Lead-Lag relationship between Libar-ois spread and CDS spread

(1)

1

UNIVERSITY OF AMSTERDAM

EXECUTIVE MASTER’S IN INTERNATIONAL FINANCE

MASTER THESIS 2018

LEAD-LAG RELATIONSHIP BETWEEN LIBOR-OIS

SPREAD AND CDS SPREAD

POORNIMA ANANTH

STUDENT NUMBER : 11933003

(2)

2

ABSTRACT

In this thesis, I use a Vector Auto Regression to capture the lead-lag relationship of Euro LIBOR-OIS spread and CDS spread of European banks, both of which are driven significantly by related factors and tend to move in the same direction when there is uncertainty prevalent in the whole market. I find that information flows from CDS to LIBOR, in times of market crisis/turbulence; but otherwise, they tend to move independently of each other. This is largely because of idiosyncratic factors moving the CDS spreads. The problem with using CDS spreads is its liquidity, and hence the accuracy of this study would be dependent on how liquid CDS spreads are in our sample. The issue could be mitigated by making use of liquidity-adjusted CDS spreads – a source of improvement to this research. Keywords : LIBOR-OIS spread vs CDS spread, European banks, Eurozone crisis, lead-lag relation, Vector Auto Regression

(3)

3

The periods of crisis, both the Subprime and Eurozone, have displayed interesting information about different market variables. The run-up to the crisis was marked with acute liquidity concerns and the counterparty risk was at an all-time high. Borrowing rates shot up to astronomical levels. Every firm was vulnerable to more risks than before. The usage of OTC derivatives had left the market participants so intricately interconnected that the failure of one would lead to a series of collapses along the way. This is typically one of the situations when central banks play an important role in the economy; they manage the liquidity in an economy through different mechanisms – setting of reserve requirement, open market operations and monitoring of the policy rates that come in different names in different countries. The subprime and the eurozone crisis chapters were so unique in that the Central banks (the U.S, the U.K and the ECB) decided to engage in Quantitative Easing - the Central bank essentially printing money to buy sovereign bonds from commercial banks.

This paper lays its focus on Eurozone crisis, and the banks therefore are European banks. During the Eurozone crisis, banks in the Euro region were heavily exposed to sovereign debt. Four of the five most distressed economies in the ratio had a national debt exceeding 100% of their GDP (with Greece recording a debt level of 163%). The crisis started in Greece, sending ripples across the Eurozone with other economies tumbling in quick succession. With the deterioration of credit worthiness of the sovereign governments, the banks’ assets started losing their value. This increased the risk exposure of the financial institutions, which was reflected in the high CDS spreads of European banks, thereby speaking volumes about the inherent credit and default risk. Widespread concerns over the sovereign financial stability

(5)

5 went to the extent of threatening the very existence of the single market and consequently the single currency - Euro. This fear reflected in LIBOR (although not as high as subprime levels) that increased to uncommon levels - prompted by a perceived increase in credit risk in the inter-bank market. The money market stress faced by the European banks was palpable as they were struggling to get funding, thereby arising liquidity concerns and pushing up LIBOR. Numerous stability packages including bailouts ensued in the years that followed, to bring the market back to normalcy. The austerity packages went to the extent of threatening a slowdown in the economy, and Quantitative Easing introduced to counter just that. The first tranche of QE began in 2015 and is expected to last until end of 2018. The ECB has also assured that the interest rates are to remain low through the summer of 2019.

One of the many interesting ex-post observations related to the crisis was the divergence between LIBOR and OIS rate, that had hitherto been in close alignment to one another. The widening gap between the two rates prompted scholars to study the validity and movement of the LIBOR rates and in turn, the LIBOR-OIS spread (hereafter called the LIBOIS spread). LIBOR ceased to be considered a proxy for risk -free rate, with the OIS rate taking its place, as is discussed in the paper by Hull and White (2013). Around the same time, the CDS spreads on entities sky-rocketed, as there was an increase in the perceived risk of a credit event getting triggered.

A similar behaviour/reaction from both LIBOR-OIS spread and CDS spread could imply they are driven by related (if not the same) market factors. This paper tries to explore the timing of the reactions by both the spreads. Essentially, the question intended to be addressed is: Which spread prices in new information faster than the other? Theoretically, we would assume CDS markets to change quicker than LIBOR. Not only is the LIBOR set just once a day, the process is in the hands of a few participating banks, who submit their estimates based on how much they would charge for lending funds. CDS markets are more broad-based, in the sense

(6)

6 that the entire market can trade on an entity. Therefore, intra-day events get reflected in CDS markets quicker than LIBOR-OIS spread, which should typically wait for the next day.

The second question in place is the time frame – although the focus is on the Sovereign debt crisis, the paper intends to study the behaviour of the two spreads during varying market dynamics. Theoretically, the CDS spreads should become more volatile when risk perception is high in the market. But, with the Quantitative Easing process under way, one would typically expect both the spreads to move in sync or probably the LIBOR to lead.

We use 5-year senior unsecured CDS data on 12 European banks belonging to 7 European countries and 3-month Euro LIBOIS spreads. A Vector Auto Regression model is performed on the average change in daily CDS spread(across all banks) versus daily change in LIBOIS spread . This method also necessitates estimation of the optimal lag length parameter, which is calculated by means of information criterion and LR test. The thesis studies the behaviour of both the spreads over 9 years starting late 2008 through end of 2016 – split into 7 stages and 3 phases : the events leading up to the crisis, crisis, and post-crisis phases. Once the regression is performed, regression estimates are tested for statistical significance (if significant, the lag(s) of the variable is/are identified).

The remainder of the paper is organized as follows. Section 2 describes theoretical framework underlying the study and discusses more about LIBOR, its setting process, the OIS swap rate calculation and Credit Default Swaps. Section 3 throws light on the growing literature that has dealt with similar studies, from one of which I derive motivation for this research. Section 4 presents the data used in this study and the details of the regression methodology being used. Section 5 reports the results of the regression and the accompanying analysis and how well the results hold with respect to our theoretical assumptions. Section 6

(7)

7 talks about a crucial limitation of this study, that could be a potential source of improvement going forward.

THEORETICAL FRAMEWORK

LIBOR as defined by the Intercontinental Exchange (ICE) Benchmark Administration, is a measure of unsecured funding for banks in the interbank market for a given period and in a given currency. LIBOR is set daily by the participating banks for five currencies and seven maturities. The top and bottom quartile of the rates submitted by the banks are ignored, and the average of the remaining rates forms the LIBOR for that day, maturity and currency. The process, which was formerly managed by British Bankers Association, is now being supervised by ICE. Hull and White (2013) define OIS as a swap in which a fixed rate is exchanged for a floating rate that is the geometric mean of the daily overnight rates. In case of Euros, the daily overnight rate used is Euro Overnight Index Average (EONIA).

OIS rate is calculated based on the EONIA in the Euro region.

The calculation of the payment on the floating side is designed to replicate the aggregate interest that would be earned from rolling over a sequence daily loans at the overnight rate. The n-month interest rate on the floating leg of the OIS swap is given by :

� �1 + 𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒_{360 �}𝑖𝑖 𝑡𝑡+𝑛𝑛

𝑖𝑖=𝑡𝑡

− 1

Where eoniai is set by the ECB as part of the official policy rate setting mechanism and uses the 30/360-day count convention. The payment in an OIS swap takes place at the end of the swap period, in case of swaps with a duration of 1 year or less. At maturity, the difference between floating and fixed leg interest rates are exchanged between the counterparties.

(8)

8 According to Jin Cui, Francis In, and Elizabeth Ann Maharaj (2016), the LIBOIS spread removes the effects of policy rate expectations, and solely reflects the credit/liquidity concerns in the inter-bank market.

As cited by Hull and White (2013), the spread usually hovers around 10bps during normal market conditions. However, in October 2008, the USD LIBOR-OIS spread rose to 364bps which was the highest ever recorded; and despite hitting normalcy a year later, it went back to 30bps and 50 bps amid concerns about Eurozone sovereign debt. The same is shown in Figure 1.

Figure 1. Evolution of LIBOIS spread over the years 2008-2016

The 1990-2000s witnessed a rising prominence of usage of OTC derivatives, securitized products, and structured finance. Credit derivatives(CDS), as with any other derivative instruments, are chiefly used for hedging, speculation and arbitrage. Minton, Stulz and Williamson (2008) find that roughly 6% of the 395 U.S banks used credit derivatives from 1999 to 2005, and the notional amount of credit derivatives used for hedging amounts to just 2% of the total outstanding contracts. This further suggests and is shown in the paper that a

0 40 80 120 160 200 01-08 01-09 01-10 01-11 01-12 01-13 01-14 01-15 01-16

(9)

9 position in credit derivatives is taken chiefly for dealer activities, rather than risk management.

CDS, simply put, is an insurance contract on an underlying security issued by a firm or a corporation (hereafter called Reference Entity). The CDS buyer makes regular payments in exchange for protection in the wake of a credit event. The regular payments are determined by the CDS spread, which is agreed upon when the contract is initiated. The CDS markets attracted worldwide attention since the CDSs and OTC derivatives at large were never required to publicly report until after the crisis. Little was known about the credit exposures of the market participants, and no one could, ex ante, accurately gauge the level of risk in the market. This was one of the reasons why AIG (a massive seller of CDS protection) was bailed out during 2007 crisis. The collapse of AIG would, otherwise, have triggered a series of bankruptcies in the market.

Figure 2. Notional principal of outstanding CDS contracts over 2004-2017

According to the periodic OTC derivatives statistics released by Bank for International Settlements(BIS) and as depicted in Figure 2, the notional amount of CDS markets rose to an all-time high by 2008, and decreased to $25tr by Dec 2012, to $16tr by Dec 2014. This means CDS markets are traded heavily during the period taken for this study – nevertheless, our

0 10 20 30 40 50 60 70

(10)

10 reference entities are solely European banks, who have an active CDS market. Normally, when there is a news development concerning a bank or financial markets at large, CDS spreads are expected to be quick to see its spreads increase accordingly. Otherwise, this would give rise to an arbitrage opportunity, and one could profit from trading on the underlying bond and selling protection or vice versa. The arbitrage activity, in the process,

fixes the mispricing. According to Oehmke and Zawadowski (2012), firms with a more

negative CDS-bond basis have more CDS outstanding, that hint at arbitrage opportunity associated with CDSs. Volz and Wedow (2011) observe evidence of general market discipline in CDS market, though they find a significant negative relationship between CDS spreads and size of a bank. This could be attributed to the increased probability of a bailout for a large bank. This can distort the efficiency of CDS markets to an extent, for the managers would resort to suboptimal strategies to take advantage of lower financing costs.

LIBOR is calculated based on the estimates submitted by the participating banks - these estimates approximate the interest rate a participant bank would charge to lend funds to another bank. Since the LIBOR setting process takes place once a day, it is evident that the intra-day events would not be captured by the LIBOR almost instantaneously. Though one participant bank who foresees this risk in the markets could theoretically profit from it by borrowing funds today at a lower LIBOR and lending funds tomorrow at a higher LIBOR, this does not usually happen since such an arbitrage is not risk-free. Additionally, no bank can accurately guess what tomorrow’s LIBOR would be, for it depends on the other banks’ views as well. In mid 2010s, reports surfaced about how the participating banks have been manipulating the LIBOR even before the subprime crisis. Mackenzie and Tett (2008) report that LIBOR has been lagging all other market-based measures of unsecured funding, which

(11)

11 could reasonably be assumed to be more efficient than the inter-bank market, which forms the basis of our expectations from this empirical study.

LITERATURE REVIEW

There are numerous research studies conducted around LIBOIS and CDS spread in isolation and together; the papers have been focussed on the factors driving the spreads, and how the spreads react relative to the sovereign spreads, and other swap spreads. Many of these studies employ the use of a Vector Auto Regression or a Vector Error Correction method, but additionally complemented by Impulse Response analysis.

Preliminary studies:

Ericsson, Jacobs, and Oviedo (2009) explore the relationship between theoretical determinants of default risk (leverage, volatility and risk-free interest rate) and actual CDS spreads. They estimate a multivariate linear regression and conclude that all the three determinants are statistically significant. This, in turn, implies that the CDS spread increases with increase in leverage, volatility or risk-free rates.

During times of market turbulence (with concerns over looming liquidity/default risk), there would be a spur in the demand for CDS and the CDS premia, as a result. According to the ECB (2016), sovereign credit spreads of the European countries shot up in the wake of the Eurozone crisis, for which credit risk (given by the CDS spread) played an important role. The deterioration of credit risk could be attributed to the loose fiscal policy of the governments. Beirne and Fratzscher (2013) carry out an empirical study on 31 advanced and emerging economies and observe that a deterioration of countries’ fundamentals led to CDS spreads rising during the crisis. The average Euro area CDS’s rose approximately 23x during the sovereign crisis. The European banks, that were heavily exposed to sovereign bonds,

(12)

12 witnessed the premia of the CDS’s issued on them rise in line with that of the sovereign bonds.

A question of intrigue was centred on whether the LIBOR-OIS spread during Subprime crisis was driven by liquidity risk or credit risk. The empirical study by Gefang, Koop and Potter (2011) shows that the 1-month and 3-month spreads were driven by liquidity risk and 12-month spread by credit and liquidity risk. Jin Cui, Francis In, and Elizabeth Ann Maharaj (2016) also talk about determinants of the behaviour of LIBOIS spread, throwing light on cross-country and across different economic conditions. The factors include counterparty risk, liquidity risk, leverage, credit risk, market volatility and general economy which are tested for statistical significance. Though the research has been focussed on liquidity and credit risk, the factors have been disentangled and analysed individually; and they conclude that the key drivers of the spread during the crisis were credit risk, market volatility and counterparty risk.

From the above studies, we can infer that the underlying factors triggering an effect on LIBOR-OIS spread and CDS spread are significantly related. Our question, however, is not around which spread effects a change in the other. Rather, it centres on which spread reacts quicker to the underlying factors than the other.

Other lead-lag studies involving LIBOR-OIS Spread or CDS spread, but not both:

Calice, Chen and Williams (2013) study about the cross liquidity and price discovery in sovereign bond markets and sovereign CDS markets, using a time varying VAR. Their studies reveal that there is a lagged transmission liquidity spread from the CDS market to the credit spread in the bond market and that the liquidity of the CDS markets significantly and in

(13)

13 a time-variant fashion impact the sovereign bond credit spread. The study estimates the following model for credit spread and liquidity spread:

BONDCSt = β 1,1,t BONDCSt−1 + β1,2,t CDSCSt−1 + β1,3,t BONDLSt−1 + β1,4,tCDSLSt−1 + μ1,t + u1,t

CDSCSt = β2,1,t BONDCSt−1 + β2,2,tCDSCSt−1 + β2,3,tBONDLSt−1 + β2,4,tCDSLSt−1 + μ2,t + u2,t

The study by Alter and Schuler (2012) explores the lead-lag relation between governments’ and banks’ default risk, focusing on the government aid in the form of bailouts during the crisis. The interconnectedness between the both has altered significantly due to the intervention; suggests a contagion from bank CDS spreads to sovereign spreads before government interventions; and post-intervention, the sovereign CDS spreads are a significant determinant of bank CDS spreads. They deploy a bivariate VAR and a VEC methodology to study the relation.

Murphy and Murphy (2010) analyse the temporal differences that liquidity and credit/default risk factors shocks have on swap spreads. They employ Vector Autoregression methodology to capture the impulse-response based on four variables and find that CDS spreads have an immediate effect on the swap spreads and show persistence over a longer period. This paper is loosely based on this analysis, but it departs from the methodology followed therein and the same has been tabulated below:

Murphy & Murphy This Analysis

Variables used 2-year swap spreads to

Treasury yields

3-month Euro LIBOR-OIS spread

3-month OIS rates

(14)

14 5-year North American

Investment Grade CDS

Time frame Mid-2007 to 2009-10 Late 2008 to end of 2016

Estimation of optimal lag length

Likelihood Ratio Test AIC, SC, HQC (Criterion

tests)

Region The U.S. Europe

Further, there are a set of expectations that I have with regard to what the output of my empirical analysis would be. One of the assumptions underlying my expectations was the efficient functioning of CDS markets, deriving the spreads to its real value lest there should be any arbitrage opportunity. A research by Smales (2016) precisely confirms the conviction about CDS markets, as shown below.

Paper showing the efficiency of CDS markets:

Smales (2016) analyses the relationship between news sentiment for a set of major international banks and LIBOIS spread & CDS spread. He observes a significant negative asymmetric relationship between news sentiment and CDS spreads. However, the author notes that while market determined credit measures (CDS spreads) respond to news releases, bank determined measures (LIBOIS spreads) do not. This in line with our expectations of how LIBOR market functions, and its inherent lack of flexibility in responding to information on a real-time basis. He further points out that the effect of news events for one bank on the credit measure of others reveals a certain level of interconnectedness, with news for Deutsche Bank influencing CDS spreads for all banks, whilst news for Citigroup impacting LIBOIS spreads in USD markets. The strong interconnectedness goes one step further to imply that CDS spreads could change quicker than LIBOR in response to a development in another part

(15)

15 of the world. This interconnection is justified since the developed markets are quite integrated, so shocks propagate from one to the other rather quickly.

Paper on which regression methodology of this paper is based:

According to Norden and Weber (2009), Marsh and Wagner (2012), and Hilscher, Pollet, and Wilson (2015), information flows unidirectionally from stocks to CDS. This theory was then challenged by Lee, Naranjo, and Velioglu (2017), who focussed on private and public forms over 2001-2013. Though they concur with the general conclusion that information flows from stocks to CDS when markets are functioning normally, they find that CDS markets offer good predictability when firm-specific information is involved.

�𝛥𝛥𝑆𝑆𝑖𝑖𝑡𝑡 𝑐𝑐𝑐𝑐𝑐𝑐 𝛥𝛥𝑆𝑆𝑖𝑖𝑡𝑡𝑏𝑏𝑏𝑏𝑛𝑛𝑐𝑐 � = �𝛽𝛽0,𝑖𝑖,𝑐𝑐𝑐𝑐𝑐𝑐 𝛽𝛽0,𝑖𝑖,𝑏𝑏𝑏𝑏𝑛𝑛𝑐𝑐� + 𝛴𝛴𝑘𝑘=1 𝑛𝑛 _�𝛽𝛽𝑘𝑘,𝑐𝑐𝑐𝑐𝑐𝑐,𝑐𝑐𝑐𝑐𝑐𝑐 𝛽𝛽𝑘𝑘,𝑐𝑐𝑐𝑐𝑐𝑐,𝑏𝑏𝑏𝑏𝑛𝑛𝑐𝑐 𝛽𝛽𝑘𝑘,𝑏𝑏𝑏𝑏𝑛𝑛𝑐𝑐,𝑐𝑐𝑐𝑐𝑐𝑐 𝛽𝛽𝑘𝑘,𝑏𝑏𝑏𝑏𝑛𝑛𝑐𝑐,𝑏𝑏𝑏𝑏𝑛𝑛𝑐𝑐� � 𝛥𝛥𝑆𝑆𝑖𝑖𝑡𝑡−𝑘𝑘𝑐𝑐𝑐𝑐𝑐𝑐 𝛥𝛥𝑆𝑆𝑡𝑡−𝑘𝑘𝑏𝑏𝑏𝑏𝑛𝑛𝑐𝑐 � + �€𝑡𝑡 𝑐𝑐𝑐𝑐𝑐𝑐 €𝑡𝑡𝑏𝑏𝑏𝑏𝑛𝑛𝑐𝑐� �𝑅𝑅𝑖𝑖𝑡𝑡 𝑐𝑐𝑡𝑡𝑏𝑏𝑐𝑐𝑘𝑘 𝑅𝑅_{𝑖𝑖𝑡𝑡}𝑐𝑐𝑐𝑐𝑐𝑐 � = � 𝛽𝛽0,𝑖𝑖,𝑐𝑐𝑡𝑡𝑏𝑏𝑐𝑐𝑘𝑘 𝛽𝛽0,𝑖𝑖,𝑐𝑐𝑐𝑐𝑐𝑐 � + 𝛴𝛴_𝑘𝑘=13 _�𝛽𝛽𝑘𝑘,𝑐𝑐𝑡𝑡𝑏𝑏𝑐𝑐𝑘𝑘,𝑐𝑐𝑡𝑡𝑏𝑏𝑐𝑐𝑘𝑘 𝛽𝛽𝑘𝑘,𝑐𝑐𝑡𝑡𝑏𝑏𝑐𝑐𝑘𝑘,𝑐𝑐𝑐𝑐𝑐𝑐 𝛽𝛽𝑘𝑘,𝑐𝑐𝑐𝑐𝑐𝑐,𝑐𝑐𝑡𝑡𝑏𝑏𝑐𝑐𝑘𝑘 𝛽𝛽𝑘𝑘,𝑐𝑐𝑐𝑐𝑐𝑐,𝑐𝑐𝑐𝑐𝑐𝑐 � � 𝑅𝑅𝑖𝑖𝑡𝑡−𝑘𝑘𝑐𝑐𝑡𝑡𝑏𝑏𝑐𝑐𝑘𝑘 𝑅𝑅_{𝑖𝑖𝑡𝑡−𝑘𝑘}𝑐𝑐𝑐𝑐𝑐𝑐 � + � €𝑡𝑡𝑐𝑐𝑡𝑡𝑏𝑏𝑐𝑐𝑘𝑘 €𝑡𝑡𝑐𝑐𝑐𝑐𝑐𝑐 �

This paper employs a panel VAR for the daily change in CDS spreads, stock returns and bond yield spreads. I propose to use a similar methodology, but since LIBOIS spread remains the same irrespective of the bank, VAR is done for an average taken across the banks. However, the above paper chooses a lag order of 3, whereas this paper selects the lag length depending on the values given by the information criterion.

The above study involves an event study; and the test is carried out in different windows to capture the relationship between the variables. However, my paper is not an event study in itself; instead it studies the dynamic behaviour of both the spreads as we traverse the financial cycle of the past decade.

(16)

16

DATA AND REGRESSION METHODOLOGY

Broadly speaking, this paper deals with two spreads; for which we need data on three variables: 1. CDS spread; 2. 3-month EUR LIBOR and 3. Overnight Indexed Swap rate. The daily 3-month EUR Libor is obtained from the data published by Federal Reserve Bank of St. Louis and the daily 3m OIS rates and CDS spreads of banks are downloaded from Datastream. The period covered is from 07.10.2008 to 31.12.2016. These data are not available for weekends and the entire dataset is taken as if it is a five-day week. The banks studied are: Allied Irish Bank (AIB), Banco Santander (BANSA), Barclays (BARC) , BNP Paribas (BNP), Commerzbank (COMM), Credit Suisse (CSUI), Deutsche Bank (DEU),

HSBC, ING Bank, RABO bank, Société Générale (SG) and UBS.

The CDS spreads are so chosen as to be actively traded, and since the 5-year spreads are the most traded ones, we use the CDS spreads of the same tenor. All the CDS are of senior unsecured nature.

Allied Irish Bank stands out from the rest of the group in that the CDS is not as liquid as others in the sample – the prices stay constant for a couple of days together. Further, the bank was bailed out by the ECB in the aftermath of the sovereign debt crisis. Because of these reasons, the bank is not considered for our statistical tests from Stage 4 on.

The entire 8 years is split into 7 stages, based on how the events unfolded, and regrouped into the pre-, post-sovereign crisis and the crisis period itself. The time-period is split up into smaller time frames in order to facilitate a better understanding of how the interaction between the spread changes.

A short description of each of the phases and the constituent stages is given below. 1. Pre-Sovereign Crisis phase

a. Stage 1: Deepening concern about the global impact of sub-prime crisis: 07.10.2008 to 16.03.2009

(17)

17 This period begins with the collapse of Lehman brothers, triggering the risk of a global contagion. This was closely followed by a slew of rate cuts by central banks across the world. Further, the Madoff scandal blew up in the later half of the year, affecting the banks in U.S. and Europe.

b. Stage 2: Initial signs of recovery 17.03.2009 to 30.11.2009

This period is marked by a respite from the rampant flight to quality and acute liquidity concerns. The stock market started recouping its losses, and a stronger boost could be attributed to the tailwinds of favourable macroenvironment (employment data and global industrial production). This further helps towards a gradual increase in optimism amongst the investors. This period also saw the DJIA closing above the 10,000 mark since October 2008.

c. Stage 3: Initial concerns about Eurozone crisis 01.12.2009 to 31.05.2010

Following the Dubai sovereign debt crisis, a potential crisis was brewing on some of the EU member states. This was initially fuelled by the rising debt level of Greece (c. €300bn at the time, accounting for 113% of GDP), and a string of other fiscal deficit irregularities. By May 2010, the EU decided on a bailout programme to the tune of €110bn for Greece. This was made worse when the EU published a report citing the accounting issues in Greece, following which budget deficit was revised to 12.7%. Concerns over Greece leaving the EU did spark off initial worries about the Eurozone, especially when other peripheral member states started facing the axe.

2. Sovereign Crisis

(18)

18 Soon after the first Greece bailout, other member states fell under scrutiny with Ireland and Portugal being the first ones. This was the period when European stability mechanism funding was established. Over the period, the crisis spread to Italy and Spain as borrowing costs edge higher and the accompanying sovereign downgrades. The macroeconomic outlook grew grim, with the growth slowing or standstill at best. France was also downgraded in 2012, not before the largest mortgage lender in Spain got bailed out, Bankia, required a bailout to the tune of €23bn.

b. Stage 5: ECB initiatives and announcement of bailout programs: 01.01.2013 to 28.02.2015

After initial rounds of bailouts, the government announced purchase of government bonds from the troubled banks purchase plan in the second half of 2012. The interest rates were also cut below zero to boost economic activity. 3. Post-crisis

a. Stage 6: QE 1 – March 2015 to March 2016 b. Stage7: QE2 – April 2016 to Dec 2016

In a final effort to rescue the eurozone and fight the prolonged period of very low inflation, the ECB announced QE on a massive scale, with the first phase launched in March 2015, at an average monthly price of €60bn. QE 2 had a monthly asset purchase of €80bn, and QE 3 €60bn. As per the latest directive in June 2018, monthly purchases of €30bn are expected to continue till September after which the purchase would be reduced to €15bn for October to December. However, the interest rates are expected to stay at record lows through the summer of 2019.

(19)

19 The descriptive statistics of the change in CDS spreads for each of the banks, the average change in the spread and change in LIBOIS spread are tabulated stage-wise in Appendix 1. The mean change in CDS spreads is negligibly close to zero in almost all the stages. But the standard deviation is higher in stage 1, implying the CDS spreads have been a bit more volatile during the crisis than otherwise. Interestingly, during the sovereign debt crisis in Stage 4, the standard deviation is lesser than during the run-up to the crisis and the sub prime crisis. This implies that CDS spreads did not change as drastically as it did during 2007-09 crisis. This could be partially because the banks in the panel were not the direct victims of weakening asset bases. For example, Commerz Bank has a mean of close to 0 through 2008-2016, but the standard deviation of the change in CDS’s is 6.71% in late 2008-early 2009 (Stage 1), 3.33% during the initial recovery phase (Stage 2), 5.05% for Stage 3, 4.49% during the sovereign crisis, 2.65% after the crisis, before rising back to 3.47% and 3.16% in 2016 and 2017. The sudden surge in 2016 and 2017 could be explained by the poor performance given by the European banks.

The average change in CDS spreads also follow a similar pattern, having a mean close to zero and standard deviation close to 4.9% during subprime crisis, and 4.5% during the period marked by concerns about Eurocrisis. After the crisis, it stayed close to 2.7%, but rose to 3.5% in 2015-early 2016 and 3.0% for 2016 when CDS spreads of European banks were rising on the back of unfavourable earnings results.

The change in Euro LIBOIS spreads have a mean value close to 0 for all the 7 stages; but the standard deviation is much higher during the sovereign crisis than other stages – the statistic stood at 6.65% and 14.64% for 2010-2012 and 2013-2015, in relation to the remaining times when it had a mean value of c.3% on average. This is reasonable since Euro lending got riskier with the onset of the Eurozone crisis, only to be followed by prolonged uncertainty surrounding the ECB’s measures to fight the same.

(20)

20 As already discussed, AIB is not a part of the dataset throughout the entire time frame, while we study the behaviour of spreads. The presence of idiosyncratic factors involved in the volatility of the CDS spread might skew the results. We analyse the behaviour of spreads for by taking an average of the daily change in CDS spreads of all banks and analyse the same in relation to the theoretical expectations.

The type of relationship being studied is the lead-lag dynamics of the two spreads; and how it varies as the economy goes through a financial cycle (with a special focus on Eurozone crisis). Since the lead-lag relation does not entail co-movement, correlation matrix is not relevant for this case. Although the two spreads must be correlated in the long run, it is their reaction time we look to focus on. Further, if there is a lag between the two variables, then the correlation would exist only with the lagged variable, and not the original one.

For the lead-lag relationship, we make use of Vector Auto Regression – a statistical methodology that lets us introduce us as many lagged variables as the situation warrants. Here, we do a bivariate vector autoregression between daily changes in LIBOIS spread and CDS spread across all the banks on average. Vector Auto Regression methodology has been very popular amongst academicians who look at tracking interdependencies between time series, even with little theoretical information about the relationship between the variables. The accompanying impulse-response analysis also helps the research examine the impact of a shock to each of the regressor variables on the dependent variable.

The VAR is not on the spreads themselves, rather on the daily changes of the spreads. We track the movement in the change of both the spreads. The VAR is represented as follows : The number of parameters to be tested for statistical significance depends on the lag parameter(n), which indicates how far into the past we intend to go. The optimal lag length is estimated by choosing the lag that minimizes the values of Akaike Information Criterion, Schwarz Criterion and Hannan Quinn Criterion.

(21)

21 �𝛥𝛥𝑐𝑐𝑐𝑐𝑐𝑐𝑖𝑖,𝑡𝑡 𝛥𝛥𝑙𝑙𝑒𝑒𝑒𝑒𝑡𝑡 � = �𝛽𝛽0,𝑖𝑖,𝑐𝑐𝑐𝑐𝑐𝑐 𝛽𝛽0,𝑙𝑙𝑖𝑖𝑏𝑏 � + 𝛴𝛴𝑘𝑘=1𝑛𝑛 � 𝛽𝛽𝑘𝑘,𝑐𝑐𝑐𝑐𝑐𝑐,𝑐𝑐𝑐𝑐𝑐𝑐 𝛽𝛽𝑘𝑘,𝑐𝑐𝑐𝑐𝑐𝑐,𝑙𝑙𝑖𝑖𝑏𝑏 𝛽𝛽𝑘𝑘,𝑙𝑙𝑖𝑖𝑏𝑏,𝑐𝑐𝑐𝑐𝑐𝑐 𝛽𝛽𝑘𝑘,𝑙𝑙𝑖𝑖𝑏𝑏,𝑙𝑙𝑖𝑖𝑏𝑏� � 𝛥𝛥𝑐𝑐𝑐𝑐𝑐𝑐𝑖𝑖,𝑡𝑡−𝑘𝑘 𝛥𝛥𝑙𝑙𝑒𝑒𝑒𝑒𝑡𝑡−𝑘𝑘 � + �𝑢𝑢𝑡𝑡 𝑐𝑐𝑐𝑐𝑐𝑐 𝑢𝑢𝑡𝑡𝑙𝑙𝑖𝑖𝑏𝑏 � , Where

𝛥𝛥𝑙𝑙𝑒𝑒𝑒𝑒𝑡𝑡 = Change in LIBOR-OIS spread at time ‘t’ (stays the same for both the subsets) 𝛥𝛥𝑐𝑐𝑐𝑐𝑐𝑐𝑖𝑖𝑡𝑡 = Change in CDS spread at time ‘t’ for ith bank

n = optimal lag length given by the information criterion

𝛽𝛽𝑘𝑘,𝑥𝑥,𝑦𝑦 = Regression slope that shows the linear relationship between x and y 𝛽𝛽0,𝑖𝑖,𝑐𝑐𝑐𝑐𝑐𝑐 = Regression intercept

𝑢𝑢𝑡𝑡𝑐𝑐𝑐𝑐𝑐𝑐 , 𝑢𝑢𝑡𝑡𝑙𝑙𝑖𝑖𝑏𝑏 = Error terms/Residuals

The null hypothesis for each of the regressions is given below: Ho: 𝛽𝛽1,𝑙𝑙𝑖𝑖𝑏𝑏,𝑐𝑐𝑐𝑐𝑐𝑐 = 𝛽𝛽2,𝑙𝑙𝑖𝑖𝑏𝑏,𝑐𝑐𝑐𝑐𝑐𝑐 = ⋯ = 𝛽𝛽𝑛𝑛,𝑙𝑙𝑖𝑖𝑏𝑏,𝑐𝑐𝑐𝑐𝑐𝑐 = 0 vs. Ha : 𝛽𝛽1,𝑙𝑙𝑖𝑖𝑏𝑏,𝑐𝑐𝑐𝑐𝑐𝑐 ≠ 0 𝑂𝑂𝑅𝑅 𝛽𝛽2,𝑙𝑙𝑖𝑖𝑏𝑏,𝑐𝑐𝑐𝑐𝑐𝑐 ≠ 0 𝑂𝑂𝑅𝑅 … 𝑂𝑂𝑅𝑅 𝛽𝛽𝑛𝑛,𝑙𝑙𝑖𝑖𝑏𝑏,𝑐𝑐𝑐𝑐𝑐𝑐 ≠ 0 and Ho: 𝛽𝛽1,𝑐𝑐𝑐𝑐𝑐𝑐,𝑙𝑙𝑖𝑖𝑏𝑏 = 𝛽𝛽2,𝑐𝑐𝑐𝑐𝑐𝑐,𝑙𝑙𝑖𝑖𝑏𝑏 = ⋯ = 𝛽𝛽𝑛𝑛,𝑐𝑐𝑐𝑐𝑐𝑐,𝑙𝑙𝑖𝑖𝑏𝑏 = 0 vs. Ha : 𝛽𝛽1,𝑐𝑐𝑐𝑐𝑐𝑐,𝑙𝑙𝑖𝑖𝑏𝑏 ≠ 0 𝑂𝑂𝑅𝑅 𝛽𝛽1,𝑐𝑐𝑐𝑐𝑐𝑐,𝑙𝑙𝑖𝑖𝑏𝑏 ≠ 0 𝑂𝑂𝑅𝑅 … 𝑂𝑂𝑅𝑅 𝛽𝛽1,𝑐𝑐𝑐𝑐𝑐𝑐,𝑙𝑙𝑖𝑖𝑏𝑏 ≠ 0

For instance, if we deduce that 𝛽𝛽2,𝑙𝑙𝑖𝑖𝑏𝑏,𝑐𝑐𝑐𝑐𝑐𝑐 is statistically significant at the α level, this implies the following: If a change in CDS spreads is observed on day t due to some negative information (possibly arising out of a widespread fear prevalent in the market), LIBOR-OIS spreads would price in the information on day ‘t+2’.

OIS rates are set at night, and LIBOR in the mornings at 11 AM. If we use the same day rates for both the spreads, there is a high chance that OIS rates would be influenced by the LIBOR

(22)

22 of that day. To reduce this dependency, we calculate the spread by subtracting OIS of day i from LIBOR of day i+1.

Each of the regression variable(Δlibois spread and Δcds spread) is tested for stationarity, before proceeding with the VAR. The stationarity testing is done using Automated Dickey Fuller Unit Root Testing. The stationarity ensures that the seasonality and the trend is removed from the data. All the variables are stationary at level, for all the 7 stages.

The optimal lag length selected for each of the regression in the 7 stages are tabulated below in Table 1. These values are selected from that chosen by 3 information criterions, and the one that minimizes the value of the criterion is chosen.

Table 1. Optimal lag length as selected by the information criterion (stage-wise)

STAGE 1 STAGE 2 STAGE 3 STAGE 4 STAGE 5 STAGE 6 STAGE 7

CHAVGCDS ₂ ₁ ₁ ₃ ₁ ₈ ₂

RESULTS

An asset-weighted average of the change in CDS spreads is taken across all banks and vector auto-regressed against change in LIBOR-OIS spread.

Stage 1: 07.10.2008 to 16.03.2009

Table 2. Stage 1 Results of Vector Autoregression performed on changes in LIBOIS (Libor-OIS Spread) and average change in CDS spread. The estimated regression parameters in the respective regression are given below, with the corresponding t-statistics in parentheses. *,** and *** correspond to 1%,5% and 10% significance level, respectively. The first row consists of regression with both libois spread and cds spread as the dependent variable.

CHLIBOIS CHAVGCDS CHLIBOIS (-1) 0,154317 -0,03315 (1,90802**) (-0,26535) CHLIBOIS (-2) 0,040738 -0,22823 (0,50258) (-1,82303*) CHAVGCDS (-1) 0,19005 0,447529 (3,13215***) (4,77534***) CHAVGCDS (-2) -0,05961 -0,21782

(23)

23 (-1,01754) (-2,40726**)

As can be seen from Table 2, CDS of banks on average leads the LIBOIS spread by 1 day. The relationship is economically and statistically significant at 1%, and the nature of the relationship is positive. This means that on average, an increase in CDS spread today is followed by an increase in LIBOIS spread tomorrow. The CDS spread of lag 2 does not have a significant relationship with the LIBOIS spread of present day.

The LIBOIS from 2 days back also has a statistically significant (at 10%) relationship with the CDS spread today, and the direction of the relationship is in the negative territory – meaning, if LIBOIS spread decreases today, the CDS spread increases 2 days from today. This need not be always economically significant, however. But, the LIBOIS from 1 day back has an insignificant relationship with the CDS spread of the present.

Drawing on both the arguments above, we could conclude that CDS leads LIBOR and not the other way around. Such a behaviour could be reasonably expected, since this stage covers the time frame that was fraught with concerns over a world-wide recession. Further, all the CDS spreads leading by 1 day reinforces our theory that the LIBOR would have to wait a whole day for it to change, while CDS captures it in real-time.

Stage 2: 17.03.2009 to 30.11.2009

CHLIBOIS CHAVGCDS

CHLIBOIS (-1) 0,015803 -0,0577

(0,21269) (-1,01223)

CHAVGCDS (-1) 0,176716 0,305573

(24)

24 Table 3 shows that CDS spreads, on average, leads the LIBOIS spreads by 1 day; and the same is statistically significant at 10%, with a positive relationship between the two spreads. This means that an increase in CDS spreads today is followed by an increase in LIBOR-OIS spread the next day.

The LIBOIS spread does not have a statistically significant relationship with the CDS spread. The direction of the relationship is negative – which is not theoretically accurate. This implies that the CDS spreads lead the LIBOIS spread as is cited in the below example:

CDS decreases on day t -> LIBOIS decreases on day t+1 -> CDS increases on day t+2 -> LIBOIS increases on day t+3

The LIBOIS decreasing on day t+1 and CDS increasing on t+2 is not a direct consequence of a lead-lag relationship; but is in fact proof that CDS reacts to new information first, moves accordingly and making LIBOIS move in the same direction a day later.

This time period showed initial signs of recovery from the global recession – meaning the markets are gradually moving out of the turmoil. Since the relationship is significant only at 10% for this period, this can be characteristic of a market recovering from a flustered state. The markets are relatively calmer than Stage 1.

Stage 3: 01.12.2009 to 31.05.2010

CHLIBOIS CHAVGCDS

CHLIBOIS (-1) 0,091186 -0,4405

(0,94891) (-3,06494***)

CHAVGCDS (-1) -0,15928 0,211036

(25)

25 From the regression output for this stage(given by Table 4), the lags of CDS spreads are statistically significant at 5% and the lags of LIBOIS spreads are statistically significant at 1%.The nature of relationship for both the regressions are negative. This means that an increase in one spread is followed by a decrease in the second and in turn an increase in the first. This seems to imply that both the spreads are moving independent of each other. This is possible when there are a lot of other factors influencing the CDS spreads that the common factor underlying both the spreads seems to have little impact on the movement of the CDS spreads.

The period covered by this stage relates to the initial concerns about a potential crisis brewing in the Eurozone. This was triggered by the high levels that Greece’s debt had reached. Bond yields started to soar, relative to the German bonds – meaning the crisis, in the initial stages, affected Greece more than the rest of the countries. With the exception of Ireland (AIB), the data set does not comprise of nations that were hit hard during the Sovereign crisis. The asset-weight corresponding to AIB is relatively smaller (c.1%), and hence its impact in the overall average change in CDS is minuscule. The regression output, therefore, is in line with what we expect to see based on our theoretical assumptions.

Stage 4: 01.06.2010 to 31.12.2012

CHLIBOIS CHAVGCDS CHLIBOIS (-1) -0,0824 0,014209 (-2,10882**) (0,77402) CHLIBOIS (-2) 0,126681 0,010007 (3,25869***) (0,54797) CHLIBOIS (-3) -0,03941 -0,00559 (-1,01750) (-0,30690) CHAVGCDS (-1) 0,371856 0,312099 (4,50491***) (8,04838) CHAVGCDS (-2) -0,14189 -0,04343 (-1,62465) (-1,05846) CHAVGCDS (-3) 0,009996 -0,12534

(26)

26 -0,1198 (-3,19786***)

Allied Irish Bank is excluded from the data set from this stage on. This period corresponds to the Sovereign debt crisis, with a series of downgrades, and one after another the countries falling prey to the high debt levels and loose fiscal policies. This gave rise to serious concerns about the viability of the monetary union.

The regression results in Table 5 show that CDS spreads lead on average by 1 day. The relationship is economically and statistically significant at 1%; also, the spreads are positively related. However, the prior lags are not significant, meaning LIBOIS catches up to pricing in the new market development.

None of the LIBOIS lags are significant, lending credence to our assumption that CDS markets are efficient (also studied by Smales(2016)). During times of crisis, the market participants would want to shed risk as quickly as possible, and hence, it is reasonable to expect that CDS spreads lead on average. Also, LIBOR setting is a prerogative of only a handful of market participants, and they are set on a daily basis making LIBOR unable to capture the intra-day events in a timely manner.

The result showing CDS spreads leading the LIBOIS spread shows how the sovereign debt crisis fuelled a contagion effect on all the member states of the EU.

Stage 5: 01.01.2013 to 28.02.2015

CHLIBOIS CHAVGCDS

CHLIBOIS (-1) -0,20691 -0,01098

(-4,91482***) (-1,39328)

CHAVGCDS (-1) 0,054827 0,15907

(27)

27 The regression results as tabulated in Table 6 shows that lags of both LIBOIS spread and CDS spread bear an insignificant relationship with the present day CDS spread and LIBOIS spread. Further, a decrease in LIBOIS on day t is expected to be followed by an increase in CDS spread on day t+1. This is again indicative of other factors driving the CDS spreads up/down – these factors may be unique to a few players or rather, to a particular sector. This is in line with our assumption since this stage corresponds to the period when the ECB laid ground for QE and issued forward guidance on interest rates. For one, the investors were not ready to take on more risk than they were capable of. Secondly, everyone was better off keeping track of ECB’s policy measures, and hence the markets were relatively calmer with both the LIBOIS and CDS moving independent of one another (triggered chiefly by the central bank announcements and long-term policy decisions; and variability in CDS spreads additionally driven by the idiosyncratic factors).

Stage 6: 01.03.2015 to 31.03.2016

CHLIBOIS CHAVGCDS CHLIBOIS (-1) -0,04491 -0,00366 (-0,71712) (-0,06429) CHLIBOIS (-2) 0,006063 0,047788 (0,09787) (0,84837) CHLIBOIS (-3) 0,125344 0,009954 (2,04135**) (0,17829) CHLIBOIS (-4) -0,05306 0,053758 (-0,86478) (0,96356) CHLIBOIS (-5) -0,14856 -0,07322 (-2,42062**) (-1,31209) CHLIBOIS (-6) -0,07619 -0,07144 (-1,23695) (-1,27547) CHLIBOIS (-7) 0,137433 -0,05623 (2,21863**) (-0,99843) CHLIBOIS (-8) -0,14347 0,078583

(28)

28 (-2,30688**) (1,38965) CHAVGCDS (-1) 0,109482 0,350577 (1,57334) (5,54102***) CHAVGCDS (-2) 0,044009 -0,18685 (0,59414) (-2,77436***) CHAVGCDS (-3) -0,01286 0,210066 (-0,17119) (3,07688***) CHAVGCDS (-4) 0,119296 -0,17711 -1,56324 (-2,55248**) CHAVGCDS (-5) -0,12963 0,013436 (-1,69594*) (0,19332) CHAVGCDS (-6) 0,155244 -0,04839 (2,06029**) (-0,70630) CHAVGCDS (-7) -0,08369 -0,00268 (-1,12309) (-0,03957) CHAVGCDS (-8) 0,098085 -0,03076 (1,39445) (-0,48100)

As shown in Table 7, the lags of LIBOIS spreads have an insignificant relationship with the CDS spreads on average. The initial lags of CDS spreads too bear an insignificant relationship, but the 5th and 6th lag are statistically significant at 10% and 5% respectively. The 5th_{lag is negatively related and the 6}th_positively.

This period corresponds to the first phase of the Quantitative Easing process and one would expect the spreads to move in tandem since everyone, during this time, looks up to the central bank to make decisions and set expectations for the market.

These relationships do not seem to be economically significant; for one, the CDS spreads leading the LIBOIS spread by 6 days does not reflect the actual way of functioning of markets.

LIBOR not moving much to reflect the state of markets could make us think that LIBOR was artificially kept low/high. However, the LIBOR scandal was put out in early 2014 – when the entire administration went through an overhaul.

Further, if the CDS spread movements are driven by reference entity-specific concerns, it could take a while before the rest of the market sense contagion and in turn, the LIBOIS spread to price in the systemic risk. In early 2016, the earnings announcements(poor

(29)

29 performance) by the European banks caused an unusual surge in the CDS’s traded on these banks. This development could have partly influenced the results observed in our regression.

Stage 7: 01.04.2016 to 31.12.2016

CHLIBOIS CHAVGCDS CHLIBOIS (-1) -0,1854 0,000995 (-2,58821***) (0,03964) CHLIBOIS (-2) -0,26005 0,013406 (-3,65648***) (0,53804) CHAVGCDS(-1) 0,510287 0,349495 (2,44479**) (4,77942***) CHAVGCDS(-2) 0,031563 -0,17546 -0,14901 (-2,36451**)

From Table 8, we can infer that all the relationships are positive; but the lags of LIBOIS are

statistically insignificant. The 1st_{lag of CDS spreads, however, are economically and}

statistically significant at 5%. An increase in LIBOIS spread follows an increase in CDS spread of the previous day.

Though QE II was ongoing during this phase, CDS markets went through a turbulent phase during 2016. Typically, in the first few years following a crisis when the economy is still recovering, one would expect the markets to remain calm or rather the LIBOR to lead. But these results seem to be in contrast with our expectation – this could be attributed to concerns over Brexit and dismal earnings shows by European banks. Poor performance was a direct result of having lesser revenues (generated by the low interest rate environment).

LIMITATION & POTENTIAL IMPROVEMENT OF CURRENT

RESEARCH

(30)

30

Liquidity of Credit Default Swaps:

Over the past 10 years, the number of active CDS has been steadily decreasing. This reflects the unwillingness of the market participants to invest in risky and arcane instruments. This study involves the daily change in CDS for capturing the lead-lag relationship with change in the LIBOIS spread. This inherently assumes that CDS are liquid and actively traded.

As studied by Irresberger, Weiß, Gabrysch and Gabrysch(2018), there is a time-varying liquidity tail risk explaining the variability of CDS spreads. Their results show that liquidity crunch is a crucial driver of CDS spreads, with the sellers of CDS demanding a premium for taking on the risk. The Liquidity CAPM model developed by Acharya and Pedersen(2005) has been the basis for several research papers regarding factoring in liquidity in derivatives pricing. While Bongaerts, De Jong and Driessen(2011) do confirm that there is a liquidity premium earned by the protection seller the effect of liquidity risk is economically small. On the other hand, Badaoui, Cathcart and El-Jahel(2014) explain the implied liquidity risk premium in the term structure of sovereign CDS spreads. The same could be extended to corporate CDS markets, but we might run a risk of not having a rich data set since corporate CDS’s are not actively traded in shorter.

Bearing the above concerns in mind, I have chosen the most actively traded CDS’s of the largest banks in Europe. This implies that the results would be heavily contingent on the liquidity of underlying CDS’s , the reference entities being included and the markets in which these CDS trade. Further, the liquidity of a CDS is partially dependent on the reference entity per se. Having a different dataset could yield a conflicting set of results, which might not be in line with the theoretical expectations. Particularly, should the sample included fragile banks with a high risk of bankruptcy, the CDS spreads would be far more volatile than usual. This could typically skew the results into becoming less meaningful. The problem could be

(31)

31 partially mitigated by introducing a liquidity-adjusted CDS spread for the banks. The study based on the liquidity CAPM and associated studies could serve as a pointer for calculating the liquidity premium.

CONCLUSION

Studies have shown that LIBOR-OIS spread and CDS spreads are affected by related market factors. We have witnessed LIBOR surging to high levels during times of liquidity crunch. A Credit Default Swap is, by design, expected to work that way – it being equivalent to betting against the company. In times of market-wide or systemic turmoil, the CDS spreads also increase. Though there have been studies surrounding how each of the spreads react relative to other market variable, little has been done towards studying the dynamics of LIBOIS spread and CDS spread relative to one other.

This thesis studies these spreads and their behaviour over a time span of 9 years, beginning late 2008 to the end of 2016. The spreads cannot be expected to behave in a pre-specified way; instead, the relationship changes with time depending on the macro environment. The period chosen is interesting in that it encompasses the latter half of the subprime crisis, the run-up to the sovereign crisis, the eurozone crisis and the post-crisis QE phases. The daily change in LIBOIS spread is regressed against the lags of CDS spread and that of its own. A similar regression is run for average daily change of CDS spread. The results of the regressions reveal that on average, CDS spreads do lead LIBOIS spread during turbulent times. However, during times of QE the markets are expected to remain rather calm and hence, both the spreads to move together. In 2016 and 2017, CDS spreads were more volatile than usual due to other factors at play at the time. Since liquidity is an important pre-requisite in the selection of CDS issues, this study could produce different results when the securities chosen are less liquid, or belong to smaller banks, or the banks are in markets other than

(32)

32 Europe and NA (where CDS trading is not that popular). The use of liquidity-adjusted CDS spreads would be reliable enough to study the behaviour between these spreads.

REFERENCES

Acharya, V. V., & Pedersen, L. H. (2005). Asset pricing with liquidity risk. Journal of Financial Economics,77(2), 375-410. doi:10.1016/j.jfineco.2004.06.007

Alter, A., & Schüler, Y. S. (2012). Credit spread interdependencies of European states and banks during the financial crisis. Journal of Banking & Finance,36(12), 3444-3468. doi:10.1016/j.jbankfin.2012.08.002

Badaoui, S., Cathcart, L., & El-Jahel, E. (2016).Implied liquidity risk premium in the term

structure of sovereign credit default swap and bond spreads. The European Journal of

Finance,22:10,825-853, doi:10.1080/1351847X.2014.996297

Bank for International Settlements. Statistical Release: OTC Derivatives statistics as of December end for years 2004 to 2017. https://www.bis.org/

Beirne, J., & Fratzscher, M. (2013). The pricing of sovereign risk and contagion during the European sovereign debt crisis. Journal of International Money and Finance,34, 60-82. doi:10.1016/j.jimonfin.2012.11.004

Bongaerts, D., De Jong, F., & Driessen, J. (2011). Derivative pricing with liquidity risk: Theory and evidence from the credit default swap market. Journal of Finance, 66, 203–240.

Calice, G., Chen, J., & Williams, J. (2013). Liquidity spillovers in sovereign bond and CDS markets: An analysis of the Eurozone sovereign debt crisis. Journal of Economic Behavior & Organization,85, 122-143. doi:10.1016/j.jebo.2011.10.013

Cui, J., In, F., & Maharaj, E. A. (2016). What drives the Libor–OIS spread? Evidence from five major currency Libor–OIS spreads. International Review of Economics & Finance,45, 358-375. doi:10.1016/j.iref.2016.04.002

De Santis, R., Stein, M. (2016). Correlation changes between the risk-free rate and sovereign yields of euro area countries. Working Paper Series, No 1979, European Central Bank. https://www.ecb.europa.eu/pub/pdf/scpwps/ecbwp1979.en.pdf?66cb3ed486dcb6e2e8624bd6 a5a41324

(33)

33

European Central Bank. Expanded Asset Purchase Programme. https://www.ecb.europa.eu/mopo/implement/omt/html/index.en.html

Ericsson, J., Jacobs, K., & Oviedo, R. (2009). The Determinants of Credit Default Swap Premia. Journal of Financial and Quantitative Analysis,44(01), 109-132. doi:10.1017/s0022109009090061

Gefang, D., Koop, G., & Potter, S. M. (2011). Understanding liquidity and credit risks in the financial crisis. Journal of Empirical Finance,18(5), 903-914. doi:10.1016/j.jempfin.2011.07.006

Gorton, G., & Metrick, A. (2012). Securitized banking and the run on repo. Journal of Financial Economics, 104(3), 425-451. doi:10.1016/j.jfineco.2011.03.016

Hilscher, J., Pollet, J. M., & Wilson, M. (2015). Are Credit Default Swaps a Sideshow? Evidence That Information Flows from Equity to CDS Markets. Journal of Financial and Quantitative Analysis,50(03), 543-567. doi:10.1017/s0022109015000228

Hull, J. C., & White, A. (2012). Libor vs. OIS: The Derivatives Discounting Dilemma. SSRN Electronic Journal. doi:10.2139/ssrn.2211800

Irresberger, F., Weiß, G. N., Gabrysch, J., & Gabrysch, S. (2018). Liquidity tail risk and credit default swap spreads. European Journal of Operational Research, 269(3),1137-1153, doi:https://doi.org/10.1016/j.ejor.2018.02.030

Lee, J., & Velioglu, G. (2017). When Do CDS Spreads Lead? Rating Events, Private Entities, and Firm-Specific Information Flows. SSRN Electronic Journal. doi:10.2139/ssrn.2933052 Lewin, J. (2016). Gloom gathers around European bank shares and CDS. Financial Times (08-Feb-16).

Mackenzie, M., Tett, G. (2008). LIBOR remarks fail to put unease to rest. Financial Times (02-Jun-08).

Marsh, I. W., & Wagner, W. (2012). Why is Price Discovery in Credit Default Swap Markets News-Specific? SSRN Electronic Journal. doi:10.2139/ssrn.1980860

Minton, B. A., Stulz, R., & Williamson, R. (2008). How Much Do Banks Use Credit Derivatives to Hedge Loans? Journal of Financial Services Research,35(1), 1-31. doi:10.1007/s10693-008-0046-3

(34)

34

Murphy, F., & Murphy, B. (2010). A vector-autoregression analysis of credit and liquidity factor dynamics in US LIBOR and Euribor swap markets. Journal of Economics and Finance, 36(2), 351-370. doi:10.1007/s12197-010-9122-2

Norden, L., & Weber, M. (2009). The Co-movement of Credit Default Swap, Bond and Stock Markets: An Empirical Analysis. European Financial Management,15(3), 529-562. doi:10.1111/j.1468-036x.2007.00427.x

Randow, J., Diamond J.S., & Warren, H. (2017). When will ECB pull trillions from the markes? Bloomberg (20-July-2017)

Smales, L. A. (2016). News sentiment and bank credit risk. Journal of Empirical Finance, 38, 37-61. doi:10.1016/j.jempfin.2016.05.002

Völz, M., & Wedow, M. (2011). Market discipline and too-big-to-fail in the CDS market: Does banks size reduce market discipline? Journal of Empirical Finance,18(2), 195-210. doi:10.1016/j.jempfin.2011.01.001

Warden,G., (2014). Europe in turmoil : five years of economic crisis. The Guardian (17-October-2014)

Zawadowski, A., & Oehmke, M. (2012). The Anatomy of the CDS Market. SSRN Electronic Journal. doi:10.2139/ssrn.2023108

APPENDIX 1 : Stage-wise Descriptive Statistics

Stage 1:

Variable # Obs Mean Std. Dev. Min Max

CHAIB 111 0,015585 0,063779 -0,22727 0,333333 CHBANSA 111 0,006401 0,069274 -0,15049 0,2088 CHBARC 111 0,005144 0,067436 -0,27485 0,125828 CHBNP 111 0,007975 0,070548 -0,30772 0,224594 CHCOMM 111 0,004106 0,067092 -0,38125 0,283145 CHCSUI 111 0,009466 0,051418 -0,17213 0,26 CHDEU 111 0,002532 0,067177 -0,27111 0,270718 CHHSBC 111 0,006812 0,058013 -0,17332 0,199041 CHING 111 0,001444 0,058759 -0,19061 0,228669 CHLIBOIS 111 -0,00414 0,033519 -0,08205 0,172431 CHRABO 111 0,007014 0,052849 -0,17488 0,174181 CHSG 111 0,004279 0,055096 -0,17668 0,212963 CHUBS 111 0,004885 0,041746 -0,17454 0,094535

(35)

35 CHAVGCDS 111 0,005445 0,049101 -0,179582 0.088442

Stage 2:

Variable #Obs Mean Std. Dev. Min Max

CHAIB 183 -0,00385 0,036638 -0,15789 0,131553 CHBANSA 183 -0,00223 0,037062 -0,17127 0,13118 CHBARC 183 -0,00469 0,035075 -0,11111 0,18469 CHBNP 183 -0,00289 0,03924 -0,14051 0,153711 CHCOMM 183 -0,00271 0,033079 -0,13858 0,123991 CHCSUI 183 -0,00603 0,035866 -0,18542 0,116839 CHDEU 183 -0,00219 0,038115 -0,16685 0,130952 CHHSBC 183 -0,00491 0,03138 -0,125 0,063621 CHING 183 -0,00345 0,034594 -0,16 0,153846 CHLIBOIS 183 -0,00625 0,035224 -0,13592 0,097938 CHRABO 183 -0,00498 0,029868 -0,14442 0,084746 CHSG 183 -0,00179 0,028822 -0,11028 0,091503 CHUBS 183 -0,00655 0,032096 -0,18974 0,077334 CHAVGCDS 183 -0,003569 0,028189 -0,105598 0,076914 Stage 3:

CHAIB 125 0,003145 0,044205 -0,25979 0,163983 CHBANSA 125 0,007276 0,064183 -0,34156 0,205705 CHBARC 125 0,005855 0,055444 -0,24493 0,173663 CHBNP 125 0,006455 0,056078 -0,24621 0,218175 CHCOMM 125 0,005904 0,050475 -0,25667 0,145954 CHCSUI 125 0,005989 0,052443 -0,27348 0,179117 CHDEU 125 0,006283 0,053808 -0,26104 0,188742 CHHSBC 125 0,004722 0,041747 -0,1677 0,144962 CHING 125 0,004836 0,04202 -0,14474 0,147004 CHLIBOIS 125 0,000246 0,029937 -0,07202 0,173028 CHRABO 125 0,002764 0,036427 -0,16345 0,114209 CHSG 125 0,005797 0,05657 -0,30177 0,222401 CHUBS 125 0,004531 0,043582 -0,14975 0,181762 CHAVGCDS 125 0,005713 0,045424 -0,22478 0,118657 Stage 4:

CHAIB 668 0,0024554 0,029512 -0,22936 0,296

CHBANSA 668 0,0013618 0,036742 -0,18058 0,155341

(36)

36 CHBNP 668 0,001239 0,044362 -0,13279 0,202053 CHCOMM 668 0,0011917 0,044897 -0,19024 0,585508 CHCSUI 668 0,0000122 0,034644 -0,15334 0,197448 CHDEU 668 0,0000903 0,042959 -0,15544 0,281143 CHHSBC 668 -0,0000126 0,029431 -0,10765 0,168791 CHING 668 0,000305 0,030984 -0,16278 0,125777 CHLIBOIS 668 0,0001362 0,066534 -0,30899 0,366976 CHRABO 668 -0,0001319 0,027444 -0,11938 0,125067 CHSG 668 0,0011028 0,041584 -0,1304 0,234516 CHUBS 668 -0,0001418 0,033704 -0,1347 0,16984 CHAVGCDS 668 0,0005631 0,0323439 -0,12147 0,17422 Stage 5:

CHAIB 547 -0,00243 0,039402 -0,86012 0,163853 CHBANSA 547 -0,00197 0,034439 -0,09994 0,17175 CHBARC 547 -0,00163 0,030496 -0,10648 0,112946 CHBNP 547 -0,00111 0,034761 -0,11637 0,179357 CHCOMM 547 -0,00123 0,026542 -0,10703 0,103821 CHCSUI 547 -0,00079 0,027789 -0,10329 0,134371 CHDEU 547 -0,00053 0,029503 -0,10994 0,124622 CHHSBC 547 -0,00033 0,035721 -0,14297 0,150177 CHING 547 -0,00102 0,029062 -0,12651 0,129602 CHLIBOIS 547 0,007571 0,146364 -0,75579 2,791627 CHRABO 547 -0,00056 0,027812 -0,10274 0,17966 CHSG 547 -0,00127 0,030852 -0,11307 0,141828 CHUBS 547 -0,00117 0,028883 -0,12508 0,159629 CHAVGCDS 547 -0,00110 0,027354 -0,08816 0,111073 Stage 6:

CHAIB 274 0,001103 0,012115 -0,05691 0,101137 CHBANSA 274 0,002973 0,0382 -0,18329 0,161965 CHBARC 274 0,005011 0,040277 -0,13987 0,236583 CHBNP 274 0,002077 0,039521 -0,16463 0,164404 CHCOMM 274 0,002479 0,034743 -0,21148 0,12538 CHCSUI 274 0,004572 0,032616 -0,11938 0,157098 CHDEU 274 0,005099 0,041413 -0,21649 0,177516 CHHSBC 274 0,003884 0,037305 -0,15708 0,162604 CHING 274 0,001563 0,034401 -0,15514 0,191318 CHLIBOIS 274 0,00104 0,037523 -0,2761 0,116991 CHRABO 274 0,002443 0,037278 -0,15154 0,225371 CHSG 274 0,001477 0,039125 -0,15387 0,147008 CHUBS 274 0,002982 0,038226 -0,17364 0,187468

(37)

37 CHAVGCDS 274 0,003203 0,0346525 -0,16209 0,148257

Stage 7:

CHAIB 190 -0,0005 0,035686 -0,22488 0,220414 CHBANSA 190 1,12E-05 0,035004 -0,07118 0,24393 CHBARC 190 -0,00183 0,051369 -0,13594 0,542248 CHBNP 190 0,000668 0,032839 -0,08455 0,246146 CHCOMM 190 0,000666 0,031612 -0,07912 0,2494 CHCSUI 190 -0,00095 0,027431 -0,07494 0,22883 CHDEU 190 -0,00023 0,036934 -0,09569 0,22276 CHHSBC 190 -0,00182 0,039881 -0,10386 0,32649 CHING 190 0,000265 0,035261 -0,07911 0,322828 CHLIBOIS 190 -0,00752 0,085988 -0,36935 0,456607 CHRABO 190 0,000168 0,036504 -0,07604 0,337593 CHSG 190 0,00048 0,035626 -0,09469 0,270288 CHUBS 190 -0,00065 0,032741 -0,10163 0,212641 CHAVGCDS 190 -0,000302 0,030207 -0,07358 0,160277

Lead-Lag relationship between Libar-ois spread and CDS spread

UNIVERSITY OF AMSTERDAM

EXECUTIVE MASTER’S IN INTERNATIONAL FINANCE

MASTER THESIS 2018

LEAD-LAG RELATIONSHIP BETWEEN LIBOR-OIS

SPREAD AND CDS SPREAD

POORNIMA ANANTH

STUDENT NUMBER : 11933003

ABSTRACT

Table of Contents

INTRODUCTION

THEORETICAL FRAMEWORK

LITERATURE REVIEW

DATA AND REGRESSION METHODOLOGY

RESULTS

LIMITATION & POTENTIAL IMPROVEMENT OF CURRENT

RESEARCH

CONCLUSION

REFERENCES

APPENDIX 1 : Stage-wise Descriptive Statistics