sachs Group, Inc. 82 54 39 64 69 45 67 The PNC Financial Services Group, Inc. 80 65 85 83 70 82 77 U.S. Bancorp 90 77 72 71 77 74 77 Wells Fargo & Company 94 80 77 72 84 91 82 Zions Bancorporation 95 90 96 89 Note: Trading volume for the day after the release of the CCAR results expressed as percentiles of the distribution of daily trading volumes for each bank for the six months preceding the release of the results. The percentiles reported are based on calculations by the authors on data from CRSP. 18

Figure 3: Extreme values of Overnight Stock Returns for Stress-Tested Banks Are Prevalent When Results Are Announced: Reactions to DFAST Results Across Cycles The magnitude of overnight returns the day before the release 20 2013 2014 15 2015 2017 2018 2019 10 5 0 0 10 20 30 40 50 60 70 80 90 100 The magnitude of overnight returns straddling the release 20 15 10 5 0 0 10 20 30 40 50 60 70 80 90 100 The magnitude of overnight returns the day after the release 20 15 10 5 0 0 10 20 30 40 50 60 70 80 90 100 Percentile of the realized distribution over the prior 6 months Note: The absolute value of the overnight returns shown are expressed as percentiles of their realized distribution for the six months prior to the release of the stress test results. The middle panel shows the absolute value of overnight returns based on stock prices at the market closing and opening straddling the announcement of DFAST results. For comparison, the top and bottom panel show analogous returns for the day before and for the day after the announcement of DFAST results, respectively. The percentiles shown are based on calculations by the authors on stock price data from CRSP. 19