1
How Does The Stock Market React To The
US Stress Test Conducted in 2015-2017
Thi Ha Vi Nguyen
1 July 2018
Supervisor: Prof.Dr. Tanju Yorulmazer E-mail: thi.ha.vi.nguyen@student.uva.nl Student number: 11813989
2 Statement of Originality
This document is written by Thi Ha Vi Nguyen who declares to take full responsibility for the contents of this document.
I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.
The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
3
Abstract
This paper investigates the stock market reactions to the US stress tests conducted over the period 2015-2017. Applying the standard event study methodology, I found that with the exception of the announcement of the 2016 stress tests and the result publication of DFAST 2016, all other stress test events had significant impact on stock prices of not only tested but also untested bank holding companies. These findings suggest that the investigated stress tests contained valuable information to the market about the financial health of both the individual tested banks as well as the whole banking system. Furthermore, using the CAMELS rating system to measure the financial health of banks, the paper also examines the relationship between stock market reactions and banks’ financial conditions at the test’s time. There is evidence that the market responded more strongly to the “passing the test” news of less healthy banks. This holds true even for untested banks though it is much less statistically significant. Finally, the detailed results of stress tests did not seem to affect market reactions in anyway. Rather the market seemed to only care about whether the banks failed or passed.
4 1. Introduction
The financial crisis 2007-2008 can be considered a turning point for policy changes in the banking system. This is because the authorities have learned how microprudential policy (i.e. firm-level regulations) failed to prevent the crisis and how capital adequacy plays an increasingly important role in maintaining a healthy banking system. In order to prevent another crisis which will lead to catastrophic consequences for the real economy, financial regulators have come up with a number of unprecedented regulations. Among those, macroprudential stress test has emerged prominent considering both of its effectiveness and large-scale application. The US policymakers brought forward the first formal macroprudential stress test which is often referred to as SCAP (Supervisory Capital Assessment Program) in 2009, at the height of market uncertainty resulted from the crisis. Since then, through the Dodd- Frank Act, the macro stress test has become mandatory for all the bank holding companies (BHCs) that were deemed to be qualified for it by the Federal Reserve. The stress tests are conducted at a system-wide level to determine whether participating BHCs have sufficient capital to survive certain outrageous economic conditions. If any banks fail the stress test, they are required to submit a revised capital distribution plan and/or raise fresh capital.
Stress test has long been a subject of interest for academics, however existing literature on stress test is mostly theoretical, focusing on its background, design, implementation and criticism of some certain frameworks. There have been only a few empirical papers investigating the implications and effectiveness of stress tests by analysing market reactions to the disclosure of the test results in both the EU and the US. Fernandez et al. (2017) and Flannery et al. (2017) are among the most prominent papers approaching this aspect of the US stress tests up to 2015. Some others papers including Petrella and Resti (2013) and Carboni et al. (2017) focus on the EU instead. While it might be interesting to look at the EU stress tests because of the heterogeneous regulations applied to banks in different countries, in this paper, I will focus only on the US and aim to answer the question “How does the stock market react to the US stress test conducted in 2015-2017 and what are its implications?”.
First, to the best of my knowledge, this is the first empirical paper that examines the market reactions to the six US stress tests over that three-year period (there are two stress tests namely CCAR and DFAST conducted each year). The results of these tests have revealed a similar yet intriguing pattern, i.e. almost all banks passed the tests. In 2015, the Federal Reserve rejected the Capital Distribution plan of only two out of thirty one BHCs. In 2017,
5 the results were even better when all thirty four banks passed the tests. Hence, some analysts have been skeptical about the results and raised a concern whether the Federal Reserve has been easing the regulations in favor of participating banks. In other words, are those tests simply too easy for banks to pass? It is beyond the scope of this paper to determine whether it is the improvement of banks’ health or the relaxed regulations that lead to the good results in the past 3 years. The question of focus here are how markets had reacted to the results in those cases and what the implications are.
Compared to the first few stress tests, especially SCAP 2009, the subsequent ones are often considered less effective, in the sense that they did not reveal much unanticipated news, demonstrated by the lower abnormal return in the stock markets around the time of result publication (Candelon, 2015). This is understandable when taking into account the fact that SCAP 2009 was carried out to clear the doubt and regain the market’s lost trust on the financial conditions of BHCs (Fernandez et al., 2017). SCAP 2009 did not go against the expectation of policymakers and the subsequent recovery of the financial system can be largely attributed to this first comprehensive stress test, according to Glasserman and Tangirala (2015). However, they also argue that the test results have become more predictable overtime as the tests became a common practice in a more stabilized environment. If the market did not react to the disclosure of the stress tests, it might suggest that these tests are uninformative, assuming Efficient Market Hypothesis holds.
Second, the paper examines whether banks’ financial conditions at the time of the tests affect how market responds to its performance in stress tests. As far as I know, this is one of the few papers that study the link between market reactions and bank characteristics. Fernandez et al. (2017) and Flannery et al. (2017) look at this link but find different results. Whilst the former fails to find any empirical evidence to support the existence of such a relationship, the latter concludes that market responds more strongly to riskier and/or high leveraged banks. The findings of this paper, therefore contribute to this unsettled debate.
Finally, this paper attempts to investigate whether the market cares about the details of the results, i.e. whether banks with higher capital excess experienced higher abnormal stock returns, under the adverse economic scenarios or just simply looks at the final pass-fail result. With the exception of Perestin et al. (2009), all other existing literatures ignore the effect of the detailed results revealed by stress tests on market reactions. Perestin et al. (2009) finds out that among the banks that were deemed to fail SCAP 2009, banks with larger capital shortfall experienced larger negative abnormal returns.
6 To answer the research question, this paper first employs the standard event study methodology which has been widely used by other empirical papers in the existing literature, on daily stock returns. The event study investigates whether there are significant abnormal stock returns for stress tested BHCs around the announcement and result publication of the stress tests. Due to the potential contagion effect and common economic exposure, the market may also draw information from those test results on non-stress tested banks. As a result, those banks’ stock prices may as well be affected; hence they are also included in the study. The findings suggest that the market indeed reacts to the stress test results and abnormal stock returns exists in both groups. That means the stress tests are useful in informing the market about the financial health of both the tested banks as well as the whole banking system. After that, the cross-sectional regressions are run to examine the relationship between stock market responses and bank’s financial conditions, which are approximated using indicators of the CAMELS rating system. Compared to Fernandez et al. (2017) and Flannery et al. (2017) who have a similar approach, my model incorporates more banks’ characteristics than just riskiness, opacity and liquidity used in their paper, hence hopefully is more comprehensive. Finally, in order to study how the market reactions to the stress test results are affected by the detailed result of the test, the paper also use cross-sectional regressions to find out the relationship between abnormal stock return and Projected Core Equity Tier 1 (CET1). The indicators found to significantly impact abnormal stock return in the previous regression will be used as control variables in this regression. However, as there is a lack of statistical significance, the paper cannot come to any conclusion regarding the relationship investigated. The findings of this paper might hopefully be relevant to the US financial regulators. The main purposes of stress tests since they were first implemented in 2009 are to reduce banks’ opaqueness and to restore market confidence in banking system through the result disclosure. One way to gauge the market perception of the usefulness of the test is to observe whether the market reacts to the test results. If the market does not react, it may signal that the tests may not reveal as much information about banks as the market desires. In that case, financial regulators may revisit their test conditions and framework or more extreme, put an end to the tests as carrying out such tests are costly. However, my paper finds evidence that the market does response in most cases, implying that the stress tests are indeed useful and should continue to be implemented.
The rest of this paper is organized as follows. In the Literature Review, I will review existing literature on macroprudential stress test in general and its information value in particular. Based on that, I will develop the main hypotheses for the thesis. Next, I will present the
7 methodology and data. After that, empirical findings on market reactions are presented and discussed. The final section concludes.
2. Literature Review
2.1. Overview of the stress test
The disastrous financial crisis in 2008 called for financial regulators to promptly come up with an effective supervisory tool in order to prevent a future crisis from happening. Supervisory stress test was the answer to that. Although stress test has only recently gained attention from the public, it has a history that dated back to 1996 following the Internal Rating Based Approach for Capital Requirements under Basel II. However, up until 2008, it was still only conducted at individual bank level. The financial crisis 2007- 2008 marked the birth of stress test at a system-wide level which is often referred to as the macro stress test.
According to Hirtle (2009), “a purely microprudential” stress test meaning that each bank conducts its own internal risk management in isolation from other banks, cannot assure a stable banking system, for which the financial crisis 2007-2008 demonstrates itself as a convincing evidence. Before 2007, banks were required by Basel II to conduct the stress test as part of their internal risk management to serve for their business plans. When being assessed independently and separately, banks might appear healthy, in the sense that they have enough capital to absorb losses and carry out their lending functions normally. However, the banking system as a whole might be vulnerable to adverse economic conditions. Japanese banks in 1990s serve as a prime example for this as they were individually strong but weak as a whole in the face of real estate shock (Goodhart, 2004). This is mainly because microprudential supervisory stress tests fail to take into account the effect of common exposure (i.e. the interconnectedness between financial institutions) and the fact that the systematic risk is determined by “collective behavior” of all financial institutions, including both those subject and not subject to regulations (Papademos, 2009). Hirtle (2009) claims that a macro stress test can help address these problems as its aim is to mitigate the systemic risk and thus strengthen the whole financial system as well as the real economy.
Goldstein and Sapra (2013) point out four major points that differentiate a macro stress test from a normal stress test conducted prior to the financial crisis 2007-2008. First, in a macro stress test, The Central Bank and/or any relevant authorities establish a set of hypothetical economic conditions applied to all participating banks. This lends itself to a consistent assessment across all the banks. Second, while under a micro stress test, a bank was assessed
8 based on its past performance (e.g. realized gains or losses); under the macro test, the projected losses and capital ratios over the forecast horizon in the event of economic downturn are calculated to determine whether banks hold sufficient capital survive and continue lending to businesses. This suggests that the macro stress test is forward looking. Another distinct property of macro stress test is that it focuses on left tail risks. The test covers extreme adverse economic scenarios which might happen with small probabilities and thus “puts relatively more weight on bad states”. Most importantly, while traditional stress test results are kept confidential, bilateral between individual banks and the Central Bank; under macro stress test, banks are required to publish the results. The Federal Reserve claims that the disclosure of test results plays an important role in providing the market with valuable information, mitigating opacity and enhancing “market discipline”.
Supervisory Stress tests are mainly implemented in two different approaches which are top-down and bottom-up. Top-top-down approach implies that the supervisors are in charge of performing the test, applying their own models on the data submitted by individual banks. The other type is bottom-up in which each banking organization uses its own internal risk models to evaluate its viability when facing with adverse economic scenarios designed by supervisors.
2.2. The US Stress test
2.2.1. Supervisory Capital Assessment Program
Supervisory Capital Assessment Program (SCAP) is the first macro stress test conducted by Federal Reserve in the US. Nineteen US banks, each of which had year-end 2008 total assets of greater than USD 100 billions and all of which in total accounted for two-thirds of all US banking organizations’ total assets, took part in the test (Board of Governors of the Federal Reserve System, 2009). SCAP examines the resilience of banking organizations in two different economic scenarios known as the baseline and the adverse scenarios. The former assumes that macroeconomic factors would follow expectations while the latter assumes a longer and more severe recession than expected. Out of those nineteen banks, ten failed to maintain a capital level above the required level under both of the two hypothetical economic conditions; and thus received the “Fail” grade. These banks were then required to raise additional capital from private market. Should any of them fails to do so, the government would intervene by injecting capital through Capital Purchase Program of the Troubled Assets Relief Program (TARP).
9 SCAP was conducted at the height of the financial system’s instability and investor confidence reaching the lowest level due to the financial crisis. It was an attempt of financial regulators to restore market confidence in the US banking system. Proposed by the former Secretary of the Treasury, Geithner Timothy to clear the market doubt on the true conditions of financial institutions, a system-wide stress test did not initially stand out as a promising solution among a number of proposals, but later was proven to be the most effective in ameliorating the crisis (Dugan, 2010).
2.2.2. CCAR Comprehensive Capital Analysis and Review
Following the success of SCAP 2009, Comprehensive Capital Analysis and Review is implemented starting from 2011 until now as part of the Dodd-Frank Act.
The Dodd-Frank Act came into effect on July 21, 2010. Under this act, the largest BHCs which hold at least 50 billion dollars of total assets are required to perform annual supervisory stress tests under three economic scenarios, namely baseline, adverse and severely adverse scenarios. Apart from that compulsory system-wide stress test, qualified BHCs are also obliged to carry out their own stress tests under the three scenarios issued by the Federal Reserve and two scenarios developed by the banks themselves.
CCAR comprises of two major parts, namely quantitative and qualitative assessment. All BHCs with total consolidated assets being greater than 50 billion dollars are eligible for the quantitative tests. Under this quantitative assessment, supervisors aim to examine whether these BHCs hold sufficient capital to act as a buffer for losses arising from adverse economic conditions. Also, due to the fact that after the financial crisis, many financial institutions want to resume dividend payment and common share repurchase, the quantitative tests involve assessing capital distribution plans of banks as well. If a bank manages to maintain an amount of capital above the required level even after implementing its capital plans, it can be considered passing the quantitative part of CCAR.
Initially, all participating BHCs are subject to both components of CCAR but from 2017, qualitative assessment only applies to BHCs under supervision of LISCC (Large Institution Supervision Coordinating Committee) or large and complex BHCs. The qualitative tests involve evaluating BHCs’ analysis and others internal processes for capital planning, “focusing on the areas that are most critical to sound capital planning - namely, how a firm identifies, measures, and determines capital needs for its material risk - and a firm’s controls and governance around those practices.” (CCAR 2017: Assessment Framework and Result, 2017).
10 The results of CCAR which take into account both quantitative and qualitative assessment come down to whether the Federal Reserve approves or rejects banks’ capital distribution plans. The Federal Reserve may reject a plan either because of quantitative reason (i.e. capital ratio falls below the minimum regulatory level) or qualitative reason (i.e. there are shortcomings in capital planning process). In the case of rejection, banks are not allowed to go ahead with their initial plans and can implement only the parts that receive approval from the Federal Reserve. Sometimes, the Federal Reserve gives a conditional non-objection for banks’ capital plans. If that is the case, banks are allowed to revise their capital plans and re-submit them at a later date at the discretion of the Federal Reserve. If the revised plan fails to address the weaknesses of the original one, the Federal Reserve will restrict banks’ capital distribution.
2.2.3. Dodd Frank Act Stress Test (DFAST)
Since 2013, apart from the existing CCAR, qualified banking organizations have been required to participate in another stress test with the introduction of Dodd-Frank Act Stress Test (DFAST).
CCAR and DFAST are distinct but very closely related and complementary to each other. Both use the same economic scenarios issued by the Federal Reserve but DFAST is generally more quantitative compared to CCAR.
Despite several differences, both stress tests aim to measure the degree to which BHCs can survive and continue their lending business in the event of adverse conditions. Also, the results obtained through DFAST are used as inputs for CCAR.
2.3. Related Empirical Papers
Although stress tests have been a subject of interest for academics since it was first introduced, there has been a limited number of empirical papers conducted to investigate the market reactions to the disclosure of stress test results. Also, the empirical findings drawn from these existing papers have not been very conclusive. Nonetheless, some studies have found empirical evidence supporting the view that market can obtain valuable information from the test results.
Petrella and Resti (2013) collect data on stock prices of banks participating in the 2011 EU-wide stress tests to investigate if the information from the test results were relevant to
11 investors. After finding out that markets did react to the results revealed, they came up with two hypothesis, namely “zoom hypothesis” and “stress hypothesis” and examined which caused the stocks’ abnormal returns of stress tested banks. The authors ultimately accept both hypotheses. This can be interpreted as an indication that market reactions were driven by the tests’ outputs, including both indicators on historical data of banks (zoom hypothesis) and indicators on viability of banks under stress scenarios (e.g. capital ratio) (stress hypothesis). Ellahie (2012) observes the movement of several indicators including equity prices, short and long-term bond prices and bid-ask spreads of tested banks to explore the implications of the announcement of the 2011 EU-wide stress tests and the result publication for Information Asymmetry (IA) and Information Uncertainty (IU) of tested banks. He does not find any evidence for the effect of stress test announcement on both IA and IU. However, he arrives at the conclusion that the publication of the test results significantly reduces IA but it comes at the cost of increasing IU.
Morgan et al. (2014) employ the standard event study technique to assess if there are abnormal stock returns resulting from the disclosure of SCAP results for tested BHCs. The findings of this paper suggest that investors have already identified banks which would fail the tests even before the results are published. However, investors still obtain valuable information from the failed banks’ capital gaps disclosed. The larger the capital gap, the more negative the abnormal stock returns are.
Carboni et al. (2017) has a similar approach to that of Morgan et al. (2014), looking at the unexpected or surprise component of the test results. The paper focuses on the Comprehensive Assessment (CA) in the EU. The finding is that the CA was successful in enhancing the market’s transparency. This is because the market was able to identify weak banks at the time of the announcement and the CA’s results were valuable as they unveiled capital shortfall. Neretina et al. (2015) look at how a variety of indicators including stock prices, credit risks and systematic risks evolve in order to investigate the effect of different events of the US stress tests, i.e. announcement and publication of the tests from 2009 to 2013. With the exception of SCAP 2009, the US stress tests had a weak effect on equity prices. However, Credit Default Swap (CDS) spread as a proxy for credit risks experienced a significant decrease following the publication of test results. A similar finding is also found for systematic risk, which is approximated by the market beta. Overall, the findings of this paper are in line with the view that stress tests contain valuable information to the market.
12 Candelon and Sy (2015) also use event study method to study how market acts in response to the publication of stress test results, but the paper provides a more thorough view by including and comparing the stress tests from 2009 to 2013 in both the US and EU. In the case of the US stress tests, SCAP 2009 resulted in a significant and positive abnormal stock returns for tested banks. However the impact of the following stress tests diminished gradually and SCAP 2009 is often considered the most successful test. This finding is not surprising considering the objectives and background of SCAP. It is also consistent with others empirical papers on US stress tests, including Neretina et al. (2015). Meanwhile, for EU-wide stress tests, the publication of 2011 test results had a significant negative impact on the banks’ stock returns while the two stress tests in 2009 and 2012 both produced positive market reactions reflected by positive abnormal stock returns. The authors attribute the variance in market reactions to the difference in the design and governance of each stress test. Furthermore, this paper also examines the reaction in stock market for untested banks around the result dates but with the exception of the 2011 EU-wide stress test, there is no evidence for the existence of such reaction.
Flannery et al. (2017) observe the movement of stock prices and trading volume of both tested and non-tested banks around the disclosure date of the US stress tests conducted in 2009-2015. Applying event study method, the authors detect significant absolute abnormal returns and absolute trading volume of both groups of BHCs. Therefore, they come to the conclusion that stress tests provide market with valuable information not only about BHCs participating in the tests but also about the whole banking system. However, the information value appears to depreciate after SCAP 2009. Furthermore, the paper finds that the disclosure of stress test results produces stronger market reaction for riskier and more leveraged BHCs.
Fernandes et al. (2017) also finds the empirical evidence to support the view that the US stress tests carried out in 2009-2015 contained new information about tested BHCs and banking industry. However, the paper fails to prove that the effect of stress test disclosure is stronger for riskier and more opaque BHCs as there is a lack of statistic significance.
2.4. Hypothesis Development
In order to answer the research question “How does the stock market react to the US stress test conducted in 2015-2017?” the following three hypotheses are developed:
Hypothesis I: The stock market does not react to the US stress tests conducted in 2015-2017
13 Under the semi-strong form of the Efficient Market Hypothesis, the market is assumed to be efficient and will incorporate all the public information into valuing stocks, hence stock market prices are truly reflective of all the relevant available information. Hence we would expect the stock market to react to the announcement and the result publication of stress tests only if those events carry new and valuable information about banks. Fernandez et al. (2017) suggest that market reacts to the announcement of stress tests as the economic scenarios issued by the Federal Reserve for each stress test contain new information for the market.
In this paper, I would like to examine if those stress tests do indeed contain meaningful information about banks by testing if abnormal stock returns exist as a result of stress tests. Under the first hypothesis, the paper investigates the stock market reactions to two events - namely the announcement and the result publication of these stress tests. Due to the potential information contagion and common exposure, the paper also looks at the movement of untested banks’ stock prices around the event dates. Benoit et al. (2017) claim that there is an information link between banks if the market interprets the failure of a bank as the signal for the failure of other banks. Dasgupta (2004) argues that banks are connected through interbank lending and thus, negative news about one bank can adversely affect other banks linked to it. The bankruptcy of Lehman Brother in 2008 demonstrates how the information contagion and/or common exposure could lead to the virtual collapse of the banking industry. Information contagion therefore suggests that the results of stress tests might not only be bad/good news to the tested banks but also to other banks in the industry. Therefore, it is expected that there would be significant abnormal stock returns for untested banks as well. This view is supported by Fernandez et al. (2017) and Flannery et al. (2017), both of which found the significant market reactions to untested banks. Flannery et al. (2017) argues that another possible explanation is that stress tests reveal information regarding the systematic risks, which are relevant to the whole banking industry.
To test the aforementioned hypothesis, my paper employs the standard event study methodology which is widely used in the existing literature.
Hypothesis II: The stock market reactions to the stress test result are not affected by banks’ current financial health.
According to Flannery et al (2017), as stress tests become common practices and the result date are publicly known, their information value must be assessed with the market’s prior
14 beliefs about the banks’ financial conditions being taken into account. This view is supported by Peristiani et al. (2010) in which the market is found to significantly respond to the unexpected gap in the capital shortfall of failed banks in SCAP 2009. Before the result was released, market already had its own estimation of the capital shortfall that banks would experience. Later, when the results were published, the market only reacted if there was a significant difference in the expected shortfall and the actual shortfall revealed by the stress tests.
The US stress test conducted in 2015-2017 had good outcomes overall with almost all the participating BHCs passing the tests. However, this is not necessarily good news for all banks from the market perspective. Following the idea of Flannery et al. (2017), the market already had its own belief about whether banks would have a high chance of passing the test. Their belief is mostly based on banks’ current financial conditions. Therefore, for a healthy bank passing the test, there might be no news or less good news if the market already expected that bank to be able to pass the test. In other words, stress test results did not reveal new information to the market and thus there is no adjustment in its belief. In contrast, passing the test would be considered surprisingly good news for unhealthy banks which are believed by the market to have a high chance of failing the test.
In order to test the second hypotheses, the paper uses the cross-sectional regressions, using a similar model to that of Flannery (2017). However, the paper hopefully offers a more comprehensive analysis as the model includes indicators that capture banks’ conditions in various dimensions based on the CAMELS rating system.
Hypothesis III: The stock market reactions to the stress test result are not affected by banks’ projected Core Tier 1 Ratio (CET1)
Under this hypothesis, I would like to test if the market does care about the detailed results of the stress tests or simply take into account only the final outcome whether banks fail or pass the tests. This is because a healthier bank which passes the test more easily may be rewarded more by the stock market than a bank that just narrowly passes the test. The third hypothesis can also be tested using the cross-sectional regression. All the indicators that found to have significant impact on market reactions will be included as the control variables in the regression.
3. Methodology
15
3.1.1. The determination of the event and the estimation period
Following most of the papers in the existing literature, in my thesis I will employ the standard event study methodology to detect abnormal stock returns resulting from the announcement and the publication of the test results of the US stress tests. Event study is one of the most commonly used techniques in research especially in the field of finance. Event study was first introduced by Fama (1969). Before abnormal stock returns can be estimated, some elements of a standard event study such as the event and the estimation period must be defined.
Since 2013, CCAR and DFAST have been conducted in parallel with the results of DFAST being released around one week earlier than that of CCAR. Therefore, from 2015 to 2017, there was a total of six US stress tests (i.e. 3 DFAST and 3 CCAR) carried out by the Federal Reserve. In my thesis, I will investigate the impact of all these 6 stress tests on stock market returns. The table below provides details including names of tested BHCs, the announcement date, the result publication date and the overall results for each stress test conducted over the past three years.
There are two important estimation periods, namely estimation window and event window that need to be determined. Fernandes et al. (2017) claim that choosing an event window length involves facing a dilemma: a short one might fail to capture the entire abnormal return while a long one might pick up abnormal return resulting from a different event. Following Petrella (2013) and Flannery et al. (2017), this paper takes a 3-day event window (t-1, t+1) with t being the event date. Morgan et al. (2014) also claims that such a short event window would work well for an event like the result publication. With this event window, the estimation will include the abnormal stock returns one day before the disclosure date in case there is information leakage and one day after the event date in case market responds slowly to the results. The normal stock returns are estimated in a 120-day estimation window (t-150, t-30) as suggested by McKinlay (1997). The estimation window and event window are 30 days away from each other, preventing the estimation of normal returns from being influenced by the abnormal return around the event date. The figure below demonstrates the 3-day event window and the 120-day estimation window ending 30 days before the event date.
16
3.1.2. Model
In the next step, normal return which is defined as the return that a stock would actually have should the event investigated did not happen is estimated. There is a number of models that can be used to estimate normal return but the most popular ones are the three-factor model developed by Fama and French (1993) and the market model. Following Petrella and Resti (2013) and Fernandes et al. (2017), in this paper I will use market model.
Ri,t = α + β1 Rmi,t + εi,t
Ri,t is the daily stock return of BHCs
Rmi,t is daily market return proxied by S&P 500 index
The model assumes a linear relationship between market returns and stock returns. Regressing individual banks’ stock returns on market returns during the 120-day estimation window would give us β, which is then used to predict the normal returns in the 3-day event window. Next, abnormal returns can be calculated by subtracting predicted returns from actual returns for each day covered in the event window. Basically, abnormal returns are the residuals of model (1).
ARi,t = Ri,t – α – β1 Rmi,t
After that, statistical tests and sign tests will be employed to check the statistical significance of abnormal returns. A lack of statistical evidence implies that there are no market reactions to the disclosure of stress test results and vice versa.
CAAR is calculated as
CAAR = (CARi /T)
Where T is the number of days in the event window [t-1, t+1] which is three in this study. The null hypothesis that the event has no impact on abnormal returns is tested for the sample as a whole. A standard t-test using robust standard errors is used to test whether the CAARs are significantly different from zero at a 5% level (Coutts, 1994):
H₀: CAAR = 0 versus H₁: CAAR ≠ 0, where ARi,t ̴N(0, σ² εi)
17 After confirming the existence of abnormal stock returns following the release of stress test results, this paper aims to investigate whether the market reactions depend on certain bank characteristics. The Cumulative Abnormal Return (CAR) obtained from the event study are regressed on banks’ financial indicators. Following Flannery et al. and (2017), the model is as follows:
is the cumulative abnormal return of Bank i following the result disclosure of the stress test in Year t.
is the financial indicators that reflects bank’s financial condition of bank i in Year t
First, CAR will be regressed on each indicator in univariate regressions. After that, based on their significance in univariate models and the correlations among them (this will be presented in Data description section later), some indicators will be included in the multivariate model.
As mentioned in the Hypothesis Development, I will use CAMELS rating system to capture the characteristics that can classify good and bad banks. CAMELS rating system was first introduced by the US financial regulators in 1979 and since then has been a subject of intense debate on whether it is really effective in predicting financial institutions’ distress and failure. Despite that, recent studies have been in favor of CAMELS. A number of papers focusing on the US market find evidence that the combination of CAMELS with some factors such as banks’ internal risk management and audit quality works well as an early distress warning tool for financial institutions. (Betz et al., 2014). CAMELS has 6 letters with each of them representing a different dimension. Following Flannery (1998), I use the following indicators as proxies for CAMELS.
C as in Capital Adequacy is represented by Tier 1 Capital Ratio (calculated as Tier 1 Capital over Risk Weighted Assets) and Equity to Asset ratio. A bank with higher capital ratios is more likely to be a healthy bank and to have a higher chance of surviving adverse economic conditions.
A which implies Asset Quality can be proxied by Return on Average Assets (ROAA) and Loan Loss Reserve Ratio (LLR). ROAA measures how much profits a bank can generate from its assets and thus, a higher ROAA is often associated with a lower probability of default. Loan Loss Reserve Ratio is calculated as loan loss reserve divided by gross loan. Banks use LLR to cover uncollected debts resulting from defaults or late payments, thus LLR
18 provides useful information regarding the quality of bank’s loans. A high LLR suggests that bank is highly concerned about the collectability of it loans.
Costs to income ratio (CTI) can be used to assess the Management (M) of a bank. The numerator of this ratio is the non-interest expenses exclusive of bad debt expenses. This cost is considered the easiest one to put under control and most sensitive to the management actions among all types of banks’ costs (David Tripe, 1998). An efficient management can lead to a decline in non-interest expenses, which in turn decreases CTI.
Earnings (E) are proxied by two ratios, namely Return on Equity (ROE) and Net Interest Margin (NIM). A high value of these two ratio indicates a healthy bank.
L indicating Liquidity is represented by short-term borrowings as percentage share of total liabilities and liquidity assets over short-term deposits and borrowings ratio. The higher these ratios are, the higher the chance that a financial institution will fail is.
Finally, S which implies the Sensitivity to market risks was added to CAMELS in 1998. Share of trading income can be used as the proxy for it, however due to the ambiguous impact of this ratio on bank’s financial performance, the model will not include it and only takes into account the indicators representing CAMEL.
Overall, all the indicators that have positive effects on bank’s profitability and viability are expected to negatively correlated with market reactions and vice versa.
4. Data and descriptive statistic
4.1. Data
The paper requires data from multiple sources. First, in order to calculate abnormal stock returns around the disclosure date, daily stock returns of the US bank holding companies and daily market index are required. There are a total of 31, 33 and 34 BHCs participating in the US stress tests conducted in 2015, 2016 and 2017, respectively. However, I remove the US branches of foreign BHCs from the sample since their stock data is not available. The resulting sample hence consists of 25 BHCs in 2015 and 2016 and 26 BHCs in 2017. As the paper also investigates if the US stress tests provide the market with relevant information about the whole banking system, it is necessary to obtain data on stock returns of non-tested BHCs. The untested BHCs sample covers 41 largest US BHCs (as of 2016 year end) apart from those joining the stress tests. Stock returns of both group of BHCs can be extracted from
19 Datastream (Thomson Reuters). The market index used as benchmark in this paper is S&P 500 index and it is retrieved from CRSP (WRDS).
Second, in order to test the second hypothesis regarding the relationship between a range of BHCs’ characteristics and market response, I take the data on financial indicators of BHCs from Osiris database. Finally, the stressed Core Tier 1 capital ratios (i.e. the capital ratio under simulated economic conditions in stress test) used for the third hypothesis come from the “Assessment Framework and Result” Report published by the Federal Reserve for every stress test.
4.2. Descriptive Statistic
Table 1provides an descriptive statistic for the indicators used as independent variables in the cross-sectional regressions. It can be seen from the Table that stress tested BHCs hold more assets than non-stress tested companies. It is not surprising considering that stress tests are conducted on the largest BHCs. Regarding Tier 1 Capital Ratio, tested BHCs have higher ratio than untested one.
Table 1: Descriptive statistic of bank’s financial data
All Tested Non-tested Difference
p-value Mean St.Dev. Mean St.Dev. Mean St.Dev.
Total Assets 219.21 8.51 522.83 8.43 20.19 9.01 0.00 Tier 1 Ratio 12.54 2.31 12.97 1.67 12.18 2.68 0.04 MarketCap/Total Asset 15.97 8.08 13.95 8.43 17.32 7.58 0.004 Equity/Assets 11.90 2.67 11.65 2.09 12.07 2.99 0.29 ROAA 1.04 0.51 1.04 0.66 1.04 0.26 0.96 LLR 1.12 0.44 1.26 0.61 1.03 0.26 0.00 CTI 63.20 13.25 66.04 11.97 61.36 13.76 0.02 ROE 3.11 1.56 3.06 2.03 3.15 1.17 0.68 NIM 3.17 1.25 2.81 1.76 3.40 0.68 0.02 Interest Expense/TotalLia 0.12 0.09 0.15 0.13 0.99 0.04 0.05 LiquidAssets/Dep&Borrowing 13.43 20.72 24.28 29.24 6.38 5.52 0.00 STFund/TotalLia 4.12 3.79 3.89 3.16 4.28 4.21 0.51
Looking at the paper by Fernandez et al. (2017) which focuses on the US stress tests conducted from 2009 to 2015, on average tested BHCs have improved their capital positions relative to untested ones. However, with the exception of ROAA and ROE, there is statistical evidence that untested bank holding companies were having better financial performances overall even though the differences are not extremely noticeable.
A high correlation between some pairs of variables can easily be spotted. This is understandable considering that some indicators are used as the proxies for the same category
20 and/or have an overlapping component in their measurement such as ROE and Net interest margin (NIM). In order to avoid the multicollinearity problem which would lead to the imprecision in the estimation of at least one of the coefficients (Watson, 2012), the correlations among variables would be given careful consideration in the process of choosing explanatory variables for the multivariate model.
5. Results and analysis
In this section, the results will be presented in turn following the order of the hypotheses developed in the previous section.
5.1. Event study
First, I will report the results obtained from the event study and conclude whether there are market reactions following the announcement and the result publication of the US stress tests.
5.1.1. Market reactions to the announcement of stress test
Table 2 reports the abnormal stock returns of both tested and untested BHCs after the announcement of the stress tests. As can be seen from the table 2 there are significant and positive
Table 2 – Stock market Reaction around the announcement
Cumulative Abnormal Return
Tested Untested Differences
2015 -0.4 (0.21) -0.35 (0.38) -0.05*** (0.015) 2016 1.04* (0.23) 1.18** (0.27) 0.14*** (0.008) 2017 0.56** (0.41) 0.61** (0.35) 0.05*** (0.012) Standard errors are in parentheses
* Significant at the 10-percent level. ** Significant at the 5-percent level. *** Significant at the 1-percent level.
abnormal returns resulted from the announcement of the stress tests in 2016 and 2017. This finding can be explained by “Transparency” theory proposed by Petrella (2013) in which the author attributes the abnormal stock returns to the rising anticipation of market participants about the enhanced transparency regarding tested BHCs. The compulsory disclosure of stress test results has been an important component in the US stress test frameworks. Even though the results would not be released until approximately three months after the announcement,
21 the market is assured that the results will be published because of strict regulations imposed on a standard stress test.
On the other hand, the abnormal stock returns after the announcement of the 2015 tests are negative but insignificant, which implies that market did not respond to the announcement of the stress tests in 2015. One possible explanation is that in 2015, there were thirteen more banks required to take the tests compared to 2014. Hence the market may be await for further information and clarity rather than responding immediately to the announcement. It is possibly one of the shortfalls of the event study methodology that the reaction happens outside the event window investigated.
Looking at the abnormal returns of untested bank holding companies, it is noticeable that they have the same significance and sign as the abnormal returns of tested companies. This can be seen as an indication that the market perceived stress tests as a tool to improve the transparency of the whole banking system. A similar conclusion can be found in Fernandez et al (2017) where the market reactions to the stress test announcement of tested and untested banks are in the same direction.
5.1.2. Market reactions to the publication of stress test result.
Moving onto the next event date which is the release of stress test results. The cumulative abnormal returns following the result disclosure for tested and untested BHCs are reported Table 3.
Table 3 – Reaction around the result disclosure
Cumulative Abnormal Return
Tested Untested Differences
DFAST 2015 2.49*** (0.34) 2.9*** (0.22) 0.41*** (0.012) CCAR 2015 2.09*** (0.36) 2.28*** (0.33) 0.19*** (0.015) DFAST 2016 -0.99*** (0.25) -0.13 (0.24) 1.12*** (0.014) CCAR 2016 -0.23 (0.24) 0.21 (0.25) 0.44*** (0.01) DFAST 2017 1.97*** (0.14) 2.098*** (0.18) -0.12*** (0.015) CCAR 2017 5.29*** (0.22) 5.68*** (0.23) 0.39*** (0.02) Standard errors are in parentheses
* Significant at the 10-percent level. ** Significant at the 5-percent level. *** Significant at the 1-percent level.
There are a total of six disclosure dates corresponding to six stress tests in three years. With the exception of DFAST 2016, the results of all other stress tests appear to contain relevant
22 and valuable information for market participants. It is reflected by the statistically significant abnormal stock returns around the disclosure dates.
Both DFAST and CCAR 2015 had significant positive impact on the stock returns of tested and untested bank holding companies. The result is similar to that of both Fernandes et al. (2017) and Flannery (2017) regarding not only the sign but also the magnitude of the cumulative abnormal return, despite the differences in the model and the estimation period used to calculate the normal returns. This helps confirm the robustness of the results of the event study in this paper.
The positive market response to the results of the 2015 DFAST and CCAR could be explained by the improvement in the overall results of the 2015 stress tests compared to those of previous years. In 2014, there were five BHCs that had their capital distribution plans rejected by the Federal Reserve. Among them, four were deemed to have weaknesses in their plans (qualitative reason) and one had the capital ratio below the required level (quantitative reason). In 2015, there were only two banks being rejected for their capital plans and both were due to failing the qualitative assessment.
The way the stock market behaved in response to the disclosure of two stress tests in 2016 is not as simple to interpret. Considering the favorable results of 2016 DFAST and CCAR, one would expect the stock market to react positively but following 2016 DFAST, the abnormal stock returns of tested banks were slightly negative while the figures for untested banks are not statistically significant. This could be possibly explained by the fact that despite an overall good outcome, nineteen smaller banks performed worse in 2016 DFAST compared to previous years as their minimum Core Tier 1 ratio decreased by 80 basis point on average. Also, apart from real estate loans, the loss rates of various types of loans increased for the first time ever since the stress test began in 2009 (Five key points from Stress Test, Ryan, 2016, PwC).
Regarding 2016 CCAR, the result appears to be uninformative to the market since the event study could not find the significant abnormal stock returns for both groups of banks around the result publication. This might be partly because the market already expected that no banks would fail the quantitative assessment of CCAR considering the result of DFAST (PwC report, 2016).
Finally, the stock market reacted positively to the results of both DFAST and CCAR 2017 and the magnitude of the reactions was also the highest among the three investigated years. This is
23 not surprising since the 2017 test results are considered the best among all the tests since 2009. Furthermore, the 2017 stress tests were probably perceived as highly credible by the market as the adverse scenarios were even stricter than those in 2016 and put a heavier emphasis on real estate by including a sharp decline in real estate price. The combination of the good results and the increasing credibility of the tests led to a significantly positive market reaction following the result publication.
While it is difficult and beyond the scope of this paper to explain the different reactions of the stock market to the results of these six stress tests, there is enough evidence to conclude that they contain new information to market participants. When the results were released, the market adjusted its prior belief based on new information, producing abnormal stock returns. This also holds true for untested banks. With the exception of DFAST 2016, the stock market reactions towards tested and untested BHCs were similar. Hence, the stress tests seem to have revealed some broader information about the whole banking system. This finding is consistent with Flannery et al. (2017) and Fernandez et al. (2017).
In conclusion, the first null hypothesis that the stock market did not react to the US stress tests in 2015-207 can be rejected.
5.2. Cross-sectional Analysis
After confirming that the stock market indeed responds to the results of stress tests, the paper now will look at the relationship between banks’ financial conditions and market reactions. Table ? provide the results obtained from cross-sectional regressions for two samples of tested and untested bank holding companies, respectively. The dependent variable is Cumulative Abnormal Return (CAR) of individual banks around the publication date of the US stress tests in 2015-2017. The explanatory variables are financial indicators that can partly differentiate a good bank from a bad one. The regressions take into account year fixed effects and standard errors are clustered at bank holding company level.
24 Table 4: The result of cross-sectional analysis for tested banks
(1) (2) (3) (4) (5) (6) (7) (8) (9) Capital Tier 1 Ratio -0.15** (0.12) -0.29** (0.14) Equity/Assets 0.19 (0.21) Asset LLR Ratio 0.19 (0.27) 0.18 (0.31)) ROAA -0.39** (0.25) Management Cost to Income 0.009 (0.01) Earning ROE -0.095*** (0.08) -0.17** (0.09) Net Interest Margin 0.05 (0.08) Liquidity STFund/TotalLia 0.027 (0.58) 0.17** (0.42) LA/STDep&Borr -0.005* Observations 152 152 152 152 152 152 152 152 152 Adjusted R2 0.158 0.161 0.161 0.20 0.18 0.178 0.16 0.178 0.183 Year FE
Standard errors are in parentheses
* Significant at the 10-percent level. ** Significant at the 5-percent level. *** Significant at the 1-percent level.
In Table 4, each column from (1) to (8) reports the coefficient and statistical significance of an indicator specified in the Methodology section. It can be seen that 7 out of 8 indicators have their expected signs. However, only 4 of them - namely Tier 1 Ratio, over Assets ratio, ROAA, ROE and Liquidity Assets over Short term Deposits and Borrowings are found significant. It is noticeable that with the exception of Management (M), all dimensions of CAMEL have at least one indicator that significantly affected the stock market behaviors around the publication of stress test results.
Moving onto column (9) of Panel A, the coefficients and statistical significance of the indicators included in the multivariate regression are reported. The multivariate regression has four independent variables, each of which represents a category of CAMEL except for Management (M).
When deciding which indicators to be included as the explanatory variables in the multivariate regression, I take into account both their significance in the corresponding univariate regressions and the correlations among these indicators. Costs to Income as the
25 proxy for Management is not included to avoid the multicollinearity as it is found to be highly correlated with the indicators covered in Earnings (E). For Asset quality (A), Loan Loss Reserve Ratio (LLR) is selected instead of ROAA despite the statistical significance of ROAA’s coefficient because it is also highly correlated with Earnings indicators. Under Capital Adequacy (A) and Earnings (E), Tier 1 Ratio and ROE are included because of their high significance. Finally, among the three ratios representing Liquidity (L), only Liquidity Assets over Short-term Deposits and Borrowings ratio is significant but due to its high correlation with Tier 1 Ratio, Short-term funding to Total Liabilities ratio is selected instead. The results of multivariate regressions are comparable to the univariate ones. The coefficients of all independent variables still get the expected signs. Tier 1 Ratio and ROE remain significantly negative which suggests that the stock market indeed reacted less strongly to the passing result of a well capitalized and profitable bank. The coefficient on Loan Loss Reserve Ratio is positive as expected in both the univariate and multivariate models but it is not statistically significant. Finally, the coefficient on Short-term funding to Total Liabilities ratio is still positive but becomes statistically significant in the multivariate model. This means that a less liquid bank experienced a more positive stock return than a liquid one following the release of stress test results.
After the multivariate regression is ran, VIF (Variance Inflation Factor) test is employed and it confirms that the model has a low chance of suffering from the multicollinearity problem. F-test is also conducted and the null hypothesis that the coefficients of all explanatory variables are equal zero are strongly rejected.
All in all, it can be concluded that the reactions of the stock market to the passing result of a bank partly depend on the financial conditions of the bank at the time. The market believed that a healthier bank would have a higher chance of passing the stress tests so it did not react as strongly as to the “passing” news of less healthier banks that surprisingly passed the test. It can also be inferred that the results of stress tests contain relatively more valuable information about a bad and/or risky bank than a good one. This finding is consistent with Flannery et al. (2017).
Following a similar process to that of tested bank sample, the paper finds that the result of stress tests appears to have less strong impact on stock returns of untested banks with high Tier 1 Capital Ratio as well. However, the relationship between the stock market responses and banks’ financial conditions is less significant compared to the case of tested banks. It can also be seen that other independent variables are not statistically significant and some even get
26 the opposite signs. This is reasonable given that that the results of stress tests do not have a direct effect on untested BHCs.
The second null hypothesis that the stock market reactions to the stress test results are not affected by banks’ financial conditions is thus rejected.
Table 5 reports the results of the cross-sectional regressions used to test Hypothesis III. Table 5: The result of cross sectional model with CET1 as independent variables.
(1) (2) CET 1 -0.05 (0.12) CET1 (with control variables) 0.09 (0.21)
Standard errors are in parentheses
* Significant at the 10-percent level. ** Significant at the 5-percent level. *** Significant at the 1-percent level.
The independent variable is CAR of individual tested banks. The explanatory variable is the Projected Core Tier 1 Ratio revealed in the stress test results. The year fixed effect is taken into account and standard errors are clustered at BHC level. As can be seen from column (2), the coefficient of CET1 is negative and not statistically significant. Column (3) reports the results after including control variables in the regression. They are variables that are found to significantly impact CAR under Hypothesis II. The coefficient of CET1 becomes positive, which makes sense since banks with higher CET1 ratio are considered more sustainable under adverse economic scenarios and thus market will react more strongly to these banks. However, the coefficient is not statistically significant so the null hypothesis III cannot be rejected.
6. Robustness Check
In this section, the results from robustness check will be presented. The main finding of this paper is obtained from the event study so first, I will check the robustness of the event study. The 120-day estimation window is shorten to 90 and 60-day windows with everything else being kept constant. Table 6 reports the results of the event study with 90-day estimation widow (t-30, t-120) and 60-day estimation window (t-30, t-90), both ending thirty days before the event window.
Table 6
Cumulative Abnormal Return
Tested Untested Differences
DFAST 2015 2.64*** (0.31) 3.31*** (0.36) 0.78*** (0.01) CCAR 2015 2.13*** 2.51*** 0.38***
27 (0.36) (0.33) (0.013) DFAST 2016 -1.44*** (0.21) 0.15 (0.23) 1.24*** (0.007) CCAR 2016 -0.09 (0.26) -0.48 (0.25) 0.44*** (0.012) DFAST 2017 1.91*** (0.2) 2.07*** (0.18) -0.12*** (0.015) CCAR 2017 5.29*** (0.21) 5.64*** (0.16) 0.39*** (0.011) Standard errors are in parentheses
* Significant at the 10-percent level. ** Significant at the 5-percent level. *** Significant at the 1-percent level.
It can be seen that for every stress test conducted in 2015-2017, the magnitude and significance of the abnormal stock returns obtained from the event study with shorter estimation windows are comparable to that of the original event study with 120-day estimation window. Next, in order to seek for further confirmation on the robustness of the event study, the length of event window is increased to five days (t-2, t+2), which means two days before and two days after the event date. This longer event window is taken to make sure that the original event study does not miss out any abnormal stock returns.
Table 7 reports the results of the event study with five-day event window and 60-day estimation window ending 30 days before the event. The adjusted event studies result in the same sign and significance for the cumulative abnormal stock returns following the result publication of all the six stress tests. However, the magnitude of the abnormal returns slightly increased. All in all, it
Cumulative Abnormal Return
Tested Untested Differences
DFAST 2015 2.64*** (0.41) 3.05*** (0.38) 0.41*** (0.13) CCAR 2015 2.3*** (0.38) 2.31*** (0.43) 0.01*** (0.08) DFAST 2016 -1.5*** (0.21) -0.22 (0.23) 1.28*** (0.007) CCAR 2016 -0.41 (0.26) -0.88 (0.25) 0.48*** (0.009) DFAST 2017 2.02*** (0.2) 2.26*** (0.18) 0.24*** (0.17) CCAR 2017 5.21*** (0.21) 5.38*** (0.16) 0.17*** (0.14)
can be concluded that the event study with 3-day event window and 120-day estimation window is robust against the changes in the estimation period.
The second hypothesis is tested using panel data with the explanatory variables being bank’s financial indicators which are usually subject to outliers. The outliers if exist can cause
28 serious problems to the estimation of the coefficients in the model. To deal with the problem, I winsorized all the financial indicators. Winsorizing assigns all outliers’ values to a specific limit of the distribution, which will make it closer to a normal distribution. I looked at the summary statistics as well as histograms of all the different ratios to observe whether and at what level they should be winsorized. I observed the distance between each ratio’s mean and median, and between its minimum and maximum values in order to decide if winsorizing is necessary. Furthermore, I examined the frequency distribution of each ratio after winsorizing at different levels, in order to determine which level makes the distribution resemble a normal one the most. Through trial and error, I decided to winsorize all the ratios at the 99th percentile.
Following the same process reported in the Methodology sections, nine univariate regressions corresponding to nine indicators will be ran first. After that, four variables will be picked as independent variables in the multivariate models. The result shows that the coefficients of all the explanatory variables have the same signs and statistical significance as in the regressions running on the original data.
Regarding the multicollinearity problem, as explained in the previous section, any two variables having the absolute value of correlation greater than 0.3 are not included together in the multivariate regression. Also with the Variance Inflation Factor (VIF) test as reported in the Result section, it is reasonable to claim that the multivariate regressions do not suffer from the multicollinearity problem. Overall, it can be concluded that the cross-sectional regressions in the paper are robust and therefore the findings are reliable.
The findings of this paper might hopefully be relevant to the US financial regulators. The main purposes of stress tests since they were first implemented in 2009 are to reduce banks’ opaqueness and to restore market confidence in the banking system through more transparent result disclosure. One way to gauge the market perception of the usefulness of the test is to observe whether the market reacts to the test results. If the market does not react, it may signal that the tests may not reveal as much information about banks as the market desires. In that case, financial regulators may revisit their test conditions and framework or more extreme, put an end to the tests as carrying out such tests are costly. However, my paper finds evidence that the market does response in most cases, implying that the stress tests are indeed useful and should continue to be implemented.
29 First introduced in 2009 with the aim to restore market confidence on the banking system after the financial crisis, since then the US macro stress test has become an important banking supervisory tool. In recent years which can be considered a tranquil time for the financial system, it is suggested by a number of papers that stress tests have a weaker impact on the market compared to the precedent ones. This is understandable considering that stress test has become more and more established and in a stabilized environment, hence there are less needs for addressing uncertainty.
This paper investigates how the stock market reacted to both the announcement and the result disclosure of the US stress tests conducted in 2015-2017. First, by employing the standard event study methodology, it finds out that there are significant abnormal stock returns for both tested and untested bank holding companies following seven out of nine events. The market indeed responded to both the announcement and the result of these stress tests. This suggests that these stress tests revealed valuable and relevant information about the tested BHCs and also the banking industry as a whole probably due to the information contagion. This finding is in line with several papers including Fernandez et al. (2017).
Next, the paper studies the relationship between the stock market reactions and banks’ financial conditions. Using a comprehensive range of indicators based on CAMELS rating system to reflect banks’ financial health, the cross-sectional regressions produce reliable findings which confirm the existence of such a relationship. Indeed, the market already had their own prediction on which banks would pass the test based on their financial performance before the result was even released. A good bank passing the stress test would be considered less good news compared to a worse bank surprisingly passing the test. Therefore, the market would react less strongly to the good banks. It can also be said that these stress tests reveal more information about unhealthy banks. This finding is consistent with Flannery et al. (2017).
Finally, the paper examines whether the market would care about the detailed results of the stress tests, i.e. to what extent a bank passed or failed the stress test. In order to test this, the paper regressed each bank’s cumulative abnormal return resulting from the result publication of each stress test on that bank’s projected CET1 Ratio. However, since the regression result is statistically insignificant, it suggests that the market only reacted to the final outcome whether the banks passed or failed the tests, but not the detailed results.
The robustness check was conducted for both the event study and the cross-sectional regressions. The event study results remain consistent regardless of the length of the event
30 window and the estimation window. The cross-sectional regressions have a low chance of being subject to outliers and multicollinearity problem. Despite that, the paper still has a number of limitations. First, the US subsidiary of some foreign BHCs that were subject to the US stress tests conducted in 2015-2017 are not included in this study as their stock data is not available. Second, there might be other variables that affect a bank’s financial condition and CAMELS indicators might not be able to fully reflect it. Third, there might be omitted variable problems for the multivariate regressions.
The findings of this paper might hopefully be relevant to the US financial regulators. The main purposes of stress tests since they were first implemented in 2009 are to reduce banks’ opaqueness and to restore market confidence in banking system through the result disclosure. One way to gauge the market perception of the usefulness of the test is to observe whether the market reacts to the test results. If the market does not react, it may signal that the tests may not reveal as much information about banks as the market desires. In that case, financial regulators may revisit their test conditions and framework or more extreme, put an end to the tests as carrying out such tests are costly. However, my paper finds evidence that the market does response in most cases, implying that the stress tests are indeed useful and should
