Credit default swap spreads as proxy of banking performance

› Studentnr: s1883224

› Name: Marjolein de Vries

› Study Program: Msc Finance

› Supervisor: dr. M. Hernandez-Tinoco

› Date: 12-01-2015

2

In early 2007 the financial system that seemed to be operating so well began to show some

small cracks. Many companies that dealt in subprime-related activities announced that their

earnings were expected to drop and along these announcements many hedge funds were

liquidated after they had experienced severe losses. These events were consequences of the

trend that had developed in the banking industry. Many banks switched from the traditional

banking model in which they hold loans on the balance sheet to the “originate and distribute”

model. The latter model led to a large reduction in the standards related to lending according

to Brunnermeier (2009).

In the “originate and distribute” banking model the banks had the opportunity to create new

types of products, namely the so-called “structured” products. First, diversified portfolios are

created from different mortgages, corporate bonds and other assets. Next, the portfolios are

divided in different tranches and are then sold to investors according to the preferences for

risk of these investors, that is, the riskier tranches are sold to risk-seekin investors. The

investors that have bought a tranche can insure themselves for the risk that they bear. In order

to insure themselves against the risk of these tranches investors can purchase a credit default

swap. The first credit default swap was negotiated in the mid-1990’s. After that first contract

the credit default swap market has grown exponential in size. Lots of firms arose that traded

in credit default swaps. According to Ericsson, Jacobs and Oviedo (2009) the amount of

outstanding principal was more than 20 trillion dollars in 2006 and the credit default swaps

accounted for a third of the trading activity.

Credit default swaps can be seen as an insurance contract between two parties. The seller,

often a hedge fund, of the credit default swap provides the buyer protection against a default

on the tranche for an upfront payment and a periodic fee. This periodic fee can also be

considered as the spread. Most credit default swaps have a life of five years and they are only

exposed to the collateral for which they are created. In this study, a default in the credit

default swap means bankruptcy or a restructuring of the entity that is supposed to pay the

obligation, or a simple failure to pay. When the default occurs, the seller has the obligation to

pay the insurance to the buyer. The credit default swap transfers risk to the party that is most

willing to bear it. So the initial purpose of the credit default swap is to serve as hedging

device.

3

Brunnermeier (2009) argues that it was believed that any investor that bought both a AAA-

rated tranche of an asset-backed debt obligation and a credit default swap bore low risk

because the chance of a default by the CDS counterparty was low.

The credit default swap is issued with risk meaning that they are exposed to counterparty risk.

It may be possible that the party that provides the insurance does not have the means to pay

the insurance to the buyer even though it has received the upfront payment and the fees. Risk

speculators who did not own the underlying asset saw an opportunity and wanted exposure to

particular assets, bonds and loans. They began speculating on certain firms of which they

thought that were not able to pay back the bond holders. They also began speculating on

overrated subprime mortgage pools that banks, insurance companies and hedge funds

possessed. Speculators eventually traded in trillions of dollars of insurance and speculated on

the idea that the subprime mortgage pools would not default. So these instruments are the

reason of massive write-downs at insurance companies, investment banks and banks.

In this paper the focus will be on the credit default swaps. Credit default swaps are not

transparent, are not regulated and are not traded on any exchange so therefore credit default

swaps are not standardized instruments. The initial purpose of this derivative was to serve as a

hedging device in the way that market participants are allowed to trade the risk that comes

along with debt-related activities. However, credit default swaps are often used for other

purposes. For example, the fee of the derivative is commonly used to measure the default

component in the corporate spread and many times credit default swaps are used for

speculating instead of hedging of risk. Because of this alternative use of derivatives, they are

very often named as weapons of destruction. Also, credit default swaps are regularly linked to

the origin of the financial crisis of 2007. Thus, the aim of this paper is to examine the role of

the derivative credit default swaps in the financial crisis by measuring to what extent the

credit default swaps have an effect on the performance of certain banks that trade in these

products. This reasoning leads to the following research question.

“Does a relationship exist between the credit default spread and banking performance for

European international banking groups?”

4

One of the main findings of this study is that, instead of the crisis dummy, the pre-crisis

dummy and the post-crisis dummy have a significant impact on the level of credit default

swap spreads. This unexpected finding is mainly due to the fact that the crisis period consisted

of a small sample of banks. For the whole period all the categories, asset quality, capital,

operations and liquidity are considered important estimators of the performance of banks.

Investors associate changes in these variables with the power of a bank to absorb losses and

will act accordingly. In the whole period in which the dummy variables are one by one

included asset quality is a significant estimator of the performance of the banks. Investors

worry only in the crisis and in the post-crisis period about the quality of the portfolios a bank

holds. Capital is in the pre-crisis period and in the crisis period of importance when investors

decide about buying insurance against the probability of default. The category operations

show significant results in all three different periods. Investors view operations as a reliable

indicator for estimating the strength of the bank and whether credit default swaps should be

bought. The market does also view liquidity as an important estimator in the crisis period.

The setup of this paper is as follows. In the following section, the literature review, the

relevant literature for this paper is reviewed. After the literature review the sections data and

methodology describe how the model used in this paper is set up. Further, the results of the

tests performed will be analyzed followed by a conclusion based on the analysis.

5

I. Literature review.

Structured products had completely taken over the traditional manner of funding in the

banking industry. This is mainly due to the benefit that the product provides for the banks that

create them. A large part of the risk that comes along with the product will be passed on to the

financial institution that buys the product. The designer of the structured product is thus

exposed for a short amount of time to “pipeline risk” as mentioned by Brunnermeier (2009).

This led to falling lending standards and due to lower lending standards there existed a

substantial amount of cheap credit in the market. In July 2007 many banks that traded in these

structured products and had a large exposure to this type of financial instruments on their

balance sheet were uncertain about how to value these products. In a reaction to this

uncertainty people began to lose trust in the reliability of the credit ratings. Credit ratings are

supposed to give a correct reflection of the probability of default.

The price of a credit default swap, often known as the spread, reflects the amount of the fee

paid by the buyer. The spread reacts faster to changes in the probability of default than credit

ratings do. The spread of such a swap also gives an indication about how the market perceives

the chance of a default related to the underlying engagement. A change in the perception of

default risk by the market leads to a change in the spread. For example, a perceived increase

in the default risk by the market leads to a higher demand for the credit default swaps and thus

leads to an increase in the spread. It is also confirmed by Acharya & Johnson (2007) that

credit default swap markets reflect valuable information. They find that the credit default

swap market incorporates non-public information in publicly traded types of securities such as

stocks. Flannery, Houstond & Partnoy (2010) confirm this by arguing that the overall market

participants believe that the credit default swap spreads include information about the buyers’

perception of the risk of a default. In their article they find that the changes in the credit

default swap spreads immediately incorporate information in contrast to changes in credit

ratings. The results of their study shows that this is the case even in times of distress.

The impact of earnings, accruals and cash flows created by a company has an impact on the

credit risk that is included in the credit default swap. An increase in the profitability of the

entity leads to a reduction in credit risk because with an increase in profitability the entity has

more wealth and thus has a lower chance to default on an obligation.

6

Callen, Livnat & Segal (2009) state that most investors mainly use earnings to evaluate the

performance and the future wealth of the firm they have an interest in because earnings are a

major determinant of credit risk. Furthermore, earnings can be used to estimate the true asset

dynamics of the reference entity, and so its credit risk. Callen, Livnat & Segal (2009) confirm

this with the findings in their paper. They find that earnings in the structure of cash flows and

accruals that are created by the reference firms are negatively and significantly correlated with

the level of the credit default swap spread. This finding is consistent with the insight that

earnings transmit information on default risk. For example, their results show that credit

default swap rates are significantly reduced, by 5 percent, due to a 1 percent increase in the

ROA. Due to these findings the variable earnings will also be an important factor in this

paper.

As mentioned earlier and also argued by Duffee and Zhou (2001), credit derivatives are used

as a tool by banks to get round the “lemons” because they can easily be used to transfer risks

and thus to manage credit risk exposure. Many banks use swaps for a short amount of time to

transfer the risk of their loans to other parties. By doing so, these banks reduce their risk

exposure and thus the probability that they will end up in financial distress when defaults on

loans occur. Many banks that used these risk-reducing tools, like a credit default swap, were

allowed to hold less capital under the Basel II Accord. This accord gives banks the

opportunity to keep on supporting their risky assets and still hold the same level of regulatory

capital ratios when they use credit default swaps for capital-intensive activities or for

increasing the level of their asset bases. Shan, Tang and Yan (2014) find in their results that

banks that make use of credit default swaps do not have a different capital ratio than banks

that do not make use of this instrument.

Banks that are not that well capitalized are more likely to improve their capital ratios by using

credit default swaps. Banks with lower capital ratios are more likely to use credit default

swaps for other activities instead of credit risk management. Shan, Tang and Yan (2014)

confirm this by measuring the Tier 1 ratio as proxy of bank capital quality. Tier 1 capital for

banks that are active in the credit default swap market is lower than for banks that are not

active in this market. A lower capital ratio may indicate that the bank is involved with more

aggressive risk taking, but that does not necessarily constitute a problem. A lower capital ratio

may also be the result of more efficient banking, meaning that the bank may have an

advantage in access to bank credit.

7

In this situation the bank contains the same or a greater number of loans. When a bank is

involved in more risk taking by lending more aggressively it can use credit default swaps for

hedging a part of the credit risk exposure. So whether a bank is more efficient or risk taking

depends upon its lending activity.

Banks can use credit default swaps in order to increase lending, but a larger number of lenders

also leads to higher bankruptcy risk. Subrahmanyam, Tang and Wang (2014) have also found

in their results that credit risk is economically significant affected by credit default swap

trading. They found that for an average firm the likelihood of bankruptcy is doubled after

introducing credit default swap trading in this firm. So credit default swaps can also

contribute to an increase in the probability that a borrower will default, even though they are

designed to protect against defaults. Lending activity increases after the introduction of credit

default swap trading indicating that more aggressive lending activities are adopted. Thus the

fact that credit default swaps are recognized in bank capital regulations leads to banks

adopting a larger amount of risk. In particular, both lower capital ratios and an increase in

aggressive lending leads to riskier banking. From this review the following hypothesis is

derived.

H0: No relationship exists between credit default spread and banking performance

Many papers can be found related to the structural models that examine the credit default

swap spreads. This paper will follow the pattern of existing literature that is focused on

structural models related to credit default swap spreads. Among other papers, this paper will

analyze the factors that determine the credit default swap spreads and thus build on the

contribution of Annaert et al. (2013) and Chiaramonte and Casu (2013). The contribution of

this paper to the existing literature is the fact that it is focused on a longer time period. Also,

the period used in this paper is divided into three sub periods which gives the advantage that

the impact of the explanatory variables on the credit default swap spread in the three periods

can be compared with each other. The inclusion of the macroeconomic conditions in the form

of dummy variables provides the opportunity to give an insight in the effect of the different

macroeconomic conditions in the three different periods on the relation between balance sheet

explanatory variables and the credit default swap spreads. Also, the paper will focus on

developed markets, namely the American and European market which are highly correlated

with each other.

8

The correlation between the two markets means that a downturn in the American market will

very likely, due to financial linkages and the amount of trade between the American and the

European market, lead to a downturn in the European market. The level of correlation

between the two markets is of importance for the study because by combining the banks

positioned in these two markets in the sample will grow in strength. Further, by combining the

American and the European market the sample consists of banks that face the same

developments in the markets and this will lead to more credible outcomes.

9

II. Data and methodology

A. Data and sample

For this paper the mid-tier and top-tier international banking groups with 5-year senior credit

default swaps are used. The focus of the paper will be on the United States and the European

area. The decision to include American and European international banking groups in the

paper is mainly due to the fact that the American and the European market, and also the

countries within the European market, are linked with each other. Between the American and

the European bank there is a high degree of interbank linkages so contagion is more easily

spread from one market to another. As mentioned earlier, the high level of correlation

between the two markets will lead to a stronger sample. Another reason for including both

American and European banks is the fact that in these two markets issuance and trade in

credit default swaps is largest.

The time horizon of the paper will start at 1 January 2005, due to the fact that from this year

on the European banks were required to use the International Accounting Standards when

preparing the consolidated financial statements. The period ends at 30 June 2014. The reason

for 30 June 2014 as ending date of the period is the fact that there is no data available for

dates after this period. Furthermore, in the paper of Chiaramonte & Casu (2013) the results

show that in the period leading to 30 June 2014 the credit default spread values were

declining and also a sign of recovery is noticed in this period.

Moreover, the time period will be divided into three sub periods. The first sub period runs

from 1 January 2005 to 30 June 2007, and is the pre-crisis period. The second period is the

crisis period. The period starts at 1 July 2007 and ends at 31 March 2009. 1 July 2007 is

chosen as start point of the crisis because it is widely accepted that this is the date that the

crisis exploded. This is also proven by the development of the credit default swap spreads in

figure I, which is taken from the BIS Annual Report (2010). In the figure it can be noticed that

the spread of the credit default swaps were at a peak around 2007. The analysis of credit

default spreads also demonstrate that in this period their values were at a peak.

10

Figure I

Development credit default swap spreads in years This table indicates the development of the credit default swap spreads in years. On the x axis the date in years is

given and on the y axis the spread in basis points of the credit default swap is depicted. The three regions, United

States, Europe and Asia are outlined by three different colors.

Source: Bank for International Settlements., 2010, “80th Annual Report”, Basel, 28 June 2010, pages 1-206

31 March 2009 is chosen as end date of the crisis because from this date on the credit default

spreads shrank in values, but the level of the values of the credit default spread remained

higher than the credit default spread level in the pre-crisis period.

The third period is the post crisis period. This period starts at 1 April 2009 and ends at 30 June

2014. In this period the credit default spread values were declining and also a sign of recovery

is noticed. This development can also be noticed from the figure included above. The graph in

figure I shows that from 2009 on the spreads of all three markets decline and stay at lower

levels.

The dependent variable that will be used in this paper for the regression is the 5-year senior

credit default swap spreads of different banks. As mentioned earlier, credit default spreads are

a good indicator of whether the market expects if a certain firm has a high default risk and

will therefore be used as proxy for bank performance. The type of credit default swap spread

that will be used is the “CDS Premium Mid” and this variable is similar to the average of the

“CDS premium bid” and the “CDS premium offered”. The mid-rate spread is expressed in

basis points.

11

B. Explanatory variables

In table I the independent variables used in the paper are outlined along with the expectation

that the independent variable will have on the dependent variable, the credit default swap

spread.

Table I

Independent variables and predicted sign

This table outlines the main independent variables used in the paper. The independent variables are divided into

four subgroups, namely, asset quality, capital, operations and liquidity. The column ‘Description’ gives an

explanation of the variables used in the paper. The predicted sign indicates the expectations of the effect of the

independent variables on the dependent variable, the credit default swap spread.

Variable Description Predicted sign Asset quality

Qa1 Loan loss reserve/gross loans (%) +

Qa2 Unreserved impaired loans /equity (%) +

Capital

Pat1 Tier 1 Ratio (%) -

Pat2 Leverage: equity/total assets (%) -

Operations

Op1 ROA = net income/average total assets (%) -/+

Op2 ROE = net income/average equity (%) -

Liquidity

Liq1 Net loans/deposits and short-term funding (%) +/-

Liq2 Liquid assets/deposits and short term funding (%) -

12

The explanatory variables used in the paper are balance sheet ratios. Eight balance ratios will

be used for the paper. For asset quality two ratios are determined. The first ratio is the loan

loss/gross loans (Qa1) expressed in percentage. The ratio shows the relation between the

value of total credits fitted and the depreciation fund. It can be expected that as the ratio

increases, the quality if the loan portfolio decreases. The second ratio for the asset quality is

the unreserved impaired loans divided by equity (Qa2). The ratio can be regarded as the

capital impairment ratio and the higher the ratio, the higher the probability that a firm will

default.

The second category is capital and both the ratios TIER 1 (Pat1) and leverage (Pat2) can be

assigned to this category. The first ratio, TIER 1, measures the capital adequacy of a

particular bank to determine whether the bank can absorb losses. An increase in the ratio

means that the bank can absorb losses and the credit default spread decreases. The second

ratio, leverage, is calculated by dividing equity by total assets. For this ratio, the leverage of

the balance sheet is used because this one is the most transparent. When banks acquire more

assets by borrowing funds this will lead to an increase in the balance sheet leverage. An

increase in the balance sheet leverage will immediately signal investors the change in the

probability of default of a bank. Leverage has a negative relationship with the risk of a default

because when equity decreases, the amount of debt increases. And with a constant amount of

total assets, the risk of a default increases. Investors will respond to this development and buy

credit default swaps which will lead to an increase in the spread.

For the third category, operations, the ratios ROA (return on average assets) and ROE (return

on average equity) are used. ROA (Op1) indicates the return on the investments of a firm. It is

hard to determine the relationship between the ROA and the risk of default. When a bank has

an amount of investments and thus a low ROA, the bank can be seen as being risky because

the low ROA corresponds with high credit default spreads. But investments can also be seen

as positive, because investments create future cash flows and income. On the other hand, it is

also possible that operating income decreases and the level of investments stays the same.

Then a reduction in the ROA will lead to an increase in the credit default swap. The ROE

(Op2) measures the return on equity. Evidently, the higher this ratio, the lower the chance that

a firm will default.

13

The last category is liquidity. This category consists of the ratios net loans/deposits and short-

term funding (Liq1) and liquid assets/deposits and short-term funding (Liq2). The net loans

divided by deposits ratio measures the liquidity. The relationship between the ratio and the

risk of default by a term can be regarded positive or negative because in case a bank contains

a smaller amount of deposits and lower equity, then this is not positively perceived by the

market. This will lead to an increase in the credit default swaps. A positive response by the

market is caused when the bank has a high number of loans and the same level of deposits.

The relationship is in this case negative and will lead to a decrease in credit default spreads. It

is positively perceived by the market because the loans are the core business of commercial

banks. The ratio liquid assets to deposits and short-term funding has a negative relationship

with the risk of default. An increase in the ratio means that the bank is highly liquid and

therefore not very vulnerable.

The data needed for the examination will be collected from the databases DataStream and

Bankscope. Two databases will be used due to the fact that some variables are not available in

one of the two databases, but are available in the other database. After all the data has been

found, merging of the data has to take place.

14

Table II

Summary statistics on the explanatory variables for the sample banks

This table shows the summary statistics of the sample banks that will be used throughout the paper. For this

paper American and European international banking groups are used. The sample period starts at 1 January 2005

and ends at 30 June 2014 and is based on quarterly data. The dependent variable in this sample is the Credit

default swap spread. The explanatory variables are divided into four categories. Loan loss reserve/gross loans

(Qa1) and Unreserved impaired loans /equity (Qa2) belong to the category asset quality. Tier 1 Ratio (Pat1) and

Leverage = equity/total assets (Pat2) can be classified into the category capital. The variables ROA = net

income/average total assets (Op1) and ROE = net income/average equity (Op2) are part of the operations

category. Last, Net loans/deposits and short-term funding (Liq1) and Liquid assets/deposits and short term

funding (Liq2) can be classified in the liquidity category. All variables are divided into an overall, between and

within estimation. The overall estimation explains the results over time and banks. The between estimation

explains the results between banks and the within estimation explains the change within banks over time. The

number of observations per explanatory variables are shown and also the accompanying standard deviation,

minimum value and the maximum value. The number of banks in this sample is 55.

Variable Mean Std. Dev. Min Max Observations Asset quality

Qa1 Overall 137.5212 122.6414 1 375 N= 2145

Between 90.7705 1 316.2308 n= 55

Within 83.3527 -177.7096 394.9571 T= 39

Qa2 Overall 106.3259 184.6743 1 646 N= 2145

Between 64.2649 1 236.1795 n= 55

Within 173.3431 -128.8536 714.9669 T= 39

Capital

Pat1 Overall 85.2991 65.6749 1 200 N= 1879

Between 38.8530 2 123.7436 n= 55

Within 59.0025 -36.4445 215.7350 T= 34.1636

Pat2 Overall 139.2529 98.3434 1 318 N= 1993

Between 63.7861 4 251.3846 n= 55

Within 81.6510 -108.1317 365.5862 T= 36.2364

Operations

Op1 Overall 852.3124 515.2334 1 1687 N= 2145

Between 364.0099 181.0513 1644.4870 n= 55

Within 367.8460 -497.1748 2261.3120 T= 39

Op2 Overall 367.7580 245.4249 1 826 N= 2145

Between 128.3674 146.7692 670.7436 n= 55

Within 209.8743 -173.9855 1042.9890 T= 39

Liquidity

Liq1 Overall 135.8583 217.3556 1 730 N= 2145

Between 63.4747 1.4103 263.1538 n= 55

Within 208.0524 -126.2956 810.1660 T= 39

Liq2 Overall 139.1333 220.4223 1 742 N= 2145

Between 65.0619 22.6410 273.5641 n= 55

Within 210.7795 -133.4308 841.4923 T= 39

15

In table II the summary statistics of the independent variables regarding the regression are

depicted. The first two variables related to asset quality show an overall low mean which

indicates good portfolio quality of 68% of the companies in the sample. From Table II, it can

also be noted that for the qa1 variable there is less within variation than between variation.

This suggests that there is more variation between the banks in the sample. For qa2 the

opposite is true. The between estimation shows that between the banks occurs less variation

than within a particular bank over time. With regard to the second category, capital, the mean

regarding to Pat1 is low which signals that on average many banks have a low risk buffers.

The variation of the position of the banks included in the sample on the other hand is large

which indicates that some banks have high risk buffers which in turn has a positive effect on

the level of the credit default swap spread of those banks. The mean of Pat2 suggests that

banks in the sample have average debt levels. The variation of the position of the banks

included in the sample is moderate and the within estimation is larger than the between

estimation. Therefore, there is more variation of individual banks’ characteristics (with regard

to capital) over time than there is across banks. The summary statistics of the category

operations are also outlined in the table and show average values for the variables. The results

regarding to the first variable Op1 signal that that the market does not perceive the banks in

the sample as risky. The between estimator for Op1 is very large and shows that for this

variable there is a large difference between banks. For the individual banks there is not much

variation over time as shown by the within estimator shows. For the Op2 variable the opposite

holds. The within estimator is slightly larger than the between estimator. Last, the liquidity

variables indicate low average values for both variables in the liquidity class. For both

variables, Liq1 and Liq2 the between estimators are considerably lower than the within

variables. There is a higher variation within the individual banks with regard to level of

liquidity of banks.

16

Table III below depicts the correlation matrix regarding the explanatory variables used in the

study. The level of correlation between the variables is of importance for the study due to the

fact that a high level of relation between two variables can lead to changes in the pattern.

Table III

Correlation matrix regarding the explanatory variables This table shows the level of correlation between the explanatory variables used in this study. The explanatory

variables are divided into four categories. Loan loss reserve/gross loans (Qa1) and Unreserved impaired loans

/equity (Qa2) belong to the category asset quality. Tier 1 Ratio (Pat1) and Leverage = equity/total assets (Pat2)

can be classified into the category capital. The variables ROA = net income/average total assets (Op1) and ROE

= net income/average equity (Op2) are part of the operations category. Last, Net loans/deposits and short-term

funding (Liq1) and Liquid assets/deposits and short term funding (Liq2) can be classified in the liquidity

category.

Qa1 Qa2 Pat1 Pat2 Op1 Op2 Liq1 Liq2

Qa1 1.0000

Qa2 -0.1625 1.0000

Pat1 0.0694 0.2870 1.0000

Pat2 0.2060 0.0440 0.2031 1.0000

Op1 -0.0910 0.1694 0.2408 0.1566 1.0000

Op2 -0.0901 0.2165 0.2822 0.1017 0.7036 1.0000

Liq1 -0.0771 0.5488 0.2657 0.0677 0.2260 0.3135 1.0000

Liq2 -0.1608 0.8983 0.2584 0.0459 0.2075 0.2320 0.6068 1.0000

All variables show low correlation values with each other except for the variables Qa2 and

Liq2. These two variables show a correlation value of 0.8983 which is relatively high. The

VIF estimator shows a value of 1.44 and for this reason both variables will not be eliminated

from the regression. However, robust standard deviation will be used in the study because the

robust standard deviation has the advantage that it minimizes the variance in the values of the

variables.

C. Methodology.

For all the independent variables in each of the three different periods the descriptive statistics

consisting of the mean, the standard deviation, the maximum and the minimum will be

calculated. The five-year quotes will be used as the benchmark. Quarterly data is preferred to

annual date because with quarterly data the number of observations increases.

17

A panel regression will be exercised in order to determine whether the balance sheet can

explain the bank its credit default spreads. Panel data is considered for this examination

because it provides two dimensions. First, panel data provides a cross sectional dimension, in

this case banks, and panel data provides a time series dimension, in this paper quarters. In this

model it is also assumed that there is correlation in the credit default swap spreads over time

for a given bank, but the credit default swap spreads are independent across banks. Another

reason for the choice of panel data is because it allows to control for unobservable variables,

like certain variables that do not change across banks but do change over time.

The following generic model will be followed to determine the relationship.

CDSit = α + βLoan Loss Reserve/Gross Loansit + βUnreserved Impaired Loans/Equityit +

βUnreserved Impaired Loans/Equityit + βTier1Ratioit + βEquity/Total Assetsit + βROAit +

βROEit + βNet Loans/Deposits and Short Term Fundingit + βLiquid Assets/Deposits and

Short Term Fundingit + Dcrisis + εit

Which can also be written as

CDSit = α + β(Bankratios)it + dcrisis + εit

With

CDSit = credit default spread of bank i in quarterly time period t

α = constant

(Bankratios)it = The bank specific explanatory variables

Dcrisis = dummy variable that stands for the outbreak of the recent financial crisis which

started at 1 July 2007.

εit = error

i identifies the bank and the t indicates the relevant time period. The time period is expressed

in quarters. The time-varying bank specific explanatory variables are included in the model.

The regressions will be executed for the entire time horizon which starts at 1 January 2005

and ends at 30 June 2014. The tests will be performed using the statistical software STATA.

18

III. Results

In this chapter the results regarding the performed regressions will be analyzed. In table IV

the results regarding the Hausman test for the fixed versus random effects model. This test is

used in order to determine whether the random effects or the fixed effects model should be

used for analyzing the results of this paper.

Table IV

Hausman test for the fixed versus random effects model

This table represents the Hausman test for the fixed versus random effects model. The variables outlined in the

table are defined in section 2B. For all the explanatory variables in the paper the coefficients of the fixed and the

random model and their difference are given. Of this difference the accompanying standard deviation is shown in

the last column of the table. Further, explanations of the b and the B coefficients are given and the null

hypothesis which will be used for the test is also included. Below these explanations the results of the Hausman

test are outlined and indicate a significant result. The number of banks included in the sample is 55.

Coefficients

Variable Fixed (b) Random (B) Difference (b-B) Sqrt S.E.

Asset Quality

Qa1 0.6610 0.3183 0.0427 0.0727

Qa2 0.3759 0.2810 0.0949 0.0673

Capital

Pat1 -0.9243 -0.9506 0.0263 0.0582

Pat2 -0.3926 -0.3452 -0.0474 0.0646

Operations

Op1 -0.0618 -0.2002 0.1384 0.0485

Op2 0.1732 0.13308 0.0401 0.0356

Liquidity

Liq1 -0.1401 -0.2267 0.0866 0.1508

Liq2 -0.3926 -0.2749 -0.1177 0.0547

b= consistent under H0 and Ha; obtained from xtreg

B= inconsistent under Ha, efficient under H0; obtained from xtreg

Test: H0: difference in coefficients not systematic

Chi2(10) = (b-B) ‘ [(V_b-V_B)^(-1)] (b-B)

= 25.45

Prob>chi2 = 0.0046

(V_b-V_B is not positive definite)

19

From the “difference” column in the table can be noticed to what level the results from the

fixed effects model differ from the random effects model. In order to measure this a null

hypothesis has been drawn up for the Hausman test which states that the differences in the

coefficients of both models are zero. The result of the Hausman test is shown in the table in

the form of a Chi score with a value of 25.45. The p-value of this Chi score is very small with

a significant value of 0.0046 which indicates that the coefficients of the fixed effects model

differ from those of the random effects model. The significant p-value of 0.0046 leads to

strong rejection to of the null hypothesis that a Random Effects model provides consistent

estimates that the random effects estimator is fully efficient. Therefore, this study employs a

Fixed Effects Model for further empirical testing.

The variables used for testing in this paper vary over time and also, it is in the interest of the

paper to analyze the impact of these varying variables. For that reason it is interesting to use

the fixed effects model in the panel regression. Following the results of the Hausman test the

results of the fixed effects model are given in table V.

20

Table V

Panel regressions for the whole and the different periods

In this table the results of the fixed effects (within) regression are outlined. The group “Whole period” starts at 1

January 2005 and ends at 30 June 2014. The “Pre-Crisis period” begins at 1 January 2005 until 30 June 2007.

The “Crisis period” runs from 1 July 2007 until 31 March 2009. The Post-Crisis period starts at 1 April 2009 and

ends at 30 June 2014. The dependent variable in this regression is the credit default swap spread which is defined

in section 2A. All the independent variables which are already explained in section 2B are shown in the table

along with their coefficients and their robust standard errors resulting from the regression. The robust standard

errors are denoted in parentheses. In this panel regression are also the dummies Pre-Crisis period, Crisis period

and Post-Crisis period included. The number of observations and the number of banks included in the sample is

represented in the table. Among the results the R2 broken down in the within, the between and the overall

estimator can be found in the table. Last the F-statistics and the rho for all the panel regressions are shown in the

table.

*** Reports coefficients statistically significant from zero (1% level, two-tail test)

** 5% level

* 10% level

Variable Whole

period

Whole

period with

pre crisis

dummy

Whole

period with

crisis

dummy

Whole

period with

post-crisis

dummy

Crisis

period

Post crisis

period

Asset quality

Qa1 0.6850*** 0.4607*** 0.7033*** 0.3838** 1.0346** 0.3229

(0.1662) (0.1473) (0.1636) (0.1528) (0.4346) (0.2011)

Qa2 0.4456** 0.3432 0.4783** 0.0980 0.3231 -0.1601

(0.2063) (0.2100) (0.2091) (0.2121) (0.2030) 0.1928

Capital

Pat1 -0.8685*** -0.5582*** -0.9144*** -0.2156 -0.0137 -0.2450

(0.2470) (0.1956) (0.2467) (0.1978) (0.5564) (0.2958)

Pat2 -0.4539** -0.4430*** -0.4554** -0.4665** -0.8690 -0.3549

(0.1910) (0.1577) (0.1892) (0.1774) (0.4764) (0.2353)

Operations

Op1 -0.0584 -0.0461 -0.0574 -0.0054 -0.0385 -0.1784**

(0.0585) (0.0411) (0.0582) (0.0512) (0.1324) (0.0718)

Op2 -0.5552*** -0.2279*** -0.5508*** -0.4458*** -0.8282*** 0.1074

(0.0978) (0.0736) (0.0962) (0.0894) (0.1913) (0.1430)

Liquidity

Liq1 -0.1415** -0.0237 -0.1437** -0.0336 0.0818 -0.0486

(0.0587) (0.0462) (0.0581) (0.0529) (0.0839) (0.0960)

Liq2 -0.4543** -0.3283* -0.4874*** -0.0931 -0.1366 0.1909

(0.1708) (0.1759) (0.1717) (0.1732) (0.1681) 0.1465

Pre-Crisis

Dummy

-

363.3363***

(29.3029)

Crisis Dummy 34.2537

(29.4845)

Post-Crisis

Dummy

284.5633***

(45.9242)

N observations 1879 1879 1879 1879 336 1008

N sample

banks

55 55 55 55 48 48

Adjusted R2

within

0.3059 0.3917 0.3068 0.3690 0.2086 0.0330

Adjusted R2

between

0.0805 0.3224 0.0754 0.0897 0.0011 0.0322

Adjusted R2

overall

0.2382 0.3539 0.2391 0.3248 0.0630 0.0055

rho 0.3048 0.1383 0.3035 0.1845 0.3858 0.1679

F-statistic 43.02*** 77.12*** 42.92*** 55.76*** 11.70*** 3.08***

21

In table V the results from the several panel regressions are shown. Several panel regressions

have been performed with and without the dummies on the whole time period from January

2005 until 30 June 2014 and on the three specific time periods. The first subgroup, the pre-

crisis period runs from 1 January 2005 until 30 June 2007. The crisis period includes the

period from 1 July 2007 to 30 March 2009. Last, the post-crisis period runs from 1 April 2009

to 30 June 2014.

For all these periods the regressions are run in order to see whether the explanatory variables

have a statistically significant influence on the dependent variable, the credit default swap

spread. In the whole period one by one the pre-crisis, the crisis and the post-crisis dummy

have been included in order to examine whether the dummies have a statistically significant

influence on the credit default swap spread. In the whole period the sample consisted of 55

banks with an accompanying 1879 observations and in the crisis and the post-crisis period 48

banks of the total 55 banks were included in the sample. For the crisis period the number of

observations was quite lower than the number of observations in the post-crisis period. The

reason for this difference is the length of the time period in which the observations are

measured. The crisis period consisted of only two years and the post-crisis period included

five years.

A. Panel regression results of the whole period

For the whole period, almost all explanatory variables used in the regression have a

statistically significant impact on the dependent variable, the credit default swap spread. Also

the model has a statistically significant effect. The rho shows a value of 0.3048 so 30,48% of

the variation in the results can be explained by the individual bank characteristics. The other

part of the variation in the results is due to factors outside the model. In this model the

adjusted R2 is broken down in three separate estimators. The adjusted R2within estimator

shows that 30,59% of the variation of the banks over time can be explained by the model.

Only 8,05% of the variation between the banks in the sample can be explained by the model.

Overall the adjusted R2 model is a fairly weak estimator and shows that only 23,82% of the

variation can be explained by the model. However, the study will be more focused on the

effects of the individual covariates on the credit default swap spreads than on the effects of

the model on the credit default swap spreads. So the low overall result does not have severe

consequences on the course of the study.

22

In the asset quality category, both explanatory variables show a significant positive relation to

the credit default swap spread. For the first variable, Qa1, the standard deviation is lower

which illustrates a low variation in the values of this variable among the banks. Many banks

of the total banks in the sample remain around the average value. The sign of the coefficient is

positive as expected prior to the study and indicates that an increase in the ratio leads to

higher levels of credit default swap spreads. This means that when in this ratio the reserves of

loan losses increases faster than the value of gross loans the bank is not performing well. A

higher loan loss reserve signals to the market that the probability of default on loans increases.

So investors will respond to this signal and buy credit default swaps.

For the second variable included in the asset quality category, Qa2, the variation in the values

of this variable of the banks in the sample is a bit larger which means a larger variation in the

values in Qa2 of the values of the banks. The positive coefficient is in line with the

expectations prior to the study regarding this variable, that is when the ratio increases the

spread of the credit default swaps will also increase. The increase in the ratio will be

interpreted by the market as a higher probability of default. The market will adjusts its

expectations in the way that they will expect a decrease in the earnings of the banks. In

response to these market expectations the investors will buy more credit default swaps and

this will have an impact on the level of the spreads.

The results of the second category, capital, are in line with the study’s ex ante expectations.

Both variables have a significant negative impact on the level of the credit default swap

spreads and thus the performance of the bank. An increase in the ratio will be followed by a

decrease in the level of the credit default swap spreads of the banks included in the sample.

Pat1 can be seen as a measure of capital adequacy of the bank. With regard to the banks in the

sample, the variation in the values of this variable is somewhat high which means that the

variation in the values of Pat1 of these banks is large. The reason for this somewhat large

spread is that some banks are more able to absorb losses than other banks or that some banks

engage in higher risk taking than other banks. The market immediately responds to changes in

the Pat1 variable. When a bank has an increased ability to absorb losses or incurs lower risk

taking this will be perceived by the market as an increase in the risk buffers of the bank, a

lower probability of default, and thus investors will be less inclined to buy protection.

23

Pat2, also included in the category capital, has low variation in the values which means that

the range in which the values of the banks lay is not very large. This variable is based on the

balance sheet concepts of the participating banks in the sample. The choice for the balance

sheet concept is established on the fact that balance sheet leverage is widely accepted and

most visible because it also the indebtedness of the bank. Banks for which this ratio increases,

especially when due to a decrease in the level of equity, are viewed as more risky by the

market than other banks. Investors in the market then respond to this increase in default risk

and will protect themselves by buying credit default swap spreads.

The performance of the banks is also significantly affected by the Op2 of the category

operations. The Op1 does not have a significant impact on the level of the credit default swap

spreads. For Op1 both a positive or a negative sign are suitable in this regression. A positive

Op1 would mean that the market views high level of investments as generators of future

income and cash flows. A negative Op1 implies that a high level of investments is viewed as

risky. In this regression the Op1 turns out to be negative and does not have a significant

impact on the dependent variable. The performance of the banks is not affected by the level of

Op1. As expected, the Op2 in this regression is negative and has a significant impact on the

level of the credit default swap spreads of the banks in the sample. The standard deviation of

this variable is very low which indicates that the variation in the values of Op2 of the banks in

the sample is not high. 68% of the banks have a Op2 value around the average of all banks.

An increase in the ratio leads to a reduction in the credit default swap spreads since a higher

Op2 is associated with a lower probability of default by the market. Investors interpret the

increase in the ratio as an increase in the earnings and thus in the profitability of the firm.

Liq1 measures the level of liquidity in the firm. As stated earlier, both coefficient signs are

suitable for this variable. However, the results show a significant negative coefficient. The

core business of most of the banks in this sample is providing loans to its customers and an

increase in the ratio signals the market that the bank is performing well and thus has a lower

probability of default. Investors will not change their behavior as a consequence. The standard

deviation of Liq1 is very low which means that the values of all banks lay in the same range

and not far away from the average value of all the banks included in the sample. The second

explanatory variable, Liq2, has also a negative significant impact on the level of credit default

swap spreads which is in line with the prior expectations. An increase in liquidity indicates

that the bank has more wealth and is less vulnerable to liquidity shocks.

24

Investors respond to this information by not buying insurance because the bank is viewed as a

strong entity. The standard deviation for this variable is high. The values of Liq2 are spread

over a larger range thereby much variation exists among the banks included in the sample.

B. Panel regression results of the dummy variables in the whole period

The low number of observations in the crisis period can also be seen as the main reason for

the interesting result that the crisis dummy variable has no significant influence on the credit

default swap spread. The pre-crisis dummy and the post-crisis dummy however do have a

statistically significant influence on the credit default swap spread. First, the pre-crisis dummy

has a p-value of 0.000 and a low standard deviation which means that the range of the credit

default swap spreads of the individual banks is not too large. Also, the results of the model as

shown by the F-statistic show a significant outcome. The model thus has a significant effect

on the level of the credit default swap spreads. The rho depicts a low value of 13,83% which

means that a lot of the variation in the model can be attributed to factors other than bank

specific characteristics. The adjusted R2within estimator is fairly high. 39,17% of the variation

of the individual banks over time can be defined by the model. The same holds for the

adjusted R2between estimator. This estimator has a strong value of 32,24% so a lot of the

variation between the banks included in the sample can be attributed to bank specific traits. So

overall the adjusted R2 is fairly strong in this model. The significant value of the pre-crisis

dummy indicates that the performance of the bank is affected by the change in the financial

circumstances. The coefficient has a negative value which indicates that the dummy has a

negative influence on the level of the credit default swap spreads of the banks. This has in

turn a positive impact on the profitability of the bank. Investors perceive a decrease in the

default risk of the bank and are less inclined to buy credit default swaps. Due to this

perception of investors the demand for the default swaps is low in this period and this decline

in demand leads to low default swap spreads.

The model in which the crisis dummy variable is included has a significant influence on the

level of the credit default swap spreads in the crisis period as indicated by the F-statistics in

the table. The rho results in a value of 30,35% and is not very strong. A part of the variation in

the results can be characterized as bank specific variation due to bank specific factors but the

main part of the variation is due to idiosyncratic risk. Also the adjusted overall R2 estimator is

moderate. Only 23,91% of the variation in the results can be explained by the model. The

same yields for the adjusted R2within and the R2between estimator.

25

The adjusted R2within estimator shows a moderate value of 30,68% and the R2between

estimator shows a very weak value of 7,54%. The crisis dummy itself has not a significant

effect on the level of the credit default swap spreads. The accompanying standard deviation is

very large which means that the spread of the credit default swaps varies large among the

banks included in the sample. This insignificant result indicates that it is not possible to make

any reliable inferences or conclusions with regard to the effects of the crisis on the banks’

performances. However, it must be noted that the small sample size should be taken into

consideration when assessing this result. It could be very well the case that the small sample

size is the main reason for the insignificant outcome.

The model in which the post-crisis dummy is included has a significant effect on the level of

the credit default swap spreads. The rho is very low in this period with a result of 18,45%.

The adjusted R2within and the R2between estimator depict a moderate and a very weak value

of 36,90% and 8,97%. A low part of the variation in the individual banks and an even lower

part of variation between banks is due to bank specific characteristics. The main variation in

the values is due to idiosyncratic risk. Overall, the R2 has a value of 32,48% and thus the

explaining power of the model for the variation is moderate. Furthermore, the post-crisis

dummy shows a statistically significant effect on the dependent variable. In this period the

standard deviation also shows a low value which indicates that the level of the credit default

swap spreads for the banks are not spread out over a large range. The post-crisis dummy is

significantly positive related with the level of the credit default swap spreads. When the

market perceives the banks as riskier in this period it will lead to an increase in the demand

for credit default swaps and this in turn leads to higher spreads. Investors view earnings as a

major determinant of risk and associate the developments in this variable with a higher or

lower probability of default.

Asset quality has the expected positive sign and Qa1 has a significant impact on the

dependent variable in all periods. For Qa1, the crisis period and the post-crisis period standard

deviations are higher than in the pre-crisis period. In these two periods the values of asset

quality vary more among the banks included in the sample than in the pre-crisis period. Also,

in this period the coefficient is higher which means that a change in the ratio leads to a higher

effect on the dependent variable than when the same change happens in the other two periods.

An increase in the ratio for the crisis period results in investors buying more credit default

swaps than in the other two periods.

26

Qa2 has only in the crisis period a significant impact on the level of the credit default swap

spreads. The standard deviation remains around the same value in all three periods. The

significant result means that investors only in the crisis period worry about the low quality of

the bank loan portfolios held by the banks in the sample. In this period an increase in the ratio

results in an increase in demand for credit default swaps and in turn the accompanying spread

increases.

In the capital category Pat1 has the expected negative impact and on the dependent variable

and has only in the pre-crisis and in the crisis period a significant influence on the dependent

variable. The standard deviation has a higher value in the crisis period. In this period the

values of the capital adequacy of banks have a larger variation in the values of Pat1. The

reason for this can be that some banks of the sample are more active in aggressive risk taking

which the market views as risky. This behavior of banks will as a results be corrected with a

higher levels of credit default swap spreads. This result is also proven by the higher

coefficient in the crisis period. Changes in the ratio have the most impact on the dependent

variable in the crisis period.

The second variable of the capital category, Pat2, is significant for all three periods. For this

variable the sign is negative as expected and the coefficient is slowly increasing which means

that the impact of a change in the indebtedness of the bank gains more strength in time. When

the level of liabilities increases the market views the entity as risky and the investors response

to this change by insuring themselves against the increased change of a default. The standard

deviation of the variable has remained constant in the three sub-periods. In the crisis-period

the deviation increased a little, but in the post-crisis period the deviation lowered again which

means that the spread of the values also decreased.

Earnings have an important influence on the level of credit risk included in the credit default

swap. This assumption is also proven by the significance of the variables Op1 and Op2 in the

results in the table. However, the coefficients of the variable Op1are not very large. Changes

in the variable only have a small negative influence on the dependent variable for all three sub

periods. In the crisis period the coefficient and the standard deviation of the variable show an

increase in the values and both decrease slightly in the post-crisis period. Thus, in all three

periods a change in Op1 leads to a same decrease in the level of the credit default swap

spread. Op2 is significant in all three periods and has a low standard deviation in all three

periods. The standard deviation shows a little bit of fluctuation in the three periods.

27

An interesting result is the fact that coefficient shows a large increase in value in the crisis

period. Earnings are a major indicator of credit risk included in the credit default swaps. In the

pre-crisis period investors do not seem to be paying a lot of attention to earnings, but in the

crisis period the market puts a lot of value on the level of the earnings in order to determine

the probability of default. As expected, an increase in the variable immediately leads to a

reduction in the dependent variable.

The negative sign of the liquidity category is in line with the expectations prior the analysis.

Of the liquidity category, Liq1 is only negatively significant in the crisis-period. In the crisis

period investors pay a lot more attention to the level of liquidity of the banks in the sample.

Based on the level of Liq1 investors form their expectations with regard to the probability of a

default. In the post-crisis period investors don’t value the liquidity anymore as much as they

did in the crisis period. Changes in this ratio don’t have a significant effect on the dependent

variable and over time the standard deviation shows stable values. The standard deviation of

Liq2 also depicts stable values over time. The spread of the banks thus remains the same in

the period. The variable is only slightly significant in the pre-crisis period. The market views

liquidity as a less important indicator of credit risk. In the crisis period the variable gains more

importance with regard to the probability of default, but after the crisis period the variable

loses its importance and does not have a significant effect on the dependent variable.

C. Panel regression results of the three sub-periods

As can be seen in table V, in the columns crisis period and post-crisis period, two further

panel regressions were conducted in order to understand whether the relationship between the

balance sheet explanatory variables and the credit default swap spreads changes with macro-

economic conditions. Due to too few observations in the pre-crisis period a panel regression

for was not feasible for this period. So the focus will be on the panel regression results of the

crisis period with 336 observations and the post-crisis period with 1008 observations.

28

The model has a significant impact on the level of credit default swaps in the crisis period as

shown by the F-statistic. The rho shows a value of 0.3858 which indicates that 38,58% of the

variation in the results can be explained by individual bank characteristics. The value is not

very high so a lot of the variation is due to idiosyncratic risk. For this period the adjusted R2 is

broken down into an adjusted within, between and overall estimator. All three estimators

show low values which means that the model does not explain much of the variation of an

individual bank over time and between banks. Overall, the adjusted R2 indicates that the bank

credit default swap spreads reflect a low amount of risk expressed by the explanatory

variables. In the crisis period only the categories asset quality and operations show a

significant influence on the credit default swap spreads and thus the performance of the banks.

For the asset quality category, Qa1 has the expected positive significant effect on the

dependent variable. In this period the market values the quality of the loan portfolios of banks.

The higher the ratio, the lower the quality of the portfolio and the market will perceive this

low quality as risky. Investors will react to this change by buying credit default swaps as

insurance. In the operations category Op2 has a negative statistical significant effect on the

level of the credit default swap spreads. This is in line with the expectations. This ratio

reflects the return on the banks’ own equity. An increase in the ratio means that the market

views the bank as a strong and well performing entity. In reaction to this perception the

demand for credit default swaps lowers.

In the post-crisis period the model has an important impact on the dependent variable with a

statistical significant F-value. The rho has decreased in value in comparison to the crisis

period and thus more variation in the results of the individual banks is due to idiosyncratic

risk. The adjusted R2within estimator has decreased in explanatory power for this period. The

adjusted R2within estimator shows that in this period only 3,30% of the variation in the credit

default swap spread for the banks over time can be explained by the model. For the adjusted

R2between estimator the model explains only 0,55% of the variation between banks. Overall

the R2 is a weak estimator in this period. A lot of the variation cannot be explained by the

model and must be attributed to idiosyncratic risk. Only the operations category has a

significant influence on the level of the credit default swap spreads. In this category, the Op1

shows a negative statistical value. A decrease in operating income leads to a lower ROA for

the same level of investments and this signals a higher probability of default. As a

consequence, the demand for credit default swaps will increase.

29

IV. Conclusion

The aim of this study is examine whether credit default swap spreads are a suitable indicator

of the performance of banks. In order to determine whether credit default swaps are a good

proxy a sample of international American and European banks is used. From these banks

senior credit default swap spreads with a maturity of five years were used as the dependent

variable. For the explanatory variables balance sheet estimators of the banks included in the

sample were used. The variables Loan loss reserve/gross loans (Qa1) and Unreserved

impaired loans /equity (Qa2) are used to proxy the asset quality. Tier 1 Ratio (Pat1) and

Leverage = equity/total assets (Pat2) can be classified into the category capital. The variables

ROA = net income/average total assets (Op1) and ROE = net income/average equity (Op2)

are part of the operations category. Last, Net loans/deposits and short-term funding (Liq1) and

Liquid assets/deposits and short term funding (Liq2) can be classified in the liquidity

category.

The time horizon in the study is the period from 1 January 2005 until 30 June 2014. This

period is subsequently divided into three sub-periods namely the pre-crisis period which starts

at 1 January 2005 and ends at 30 June 2007, the crisis period which runs from 1 July 2007

until 30 March 2009 and the post-crisis period which includes the period from 1 April 2009

until 30 June 2014.

Several regressions are run to interpret whether the credit default swap spreads are good

indicators of the performance of banks. First, a panel regression over the entire period is

performed. The results from this regression indicate that almost all categories have a

significant effect on the height of the credit default swap spreads. The variable Op1 does not

have any explanatory power in the changes of the dependent variable. For the rest of the

variables, the market immediately responds to changes in the variables by increasing or

decreasing the demand for credit default swaps.

In addition to this regression over the whole period, three regressions are run and each

including a dummy variable. The outcomes show an interesting result, namely that the

dummy variable crisis has no significant effect on the level of credit default swap spreads, but

the dummy variables pre-crisis and post-crisis do. The main reason for this unexpected result

can be found in the number of observations in the crisis period.

30

The pre-crisis dummy has a significant impact on the level of the credit default swap spreads.

The negative influence of the dummy leads to lower level of credit default swap spreads, but

to an increase in the performance of the bank. Investors are less inclined to buy credit default

swaps because they perceive a decrease in the default risk of the banks. As a consequence of

this behavior the demand for insurance decreases and this reduction leads to a decrease in the

credit default swap spreads. The decrease in the spreads signals a stronger performance of a

particular bank. Also the post-crisis dummy has a significant influence on the level of the

credit default swap spreads in the sense that demand for credit default swaps increases when

the market perceives the banks as riskier. This in turn will lead to higher spreads.

In the asset quality category Qa1 is in all three sub periods an important estimator for the

performance of banks. In all three periods investors view the quality of the portfolio as an

important indicator of probability of default. When the quality of the portfolio of a bank

lowers, the ratio of the variable increases and investors will be buying more credit default

swaps which results in a higher spread and lower bank performance. The variable Qa2 has

only in the crisis period a significant effect. As expected, the capital category has a negative

significant influence on the dependent variable. The variable Pat1 shows only in the post-

crisis period an insignificant value. Pat2 has in all three periods a negative significant effect

on the performance of the bank. Investors associate an increase in the ratio with a weaker

entity and will be buying credit default swaps to protect themselves against default risk. As a

consequence the spread will increase. The assumption that earnings have an important

influence on the level of credit risk is proven by the results of the operations category. Op2

has a significant influence in all three different periods. An increase in earnings signals

strength to the market and investors will immediately change their behavior by decreasing

their demand for insurance. For the last category, liquidity, Liq1 is only negatively significant

in the crisis period. Investors add more weight to the level of liquidity of banks in the sample

in the crisis period. In the post-crisis period investors do not pay a lot attention to the value of

liquidity as they did in the crisis period. Liq2 is only slightly significant in the pre-crisis

period, but more significant in the crisis period. Leading up to the crisis period, the variable

gains more importance.

31

In order to understand the relationship between the balance sheet explanatory variables and

credit default swap spreads when the macro economic conditions change two further panel

regressions were conducted. Due to few observations results of the pre-crisis period could not

be examined. The results of these two regressions show that only the asset quality and the

operations category have a significant effect on the level of credit default swap spreads. In

particular, only in the crisis period investors worry about the quality of the portfolios held by

banks. Liq1 shows only in the post-crisis period a significant result and the variable Liq2

shows only in the crisis period a significant value. Overall the market regard liquidity as an

important estimator for the performance of banks.

The results of the several regressions show that the four categories used in the study have a

significant influence on the dependent variable, the credit default swap spreads. The market

responds to changes in the variables by increasing and decreasing the demand for insurance.

These changes in the variables signal the strength of a particular bank and thus its

performance. It can be stated that the credit default swaps are a good indicator of the

performance of the bank and the H0 hypothesis can thus be rejected.

32

V. References

Acharya, V.V., Johnson, T.C., 2007, “Insider trading in credit derivatives”, Journal of

Finance Economics, vol.84, pages 110-141

Annaert, J., De Ceuster, M., Van Roy, M., & Vespro, C., 2013, “What determines euro area

bank CDS spreads?”, Journal of International Money and Finance, vol. 32, pages 444-461

Bank for International Settlements., 2010, “80th Annual Report”, Basel, 28 June 2010, pages

1-206

Brunnermeier, M.K., 2009, “Deciphering the Liquidity and Credit Crunch 2007-2008”,

Journal of Economic Perspectives, vol. 23(1), pages 77-100

Callen, J.L., Livnat, J., Segal, D., 2009, “The Impact of Earnings on the Pricing of Credit

Default Swaps”, The Accounting Review, vol.84(5), pages 1363-1394

Chiaramonte, L., Casu, B., 2013, “The determinants of bank CDS spreads: evidence from the

financial crisis”, The European Journal of Finance, Routledge, vol. 17(9), pages 861-887

Duffee, G.R., Zhou, C., 2001, “Credit derivatives in banking: Useful tools for managing

riks?”, Journal of Monetary Economics, vol. 45, pages 25-54

Ericsson, J., Jacobs, K., & Oviedo, R., 2009, “The Determinants of Credit Default Swap

Premia”, Journal of Financial and Quantitative Analysis, vol.44(1), pages 109-132

Flannery, M.J., Houston, J.F., Partnoy, F., 2010, “Credit Default Swap Spreads as viable

substitutes for credit ratings”, Univerity of Pennsylvania Law Review, vol.158(7), pages 2085-

2123

Shan, S.C., Tang, D.Y., Yan, H., 2014, “Did CDS Make Banks Riskier? The Effects of Credit

Default Swaps on Bank Capital and Lending”, pages 1-80

Subrahmanyam, M.G., Tang, D.Y., Wang, S.Q., 2014, “Does the Tail Wag the Dog?: The

Effect of Credit Default Swaps on Credit Risk”, The Review of Financial Studies, vol.27(10),

pages 2926-2960