determinants of bank credit default swap spreads: the role of the housing sector

North American Journal of Economics and Finance 24 (2013) 243– 259

Contents lists available at SciVerse ScienceDirect

North American Journal ofEconomics and Finance

Determinants of bank credit default swap spreads: The roleof the housing sector

Nadia Benbouzid, Sushanta Mallick ∗

Queen Mary University of London, School of Business and Management, Mile End Road, London E1 4NS, UK

a r t i c l e i n f o

JEL classification:G01G15

Keywords:Corporate CDS spreadsHousing marketCredit crisisDefault riskLiquidity risk

a b s t r a c t

This paper relates credit spreads (CDS prices) in the UK bankingsector with the performance of the housing sector. Using data onbanking sector CDS spreads for the period January 2004 to April2011, we find that house price dynamics are a key driving factorbehind the increase in credit spreads as reflected in CDS prices. Alsowe find that as stock prices increase, both bank capital and bankborrowing capacity increase that in turn decreases credit risk. Fur-thermore as banking sector liquidity increases banks tend to lendto less credit-worthy (subprime) borrowers that in turn increasescredit risk in the banking sector. Collectively the results shed lighton the determinants of credit risk in the banking sector.

© 2012 Elsevier Inc. All rights reserved.

1. Introduction

The period from 2001 and 2006 witnessed low quality underwriting standards and a higher thannormal default rate on home mortgages (see Klomp, 2010; Taylor, 2009). Financial engineering aswell as securitization allowed banks and other financial institutions to expand their lending whileat the same time satisfying regulatory capital requirements. The subsequent mortgage crisis thatcommenced in the summer of 2007 led to turmoil in the mortgage markets. Large banking corporationsand other financial institutions were obliged to write off losses on many of the structured derivativesand securitized assets on their balance sheet. This paper is a study of the relationship between housingprices in the UK and credit spreads in the UK Banking sector.

The sudden and sharp decline in the house prices in the aftermath of the crisis has direct implica-tions for the credit default swap (hereafter CDS spread) of financial institutions. During the period of

∗ Corresponding author. Tel.: +44 20 7882 7447.E-mail addresses: [email protected] (N. Benbouzid), [email protected] (S. Mallick).

1062-9408/$ – see front matter © 2012 Elsevier Inc. All rights reserved.http://dx.doi.org/10.1016/j.najef.2012.10.004

dx.doi.org/10.1016/j.najef.2012.10.004

http://www.sciencedirect.com/science/journal/10629408

mailto:[email protected]

mailto:[email protected]

dx.doi.org/10.1016/j.najef.2012.10.004

244 N. Benbouzid, S. Mallick / North American Journal of Economics and Finance 24 (2013) 243– 259

credit expansion between 2000 and 2006, credit underwriting standards of mortgage securities wereassociated with lax supervision by financial authorities. With the continuous rise in house prices andthe increasing securitization activities, mortgage lending expanded substantially. As long as the realestate prices increased, lending to lower quality borrowers did not pose a problem for financial insti-tutions as they could always resell houses at a higher price in the secondary market. However, whenresidential prices drastically plummeted and mortgage rates substantially increased, with borrowers’personal income growth reaching its lowest level, sub-prime mortgages fell in value and resulted inhuge financial losses. This caused a rise in both the overall credit risk of the lending institutions andtheir CDS spread. The CDS market had an outstanding notional amount of $631.5 billion in 2001 andgrew substantially over the following years. However at the end of 2008, due to the financial crisis themarket experienced a sharp decline in trading volume and notional amount outstanding (ISDA, 2010).

The CDS spread is usually interpreted as the price of the credit default risk of the underlying asset(Ötker-Robe & Podpiera, 2010). A CDS contract is similar to insurance contracts, meaning that the buyerof the contract, also referred to as the protection buyer, makes a series of payments, i.e. the spread,to the protection seller of the CDS. In case a credit event occurs, such as a default, a restructuringor bankruptcy of the financial institution involved, the protection buyer is entitled to receive a pay-off from the protection seller. The payment is usually equal to the par value of the underlying asset,typically a bond. If no credit event occurs, the protection seller receives quarterly premium payments(also referred to as the CDS spread) from the protection buyer.

The literature on credit risk, more specifically the CDS spread, focuses on analysing the key struc-tural determinants of CDS spreads. These include the risk free rate and the yield spread, but not theunderlying economic factors such as the housing market.1 The risk free rate and the yield spreadare significant factors in explaining the CDS spread (Alexander & Kaeck, 2008; Bevan & Garzarelli,2000; Duffie & Singleton, 1999; In, Brown, & Fang, 2003; Lekkos & Milas, 2001; Naifar, 2010). Otherresearchers including Collin-Dufresne, Goldstein, and Martin (2001), Campbell and Taksler (2003),and Benkert (2004) analyse the CDS spread by focusing on firm level data and incorporate financialvariables and volatility. Recent research studies the impact of credit ratings on the CDS spreads anddemonstrates that it is important in determining the spread at the firm-level (Fabozzi, Cheng, & Chen,2007; Hull, Predescu, & White, 2004).

With the financial crisis, it became clear that credit risk is not only related to interest rates, yieldspread and financial leverage, but most defaults that occurred during the crisis were driven by anunderlying factor that is closely rated to the house prices. This study investigates whether and towhat extent housing sector prices feedback into the financial institution profits and the bank CDSspread. This is a key contribution of our paper, as no previous research has conducted such analysis.We therefore analyse the impact of housing market on the CDS spread in the UK banking sector, alongwith other factors such as money market yield spreads and the stock market index in the UK.

Three alternative empirical methods are used in this paper to determine the presence of a co-integrating relationship between the variables. We use the Johansen’s method and the Dynamic OLSapproach (Stock Watson) in order to establish the key determinants of the CDS spread in the long run.We then employ a structural vector-auto-regression (SVAR) model in order to analyse the short termdeterminants of the CDS spread. The findings suggest that the house price dynamics are a key drivingfactor behind the recent collapse of corporate CDS market influencing credit risk. Both Johansen’s andStock Watson’s methods indicate the presence of a negative relationship between the house prices andthe CDS spread. This finding implies that both the banking sector credit market and the housing marketare strongly related. Thus, financial distress in the housing market is highly likely to get transmittedto the credit market and impact related markets.

Furthermore, we find a negative relationship between the CDS spread and the yield spread in thelong run, implying that as investors demand higher yield to compensate for bearing extra risk, itcould reflect lower likelihood of credit default in future. In addition, the FTSE 100 index appears to bepositive and significant under the DOLS method. This implies that as the stock index increases, both

1 See for example Hammoudeh and Sari (2011) for sectoral CDS focusing on the financial sectors and examining the linkageof such sectoral CDS with interest rates and stock market only.

N. Benbouzid, S. Mallick / North American Journal of Economics and Finance 24 (2013) 243– 259 245

banks’ capital and its borrowing capacity rise, leading to higher credit risk. Liquidity (TED) was foundto be insignificant in the long run, but in the short run, the structural VAR model indicates that it issignificant and positively related to the CDS spread. Therefore, as liquidity goes up, banks tend to lendmore to subprime borrowers, thus increasing credit risk and the overall CDS spread. In fact, from thestructural VAR model, the only variable that is statistically significant in the short run is TED.

The remaining part of this paper is structured as follows. Section 2 discusses the literature relatedto the determinants of the CDS spread. Section 3 describes the data used for different determinants.Section 4 is divided into 3 subsections with Section 4.1 focusing on the unit root tests (the AugmentedDickey Fuller test and the Zivot–Andrews test for structural breaks). In Section 4.2, we discuss thecointegration tests (Johansen’s method and the Stock and Watson’s Dynamic OLS method), while inSection 4.3 we focus on the short-term effects by applying a recursive structural VAR model. Section5 concludes the paper.

2. Determinants of the CDS spread

Understanding the determinants of credit spreads is crucial for financial regulators, traders, andpolicy makers, given the growth in the credit derivative market over the last decade. A large body ofliterature has focused on analysing credit defaults and investigating the reasons why the CDS spreadis considered to be a better proxy for default risk compared to bond spreads. In fact, researchers earlierrelied on bond spreads in order to get an approximation of the level of credit default risk. Duffie andSingleton (1999) and Hull et al. (2004) demonstrate that the CDS spreads are related to the creditspread implicit in bond prices. Bhanot (2001) highlights problems associated with bond spreads asindicators of corporate risk, discussing the lack of reliability of the Moody’s index as a corporate defaultindicator.

There are a number of reasons that make the CDS spread a better proxy for default risk. First, CDSspreads can be easily observed, while bond spreads have to be derived using a risk free benchmarkrate. Furthermore, it can be sometimes difficult to choose the appropriate risk free rate (Houweling& Vorst, 2005). In addition, the estimation of the credit premium in bond spreads is usually affectedby financial market variables.2 Moreover, CDS spreads react more rapidly to information related tothe credit quality of the underlying reference entity compared to bond spreads (Blanco, Brennan, &Ian, 2005; Hull et al., 2004; Zhu, 2006). Furthermore, credit ratings of long term bonds are influencedby CDS spread fluctuations (Norden & Weber, 2004). Given all the limitations of bond spreads, in thispaper, the CDS spread will be used as a proxy for credit default risk.

Evidence suggests that default losses account only for a small fraction of the credit spread (Anderson& Sundaresan, 2000). Duffie and Singleton (1997), Duffie (1998), Lekkos and Milas (2001) and In et al.(2003) find that the risk free rates and their term structure are important determinants of the creditspread. These results were supported by early findings conducted by Fama (1984) and Estrella andHardouvelis (1991). Friedman and Kuttner (1992) demonstrate that interest rate variations are thestrongest factor that predicts default risk. Furthermore, Longstaff and Schwartz (1995) show evidenceof a negative relationship between the probability of default and the interest rate. They explain thathigher interest rates cause the risk neutral drift of the company’s value to go up, thus leading to alower probability of default and consequently a narrow credit spread.

This negative relationship between credit spreads and interest rates was also observed (Bevan &Garzarelli, 2000) in the short run, eventually turning into a positive relationship in the long run. Estrellaand Hardouvelis (1991), Fehle (2003) and In et al. (2003) all found a negative relationship betweenthe credit spread, interest rate and the yield curve. In fact, during economic downturns or companyinsolvencies, the interest rate level tends to be low. Thus, if a yield curve was steeper than usual, thiswould indicate expectations of a better economic performance in the future. In addition, Friedmanand Kuttner (1998) analyse the CDS spread and the business cycle, demonstrating that credit defaulttends to increase in times of recessions. Their results indicate that variations in the business cycle

2 For example, bond spreads can be negatively affected by illiquidity issues (Chen, Lesmond, & Wei, 2007; Sarig & Warga,1989), different tax treatments, and other market microstructure factors including coupon and maturity effects.


tend to impact on the aggregate economy and investors’ behaviour, affecting the level of defaults.Minton (1997) on the other hand found that the yield curve had a positive relationship with the swapspread.

Stokes and Neuburger (1998) find that inflation is another factor that greatly affects credit defaultrisk through its impact on input and output prices. This implies that if a firm is facing higher costsas a result of inflation, the firm might find it hard to carry on with its daily business obligations andachieve the targeted profit. If inflation reaches extreme levels, this may lead the company to defaulton its obligations, thus increasing default risk. Duffie, Leandro, and Ke (2007) find that macroeconomicvariables and industrial production growth tend to be good indicators for predicting and understandingthe future fluctuations of credit risk. Similarly, Carling, Jacobson, Linde, and Roszbach (2007) show intheir research that both macroeconomic and various financial ratio variables are essential in predictingand determining default risk (Tang & Yan, 2010).

Collin-Dufresne et al. (2001) focused on analysing bond spread instead of the CDS spread and foundthat financial leverage, volatility and the yield spread have significant power in explaining bond spread.Similarly, Campbell and Taksler (2003) use company bond spreads and find that volatility explainsbond spreads. In a similar vein, Benkert (2004) used various volatility measures to analyse the CDSspread, and found the presence of a negative relationship. Another stream of research incorporatedcredit ratings as determinants of the CDS spread fluctuations. Studies such as Hull et al. (2004), Fabozziet al. (2007) and Karagozoglu and Jacobs (2010) investigate the relationship between credit ratings andCDS spreads and concluded that ratings were important in explaining credit risk. Also for sovereignCDS spread, similar analysis is used for pricing of credit risk (Alper, Forni, & Marc, 2012).

In the light of the above literature, we investigate both the macroeconomic and financial determi-nants of CDS spread in the UK banking sector by considering the house price, the yield spread, TED(difference between the three-month UK T-Bill and the three-month LIBOR rates) and the FTSE 100index.

3. Data

CDS: The data covers the period from January 2004 to April 2011. CDS data are obtained fromThomson Reuters Datastream, the world’s largest financial statistical database published by the CreditMarket Analysis (CMA) Group. The CDS banking sector data were first launched by the CMA groupin 2004. This dataset is ideal as it allows the analysis of the CDS spread both before and after thefinancial crisis, including the period when the credit market was booming. We use the monthly 5-Year Credit Default Swap (CDS) Index as a proxy for credit risk. The index includes the banking sectorin the UK and is denominated in the British Pound currency (£). We use log of CDS prices in ouranalysis. The data include indices with a five year maturity because they are the most liquid type ofCDS.

House price index: The UK house price index is obtained from Datastream Thompson Reuters andpublished by Nationwide Anglia Building Society under the reference ‘Nationwide House Price Index’,which is a monthly average and denominated in British Pounds. It is seasonally adjusted and rangesfrom January 2004 to April 2011. Again we use a log transformation for the empirical tests. Given thathouses are not similar, a simple average of all house prices in a specific sample would lead to misleadinginferences. For this reason, Nationwide adopts a statistical method, which uses the constantly varyingsample of mortgage approvals to produce a consistent index, capturing price fluctuations on a regularbasis.

Yield spread: The 3-month Treasury bill is obtained from Datastream Thompson-Reuters database.It is the middle market closing rate (i.e. the mean of bid and offer) as recorded by the Bank of Englandin the late afternoon. Also the 30-year Treasury bond yield has been obtained from Datastream. Itrepresents the UK Government 30-year benchmark bid yield, denominated in the UK Sterling currency.The yield spread variable is calculated as the yield of a 30-year UK Treasury bond minus 3-month UKTreasury bill. The data frequency for both variables is monthly, ranging from January 2004 to April2011.

TED spread: The liquidity spread in this paper is represented by TED. The acronym is formed fromT-Bill and ED – the ticker symbol for the Eurodollar futures contract. Instead of the Eurodollar, we use


the LIBOR rate. The series were obtained from Datastream. It is monthly, ranging from January 2004to April 2011. TED is defined as follows:

TED = 3 Month LIBOR − 3 Month UK Treasury Bill.

FTSE index: Given that this paper focuses on the UK financial market, FTSE-100 index was chosenas a benchmark for stock prices. The FTSE-100 is a share index of the 100 most highly capitalizedUK companies listed on the London Stock Exchange. The data were obtained from Thompson ReutersDatastream, at monthly frequency, ranging from January 2004 to April 2011.

3.1. Descriptive Statistics

3.1.1. CDS spreadFig. 1 shows that before July 2007 when the crisis began, the CDS spread was very low, averaging 13

basis points. The reason why the level of the CDS spread was not high prior to the financial crisis wasdue to the low perception of credit default risk in the financial system. However, in July 2007, the CDSspread started to increase dramatically, going up from 9 bps in July 2007 to 176 bps in March 2008 andpeaking at 200 bps in March 2009. The dramatic and sudden increase in the CDS spread was a reflection

1

2

3

4

5

6

2004 2005 2006 2007 2008 2009 2010 2011

LOGCDS

5.60

5.64

5.68

5.72

5.76

5.80

5.84

5.88

5.92

2004 2005 2006 2007 2008 2009 2010 2011

LOGHOUSEPRICE

-2

-1

0

1

2

3

4

5

2004 2005 2006 2007 2008 2009 2010 2011

YIELDCURVE30

8.2

8.3

8.4

8.5

8.6

8.7

8.8

8.9

2004 2005 2006 2007 2008 2009 2010 2011

LOG(FTSE100)

0.0

0.4

0.8

1.2

1.6

2.0

2004 2005 2006 2007 2008 2009 2010 2011

TED

Fig. 1. Graphical illustration of the non-stationarity of the CDS spread, House Price Index, Yield Spread and the FTSE 100 indexseries. Notes: The plots of these series clearly reflect the non-stationarity nature of the data involved which need to be paidattention before confirming a relationship between credit risk (CDS) and its determinants.


of the subprime mortgage crisis that affected the financial markets. In fact, the period between July2007 and March 2009 was marked by the bankruptcy of large investment banks, including AIG, LehmanBrothers, and Bear Sterns among other major banks that threatened the stability of the overall financialsystem.

In February 2011, the CDS spread for the UK banking sector averaged 156 bps. Thus, although thelevel of credit risk started to gradually decrease after March 2009 as a result of the slow economicrecovery, its level was still high relative to the period preceding the financial crisis. This reflects thevulnerability of the financial system to external shocks.

3.1.2. House price indexFrom Fig. 2, it can be clearly observed that during the period before October 2007, the UK house price

index gradually increased from 273 bps in January 2004 to 371 bps in October 2007. That represents analmost 100 bps increase over a 3-year period. However, with the beginning of the subprime mortgagecrisis, the house price index started to fall drastically, reaching 300 bps in February 2009. Although,the house price index started to progressively increase again, reaching 337 bps in June 2010, it fellagain to 330 bps in March 2011. Thus, despite the marginal increase in house price index after thecrisis, the level of the index never came back to its peak level of 371 bps that was recorded just beforethe beginning of the financial crisis.

Before the financial crisis, consumers borrowed heavily due to the low interest rates that in turnwere partly a result of foreign inflows. This created a massive credit expansion and easy borrowingto low income consumers, leading to a housing boom in the economy. In fact, most of the borrowershad very low credit ratings and were still able to obtain mortgages. Furthermore, given the increasedsecuritization activities in the financial system, banks were not very concerned about the qualityof borrowers. Due to the sophisticated financial engineering practices and securitization activities,banks which were granting mortgages were repackaging these mortgage obligations into syntheticstructured products such as Mortgage Backed Securities (MBS), Retail Mortgage Backed Securities(RMBS) or Collateralized Debt Obligations (CDOs) among other more complex structured products.The repackaging process of mortgages and other instruments were achieved with the help of creditrating agencies and Special Purpose Vehicles (SPV) which effectively trenched the MBS and RMBs,reselling them to other parties who were better equipped to handle risk. This has allowed a betterrisk diversification while resulting in low lending standards. Given the easy borrowing in the financial

Fig. 2. The transmission channel leading to the credit crisis. Notes: This figure demonstrates how the housing bubble thatoriginated in the US evolved into subprime mortgage crisis, which later turned into the global financial crisis affecting theworld economy. It can be observed that the subprime mortgage crisis suggests a spill over effect, creating volatility in othermarkets, thus affecting the CDS spread.


market, the demand for real estate purchases dramatically increased, pushing up the house prices. Asa result, borrowings increased and savings decreased. This has in turn led to a housing boom, followedby a surplus of unsold houses driving the real estate prices down. This period corresponds to thebeginning of the subprime mortgage crisis.

At the beginning of 2007, sub-prime mortgage borrowers started heavily defaulting on their mort-gage obligations, and given that house prices were already decreasing, banks were no longer able torecover their loans by reselling the properties. It should be noted that financial institutions were notthe only parties who suffered from the sub-prime mortgage crisis. Primary borrowers who defaultedwere also in distress as a result of losing their primary residence. In addition, there was another classof borrowers who were taking mortgage loans to later resell it into the secondary market when realestate prices were higher, with the aim to achieve a profit (also referred to as remortgaging activities).Following the crisis, these borrowers were now holding negative equity. This has caused a furtherincrease in the number of defaults and houses for sale. Thus, all securitized products were drasticallyfalling in value, with most of the mezzanine tranches defaulting first. In addition, the CDS spreadon Mortgage Backed Securities (MBS) was extremely high, leading to an overall increase in the CDSspread. The high CDS spread was a reflection of the rising default risk in the financial system. Thetransmission channel leading to the credit crisis is explained in a Flow Chart in Fig. 2.

3.1.3. Yield spreadIn Fig. 1, the yield spread is split into two different time periods: the period preceding the financial

crisis (January 2004 until August 2007), and the period following the financial crisis (August 2007 toApril 2011). In the first period, before the crisis emerged, the yield curve was downward sloping, andprogressively decreasing. However, from the beginning of the financial crisis, the direction of the yieldcurve has changed, becoming upward sloping. The yield spread can increase either as a result of higheryields being offered on long term bonds (yield on the 30 year Treasury bond) or due to the decreasein short term yields (yield on 3-month Treasury bills).

The yield on the long term bond can increase following higher perception of credit risk in thegovernment sector resulting from enormous fiscal deficit. The higher Sovereign credit risk can gettransmitted to the private sector, given the government’s borrowing requirement from financial mar-kets. In addition, inflation rate is another factor that could influence the shape of the yield curve. Ahigher public sector borrowing requirement and inflation risk would lead to an increase in the yieldof the long term bond. Furthermore, it can be observed from Fig. 1 that the yield curve steepened atthe beginning of the credit crisis, while from March 2009 onwards the yield curve started to flatten,reflecting the beginning of an economic recovery.

3.1.4. Stock pricesDuring the period preceding the financial crisis, the FTSE 100 index was gradually increasing,

reflecting a constantly improving economic performance of the UK 100 most capitalized companies(see Fig. 1). After August 2007, with the beginning of the financial crisis, the FTSE-100 index started todrastically fall, reflecting deteriorating market conditions. Given that most of the financial institutionsare interlinked, defaults in one financial institution may spill over to other banks, companies and sec-tors. This was exactly what happened in the recent financial crisis. With the beginning of the sub-primemortgage crisis, major investment banks collapsed, and investors lost confidence in the financial sys-tem and started heavily withdrawing all their funds. This has caused bank runs (for example, the caseof Northern Rock) and the drying up of liquidity in the financial sector. Many banks went bankruptas they faced liquidity crisis and were unable to repay their debt obligations. The latter turned intoglobal financial crisis affecting not only the housing and financial sectors, but also causing distress inother sectors as reflected in rising unemployment rates.

3.1.5. TED spreadFig. 1 indicates the fluctuation of the liquidity spread between January 2004 and the April 2011, in

the UK money market. Following the financial crisis, liquidity in the financial markets started to declineconsiderably. The dramatic collapse in liquidity was recorded around September 2008, when capital


Table 1Explanatory variables and expected signs on the coefficients.

Variables Type of the variable Description of the variables Predicted sign

ln(CDS) Dependent Natural logarithm of the CDSln(HP) Explanatory Natural logarithm of the House Price Index (−)Yield Spread Explanatory 30 year Treasury bond − 3 month Treasury bill (−)TED Explanatory 3 month LIBOR − 3 month Treasury bill (+)ln(FTSE100) Explanatory Natural logarithm of the FTSE100 Stock Price Index. (+)

Notes: This table includes the expected signs of the coefficients of the explanatory variables. The first column indicates thetype of the explanatory variable, followed by a description of the variables, while the last column indicates the predicted signbetween the dependent variable and the independent variables.

markets froze and investors started withdrawing their funds from financial institutions, causing bankruns. Table 1 includes the expected signs of the above four variables.

In Table 2, column 1 indicates the variations of the CDS spread between January 2004 and April2011. The CDS spread ranges from a minimum of 1.584 bps to a maximum of 5.396 bps, with the overallmean of 3.511 reflecting high default risk. Column 2 describes the variations of the house price indexbetween January 2004 and April 2011, ranging from 5.613 bps to a maximum of 5.917 bps, with anoverall mean of 5.785. Columns 3 and 4 describe the variations of the yield spread and the FTSE-100index between January 2004 and April 2011, while Column 5 describes the variations of the TED overthe same time period.

4. Methods and results

In this section, we discuss the three methods mentioned earlier. Our first objective is to establishif a long-run relationship exists between CDS spread, house price index, yield spread, TED and stockprices. Each of these variables represents different markets, namely credit, housing, money and finan-cial markets. In order to identify a long-run relation, it is essential to test each of the variables forstationarity. There are various tests that exist to identify the presence of a unit root, including thebenchmark Augmented Dickey-Fuller (ADF) test. Fig. 2 clearly illustrates that the CDS, House PriceIndex, Yield Spread and the FTSE 100 index are all non-stationary, i.e. all of the variables do not have aconstant mean, a constant variance and constant auto-covariances for each given lag. The stationaritytest confirms whether unit root exists in the series under consideration. The Johansen method and theDynamic Ordinary Least Squares (DOLS) approach are used to investigate which of our variables hasbetter predictive power to explain the behaviour of the CDS spread. Both methods help uncover if along run relationship exists between the variables, whereas in order to analyse the short run effects,a vector error correction model (VECM) is derived and a structural VAR model (SVAR) is employed toexamine the impact of unanticipated shocks.

4.1. Unit root tests

4.1.1. Augmented Dickey-Fuller (ADF)The non-stationarity phenomenon is very common in time series data. In order to avoid the problem

of spurious regression that non-stationarity brings, it is essential to correct this problem through

Table 2Descriptive statistics of the CDS spread and its determinants.

Log(CDS) Log(House Price Index) Yield Spread Log(FTSE100) TED

Mean 3.510696 5.784637 1.000155 8.576807 0.396163Median 3.126910 5.786867 0.082315 8.588579 0.184820Std. Dev. 1.371441 0.066932 2.041531 0.142085 0.413779Minimum value 1.584120 5.612982 −1.812130 8.214414 0.100070Maximum value 5.396448 5.917414 4.108360 8.813519 1.946920


Table 3.1ADF and DF-GLS Unit Root Tests Results.

Variables ADF(�) ADF(�) DF-GLS Inference on integration

ln(CDS) −0.586490 −1.998058 −0.398609 I(1)ln(House Price Index) −2.237199 −2.314131 −0.909877 I(1)Yield Spread −0.870173 −1.874629 −0.886743 I(1)TED −2.135415 −2.118459 −2.011065 I(1)ln(FTSE 100) −1.934734 −1.922138 −1.274281 I(1)

Notes: (a) ADF(�) and ADF(�) are tests of the unit root null hypothesis where the test regression contains a constant and nodeterministic components, and a constant and a trend, respectively. (b) The asymptotic 1 per cent critical values for the ADFtest are: −3.51 (with constant) and −4.07 (with a constant and a linear trend). The critical value at 1 percent level for theElliott–Rothenberg–Stock DF-GLS, with a constant is: −2.60. All series are non-stationary.

differencing or detrending the series. The Augmented Dickey-Fuller (ADF) unit root test is given asfollows:

CDSt = �CDSt + εt (1)

where � is the parameter to be estimated and εt represents white noise. If |�| ≥ 1, then CDS is a non-stationary series and the variance increases with time and approaches infinity. If on the other hand|�| < 1, then the time series is trend stationary. Accordingly, the hypothesis of stationarity can be testedby investigating whether the absolute value of � is strictly less than 1. The null and the alternativehypotheses under the ADF test are the following:

H0. � = 1

H1. � < 1

The ADF test builds a parametric correction for higher correlation by assuming that the series followan auto-regressive AR(p) process and adding p lagged difference terms of the CDS variable:

CDSt = + CDSt−1 + �

n∑

i=1

CDSt−i + εt (2)

This augmented specification represented in Eq. (2) is utilized in order to test the null and thealternative hypothesis of ADF test based on the t ratio. The results from the ADF test and theElliott–Rothenberg–Stock DF-GLS test are summarized in Table 3.1. The ADF Test was conducted withalternative assumptions (with a constant and no deterministic component, and with a constant anda trend). This is presented in the first two columns of Table 3.1. The results indicate that at 1% level,all the variables [ln(CDS), ln(House Price Index), Yield Spread, TED and ln(FTSE 100)] contain a unitroot and are therefore non-stationary. When the time series is tested using Elliott–Rothenberg–StockDF-GLS test – as a more powerful test relative to the ADF test, the results proved to be the same, con-firming the existence of a unit root (or non-stationarity) in the variables. The last column of Table 3.1shows the order of integration of the variables. The presence of unit root makes all the variables asintegrated of order 1.

Table 3.2 indicates that after considering ln(CDS), Yield Spread, TED and ln(FTSE 100) in first differ-ences, the series become stationary (integrated of order zero) being statistically significant at 1% level,implying that we can safely reject the null hypothesis of a unit root. The only variable that remainsnon-stationary is ln(House Price Index). As the non-stationarity of this variable can be driven by possi-ble structural break that occurred during the financial crisis, we conduct Zivot and Andrews unit roottest that considers structural breaks.

4.1.2. Zivot and Andrews unit root testThe Zivot and Andrews unit root test is performed by allowing a break at an unknown point in

either the intercept, the linear trend or in both. The test is based upon the recursive estimation ofa test regression. The test statistic is interpreted as the minimum t-statistic of the coefficient of thelagged endogenous variable. The outcome of the Zivot–Andrews unit root test is presented in Table 4.


Table 3.2ADF and DF-GLS Unit Root Tests Results when the data is taken in first differences.

Variables ADF(�) ADF(�) DF-GLS Inference on integration

�ln(CDS) −7.275264** −7.251929** −6.158968** I(0)�ln(House Price Index) −2.710530 −2.544836 −1.134453 I(1)�Yield Spread −5.340040** −5.348152** −4.386833** I(0)�TED −9.310985** −9.270561** −9.365047** I(0)�ln(FTSE 100) −8.729111** −8.682519** −8.158638** I(0)

Notes: (a) ADF(�) and ADF(�) are tests of the unit root null hypothesis where the test regression contains a constant and nodeterministic components, and a constant and a trend, respectively. (b) The asymptotic 1 per cent critical values for the ADFtest are: −3.51 (with constant) and −4.07 (with a constant and a linear trend). The critical value at 1 percent level for theElliott–Rothenberg–Stock DF-GLS, with a constant is: −2.59. All series in first differences are stationary, except house priceindex.

** t-Values being significant at 1% level, implying no unit root in the series.

From Table 4, it can be clearly observed that the variables ln(CDS), ln(House Price Index), and theYield Spread are all stationary, contrary to the earlier result for house price. Using this test, it is nowpossible to reject the null hypothesis of a unit root at 5% level. The only variable that contains a unitroot even after allowing for a structural break is ln(FTSE 100). Conducting the Zivot–Andrews unitroot for ln(FTSE 100) in first differences, the series became stationary being significant at 1% level asshown in Table 4. Given the presence of unit roots, in the next section we will test whether there existsa cointegrating relationship between ln(CDS), ln(House Price Index), Yield Spread and ln(FTSE100),along with considering TED.

4.2. Cointegration tests

In this section we will be using the Johansen’s cointegration test (Johansen, 1991) in order toidentify any cointegrating relationship. Before conducting this test, we first need to establish theoptimal number of lags to be used in our model. For this purpose, we use the VAR Lag Order SelectionCriteria. The results are presented in Table 5.

From Table 5, the Sequential Modified LR test Statistic, Final prediction error, Akaike and Hannan-Quinn information criteria all indicate that the optimal lag length for our VAR model is 3 lags. Weestimate a VAR model with 3 lags as found to be optimal by the VAR Lag Selection Criteria (as shownin Table 5). The Johansen test provides evidence that there exists one cointegrating relation using alinear model (with intercept and trend). In Table 6, the Trace Test statistics indicates the existenceof one cointegrating equation at 5% significance level. Given that the null r = 0 is rejected at a 5%significance level implies that there is one meaningful long-run relation between ln(CDS), ln(HousePrice Index), Yield Spread and ln(FTSE 100 index).

The normalised cointegrating equation can be written as follows:

ln CDSt = 229 − 37.98 ln HPt[3.989]

− 1.20 YSt[3.605]

− 1.38 ln FTSEt[0.465]

+ 0.15T[5.315]

(3)

Table 4Zivot–Andrews unit root test.

Variables Test statistics Structural break point

Ln(CDS) −7.50618** August 2008Ln(House Price Index) −5.08262* March 2008Yield Spread −5.69783** October 2008TED −3.74271 February 2008�TED −7.56039** January 2009Ln(FTSE 100) −3.27453 September 2009�ln(FTSE 100) −5.99628** April 2009

Notes: Critical values at 1% level: −5.57 and at 5% level: −5.08; While CDS, House price, and Yield Spread are stationary in levelsallowing for structural breaks in the intercept and trend, TED and FTSE turn stationary only in first differences.

* Significance at 5% level.** Significance at 1% level.


Table 5VAR Lag Order Selection.

Lag LR FPE AIC HQ

0 NA 4.21e−05 4.21e−05 1.3245391 819.1956 6.26e−10 6.26e−10 −9.5962742 48.53018 4.64e−10 4.64e−10 −9.7034533 39.48386* 3.81e−10* 3.81e−10* −9.713028*

4 19.63113 4.23e−10 4.23e−10 −9.4286075 19.30998 4.67e−10 4.67e−10 −9.1625456 14.97610 5.54e−10 5.54e−10 −8.8390417 15.83245 6.40e−10 6.40e−10 −8.5587508 16.84717 7.17e−10 7.17e−10 −8.3333929 19.20207 7.48e−10 7.48e−10 −8.208600

10 19.42243 7.61e−10 7.61e−10 −8.146374

Notes: LR: sequential modified LR test statistic (each test at 5% level), FPE: final prediction error, AIC: Akaike informationcriterion, HQ: Hannan-Quinn information criterion. The optimal lag length is found to be 3 months across different lag selectioncriteria.

* Lag order selected by the criterion.

where CDS: the Credit Default Swap Spread, HP: House Price Index, YS: Yield Spread, FTSE: FTSE-100Index, T: Time trend. It should be noted that the CDS spread, the house price index, and the FTSE-100index are all taken in natural logarithms.

In Eq. (3), figures in brackets represent the t-values. Almost all the key variables [ln(House PriceIndex), Yield Spread and the Time Trend] included in the normalised cointegrating relation are sta-tistically significant and the signs of the coefficients turned out to be as expected apriori. The onlyvariable that is statistically insignificant is ln(FTSE-100). The long-run relationship can be explainedas follows. First, an increase (decrease) in the house price index by 1 percent is associated with adecrease (increase) in the CDS spread by 37.98 percent in the sample period. Furthermore, an increase(decrease) in the yield spread by 1 percent is associated with a 1.2 per cent decrease (increase) in theCDS spread. Finally, a significant time trend suggests an increase in the CDS spread over the sampleperiod. Given that the t-statistic for the FTSE-100 index is not significant, our aim is to only focus onthe impact of the house price index and the yield spread on the CDS spread.

From Eq. (3), it can be clearly observed that there is a negative relationship between the CDS spreadand the house price index as expected. CDS contracts work as insurance contracts. The CDS spread isthe price that the protection buyer pays to the protection seller in order to benefit from a guaranteethat in case of a default, the holder of the CDS will be covered against a default. Thus, as risk increases,the premium on the CDS contract also rises in order to reflect the higher risk. Banks and other financialinstitutions that were issuing CDS contracts backed on mortgages priced their CDS contracts based onhouse prices. In fact, when house prices were high, even if a lender defaulted on his/her payment, itwas relatively easy for the bank to recover the initial cost of the property given the continuous upwardtrend in the house prices. Thus, the CDS spread was narrow reflecting a low default risk. However bythe end of 2006, with the beginning of the housing bubble when house prices started to collapse, thenumber of defaults suddenly went up. This was reflected in a much larger CDS spreads. In Eq. (3), the

Table 6Unrestricted cointegration rank test (trace).

Hypothesized No. Of CE(s) Eigenvalue Trace statistic 0.05 Critical value Prob.**

None* 0.343598 71.73126 63.87610 0.0094At most 1 0.197909 37.21076 42.91525 0.1655At most 2 0.141312 19.12702 25.87211 0.2733At most 3 0.077721 6.634399 12.51798 0.3842

Notes: Trace test indicates 1 cointegrating eqn(s) at the 0.05 level. The test confirms existence of one long-run relation.* Rejection of the hypothesis at the 0.05 level.

** MacKinnon–Haug–Michelis (1999) p-values.


Table 7Determinants of the CDS spread using the Johansen’s approach.

Model 1 Model 2 Model 3 Model 4 Model 5

Log(House Price Index) −16.16920 −43.19819 −37.97884 −12.81836 −12.31915[5.99389]** [4.94263]** [3.98866]** [5.42143]** [2.31267]**

Yield Spread −1.201765 −1.204231[3.09413]** [3.60466]**

TED 0.370977 0.935764[0.085679] [1.84912]

Log(FTSE100) −1.379733 0.245253[0.46549] [0.09714]

Trend 0.055161 0.157168 0.151729 0.050003 0.048939[8.23317]** [4.69568]** [5.31507]** [8.58935]** [9.07686]**

Constant 94.68693 247.7430 229.6091 75.36537 70.18846

Notes: The long-run relationship is estimated between credit default swap (CDS), housing, money, bond and stock marketsduring January 2004 to April 2011, using Johansen’s cointegration approach. It is found that the CDS market is more sensitiveto the housing market or bond market than the stock market or money market.

** Significance at 1% level.

long run equation confirms this negative relationship between the CDS spread and the house priceindex.

Another observation from Eq. (3) is the negative relationship between the yield spread and theCDS spread. As it was previously mentioned in the literature review, the steepness of the yield curveindicates future economic activity. Thus, the steeper the yield curve, the higher is the expected futureinterest rate. In fact, in times of a recession or when firms start defaulting, interest rates tend to be verylow, but the CDS spread tends to increase given the higher credit risk. The higher credit risk directlyimpacts the CDS spread as it is now more expensive for investors to benefit from a protection againstdefault. This is reflected in a rising CDS spread. Therefore, there should be a negative relationshipbetween the yield curve and the CDS spread. This negative relationship between the slope of theterm structure and the swap spreads was empirically supported by Fehle (2003), In et al. (2003),Fama (1984) and Estrella and Hardouvelis (1991). Another line of thinking that supports this negativerelationship between swap spreads and the slope of the risk free term structure was found by Friedmanand Kuttner (1992). The authors explained that the slope of the risk-free term structure was foundto be pro-cyclical while the credit spreads could be counter-cyclical. For this reason there shouldbe a negative relationship between swap spreads and the slope of the risk-free term structure. AlsoLongstaff and Schwartz (1995) found that as interest rates increase, there could be lower probabilityof default, thus narrowing of credit spread. The results obtained from the Johansen’s approach aresummarized in Table 7.

When we estimate the CDS relation by using the TED variable instead of the yield spread, ourresults remain robust, still showing a negative relationship between the house price index and theCDS spread. However, both TED spread and the FTSE-100 index turn out to be insignificant.

Given the long-run relation derived in Eq. (3), it is now important to examine the short-run dynam-ics of the CDS spread. This can be achieved by employing the Vector Error Correction Model (VECM).While deriving the short-run equation, the residual or error correction (EC) term obtained from thelong-term relationship is incorporated into the short run dynamics.

� ln CDSt = −0.0213 ECt−1[−0.76948]

+ 0.43� ln CDSt−1[3.36706]

− 0.18� ln CDSt−2[−1.29722]

+ 0.39� ln CDSt−3[2.76105]

+ 2.56� ln HPt−1[0.67165]

− 8.20� ln HPt−2[−2.36137]

+ 2.26� ln HPt−3[0.59556]

+ 0.02� YSt−1[0.16184]

− 0.13� YSt−2[−1.13510]

+ 0.17� YSt−3[1.61058]

+ 1.13� ln FTSEt−1[1.97484]

+ 0.50� ln FTSEt−2[0.83069]

+ 1.27� ln FTSEt−3[2.08716]

+ 0.003[0.125]

(4)


Eq. (4) shows that the t-values (in brackets) are not always significant for each lagged value of theincluded variables in the short-run. The EC term being insignificant indicates that there is no shortrun adjustment between the CDS spread and House price Index (HP), the Yield Spread (YS) and theFTSE. In fact, the only variables that are statistically significant from equation (4) are: �ln CDSt−1,�ln CDSt−3,�ln HPt−2 and �ln FTSEt−3. The fact that the adjustment coefficient in the short run is notsignificant suggests that any deviation from the long-run is not getting corrected in the next period.Thus, equilibrium can be only achieved in the long run as it may take a considerable amount of timefor the CDS spread to adjust and stabilize.

4.2.1. Robustness check: Stock Watson’s dynamic OLS modelStock and Watson (1993) introduced the dynamic OLS (DOLS) method that allows for variables

being integrated of different orders (i.e. a higher order of integration) and deals with the issue ofsimultaneity that may arise amongst the regressors. The DOLS approach is considered to be a morecomprehensive method than the OLS as it has the ability to deal with small sample and dynamic sourcesof bias. In the Johansen approach, the parameter estimates in one equation can be influenced by mis-specification that may exist in other equations, leading to wrong inferences. The Stock Watson DOLSmethod of estimation corrects for regressor endogeneity by including leads and lags of first differencesof the regressors, and for serially correlated errors by a GLS procedure which has similar asymptoticoptimality properties as the Johansen method. The DOLS method augments the cointegrating regres-sion with lags and leads of �Xt so that the resulting error term from the cointegrating equation isorthogonal to the entire history of the stochastic regressor innovations. This can be demonstrated inthe following equation:

CDSt = X ′t + D′

1t�1 +r∑

j=−q

�X ′t+jı + u1t (5)

where X ′t+j

is the vector of independent variables and D1t represents a drift. Adding q lags and r leadsof the differenced regressors soaks up all of the long-run correlations between u1t and u2t.

Aside from Johansen’s cointegration method in the first part of our analysis, the dynamic OLSmethod is now implemented as a robustness check of earlier results. Stock Watson’s DOLS results arepresented in Table 8. The DOLS results further confirm a negative and significant relationship betweenthe CDS spread and the house price index as in Johansen’s method. Thus, as house prices go down,the CDS spread goes up, reflecting higher credit risk as previously discussed. Once the liquidity spread(TED) and the FTSE 100 index are added to the model, the yield spread turns out to be insignificant whileboth TED and the FTSE are positive and significant (see Model 4 in Table 8). The positive relationshipbetween the TED spread and the CDS spread follows the economic logic that when liquidity is high

Table 8Determinants of the CDS spread using the DOLS method.

Model 1 Model 2 Model 3 Model 4

Log (House Price Index) −14.02065 −42.64619 −23.39485 −25.59222[9.176418]** [6.070436]** [2.040168]** [2.987228]**

Yield Spread −1.162395 −0.402860 0.025576[4.160006]** [0.771487] [0.059925]

TED 1.172692 2.158376[1.079357] [2.310708]**

Log (FTSE100) 6.649284[3.060539]**

Trend 0.053776 0.152989 0.084244 0.051900[15.35852]** [6.279349]** [1.916873] [1.421926]

Constant 82.53469 245.3135 135.6079 91.70514[9.362419]** [6.139234]** [2.075008]* [1.659310]

Notes: The lead-lag relationships are estimated between credit default swap (CDS), housing, money, bond and stock marketsduring January 2004 to April 2011, using Dynamic OLS method due to Stock–Watson. It is found that the CDS market is moresensitive to the housing market than the stock market.


in the financial market, banks tend to lend more to subprime borrowers, thus increasing credit riskand the overall CDS spread. Furthermore, the CDS spread and FTSE index are positively related. Asthe stock index increases, both banks’ capital and its borrowing capacity rises, resulting in a highercredit risk and wider spread. Having analysed the CDS determinants in the long run, using both theJohansen’s and the Stock–Watson’s Dynamic OLS method, in the next sub-section we will look at theshort-run impact of unexpected shocks using a structural VAR model.

4.3. The structural VAR model

Having established the variables that explain the CDS spread in the long run, it is now essential toanalyse the significance of unanticipated movements in these determinants in the short run. The short

Fig. 3. The determinants of the CDS spread using structural VAR model. Notes: This figure indicates the outcome of the impulseresponse based shock analysis via a vector auto-regression (VAR) model. The SVAR model is formulated as follows: TED spread(shock 1), Yield Spread (shock 2), LOG (House Price Index) (shock 3), LOG (FTSE 100) (shock 4) and LOG (CDS spread) (shock 5). Thegraphs that are plotted horizontally demonstrate the response of each of the variables to different shocks. The only significantunexpected structural shock in the short run is the liquidity premium (TED spread), which is demonstrated in the first column,fifth row. Thus, the CDS spread responds positively to a shock in the liquidity premium. The rest of the shocks are insignificantin influencing CDS spread as the CDS spread can adjust only in the long run.


run analysis can be achieved by undertaking an impulse response based shock analysis via a vectorauto-regression (VAR) framework.

The main purpose of structural VAR (SVAR) estimation is to obtain orthogonalisation of the errorterms for impulse response analysis. A restricted structural VAR model will define the relationshipbetween the VAR residuals, that is, the unexpected shocks, and the structural shocks, which areexogenous and uncorrelated with each other. This can be defined as follows:

Aet = But, with E[utu′t] = I (6)

where ut is the vector of structural shocks, while A and B are the matrices that define the linearrelationship between the structural shocks and the VAR residuals (et).

The optimal lag selection was found to be 3 for the VAR model (see Table 6). The structural VARmodel is formulated with the following ordering of the endogenous variables: TED spread, YieldSpread, LOG (House Price Index), LOG (FTSE 100) and LOG (CDS spread). The reason behind this specificordering stems from the theoretical ordering of the variables that should run from the more exoge-nous to the less exogenous variables. In fact, both the liquidity spread and the yield spread are likelyto be determined by the monetary policy and the state of the economy prevailing at a specific point intime. However, the house price index, FTSE 100 index and the CDS spread are less exogenous than theTED spread and the yield spread. Thus, they are placed later in the ordering. The CDS spread is likelyto be affected by all of the previously included variables, staying last in the ordering.

Fig. 3 shows the responses from the structural VAR model. It appears that the only unexpectedstructural shock that is significant in the short run is the liquidity premium (TED spread). This can beobserved from the first column, fifth row of Fig. 3. It indicates that the CDS spread responds positivelyto a shock in the liquidity premium in the short-run, in line with the long-run DOLS result. All theremaining variables that were significant under both the Johansen’s and the Dynamic Stock Watson’sOLS methods do not seem to be able to explain the CDS spread movements in the short run. This canbe explained by the time it takes for the CDS spread to adjust, which is only achievable in the long run.

5. Conclusion

This paper analyses the factors that determine CDS spreads in the UK banking sector. In partic-ular we analyse the impact of the house price index, the liquidity premium (TED), the yield spreadand the FTSE 100 index. Each of these variables represents four different sectors, the credit market,the housing market, the money market and the stock market, respectively. Two long-run methods(Johansen’s Co-integration and Stock & Watson’s dynamic OLS) and two short-run methods (VECMand SVAR for anticipated and unanticipated impacts, respectively) are used in the analysis. We findstrong evidence that the house price dynamics are a key-driving factor behind the recent collapse ofcorporate CDS market that captures the credit risk of the banking sector. Specifically, there exists anegative relationship between the CDS spread and the house price index in the UK after controlling forother determinants of credit risk. In addition, a negative relationship exists between the CDS spreadand the yield spread that reflects investors’ risk aversion following the financial crisis. In addition, asliquidity goes up, banks tend to lend more to subprime borrowers thus increasing credit risk and theoverall CDS spread. As the stock index increases, both banks’ capital and its borrowing capacity rise,resulting in a lower credit risk.

In order to analyse the short run impact of unexpected structural shocks on CDS spread, we usea structural VAR approach, in which the only unexpected shock that turns significant is the liquidityspread (TED), while in the VECM framework that helps uncover anticipated effects, both house priceand stock price are significant. The credit market and the housing market are strongly related both inthe short- and long-run. Thus, financial distress in the housing market is highly likely to get transmittedto the credit market and cause a contagion to other markets. In order for the CDS spread to adjust andreach an equilibrium level, a longer time period is required. Thus, possible equilibrium between thecredit market, the housing market, the money and the stock markets can only be achieved in the longrun.


Acknowledgements

The authors gratefully acknowledge the constructive comments made by the editor, Karan Bhanot,on earlier versions of this article. Thanks are also due to Andros Gregoriou and Ricardo Sousa forcomments at the initial stage of this work. An early version of this paper was presented at the 2ndInternational Conference of the Financial Engineering and Banking Society (FEBS) on 7–8 June, 2012at the ESCP Europe London campus; the 19th Annual Global Finance Conference at Chicago, IL, USA,May 23–25, 2012; and the 2nd International Symposium in Computational Economics and Finance(ISCEF), March, 15–17, 2012, Tunisia.

References

Alexander, C., & Kaeck, A. (2008). Regime dependent determinants of credit default swap spreads. Journal of Banking & Finance,32(6), 1008–1021.

Alper, C. E., Forni, L., & Marc, G. (2012). Pricing of sovereign credit risk: Evidence from advanced economies during the financial crisis.IMF Working Paper, IMF.

Anderson, R., & Sundaresan, S. (2000). A comparative study of structural models of corporate bond yields: An exploratoryinvestigation. Journal of Banking and Finance, 24, 255–269.

Benkert, C. (2004). Explaining credit default swap premia. Journal of Futures Markets, 24(1), 71–92.Bevan, A., & Garzarelli, F. (2000). Corporate bond spreads and the business cycle. Journal of Fixed Income, 9, 8–18.Bhanot, K. (2001). Dynamics of corporate spreads – A non parametric analysis. Journal of Fixed Income, 11(2), 28–35.Blanco, R., Brennan, S., & Ian, W. (2005). An empirical analysis of the dynamic relationship between investment-grade bonds

and credit default swaps. Journal of Finance, 60(5), 2255–2281.Campbell, J. T., & Taksler, G. B. (2003). Equity volatility and corporate bond yields. Journal of Finance, 58, 2321–2349.Carling, K., Jacobson, T., Linde, J., & Roszbach, K. (2007). Corporate credit risk modelling and the macroeconomy. Journal of

Banking and Finance, 31(3), 845–868.Chen, L., Lesmond, D. A., & Wei, J. (2007). Corporate yield spreads and bond liquidity. Journal of Finance, 57, 119–149.Collin-Dufresne, P., Goldstein, R. S., & Martin, J. S. (2001). The determinants of credit spread changes. Journal of Finance, 56(6),

2177–2207.Duffie, G. R. (1998). The relation between treasury and corporate bond yield spreads. Journal of Finance, 53, 2225–2241.Duffie, D., Leandro, S., & Ke, W. (2007). Multi-period corporate failure prediction with stochastic covariates. Journal of Financial

Economics, 83(3), 635–665.Duffie, D., & Singleton, K. J. (1997). An econometric model of the term structure of interest-rate swap yields. Journal of Finance,

52, 1287–1321.Duffie, D., & Singleton, K. (1999). Modelling term structures of defaultable bonds. The Review of Financial Studies, 12(4), 687–720.Estrella, A., & Hardouvelis, G. (1991). The term structure as a predictor of real economic activity. The Journal of Finance, 45(2),

555–576.Fabozzi, F. J., Cheng, X., & Chen, R. R. (2007). Exploring the component of credit risk in credit default swaps. Finance Research

Letters, 4, 10–18.Fama, E. (1984). The information in the term structure. Journal of Financial Economics, 13(4), 509–528.Fehle, F. (2003). The components of interest rate swap spreads – Theory and international evidence. Journal of Futures Markets,

23(4), 347–387.Friedman, B. M., & Kuttner, K. N. (1992). Money, income, prices, and interest rates. American Economic Review, 82(3), 472–492.Friedman, B. M., & Kuttner, K. N. (1998). Indicator properties of the paper-bill spread: Lessons from recent experience. Review

of Economics and Statistics, 80, 34–44.Hammoudeh, S., & Sari, R. (2011). Financial CDS, stock market and interest rates: Which drives which? The North American

Journal of Economics and Finance, 22(3), 257–276.Houweling, P., & Vorst, T. (2005). Pricing default swaps: Empirical evidence. Journal of International Money and Finance, 24(8),

1200–1225.Hull, J., Predescu, C., & White, M. A. (2004). The relationship between credit default swap spread spreads, bond yields, and credit

rating announcements. Journal of Banking and Finance, 28, 2789–2811.International Swaps and Derivatives Association (ISDA). (2010). Market Survey; National Amounts Outstanding at year-end, all

surveyed contracts, 1987-present. London.In, F., Brown, R., & Fang, V. (2003). Modelling volatility and changes in the swap spread. International Review of Financial Analysis,

12(5), 545–561.Johansen, S. (1991). Estimating and hypothesis testing of cointegrating vectors autoregressive models. Econometrica, 59,

1551–1589.Karagozoglu, A., & Jacobs, M. (2010). Measuring credit risk: CDS spreads vs. credit ratings. Working Paper, Office of the Comptroller

of the Currency and the U.S. Department of Treasury.Klomp, J. (2010). Causes of banking crises revisited. The North American Journal of Economics and Finance, 21(1), 72–87.Lekkos, I., & Milas, C. (2001). Identifying the factors that affect interest-rate swap spreads: Some evidence from the United States

and the United Kingdom. Journal of Futures Markets, 21, 737–768.Longstaff, F. A., & Schwartz, E. S. (1995). A simple approach to valuing risky and floating rate debt. Journal of Finance, 50, 789–819.Minton, B. (1997). An empirical examination of basic valuation models for plain vanilla U.S. interest rate swaps. Journal of

Financial Economics, 44(2), 251–277.Naifar, N. (2010). What explains default risk premium during the financial crisis? Evidence from Japan. Journal of Economics and

Business, 63(5), 412–430.


Norden, L., & Weber, M. (2004). Informational efficiency of credit default swap and stock markets: The impact of credit ratingannouncements. Journal of Banking and Finance, 28, 2813–2843.

Ötker-Robe, I., & Podpiera, J. (2010). The fundamental determinants of credit default risk for European large complex financialinstitutions. IMF Working Paper, WP/10/153. IMF, Washington, DC.

Sarig, O., & Warga, A. D. (1989). Bond price data and bond market liquidity. Journal of Financial and Quantitative Analysis, 24(3),367–378.

Stock, J. H., & Watson, M. (1993). A simple estimator of cointegrating vectors in higher order integrated systems. Econometrica,61(4), 783–820.

Stokes, H. H., & Neuburger, H. M. (1998). New methods in financial modelling: Explorations and applications. Westport, CT: QuorumBooks.

Tang, D. Y., & Yan, H. (2010). Market conditions, default risk and credit spreads. Journal of Banking & Finance, 34, 743–753.Taylor, J. B. (2009). Getting off track: How government actions and intervention caused, prolonged and worsened the financial crisis.

Stanford, CA: Hoover Institution Press.Zhu, H. (2006). An empirical comparison of credit spreads between the bond market and the credit default swap market. Journal

of Financial Services Research, 29, 211–235.

determinants of bank credit default swap spreads: the role of the housing sector

Documents