+ All Categories
Home > Documents > credit default swap market

credit default swap market

Date post: 24-Jan-2016
Category:
Upload: testetetetetet22
View: 24 times
Download: 0 times
Share this document with a friend
Description:
Trade matching forms in the credit default swapmarket
Popular Tags:
39
Electronic copy available at: http://ssrn.com/abstract=2665511 How trade matching forms in the credit default swap market * Jun Sung Kim Bonsoo Koo Zijun Liu § September 2015 Abstract We investigate the pairing of dealers and customers in credit default swap (CDS) transactions. Specifically, we analyse how a participant in a CDS transaction de- termines its trading partner in a matching game frameworks. Using comprehensive transaction reports from the Depository Trust & Clearing Corporation (DTCC), we show that the type and size of participating organisations, the degree of participation and intermediation in CDS transactions and counterparty risk are all important fac- tors for trade matching in the UK CDS market. We find that different types of market participants have different trade matching payoffs. For example, unlike other institu- tions, hedge funds prefer trading with risky counterparties during some periods. This finding is significant for policy-makers because such incentives can potentially lead to contagion risk. JEL: G10, G20, C13, D85 Keywords: Credit default swap; Matching; Counterparty risk; Financial network * The authors would like to thank Evangelos Benos, Yalin Gündüz, Terence Johnson, Michalis Vasios, and Zhuo Zhong for helpful conversations. The authors also thank all seminar audiences at the 2015 EEA Congress, the 7th International IFABS Conference, the IAAE 2015 Annual Conference, and Monash-Warwick Workshop for comments and suggestions. This paper is partially supported by the Monash Warwick Alliance Fund. The views expressed in this paper are the responsibility of the authors only and should not be representing those of the Bank of England. Monash University, Econometrics and Business Statistics, e-mail: [email protected] Monash University, Econometrics and Business Statistics, e-mail: [email protected] § Bank of England, e-mail: [email protected] 1
Transcript
Page 1: credit default swap market

Electronic copy available at: http://ssrn.com/abstract=2665511

How trade matching forms in the credit default swap

market∗

Jun Sung Kim† Bonsoo Koo‡ Zijun Liu§

September 2015

Abstract

We investigate the pairing of dealers and customers in credit default swap (CDS)transactions. Specifically, we analyse how a participant in a CDS transaction de-termines its trading partner in a matching game frameworks. Using comprehensivetransaction reports from the Depository Trust & Clearing Corporation (DTCC), weshow that the type and size of participating organisations, the degree of participationand intermediation in CDS transactions and counterparty risk are all important fac-tors for trade matching in the UK CDS market. We find that different types of marketparticipants have different trade matching payoffs. For example, unlike other institu-tions, hedge funds prefer trading with risky counterparties during some periods. Thisfinding is significant for policy-makers because such incentives can potentially lead tocontagion risk.JEL: G10, G20, C13, D85Keywords: Credit default swap; Matching; Counterparty risk; Financial network

∗The authors would like to thank Evangelos Benos, Yalin Gündüz, Terence Johnson, Michalis Vasios,and Zhuo Zhong for helpful conversations. The authors also thank all seminar audiences at the 2015 EEACongress, the 7th International IFABS Conference, the IAAE 2015 Annual Conference, and Monash-WarwickWorkshop for comments and suggestions. This paper is partially supported by the Monash Warwick AllianceFund. The views expressed in this paper are the responsibility of the authors only and should not berepresenting those of the Bank of England.†Monash University, Econometrics and Business Statistics, e-mail: [email protected]‡Monash University, Econometrics and Business Statistics, e-mail: [email protected]§Bank of England, e-mail: [email protected]

1

Page 2: credit default swap market

Electronic copy available at: http://ssrn.com/abstract=2665511

1 Introduction

The credit default swap (CDS) market allows participants to speculate on or hedge against

credit risk. There are primarily two types of participants: dealers whose main purpose

is to serve as intermediaries in the market and customers who buy or sell risk protection

depending on their trading motives. Similar to other over-the-counter (OTC) derivatives,

CDS transactions give rise to counterparty credit risk and therefore the choice of trading

partner matters. This paper studies how dealers and customers are paired in the CDS market

and how counterparty risk is related to this pairing.

In the last two decades, the theoretical literature on the CDS market has expanded sig-

nificantly. See Augustin, Subrahmanyam, Tang, and Wang (2014) for an extensive survey

on CDS-related papers. In contrast, Arora, Gandhi, and Longstaff (2012) note that the

empirical literature, particularly empirical studies on how CDS transactions take place be-

tween market participants, is limited. This is a reflection of limited data availability and the

opaque nature of the CDS market that was more pronounced prior to the financial crisis.

Recently, however, improved market transparency and data accessibility have allowed

papers such as Alexander and Kaeck (2008), Forte and Pena (2009), Arora et al. (2012) and

Benos, Wetherilt, and Zikes (2013), to focus the on empirical aspects of the CDS market (e.g.

the treatment of credit risk in CDS pricing). The extant literature sheds light on several

unique features (stylised facts) of the CDS market (see Section 2 for more details) and CDS

spreads, see Corò et al. (2013) and Galil et al. (2014) among others. However, to the best

of our knowledge, the trading partner choice of CDS market participants has not yet been

investigated. This type of analysis requires observations of the entire structure of the CDS

network and the underlying characteristics of CDS market participants.

Understanding the drivers behind trade matching between CDS market participants is

important because it is in the interests of policy-makers to maintain a well-functioning CDS

market. For this reason, a number of regulatory reforms have been introduced since the

2

Page 3: credit default swap market

crisis as described in Vause (2010). If market participants do not pay sufficient attention

to the credit worthiness of their counterparties, this may lead to the accumulation of po-

tential contagion risk. Considering this, we analyse the trade matching decisions in the UK

CDS market based on transaction-level data obtained from the DTCC. We define the UK

market as the market of CDS contracts with UK-based counterparties (see Section 2 for an

explanation).

We employ a matching game approach as in Fox (2010b) because any bilateral relationship

between a buyer and a seller of credit risk protection can be considered as a match. CDS

transactions can be viewed as a typical example of two-sided, many-to-many matching games.

We assume that realised CDS transactions between participants are equilibrium outcomes to

a competitive market and these outcomes fully reflect participants’ preferences. Hence, each

pair of participants does not want to deviate from their current trading partners. Section 3

provides more details on our methodology.

Our findings are summarised as follows. First, we find that CDS market participants

receive positive benefits when trading with counterparties of larger size, except for pairs

between dealers and asset managers, and when trading with counterparties that have large

gross CDS positions for all pair types. This is consistent with the intuition that CDS market

participants with larger assets or gross positions may be considered less risky or able to offer

lower transaction costs.

Second, we find that CDS market participants trade with less risky counterparties, except

for pairs between dealers and hedge funds. Hedge funds may prefer trading with more risky

dealers because hedge funds and risky dealers tend to place less weight on counterparty risk

when trading CDS contracts. Finally, CDS market participants trade with counterparties

that are more involved in intermediation (i.e. they hold a large gross position relative to the

net position), which is likely to be driven by transaction cost factors.

The finding that hedge funds prefer trading with risky dealers has significant policy

3

Page 4: credit default swap market

implications. If risky dealers and risky hedge funds tend to trade with each other, CDS

positions could build between them to the extent that a shock to one vulnerable organisation

could spread to other vulnerable organisations in the network. Such interconnectedness

between vulnerable organisations increases systemic risk.

In the remainder of this paper, Section 2 describes the CDS market and the datasets.

Section 3 explains our methodology for estimating matching models. The empirical results

and their implications are discussed in Sections 4 and 5, and the final section concludes. The

formal model and derivations are presented in the Appendix.

2 Data

2.1 A brief description of the CDS market

Credit default swaps (CDS) are swap contracts in which the buyer makes a series of payments

to the seller and receives a payoff if the underlying reference entity suffers a credit event, for

example, default. By purchasing CDS contracts, the buyer effectively protects against any

counterparty credit risk arising from exposure to the underlying reference entity. However,

the seller of CDS contracts obtains a synthetic credit exposure to the reference entity without

having to hold bonds directly.

Since the trade of the first CDS contract initiated by JP Morgan in 1994, two decades

have passed. During this period, the CDS market grew rapidly prior to the financial crisis

to approximately $58.2 trillion at the end of 2007, according to the Bank for International

Settlement (BIS), before falling to $16.4 trillion at the end of 20141. However, before the

crisis, there are few empirical studies on the structure of the CDS market, largely because

of data constraints. Recently, increasing literature has focused on the CDS market, such as1Much of the reduction in gross positions post-crisis may be caused by compression trades as explained

in Benos et al. (2013).

4

Page 5: credit default swap market

Arora et al. (2012), Benos et al. (2013), and Brunnermeier, Clerc, and Scheicher (2013). We

introduce some features and stylised facts on the CDS market identified in those papers that

are particularly relevant to our study.

First, dealers are the most important players in the CDS market. Dealers are financial

intermediaries, typically large investment banks that facilitate CDS transactions between

customers (non-dealers). According to Benos et al. (2013), 99% of transactions in the UK

single-name CDS market involve at least one dealer. Moreover, CDS market activity is

concentrated in a small number of dealers. Brunnermeier et al. (2013) find that the ten most

active dealers accounted for 73% of gross CDS volumes based on a global sample of single-

name CDS transactions. Atkeson et al. (2013) argue that dealers exist mainly to provide

intermediation services while customers participate in the CDS transactions mainly to share

risks. Atkeson et al. (2013) also argue that this heterogeneity in participants manifests as

the difference between gross and net volumes of CDS market trade. The authors find that

dealers have large gross positions but small net positions (on a relative basis), whereas the

gross and net positions of customers are close to each other.

Second, CDS market customers are heterogeneous. Typically, an institution that is not

a dealer may trade CDS contracts for two purposes: hedging an existing credit exposure or

executing a speculative trading strategy (e.g. taking a view on the credit quality of a given

reference entity or an arbitrage between different financial instruments). Banks, hedge funds

and asset managers are the largest non-dealer participants in the CDS market, according

to the findings of Brunnermeier et al. (2013) and Benos et al. (2013). Different types of

customers may trade CDS for different purposes. For example, both papers find that asset

managers, which typically have zero or low leverage, tend to be net buyers of CDS protection,

consistent with their hedging motive. Interestingly, hedge funds are also net buyers of CDS

protection according to the two papers above, which suggests that the majority of hedge

funds adopt a short view on credit during the sample period. Hedge funds in our sample

are also net CDS buyers (see Table 1). However, asset managers are net CDS sellers in our

5

Page 6: credit default swap market

sample, mostly driven by one outlier.

*****Table 1 Here *****

Third, counterparty risk plays an important role in the CDS market. As Arora et al.

(2012) note, counterparty credit risk is significantly priced in the CDS market, particularly

since the financial crisis. This implies that the less creditworthy dealers charge lower spreads

when selling CDS protection. Although Arora et al. (2012) focus on the credit quality of

dealers, counterparty credit risk concerns should go both ways, that is, dealers may charge

higher spreads when trading with less creditworthy counterparties. This paper studies the

trade matching between CDS market participants in the presence of counterparty credit

risk concerns. When trading CDS contracts, market participants may be inclined to choose

partners that reduce their counterparty risk. In such cases, observable characteristics of

trading partners, such as total assets, could be significant factors as suggested in Atkeson

et al. (2013). Section 3 provides more details on the determinants of participants’ payoff.

2.2 DTCC data

Prior to 2008, the CDS market was over-the-counter, largely unregulated, and opaque. How-

ever, a series of regulatory reforms have been implemented since the crisis2, and the trans-

parency of the CDS market has improved considerably, for example, by the data provided by

the Trade Information Warehouse (TIW) at the Depository Trust & Clearing Corporation

(DTCC). DTCC is a US post-trade financial services company providing clearing and set-

tlement services to the financial markets. The TIW was established in 2006 and became the

market’s first and only centralised global repository for trade reporting of CDS contracts.

Since the crisis, the regulators have worked together with the industry to expand the cov-2First, major changes have occurred in the settlement process of CDS contracts, known as “Big Bang” and

“Small Bang” in 2009. Second, global regulators have agreed that standardised OTC derivatives, includingCDS, should be subject to mandatory central clearing. Other broad-ranging regulatory initiatives, suchBasel III and the Dodd-Frank Act, also have implications for the CDS market.

6

Page 7: credit default swap market

erage of the DTCC, and 98% of outstanding CDS contracts worldwide are now recorded in

TIW.

Most recent empirical literature on the CDS market is based on DTCC data. For instance,

Chen, Fleming, Jackson, Li, and Sarkar (2011) analyse the CDS market composition, trading

dynamics and level of standardisation based on DTCC data and provide a framework for

improvements in the design of public reporting and data collection. Benos et al. (2013) study

a subset of DTCC data, which includes all transactions on single-name CDS contracts on

the 126 most heavily traded UK reference entities from January 2007 to December 2011.

The authors’ primary interest is to understand the structure and dynamics of the UK single-

name CDS market. Brunnermeier et al. (2013) use DTCC data to construct a network of

counterparties’ bilateral notional CDS exposures to 642 sovereign and financial reference

entities to analyse the network structure of the CDS market.

Similarly, our dataset is obtained from the DTCC and covers outstanding CDS trans-

actions of nine major UK dealers, which accounts for over 95% of transactions involving

UK-regulated entities (i.e. the UK market). The UK market accounts for approximately

50% of outstanding CDS transactions globally. Therefore, our dataset covers nearly half of

all transactions worldwide, although transactions taking place between non-UK regulated

entities are not captured in our dataset. We consider that the incomplete coverage of our

dataset should not lead to a significant bias in the trade-matching decisions, assuming that

UK-regulated entities tend to trade with UK-regulated entities.

Note that the ‘UK CDS market’ can be interpreted in different ways, such as the market

of CDS contracts with UK-based reference entities or the market of CDS contracts with

UK-based counterparties. Our paper uses the latter, that is, we study all CDS contracts

with at least one UK-based counterparty for the following reasons. First, this represents the

largest sample we could obtain because UK authorities do not have access to CDS contracts

data between non-UK counterparties. Second, this paper focuses on the relationship be-

7

Page 8: credit default swap market

tween trade matching and institutional characteristics, such as CDS positions. By limiting

the scope to CDS contracts with UK-based reference entities only, we cannot fully capture

the institutions’ CDS positions. Third, CDS transactions with UK-based reference entities

account for less than 1% of our sample in terms of gross notional and, therefore, may not be

representative of the wider market.

Our dataset consists of CDS transactions by 1,357 counterparties (including nine dealers)

on approximately 1,500 reference entities from the first half of 2012 (2012H1) to the second

half of 2014 (2014H2). We exclude duplicate trades (where both counterparties reported)

and compression trades (which do not have a genuine economic purpose - see Benos et al.

(2013)). Unlike Benos et al. (2013) and Brunnermeier et al. (2013), our dataset includes both

single-name and index transactions (excluding index tranches). Table 2 shows additional

details.

Market participants are broadly divided into dealers and customers. Every transaction

in our dataset involves at least one of the nine major dealers who are at the core of the

CDS trading network. The CDS network has a core-periphery structure, where links are

highly concentrated among a few dealers, while other market participants have limited links.

Figures 1 and 2 show the core-and-periphery structure of the CDS trading network and the

degree distributions of dealers and customers in 2014H2, respectively.

*****Figure 1 Here *****

*****Figure 2 Here *****

*****Table 2 Here *****

2.3 Organisation characteristics

We classify non-dealer market participants into four categories: banks, asset managers, hedge

funds and others, following Benos et al. (2013). Our classification differs only in that we

8

Page 9: credit default swap market

combined insurance companies with other types of organisations because the number of

CDS trades by insurance companies was too low to be independently categorised. Table 1

shows ‘other’ only accounts for less than 5% of total transaction volumes in our dataset.

Additionally, dealers and banks have relatively low net transaction volumes (the absolute

difference between buy volumes and sell volumes) while asset managers have large positive

net volumes (net CDS buyer) and hedge funds have large negative net volumes (net CDS

seller).

We also collect data on organisations’ total assets from Capital IQ. Among the 1357

institutions in our dataset, data on total assets were available for 564 institutions. The

majority of institutions with missing total assets are asset managers and hedge funds because

of a lack of disclosure by these types of institutions. Other characteristics such as capital

ratio and return on equity were available for banks, but were not included in our analysis

because they were not available for other entity types. Table 3 shows that dealers and banks

have the largest total assets in our sample while other entity types are relatively small in

size.

*****Table 3 Here *****

3 Methodology

Our methodology is based on matching games, which has wide applications across different

disciplines.3 For example, Gordon and Knight (2009) study teacher to school matching.

Fox (2010a) investigates automotive assembler car parts portfolios. Yang, Shi, and Goldfarb

(2009) estimate the value of alliances between professional sports teams and players. Baccara,

İmrohoroğlu, Wilson, and Yariv (2012) investigate the matching of faculty members to office3Empirical models of matching games were first investigated by Choo and Siow (2006) who study one-

to-one matching with transferable utility and provide an estimator that exploits a logistic distribution as-sumption on error terms. The authors apply the model to the US marriage market to investigate changes inmatching patterns.

9

Page 10: credit default swap market

buildings. Sørensen (2007) studies matching between venture capitalists and entrepreneurs

in the finance field, and Chen (2013) investigates the loan spread equation with endogenous

matching between banks and firms.

To the best of our knowledge, this paper is the first to apply matching games to the CDS

market. Our approach follows Fox (2010b), which studies what economic parameters can be

identified from matching data. Assuming the matching outcome satisfies pairwise stability,

the author shows that features of the sum of production functions are identified. His model

can be considered a close benchmark of our empirical framework.

3.1 Dealer and customer matching

We analyse the CDS market under a framework of a two-sided, many-to-many matching

game with transfers. One side is a set of dealers, typically large investment banks, and the

other side is a set of customers including non-dealer banks, asset managers, hedge funds,

and other entities. A customer chooses a dealer to buy or sell CDS protection and vice

versa. A match is a pair consisting of a dealer and a customer who make one or more CDS

transactions with each other. There may be multiple transactions in both directions (buy or

sell) between each pair. However, we simplify the model by focusing on the binary pairing

decision.

Because dealers trade CDS with each other, the CDS market is potentially more com-

plicated than two-sided matching. Although a general form of network can be constructed

among multiple organisations, we choose a two-sided matching model for a number of reasons.

First, dealers and customers have different characteristics. Dealers are typically investment

banks while customers exhibit strong heterogeneity both in organisation type and size. Sec-

ond, dealers and customers have different trading motives. Customers are actively seeking

trading partners to transfer risk while dealers have relatively matched books and profit on

the bid-ask spread. Dealers tend to have a small net CDS position relative to the gross

10

Page 11: credit default swap market

position while customers tend to have a net position that approximates the gross position.

Additionally, our estimation procedure involves hypothetical trading partner exchanges.

The inter-dealer CDS network is complete and, therefore, cannot be estimated using our

approach. Nonetheless, inter-dealer transactions play an important role in the CDS market,

and their pay-off functions may be estimated using other approaches, such as one based on

weighted networks, which we leave for future research.

3.2 Pairwise stable matching

Pairwise stability is a necessary condition for a matching equilibrium, which is the foun-

dation of our matching game estimation. Specifically, we assume that the observed trade

matching allocation is an equilibrium outcome of the game, which satisfies pairwise stabil-

ity with transfers defined by Fox (2010b). Each participant in the CDS market maximises

payoff by choosing appropriate trading partners. Payoff maximisation for each participant

is manifested in the form of the observed trade matching in the CDS market.

In the CDS market, the payoff of each participant may consist of three components: the

counterparty risk associated with the chosen trading partner, the cost associated with the

transaction, and the value of the future monetary transfers. The usual assumption of a quasi-

linear form of monetary transfers implies that the transfer component will be cancelled out in

a pair and, therefore, we can focus on the counterparty risk component and the transaction

cost component.

Let R(di, cj) be the matching payoff, that is, the sum of match-specific payoffs to dealer

i and customer j from their match, where di, i = 1, · · · , ND. and cj, j = 1, · · · , NC are

indices for dealers and customers, respectively. The matching payoff may be heterogeneous

with respect to pair types, for example. dealer-bank, dealer-asset manager, dealer-hedge fund

and dealer-other. Note that in this matching payoff, the transfer between two participants

in a match is cancelled out. This is a desirable feature of the model because we do not

11

Page 12: credit default swap market

need to consider the transfer information that is incompletely observed and endogenously

determined in a matching equilibrium.

*****Figure 3 Here *****

A trade match is pairwise stable if the following conditions hold. First, for any two

pairs of current trade matches, no pairs are willing to exchange their partners. Second,

no pairs of organisations want to cut their current trade link. Third, no pair is willing

to build a new trade link. From the conditions of pairwise stability, we derive a payoff

inequality whereby the total payoffs to any two observed pairs are at least as large as the

total payoffs when the partners are exchanged. Fox (2010b) refers to this as the local

production maximisation inequality. In the CDS market, each organisation chooses their

trading partners to maximise payoff or to minimise risk. Hence, we call this the local payoff

maximisation (LPM) inequality.

Figure 3 illustrates the pairwise stable matching and construction of the LPM inequality.

Current match allocation A consists of a match between Dealer 1 and Customer 1, and

another match between dealer 2 and customer 2. If partners are exchange, the new allocation

A′ would have a match between dealer 1 and customer 2 and a match between dealer 2

and customer 1. The matching payoffs to new pairs from the exchange of pairs should

not be greater than the matching payoffs to the original pairs from the current allocation.

Otherwise, the currently observed match(es) would be broken and new matches will be

formed. Consequently, the researcher could not observe the current matches.

With the matching payoff, we formalise the local payoff maximisation inequality as fol-

lows. When di is matched to cj, and dk is matched to ch, but neither di and ch nor dk and

cj are paired,

R(di, cj) +R(dk, ch) ≥ R(di, ch) +R(dk, cj). (3.1)

The inequality (3.1) implies that if current two pairs exchange their partners, the matching

12

Page 13: credit default swap market

payoff of all four organisations with this exchange does not exceed the matching payoff with

the original pairing. For example, Figure 3 describes the local payoff maximisation inequality

when two currently matched pairs, <di, c1>, <d2,c2> exchange their partners. That is,

R(d1, c1) +R(d2, c2) ≥ R(d1, c2) +R(d2, c1). (3.2)

For estimation, we construct this LPM inequality for many randomly selected exchanges of

pairs.

3.3 Payoff

We allow for heterogeneity in the matching payoff, R(di, cj) across different pair types. Deal-

ers are homogeneous except for their observed characteristics, and customers are classified

into four types: banks, asset managers, hedge funds, and others. Hence, there are four

different pair types: dealer-bank (DB), dealer-asset manger (DA), dealer-hedge fund (DH),

and dealer-other (DO). Figure 4 shows the snapshot of trading volumes by pair types in

2014H2, and Table 4 shows the established percentage of links by pair types throughout our

sample periods. We use superscript s to denote heterogeneity in the payoff parameters. For

example, s = DB if a given pair between a dealer and a bank are matched. We specify the

matching payoff R(di, cj) as follows.

R(di, cj) = βs1assetsdi

× assetscj+ βs

2grossdi× grosscj

+βs3netdk

× netcj+ βs

4ratiodk× ratiocj

+ εdicj, (3.3)

where εdicjis a match-specific unobserved variable. Note that if dealer i and customer

j have a transaction in another over-the-counter market, then εdicjcaptures this type of

unobserved heterogeneity. The matching payoff only displays the interaction terms, or the

products of two organisations’ characteristics. This is because the intercept and level terms

13

Page 14: credit default swap market

will be cancelled out in each LPM inequality. For a similar reason, the matching payoff can

accommodate individual and country fixed effects.

*****Figure 4 Here *****

*****Table 4 Here *****

3.4 Explanatory variables for CDS matching analysis

We briefly discuss the interpretation of our explanatory variables for the estimation of CDS

matching. The variable assets measures the size of an organisation. If the coefficient asso-

ciated with this variable is positive for a certain pair type, we infer that the larger the size

of the organisations in this pair type, the higher the matching payoff. That is, a participant

finds the bigger trading partner more attractive, ceteris paribus. Intuitively, larger organi-

sations may be considered less risky or able to offer lower transactions costs4, which would

lead to a higher matching payoff in both cases.

The variable gross is the gross notional CDS position of an organisation regardless of the

direction of trading and represents the presence of an organisation within the CDS market.

If the coefficient associated with the gross variable for a certain pair type is positive, we

infer that the more active the organisations are in the CDS market, the higher the matching

payoff. That is, a participant finds a partner who trades more actively in the CDS market

more attractive. Again, we expect an organisation with a large gross CDS position to have

low risk or transaction costs.

The variable net represents the (absolute value of the) difference between CDS contracts

bought and sold by an organisation. This is interpreted as the degree of the riskiness of the

organisation (or the degree of counterparty risk faced by its trading partner) because the net4Transaction cost in OTC markets may take many forms (see Duffie, Gârleanu, and Pedersen (2005)).

For example, larger organisations may have lower search costs or financing costs from economies of scale.

14

Page 15: credit default swap market

CDS position reflects the level of market risk assumed by each organisation. If the coefficient

associated with this variable is negative, it implies that the more risky the counterparties,

the lower the matching payoff. We expect the net variable to be associated with riskiness

only (not transaction costs).

Note that the net variable is only an approximate proxy for counterparty risk for two

reasons. First, market participants may have bought CDS protection for one reference entity

but sold CDS protection for another. This leads to a small net position but not necessarily a

small market risk given that the two reference entities may not be perfectly correlated with

each other. Second, there are many other factors to consider when assessing counterparty

risk including collateral arrangements and counterparty risk management. Despite these

caveats, we consider the net variable a reasonable proxy for counterparty risk at least within

the same type of organisations.

Finally, the variable ratio is the ratio of net CDS position to gross CDS position of an

organisation. For intermediaries in the CDS market, this ratio should be small and close

to zero. Thus, we interpret this variable as the degree of intermediation of an organisation.

Although dealers in our dataset are primarily intermediaries, they may still hold an ‘inven-

tory’ of net position if they fail to find customers willing to trade (in the opposite direction

of dealers’ net position), or if they are trading CDS for their own purposes. Dealers that

specialise in intermediation may have lower transaction costs because of dealer expertise and

customer networks. If the coefficient associated with ratio is positive, this indicates that the

more the organisations resemble intermediaries, the higher the matching payoff. We expect

the ratio variable to be associated with transaction costs only (not riskiness, which will be

captured by the net variable).

15

Page 16: credit default swap market

3.5 Matching maximum score estimation

We apply the specification (3.3) to the LPM inequality (3.1). Then, we have the deterministic

part of the LPM inequality as follows.

βs1x1,dicjdkch

+ βs2x2,dicjdkch

+ βs3x3,dicjdkch

+ βs4x4,dicjdkch

≥ 0, (3.4)

where

x1,dicjdkch= assetsdi

(assetscj− assetsch

) + assetsdk(assetsch

− assetscj),

x2,dicjdkch= grossdi

(grosscj− grossch

) + grossdk(grossch

− grosscj),

x3,dicjdkch= netdi

(netcj− netch

) + netdk(netch

− netcj),

x4,dicjdkch= ratiodi

(ratiocj− ratioch

) + ratiodk(ratioch

− ratiocj).

For simplicity, we let m = 1, · · · ,M be an index for each exchange of pairs. Then, the

deterministic part of the LPM inequality can be written as x′mβ ≥ 0, where xm is a vector of

organisation characteristics for each exchange of pairs as described. We apply the maximum

score estimation method first proposed by Manski (1975) and extended by Fox (2010a) to

matching models, and find the payoff function that maximises the total number of inequalities

satisfied. The objective function can be written as

QM(β) = 1M

∑m

1 [x′mβ ≥ 0] , (3.5)

where 1[·] is the indicator function that takes one when the inequality in the bracket is

satisfied and zero otherwise. The maximum score estimator is consistent under certain

conditions which are satisfied in our model. Appendix A.5 shows the detailed explanations

on the identification and consistent estimation.

16

Page 17: credit default swap market

4 Results

This section discusses the results of our estimation for our CDS matching model. Note that

the model has two features. First, the model is based on qualitative outcomes of equilibrium

matches and the parameters are identified up to positive monotonic transformation. Hence,

it is natural to interpret the results by investigating signs and ordinal ranking of parameters

rather than their cardinal values. Second, we estimate the matching payoff to a matched

pair where each explanatory variable is an interaction term. Therefore the results should be

interpreted from the perspective of both dealers and customers.

4.1 Estimation results and implications

Section 2 explained that the data covers CDS transactions of nine major UK-based deal-

ers between the years 2012 and 2014. We aggregate the transactions to six data points at

six-monthly frequencies and estimate the model for each data point. The gross and net posi-

tions of organisations, which are explanatory variables, are estimated based on outstanding

positions rather than new transactions. We choose a six-monthly frequency because this is

a reasonable length of time during which material changes in market participants’ trading

behaviour could occur.

We run our estimation procedure assuming homogeneity in the matching payoff across

different pair types. Table 5 shows the estimation results under this assumption. However,

acknowledging the presence of heterogeneity and its impact on the estimation results, we

also consider the case in which customers are divided into four groups: (non-dealer) bank,

asset manager, hedge fund and other. Table 6 shows our estimation results for each pair

type from 2012H1 to 2014H2 along with maximum scores. The maximum scores in Tables

5 and 6 show that the homogeneous model does not yield the best outcome because of the

presence of matching payoff heterogeneity across pair types.

17

Page 18: credit default swap market

*****Table 5 Here *****

*****Table 6 Here *****

Size Recall that the variable assets measures the size of an organisation, and a positive

coefficient indicates that larger organisations are more attractive, because of lower riskiness

or transaction costs (or both). Table 6 shows that coefficients on assets are positive and

significant for dealer-bank pairs except in 2013 and positive and significant for dealer-hedge

fund and dealer-other pairs except in 2012. In contrast, dealer-asset manager pairs tend to

have a negative coefficient on assets where it is significant. The reason that Dealer-Asset

manager pairs do not have a positive coefficient on assets may be that asset managers

are only operating on an agency basis with limited leverage and, therefore, do not create

significant counterparty risk. Overall, the results show that CDS market participants receive

positive benefits when trading with counterparties of larger size, although this may not hold

across all the periods.

Degree of participation The variable gross captures each organisation’s degree of par-

ticipation in the CDS market. Coefficients on the variable gross are positive for all pair

types where they are significant. This is consistent with the intuition that CDS market

participants receive higher payoffs when trading with counterparties that are more active

because participants with greater market presence are considered less risky or have lower

transaction costs. Therefore, gross is a more consistent indicator of riskiness compared to

assets, which may include business outside the CDS market.

Counterparty risk The variable net is a proxy for the degree of counterparty risk in

the CDS market. Table 6 shows that the coefficients on net are negative and strongly

significant for dealer-bank and dealer-other in all six periods. This indicates that CDS

market participants tend to trade with counterparties that are less risky, consistent with

18

Page 19: credit default swap market

the stylised fact that counterparty risk is an important consideration in the CDS market.

In contrast, dealer-hedge fund pairs receive higher payoff when their net variable is higher,

until 2014. This might suggest that hedge funds, which are speculative traders, do not place

significant weight on counterparty risk when trading in the CDS market (or even prefer more

risky dealers who are also likely to place a lower weight on counterparty risk; see Section 5

for further explanations). However, the coefficient on net for dealer-hedge fund pairs became

negative or insignificant in 2014. This may be explained by a number of factors, such as

regulatory reforms that constrained dealers’ ability to take counterparty risk.5 Finally, there

is no consistent and significant relationship between matching payoff and the net variable

for dealer-asset manager pairs, probably because asset managers do not cause significant

counterparty risk.

Degree of intermediation The variable ratio represents the degree of intermediation of

an organisation. Table 6 shows that the coefficient on ratio is almost always negative and

significant across all periods. The only exception is dealer-other in 2012. We cannot explain

this exception without granular information on the organisations included in the ‘other’

category. However, overall we have strong evidence that CDS market participants tend to

trade with intermediation-focused counterparties that are able to offer lower transaction

costs.

5 Policy implications

We stated that counterparty risk is an important consideration of CDS market participants.

However, there is also evidence (Arora et al. (2012)) that counterparty risk may not be

priced in CDS transactions with dealers that are too-big-to-fail. Since the crisis, central5For example, the Basel Committee on Banking Supervision published the final Basel III leverage ratio

framework in January 2014 and the final standard on the standardised approach for measuring counterpartycredit risk exposures in March 2014. See http://www.bis.org/bcbs/publications.htm.

19

Page 20: credit default swap market

counterparties have been introduced to reduce the potential risk of counterparty contagion

in the CDS (and other OTC derivatives) market, and bank regulations have been made more

stringent to ensure effective bank management of counterparty risk. Reassuringly, based on

our results, dealer-bank pairs tend to trade with counterparties that are less risky, reflecting

prudent risk manage practices.

However, we also find evidence that dealer-hedge fund pairs tended to trade with counter-

parties that are more risky, during the years 2012 and 2013. Among all types of customers,

hedge funds are considered the most risky and the least risk averse. Dealers with poor risk

management standards are likely to take more risks and have a high tolerance for counter-

party risk in the CDS market. Such dealers may be attracted to risky hedge funds because

they may find it more costly to trade with anyone else. This could explain the positive

coefficient on the net variable for dealer-hedge fund pairs. The coefficient became negative

or insignificant in 2014, possibly because of tighter regulatory constraints on dealers.

If risky dealers and risky hedge funds tend to trade with each other, CDS positions could

build between them to the extent that a shock to one vulnerable organisation could spread to

other vulnerable organisations in the network. Such interconnectedness between vulnerable

organisations increases systemic risk. Although the coefficient on the net variable for dealer-

hedge fund pairs was no longer positive in 2014, this should be monitored closely to prevent

accumulation of contagion risk in the CDS market.

6 Conclusion

We analyse how a participant in a CDS transaction determines its trading partner in the

framework of matching games, using comprehensive transaction reports from the DTCC. Our

findings show that the type and size of participating organisations, the degree of participation

and intermediation in CDS transactions and counterparty risk are all important factors

20

Page 21: credit default swap market

for trade matching in the UK CDS market. Particularly, we find that different types of

market participants have different trade matching payoff functions. For example, unlike

other institutions, hedge funds prefer trading with risky counterparties during some periods.

This is important for policy-makers because such incentives can potentially lead to contagion

risk.

Our study could be extended in many ways. For example, our methodology could be

applied to other OTC derivative markets where counterparty credit risk is present. Ad-

ditionally, estimating the model using a more refined classification of counterparties, or

different types of reference entities, could yield interesting results. We leave this to future

research.

21

Page 22: credit default swap market

References

Alexander, C., Kaeck, A., 2008. Regime dependent determinants of credit default swap

spreads. Journal of Banking & Finance 32 (6), 1008–1021.

Arora, N., Gandhi, P., Longstaff, F. A., 2012. Counterparty credit risk and the credit default

swap market. Journal of Financial Economics 103 (2), 280–293.

Atkeson, A. G., Eisfeldt, A. L., Weill, P.-O., 2013. The market for otc derivatives. Tech. rep.,

National Bureau of Economic Research.

Augustin, P., Subrahmanyam, M. G., Tang, D. Y., Wang, S. Q., 2014. Credit default swaps–a

survey. Regulations and Trends in Finance, Forthcoming.

Baccara, M., İmrohoroğlu, A., Wilson, A. J., Yariv, L., 2012. A field study on matching with

network externalities. American Economic Review 102 (5), 1773–1804.

Benos, E., Wetherilt, A., Zikes, F., 2013. The structure and dynamics of the uk credit default

swap market. Bank of England, Financial Stability Paper (25).

Brunnermeier, M., Clerc, L., Scheicher, M., 2013. Assessing contagion risks in the cds market.

FSR FINANCIAL, 123.

Chen, J., 2013. Estimation of the loan spread equation with endogenous bank-firm matching.

Advances in Econometrics 31, 251–290.

Chen, K., Fleming, M. J., Jackson, J. P., Li, A., Sarkar, A., 2011. An analysis of cds

transactions: Implications for public reporting. FRB of New York Staff Report (517).

Choo, E., Siow, A., 2006. Who marries whom and why. Journal of Political Economy 114 (1),

175–201.

Corò, F., Dufour, A., Varotto, S., 2013. Credit and liquidity components of corporate cds

spreads. Journal of Banking & Finance 37 (12), 5511–5525.

22

Page 23: credit default swap market

Duffie, D., Gârleanu, N., Pedersen, L. H., 2005. Over-the-counter markets. Econometrica

73 (6), 1815–1847.

Echenique, F., Oviedo, J., 2006. A theory of stability in many-to-many matching markets.

Theoretical Economics 1, 233–273.

Forte, S., Pena, J. I., 2009. Credit spreads: An empirical analysis on the informational

content of stocks, bonds, and cds. Journal of Banking & Finance 33 (11), 2013–2025.

Fox, J. T., 2010a. Estimating matching games with transfers. Tech. rep., National Bureau

of Economic Research.

Fox, J. T., 2010b. Identification in matching games. Quantitative Economics 1 (2), 203–254.

Galil, K., Shapir, O. M., Amiram, D., Ben-Zion, U., 2014. The determinants of cds spreads.

Journal of Banking & Finance 41, 271–282.

Gordon, N., Knight, B., 2009. A spatial merger estimator with an application to school

district consolidation. Journal of Public Economics 93 (5), 752–765.

Hatfield, J. W., Kominers, S. D., 2012. Matching in networks with bilateral contracts. Amer-

ican Economic Journal: Microeconomics 4 (1), 176–208.

Horowitz, J. L., 1992. A smoothed maximum score estimator for the binary response model.

Econometrica: journal of the Econometric Society, 505–531.

Manski, C., 1975. Maximum score estimation of the stochastic utility model of choice. Journal

of Econometrics 3 (3), 205–228.

Manski, C., 1985. Semiparametric analysis of discrete response:: Asymptotic properties of

the maximum score estimator. Journal of Econometrics 27 (3), 313–333.

Manski, C., 1988. Identification of binary response models. Journal of the American Statis-

tical Association 83 (403), 729–738.

23

Page 24: credit default swap market

Sørensen, M., 2007. How smart is smart money? a two-sided matching model of venture

capital. The Journal of Finance 62 (6), 2725–2762.

Vause, N., 2010. Counterparty risk and contract volumes in the credit default swap market.

BIS Quarterly Review, December.

Yang, Y., Shi, M., Goldfarb, A., 2009. Estimating the value of brand alliances in professional

team sports. Marketing Science 28 (6), 1095–1111.

24

Page 25: credit default swap market

Appendix

A Matching model

A.1 Set-up

Let {d1, · · · , dNd} and {c1, · · · , cNc} be sets of dealers and customers, respectively. Each

organisation knows all payoff-relevant information and chooses trading partners. Hence,

the model is of complete information. A dealer (a customer) receives payoff Πd (Πc) from

their matches. Each participant maximises payoff arising from matches. When dealer i and

customer j are matched, we denote the realised match as < di, cj >. Let Ndiand Ncj

be the

sets of current trading partners of di and cj, respectively. We use τdicjto denote the transfer

from di to cj, and τdito denote the sum of all transfers from di, i.e. τdi

= ∑cj∈Nd

τdi,cj.

Customer transfer τcjis similarly defined, i.e. τcj

= ∑di∈Ncj

τdi,cj. Then, the total payoffs to

dealer di and customer cj are written as

Πd(di,Ndi, tdi

) : = rd(di,Ndi)− τdi

, (A.1)

Πc(cj, Ncj, tcj

) : = rc(cj,Ndi) + τcj

, (A.2)

where πd(·) and πc(·) are the payoffs related to risk and transaction costs for a dealer and a

customer, respectively.

A.2 Pairwise stability with transfers

The set of all observed matches is an equilibrium outcome to a competitive market and is

used to estimate the matching payoff that both customers and dealers contribute to. The

observed matching assignment satisfies stability as a necessary condition for an equilibrium.

25

Page 26: credit default swap market

There are several ways to define stability in matching, especially many-to-many matching.6

For the CDS matching market, we adopt pairwise stability with transfers defined by Fox

(2010b).

Definition. (Fox, 2010b) An equilibrium outcome satisfies pairwise stability if the following

conditions hold.

1. For any cj ∈ Ndiand ch ∈ Ndk

with cj /∈ Ndkand ch /∈ Ndi

,

πd(di,Ndi)−

∑cj∈Ndi

tdi,cj≥ πd(di, (Ndi

\{cj}) ∪ {ch})−∑

cj∈Ndi\{cj}

tdi,cj− t̃di,ch

, (A.3)

where t̃di,ch= πc(ch,Nch

) + tdk,ch− πc(ch, (Nch

\{dk}) ∪ {di}).

2. For any cj ∈ Ndkwith cj /∈ Ndi

,

πd(di,Ndi)−

∑cj∈Ndi

tdi,cj≥ πd(di,Ndi

∪ {ch})−∑

cj∈Ndi

tdi,cj− t̃di,ch

, (A.4)

where t̃di,chis the same as above.

3. For any cj ∈ Ndi,

πd(di,Ndi)−

∑cj∈Ndi

tdi,cj≥ πd(di,Ndi

\{cj})−∑

cj∈Ndi\{cj}

tdi,cj, (A.5)

πc(cj,Ncj) +

∑cj∈Ndi

tdi,cj≥ πc(cj,Ncj

\{di}) +∑

cj∈Ndi\{di}

tdi,cj. (A.6)

The first condition indicates that each dealer di prefers its current partner cj instead of an

alternative ch at the transfer t̃di,ch, when ch is not di’s current trading partner. The transfer

t̃di,chis the amount that makes the customer ch to switch its trading partner from dealer k

to dealer i. The second condition implies that even if a dealer has an extra capacity to make

a CDS transaction, it will not add a new partner. Finally, the last condition represents that6See Echenique and Oviedo (2006) and Hatfield and Kominers (2012) for the various definitions of stability

in matching.

26

Page 27: credit default swap market

each agent would not receive strictly larger payoff by cutting off one of its existing trading

links.

A.3 Derivation of LPM inequality

From pairwise stability with transfers, we derive the LPM inequality. That is, if two pairs of

dealers and customers exchange their partners, matching payoffs to the new matches must be

smaller than or equal to matching payoffs to the original matches. For example, suppose that

the current matching allocations are dealer 1-customer 1, and dealer 2-customer 2, denoted

by < d1, c1 > and < d2, c2 >. We further assume that d1 is not matched to c2, and d2 is not

matched to c1. Define N ′d1 , N′d1 , N

′d1 , and N

′d1 as the new sets of their partners such that

N ′d1 = (Nd1\{c1}) ∪ {c2},

N ′d2 = (Nd2\{c2}) ∪ {c1},

N ′c1 = (Nc1\{d1}) ∪ {d2},

N ′c2 = (Nc2\{d2}) ∪ {d1},

where A\B denotes {x ∈ A : x /∈ B}.

Since the sum of total payoffs to the current pairs {< d1, c1 >,< d2, c2 >} are at least

as large as those to the new matches {< d1, c2 >,< d2, c1 >}, we have

{πd(d1,Nd1) + πc(c1,Nc1)}+ {πd(d2,Nd2) + πc(c2,Nc2)} ≥{πd(d1,N ′d1) + πc(c2,N ′c2)

}+

{πd(d2,N ′d2) + πc(c1,N ′c2)

}. (A.7)

In other words, each participant prefers trading with its current partner over exchanging

partners. Note that transfers are all cancelled out since they are additive across partners.

27

Page 28: credit default swap market

To formally derive the LPM inequality, we first consider the inequality (A.3). We replace

t̃di,ckwith t̃di,ch

= πc(ch, (Nch\{dk}) ∪ {di})− {πc(ch,Nch

) + tdk,ch}. Then, we have

πd(di,Ndi)− tdi,cj

≥ πd(di, (Ndi\{cj}) ∪ {ch})

− πc(ch,Nch)− tdk,ch

+ πc(ch, (Nch\{dk}) ∪ {di}). (A.8)

Since there is another match < dk, ch > which involves in this exchange of partners, we have

the same inequality for the pair. That is,

πd(dk,Ndk)− tdk,ch

≥ πd(dk, (Ndk\{ch}) ∪ {cj})

− πc(cj,Ncj)− tdi,cj

+ πc(cj, (Ncj\{di}) ∪ {dk}). (A.9)

Adding the above two inequalities (A.8) and (A.9), we have

πd(di,Ndi)− tdi,cj

+ πd(dk,Ndk)− tdk,ch

≥ πd(di, (Ndi\{cj}) ∪ {ch})

+ πd(dk, (Ndk\{ch}) ∪ {cj})− πc(ch,Nch

)− tdk,ch+ πc(ch, (Nch

\{dk}) ∪ {di})

− πc(cj,Ncj)− tdi,cj

+ πc(cj, (Ncj\{di}) ∪ {dk}) (A.10)

After rearranging terms in the inequality (A.10), we have the extensive version of the

LPM inequality (A.7).

A.4 Empirical specification of individual payoff

We specify the payoff functions πd(di,Ndi) and πc(cj,Ncj

) as follows:

πd(di, Ndi) = αd0 + αdxdi

+ αd,c

∑cj∈Ndi

xcj+ βxdi

∑cj∈Ndi

xcj+ ηdi

+∑

cj∈Ndi

εdi,cj(A.11)

28

Page 29: credit default swap market

and

πc(cj, Ncj) = αc0 + αcxcj

+ αc,d

∑di∈Ncj

xdi+ βxcj

∑di∈Ncj

xdi+ νcj

+∑

di∈Ncj

εdi,cj, (A.12)

where xdiand xcj

are (possibly vectors of) characteristics such as total assets, of dealer i and

customer j respectively. Dealer (customer) fixed effects are captured by ηdi(νcj

), and εdi,cj

is a match-specific error. The coefficient αd (αc) captures the effect of the dealer (customer)

characteristic on its own payoff. The next coefficient αd,c (αc,d) shows the effect of partners’

characteristics on dealer’s (customer’s) payoff. The last parameter β captures the interaction

effects.

Since all variables are additive in the payoffs and the errors are match-specific, we can

derive the marginal payoff of having cj (di) as a trading partner to di (cj). Customer j’s

contribution to dealer i’s payoff is αd,cxcj+ βxdi

xcj+ εdi,cj

. Similarly, dealer i’s contribution

to customer j’s payoff is αd,cxdi+ βxdi

xcj+ εdi,cj

. Then, the matching payoff R(di, cj) to

dealer i and customer j is the sum of the above marginal payoffs. That is,

R(di, cj) = αc,dxdi+ αd,cxcj

+ 2βxdixcj

+ 2εdi,cj. (A.13)

Note that the individual terms, αc,dxdiand αd,cxcj

are cancelled out in the LPM inequal-

ity and that the parameter vector is identified up to a positive monotonic transformation.

Therefore, without loss of generality, we can write the matching payoff as

R(di, cj) = βxdixcj

+ εdi,cj, (A.14)

where xdi,cjincludes asset, gross, net, and ratio in our estimation.

29

Page 30: credit default swap market

A.5 Identification and Consistent Estimation

There are three main conditions for identification and consistent estimation. First, since the

model is based on inequalities, the model is identified up to a positive monotonic transfor-

mation. We have done scale-normalization of the parameter vector such that the norm of

the parameter vector equals to one, i.e. ||β|| = 1.

Second, the stochastic structure of the model satisfies the rank order property studied

by Fox (2010b). The rank order property implies that the stochastic structure of the model

preserves the ranking of payoffs in the LPM inequalities. Fox (2010a and 2010b) provides

simulation results that the rank order property is closely captured by match-specific i.i.d

errors across pairs.

Third, there must be at least one special regressor which has large support (continuous

support in R would suffice) provided by Manski (1985 and 1988) for point identification.

Our explanatory variables such as assets, gross and net volumes satisfy this condition.

The maximum score estimator is consistent under the above conditions. The maximum

score estimation is convenient since it exploits the LPM inequalities which represent revealed

preference relations rather than the full solution concept of a matching equilibrium. Since we

just need to check each revealed preference condition per se, the estimation does not require

solving or checking equilibria among large set of potential match allocations. In addition,

the maximum score estimation avoids performing integrals. These are the most desirable

features of the maximum score estimation.

Since the asymptotic distribution of the maximum score estimator is unknown, we use

the smoothed version of maximum score estimator proposed by Horowitz (1992), that is,

QM(β) = 1M

∑m

K(x′mβ

σM

), (A.15)

where K(·) is the standard normal kernel. We use subsampling to obtain standard errors.

30

Page 31: credit default swap market

For computation, we employ the simulated annealing algorithm to find the global maximum

of the maximum score objective function. We try many starting values to ensure that the

optimisation process reaches the global maximum.

31

Page 32: credit default swap market

B Figures and Tables

B.1 Figures

Dealer Customers Trade link

The size of an orange circle: gross volume of the corresponding dealer

Figure 1: CDS Trading Network Based on New Transactions in 2014H2

32

Page 33: credit default swap market

Figure 2: Degree Distributions

d2

d1

c2

c1

Current Matches A

d2

d1

c2

c1

Exchange of Partners A′

Pairwise stability⇒Market is locally most efficient under current matches

i.e. R(d1, c1) + R(d2, c2) ≥ R(d1, c2) + R(d2, c1)

Figure 3: Pairwise Stability and Local Payoff Maximisation

33

Page 34: credit default swap market

Figure 4: Trading volume by Type-pair in 2014H2

34

Page 35: credit default swap market

B.2 Tables

2012H1 2012H2 2013H1 2013H2 2014H1 2014H2

Buy Sell Buy Sell Buy Sell Buy Sell Buy Sell Buy Sell

Dealers 29% 32% 34% 35% 34% 38% 30% 30% 29% 25% 27% 26%

Banks 42% 43% 38% 37% 32% 34% 33% 32% 33% 34% 33% 30%

Asset Managers 8% 11% 9% 13% 12% 12% 11% 17% 11% 20% 13% 19%

Hedge Funds 19% 11% 16% 12% 18% 14% 22% 18% 24% 18% 21% 21%

Other 2% 3% 3% 3% 3% 2% 4% 4% 3% 3% 5% 4%

Note: Numbers in the Buy side (Sell side) column denote the percentages of bilateral trade volume when

each type of market participants buy (sell) CDS from (to) the nine major dealers, over total trading volume.

Table 1: Trade Position by Type

2012H1 2012H2 2013H1 2013H2 2014H1 2014H2

Reference Entities 1716 1609 1507 1473 1638 1571

Counterparties 853 845 762 708 737 700

New Count 162,268 160,277 135,292 121,222 142,688 105,063

New Trading Volumes 4212.2 5200.8 4307.9 3340.9 3439.7 3209.3

(1) Number of reference entities and counterparties involving to new CDS transactions in each period

(2) Number of total counterparties: 1,357

(3) New Count: number of new CDS transactions

(4) Unit of trading volume: billion USD

Table 2: Overview of UK CDS Market in 2012H1-2014H2

35

Page 36: credit default swap market

Number of Assets Gross Net Ratio

Organisations Mean SD Mean SD Mean SD Mean SD

Dealers 9 5,411.7 12,210.2 1143.5 1214.5 16.109 27.898 0.025 0.034

Banks 169 4,781.6 21,507.0 23.998 120.910 0.942 3.130 0.552 0.415

Asset Managers 318 11.9 44.7 1.565 7.815 0.617 5.032 0.651 0.350

Hedge Funds 370 2.7 13.4 1.723 12.647 0.566 2.646 0.727 0.336

Other 121 105.8 173.2 1.113 2.649 0.356 0.754 0.719 0.351

(1) Number of organisations which have outstanding CDS transactions in 2014H2

(2) Unit: billion USD (assets, gross, net)

Table 3: Organisation Characteristics by Types in 2014H2

36

Page 37: credit default swap market

2012

H1

2012

H2

2013

H1

2013

H2

2014

H1

2014

H2

#Li

nks

%#

Link

s%

#Li

nks

%#

Link

s%

#Li

nks

%#

Link

s%

Dea

ler-

Ban

ks60

043

.86

556

46.8

0%47

744

.53

441

41.1

742

141

.39

397

39.0

3

Dea

ler-

Ass

etM

anag

ers

713

38.0

879

631

.14

667

26.5

662

128

.99

641

28.7

158

727

.28

Dea

ler-

Hed

geFu

nds

827

45.0

493

032

.80

703

29.5

871

130

.15

736

28.8

966

629

.24

Dea

ler-

Oth

er24

135

.23

248

27.0

119

924

.84

177

25.2

119

826

.82

200

26.4

5

Tota

l23

8131

.12

2530

33.7

420

4630

.27

1950

31.0

819

9630

.54

1850

29.8

3

(1)Fo

reach

period,

thefirstcolumnshow

sthenu

mber

oflink

sestablishedforeach

pair

typein

that

period,

andthesecond

columnshow

swha

tpercentageam

ongallpossiblelink

sha

sbeenestablishedin

that

period.

For

exam

ple,

in2014H2,

amon

g1197

possiblelink

sbetweendealersan

dba

nks,

397link

s(33.17%)ha

sbeenestablished

Alldealersareconn

ectedwitheach

other,

sowedo

notcoun

ttheirlink

s.

Table4:

Establish

edPe

rcentage

ofTr

adingLink

sby

Typ

e-pa

irs

37

Page 38: credit default swap market

Variables 2012H1 2012H2 2013H1 2013H2 2014H1 2014H2Assets 0.9288∗∗

(0.0161)0.8665∗∗

(0.0027)0.2108

(0.0057)0.3214∗∗

(0.0467)0.8521∗∗

(0.0053)0.7544∗∗

(0.0108)Gross 0.0002∗∗

(0.00001)−0.00002∗

(0.00001)0.0238∗∗

(0.00002)0.0111∗∗

(0.0008)0.0092∗∗

(0.0001)−0.0004∗∗

(0.00002)Net −0.3621∗∗

(0.0293)0.0824

(0.0082)−0.5406∗∗

(0.0155)−0.4145∗∗

(0.0549)0.4816∗∗

(0.0247)0.4088∗∗

(0.0192)Ratio −0.0781∗

(0.0352)0.4923∗∗

(0.0758)0.8141∗∗

(0.2704)0.8513∗∗

(0.1986)−0.2049∗∗

(0.0368)−0.5135∗∗

(0.0767)Max Score 5417 5578 5844 5523 5589 5753

(1) Column 1: product variables. For example, the product of total assets of dealer i and customer j.

(2) Max Score: the value of objective function, which represents the fit of each specification.

(3) *: 5%, **: 1% level of significance.

Table 5: Estimation Results with Homogeneous Customers From 2012H1-2014H2

38

Page 39: credit default swap market

Variables 2012H1 2012H2 2013H1 2013H2 2014H1 2014H2Dealer-Bank

Assets 0.6756∗∗

(0.0154)0.6891∗∗

(0.0252)−0.0150(0.0115)

0.0120(0.0103)

0.3219∗∗

(0.0032)0.4128∗∗

(0.0129)Gross 0.0008

(0.0036)0.0010

(0.0080)0.0003∗∗

(0.00002)0.0007

(0.0036)0.0030∗∗

(0.00004)−0.00001(0.0054)

Net −0.1874∗∗

(0.0052)−0.3553∗∗

(0.0276)−0.4724∗∗

(0.0303)−0.4145∗∗

(0.0211)−0.2041∗∗

(0.0063)−0.2471∗∗

(0.0136)Ratio −0.2322∗∗

(0.0115)−0.1669∗∗

(0.0133)−0.0829∗∗

(0.0138)−0.1005∗∗

(0.0111)−0.2850∗∗

(0.0171)−0.3406∗∗

(0.0189)

Dealer-AM

Assets −0.0654(0.0379)

0.0163(0.0242)

−0.0728∗∗

(0.0102)−0.0217(0.0116)

−0.4382∗∗

(0.0940)0.0014

(0.0237)Gross 0.0002∗∗

(0.00001)0.0002

(0.0004)0.0001

(0.0034)0.0004∗∗

(0.00002)−0.0005(0.0050)

−0.00006(0.0011)

Net −0.0003(0.0075)

−0.0036(0.0076)

−0.0172(0.0160)

−0.0460∗∗

(0.0083)0.3204∗∗

(0.0146)0.0062

(0.0063)Ratio −0.4050∗∗

(0.0177)−0.3730∗∗

(0.0285)−0.0158(0.0143)

−0.0231∗∗

(0.0091)−0.2314∗∗

(0.0127)−0.2930∗∗

(0.0169)

Dealer-HF

Assets −0.1961∗∗

(0.0606)−0.1957∗∗

(0.0558)0.2397∗∗

(0.0334)0.1536∗∗

(0.0110)0.4225∗∗

(0.0061)0.5032∗∗

(0.0170)Gross 0.2857∗∗

(0.0040)0.2268∗∗

(0.0034)0.0034

(0.0063)0.0318

(0.0317)0.2750∗∗

(0.0027)0.3264∗∗

(0.0044)Net 0.2745∗∗

(0.0109)0.2165∗∗

(0.0164)0.3232∗∗

(0.0275)0.4351∗∗

(0.0249)−0.0995∗∗

(0.0093)−0.0018(0.0046)

Ratio −0.1918∗∗

(0.0085)−0.2144∗∗

(0.0132)−0.1235∗∗

(0.0171)−0.1203∗∗

(0.0082)−0.1503∗∗

(0.0087)−0.1819∗∗

(0.0096)

Dealer-Other

Assets −0.0470(0.0352)

−0.1271∗∗

(0.0440)0.7031∗∗

(0.1220)0.6549∗∗

(0.0691)0.1927∗∗

(0.0045)0.1987∗∗

(0.0086)Gross 0.1091∗∗

(0.0026)0.1025∗∗

(0.0055)0.1085∗∗

(0.0144)0.2172∗∗

(0.0067)0.0096∗∗

(0.0038)0.0097∗∗

(0.0037)Net −0.1846∗∗

(0.0093)−0.1323∗∗

(0.0115)−0.2010∗∗

(0.0264)−0.2870∗∗

(0.0447)−0.2841∗∗

(0.0106)−0.3232∗∗

(0.0154)Ratio 0.0761∗∗

(0.0061)0.0718∗∗

(0.0103)−0.2004∗∗

(0.0195)−0.1677∗∗

(0.0129)−0.1443∗∗

(0.0069)−0.1736∗∗

(0.0125)Max Score 6009 6147 6242 6368 6286 6042

(1) AM: Asset Manager, HF: Hedge Fund, Other: Insurers and other types of participants.

(2) Column 1 represents types of organisations in a match. For example, rows 2-5 correspond to the matching payoff to a dealer-bank pair.

(3) Column 2 shows product variables. e.g. the second column of the first row indicates the product of total assets of dealer i and bank j.

(4) The last row shows the maximum value of objective function for each specification. These values represent the fit of each specification.

(5) **: 1% level of significance.

Table 6: Estimation Results From 2012H1-2014H2

39


Recommended