Bank stress testing and market confidence : a comparison between Europe and the United States

(1)

Bank Stress Testing and Market Confidence

A Comparison Between Europe and the United States

Sophia Rozemarijn Borra

15 July 2015

Supervisor: Dr. Kostas Mavromatis

E-mail: Sophie.borra@student.uva.nl

(2)

Statement of Originality

This document is written by Student Sophie Borra 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

The purpose of this study is to provide useful insight into the effectiveness of stress testing exercises. Bank stress testing became an important tool to restore confidence in financial markets after the recent financial crisis in 2008. The regulatory community uses these stress tests to provide market participants with valuable information about the opaque banking system by openly disclosing stress test approaches and results. This study measures the actual effect of stress testing exercises in Europe and the United States (US) on market confidence through analysis of different confidence indicators. An event study approach as well as regression models are used to measure the effect. The findings reveal that stress test events have a significant impact on market confidence, which can be either positive or negative. In both Europe and the US publications of test results usually helped to restore confidence in the financial system, however the effect was less positive when stress tests were announced. Especially in the US, announcements of stress test exercises resulted in ambiguous or even negative effects.

(4)

1. Introduction

After the recent financial crisis in 2008, stress testing became an important tool to restore market confidence in both the US and Europe. The relatively new stress testing technique has been increasingly used by countries’ authorities to regain the trust of investors, and other market participants, in the financial system. This study aims to provide useful insight into the actual effectiveness of stress testing and thereby focuses on the market confidence aspect of the tests. The effectiveness of stress testing is investigated by testing whether stress test events have a significant impact on market confidence in both Europe and the US. In addition a comparison between the effects in both countries is made. Based on these results the main research question is answered: What is the effect of bank stress testing on financial market confidence?

The study is based on two different methods to enhance robustness of the results. First an event study is conducted to test the effect of the events. Afterwards the same effect is estimated by using portfolio and panel regression models including event time dummies. Different confidence indicators are used to measure changes in financial market confidence. Bank stock returns function as one indicator of market confidence, and additionally the effect of stress testing on bank and sovereign Credit Default Swaps (CDS) spread changes is assessed. Both methods use daily panel data covering listed banks in Europe and the US. In addition, a panel consisting of different European countries is used to estimate the effects on sovereign spreads in Europe. This European sovereign sample is split up to be able to address different effects for different European regions.

The findings of this study reveal that stress testing has a significant impact on market confidence through both the announcement of the stress testing exercise as well as the publication of test results. In most cases publication of the results led to an improvement of market confidence, which is supported by the findings of Candelon and Sy (2015) in the current literature. Although, it appears that effects are not always positive as some stress test events resulted in a downturn of financial market confidence. Hence, to be able to prevent a negative outcome of the stress testing exercise, different factors have to be taken into consideration when conducting the tests. For example, stress test design and macroeconomic and political factors have been shown to have a significant impact on the actual effect of the test (Ong and Pazarbasioglu, 2013).

(6)

2

The results of this study are highly relevant for policy makers and country authorities as it shows that stress testing can be an important tool for crisis management. Most research on stress testing in the current literature is focused on analysing the design and execution of the stress tests and their importance for effectiveness (Borio et al., 2012; Greenlaw et al., 2012; Hirtle et al., 2009; Jobst et al., 2013; Ong and Pazarbasioglu, 2013). However, empirically investigating the actual effect of stress testing on market confidence is rarely performed, or if so, only focused on one stress test event. Since this study analyses the effects of all different stress tests conducted in Europe and the US after the crisis in 2008, it significantly contributes towards new information on the topic of stress testing.

The finding of this study are usually supported by anecdotal evidence and current literature. Bernanke (2013) stated in his speech that the stress test of 2009 as part of the Supervisory Capital Assessment Program (SCAP) in the US provided investors with credible information about banks’ financials and that the public disclosure of test results helped to restore confidence in the banking system. Research by Greenlaw and others (2012), Morgan and others (2013) and Ong and Pazarbasioglu (2013) showed that the SCAP stress test in 2009 indeed revealed new information, which improved investor sentiment. However, the effects of subsequent stress tests undertaken in Europe and the US were less clear and convincing. The results of this study confirm the positive effect of the publication of results of the SCAP test in the US and in addition show that the announcement of the test had a negative impact. Furthermore, this study provides some important insight into the effects of the subsequent stress tests. Research by Candelon and Sy (2015) shows that the publication of the European stress test results in 2011 had a negative effect on bank equity. This is in line with the results provided by this study when CDS spread changes are also taken into account. These differences in effects underline the statement by Jobst and others (2013) that the increased use of crisis stress testing creates a trade-off for financial institutions and supervisors. The fact that several stress test events resulted in a negative effect on market confidence shows the importance of finding a balance between enhancing transparency and avoiding to unduly alarm the markets and thereby creating self-fulfilling prophesies (Jobst et al., 2013).

The following structure is maintained throughout this study. Section 2 describes the different types of stress testing and how they can contribute to restoring market confidence. In section 3 the effects of stress testing on market confidence in both Europe and the US are tested by using both an event study approach and regression models. Daily data on bank stock returns and bank and sovereign CDS spreads is acquired from Datastream and used to test a

(7)

3

Market Model. In section 4 the results are analysed and the effects in Europe and the US are compared. Section 5 contains the discussion of the methodology and the outcomes and finally, section 6 concludes.

2. Literature Review

2.1 The Development of Stress Testing

The International Monetary Fund (IMF) concluded after its Financial Stability Assessment of Iceland in August 2008 that the system was resilient. This shows an example of the general messages provided by stress tests conducted all over the globe before the most recent financial crisis. The fact that the worst financial crisis in history could evolve, even though the testing mechanism predicted the financial system to be sound and resilient, highlights the importance of improvement of the policy framework for financial stability (Borio et al., 2012). Bank stress testing is one important aspect of this policy framework and developed rapidly over the past ten years.

Four different types of stress tests developed over time to evaluate the resilience of the financial sector, and which can be distinguished based on their ultimate objectives. The first stress tests, originating from the mid-1990s, had a purely micro perspective and were conducted by individual financial institutions themselves as an internal risk management tool. These stress tests were only used as an internal input for business planning and did not disclose any results to the public. From these original stress tests, microprudential or supervisory stress testing evolved, which was used by supervisors to assess the health of individual financial institutions as a part of the Basel II framework (IMF, 2012). According to the Basel Committee for Banking Supervision (BCBS) (2012) these tests were increasingly used to set capital requirements and explicit buffers for specific banks or to limit capital distribution. Basel III and Solvency II (which is the insurance regulation in Europe) use this kind of stress testing as a fundamental part of their regulatory framework for liquidity (BCBS, 2012).

The lessons learned from the crisis led to the development of macro stress testing which also became a tool to restore market confidence in the US and Europe (Čihák et al., 2012; Henry and Kok, 2013; Jobst et al., 2013). These tests with a more system wide perspective are used as a tool to measure the extent to which the financial system as a whole is capable of withstanding external shocks (Borio et al., 2012). The IMF was one of the first institutions to launch stress tests with a macroprudential perspective following the Asian

(8)

4

financial crisis in the late 1990s. This crisis showed that instability within the financial sector can spread fast and widely, thereby triggering a new focus on systemic risks and a system-wide stability perspective. Afterwards, the recent financial crisis in the US and Europe deepened the use of macro stress testing by using it not only for surveillance purposes but also as a tool for financial crisis management (Henry and Kok, 2013). Hence two different types of these macro stress tests can be distinguished based on their main objectives.1

The first macro stress tests were organized by the IMF, central banks and other macroprudential authorities for surveillance and system-wide monitoring purposes. The aim of these tests is to reveal the sources of systemic risk in addition to institution-specific risk and to identify potential threats to overall stability. This should lead to a reduced likelihood of costs spreading to the real economy as a result of overall financial distress (Jobst et al., 2013). According to Acharya et al. (2014), this can be achieved when the macro stress tests ensure that the financial system is sufficiently capitalized in order to continue financial intermediation when an economic downturn hits the industry. This implies that stress tests should evaluate the, usually closely related and interdependent, solvency and liquidity performances of financial institutions (IMF, 2012). However, the focus on liquidity stress testing only intensified after the recent financial crisis as this event showed the substantial costs at which neglecting liquidity risk comes, whereas solvency regulations already received a lot of attention under the influence of the BCBS in the recent decades before the crisis. The main problem was however, that the BCBS did not take into consideration that banks became more reliant on short-term wholesale funding, especially in interbank markets. So when the crisis erupted and uncertainty about the solvency of different banks increased, the wholesale funding market dried up resulting in liquidity problems for many banks (Hesse et al., 2012). This led to regulatory responses after the crisis and the requirement of higher liquidity buffers. In addition, recent efforts are made to include liquidity risk, funding risk, counterparty risk and vicious fire sales in bank stress tests (Henry and Kok, 2013). Ultimately, this will lead to a reduction of the costs of fire sales, credit crunches and systemic defaults on the overall economy (Greenlaw, 2012).

The second type of macro stress tests developed after the recent financial crisis as a tool for crisis management (Ong and Pazarbasioglu, 2013). The first stress test with the main purpose of crisis management was conducted in 2009 by the authorities of the US in the form of the SCAP (Schuermann, 2014). Later the European Banking Authority (EBA) also

(9)

5

conducted these kind of stress tests in Europe. These crisis stress tests can be used as an input for bank recapitalizations and business restructuring plans (IMF, 2012) but their main objective is to regain public’s market confidence and trust. Crisis stress tests can in fact be used for the same purposes as the surveillance stress tests. Both can be used for the determination of adequate capital buffers, to quantify potential fiscal costs or to test the soundness of banks and the system. However, as these crisis stress tests are part of the overall strategy to rebuild confidence in the banking system, executing these tests should reduce uncertainty about banks’ balance sheets through widespread transparency and public disclosure of test results and further implications (Ong and Pazarbasioglu, 2013).

2.2 Crisis Stress Testing in Europe and the United States

After the success of the SCAP test in the US, the Committee of European Banking Supervision (CEBS) started to carry out crisis stress tests in Europe as well in order to restore confidence in financial markets. The first EU-wide exercises took place in 2009 and 2010 and stood in strong contrast to the SCAP exercise. At the beginning, these tests were conducted against the background of increased concerns about a sovereign debt crisis in Southern European countries. Therefore exposures to sovereign debt were incorporated into the stress tests (Schuermann, 2014). Furthermore, coordination of the stress testing procedures in Europe is more complex in comparison to the US due to the large number of different countries involved in the tests. Four different types of authorities participated in the execution of the EU-wide tests including all different national supervisory authorities, whereas the US benefits from a common national framework (Ong and Pazarbasioglu, 2013). In addition, a higher level of heterogeneity can be expected among European banks since they have no common backstops or fully consistent accounting rules (Candelon and Sy. 2015). This can make the interpretations of results for Europe as a whole less straightforward.

After the two crisis stress tests conducted by the CEBS, the EBA became responsible for conducting the subsequent stress tests in Europe, all with the main objective of crisis management. These tests featured a higher degree of disclosure during the whole stress testing process. In the US annual supervisory stress tests were incorporated into the broader Comprehensive Capital Analysis and Review (CCAR) program. The CCAR requires large banks to submit their own capital plans annually, which are based on the results of their own base and stress scenario, as well as the ones from the common supervisory stress scenario. This approach reveals the vulnerabilities and sensitivities of individual bank portfolios and

(10)

6

provides supervisors with a structured and comparative way to assess the banks’ ability to manage risks (Schuermann, 2014).

2.3 Stress Testing and Market Confidence

Ong and Pazarbasioglu (2013) found in their research that stress tests should fulfil multiple criteria to be credible and thereby able to positively influence market confidence. The SCAP stress test was initially met with scepticism as it was announced in the darkest-days of the sub-prime loan market crisis. However, eventually markets were reassured and investor sentiment recovered and stabilized. This success of the SCAP stress test was due to the fact that it fulfilled different criteria which improved its credibility. First, the public dimension of crisis stress tests requires the test to be designed in such a way that its elements will survive intense investigation.2 Additionally, the scope of coverage and scenario design should be rigorous enough to make the results convincing and other activities should be undertaken to support the stress test results and thereby enhance credibility. The state of the political economy is another element that might influence the effect of the stress test given the potential economic and reputational implications of the findings of the test. Ultimately, Ong and Pazarbasioglu (2013) address the importance of country authorities’ full commitment when undertaking a crisis stress test. This means that the authorities should have a clear and transparent objective, be able and willing to conduct a total examination of their banking system and be willing to undertake follow-up actions when necessary. Executing stress tests will only be able to improve market confidence when the authorities have sufficient resources to back the financial institutions. Completing crisis stress tests can otherwise even result in a loss of market confidence with a potential damaging effect for the overall real economy.

To investigate whether stress tests are capable of restoring confidence in financial markets different variables can be used as proxies for financial market confidence. First, changes in bank stock returns are used as a confidence indicator. This variable is a bellwether indicator for financial market confidence as shareholders are the first investors to bear the losses of default. Furthermore, the fact that new publicly available information is incorporated into equity prices at a rapid rate, makes the return on bank stocks a reliable indicator for the market’s assessment of the soundness of the banking system (Ong and Pazarbasioglu, 2013). When investors believe that bank stress testing will create value, their confidence increases which is reflected in the abnormal returns of the bank’s stocks around the event date. Other

2_{Specific elements are: the design, such as the timing of the test, its governance, the objective of the exercise,}

(11)

7

indicators of market confidence are sovereign and bank CDS spreads, which provide information about the creditworthiness of a country or bank respectively. A CDS spread is a periodic rate paid by the buyer of the CDS contract to the seller for transferring credit risk. Therefore, CDS spreads can be seen as reflections of the market’s perception about the health of (financial) institutions (Annaert et al., 2013). A decline in the CDS spread means that the perception about the creditworthiness of the institution increases. From this it follows that an aggregate decline in bank and sovereign CDS spreads can imply an increase in overall market confidence. This view is supported by research conducted by Ferdinand and Sun (2014) who find that changes in CDS spreads are primarily driven by investor sentiment. Furthermore, the effect of stress testing on sovereign CDS spreads is studied since the creditworthiness of a country is considered to be closely linked to the health of its financial sector. This follows from the fact that a collapse within the banking sector usually requires government support and increases default risk. Hence changes in sovereign CDS spreads also provide an indication of confidence in financial markets (Ong and Pazarbasioglu, 2013; Greenlaw et al., 2012).

3. Methodology

The results are obtained through two different methods for robustness reasons. Both an event study as well as regression models including event dummies are used to measure the effect of stress testing on market confidence in Europe and the US.

3.1 Event Study

First the standard event study methodology is used to analyse the effect of the stress testing events on financial market confidence. This methodology was introduced by Fama et al. in 1969 and examines the behaviour of security prices around events (Binder, 1998). The impact of the event is measured in terms of abnormal returns (εi,t) by comparing the actual or realised

return around the event date (Ri,t) with the normal return (RNi,t):3

Ri,t = RNi,t + εi,t

Normal returns are the expected or predicted returns based on a particular model (Kothari, 2006). In this study the Market Model is used to calculate normal returns, which creates the

3_{abnormal returns (ε}_i,t_{) are indicated by AR}_i,t_{in the following sections as ε}_i,t_{will be used to indicate the error}

(12)

8

possibility to isolate the impact of the event from other general market movements (Fama, 1969).

3.1.1 Events

All US- and EU-wide stress tests conducted from 2009 until April 2015 are included in this study.4 Both the announcement and the publication of the test results are considered as events. In Europe four stress tests are examined, two conducted by the EBA and two by its processor: the CEBS. The capital exercise conducted by the EBA in 2011 and 2012 is also included as a European stress test event. Even though this exercise was technically not a stress test it is still important to include this event since it was performed as a part of a series of coordinated policy measures to restore confidence in the European banking sector (EBA, 2012). Furthermore, six stress tests conducted in the US are also included in this study. The first one was the SCAP led by the Federal Reserve (Fed), which tested the largest 19 bank holding companies in the US. In addition, five stress tests performed as part of the CCAR in 2011 until 2015 are included. These tests evaluate the capital adequacy, the capital adequacy processes and capital distributions of bank holding companies.

3.1.2 Estimation Period

The announcement of the stress test and publication of results are all one-day events of which the actual date is defined as t. However, the event period is expanded to be able to capture the full effect of the event (Neretina, 2014). A three-day event window [t-1, t+1], including one day

before and after the event, is used to calculate abnormal returns. This time span incorporates the risk of information becoming available one day before the actual event date and also captures the effects after the market closes due to a slow response by investors (Petrella, 2012). Other event windows can be considered but usually provide very similar results and hence have not been treated (Morgan et al., 2013; Neretina, 2014). The parameters of the normal return model are estimated over an estimation window [t-150, t-30], covering 120 trading

days prior to the event as suggested by MacKinlay (1997). This period is long enough to let the sampling error of the parameters vanish and the abnormal returns to be independent over time. To avoid influence of the returns around the event on the estimated parameters of the normal return model, the event window and the estimation window do not overlap. This allows for the event impact to be only captured by the abnormal returns (MacKinlay, 1997).

(13)

9

The figure below shows an overview of the estimation period with an estimation window ranging from t-150 to t-30 and an event window from t-1 to t+1:

3.1.3 Models and Methods 3.1.3.1 Stock Returns

The standard event study approach described by for example Brown and Warner (1985) and MacKinlay (1997) is followed to analyse the impact of stress testing on stock returns. This method estimates the effect of the stress test announcement and result publication on stock returns under the assumption of rationality in the marketplace (MacKinlay, 1997) and consists of two stages.

In the first stage the normal return parameters of the model, α and β1, are estimated

over the estimation window by Ordinary Least Squares (OLS) using a single index Market Model. This model shows the linear relationship between the return of a given security and a market portfolio by the following formula:

Ri,t = α + β1 Rmi,t + εi,t

E[εi,t ]= 0 and var(εi,t) = σ² εi

Where Ri,t is the observed return on stock i at date t, Rmt is the return on a broad reference

stock market index at date t and εi,t is the error term. Here the return is calculated as the daily

change in the Return Index, assuming that dividends are reinvested.5 The Market Model is the most popular and commonly used model to benchmark normal returns in event studies (Coutts et al., 1994) and provides a reliable measures of returns (Binder, 1998). One of the benefits of this method is that it accounts for the portion of the return that is caused by the variation in market return. By removing this portion from the total return the variance of the abnormal returns is reduced which increases the ability to detect event related effects due to more powerful statistical tests (MacKinlay, 1997). Furthermore, this method produces smaller

5_{The Return Index is calculated by the following formula: RI}_{t =}_RI_t-₁_{* (PI}_t_{/ PI}_t-₁_{) * ( 1 + DY}_t_{/100 * 1/N) where}

RI is the Return Index, PI is the price index, DY is the dividend yield and N is the number of working days

t -30

Estimation Window Event Window Post-event Window

(14)

10

correlations across security abnormal returns which makes standard statistical tests more conform (Strong, 1992). Finally MacKinlay (1997) argues that using economic models to estimate normal returns provides very little gain. For example, results of studies that used the Capital Asset Pricing Model (CAPM) may be sensitive to the specific restrictions that this model imposes. This can easily be avoided by using the Market Model. Also, the gains from using an Arbitrage Pricing Theory (APT) motivated model versus a Market Model are small since the largest explanatory factor in these APT models behave like a market factor and the additional factors have little explanatory power.

In the second stage abnormal returns (ARi,t) are calculated over the event window by

taking the difference of between realized stock return (Ri,t) and the normal returns given the

absence of the event:

ARi,t = Ri,t – E[Ri,t | Ωt]

ARi,t = Ri,t – α – β1 Rmi,t

Where E[Ri,t | Ωt] = RNi,t as Ωt is the conditioning information of the Market Model which is

used to estimate the OLS parameter estimates of the normal return model α and β1.

Afterwards cumulative abnormal returns (CAR) and average abnormal returns (AAR) for each individual firm in the sample are calculated by the following formulas respectively:

CARi =∑1_𝑡=−1𝐴𝑅i,t

AARi = (CARi/T)

Where T is the number of days in the event window [t-1, t+1] which is three in this study.

CARs are calculated over the three-day event window to capture all event related effects. Afterwards the AARs are calculated as a cross-sectional average to isolate the effect of the event as other information, unrelated to the event, cancels out on average (Cardinali, 2011). Also cumulative average abnormal returns for the industry as a whole are calculated by:

AARt = (∑𝑁_𝑖=1𝐴𝑅i,t)/N

CAAR = ∑1_𝑡=−1𝐴𝐴𝑅t = (∑𝑁_𝑖=1𝐶𝐴𝑅i)/N

Where N is the number of firms in the sample. Finally, 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):

(15)

11

3.1.3.2 Credit Default Swaps

To evaluate the effect of bank stress test events on credit risk the same approach as for equity markets is used. Abnormal CDS spread changes are calculated which measure the abnormal change in the premium of a newly issued CDS contract (Andres et al., 2013). This event study method also consists of two stages.

Again the Market Model is used in the first stage, to estimate the parameters for normal spread changes over the estimation window:

ΔSreli,t = α + β1 ΔSindexi,t + εi,t

E[εi,t ]= 0 and var(εi,t) = σ² εi

Where the dependent variable (ΔSreli,t) is the relative change in CDS spreads of entity i at

time t. This value is calculated by taking the percentage change of daily CDS spreads: ΔSreli,t = ln(Si,t) – ln(Si,t-1)

Relative changes are used since it seems more plausible to expect the stress test induced effect on CDSs to be proportional to the firm's initial probability of default and resulting losses (Andres et al., 2013). ΔSindext is the spread change of the CDS market index at time t.

In the second stage abnormal spread changes are calculated by taking the difference between the realized absolute spread change and the normal spread change estimated by using the parameters of the market model:

ΔASi,t = ΔSreli,t – E[ΔSi,t | Ω,t]

ΔASi,t = ΔSreli,t – α + β1 ΔSindexi,t

Where α and β1 are again the OLS estimators from the Market Model reflecting normal spread

changes. Afterwards, the AARs and CAARs are calculated over the event window and the null hypothesis is tested.

3.2 Regression models

Another approach to test whether the event had a significant effect on stock returns and CDS spreads is the use of regression models. Event dummies are included into the standard Market Model to measure the effect of the event.

3.2.1 Stock Returns

In this study the event windows for the different banks overlap as the stress test event affects multiple banks at the same time. This clustering causes securities to be cross-correlated and

(16)

12

therefore the covariance between abnormal returns to be non-zero. This causes the assumption of independent and identically distributed residuals to be violated. To account for this problem abnormal returns can be directly estimated within the return equation (Binder, 1985). This method uses an event time dummy (Dt), which equals one on the days within the event

window (Binder, 1985; Norden, 2004), and includes the effect of the event in the individual return equations. Izan (1980) was the first one to use this method when examining a portfolio of firms that all experienced the event at the same time, and he proposes the following regression model:

Rp,t = αp + βp,1 Rmt + βp,2 Dt + βp,3 Dt Rmt + εp,t

The equally weighted portfolio return (Rp,t) is the dependent variable and βp,2 is the estimator

of the average abnormal returns across the stocks in the portfolio. βp,3 is the coefficient of the

interaction term which allows for different intercepts and different slopes of the regression. This model is estimated by the Prais-Winsten and Cochrane-Orcutt linear regression method, which uses Generalized Least Squares (GLS) and corrects for serial correlation when estimating the parameters of the linear regression model.6

To be able to account for omitted variables that vary across banks but are constant over time the following fixed-effect model is estimated as well by use of the within regression estimator:

Ri,t = β1 Rmi,t + αi + β2 Di,t + β3 Di,t Rmi,t + εi,t

Where αi is the company specific intercept in this fixed-effect regression model.7

3.2.2 Credit Default Swaps

The Market Model assumes that the CDS spread change can be sufficiently explained by common pricing factors (Andres et al., 2013). Therefore the same approach is followed to estimate the effect on CDS spread as for equity returns. The abnormal returns are directly estimated by including binary variables into the portfolio model:

ΔSrelp,t = αp + βp,1 ΔSindext + βp,2 Dt + βp,3 Dt ΔSindext + εp,t

Where the dependent variable (ΔSrelp,t) is the equally weighted portfolio average of the CDS

spread change.

6_{A Durbin-Watson test is used to test for serial correlation. The H₀ : no serial correlation, could be rejected at a}

1% significance level. A Cochrane-Orcutt regression is run to correct for serial correlation where εt = ρ εt-1 + et

7_{A Hausman test was conducted to test whether the model has fixed effects. The H₀ : random effects could be}

rejected at a 5% significance level so H₁: fixed effects in the model is accepted. See table 1 in appendix III for more details

(17)

13

However, different studies have found that there are additional factors affecting CDS spread changes (Ericsson et al., 2009; Collin-Dufresne et al., 2001; Alexander and Kaeck, 2008; Greatrex, 2008). From this it follows that estimating the normal CDS spread change by using a Market Model similar to the one used in the event study is not optimal since these models are poorly specified. Therefore a CDS factor model is constructed based on the findings of previous research. This new factor model includes additional explanatory variables of CDS spreads as control variables to reduce omitted variable problems and is expressed by the following formula:

ΔSreli,t = α+ β1 ΔSindexi,t + β2 lnLeveli,t + β3 lnVolai,t + β4 lnSlopei,t + εi,t

The first part follows from the Market Model where the dependent variable (ΔSi,t) is again the

realized change CDS spreads of entity i and ΔSindext is the spread change of the CDS index

at time t. Additionally, Ericsson and others (2009) found that equity implied volatility and the level of the treasury yield curve are the major determinants of credit spread changes. Therefore the natural logarithms of the level of the risk-free yield curve (lnLevel)and equity implied volatility (lnVola) are included into the Factor Model. Furthermore, equity returns can also be a determinant of CDS spread changes, however as Andres and others (2013) find that equity returns and equity implied volatility are highly correlated, equity returns are excluded from the model to avoid multicollinearity problems. Additionally, Alexander and Kaeck (2008) find in their research that the slope of the risk-free yield curve has a significant effect on spread changes and is therefore included into the model as the lnSlope variable. Finally the event dummy (Dt) and interaction term are added to evaluate the impact of the

event. This results in the following formula:

ΔSreli,t = α+ β1 ΔSindexi,t + β2 Di,t + β3 Di,t Rmi,t + β4 lnLeveli,t + β5 lnVolai,t + β6 lnSlopei,t +

εi,t

This model is estimated by Random Effects GLS regression as the null hypothesis of the Hausman test cannot be rejected at a 5% significance level. Heteroskedasticity-robust standard errors are used in all regression estimations to account for heteroskedasticity.

3.3 Sample and Data

Panel data is used to test the effect of stress testing on daily equity returns and CDS spread changes.8_{The data covers a time period from the beginning of 2009 until May 15}th_2015.

(18)

14

According to Morse (1984), using daily return data is favoured over monthly data since it provides a more powerful test statistic for average abnormal returns. Therefore, banks for which daily data after 2008 is not available, are excluded from the sample. This results in a sample that consists of 149 European and 102 US banks.9 The return on the S&P 500 index, which includes the reinvestment of dividends, is used as a proxy for the US equity market and the return on the STOXX Europe 600 is used as a proxy for the European stock market returns.

The CDS samples consist of the European and US banks for which daily CDS spread data was available. The samples consist of 59 European and 15 US banks. The Markit iTraxx Europe spread is used as the index for the CDS market in Europe and the proxy for the US CDS market is constructed manually by taking the average of 9 different sector indices.10 A sample of 20 European countries is constructed to test the effect of stress testing on sovereign spreads. All countries include banks that underwent at least one stress testing exercise. A division is made between Southern European and North-Western European countries to test for differences in effects of stress testing between those two regions. Different effects might occur due to the fact that these regions are exposed to a disparate level of sovereign risk and market confidence. For the US the spreads on CDSs of the Department of Treasury are used to analyse sovereign spread changes. For all CDS data 5-year senior unsecured CDS spreads are used because these contracts are the most liquid (Jorion and Zhang, 2007). In the CDS Factor Model stock market volatility is measured by the VSTOXX Volatility Index for Europe and the VIX Index for the US. The level of the risk free yield curve is measured by the yield to maturity on 5 year government bonds as this maturity matches the maturity of the CDS contracts used in this study. The yields on German bonds are used as a proxy for the risk free yield level in Europe. The difference between the yields on 10 and 2 year government bonds are used as a proxy for the slope of the curve. All data is obtained from Datastream.

4. Results and Analysis

The results reveal that the event study usually provides similar results to those acquired from the regression analysis, since they both show significant effects for the same stress test

9_{For European banks the list G#LBANKSER in Datastream is used, 3 missing important banks were added}

manually. The US banks are selected from the S&P1500 using the LSPSGBNK code in Datastream, 9 missing important banks were added manually

10 The US sectors on which the average is calculated are: banks, consumer goods, electric power, energy

(19)

15

events.11_{Furthermore, the signs of significant results are usually consistent, being either}

positive or negative under all methods and the orders of magnitude of the effects tend to be similar as well. Results are excluded from the analysis when it is ambiguous whether the significant effect is positive or negative. Moreover, results are also not included in the analysis when only the interaction term is significant since the interaction term does not reveal whether the effect is positive or negative. Additionally, interaction effects are too small to change the sign of the effect reported by the dummy coefficient.12

Stress testing is considered to restore market confidence when a positive and significant effect on equity returns and negative significant effects on CDS spreads are observed. The significant and consistent effects for Europe and the US are discussed below, after which a comparison between the results for the two regions is made.

4.1 Europe

Results for the stress tests conducted in Europe are presented in table 2 below and show that all stress tests had a least some effect on market confidence through either a change in equity returns or CDS spreads.13 A plus indicates a positive significant effect whereas a minus shows a negative effect. Moreover, n/a indicates that no significant effect was observed or that the result is excluded from the analysis due to contradicting outcomes obtained via the different methods.

The results show that the announcement of a stress test usually has the same effect as the publication of the results of the preceding stress test. The effects of the individual stress test events are discussed and analysed below.

11_{A complete overview of the results can be found in appendix V}

12_{With average Rm and ΔSindex being 0.00062 and -0.00065 respectively for Europe and 0.00066 and -0.00092}

respectively for the US

13_{These are aggregate results acquired from all three different methods. See appendix VI for an overview of how}

the aggregate results are obtained

Table 2 - Significant effects of stress test annoucement and publication of results in the EU

Ann. Pub. Ann. Pub. Ann. Pub. Ann. Pub. Ann. Pub.

Bank equity - + + + + - - + n/a n/a

Bank CDS spreads + - - - - + + n/a + n/a

Sovereign CDS spreads + n/a - - - + + - +

West & Northern Europe + n/a - n/a - + + - +

Southern Europe & IRL n/a n/a - - - + + n/a n/a n/a

Ann. = Stress Test Announcement Pub. = Publication of Results

(20)

16

4.1.1 CEBS Stress Testing

The announcement of the first EU-wide stress test conducted by the CEBS in 2009 had a negative effect on market confidence whereas the publication of results had a positive effect. The negative effect of the announcement can be explained by the fact that this test was the first EU-wide exercise in the context of risk-assessment of the aggregate banking system and investors had no past experience with the results or effects of these kind of tests. Therefore it might be plausible that investors expected a negative outcome, especially as this test was conducted only one year after the Lehman Brothers collapse, which resulted in a large amount of uncertainty and instability within the financial system and economy as a whole. The publication of the results of this test was able to raise market confidence to some extend as abnormal returns on bank equity increased and bank CDS spreads declined. This increase in confidence can be explained by the fact that the publication of results revealed resilience of the banking system. It showed that the financial positions of banks were sufficient enough to maintain an adequate level of capital under adverse economic conditions. The aggregate Tier 1 ratio remained above 8% and no individual Tier 1 ratio dropped below 6% as a result of the adverse scenario (CEBS, 2009). Furthermore, it was stated in the results that Ministers and Governors intended to closely monitor the financial sector and would react appropriately if necessary. This fulfils the criterion of country authorities being fully committed for the stress test to be credible. The insignificant impact of the test results on sovereign CDS spreads can be due to the fact that individual bank results were not disclosed, which might have caused some uncertainty about sovereign risk related to the banking sector.

The stress test conducted in 2010 was highly effective at restoring market confidence, both through its announcement and publication of results. Equity returns increased and CDS spreads decreased for both the banking industry as well as for European countries. It is noteworthy that this test had a significant effect on sovereign CDS spreads through both its announcement and publication of results since this was not the case for the previous test in 2009. A possible reason for this result could be the fact that this test incorporated the exposure to European sovereign debt (CEBS, 2010). The positive effect of the announcement on restoring market confidence could be explained by the statement that results would be publicly disclosed. This public disclosure of results aligns with the test’s transparency requirement in order for it to be credible and therefore able to influence market confidence. Furthermore, the positive results of the stress test in 2009 could have also contributed to the increase in confidence when investors believed the 2010 stress test to cause the same effect. The publication of results increased confidence as well, which can be assigned to the fact that

(21)

17

the aggregate results show the banking system to be strong and resilient. Specific follow-up actions were taken to support close contact between national supervisors and the individual institutions that failed the test. Furthermore, bank recapitalisation plans were assessed by the national supervisors. The increased amount of supervision could contribute to the increase in confidence in the banking industry. The decrease in sovereign CDS spreads can be explained by the decision to create European bailout funds which ruled out the possibility of a Greek sovereign debt default (Candelon and Sy, 2015). In addition, the fact that the positive test results only significantly affected Southern European CDS spreads could be related to the higher perception of sovereign default risk in those southern countries. The positive test results could have provided the market participants with new, credible information.

4.1.2 EBA Stress Testing and the Capital Exercise

The announcement of the stress test conducted by the EBA in 2011, also had a positive effect on market confidence since bank equity returns increased and all CDS spreads decreased. Again, this result could be explained by the positive effect of the publication of results in 2010. However, the publication of results had the opposite effect, a decline in market confidence, which agrees with previous research (Candelon and Sy, 2015) and is consistent with what was observed in financial markets. The first trading day after the publication of results showed a drop in the Euro Stoxx 600 Equity index of nearly 2% with bank stocks decreasing the most among all sectors.14_{Furthermore, increased sovereign risk was observed}

since 10-year Italian and Spanish bond yields reached their highest level since the introduction of the Euro (Candelon and Sy, 2015). This negative effect could have been caused by a decrease in stress testing credibility due to the bailout request by Ireland even though the 2010 stress test indicated their financial system to be resilient. However, the positive effect from the announcement does not reflect this downfall of credibility (Ong and Pazarbasioglu, 2013). Another explanation for the negative effect on Southern European CDS spreads could be the increased probability of sovereign defaults in Greece as yields on Greek 10 year bonds rose by 5.81 percent point between publications of results in 2010 and 2011. The overall deterioration of market sentiment about sovereign risk and the fear for potential spillovers to the banking sector are reflected in a widening of bank and sovereign CDS spreads. Finally, the decrease in confidence could be due to the lower capital adequacy threshold of 5% Tier 1 ratio versus 6% in 2010 and the disclosure of highly detailed

14_{The price level of the Euro Stoxx 600 decreased from 266.91 on the 15th of July 2011 to 262.10 on the next}

(22)

18

individual bank data. This data availability allowed market participants to come up with their own recapitalization estimations which could easily turn out to be more negative than actually appropriate (Candelon and Sy, 2015).

The Capital Exercise is different from previous stress tests as it was a recapitalization exercise, without macroeconomic scenario and only focused on sovereign risk. Again investors based their expectations on previous experiences as the announcement of the Capital Exercise resulted in a decrease in market confidence which is similar to the effect of the publication of results of the preceding stress test. This loss of confidence could also be a result of the significant restructuring which systemic banks Dexia and Bankia required after passing the 2011 stress test (Ong and Pazarbasioglu, 2013). On the other hand, the publication of results caused a positive effect on market confidence through an increase in bank equity returns and a decrease in sovereign CDS spreads. This positive effect can be explained by the overall increase in the banks’ capital positions resulting from the required temporary buffer of 9% Core Tier 1 capital ratio that banks had to reach by June 2012. Moreover, the introduction of the ECB‘s Long-Term Refinancing Operation in December 2011 and the Outright Monetary Transactions in August 2012 further improved market confidence in the whole region. Finally, the EBA undertook strong follow-up actions to ensure that banks did not make strategic use of the accumulated capital and hence retained the ability to deal with unexpected losses. This contributed to the credibility of the Capital Exercise (EBA, 2012).

The stress test conducted in 2014 was the least effective of all the European-wide stress tests over the past six year since it only affected CDS spreads. Bank and sovereign CDS spreads increased as a result of the announcement of the test, reflecting a decrease in market confidence. This result is not in line with previous findings, which suggest that the effect of the announcement is similar to the effect of result publication of the previous test. The negative effect could be a result of the global market turmoil at the time of the announcement. During this time investors were concerned with hidden problems on bank balance sheets and officials stated that short-term pain was necessary to rebuild a healthy and stable European economy (Ewing and Thomas, 2014). The publication of results, however, had a positive effect on market confidence through its decreasing effect on sovereign CDS spreads. The results showed that banks made significant progress in strengthening their capital positions, which reduces the risk of financial institutions requiring government bailouts (EBA, 2014). The publication did not have any effect on Southern European and Irish CDS spreads, nor did it significantly affect bank equity and CDS spreads.

(23)

19 4.2 United States

Results for the stress tests conducted in the US are presented in table 3 below and show that all stress tests had at least some significant effect.

4.2.1 The SCAP Stress Test

These results show that the SCAP stress test in 2009 was the most successful stress test as it was able raise abnormal equity returns and decrease CDS spreads through publication of its results. This is in agreement with findings in previous literature such as the finding that the 2009 stress test provided investors with credible information about the opaque banking system (Candelon and Sy; 2015; Hirtle et. al., 2009; Morgan et. al., 2013; Neretina et. al., 2014; Ong and Pazarbasioglu, 2013) and with Bernanke’s statement in 2013 that “the public disclosure of results helped to restore confidence in the banking system” (Bernanke, 2013). The fact that the SCAP stress test had the clear objective of crisis management whereas the subsequent CCAR tests had a mainly supervisory objective, could have caused the difference in effectiveness among the tests. The announcement of the SCAP was, however, not able to raise market confidence as it had no effect on bank equity returns and sovereign CDS spreads, but did significantly increase bank CDS spreads. This negative effect on market confidence could be assigned to the fact that the stress test was announced in a period of high uncertainty and lack of confidence in the financial system, making investors suspicious about the outcome of the test. However, the success of the publication of results of the SCAP can also be assigned to the fact that the test was conducted in this period of high uncertainty. Investors had high perceptions of tail risks before the result publication. The results showed that capital needs were manageable under an adverse scenario which led to the reassurance of market participants, an increase in bank equity returns and a decrease of CDS spreads (Tarullo, 2010). Furthermore, the high level of transparency and disclosure of details through the whole stress test procedure made results credible. Finally, the effective communication strategy adopted by the Fed and the fact that the Treasury stood ready to provide capital to any SCAP

Table 3 - Significant effects of stress test annoucement and publication of results in the US

Ann. Pub. Ann. Pub. Ann. Pub. Ann. Pub. Ann. Pub. Ann. Pub.

Bank equity n/a + - n/a - + + + + - - +

Bank CDS spreads + - - - + - + n/a n/a n/a + n/a

Sovereign CDS spreads n/a - n/a n/a + n/a n/a n/a n/a - - n/a

Ann. = Stress Test Announcement Pub. = Publication of Results

2015 CCAR

(24)

20

bank with capital needs resulted in a significant increase in credibility and market confidence (Candelon and Sy, 2015).

4.2.2 CCAR Stress Testing

The results of the CCAR stress tests conducted in the years after the SCAP were less convincing. The announcement of the CCAR in 2011 had an ambiguous effect on confidence as it led to a decrease in both bank equity returns and CDS spreads and does not reflect the same effect as the results publication of the SCAP. The publication of results in 2011 only reduced bank CDS spreads but had no effect on equity returns or sovereign spreads. The reason for the small and ambiguous effect could be the low level of disclosure compared to the previous SCAP stress test. No firm-specific results were disclosed and no information about the base or stress scenario was provided. This made direct comparisons between banks and an overall assessment of financial sector stability impossible (Schuermann, 2014; Fed, 2011).

Again, the announcement of the 2012 CCAR stress test did not reflect the previous stress test results as it caused a decrease in equity returns and a widening of all CDS spreads. This significant deterioration of market confidence could be an effect of negative result expectations since the announcement revealed that individual bank results would be published, in contrast to the previous exercise. Market participants could have based their negative expectations on the fact that this stress test took the European situation into account, including potential severe price movements in European sovereign and financial markets (Fed, 2011). Another contribution to the downfall of confidence could have been the more severe stress scenario, as the adverse growth shock increased by 1.1 standard deviation compared to the previous scenario in 2011 (Ong and Pazarbasioglu, 2013).15 On the contrary, the actual publication of results caused an increase in market confidence through higher equity returns and a decrease of bank CDS spreads. The positive effect could be due to increased transparency as the results revealed that the majority of US banks would still meet supervisory expectations for capital adequacy under an extremely adverse scenario.

The effect of the announcement of the CCAR test in 2013 is ambiguous since both bank equity returns and CDS spreads increased. No significant effect on sovereign CDS spreads was found and the announcement effect again does not reflect the results from the previous exercise. The publication of results had a small positive effect through an increase in

(25)

21

bank equity returns as the results showed that financial institutions substantially increased their capital since the first stress test in 2009.

The announcement of the stress test in 2014 had the same impact as the publication of results in 2013, namely a significant increase in bank equity returns. The effect of the publication of results is ambiguous, as it led to a decrease of both equity returns as well as sovereign CDS spreads. The decrease in CDS spreads could be explained by the ongoing accumulation of bank capital and the statement made by the Fed that this trend was expected to continue. However, the decrease in equity return cannot be explained by this phenomenon. The differences in projections of income loss under the adverse scenario between the banks and the Fed could have contributed to the negative effect on abnormal returns as banks’ had projected a more positive outcome (Hirtle etl al., 2014).

The announcement of the CCAR in 2015 resulted in a significant decrease in equity returns as well as in sovereign CDS spreads, which is similar to the effect of the publication of results in 2014. Moreover, the announcement in 2015 caused bank CDS spreads to widen as well. Therefore the effect of the announcement in 2015 on market confidence is not clear. The publication of results only caused a significant increase in bank equity returns which can be addressed to the better test results compared to those in 2014. The CCAR 2015 revealed that 29 out of 31 banks did not receive an objection to their capital plans in comparison to 24 out of 30 in 2014. Furthermore, the adverse scenario in 2015 resulted in a loan loss rate of 6.1 compared to 6.9 in 2014 and the decline in the aggregate Tier 1 ratio was also less in 2015 (Fed, 2014; Fed, 2015).

The overall decrease in effectiveness of the stress tests conducted after the SCAP test can be explained by the fact that the outcomes of the tests have become more predictable and therefore less informative. Glasserman and Tangirala (2015) found in their research that projected losses for the CCAR tests in 2013 and 2014 were almost perfectly correlated.

4.3 Europe versus the United States

Each stress test event in each country had at least some effect on market confidence. The results show that all European stress tests had a clear and consistent effect, which was either positive or negative. The influence of US stress tests on market confidence is not as clear since the announcements in 2011, 2013 and 2015 and the publication of results in 2014 showed contradictory results for equity and CDS spreads. Furthermore, announcements in Europe usually reflect the outcome of the results publication in the previous year, which could indicate that investors base their expectations on previous test’s results. This pattern cannot be

(26)

22

observed in the US. This outcome is puzzling as the stress tests conducted in the US cover banks in one single country with a common monetary and fiscal policy whereas European stress tests covered approximately 90 banks in more than 20 different countries. Hence more heterogeneity is expected among the European banks, which makes US test results more likely to be consistent.

The results also show that sovereign CDS spreads from the department of Treasury in the US are, in most cases, not significantly affected by the stress testing exercises. Moreover, the effects on US sovereign spreads sometimes differ from the effects on bank CDS spreads which is never the case for European spreads. This difference in consistency of CDS spread movement could be explained by the difference in correlation between sovereign and bank CDS spreads in the two countries. Bank and sovereign CDS spreads in Europe show a strong correlation of 0.8 whereas the correlation between these two variables in the US is trivial.

In addition, the results show that publications of test results usually led to an increase in market confidence in both countries, whereas stress test announcements more frequently caused negative or ambiguous effects. European exercises had a slightly higher success rate since more than half of the events, either announcements or publications, were capable of restoring market confidence to some extent. In the US, half of the tests managed to reassure markets while the other half failed to do so.

No evidence can be found to support that stress tests in Europe and the US affect one another. Subsequent stress tests conducted in both regions usually had different effects on market confidence. From this it follows that the outcome of tests in the US usually did not affect expectations in Europe when a new exercise was announced and vice versa. Only the negative result of the EBA stress test in 2011 was followed by a negative outcome from the announcement of the CCAR on November 22, 2011. However it is not certain whether the effect in one country caused the other since there are several additional factors that could have influenced the results.

5. Discussion

The analysis of the effect of stress testing on CDS spreads could have been affected by the lack of availability of CDS data. CDS data is not available for all banks included in the sample used to assess the effect on equity returns. Therefore, estimations of the effects on CDS spreads are based on a smaller sample of banks for both Europe and the US. The slightly ambiguous US stress test results compared to European stress test results could partially be

(27)

23

due to the lack of CDS data for the US market. For example, the North America CDX Index would be a more reliable proxy for the CDS market index than the manually constructed one used in this study. However, it was impossible to use this better CDS market index due to a lack of access to a database that provides this data. Another reason for the less convincing US stress test results could be the fact that Dodd-Frank Act Stress Testing (DFAST) was not taken into consideration in this study. The DFAST exercises were left out because they are only complementary to the CCAR tests and in order to avoid overlapping estimation windows during event periods. However, DFAST exercises help assess whether banks have sufficient capital in order to absorb losses and continue operations when an adverse economic scenario occurs. These tests could therefore also influence market confidence and contribute to the overall effect of the CCAR in a particular year.

A threat to the internal validity of this study could be the potential of omitted variable bias. This problem occurs when variables that effect equity returns or changes in CDS spreads and additionally are correlated with the market index, are excluded from the model. This bias can be solved by including those variables in the multiple regression models. Another technical improvement could be the use of the test statistic developed by Boehmer, Musumeci and Poulsen (1991) since this test corrects for the additional variance of returns caused by the event. However, Candelon and Sy (2015) found that the results of the simple t-stat are comparable to the ones acquired by using the statistic proposed by Boehmer, Musumeci and Poulsen. Therefore, the significance of this study is tested by using the standard t-statistic.

Another technical improvement could be the use of a Multivariate Regression Model (MVRM) to estimate the panel regressions with the event dummies. This method is commonly applied in the case of total clustering, meaning that the event period is the same for all firms, as it accommodates hypothesis testing with differences in individual returns. However, MacKinlay (1997) mentions that this method has two drawbacks as the test statistic could have poor finite sample properties and has little power against alternatives. Furthermore, because the estimated coefficients and standard errors produced by the MVRM will be identical to the ones obtained through a simple OLS estimation, there is no gain in efficiency through use of the MVRM. Therefore this study did not rely upon this approach but instead used fixed or random effect estimation techniques to be able to account for individual heterogeneity across banks.

(28)

24

6. Conclusion

From 2009 onwards several stress testing exercises have been conducted in both Europe and the US. The main objective of these stress tests was to restore confidence in the financial system after the financial crisis in 2008. However, it was not exactly clear to what extend these stress tests were capable of rebuilding confidence. This study examined the impact of different stress testing exercises on market confidence and reveals that stress testing can be an important tool for crisis management.

The results of this study show that stress tests are able to significantly affect financial market confidence through both their announcement and publication of the results. Every EU- and US-wide stress test examined in this study had some significant impact on equity returns or CDS spreads. For the majority of the stress tests, the publication of test results increased market confidence in Europe and the US. The only negative impact due to the publication of results was observed for the 2011 EBA test and the 2014 US test had an ambiguous effect. The remaining publications had positive effects on market confidence. Most results are in agreement with anecdotal evidence or the current literature. The publication of the results of the SCAP test led to a significant increase in bank equity returns and caused CDS spreads to narrow which agrees with the fact that the SCAP was seen as one of the most successful stress tests exercises.

In most cases announcements led to ambiguous effects in the US and to negative effects in Europe. Overall, ambiguous effects were only found for the announcement of US stress tests, whereas the impact of European tests is clear in all cases.

Analysis of the results found in this study reveals that the general effect of stress testing on market confidence is positive, as it contributes to improvement of the transparency of the financial system. However, also some negative or ambiguous effects were found which indicate that the effect of stress testing on market confidence depends on many different factors, all of which have to be taken into account in order to ensure success of the stress test. Transparency throughout the entire process is one important factor for success, but can also result in a negative outcome depending on investor sentiment and expectations. Other factors that could influence the result of the stress test are the commitment of authorities, the state of the political economy, the perception of sovereign risk, governance and coordination of the stress test and technical aspects such as scenario design, the scope of coverage and information disclosure.

(29)

25

References

Acharya, V., Engle, R. and Pierret, D. (2014), “Testing macroprudential stress tests: The risk of regulatory risk weights”, Journal of Monetary Economics, Vol. 65, pp. 36–53 Alexander, C. and A. Kaeck (2008), “Regime dependent determinants of credit default swap

spreads”, Journal of Banking and Finance 32, pp. 1209-1228

Altintas, Y., Callender, W. and Matsakh, E. (2010), “Stress Testing in Action: Lessons from the Financial Crisis”, Bank Accounting and Finance

Andres, C., Betzer, A. and Doumet, M. (2013), “Measuring Abnormal Credit Default Swap Spreads”, Available at SSRN: http://ssrn.com/abstract=2194320

Annaert, J., de Ceuster, M., van Roy, P., Vespro, C. (2013), “What determines Euro area bank CDS spreads?” Journal of International Money and Finance 32, pp. 444–461

Basel Committee on Banking Supervision (2012), “Peer review of supervisory authorities’ implementation of stress testing principles” (Basel: Bank for International

Settlements)

Bernanke, B. S. (2013), “Stress Testing Banks: What Have We Learned?”, Speech delivered at the "Maintaining Financial Stability: Holding a Tiger by the Tail" financial markets conference, Stone Mountain, Georgia on April 8, 2013

Binder, J. J. (1985), “On the Use of the Multivariate Regression Model in Event Studies”, Journal of Accounting Research, Vol. 23, No. 1, pp. 370-383

Binder, J. J. (1998), “The Event Study Methodology Since 1969”, Review of Quantitative Finance and Accounting, Vol. 11, No. 2, pp. 111-137

Blanco, R., Brennan, S. and Marsh, I. W. (2005), “An Empirical Analysis of the Dynamic Relation between Investment-Grade Bonds and Credit Default Swaps”, The Journal of Finance, Vol. 60, No. 5, pp. 2255-2281

Boehmer, E., Musumeci, J. and Poulsen, A. B. (1991), “Event-study methodology under conditions of event-induced variance”, Journal of Financial Economics, Vol. 30, pp. 253-272

Borio, C., Drehmann, M. and Tsatsaronis, K. (2012), “Stress-testing macro stress testing: Does it live up to expectations?”, BIS Working Papers, No. 396

Brown, S. J. and Warner, J. B. (1985), “Using daily stock returns - The case of event studies”, Journal of Financial Economics, Vol. 14, No. 1, pp. 3-31

Candelon, B. and Sy, A. N. R. (2015), “How Do Markets React to Stress Tests?”, IMF Working Paper, WP/15/75

Bank stress testing and market confidence : a comparison between Europe and the United States