Banking Stress Test Effects
on CDS Spreads
Master thesis by Lisanne Nijhuis1 Supervised by M.J. Gerritse
ABSTRACT
This paper uses an event study to examine whether the announcement and disclosure of the results of stress tests affects market certainty over the period 2010-2014. It examines the impact of banking stress tests in Europe on participating banks’ CDS spreads. Furthermore it makes a distinction between CDS spread reaction of PIIGS and non PIIGS banks. The data set includes Europe’s stress tests of 2010, 2011, and 2014. The empirical results indicate that only in 2011 the disclosure of the stress test reduces the market confidence, at all other significant events the CDS spreads decline; indicating more certainty. Furthermore, the results on individual bank level show that overall the non PIIGS banks react stronger to a stress test announcement or result than PIIGS banks. This finding can be explained by deposit insurance: a higher coverage ratio leads to less response to a stress test. The CAAR’s can be explained by liquidity ratios and capital ratios. Noteworthy, the poorer performance of PIIGS banks in 2014 can be partially explained by poorer liquidity ratios. Lastly, this study finds evidence that the market does not know beforehand, i.e. the announcement date, which bank will fail the stress test.
JEL classification G14; G21; G28;
Key words: stress testing, banks, CDS spreads.
MSc International Financial Management MSc International Economics & Business University of Groningen, the Netherlands Faculty of Economics & Business
1
2
Introduction
Nowadays, the reputation of banks is an important topic in international financial economics. Throughout the recent US subprime mortgage crisis and Eurozone sovereign debt crisis, the quality of assets held by European banks has declined. Uncertainty about the financial healthiness of EU member states further decreased expected growth and market confidence. This in combination with an economic downturn, became a big concern for some European countries, especially Southern Euro countries like Portugal, Ireland2, Italy, Greece and Spain (PIIGS). These countries have received a
comprehensive rescue package of 750 billion Euro; financed by the other European countries (Oliver, 2013).
To regain publics’ trust, more and more authorities have developed a stress test during the recent financial crisis (Ong and Pazarbasioglu, 2013). Following the Federal Reserve (FED), the European Central Bank (ECB) develops a stress test which tests the financial healthiness of banks under unfavorable market circumstances. The EU stress test include information about credit risk exposures of main European banks (Bischof and Daske, 2012). The aim of the test is bilateral, on one hand it is created for National Authorities as a common tool to optimize their supervisory of banks’ resilience to relevant shocks and to remove uncertainty, by identifying weaknesses, uncertainties and perform suitable actions. On the other hand, the test has to strengthen market discipline and restore confidence in the EU banking sector. Hereby transparency, and the quality of disclosure is important (EBA (a), 2014). Previous research finds that net trust3 in the ECB has declined from 40 percent before the crisis
to 10 percent nowadays (Wälti, 2012). In times of crisis, stress test have become an important tool for bank supervisors and managers. It is no longer sufficient to wait for the final stress test results, an active attitude is essential in order to prevent bad outcomes of the stress test. These management implications, before and after the stress test, could be used as a supervisory tool to evaluate the resilience of banks to adverse economic developments (EBA (a), 2014). The aim of the stress test is to improve the information environment during the recent global financial crisis. The disclosure of the results of the stress test have to increase transparency in the banking sector, with the emphasis on banks’ sovereign risk, reduce information asymmetry and restore investors’ confidence (Alves, 2014). The main question by a stress test is if the release of the stress test results add new information to the investor. Similarly, does the announcement give new information to the market? Do the results differs between PIIGS banks and Non PIIGS banks? The purpose of this paper is to investigate if the announcement and disclosure of stress tests provides new information to the European market. It will be a following up study of Cardinali and Nordmark (2011), who investigate the reaction of the stock
2
Not a Southern country. however, it is facing the same problems as the Southern countries, therefore, it is included in the PIIGS.
3
3 market after the announcement and methodology disclosure of the EU stress test in 2010 and 2011. They suggests to investigate how the CDS market reacts to the stress test. Furthermore Norden and Weber (2004) observe that CDS spreads reacts faster than stock prices to news. This study will therefore examine changes experienced in CDS spreads of European banks after the announcements and disclosure of the stress test results. An event study is used to observe the impact of EU stress test announcement and disclosure of the results on banks’ CDS spreads over the 2010–2014 period. The data set includes Europe’s stress tests of 2010, 2011, and 2014.
For several years stress testing has been an important risk management tool for supervisors. Stress test allow supervisors and managers to understand the financial risk, which is relevant for decision making. For example at managerial level, stress test provide an overview of risky assets that can be compared with other participating banks. For executive level the test is a manner to compare the portfolio of the bank with the preferences of the owners (Blaschke et al., 2001). Stress test have become therefore the new standard in financial crisis management.
This research is motivated by the following. First, The EU stress test include a rich and unique data set for each participating bank, including up to 3,400 data points (Petrella and Resti, 2012). The test, examine the financial healthiness of banks under unfavorable market circumstances. Scenarios as lower economic growth, falling housing prices and declining stock markets have provided valuable information to the ECB in order to determine how the capital position of banks would develop under these different scenarios. The data of the three EU stress tests provide an exclusive set of financial determinants, making it possible to examine how the market will react to these determinants. This market reaction is interesting to measure while, the aim of this tests is ‘to measure the resilience of financial institutions to adverse market developments, as well as to contribute to the overall assessment of systemic risk in the EU financial system’ (EBA (b), 2014).
4 assumes that the stress test could have no or only a small effect to the financial market (Marray, 2014). This raises the question if the EU stress test still is an adequate measurement, and if it provides accurate provides information to their environment.
A third motivation is the fact that the banks that have failed are nearly all located in Southern Europe. These countries will have a significant impact on the future stability of the Eurozone. Portugal, Ireland and Greece received rescue packages, Spain and Italy have to deal with massive economic troubles. Banking crises are primarily caused by micro-economic factors, however also macro- economic factors are likely to have influence on the banking system. The most hazardous factors of banks are their high reliance on creditors’ resources, risky claims on different sectors, and finally that fact that almost all assets are long term and less liquid than their liabilities. Additionally, Arpa et al. (2001) argue: ‘ Banks’ health reflects to a large extent the health of their borrowers, which in turn reflects the health of the economy as a whole.’ This statements makes it interesting to make a distinction between ‘healthy’ Eurozone countries and the PIIGS countries and investigate if the stress test have a different impact on countries with poor macro-economic conditions. According to Brazys and Hardiman (2013) there is a correlation between economic wealth and trust: the more wealthy an economic the more citizens express trust in the institutions. Moreover, it is interesting to evaluate if the stress test provides new information to investors about their healthiness. Does the impact differs between banks that are located in ‘healthy’ Eurozone countries as opposed to banks that are located in weak Eurozone countries?
5 The paper contributes to the existing literature because it takes into account all European stress test from 2010 to 2014. Especially including the 2014 test makes this research interesting, the 2014 version being more strict and extensive than ever before. Moreover it test directly the effect on CDS spreads, instead of a Alves et al. (2014), who measure the impact on returns of synthetic bonds. Besides this, it takes into account the distinction between PIIGS countries and non PIIGS countries. The paper is structured as followed. Section 2 provides a summary of related literature on stress testing. Section 3 provides an overview of the main features of European stress tests. Section 4 presents the testable hypothesis. Section 5 outlines the data and methodology. Section 6 presents the main findings, discussions and robustness checks. Section 7 concludes.
Literature Review
Literature based on the disclosure of stress test outcomes is scarce, most articles focus on the US stress test. However there are differences between the US and the EU stress test, making it interesting to examine the effect of the disclosure of the EU stress test on returns. Some critics argue that due to the more flexible definitions of core capital allowed by the EU, the EU stress test is easier to pass than the US version. The main reason for the differences in stress test approach of the US and EU, is the size of the financial system. The European bank system is three times the size of the European economy, the US banking system is around one time the size of the US economy. Bad results of the EU stress test could affect the whole economy (Barnato, 2014). According to research of the AFME (2014), bank loans remain the main source of financing of the EU economy. Around 70 percent of the EU economy is financed with debt, compared with 30 percent in US (AFME, 2014).
Following the efficient market hypothesis theory of Fama (1970), and existing literature who investigate the information component of the stress test to the market, stock prices will directly reflect all public available information. According to the theory of Fama, at all times stocks trades at fair value on stock exchanges. In efficient markets, stock prices will only change after publishing new relevant information. If stress test results improved the information environment stock prices should reflect these information.
6 methodology has impact. Their main explanation for this result is that the stress test is viewed as too easy. Furthermore theyfind that the announcement of using methodology in 2011 test gives negative CARs for tested banks while untested banks remain unaffected. Surprisingly, they did not find a difference between PIIGS and non PIIGS banks. Only significant negative abnormal returns have been found on the release day, their main explanation is unease of the market to the results.
However Beltratti (2011), who investigated if stress tests carries new information, argued that the results of the 2011 stress test provides new and relevant information to the market. His main argument is that investors could not forecast the main outcomes of the test, with the information they already know. Furthermore, research of Petrella and Resti (2012) examined whether and how the 2011 European stress test affected stock prices. They found significant market reactions to the announcement, methodology and disclosure of the stress test, which confirms the research of Beltratti, that the stress test provides new information to the market. Moreover they found that the market is not able to anticipate the stress test results.
Alves et al. (2014) investigated the effect on stock price data after the disclosure of European stress test in 2010 and 2011. Their main finding is both European stress tests results have affected stock prices of banks. The 2010 stress test reduced the volatility in stock prices while the volatility increased after the release of 2011 results. This indicates that investors attach value to the results of the stress test. A small remark must be made here, the fact that not all EU banks are listed may influence the results. Furthermore, research of Ellahie (2012) examined 2010 and 2011 European stress test on equity and credit ask spreads. They found reduced information asymmetry (i.e. equity-credit bid-ask spreads) and increased information uncertainty (measured by equity option implied volatilities and ratio of CDS spreads) of banks after the 2011 stress test. This finding is in line with the findings of Alves et al. (2014). They found evidence that the stock market reacts to information disclosure provided by the stress test, but the CDS market has a reverse reaction, which may indicate lack of trust by investors in the reasonability of 2011 stress test outcomes. This result is in line with criticism of analysts, who argue that Europe’s stress test is not strict enough. Byström (2009) confirms this reverse reaction of CDS markets, whereas they found if stock prices fall, CDS spreads have the tendency to widen. This assumes a negative relation between stock prices and CDS spreads. Moreover, research of Hintz et al. (2011) found that after the results of EU 2011 stress test, stock prices for almost all European banks declined, and, in line with the reverse relation, CDS spreads increased. Banks with increasing spreads were mainly located in Greece, Italy and Spain. Their main explanation is that the market did not expect the high capital shortfalls which were announced.
7 perception of sovereign risk. Furthermore CDS premiums are a reliable indicator of bank risk (Meine, 2013).
European Stress tests
The main idea behind the European Stress test is to give a rich set of information to the market about individual bank’s strengths and weaknesses. This has to reduce market uncertainty, stabilizing stock prices and prevent panic in the market (Petrella and Resti, 2012). The test tests the financial healthiness of banks under unfavorable market circumstances. Scenarios as lower economic growth, falling housing prices and declining stock markets have provided valuable information to the ECB in order to determine how the capital position of banks would develop under these scenarios. The aim of these tests is ‘to measure the resilience of financial institutions to adverse market developments, as well as to contribute to the overall assessment of systemic risk in the EU financial system’ (EBA (b), 2014).
The Committee of European Banking Supervisors (CEBS) published the first European stress test results in 20094. This study will not include these test beause it did not identify individual banks weaknesses nor it provides data about individual banks.
The European economic environment was changing in 2010. Concerns about a double dip recession, high unemployment rates, weak economic growth and the healthiness of countries like Greece and Spain declinesd the confidence in the Euro area even more. Furthermore, several Southern Euro countries were facing large deficits, huge public debt and downgrading credit ratings. On May 2, 2010, after the announcement of the bailout package to Greece, financial markets feared that other countries either could have problems to meet their obligations. This resulted in declining sovereign bond prices and raising yields, especially in Portugal and Spain (Blundell-Wignall and Slovik, 2010). Declining sovereign bond prices weakened the financial condition of European banks even further, since many government securities are held by banks. In order to restore the market confidence, the CEBS announced on June 18, 2010 a new European stress test. 91 banks of the European member states were selected, representing 65 percent of total assets in the EU banking sector and covering at least 50 percent of national banking sectors. (CEBS, 2010). Only systemically important banks based on asset size were selected, these are banks that are too-big-to-fail5. Different from the 2009 test,
individual results of the banks were disclosed, using detailed bank data and supervisory information. The objective of the 2010 test is to ‘provide policy information for assessing the resilience of the EU banking system to possible adverse economic developments and to assess the ability of banks in the exercise to absorb possible shocks on credit and market risks, including sovereign risks’ (CEBS, 2010). The 2010 test mainly focused on credit and market risks, including sovereign debt. Two macro-economic scenarios have been tested, the baseline and adverse scenario. The baseline scenario is based
4
The results of 2009 stress test were confidential (http://www.eba.europa.eu/-/cebs-s-statement-on-stress-testing-exercise). 5
8 on the economic forecast of the European Commission in 2009. The adverse scenario was based on ECB estimates, it reflects market conditions in the adverse scenario. The test indicates that seven banks fell below the threshold of 6 percent Tier 1 capital ratio.6 Five banks from Spain, one from
Greece and one German bank failed the 2010 test. These seven banks had to raise their capital by 3.5 billion euros, much less than expected by the market (Slater and Ortiz, 2010). The aim of the 2010 stress test was to restore market confidence after the problems with several Euro countries like Greece and Ireland, but was it successful? Critics argue that the test was too easy, doubting about the credibility of the test. Just four months after passing the test, two Irish banks had to be saved (Jenkins and Masters, 2011). Others argue that the severity of the test does not matter because the European economy is rapidly changing (Slater and Ortiz, 2010). The expectations of the market were that the capital shortfall was over 30 billion euros, instead of the resulting 3.5 billion. Furthermore, experts argue that the test did not include systematic concentration risk, a contagion effect is not measured (Balzli et al, 2010). On top of that Blundell-Wignall and Slovik (2010) argue that the 2010 test is not sufficient enough because they only test banks’ trading books and not banking books. Additionally the test makers did not take into account the possibility of a default of a country, therefore changes in banking books would not occur. However this is doubtful when taken into consideration the situation of Greece.
After the 2010 test, the European Banking Authority (EBA) conducted, in stand of the CEBS, the new stress test. The European market was still in a crisis after the results of the 2010 stress test. In August 2011, The Dow Jones dropped 4.3 percent, the biggest single-day loss since 2008, because of the market’ concerns of a new recession (double dip)7. Furthermore, Portugal and Greece received
rescue packages.8 At the end of 2010, concerns about the financial health of some euro countries led to
falling sovereign debt prices and rising CDS prices. Irish banks received, soon after passing the test, a bailout package, and concerns about the healthiness of Spain and Portugal were raising. The confidence in banking sector declines ever further. On January 13, 2011 the EBA announced a new stress test. The objective is to assess the resilience of the EU banking system and the solvency of individual institutions (European Court of Auditors, 2014). According to the EBA, this version will improve the transparency of the stress test and it took criticism of the 2010 version into account. The sample of the 2010 and 2011 stress test is similar. Again an adverse and a baseline scenario were used. However, the 2011 test takes into account the weak financial situation of national governments, which leads to a further decline of some EU bonds value which affects European banks. Furthermore, main macroeconomic indicators such as GDP, unemployment and declining housing prices were included which was not the case in 2010 test. Moreover the Tier 1 ratio is 5 percent, in stand of 6 percent used in 2010. The adverse scenario has been made more severe, but EU sovereign bond holdings in banking
6
Tier 1 capital ratio is the ratio between equity of a bank and the value of risky loans outstanding. 7
http://moneymorning.com/2014/02/13/stock-market-crash-history-dows-10-biggest-one-day-plunges/accesses 1 June, 2015. 8
9 book were again excluded (Petrella and Resti, 2012). Additionally transparency increased because communication of the sample and the details on the scenarios and methodology (EBA, 2011). On July 15, 2011 the results were disclosed. 20 banks would have fall below the threshold of 5 percent, the capital shortfall was 26.8 billion euro. However, because banks were allowed to raise their capital in the first four months of 2011, they were able to strengthen their capital position and still pass the stress test. On July, 15 the results show that eight banks fall below the threshold, with a capital shortfall of 2.5 billion. Banks who failed were located in Spain (5), Greece (2) and Austria (1). Sixteen other banks held a Tier 1 ratio between 5 and 6 percent which is alarming (EBA, 2011). Again, this stress test received a lot of criticism, while the Tier 1 ratio of 5 percent would have been too easy and the possibility of a Greek default has not been taken into account. Three months after, Dexia, a large Belgium-French bank who had passed the stress test, needed to be rescued. Three months before, the test claimed "no need for Dexia to raise additional capital" (EBA, 2011). Also Bankia, a Spanish bank, needed to be rescued just a couple of months after passing the stress test. Research of the European Court of Auditors (2014) towards the problems with the 2011 tests conclude the following. The large sample, the limited time and the fact that the EBA did not make the scenarios and calculations by themselves makes the tests more complex. Above all they did not have directly access to the banks making it difficult for the EBA to make the results reliable. Furthermore, the quality of assets were not included, nor banks’ liquidity.
On January 31, 2014 the EBA, in cooperation with the ESRB, announced a new stress test. In anticipation of the banking union, the healthiness of banks was tested again. The objective of the EU‐ wide stress test is ‘‘to assess the resilience of banks in the EU to adverse economic developments, helping supervisors assess individual banks, contributing to understanding systemic risk in the EU and fostering market discipline’’ (EBA (b), 2014). The sample differs from the 2010 and 2011 version, 123 European banks were tested, including Norwegian banks. These sample represent 70 percent of the European banking sector and 50 percent of the national banking sector. The assessment consisted of three components: an Asset Quality Review (AQR), a stress test and a prudential risk assessment (EBA (b), 2014). Combining the stress test with the AQR, should ensure the data is ‘pure’; a prerequisite for credibility. Furthermore, it increases the transparency of banks healthiness and it has to increase market confidence.
10 the opportunity to raise money. On October 27, 14 banks failed the test, with a capital shortage of 10 milliard9.
Again the results were published in a time were the Eurozone faced challenges with their financial strength. Growth forecasts had been cut and people were aware of spending money, which could lead to a triple dip recession. The EBA and ESRB attempts to restore market confidence in the financial sector, by pushing banks to provide more loans and so boost economy growth (Treanor, 2014). But then again were all these changes sufficient to restore market confidence?
PIIGS and non PIIGS
Already prior to the financial crisis the PIIGS banks were facing major imbalances and high levels of debt, these imbalances were increasing during the financial crisis. According to the Finnish Prime Minister, Jyrki Katainen: ‘the Eurozone's crisis had created a lack of trust between the Northern countries and Southern European countries.’ During the crisis, consumers and companies of the weaker euro countries have transferred their money to the northern banks because they did not trust their own bank anymore. Furthermore, Southern banks were facing deficits though investors and other banks did not want to borrow them money (Marray, 2014). Resulting in the fact that Southern banks rely more and more on the ECB, while strong northern banks have enough capital. Because of the economic crisis, Northern banks lend less to customers or companies, nevertheless Northern banks get more capital inflow from the Southern countries. Northern banks built up more buffers, whereas Southern banks have increasing debt levels. Banks are still reluctant to provide credit to the market, which have a negative impact on the recovery of the economy. This influenced the confidence in the Southern part of the Eurozone even more, already facing high levels of debt and unemployment. To protect tax payers for a potential bank bailout, the European Banking crisis have contributed to the formation of safety net agreements (Rhee, 2013). Government’s favors a deposit insurance since it confirms banking system stability and it helps to protect depositors against bank failures (Kalu, 2014). These safety nets includes deposit insurance, lender of last resort facilities of the central bank, and support of institutions like IMF (Demirgüç-Kunt et al., 2002) The aim of a safety net is to prevent a bank from bailout, bank runs and to reduce the cost of bank bailouts for the society (Gropp et al, 2011). Previous research has investigated the relationship between the attendance of a safety net and excessive risk taking of banks. Research of Dam and Koeter (2011) found that banks with more nonperforming and customer loans, higher cost-income ratios (which is the indicator for managerial inefficiency), and more low-yield liquidity taking more risk than others. Furthermore, MacDonald (1996) argues that the existence of a deposit insurance eliminates market discipline, causes excessive risk taking and depositors do not longer feel the incentive to make a proper decision were deposit money in. Wheelock (1992) found that because of the deposit insurance, which absorbs parts (or all)
9
11 of the losses at bank failure, banks are taken more and more risk. Demirgüç-Kunt et al., 2002 argue that deposit insurance reduces the encouragement of depositors to screen banks, which might lead to excessive risk-taking. Banks attracts depositors by offering high interest rates, trying to earn money with high risk loans. Because of the deposit insurance, both banks and depositors know that if the high risky loans cannot be repaid, they will be protected by the deposit insurance. This is called moral hazard: if you are protected by an insurance for the consequences of risk, you feel the incentive to take more risk. According to data of the IMF, 50 percent of the PIIGS banks outstanding loans are provided to credit unworthy companies. Furthermore, the amount of provided loans to these credit unworthy companies, have increased before and during the crisis. This is another reason why it is interesting to investigate differences between the PIIGS and non PIIGS countries. Within this study it could be the case that Southern European banks do not react to the stress test they know beforehand that they will be rescued by a potential bailout. Or perhaps it could explain Cumulative Abnormal Returns like the research of Ötker-Robe et al. (2010) applied.
In order to investigate if there are differences between PIIGS and non PIIGS banking sheets like the economic situation might suggest, the indicator nonperforming loans used by Dam and Koeter (2011), but also indicators of the stress test like Tier 1 and Regulatory Capital to Risk-Weighted-Assets have been collected from the IMF database to graph and display the possible differences. As the visual inspection shows (see Appendix), the PIIGS countries have indeed poorer ratios compared to non PIIGS banks which indicates a difference between PIIGS and non PIIGS. Therefore it would be plausible that the two groups react significantly different to the stress tests.
Hypothesis testing
12 significant positive abnormal CDS spreads will be expected when the stress test has negative information. No change in CDS spreads will be expected when the results are positive. Furthermore, no change in CDS spreads will be expected when the stress test does not provide new information to the market. To test if the disclosure of the EU stress test include new information the following hypothesis will be tested.
H1a: The CDS spreads of participated banks widens more which results in more uncertainty to the announcement of a new stress test.
H1b: The CDS spreads of participated banks widens more which results in more uncertainty at disclosure of the results of stress test.
The PIIGS countries are grouped together due their relatively high level of debt and they are the less creditworthy countries of the European Union. Currently, rescue packages have been provided to three of the five PIIGS countries, these countries were treated by bankruptcy (Cardinali and Nordmark, 2011). Uncertainty of the financial healthiness of these countries has declined growth and market confidence even further. Bischof and Daske (2012) found that after the 2011 test, CDS spreads declines for PIIGS stress test participants. Indicating that the market found the results favorable and an increase of uncertainty. However they did not make a comparison with non PIIGS countries. While PIIGS countries are heavily indebted countries, facing economic downturn, and received rescue package it is interesting to investigate if the announcement and results of the stress test has another influence in the debt market of PIIGS and non PIIGS countries. Furthermore, due the capital flows from South Europe to the Northern banks, the capital position of Southern banks have declined even further, while the Northern banks strengthen their position. The assumption can be made that there are indeed differences between the two banking groups, which may lead to a different response of the financial market to the stress test announcement and outcomes. Moreover, because of the provided loans of the Southern banks to credit unworthy companies, the results in 2014 may have a stronger impact. In 2014 the AQR has been taken into account which might influence the uncertainty in the market even more, due the balance sheet valuation. Though the PIIGS countries facing economic difficulties, the uncertainty in their financial healthiness and the most failing stress test banks are located in a PIIGS country, the expectation is that CDS spreads of PIIGS countries widen significant more than non PIIGS countries. Widening CDS spreads will increase the uncertainty in these countries even more. The following hypothesis will be tested.
H2a: the CDS spreads of PIIGS banks widens more than CDS spreads of non PIIGS banks to the announcements of a new stress test which increase the uncertainty in those countries.
13
Data
The stress test sample of 2010 and 2011 included 91 European banks from 21 countries who representing approximately 65 percent of the total assets of all European banks (Alves et al. 2014). The stress test sample of 2014 included 130 European banks (including Norway), which covered at least 50 percent of the national banking sector in each EU Member State and Norway (EMU, 2014) (see Appendix ). The banks included in this paper are only those banks that had daily CDS spreads for the particular estimation windows available in Thomson Reuters Datastream. This studies sample include 49 banks in the 2010 tests and 27 of the 2011 and 2014 tests. The Appendix list the participating banks and their results of the stress tests.
Daily data of 5-year CDS spreads are used and the iTraxx Europe index10 will be used as proxy for the market portfolio. Holidays and bank closing days are not taken into account.
In order to examine the differences between PIIGS and non PIIGS countries, the stress tested banks were divided into two portfolios, PIIGS and non PIIGS. The PIIGS countries are Portugal, Italy, Ireland, Greece and Spain. In addition, the banks from Malta and Cyprus11 are also included in this group, making both samples more equally. The sample for 2010 consist of 22 PIIGS and 27 non PIIGS banks, the 2011 and 2014 sample consists of 13 PIIGS and 14 non-PIIGS banks.
In the Appendix the descriptive statistics of the six different estimation windows are given. The descriptive results indicates12 that the distribution of realized spread changes is not normal. The descriptive statistic of abnormal CDS spread changes shows deviations from normality, due to sometimes negative or extremely high skewness, and negative and low kurtosis.
Methodology
To measure the impact of the announcement and disclosure of a stress test an event study will be used. CDS are relatively new assets in the credit market, therefore the literature lacks a clear understanding of CDS event studies (Andres et al., 2013). This study will apply a similar approach to that of Brown and Warner (1985) MacKinlay (1997) and Andres et al. (2013) to investigate CDS data of the stress tests for potential abnormal returns.
According to MacKinlay (1997): ‘given rationality in the marketplace, the effects of an event will be reflected immediately in security prices. Thus a measure of the event's economic impact can be constructed using security prices observed over a relatively short time period.’ Short term event studies are more reliable than long term event studies: They have less limitations (Kothari and Warner, 2005).
10
The iTraxx Europe index is composed of one hundred twenty five (125) liquid European entities with investment grade credit ratings that trade in the CDS market. The iTraxx is an equally weighted index, this choice has been made while an equally weighted index is more likely to identify abnormal returns. Retrieved from
http://www1.uni-hamburg.de/Kapitalmaerkte/download/QuantmethodsWiSe200405Henderson.pdf Accesses on 6 May 2015 11
This decision has been made due their geographical location and the economic situation in both countries. 12
14 In the literature there is no agreement about the length of the estimation window. Neretina et al. (2014) use an estimation window of 150 days prior to the event. Campbell et al. (2010) use 120 days prior to the event to estimate the normal returns. Furthermore the study of Peristiani et al. (2010) use daily data for one year to estimate the normal returns.
This study estimation window consists 120 days13, (-130, -10), where t = 0 represent the trading day when the results were revealed. This timeframe is sufficiently long to conduct an event study based on daily data, and short enough to avoid overlap with other stress tests. To avoid influencing outcomes, the release day of the stress test results will not be included in the estimation period (MacKinlay, 1997).
For controlling purposes, a three day event window (-1,+ 1) will be used for measuring CDS spread change. The day after the event is included while sometimes the results will revealed after closing the market, and reply delays. According to König (2011)14, some details of the stress test were already leaked beforehand which may have had an impact on the market. However, leakages cannot be avoided, the +1 will captured lagged market response. The 3 day window must be sufficient enough for all events without interfering with individual dates, above on that it includes both the risk of news leakages before the announcement as it includes slow response of investors. If a stress test result or announcement gives information to investors and traders, and is able to restore market confidence, markets will react to unfavorable news by increasing spreads and to favorable ones by decreasing spreads (Pedrescu, 2009).
To compute CDS spread changes, a logarithm of daily spread changes will be calculated15
𝑅𝑖𝑡 = ln (𝑃𝑃𝑖𝑡𝑖,𝑡−1) (1)
Where 𝑅𝑖𝑡 is the log CDS spread is change for a particular bank i at day t, 𝑃𝑖,𝑡 is todays spread and 𝑃𝑖,𝑡−1is last day’s closing spread.
Secondly, the normal return has to be calculated, which is the return that would be expected if the event did not take place. To estimate normal returns a market model would be used. This model assumes a simultaneous relationship between realized security returns and realized market returns (Brown and Warner, 1980). The model connected the spread of a bank to the spread of the market (MacKinlay, 1997). This statistical model assumed a linear relationship between the security performance and it is portfolio performance according to its beta.
13
The estimation window will include 120 days, not 120 trading days due the fact that otherwise there will be overlap between the stress test events. Furthermore due the turbulent times the stress tests events takes place, a shorter event window will be sufficient enough.
14
Retrieved from http://www.dw.de/ecb-rumored-to-fail-25-banks-in-stress-test/a-18020403. Accesses on 8 January 2015. 15
These studies used continuously compounded returns used by log. This improves the normality of the return distribution, it eliminates negative values and it is easy by compounding. Retrieved from
15 For any given security I the market model is:
∆𝐶𝐷𝑆𝑖𝑡 = 𝛼𝑖+ 𝛽𝑖 ∆𝑅𝑚𝑡+ 𝜀𝑖𝑡 (2)
∆𝐶𝐷𝑆𝑖𝑡= 𝐶𝐷𝑆𝑖𝑡− 𝐶𝐷𝑆𝑖𝑡−1 ∆𝑅𝑚𝑡= 𝑅𝑚𝑡− 𝑅𝑚𝑡−1
Where ∆𝐶𝐷𝑆𝑖𝑡 is the daily change in the CDS spread for bank I at day t. ∆𝑅𝑚 represent the daily change in spread of the iTraxx Europe index. The main motivation to select the market model in stand of the mean-adjusted model is the fact that MacKinlay (1997) argues that if the R-Squared of the market model increase, the power of the test to detect event effects increases. According to Andres et al. (2013), this argument suggest that: ‘The market model dominates the mean adjusted model in terms of size, power and better explanatory power for spread changes.’
To evaluate the announcement and results effects, abnormal returns will be measured. The abnormal return is the securities actual return over the event window minus the normal return of the security over the event window (MacKinlay, 1997). The normal returns is the expected return without taking the event into account. Abnormal CDS spread (AR) changes for i at event window t will be:
𝐴𝑅𝑖,𝑡= 𝑅𝑖,𝑡− 𝛼̂𝑖− 𝛽̂ 𝑅𝑖 𝑚,𝑡, (3)
Through this formula the abnormal return is determined by correcting the normal return by two coefficients, namely α and β. These estimates has been calculated using the following formulas.
𝛽̂ i = 𝑐𝑜𝑣(𝑅𝑖 ,𝑅𝑚)𝑣𝑎𝑟(𝑅𝑚) (4)
𝛼̂ i = Ṝ i -( Ṝm*𝛽̂ i) (5)
The estimator β (beta) determines the sensitivity of the return relative to the return on the market. In this manner we can examine whether the abnormal return of a particular bank differs from the return on the market, which will is Rm in the formula. The estimator α (alpha) relates to the risk that is taken with respect to the investments in CDS. Furthermore, the Market Adjusted Returns Model has been used:
𝐴𝑅𝑖𝑡 = 𝑅𝑖,𝑡− 𝑅𝑚𝑡 (6)
Where 𝑅𝑚𝑡 is the market index return for day t, and 𝑅𝑖,𝑡 the return of bank i at time t
16 be determined. The unweighted cross-sectional average abnormal returns (AAR) are determined using the following formula.
𝐴𝐴𝑅𝑡 = 𝑛1∑𝑛 𝐴𝑅𝑖,𝑡.
𝑖=1 (7)
Where n is the number of events over which the average is calculated, i represents the particular bank and t represents the day in the event window and AR the abnormal returns.
Furthermore, the time-series cumulative average abnormal returns (CAAR) has been calculated. Calculating the CAAR is valuable for determining the effect of the stress test in the entire event window.
The CAAR’s will be calculated using the formula of MacKinlay (1997). By adding up all the AARs in the event window. The cumulative abnormal spreads changes for over the event window is
𝐶𝐴𝐴𝑅 (𝑡1, 𝑡 − 1) = ∑𝑡−1𝑡=𝑡1𝐴𝐴𝑅𝑖𝑡 (8)
The variance of the CAAR has been calculated by the following formula
𝜎 = 1
𝑁2 ∗ ∑ 𝜎𝑖2 𝑁 𝑖=1
(𝑇0, 𝑇1)(9)
Where t0-t1 represents the variation of the CAAR in the estimation window using the market model. Furthermore the t value of the CAAR has been calculated by
𝑇 = 𝐶𝐴𝐴𝑅
σ (𝐶𝐴𝐴𝑅)0.5 (10)
Where CAAR is the sum of all AAR’s.
17 Afterwards, a parametric t-test and a non-parametric generalized sign test will be used.16 The t-test follows a student-t distribution with n-1 degrees of freedom.
t= 𝐴𝐴𝑅 [𝑇1] – 𝐴𝐴𝑅 [𝑇0−𝑇1] σ [T0−T1]/ √n (11)
Where n is the number of observations and σ is the standard deviation of the estimation window.17 𝐴𝐴𝑅 [𝑇1] represents the particular event day, 𝐴𝐴𝑅 [𝑇0 − 𝑇1] represents the average abnormal returns of the estimation window.
The generalized sign test will be used in stand of the rank test while it provides more powerful results if the event window is longer than one or two days (Cowan, 1992). Above on that Andres et al. (2013) argue that the non-parametric generalized sign test is the best technique to identify abnormal CDS spread changes due the non-normality the data follows. Above on that Campbell et al. (2010) argue that the sign test excellently performs in a multi country setting within a three day event window. The sign test can be defined as18:
t= [ 𝑁+ 𝑁 − 0.5 ]√N0.5 (12)
Where N+ represents the total amount of positive abnormal returns for the particular event day. Moreover, the expected returns has been calculated using the confidence level of all banks, to investigate if the returns are in line with the variance calculated in the estimation window. While the abnormal returns are returns that fall outside the confidence interval, these returns are calculated from the forecast rate of returns of the estimation window prior the event. The confidence level of the abnormal returns has been calculated using the following formula.
𝐶𝐼 = 𝑅𝑖,𝑡+𝑒 ± 𝑡𝑐𝑣∗ 𝑠𝑒 (𝑅𝑖,𝑡+𝑒)(13)
Where 𝑡𝑐𝑣 is the critical value, 𝑠𝑒 (𝑅𝑖,𝑡+𝑒) is the standard error of the regression 𝑖 refers to a banks’
security and 𝑡 + 𝑒 refers to the day of the event. The confidence level of 95 percent has been used in this study.
Moreover, it would be interesting to investigate the test reaction on individual bank level. Therefore the variance of all single banks are measured by their confidence level of 95 percent. To investigate if the returns in the event window can be classified as abnormal returns for each particular bank.
16
Parametric test are not used in isolation but together with non-parametric tests. A non-parametric test can be used as a robustness of conclusion based on parametric tests. Campbell, J.Y., Lo, A.W., MacKinlay, A.C., 2012. The Econometrics of Financial Markets.
17
of het market model 18
18
Results and Discussion
The findings are presented in table 1.19 The AAR, the p value of the t-test and p value of the sign test for the two events of three stress tests are presented. The length of the event window is three days, where 0 represents the event day. 20
Table 1 Stress test results for Announcements
Notes: *** - 1% ** - 5%, * - 10% significance level. This table presents the AAR and the p-statistics of the t test and the sign test for the announcements and result events of the stress tests in 2010, 2011 and 2014.
Stress test announcements
As table 1 shows, the results of the stress test announcements based on student t-test are mixed. In 2010, significant declining results on CDS spreads have been found on 1 percent level based on the AAR of all three event days. The CDS spreads of participated banks declines more than normally which results in less uncertainty. Moreover, in 2011, no event day is significant therefore the null hypothesis cannot be rejected in this case. Following Neretina et al. (2014), if there is no significant impact on CDS spreads, the information is not informative for the market. This means that the announcement was less informative to the market compared to 2010. Furthermore, in 2014, all three event days were again significantly declining on a 1 percent level. Again, the CDS spreads of participating banks declines at the announcements of the stress tests. This suggest that the market found this new method of stress testing as informative or the change of default is lower. However it could also suggest that due the fact that stress tests have not been performed in 2012 and 2013, the market appreciate the information provided by the announcement quite strongly.
Nevertheless, the student t-test assumes normality, which is not the case with the CDS spreads data. Therefore the non-parametric sign test has been used these results can also be found in table 1. The p-value of the sign test differs from the t-test in 2011 now the first event day is significant, the CDS spreads declines more than normally, but only on a 10 percent level. This may indicate that there were some rumors about the agreement of the European Commission of a new stress test, which affected the
19 As a robustness check Banks with zero CDS spread in the market model has been taken out, this does not affect the overall results. Therefore and while they do have changes in market adjusted returns, these banks are still included.
19 market certainty.21 Furthermore, only the last event day in 2014 is significantly declining compared to three significant declining days with the student t-test. While the CDS spreads declines more, this may indicate that the market certainty increased due the new stress tests method. Concluding, at all announcements, the hypothesis H1a: The CDS spreads of participated banks widens more which
results in more uncertainty to the announcement of a new stress test has not been met.
Table 2 CAAR's Stress test announcement & Results
Notes: *** - 1% ** - 5%, * - 10% significance level. This table presents the CAAR and the p value of the t test for the announcements and result events of the stress tests in 2010, 2011 and 2014.
Table 2 shows the CAAR results of the stress test announcements and results. As not all AAR’s are significant, all CAAR’s are significant for all event windows. This might suggest that the days taken separately do not appreciate the new information as valuable, however, taken all three days together, the market appreciates the new information as valuable. As can be seen in table 2, widening results have only be found by the results in 2011, meaning that in 2011 the markets were not reassured after the disclosure of the stress test results which is in line with the founding of Byström (2011). At all other events, the market certainty recovers due the European Stress Tests. The CAAR has been used to investigate if the effect has been spread upon several days22, and it provides long term information
about the effect on CDS spreads (Mackinlay, 1997). Furthermore it gives a precise valuation of the CDS spread. Especially for a stress test, this spread is likely due new leakages, delays or speculations of traders and investors. For this reason, the CAAR is a beneficial statistical analysis in addition to the AAR since it provides an overview of the total effect of the abnormal returns.23
Lastly, if all stress test announcements together have been compared with an ANOVA test (see Appendix), there is evidence that the stress tests announcements differs between the three years. Meaning that in all three years a stress test announcement has been made, the market reacts significantly different.
21
Retrieved from https://www.eba.europa.eu/-/the-eba-announced-a-new-round-of-stress-tests. Assesses 2-4-2015. On 13th January 2011 the EBA announced the stress test, however the agreement was on 12th January 2011. This may influence the market because of news leakages.
22
As a Robustness check the event window was increased by +3,-3. However this does not change the results. 23
20
Table 3 Stress test results for Announcements
Notes: *** - 1% ** - 5%, * - 10% significance level. This table presents the AAR and the p-statistics of the t test and the sign test for the announcements and result events of the stress tests in 2010, 2011 and 2014.
Stress test results
In all three years, the student t-test statistics of the disclosure of the stress test results, has not found any significant effect. The null hypothesis cannot be rejected, the CDS spreads of participated banks did not change significantly after the disclosure of the results. The main motivation for this results is the information was not new for the market, it is already incorporated in the CDS spreads.
The p-value results of the sign test are different compared to the student t-test. In 2010, all event days are significant. Indicating that the CDS spreads declines significantly, which results in less uncertainty in the market, while the likelihood of default is lower. Moreover, in 2011 the disclosure day and the day after are significant, the CDS spreads widens. The H0 can be rejected, The CDS spreads of participated banks widens more, which results in more uncertainty to the results of the stress test. This could be explained by the uncertainty in the market, concerns about a double dip and bailout packages to Greece and Portugal, which has also big impact on the stock market.24 Furthermore, in 2014, all event days are significant. The disclosure of the results in 2014 had a significant impact on the CDS spreads, they declined more, which results in less uncertainty in the market, while the change of default is lower. Notice that the disclosure of the results of the stress tests was entirely published after the markets in Europe and Asia had closed. In 2010 and 2011 the disclosure was on Friday, in 2014 the disclosure was on Sunday. Therefore investors could not directly react to the stress test results but just the day after, so they can incorporate more information. Again, only in 2011 the null hypothesis can be rejected. Bystörms (2011) main explanation for this effect in 2011 is that the market did not expect the high capital shortfalls which were announced. Furthermore, both at the announcement and results in 2014, the banks reacts the same as in 2010. Therefore, no conclusion can be made that the new version of the stress test has another influence on the market.
Finally, if all stress test results together have been compared with an ANOVA test (see Appendix), there is evidence that the market reaction on the stress test results does not change over
24
21 time. If the disclosure of the stress test has been made, the market reacts not significantly different in all three years which confirms the conclusion that the new version does not influence the investors.
PIIGS and non PIIGS
To examine whether there are any differences between PIIGS and non PIIGS countries, an ANOVA25 test has been used, which compares both groups based on their average.26 The ANOVA test results (see Appendix) shows that only the day after the disclosure of the stress test results in 2011 the null hypothesis can be rejected. At this day, the PIIGS and non PIIGS react significantly different. If the countries are taken separately, the PIIGS spreads declines much more the day after the announcement comparing to the non PIIGS. Which is remarkable while seven of the eight failed banks are located in a PIIGS country with high capital shortfall. 27
Table 4 Differences between PIIGS and non PIIGS at announcement & results
Notes: *** - 1% ** - 5%, * - 10% significance level. This table presents the CAAR and the p value of the t test for the announcements and result events of the stress tests in 2010, 2011 and 2014. Distinction has been made between PIIGS and non PIIGS countries in the event window (-1,+1) to evaluate the market reaction of both groups.
To further investigate differences between PIIGS and non PIIGS, the CAAR’s of both groups has been calculated. As table 3 shows, the 2010 announcement has significantly declined the CAAR CDS spreads of both countries. The effect of the non PIIGS is bigger than the PIIGS reaction. This could be explained by the fact that most of the failed banks are located in a PIIGS country. However, it is surprising that even when there are eight failed banks in a PIIGS country, the change of default of the whole group is seen as lower through the investors. In 2011, both groups reacted strongly to the announcement by declining the CDS spreads, the PIIGS even more than non PIIGS, investors value the change of default in a PIIGS country even lower. In 2014, no significant results have been found. Furthermore, the disclosure of the results in 2010 had a significant declining impact on the PIIGS countries, on 1 percent level, and a significant declining effect on the non PIIGS countries on 5 percent level. This translates into declining CDS spreads, more certainty and the change of default is lower. In 2011, among both groups, the CDS spreads widens slightly to the disclosure of the stress
25
The non-parametric kruskal wallis test has been used as a robustness check. However the results do not differ between these two tests.
26
The AAR of the CDS spreads has been used. 27
22 test, but only on a 10 percent level. This may indicate that the stress test increased the uncertainty in the market in relation to European banks, which is in line with the overall sample result. Finally, in 2014 the results did have a significant declining impact28 on the non PIIGS but not on the PIIGS,
whose spreads widening after the results. The main motivation for this effect is that most of PIIGS banks failed in 2014, namely 19 out of 25, therefore it is plausible that their investors reacted by widening the CDS spreads due the change of default has become higher.
Overall the non PIIGS react slightly stronger to a stress test announcement or result compared to PIIGS banks. An explanation for this effect might be the safety net certainty, which could be a motivation for investors to invest in risky banks to take more risk with the certainty that they will be saved by a default. Therefore, it is plausible that investors of risky banks do not react to a stress test.
Stress test impact on individual bank level.
To investigate stress test reaction on individual bank level, the abnormal returns are measured by their confidence level (99 percent). Furthermore, it would be interesting to investigate if the market already reacts differently to the banks that failed the stress test. In other words, if the investors and traders already know the (weak) financial position of their bank and therefore the CDS spreads widens at the announcement of a new stress test round.
The results (see Appendix) shows that almost all banks reacts significantly to the stress test announcement in 2010, their CDS spreads declines. Remarkably, more non PIIGS banks reacts to the announcement than PIIGS banks. The non PIIGS banks that do not react are mainly located in the UK, Norway or Sweden, which might be explainable by the fact that these countries are not all members of the Monetary Union. Interestingly is the fact that banks as Bank of Greece, Bank of Ireland and some Spanish banks do not react significantly different to the stress test, which might be expected due the financial problems in these countries. Furthermore, the reaction to the disclosure of the results differs, more PIIGS banks react significantly declining to the disclosure where non PIIGS reacts overall increasingly positive, their CDS spreads widens. This is surprising due the fact that all failed banks are located in Southern Europe, namely in Spain (5) and Greece (2). However investors viewed the change of default in the PIIGS countries as lower than in non PIIGS. In 2011 there are some significant reactions to the stress test announcement on the day before and the announcement day. However, the day after, no bank reacted significantly different. The same goes for the results in 2011, at the announcement day almost all banks had significant widening spreads. The day after the disclosure nearly all banks reacted with declining spreads but insignificant. In 2014, almost no bank reacted to the new version of the stress test announcement. The banks that reacted significant with widening spreads are mainly located in a non PIIGS countries. All this indicates uncertainty in the new stress test version, investors might be afraid that healthy banks in North Europe do not pass the AQR
28
23 which is part of the new stress test. The results of the disclosure show that the day after the disclosure the widening significant reacting banks are located in a non PIIGS country. The main explanation could be the fact that there has never been this amount of failed bank through the stress test, which might create uncertainty among traders and investors at Northern banks. If Southern banks cannot fulfil their capital shortcut, they are able to lend from the ECB. These loans are provided by central banks (mainly non PIIGS) of each particular country who have surpluses administered by the ECB. This might be an explanation of the reaction of the non PIIGS bank traders; it creates uncertainty among the Southern banks that can fulfill their obligations, which might influence the Northern banks as well.
To investigate if the market already know beforehand if a particular bank fails, the reaction of the announcement and the disclosure of the results of a particular bank has been compared. The 2010 stress test has been excluded from this analysis because of data limitations; none of the failed banks have been included in the sample. Nonetheless, Irish banks needed to be rescued soon after the stress test. At the announcement, only the Allied Irish bank reacted significant on the day before and the announcement day; however their spreads declines, meaning less change of default. At the results, both the Bank of Ireland and Allied Irish bank react with significant declining spreads. This indicates that the market does not know beforehand the troubles the Irish banks were facing. Furthermore, at the announcement of 2011 and 2014, the traders of the Bank of Ireland did not react significantly different. Indicating that the traders did not become nervous for new rounds of stress testing. In 2011 again none of the failed banks were included in the sample; however Dexia failed only a few months after passing the test. At the announcement, the AAR of Dexia are the day before the announcement and the day after significant different by declining their spreads. Noteworthy, the day after the announcement the AAR of Dexia is widening but insignificant, indicating that the Dexia CDS traders do slightly become nervous for the next stress test. In 2014, the sample included banks such as Monte dei Paschi di Siena, National Bank of Greece and Dexia, which all failed the stress test. At the announcement again only the AAR of Dexia is significant at the day before and the announcement day. Remarkable, these AAR are widens significantly, indicating uncertainty in the new stress test round which could be explained due the problems the bank had in the years prior to the stress test. Overall these results shows that the market does not identify beforehand which banks will fail the stress test, the only expectation is the case of Dexia. This results is in line with the findings of Petrella and Resti (2012) who found that the market is not able to anticipate the stress test results.
To investigate what may cause the significant CAAR results, various determinants are regressed against the CAAR using a simple OLS regression.29 The determinants have been selected based on
literature. Differences between PIIGS and non PIIGS were the major imbalance, high level of debt and
29
24 lack of trust. The Southern banks were facing problems by attracting money, and they had several outstanding impaired loans. Furthermore, banks that failed the stress test were banks who were facing capital shortfalls. The aim of the stress tests was to, among others, see which bank would be able to survive a further decline in economic terms. For failed banks adequate measures could be taken to avoid worse results (EBA (b), 2014). To identify the possibility of bank failure, the CAMELS indicators will be used. As already described in the literature, the visual inspection of bank sheet indicators shows differences between both groups. Hirtle and Lopez (1999) observed how effective the CAMELS rating is with predicting bank conditions. They argue that the CAMELS ratings provide further information about banking conditions, and it is a useful indicator of controlling bank conditions therefore this rating system will be applied to analyze the CAAR’s.
CAMELS is an acronym for Capital Adequacy, Asset Quality, Management Quality, Earnings Potential, Liquidity, and Sensitivity to Market Risk (Ötker-Robe et al., 2010). For Capital Adequacy, the equity-to-asset ratio, equity to customer short term funding and Tier 1 capital ratio will be used. The higher the level of capital, the higher the buffer against a potential failure (Betz et al., 2013). Asset quality will be measured by Impaired Loans to Equity. The higher the impaired loans as a share of equity, the weaker the assets the more likely a default. The management quality will be measured by the cost-to-income ratio. Earnings potential will be measured by the return on average equity and liquidity will be measured by the share of liquid assets to deposits and short term funding, the share of liquid assets to total deposits and borrowings, and the share of net loans to total assets. The higher the ratio of net short-term borrowings, the more likely a bank default (Betz et al., 2013).
The white test has been used to inspect for heteroskedasticity; in this case there is no heteroskedasticity detected. Furthermore collinearity has been taken into account. In order to detect collinearity, the variation inflation factor (VIF) has been used. Following Chatterjee and Hadi (2012), there is evidence of collinearity if the largest value of VIF is greater than ten. As the results show (see Appendix), Liquid Assets Deposits Short Term Funding (72,51), Liquid Assets Total Deposits Borrowings (38,34), Return On Average Equity (ROAE) (16,25), Net Loans Total Deposits Borrowings (16,15) and Net Loans Total Assets 2014 (14,66) are all above the threshold. Liquid assets / total deposits and borrowings has the highest value, therefore it has been deleted. Furthermore, a new round of VIF has been calculated. At this stage ROAE (12,53) is still above the threshold so this one has been deleted. Furthermore, Equity Total Assets and Net Loans Total deposits borrowings are quite high. Because there are more variables of net loans, this one has also been deleted. With these actions collinearity has been removed.
25 total assets ratio shows the amount of assets invested in a loan portfolio. This amount is not allowed to be too high, which might create liquidity problems. While this ratio has a negative coefficient, and in combination with a declining CAAR, it might be the case that in 2014 this ratio was sufficient according to investors.30 Finally, the net loans total deposits and short term funding is significant. This
ratio indicates the reserve to deposit of a bank, the higher the ratio the less liquid the bank.31 The
negative coefficient might indicate that investors viewed banks liquid by declining CAAR. The tier 1 ratio, the return on average equity and the share of liquid assets to deposits and short term funding are significant on a 10 percent level.
At the results the share of equity / customer short term funding is significant on a 1 percent level. This ratio explains the amount of permanent funding to short term, which is volatile funding.32 The ratio correlates negative with the CAAR, which might indicate that banks held less short term funding compared to their equity which makes the bank less vulnerable. The share of impaired loans to equity and the share of equity to assets are significant but only on a 10 percent level. These results are in line with the overall view of the investors; the change of default has been declined. The conclusion can be made that only at the results and announcement the CAAR’s can be explained by liquidity ratios. The remark must be made that the data of the determinants is based on yearly data provided by Bankscope, where the CAAR data is on a daily basis therefore the determinants have to be interpreted as indication. As a robustness check, one regression has been created for the announcement and one for the results, in order to investigate if the CAARs can be explained by banking condition factors. The main reason for this check is because of the small sample of this paper.33 As can be seen in the Appendix, the announcement effect of all years can only be explained
by net loans deposits short term funding, at 5 percent significance level. At all stress test years, the banking sector has been seen as less liquid. The results can slightly be explained by the Tier 1 ratio, however this is just significant on a 10 percent level. For all together years, the results cannot be fully explained by banking conditions factors.
To detect if the CAMELS ratings explain the different outcomes between the PIIGS and non PIIGS, dummy for the PIIGS has been included in the regression. Only slightly significant results on a 10 percent level has been found among the results of 2014, therefore it can be concluded that the PIIGS reacts different than the overall sample. The economic indicators, the cost-to-income ratio and the equity / customer short term funding might explain a little bit the CAAR of banks in a PIIGS country. The cost-to-income ratio provides investors a view of how efficient the bank is. The
30
Reverse reaction CAAR and the ratio, higher ratio which is more stable, means declining CAARs. 31
Retrieved from
https://books.google.nl/books?id=pbg-ULiBuosC&pg=PA157&lpg=PA157&dq=ratio+equity+to+customer+short+term+funding&source=bl&ots=DbGKcTIZOL& sig=g4PCuAsTByI1MOUtT_frqnbJq50&hl=nl&sa=X&ei=GI5IVd29OcrLaLfZgcgF&ved=0CCIQ6AEwAA#v=onepage&q =ratio%20equity%20to%20customer%20short%20term%20funding&f=false accesses on 5 May 2015.
32 See 31 33
