Master Thesis
The effect of the public disclosure of banking
stress tests on the CDS market:
A comparative study between the EU and US
January, 2016
MSc. Finance
Faculty of Economics and Business
University of Groningen
Author: Suzanne Kluifhooft Student nr.: s2169541
Email: suzannekluifhooft@gmail.com Date: January 14, 2016
2
The effect of the public disclosure of banking stress tests on the CDS market:
A comparative study between the EU and US
Abstract
To restore public trust the European Banking Authority initiated EU-wide banking stress tests in 2010 and performed subsequent tests in 2011 and 2014. This paper examines whether there is an effect of the public disclosure of banking stress test on the CDS market and whether it adds informational value to the market. By using an event study this effect is confirmed for both the EU and US sample, which both anticipated the results of the stress tests. Overall, the EBA reached its goal since the CDS spreads decreased and conjointly the perceived risk and uncertainty in the market. When examining the subsamples the CDS spreads decreased predominantly for banks which passed the test, however no effect was found for banks which failed the test. Furthermore, this study finds an asymmetric effect based on the outcome of the tests in the European sample but not in the US sample.
3
1. Introduction
With the recent financial crisis, financial risk management has become of increased importance for financial institutions (Hull, 2012). Due to opaqueness and decreased trust in the financial markets, transparency became of great importance. This transparency in the banking sector is endeavored by the European Banking Authority (EBA), since transparency reinforces market discipline and restores public trust. The EBA is responsible for the functioning and integrity of the financial markets and the stability in the EU. In this manner, the EBA attempts to help national authorities in their supervisory role, by conducting an EU-wide stress test analysis to estimate the resilience of financial institutions to adverse market shocks. The first stress test, conducted in 2010, included 91 European banks from 20 EU member states, covering 65% of total assets of all European banks. The second stress tests, conducted in 2011, covered 90 banks from the same 20 EU member states and Norway. The third stress test, performed in 2014, included 123 banks covering 21 EU member states and Norway. These banks represented 70% of total assets in all European banks. The banks subject to the stress test had to cover at least 50% of the national banking sector for the EU and Norway. This percentage is based on total consolidated assets from end 2009, 2010 and 2013, respectively. The banks were stress-tested at the highest consolidation level possible (group level) and were selected in a descending order based on market share. In addition, foreign subsidiaries were excluded, but foreign non-EU banking groups were included to reach the 50% coverage (www.eba.europe.eu).
In the United States the Supervisory Capital Assessment Program (SCAP) was introduced in 2009 as a reaction to the financial crisis. The Federal Reserve (FED) evaluated 19 US bank holding companies (BHCs) on their ability to cope with their customers credit needs in times of crises. To control for this ability the SCAP simultaneously estimated the additional capital buffers for adverse market conditions for all 19 BHCs. Each year from 2011, the FED exercises the Comprehensive Capital Analysis and Review (CCAR) for the largest BHCs operating in the US. The CCAR reviews whether these BHCs have sufficient capital to assure operations in adverse market developments and controls for robust forward-looking capital plans. From 2012 onwards the results are publicly announced. The requirements for the CCAR only apply to selected banks with assets of $10 billion or higher. The test started with evaluating 19 BHCs, however the most recent US CCAR stress test included 31 BHCs, holding 80% of all domestic assets (http://www.federalreserve.gov).
4 public disclosure of banking stress test results on the credit default swap (CDS) spreads for both European and US banks.
Credit default swaps are a type of credit derivative designed to insure against default of a third party. This swap transfers the credit exposure from the buyer to the seller of the derivative. The buyer of the swap receives credit protection by paying a premium to the seller of the swap, who guarantees the credit worthiness and will pay the losses in case of default. This premium is called the CDS spread and increases when the probability of default increases. Therefore, it is influenced by market forces and an indicator for the perception of bank risk (Weistroffer, Speyer and Walter, 2009).
Earlier studies have already examined the EU stress tests of 2010 and 2011. Alves, Mendes and Pereira da Silva (2015) studied the impact of the disclosure of stress test results on European financial stocks and CDS markets. The authors find significant abnormal returns on the stock market and a reverse reaction for the CDS market. The significance of the results increased with the second stress test. This study will investigate how these results will change with the inclusion of the third stress test. Another study by Morgan, Peristiani and Savino (2014) examines the informational value of the US SCAP stress test and finds a significant negative relation between abnormal returns and the capital gap. Moreover, they find counterintuitive results for the CDS market. CDS spreads decreased for banks, which underperformed and were expecting a capital gap, and the spreads increased for banks with no capital gap. However, research by the Dutch Central Bank on the disclosure effect in the US during 2009-2015 shows declining CDS spreads as a reaction to the disclosure of the results and the announcement of banks with an approved and non-approved capital plan (Neretina, Sahin and de Haan, 2015).
5 US market is also investigated. The study concludes that CDS spreads are negatively affected by the public disclosure of the outcome, thereby decreasing the perceived risk and uncertainty in the market. Therefore, the EBA reached its goal of achieving more transparency and reinforcing public trust on the financial markets. Furthermore, when differentiating among banks who passed and those who failed, this negative effect was predominantly present for banks who passed whereas there was no effect for those who failed. Additionally, an asymmetric effect based on the outcome of the test was confirmed in the European sample but not in the US sample.
6
2. Literature review
Current literature offers limited information concerning the disclosure of stress test outcomes and their effect on financial markets. Studies covering the effects on the CDS markets in both the United States and the European Union are especially scarce. A possible explanation for this lack of information on CDS markets is because this market is relatively young and small compared to other markets (Norden and Weber, 2004).
Studies examining EU stress tests:
7 2011 for long bond bid-ask spreads. Therefore, public disclosure leads to reduced information asymmetry, but it increases information uncertainty.
Studies examining US stress tests:
In a study investigating the informational value of the 2009 SCAP stress test on the US stock market, Morgan, Peristiani and Savino (2014) conclude that the market already knows which banks present capital gaps but not the size of this gap. Therefore, with the disclosure of the results, banks with a larger capital gap than expected had significant negative abnormal returns. On the CDS market the result were counterintuitive, the clarification and publication of the results led to a decrease in the abnormal change for banks with capital gaps, and an increase for banks with no capital gap. The methodology event was consistent with theory and the equity market. Since a capital gap should decrease the financial health of a firm therefore, the perceived credit risk increases and thus the CDS spreads increase. Furthermore, private sector analysts did not have all information, therefore the disclosure of stress test outcomes produced informational value, leading to abnormal changes in CDS spreads. A working paper of the Dutch Central Bank investigates the disclosure effect of US banking stress test on banks equity prices, credit risk and systematic risk during 2009-2015 (Neretina, Sahin and de Haan, 2015). Furthermore, they measured credit risk with CDS spreads and found declined CDS spreads as a response to the disclosure of the results in 2009, 2012 and 2013. These results suggest that the stress tests reached their goal of restoring market confidence and lowering the credit default risk of subjected banks. The authors also suggest that stress tests affect systematic risk. A related study examines the effects of regulatory reforms for both US and European banks (Schaefer, Schnabel and di Mauro, 2013). These reforms have a similar effect as the regulatory requirements of the core tier 1 ratios stated in the Basel accords which are tested in the stress tests of both the EBA and FED. The authors find significant market reactions in the stock market and the CDS market. The reforms resulted in a strong increase of the CDS spreads for systemically important banks, but had only a mildly effect on the stock market.
Rating announcements:
According to a report by the Basel Committee on Banking Supervision (2004) external credit ratings could be used in the near future to determine regulatory capital. Therefore, it could be relevant to compare the results of these studies.
9
3. Data and methodology
3.1 Data
To examine the effect of EU-wide stress testing on CDS spreads of the according banks our sample consists of 63 banks with CDS contracts, which were selected for the stress test of 2010, 2011 or 2014. 53 banks were tested in 2010 and 41 and 45 banks for respectively 2011 and 2014, summing up to 137 events. The CDS contracts are senior contracts with a maturity of 5 years, since this is the benchmark for CDS market studies, and are of the preferred restructuring type Modified-Modified (MM)1, which is the best fit for Europe (Thomson Reuters DataStream).
As a benchmark, a CDS market portfolio for European financial firms is used. For the samples of 2010 until 2014 a European financial sector benchmark is used with the specification for senior 5-year data (DS EU FIN OTHER 5Y CDS INDEX). Other banking CDS indices have not been chosen, due to overlap of banks subject to the stress test. The benchmark is also included into the control group2, a set of
financial firms which were not subject to the stress tests. The control group consists of the constituent list of firms used in the benchmark, wherein the other benchmark group (DS EU OTHER FIN 5Y CDS INDEX) was added and banks subject to the stress test were deleted. The control group shows how financial firms, which were not subjected to the stress tests, behave and are affected by news from the stress test. One goal of stress testing is to bring stability and confidence to the market, whether this goal is achieved can be seen through the control group.
This thesis examines the effect of the disclosure of stress test outcomes on the credit market, for this purpose, not stock prices but bond yields have to be examined. However compared to CDSs, bonds have higher transaction costs, are for investors with a long trading horizon, are less frequently traded and are not quoted daily (Oehmke and Zawadowski, 2014). By using a synthetic bond yield instead of the real bond yield these problems are evaded. The synthetic bond yield is formed by taking a long position in the risk free yield, which is proxied by the yield on euro area government bonds with triple A ratings and 5-year maturity, and a short position in the CDS contract. Additionally, literature suggests that this theory holds, and shows similar results (Hull, Predescu and White, 2004).
Synthetic bond yield = risk-free yield on Eurobond + CDS spread (1) The risk free yield represents the time value of money or the interest rate risk and the CDS contract takes credit risk into account, this way both types of risk are included in the thesis. For the purpose
1 Not all banks have MM CDS contracts, then the fully restructured (CR) or non-specified CDS contract is
chosen.
10 of focusing purely on the credit risk effect, the results when using solely the CDS spreads are showed in the appendix.
Furthermore, the banks in the sample are divided into groups based on whether they passed the stress test (Group A), achieved tangential results3 (Group B) or failed the stress test (Group C). By
separating these groups, differences depending on the outcomes of the stress test can be detected. In the comparative analysis the results of the CCAR for the years 2012-2015 are examined. This US stress test is an annual test performed by the Federal Reserve and analyzes whether the largest banks can survive in periods of economic and financial stress and whether they have a reliable forward-looking capital plan to deal with different types of risk. Data for the participating banks is retrieved from the CCAR reports of 2012-2015.
Over these years 33 banks were subject to the CCAR stress tests. Due to data availability, this led to a total sample of 16 banks over the period of 2012-2015. Wherein, 11 banks were tested in 2012, and 10, 13 and 14 for respectively 2013, 2014 and 2015. Consistent with the EU sample the CDS contracts have seniority and have a maturity of 5 years. However, the preferred restructuring type is no restructuring (XR), since this is the best fit for US firms (Thomson Reuters Datastream).
For the US sample a similar benchmark as the EU sample is used, i.e. the Datastream North America financial sector index (DS NA OTHER FIN 5Y CDS INDEX). This is the equivalent of the benchmark used in the European sample. Furthermore, in this comparative analysis again synthetic bond yields are used to measure abnormalities in credit default risk. Therefore, the risk free rate is proxied by the US treasury yield with a maturity of 5 years.
In line with the EU sample the CCAR stress tests also publicly announce whether the subjected banks comply with regulatory capital ratios under periods of stress or not. The requirements of these ratios are equal to the EU sample. Based on the outcome of the stress tests the Federal Reserve objects to the capital plan or not. Therefore, like the EU sample different subgroups can be determined. Namely, banks with a objection to its capital plan (Group A), banks with a conditional non-objection (Group B)4 and banks with an objection to its capital plan (Group C).
3 Tangential results are core tier one ratios between 5-6% over a 2 year horizon for the years 2010 and 2011
and between 5.5-6.5% over a 2 year horizon for the sample of 2014. These are banks who nearly passed the stress test.
11
Figure 1: Timeline of events
3.2 Methodology
This paper builds upon the methodology used in Alves, Mendes and Pereira da Silva (2015), which used an estimation window of 120 days [t-129; t-10] and multiple event windows [t-9;t-1], [t-5;t-1],
[t0;t5],[t0;t10]. These event windows account for periods before and after the disclosure of the stress
test results as shown in the figure below. This study resembles the estimation and event windows and adds a smaller event window of [t-1;t1] for the effect on the event date as used in other research
papers (Morgan, Peristiani and Savino, 2014; Neretina, Sahin and de Haan, 2015).
Figure 2: Estimation window and event window
For simplicity, in the remainder of this study, the event study terminology will be employed when referring to abnormal changes in CDS spreads as abnormal returns in CDS spreads.
15-07-2011 Disclosure of second EU stress test 26-10-2014 Disclosure of third EU stress test 13-03-2012 Disclosure of first CCAR test
14-03-2013 Disclosure of second CCAR test 16-03-2014 Disclosure of third CCAR test 11-03-2015 Disclosure of fourth CCAR test 23-07-2010 Disclosure of first EU stress test
T1= Day -9 T= Day 0 T2= Day +10
Event day
Estimation Window Event Window
12 The effect of the disclosure of stress test outcomes on CDS spreads is measured using abnormal returns of the CDS spreads by using the market adjusted returns model specified in Brown and Warner (1980):
𝐴𝐴𝐴𝐴𝑖𝑖𝑖𝑖 = 𝐴𝐴𝑖𝑖𝑖𝑖 − 𝛼𝛼�𝑖𝑖− 𝛽𝛽̂𝑖𝑖(𝐴𝐴𝑚𝑚𝑖𝑖) (2)
Here, ARit is the abnormal return in the market model of bank i on day t, Rit is the return of bank i on
day t, Rmt is the market portfolio return on day t. 𝛼𝛼� and 𝛽𝛽̂ are the estimated parameters of the
market model. Formula 4 and 5 calculate the alphas and beta’s from the data of the estimation window. In this model the abnormal return is derived from the CAPM formula shown in equation 2, in which a risk-free rate of zero is assumed. Rearranging this formula leads to the market adjusted model displayed in equation 1. Micu, Remolona and Wooldrigde (2006) concluded that abnormal returns of CDS spreads estimated with the market adjusted returns model is more robust than the absolute spread changes.
𝐴𝐴𝑖𝑖𝑖𝑖 = 𝛼𝛼�𝑖𝑖+ 𝛽𝛽̂𝑖𝑖�𝐴𝐴𝑚𝑚𝑖𝑖− 𝐴𝐴𝑓𝑓� + 𝜀𝜀 (3)
𝐴𝐴𝑖𝑖𝑖𝑖 = 𝐿𝐿𝐿𝐿(𝑆𝑆𝑆𝑆𝑖𝑖⁄𝑆𝑆𝑆𝑆𝑖𝑖−1) (4)
Equation 3 shows the computation of Rit, which is based on continuously compounded returns. SMt is
the CDS mid-rate spread in basis points on date t and SMt-1 is the mid-rate spread on the previous
trading day.
Equation 4 and 5 show the formulas for the beta and alpha, respectively. Beta (βi) is an indicator of
systematic risk of a security in relation to the market portfolio. It reflects the sensitivity of a security to market movements. µi is the average return of bank i and µm is the average return of the financial
sector benchmark, both calculated for the estimation window. The alpha (αi) indicates the difference
between the actual return and estimated return based on the benchmark. The alpha represents the excess return on the benchmark (market) return.
𝛽𝛽𝛽𝛽 =𝐶𝐶𝐶𝐶𝐶𝐶(𝐴𝐴𝛽𝛽, 𝐴𝐴𝑅𝑅)𝑉𝑉𝑉𝑉𝑉𝑉(𝐴𝐴𝑅𝑅) =∑𝑖𝑖−10𝑖𝑖−129∑ (𝐴𝐴𝑖𝑖𝑖𝑖(𝐴𝐴− 𝜇𝜇𝑖𝑖)(𝐴𝐴𝑚𝑚𝑖𝑖− 𝜇𝜇𝑚𝑚)
𝑚𝑚𝑖𝑖− 𝜇𝜇𝑚𝑚)2 𝑖𝑖−10
𝑖𝑖−129 (5)
𝛼𝛼𝑖𝑖 = 𝜇𝜇𝑖𝑖− 𝛽𝛽𝑖𝑖(𝜇𝜇𝑚𝑚) (6)
13 This event study will measure the abnormal returns over different event windows. To estimate the total abnormal return over these periods, the AARs and CAARs must be computed, which is shown in the following formulas:
𝐴𝐴𝐴𝐴𝐴𝐴𝑖𝑖 =∑ 𝐴𝐴𝐴𝐴𝑖𝑖𝑖𝑖 𝑁𝑁 𝑖𝑖=1 𝑁𝑁 (7) 𝐶𝐶𝐴𝐴𝐴𝐴𝐴𝐴𝑖𝑖 = � 𝐴𝐴𝐴𝐴𝐴𝐴𝑖𝑖 𝑖𝑖1 𝑖𝑖0 (8)
3.3 Hypotheses
This study looks into the effects of public disclosure of the outcomes of EU-wide stress tests on the CDS market. Based on previous literature and the objectives of this paper the following hypotheses are stated:
H1: The disclosure of the stress test outcomes has a significant effect on the CDS market.
First, this study will examine whether there is informational value added to the market when the outcomes of the stress tests are disclosed. The reports of the EBA present private information about asset quality, risk appetite and whether a bank meets the Basel requirements. Stress tests could convey information valuable for investors or information that changes the estimated default probabilities. Therefore, the CDS spreads could change, giving rise to an abnormal return during the event windows after the disclosure ([-1;1],[0;5] and [0;10]).
H2: CDS markets anticipate the direction of the outcomes of the stress test.
It is possible that investors and thus markets anticipate on the outcomes of the stress test, because of information leakages, market expectations or speculation by traders (Harrington, 2006). In this case, CDS spread will start to adjust in the period before the announcement. Therefore, the CAARs of [-9;-1] and [-5;-1] will be examined for significance as an indicator of anticipation.
H3a: The disclosure of the stress test outcomes has a significant (negative) effect on the banks that passed the stress test (group A).
H3b: The disclosure of the stress test outcomes has a significant (positive) effect on the banks that failed the stress test (group C).
14 spread if this is new information. The abnormal return should be negative and significant, because passing the stress test indicates a healthy bank which decreases the chance of default. A negative outcome of the stress test (group C) could lead to an increase in the CDS spread and therefore the abnormal return should be positive and significant. When banks fail the stress test, they are not financially fit and do not comply with the Basel accords. The following hypotheses will be tested for the same event windows as hypothesis 1.
H4: There is an asymmetric result whether the banks passed or failed the stress test.
Literature states there could be an asymmetric effect depending on the outcome of the tests (Hull, Predescu and White, 2004). A negative outcome, of a bank failing the test, has a deeper impact on the CDS market than a positive result. The gains of a bank passing the stress test are not proportional to the losses of a bank failing the stress test. This could be interpreted as, the decrease in CDS spreads for group A should be smaller than the increase in CDS spreads for group C. A reason could be that failing is a signal of serious issues with the financial fitness of the bank, whereas one would normally expect banks to pass the stress tests. Hypothesis 4 is checked by comparing sign and significance of positive and negative abnormal returns for both groups (A and C) and by the means of a sensitivity analysis.
H5: The effect of the disclosure of the stress test outcomes has a higher impact on banks subject to the stress tests than on financial firms not subject to the stress tests.
In this hypothesis, firms subjected and not subjected to the stress test are compared. It is interesting to evaluate if and how the control group reacts to stress tests and whether there is a difference between the control group and bank that passed the stress tests. The hypothesis will be tested using a sensitivity analysis.
H6: The credit quality of banks subject to the stress tests influences the reaction to the disclosure of the stress test outcomes.
15
3.4 Significance tests
The most common test to assess if the average abnormal returns (AARs) and cumulative average abnormal returns (CAARs) of the CDS spreads are significant is the student t-test. This parametric test has high power if strict conditions are met (Brown and Warner, 1980). One of these conditions is normality of the probability distribution, which will be tested with the Jarque-Bera statistic. The student t-test is specified in the following formula:
𝑡𝑡 =𝑥𝑥̅ − 𝜇𝜇𝑠𝑠 =(𝐶𝐶)𝐴𝐴𝐴𝐴𝐴𝐴𝑠𝑠 𝑖𝑖 (9)
Here x is the observed average abnormal return, which will be tested against µ, the hypothesized average abnormal return of zero, s is the standard deviation in the estimation window. The standard deviation is calculated using the cross-section dependence adjustment (Brown and Warner, 1980), where 𝜎𝜎𝑖𝑖 is the standard deviation for event i and N the number of events:
𝑠𝑠 = ��� 𝜎𝜎𝑖𝑖2� 𝑁𝑁 − 1� (10)
Another test is Cowan’s generalized sign test, a non-parametric test, which does not assume stringent assumptions such as normality of the dataset and has high explanatory power. This test does not require symmetry of the cross sectional abnormal return distribution and performs better under variance increases. Furthermore, it is less sensitive to a single large outlier. Moreover, Cowan’s generalized sign test is more appropriate for longer event windows as used in this study (Cowan, 1992; Campbell, Cowan and Salotti, 2010). Additionally, a non-parametric test increases the robustness of the results (MacKinlay, 1997). However, a disadvantage is that this test is best specified in case of equal proportions of negative and positive abnormal returns in the estimation window (Brown and Warner, 1980). The test statistic is:
𝑍𝑍𝐺𝐺 = 𝑤𝑤 − 𝑁𝑁𝑝𝑝̂
[𝑁𝑁𝑝𝑝̂(1 − 𝑝𝑝̂)]12 (11)
Were w is the number of positive signs in the event window, N the number of events, M the number of days in the estimation window and 𝑝𝑝̂ a parameter which estimates the proportion of positive signs in the estimation window. 𝑝𝑝̂ can be computed using equation 11.
16 As a control a second non-parametric test is used, namely the Corrado rank test. When correctly specified the rank test has more power than the generalized sign test in detecting significant abnormal returns. Even though, it is less powerful in longer event windows, for the smaller event window around the event date this test could give a definite answer regarding our hypotheses. Another disadvantage is the deteriorated performance in the case of increased variance, then the rank tests rejects true null-hypotheses more often. It also assumes that daily returns in event windows are independent (Cowan, 1992; Campbell, Cowan and Salotti, 2010). The Corrado rank statistic is calculated as follows:
𝑍𝑍𝑅𝑅=𝑑𝑑 1 2 𝐾𝐾���� − 70.5𝐷𝐷 [∑ (𝐾𝐾140𝑖𝑖=1 ��� − 70.5)𝑖𝑖 2/140] 1 2 (14)
The average rank across the n stocks and d days of the event window is reflected by 𝐾𝐾���� and 𝐾𝐾𝐷𝐷 ��� is the 𝑖𝑖
average rank across n stocks on day t. 70.5 is the mean rank and 140 is the size of the combined event and estimation window.
3.5 Sensitivity analyses
The sample is divided into subgroups based on whether they passed, nearly passed or failed the stress test. Therefore, these subsamples can also be tested individually or relative to each other. When comparing the subgroups there is an unbalanced case because the subgroups have unequal sizes, therefore an adjusted t-test should be used.5
𝑡𝑡 =|𝐶𝐶𝐴𝐴𝐴𝐴𝐴𝐴1− 𝐶𝐶𝐴𝐴𝐴𝐴𝐴𝐴2|
𝑠𝑠𝐶𝐶𝐶𝐶𝐶𝐶𝑅𝑅 1−𝐶𝐶𝐶𝐶𝐶𝐶𝑅𝑅 2 (15)
Here, CAAR1 and CAAR2 are the CAARs of the different subgroups or the control group. The standard
deviation of the mixed groups sCAAR 1 – CAAR 2, is determined using the following formulas:
𝑠𝑠𝐶𝐶𝐶𝐶𝐶𝐶𝑅𝑅 1−𝐶𝐶𝐶𝐶𝐶𝐶𝑅𝑅 2= �2 𝑆𝑆𝑆𝑆𝑀𝑀𝐿𝐿 ℎ (16) 𝑆𝑆𝑆𝑆𝑀𝑀 =𝑁𝑁𝑠𝑠12+ 𝑠𝑠22 1+ 𝑁𝑁2− 2 (17) 𝐿𝐿ℎ=1 2 𝑁𝑁1 � + 1 𝑁𝑁� 2 (18) Where, MSE is the mean squared error for the two subgroups, s is the standard deviation, N is the number of observations and nh is the harmonic mean of the sample sizes.
17
4 Results
Table 1: Descriptive statistics of abnormal returns of synthetic bonds in the estimation window Descriptive
statistics 2010 2011 2014 Control Control Control
(1) (2) (3) (4) (5) (6) Mean 0.0000 0.0000 0.0000 0.0000 -0.0002 0.0000 Median -0.0014 0.0002 0.0013 0.0004 0.0002 0.0010 Standard deviation 0.0349 0.0275 0.0413 0.0463 0.0223 0.0244 Minimum -0.2535 -0.3239 -1.9383 -1.9260 -0.2381 -0.6369 Maximum 0.3845 0.2128 0.3282 0.4763 0.2311 0.1620 Kurtosis 6.8231 6.4121 18.2509 17.1082 7.6000 15.5477 Skewness 0.1818 -0.1029 -1.1837 -0.3344 -0.3158 -1.4171 Jarque-Bera 73.7427 58.4222 1190.9657 997.4444 107.7929 827.3900 Estimation window (T) 120 120 120 120 120 120 Events (N) 53 41 45 38 43 44
Notes: This table displays the descriptive statistics in percentages for the estimation windows of the three time
samples using the synthetic bond yields.
Table 1 shows the descriptive statistics for the abnormal returns during the three time samples for the synthetic bond approach for the European sample. Overall the means are zero, which suggests an average abnormal return of zero over the estimation windows. The year 2014 shows some extreme minimum abnormal returns and this sample has a larger standard deviation. The kurtosis, skewness and Jarque-Bera statistics are high for all time samples, however 2014 shows extreme values.
18
Table 2: Percentage of statistically significant CAR (5% level)
CAR>0
CAR<0
Synthetic bonds [-9;-1] [-5;-1] [-1;1] [0;5] [0;10] [-9;-1] [-5;-1] [-1;1] [0;5] [0;10] (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) 2010 Treatment group 24.53% 9.43% 0.00% 1.89% 1.89% 13.21% 3.77% 5.66% 50.94% 62.26% Group A 27.66% 10.64% 0.00% 0.00% 0.00% 10.64% 4.26% 6.38% 51.06% 65.96% Group B 16.67% 0.00% 0.00% 16.67% 16.67% 33.33% 0.00% 0.00% 50.00% 50.00% Control group 13.16% 2.63% 5.26% 15.79% 23.68% 5.26% 0.00% 5.26% 10.53% 10.53% 2011 Treatment group 82.93% 68.29% 68.29% 4.88% 34.15% 0.00% 0.00% 0.00% 29.27% 7.32% Group A 81.82% 66.67% 69.70% 3.03% 42.42% 0.00% 0.00% 0.00% 27.27% 6.06% Group B 100.00% 100.00% 83.33% 0.00% 0.00% 0.00% 0.00% 0.00% 50.00% 33.33% Group C 50.00% 0.00% 0.00% 0.00% 0.00% 0.00% 0.00% 0.00% 0.00% 0.00% Control group 30.23% 25.58% 16.28% 2.33% 6.98% 0.00% 0.00% 0.00% 18.60% 4.65% 2014 Treatment group 8.89% 2.22% 0.00% 8.89% 11.11% 8.89% 35.56% 15.56% 13.33% 15.56% Group A 6.25% 3.13% 0.00% 6.25% 12.50% 9.38% 28.13% 15.63% 15.63% 12.50% Group B 16.67% 0.00% 0.00% 0.00% 0.00% 33.33% 50.00% 0.00% 16.67% 16.67% Group C 28.57% 0.00% 0.00% 14.29% 14.29% 0.00% 42.86% 28.57% 28.57% 42.86% Control group 0.00% 0.00% 0.00% 6.82% 2.27% 11.36% 29.55% 2.27% 2.27% 11.36%Notes: This table displays the percentage of banks with statistically significant CARs at the 5% level for the different event windows. Significance is determined using the
19
4.1 T-test
Table 3: Cumulative average abnormal returns (CAARs) and results of the t-test
Time 2010 2011 2014
CAAR P-value CAAR P-value CAAR P-value
[-9;-1] 0.0120 0.3715 0.1395 0.0000*** -0.0021 0.4798
[-5;-1] 0.0097 0.3960 0.0900 0.0011*** -0.0419 0.1575
[-1;1] -0.0083 0.4109 0.0982 0.0005*** -0.0331 0.2131
[0;5] -0.0642 0.0419** -0.0334 0.1160 -0.0172 0.3399
[0;10] -0.0932 0.0067*** 0.0270 0.1665 -0.0269 0.2586
Notes: ***, ** and * denote the statistical significance at 1%, 5% and 10%, respectively. SE are based on the
Crude Dependence Adjustment. This table contains the results of the t-test of the treatment group for the three stress tests separately. It shows the CAARs and its significance during the five event windows using the synthetic bond approach. A right sided t-test is used when the CAAR is positive and a left sided t-test is applied when the CAAR is negative. The degrees of freedom for 2010, 2011 and 2014 were respectively 52, 40 and 44. Table 3 shows the CAARs of the treatment group for the three stress tests for the five event windows. Additionally, it demonstrates the p-value and its significance when using the student t-test. The first hypothesis examines whether the stress test adds valuable information to the CDS market, leading to abnormal returns in CDS spreads or synthetic bond yields. The stress test of 2010 gives negative significant results in the post-announcement event windows, rejecting the null-hypothesis and giving evidence for abnormal returns at the CDS market. However, the smaller event window surrounding the announcement day is not significant and so are almost none of the AARs of the individual days (not included6). Furthermore, the event windows of [0;5] and [0;10] are not
significant for the other stress tests in 2011 and 2014. In contrast to 2010, the window [-1;1] in 2011 is significant but positive, suggesting an increase in the CDS spreads in the few days surrounding the disclosure of the outcomes of the 2011 stress test. This opposite reaction could be due to the fact that it is the second test, therefore it is not a new phenomenon anymore and investors are more familiar with possible outcomes and effects, also investors could not trust the outcomes or its reasonability. Another reason could be that more banks failed the second stress test, and that these banks were bigger and more prominent. Based on the results of table 3, hypothesis 1 is rejected for 2010 and 2011 but not for 2014. However, the effect is more persistent in 2010 whereas in 2011 there is no longer an effect on the days after the announcement.
Hypothesis 2 tests for anticipation of the outcomes of the stress tests by the market. Then, significant abnormal returns should be seen in the event windows prior to disclosure, so the windows [-9;-1] and [-5;-1]. Only in 2011 there were significant results for the positive CAARs of these event
6 For 2010 only the AAR of day 2 is significant at the 5% level. The tables with these analyses are available on
20 windows. CDS spreads increased in the days prior to the disclosure, signaling a bad outcome of the stress tests, because bad news about the financial health of certain banks or bank holding companies raises the interpreted default probabilities and thus the CDS spreads. Hypothesis 2 of anticipation can be rejected for 2011 but not for 2010 and 2014.
However, the Jarque-Bera statistics shown in table 1 are very high giving evidence for non-normality of the dataset. Therefore, results of the parametric student t-test do not provide enough evidence to accept or reject the stated hypotheses. The results should be complemented with an additional non-parametric test to check the results for reliability and provide further evidence.
4.2 Cowan’s generalized sign test
Table 4: Results of the sign test for the cumulative average abnormal returns (CAARs)
Time 2010 2011 2014
CAAR>0 CAAR<0 CAAR>0 CAAR<0 CAAR>0 CAAR<0
[-9;-1] 0.0372** 0.9628 0.0000*** 1.0000 0.9761 0.0239**
[-5;-1] 0.0204** 0.9796 0.0000*** 1.0000 1.0000 0.0000***
[-1;1] 0.8229 0.1771 0.0000*** 1.0000 1.0000 0.0000***
[0;5] 0.9999 0.0001*** 0.9975 0.0025*** 0.9883 0.0117**
[0;10] 1.0000 0.0000*** 0.0217** 0.9783 0.9990 0.0010***
Notes: ***, ** and * denote the statistical significance at 1%, 5% and 10%, respectively. This table contains the
results of the sign-test of the treatment group for the three stress tests separately. It shows the sign of the CAARs and its significance during the five event windows using the synthetic bond approach. The CAARs are tested for both signs with degrees of freedom for 2010, 2011 and 2014 of respectively 52,40 and 44.
Table 4 presents the significant CAARs in the different event windows based on the sign test. For the first stress test, it shows a positively significant effect for the days prior to the disclosure, [-9;-1] [-5;-1] and is negatively significant for the CAARs of [0;5] and [0;10], which confirms the results of the t-test. In 2011, the sign test confirms the t-test results for the positive abnormal returns in the windows prior to the disclosure and on the event window [-1;1]. There is also a negatively significant effect on the smaller event window after the disclosure, however on the longer window there is again a positive significant effect. Contradicting the t-test, the sign test does find significant negative effects for all event windows in 2014.
21 For all time frames, there is a significant negative effect for the event window [0;5] and [0;10] (only [0;10] is significantly positive for 2011). Therefore, there is significance leading to the rejection of the null-hypothesis for all time frames. The stress tests of 2010-2014 added valuable information to the CDS market, because the CDS spreads decreased after the announcement of the outcomes, suggesting a decrease in credit risk. Thus tests of the EBA ensured transparency and trust in the financial sector.
22
4.3 Corrado rank test
Table 5: Results of the rank test for the cumulative average abnormal returns (CAARs)
Time 2010 2011 2014
CAAR>0 CAAR<0 CAAR>0 CAAR<0 CAAR>0 CAAR<0
[-9;-1] 0.2429 0.7571 0.0508* 0.9492 0.5474 0.4526
[-5;-1] 0.2572 0.7428 0.1088 0.8912 0.8939 0.1061
[-1;1] 0.5257 0.4743 0.0026*** 0.9974 0.8632 0.1368
[0;5] 0.7537 0.2463 0.7735 0.2265 0.7196 0.2804
[0;10] 0.8043 0.1957 0.3695 0.6305 0.7310 0.2690
Notes: ***, ** and * denote the statistical significance at 1%, 5% and 10%, respectively. This table contains the
results of the rank-test of the treatment group for the three stress tests separately. It shows the sign of the CAARs and its significance during the five event windows using the synthetic bond approach. The CAARs are tested left-sided when the Z-statistic is negative and right-sided when the Z-statistic was positive. The degrees of freedom for 2010, 2011 and 2014 are respectively 52,40 and 44.
Table 5 shows the results of the Corrado rank test for the same sample as the sign test in the previous section. This test is more suitable to test the significance of very small event windows of one day, hence the focus is on the smallest event window of [-1;1]. For 2011 it finds a positive effect for the window [-1;1], which is consistent with the sign test. Nonetheless, it does not confirm the significance in 2010 or 2014. The Corrado rank test is less powerful than the generalized sign test in this specific set-up, due to the length of the event windows and possible outliers. Therefore, in the subsequent tests the Corrado rank test will be dropped and the generalized sign test will solely be used as non-parametric test.
4.4 Subgroups
23
Table 6: Cumulative average abnormal returns (CAARs) and results of the t-test for the subgroups
Time 2010 2011 2014
CAAR P-value CAAR P-value CAAR P-value
Group A [-9;-1] -0.0695 0.0278** 0.0854 0.0016*** -0.0389 0.1995 [-5;-1] 0.0016 0.4822 0.1269 0.0000*** -0.0039 0.4662 [-1;1] 0.0076 0.4153 0.0963 0.0005*** -0.0346 0.2264 [0;5] 0.0408 0.1277 -0.0315 0.1250 -0.0158 0.3649 [0;10] 0.0885 0.0080*** 0.0332 0.1128 -0.0218 0.3175 Group B [-9;-1] -0.0312 0.2686 0.2281 0.0004*** -0.0132 0.3357 [-5;-1] -0.0059 0.4528 0.1423 0.0038*** -0.0584 0.0484** [-1;1] 0.0082 0.4346 0.1425 0.0037*** -0.0164 0.3007 [0;5] -0.0460 0.1861 -0.0549 0.0890* -0.0135 0.3332 [0;10] -0.0642 0.1138 -0.0157 0.3390 -0.0152 0.3131 Group C [-9;-1] 0.0091 0.3764 0.0160 0.3204 [-5;-1] 0.0119 0.3420 -0.0419 0.1216 [-1;1] -0.0019 0.4739 -0.0411 0.1258 [0;5] 0.0141 0.3159 -0.0142 0.3399 [0;10] -0.0001 0.4987 -0.0556 0.0675*
Notes: ***, ** and * denote the statistical significance at 1%, 5% and 10%, respectively. SE are based on the
Crude Dependence Adjustment. This table contains the results of the t-test of the subgroups A, B and C for the three stress tests separately. It shows the CAARs and its significance during the five event windows using the synthetic bond approach. A right-sided t-test is used when the CAAR is positive and a left sided t-test is applied when the CAAR is negative. The degrees of freedom for 2010, 2011 and 2014 are shown in below.
Degrees of freedom 2010 2011 2014
A 47 33 32
B 6 6 6
C 2 7
24
Table 7: Cumulative average abnormal returns (CAARs) and results of the sign test for the subgroups
Time 2010 2011 2014
CAAR>0 CAAR<0 CAAR>0 CAAR<0 CAAR>0 CAAR<0
Group A [-9;-1] 0.0116** 0.9884 0.0000*** 1.0000 0.9297 0.0703* [-5;-1] 0.0116** 0.9884 0.0001*** 0.9999 0.9997 0.0003*** [-1;1] 0.8762 0.1238 0.0000*** 1.0000 0.9997 0.0003*** [0;5] 0.9999 0.0001*** 0.9979 0.0021*** 0.8694 0.1306 [0;10] 1.0000 0.0000*** 0.0063*** 0.9937 0.9655 0.0345** Group B [-9;-1] 0.9018 0.0982* 0.0095*** 0.9905 0.9096 0.0904* [-5;-1] 0.8909 0.1091 0.0095*** 0.9905 0.6980 0.3020 [-1;1] 0.8791 0.1209 0.0095*** 0.9905 0.9096 0.0904* [0;5] 0.8909 0.1091 0.7987 0.2013 0.6980 0.3020 [0;10] 0.9018 0.0982* 0.7987 0.2013 0.9096 0.0904* Group C [-9;-1] 0.3534 0.6466 0.1648 0.8352 [-5;-1] 0.3534 0.6466 0.8981 0.1019 [-1;1] 0.3534 0.6466 0.8981 0.1019 [0;5] 0.3534 0.6466 0.8981 0.1019 [0;10] 0.3534 0.6466 0.8981 0.1019
Notes: ***, ** and * denote the statistical significance at 1%, 5% and 10%, respectively. This table contains the
results of the sign-test of the subgroups A, B and C for the three stress tests separately. It shows the sign of the CAARs and its significance during the five event windows using the synthetic bond approach. The CAARs are tested for both signs with degrees of freedom as described in table 6.
25 research avenue for future studies would be to investigate the economic factors that lead to this seemingly counterintuitive results.
4.5 Sensitivity analysis
Table 8: Sensitivity analysis of the cumulative average abnormal returns using the t-test
Time 2010 2011 2014
P-value P-value P-value
Group A-B [-9;-1] 0.0018*** 0.0000*** 0.3181 [-5;-1] 0.1382 0.0000*** 0.1606 [-1;1] 0.1252 0.0004*** 0.1774 [0;5] 0.1022 0.0361** 0.4515 [0;10] 0.0226** 0.0002**** 0.3689 Group A-C [-9;-1] 0.0000*** 0.1332 [-5;-1] 0.0002*** 0.4325 [-1;1] 0.0000*** 0.3576 [0;5] 0.0554* 0.2786 [0;10] 0.0430** 0.0322** Group B-C [-9;-1] 0.0000*** 0.0461** [-5;-1] 0.0004*** 0.1624 [-1;1] 0.0002*** 0.0744* [0;5] 0.0282** 0.2190 [0;10] 0.2683 0.0132**
Notes: ***, ** and * denote the statistical significance at 1%, 5% and 10%, respectively. SE are based on the
Crude Dependence Adjustment and with an adjusted MSE. This table contains a sensitivity analysis for the differences between the CAARs of the subgroups for the three stress tests separately. It shows the p-values of the difference between the CAARs for the five event windows for the synthetic bond approach using a t-test. The CAARs are tested for both signs with degrees of freedom measure by a harmonic mean as described in equation 17.
26 In 2010, both the longer event window prior and after disclosure present a significant difference between the results of group A and B. As shown in table 6, the significance for group A was in both windows stronger than for group B. Therefore, the CDS spreads were more strongly affected for banks who passed the stress tests than for banks who nearly passed it.
27
Table 9: Sensitivity analysis difference in CAAR between group A and control group
Time Group A CAAR Control CAAR P-value
2010 [-9;-1] 0.0175 0.0181 0.4769 [-5;-1] 0.0119 0.0036 0.1778 [-1;1] -0.0109 -0.0012 0.1388 [0;5] -0.0677 0.0114 0.0000*** [0;10] -0.0977 0.0210 0.0000*** 2011 [-9;-1] 0.1354 0.0791 0.0000*** [-5;-1] 0.0880 0.0447 0.0000*** [-1;1] 0.0980 0.0441 0.0000*** [0;5] -0.0319 -0.0240 0.0814* [0;10] 0.0352 0.0161 0.0006*** 2014 [-9;-1] -0.0041 -0.0099 0.2381 [-5;-1] -0.0390 -0.0556 0.0218** [-1;1] -0.0347 -0.0054 0.0003*** [0;5] -0.0154 0.0121 0.0005*** [0;10] 0.0352 -0.0101 0.0000***
Notes: ***, ** and * denote the statistical significance at 1%, 5% and 10%, respectively. SE are based on the
Crude Dependence Adjustment and with an adjusted MSE. This table contains a sensitivity analysis for the differences between the CAARs of the control group for the three stress tests separately. It shows the p-values of the difference between the CAARs for the five event windows for the synthetic bond approach using a t-test. The CAARs are tested for both signs with degrees of freedom measured by a harmonic mean as described in equation 17.
28
4.6 Default risk quartiles
Credit quality or the riskiness of a firm can influence the effect of the disclosure of the stress test results. Hypothesis 6 determines whether riskier firms react differently to the disclosure of the outcomes than less riskier firms. Theory suggests that a riskier firm is more sensitive because it presents higher default risk and has lower credit quality. This section investigates this issue by splitting the banks of each stress test into different quartiles based on the absolute CDS spread on the last day of the estimation window, day t-10. The first quartile (q1) stands for the firms with the
lowest CDS spread in basis points, the fourth quartile (q4) includes firms with the highest CDS spreads on the last day of the estimation window. Thereafter, the CAARs of the quartiles are calculated and a sign test is applied to determine the sign and significance for the five event windows in the three time samples. The results of this test are shown in table 12.
In 2010, there is only a minor difference in significance for the less riskier and riskier firms for [0;5] The sign and significance for the other post-event windows is consistent for both quartiles. Furthermore, the sign and significance for q1 and q4 are reversed for the days prior to disclosure. It can be said that investors expected the CDS spreads of low risk firms to increase and the spreads of riskier firms to decrease. A reason could be that the risk is already known for q4 firms and that investors started to doubt the safer banks, since they had more to lose. However, at disclosure the sign for these safe banks switches and the CDS spreads decline.
The second stress test shows the opposite situation in the days prior to disclosure than the 2010 test. There is a significant negative sign for q1 and a positive sign for q4. Investors expected a bad outcome for the riskier firms and therefore the CDS spreads were increasing. This trend continued in event window [-1;1], for the safe firms spreads kept declining after disclosure but there was no significance for riskier firms on the longer term. It is interesting to see that this difference between q1 and q4 is also displayed when comparing q1 and q2. Around 75% of the data show the reversed sign, dominating the outcomes in earlier analyses.
In 2014 is a negative significant effect for the post-event windows of q1, the results for q4 are similar but less strong, the results are only marginally significant. There is no strong evidence for an effect based on riskiness. For both quartiles of banks, investors react the same, the only difference being in the anticipation period for the safest banks in the sample.
29
Table 10: Risk quartiles EU stress tests
Ha: CAAR > 0 Ha: CAAR < 0
1 2 3 4 1 2 3 4 2010 [-9;-1] 0.0089*** 0.0960* 0.0616* 0.9791 0.9911 0.9040 0.9384 0.0209** [-5;-1] 0.0229** 0.0960* 0.0245** 0.9460 0.9771 0.9040 0.9755 0.0540* [-1;1] 0.8421 0.7671 0.1430 0.8706 0.1579 0.2329 0.8570 0.1294 [0;5] 0.9984 0.8886 0.8513 0.9460 0.0016*** 0.1114 0.1487 0.0540* [0;10] 0.9984 0.9813 0.9356 0.9921 0.0016*** 0.0187** 0.0644* 0.0079*** 2011 [-9;-1] 0.9974 0.0193** 0.0230** 0.0192** 0.0026*** 0.9807 0.9770 0.9808 [-5;-1] 0.9974 0.0183** 0.0619* 0.0192** 0.0026*** 0.9817 0.9381 0.9808 [-1;1] 0.9974 0.0062*** 0.0073*** 0.0068*** 0.0026*** 0.9938 0.9927 0.9932 [0;5] 0.9974 0.9554 0.7940 0.7720 0.0026*** 0.0446** 0.2060 0.2280 [0;10] 0.9974 0.0183** 0.1601 0.3140 0.0026*** 0.9817 0.8399 0.6860 2014 [-9;-1] 0.9861 0.9275 0.8270 0.1655 0.0139** 0.0725 0.1730 0.8345 [-5;-1] 0.9983 0.9969 0.8270 0.7811 0.0017*** 0.0031*** 0.1730 0.2189 [-1;1] 0.9861 0.9969 0.9773 0.9067 0.0139** 0.0031*** 0.0227** 0.0933* [0;5] 0.7447 0.4112 0.9925 0.9067 0.2553 0.5888 0.0075*** 0.0933* [0;10] 0.9596 0.6407 0.9925 0.9067 0.0404** 0.3593 0.0075*** 0.0933*
Notes: ***, ** and * denote the statistical significance at 1%, 5% and 10%, respectively. This table contains the results of the sign-test of the quartiles based on absolute
CDS spreads for the three stress tests separately. It shows the sign of the CAARs and its significance during the five event windows using the synthetic bond approach. The CAARs are tested for both signs with degrees of freedom as stated below.
1 2 3 4
2010 13 14 13 13
2011 10 11 10 10
30
Additional Robustness check:
31
5 Comparative analysis
In this comparative analysis the comprehensive capital analysis and review (CCAR) for the years 2012-2015 is examined. This US stress test is an annual test performed by the Federal Reserve and controls if the largest banks can survive in periods of economic and financial stress and whether they have a reliable forward-looking capital plan to deal with different types of risk. Furthermore, this analysis can be used to compare the differences between the European and US banks, stress test outcomes and the reactions on the CDS market. Therefore, the same hypotheses will be tested for the US sample as can be seen in the methodology section.
Table 11: Results of the t-test for the cumulative average abnormal returns (CAAR)
Time 2012 2013 2014 2015
CAAR P-value CAAR P-value CAAR P-value CAAR P-value
Total [-9;-1] -0.0399 0.1767 -0.0313 0.1172 -0.0130 0.2317 0.0100 0.2346 [-5;-1] -0.0404 0.1738 -0.0286 0.1369 -0.0086 0.3135 0.0114 0.2054 [-1;1] -0.0342 0.2123 -0.0327 0.1073 0.0023 0.4471 0.0053 0.3484 [0;5] -0.0645 0.0728* 0.0313 0.1166 0.0059 0.3693 -0.0005 0.4841 [0;10] -0.0593 0.0890* 0.0798 0.0045*** 0.0041 0.4071 0.0316 0.0170** Group A [-9;-1] -0.0466 0.1370 -0.0249 0.1817 -0.0096 0.3167 0.0127 0.2597 [-5;-1] -0.0423 0.1589 -0.0207 0.2235 -0.0069 0.3660 0.0092 0.3197 [-1;1] -0.0349 0.2023 -0.0206 0.2243 -0.0005 0.4902 0.0036 0.4260 [0;5] -0.0677 0.0634* 0.0227 0.2025 0.0035 0.4305 -0.0118 0.2746 [0;10] -0.0702 0.0575* 0.0633 0.0214** 0.0047 0.4082 0.0138 0.2422 Group C [-9;-1] -0.0220 0.3566 0.0301 0.1602 -0.0115 0.4152 [-5;-1] -0.0125 0.4166 0.0279 0.1767 0.0270 0.3129 [-1;1] -0.0446 0.2363 -0.0042 0.4410 0.0054 0.3653 [0;5] -0.0663 0.1552 -0.0142 0.3104 -0.0006 0.4846 [0;10] -0.0585 0.1807 -0.1195 0.0054*** 0.0316 0.0724*
Notes: ***, ** and * denote the statistical significance at 1%, 5% and 10%, respectively. SE are based on the
Crude Dependence Adjustment. This table contains the results of the t-test of the treatment group and subgroups A and C for the four US CCAR stress tests separately. It shows the CAARs and its significance during the five event windows using the synthetic bond approach. A right-sided t-test is used when the CAAR is positive and a left sided t-test is applied when the CAAR is negative. The sample sizes are as described in the table below.
2012 2013 2014 2015
Total 11 10 13 14
A 8 7 9 12
C 3 1 4 2
32 2013 and 2015 and no significance in 2014. The latter is consistent with the parametric test outcomes in the EU in 2014. Also consistent with the EU sample is the negative significance of the first publicly disclosed stress test, an abnormal decrease in the CDS spread could indicate that the stress tests provides trust and stability in the market. Therefore, this results in a lower CDS spread or probability of default. Even though the event window [-1;1] is never significant, the null-hypothesis of no effect can be rejected for all years except 2014 when using the parametric t-test.
Table 12: Results of the sign test for the cumulative average abnormal returns (CAAR)
Time 2012 2013 2014 2015
CAAR>0 CAAR<0 CAAR>0 CAAR<0 CAAR>0 CAAR<0 CAAR>0 CAAR<0
Total [-9;-1] 0.8860 0.1140 0.9231 0.0769* 0.1674 0.8326 0.0135** 0.9865 [-5;-1] 0.9944 0.0056*** 0.9231 0.0769* 0.1674 0.8326 0.0344** 0.9656 [-1;1] 0.8860 0.1140 0.9915 0.0085*** 0.9469 0.0531* 0.3539 0.6461 [0;5] 0.9561 0.0439** 0.3392 0.6608 0.5397 0.4603 0.7549 0.2451 [0;10] 0.8860 0.1140 0.1519 0.8481 0.9920 0.0080*** 0.1855 0.8145 Group A [-9;-1] 0.8436 0.1564 0.6102 0.3898 0.2673 0.7327 0.0177** 0.9823 [-5;-1] 0.9799 0.0201** 0.9806 0.0194** 0.2673 0.7327 0.1235 0.8765 [-1;1] 0.6404 0.3596 0.9806 0.0194** 0.9620 0.0380** 0.6994 0.3006 [0;5] 0.8436 0.1564 0.9929 0.0071*** 0.5087 0.4913 0.9426 0.0574* [0;10] 0.8436 0.1564 0.9929 0.0071*** 0.9620 0.0380** 0.4809 0.5191 Group C [-9;-1] 0.7178 0.2822 [-5;-1] 0.8965 0.1035 0.5550 0.4450 0.4364 0.5636 [-1;1] 0.8965 0.1035 0.5550 0.4450 0.1751 0.8249 [0;5] 0.8965 0.1035 0.8338 0.1662 0.1751 0.8249 [0;10] 0.7178 0.2822 0.8338 0.1662 0.1751 0.8249
Notes: ***, ** and * denote the statistical significance at 1%, 5% and 10%, respectively. This table contains the
results of the sign-test of the treatment group and subgroups A and C for the four US CCAR stress tests separately. It shows the sign of the CAARs and its significance during the five event windows using the synthetic bond approach. The CAARs are tested for both signs with sample sizes as described in table 11.
33 Hypothesis 2 investigates whether the market anticipates abnormalities in the changes in synthetic bond yields and thus CDS spreads. Table 11 and 12 show respectively the results for the parametric and non-parametric methods to test for significance of abnormal returns in the pre-event windows [-9;-1] and [-5;-1]. Results of the t-test are not significant, indicating no anticipation of abnormal returns on the CDS market. However, the sign test shows a positive significant effect in 2015, for 2012 and 2013 a (marginally) negative significant effect is found and no effect in 2014. Therefore, the null-hypothesis should be rejected for all years except 2014. In 2015 the CDS market expected an increase in the CDS spreads, suggesting that disclosure of the results led to higher default probabilities. However, in the post-event windows no significant results are detected. In 2012 and 2013 the anticipated direction was in line with the significant signs after disclosure.
34 effect when publicly disclosing the outcomes of the test for the banks who failed. The CDS spreads and the perceived credit default risk of these banks does not abnormally change.
Table 13: Sensitivity analysis of the cumulative average abnormal returns using the t-test
Time 2012 2013 2014 2015
P-value P-value P-value P-value
[-9;-1] 0.1870 0.0028*** 0.0780*
[-5;-1] 0.1430 0.0061*** 0.1441
[-1;1] 0.3617 0.3805 0.4574
[0;5] 0.4798 0.0810* 0.2501
[0;10] 0.3339 0.0000*** 0.1441
Notes: ***, ** and * denote the statistical significance at 1%, 5% and 10%, respectively. SE are based on the
Crude Dependence Adjustment and with an adjusted MSE. This table contains a sensitivity analysis for the differences between the CAARs of the subgroups for the stress tests separately. It shows the p-values of the difference between the CAARs for the five event windows for the synthetic bond approach using a t-test. The CAARs are tested for both signs with degrees of freedom measured by a harmonic mean as described in equation 17.
Additionally, hypothesis 4 examines the possible asymmetric result depending on whether a bank passed or failed the stress test. The first discrepancy is the difference in significance in table 11 when comparing the t-test results of group A and C. For 2012, group A is significant but C not, and in 2014 and 2015 the opposite appears. However, the signs of the CAARs of both groups are somewhat similar. Table 13 shows a sensitivity analysis wherein the CAARs of the two subsamples are compared using a t-test. Only in 2014 there was a significant difference in the CAARs of group A and C for the post-event windows. The sign test displayed in table 12, as well shows the biggest difference between the groups in the year 2014 even though this test claims the CAARs to have a negative sign. The other years (disregard 2013) do not show these severe differences. The two tests present contrary situations, therefore an asymmetric effect cannot be clearly detected. It is also difficult to test these two groups due to their size. The null-hypothesis of H4 will not be rejected, indicating no obvious sign of an asymmetry reaction on the different outcomes.
35 all event windows. The CAARS for group A and the control group have opposite sign and show a different pattern. The subjected group (group A) has decreasing CAARs prior to disclosure but after disclosure they increase, for the control group the opposite is occurring. The last test, does not show a significant result for the long term [0;10] but does for the shorter term of [0;5] which presents a negative effect on group A and a positive for the control group. Overall, the null-hypotheses of no difference between group A and the control group can be rejected. There is a significant difference in the market reaction for firms subject to the stress test and those who are not directly affected. However, the sign and whether the results are contrasting is not consistent over the years.
Table 14: Sensitivity analysis differences in CAAR between group A and control group
Time Group A Control P-value
2012 [-9;-1] -0.0466 -0.0227 0.0635* [-5;-1] -0.0423 -0.0116 0.0258** [-1;1] -0.0349 -0.0131 0.0812* [0;5] -0.0677 -0.0475 0.0977* [0;10] -0.0702 -0.0375 0.0191** 2013 [-9;-1] -0.0249 -0.0486 0.0095*** [-5;-1] -0.0207 -0.0181 0.3969 [-1;1] -0.0206 -0.0103 0.1488 [0;5] 0.0227 0.0091 0.0846* [0;10] 0.0633 0.0249 0.0001*** 2014 [-9;-1] -0.0096 0.0170 0.0004*** [-5;-1] -0.0069 0.0131 0.0053*** [-1;1] -0.0005 0.0031 0.3174 [0;5] 0.0035 -0.0104 0.0359** [0;10] 0.0047 -0.0068 0.0673* 2015 [-9;-1] 0.0127 0.0011 0.0478** [-5;-1] 0.0092 0.0157 0.1721 [-1;1] 0.0036 0.0062 0.3578 [0;5] -0.0118 0.0007 0.0371** [0;10] 0.0138 0.0098 0.2841
Notes: ***, ** and * denote the statistical significance at 1%, 5% and 10%, respectively. SE are based on the
36
Table 15: Risk parts in US stress tests
Ha: CAAR > 0 Ha: CAAR < 0
Time Low High Low High
2012 [-9;-1] 0.8281 0.8560 0.1719 0.1440 [-5;-1] 0.9394 0.9819 0.0606* 0.0181** [-1;1] 0.5847 0.9499 0.4153 0.0501* [0;5] 0.8281 0.9499 0.1719 0.0501* [0;10] 0.8281 0.8560 0.1719 0.1440 2013 [-9;-1] 0.5606 0.5723 0.4394 0.4277 [-5;-1] 0.5606 0.9457 0.4394 0.0543* [-1;1] 0.9433 0.2506 0.0567* 0.7494 [0;5] 0.8318 0.5723 0.1682 0.4277 [0;10] 0.5606 0.8386 0.4394 0.1614 2014 [-9;-1] 0.7519 0.0704* 0.2481 0.9296 [-5;-1] 0.7519 0.0704* 0.2481 0.9296 [-1;1] 0.9672 0.8850 0.0328** 0.1150 [0;5] 0.9061 0.8850 0.0939* 0.1150 [0;10] 0.7519 0.9594 0.2481 0.0406** 2015 [-9;-1] 0.1918 0.0288** 0.8082 0.9712 [-5;-1] 0.4363 0.0288** 0.5637 0.9712 [-1;1] 0.7163 0.2243 0.2837 0.7757 [0;5] 0.7163 0.7510 0.2837 0.2490 [0;10] 0.1918 0.4829 0.8082 0.5171
Notes: ***, ** and * denote the statistical significance at 1%, 5% and 10%, respectively. This table contains the
results of the sign-test of the high and low risk groups based on absolute CDS spreads for the four stress tests separately. It shows the sign of the CAARs and its significance during the five event windows using the synthetic bond approach. The CAARs are tested for both signs with degrees of freedom as stated below.
2012 2013 2014 2015
Low 6 5 7 7
High 5 5 6 7
37 again presents significant positive CAARs in the pre-event windows for the high risk group, however this effect does not continue after disclosure. Summarizing, the two risk groups do not have significant differences in sign or significance, for the post-event windows. Indicating that there is no difference in sensitivity to the disclosure of the stress test outcomes for low and high risk banks. This could be due to the fact that there are only two subgroups with no clear margin between them, which could bias the results. Further research with more data could give a more decisive answer concerning the differences between low and high risk banks. However, in the days prior to disclosure there is a positive effect for high risk banks in 2014 and 2015, these results are in line with the outcomes of the European stress test.
Due to the fact that the two geographical areas have slightly different test methods and cover different time frames a direct comparison cannot be made. However, some comparisons regarding the reactions and market mechanisms could be made. In comparison with the EU sample, the US sample also confirmed the hypothesized effect of a significant impact on the CDS market by publicly disclosing the outcomes of the stress test. The overall sign was negative which indicates good news on the CDS market. The spreads decreased so the risk decreased, due to more trust and transparency on the market. Hypothesis 2, considering the anticipation of the market on the direction of the stress test outcomes, gave similar results for across the geographical areas. For both markets there was some degree of anticipation. When looking at the impact on the banks that passed the test, the effect was again similar. Both geographical areas had the hypothesized negative sign and hypothesis 3a was confirmed. For hypothesis H3b, wherein the banks who failed were examined, there was for both samples no evidence. In the EU sample hypothesis 4 is confirmed for all years, the effect was stronger for those firms who passed in 2010 and 2011, but in 2014 the effect was reversed and the banks that failed had a stronger reaction in the CDS market. Even though there were differences in sign and significance between both groups in the US sample, there was not enough evidence for an asymmetric effect in this sample. Therefore, the results between the EU and US are contrasting for this hypothesis. With regards to hypothesis 5, both countries show similar results, demonstrating significant differences in the impact between firms subject to the test and other financial firms which were not tested. Lastly, dividing the groups based on riskiness does not result in significant differences after disclosure for both samples.7 Although, in the days prior to disclosure some
differences can be observed.
38
6 Conclusion
This thesis examines the effect of both EU and US stress tests on the CDS market, it determined whether the public disclosure of the stress tests added informational value to the market and how this information was interpreted by investors. By adding the most recent EU stress test of 2014, comparing the outcomes across the two different geographical areas, adding complementary non-parametric tests, and by justifying and explaining the use of synthetic bonds, this paper is a valuable contribution to the current literature.
The empirical results show that the disclosure of the stress test outcomes has a significant effect for both the EU as US sample. With the exception of the second European test, there is a negative effect which leads to decreased CDS spreads and lower credit default risk after the public disclosure of the outcomes. Thus, excluding 2011, where a positive effect was found, the EBA and the FED achieved its goal of restoring confidence in the EU and US banking system. Due to increased transparency and the confirmation of the financial fitness of banks, uncertainty in the market decreased. The investors reacted to the announcement as positive news and as valuable information. It is also examined whether the CDS spreads in the pre-event windows anticipate the outcome of the stress tests. This could be due to information leakages, market expectations or investor speculation. For almost all years in both samples there were abnormal returns in the days prior to disclosure and, in most cases, the direction of these abnormal returns corresponded to the sign in the post-event windows. The CDS spreads already adjusts before announcement and therefore the effect could be bigger than measured in the post-event windows.
39 null-hypothesis. Further research should focus on retrieving more data to avoid these problems and provide further evidence on these hypotheses.
Furthermore, a test of the differences in effects between banks subjected to the stress and those not-subjected was performed. The hypothesis is confirmed for both the EU as the US, implying that the stress test itself has an effect on the banks subject to it. This could be due to the publication of private information which has a soothing effect and gives some reassurance concerning the financial fitness of the bank.
Lastly, it is examined whether riskier banks are more sensitive to news about its credit quality. However, only in the second EU stress test significant differences between the safe and riskier banks are found. For riskier banks the CDS spreads increase whereas the CDS spreads for safer banks decrease. This could also explain the odd outcome of the second EU stress test, possibly more riskier banks are present in this sample increasing the abnormal returns and biasing the outcomes. Similarly, in the US sample an increase of the CDS spreads for riskier banks can be found in the days prior to disclosure.
40
7 References
Alves, C., Mendes, V., Pereira da Silva, P., 2015. Do stress tests matter? A study on the impact of the disclosure of stress test results on European financial stocks and CDS markets. Applied Economics 47, 1213-1229.
Basel Committee on Banking Supervision, 2004. International Convergence of Capital Measurement and Capital Standards: A Revised Framework. June.
Brown, S.J., Warner, J.B., 1980. Measuring security price performance. Journal of Financial Economics 8, 205-258.
Cardinali, A., Nordmark, J., 2011. How informative are bank stress tests? Bank opacity in the European Union. Master’s thesis, Lund University.
Campbell, C.J., Cowan, A.R., Salotti. V., 2010. Multi-country event-study methods. Journal of Banking and Finance 34, 3078-3090.
Cowan, A.R., 1992. Nonparametric event study tests. Review of Quantitative Finance and Accounting 2, 343-358.
Ellahie, A., 2012. Capital market consequences of EU bank stress tests. Working Paper, London Business School.
Fama, E.F., 1970. Efficient capital markets: a review of theory and empirical work. The Journal of Finance 25, 383–417.
Federal Reserve, 2012. Comprehensive Capital Analysis and Review 2012: Methodology and results for stress scenario projections. Federal Reserve, Washington DC.
URL: http://www.federalreserve.gov/newsevents/press/bcreg/bcreg20120313a1.pdf
Federal Reserve, 2013. Comprehensive Capital Analysis and Review 2013: Assessment framework and results. Federal Reserve, Washington DC.
41 Federal Reserve, 2014. Comprehensive Capital Analysis and Review 2014: Assessment framework and results. Federal Reserve, Washington DC.
URL: http://www.federalreserve.gov/newsevents/press/bcreg/ccar_20140326.pdf
Federal Reserve, 2015. Comprehensive Capital Analysis and Review 2015: Assessment framework and results. Federal Reserve, Washington DC.
URL: http://www.federalreserve.gov/newsevents/press/bcreg/bcreg20150311a1.pdf
Harrington, S., 2006. Credit-default swap traders anticipated announcements of LBOs. Available at Bloomberg.com (accessed 11 November 2015).
Hull, J., Predescu, M., White, A., 2004. The relationship between credit default swap spreads, bond yields, and credit rating announcements. Journal of Banking and Finance 28, 2789-2811.
Hull, J.C., 2012. Risk Management and Financial Institutions. John Wiley & Sons, Inc., Hoboken, New Jersey.
MacKinlay, A. C., 1997. Event studies in economics and finance. Journal of Economic Literature 35, 13-39.
Micu, M., Remolona, E., Wooldrigde, P., 2006. The price impact of rating announcements: Which announcements matter? Bank for international settlements, Working paper 207, June 2006.
Morgan, D.P., Peristiani, S., Savino, V., 2014. The information value of the stress test. Journal of Money, Credit and Banking 46, 1479-1500.
Neretina, E., Sahin, C., de Haan, J., 2015. Banking stress test effects on returns and risks. De Nederlandse Bank, Working Paper 104.
Norden, L., Weber, M., 2004. Informational efficiency of credit default swaps and stock markets: The impact of credit rating announcements. Journal of Banking and Finance 28, 2813-2843.
42 Petrella, G., Resti, A., 2013. Supervisors as information producers: Do stress tests reduce bank opaqueness? Journal of Banking and Finance 37, 5406-5420.
Schaefer, A., Schnabel, I., di Mauro, B. W., 2013. Financial sector reform after the crisis: Has anything happened? CEPR Discussion Papers 9502.
Weistroffer, C., Speyer, B., Walter, N., 2009. Credit default swaps: Heading towards a more stable system. Deutsche Bank Research.
Websites:
European Banking Authority:
http://www.eba.europa.eu/risk-analysis-and-data/eu-wide-stress-testing assessed at 14-08-2015
Government yields synthetic bonds EU:
http://sdw.ecb.europa.eu/quickview.do?SERIES_KEY=165.YC.B.U2.EUR.4F.G_N_A.SV_C_YM.SR_5Y&s tart=28-04-2014&end=10-11-2014&submitOptions.x=88&submitOptions.y=3&trans=N
assessed at 6-10-2015
Government yields synthetic bonds US:
