Journal of Financial Economics: Gino Cenedese, Angelo Ranaldo, Michalis Vasios

JID: FINEC
ARTICLE IN PRESS [m3Gdc;October 3, 2019;1:7]

Journal of Financial Economics xxx (xxxx) xxx
Contents lists available at ScienceDirect
Journal of Financial Economics

journal homepage: www.elsevier.com/locate/jfec
OTC premiaR
Gino Cenedese a, Angelo Ranaldo b,∗, Michalis Vasios c
a
Fulcrum Asset Management, Marble Arch House, 66 Seymour Street, London W1H 5BT, United Kingdom
b
University of St Gallen and Swiss Finance Institute, Unterer Graben 21, St.Gallen CH-9000, Switzerland
c
Bank of England, Threadneedle Street, London EC2R 8AH, United Kingdom
a r t i c l e i n f o a b s t r a c t
Article history: Using unique data at transaction and identity levels, we provide the first systematic study
Received 17 August 2018 of interest rate swaps traded over the counter (OTC). We find substantial and persistent
Revised 18 February 2019
heterogeneity in derivative prices consistent with a pass-through of regulatory costs on to
Accepted 21 February 2019
market prices via so-called valuation adjustments (XVA). A client pays a higher price to
Available online xxx
buy interest rate protection from a dealer (i.e., the client pays a higher fixed rate) if the
JEL classification: contract is not cleared via a central counterparty. This OTC premium decreases by posting
G12 initial margins and with higher buyer’s creditworthiness. OTC premia are absent for dealers
G15 suggesting bargaining power.
G18
© 2019 The Author(s). Published by Elsevier B.V.
G20
G28
This is an open access article under the CC BY license.
(http://creativecommons.org/licenses/by/4.0/)
Keywords:
Interest rate swaps
Financial regulation
Central clearing
Over-the-counter market
Valuation adjustments
R
The views expressed in this paper are those of the authors, and are 1. Introduction
not necessarily those of the Bank of England. The authors would like to
thank for constructive comments Sam Lagfield (referee), Thom Adcock, The Group of Twenty (G20) leaders agreed in 2009 on
Evangelos Benos, Paolo Colla, Darrell Duffie, Mahmoud Fatouh, Peter Gee, an ambitious agenda to reform and strengthen the finan-
Julia Giese, Mariam Harfush-Pardo, Russell Jackson, Lukas Kuenstl, David
cial system. At the centre of the agenda was the reform of
Macdonald, Darren Massey, Tim Meggs, David Murphy, Silvia Pezzini,
Batchimeg Sambalaibat, Philip Strother, Marcel Sweys, Neeltje Van Horen, over-the-counter (OTC) derivatives. What are the effects of
Haoxiang Zhu, and seminar participants at the Bank of England, Euro- the new regulation on derivatives pricing? Has it created
pean Central Bank, European Systemic Risk Board, Hebrew University of stronger price heterogeneity? If so, how can we explain it?
Jerusalem, Goethe University Frankfurt, Norwegian School of Economics,
To address these important questions, we provide the
University of St Gallen, Stockholm Business School, and Tel Aviv Uni-
versity, as well as participants at the 2018 Western Finance Association
first empirical study of interest rate swaps (IRS) in the im-
meeting, the 2018 Annual Central Bank Workshop on the Microstructure plementation of the new regulatory regime. Using unique
of Financial Markets, the 2018 Bocconi-ESMA-CONSOB Conference on Se- and confidential data at transaction and identity (ID) lev-
curities Markets, the 2017 Federal Reserve Bank of Atlanta conference on els, we find substantial and persistent heterogeneity in
‘Financial Regulation: Fit for the Future?’, and the 2017 London School
derivative prices. Besides the standard contract risk, IRS
of Economics-Bank of England Financial Market Infrastructure Conference.
Gino Cenedese and Angelo Ranaldo gratefully acknowledge financial sup- rates vary depending on whether the transaction is cleared
port from the Canadian Derivatives Institute. Gino Cenedese started this via a central counterparty (CCP) or cleared bilaterally (non-
project while being employed at the Bank of England. CCP), if initial margin is posted, and on counterparty credit
∗
Corresponding author. risk. These price differentials that we call OTC premia are
E-mail addresses: [email protected] (G. Cenedese),
highly significant in statistical and economic terms. We ra-
[email protected] (A. Ranaldo), [email protected] (M. Vasios).
tionalise these OTC premia with increased inventory costs
https://doi.org/10.1016/j.jfineco.2019.09.010
0304-405X/© 2019 The Author(s). Published by Elsevier B.V. This is an open access article under the CC BY license.
(http://creativecommons.org/licenses/by/4.0/)
Please cite this article as: G. Cenedese, A. Ranaldo and M. Vasios, OTC premia, Journal of Financial Economics, https:
//doi.org/10.1016/j.jfineco.2019.09.010
JID: FINEC
2 G. Cenedese, A. Ranaldo and M. Vasios / Journal of Financial Economics xxx (xxxx) xxx
induced by the recent financial regulation that are passed as FX derivatives (Andersen et al., 2019), our study has ex-
on to market prices through so-called valuation adjust- ternal validity beyond the IRS market. Moreover, some cat-
ments (XVA). Also, OTC premia favour dealers in the sense egories of market participants such as pension funds and
that the extra cost in the non-CCP segment for interest rate certain nonfinancial corporates are (still) exempt from cen-
protection (i.e., paying the fixed rate) is absent for dealers tral clearing and it is not clear how the new regulation will
while it is large for clients suggesting pass-through of reg- affect them.
ulatory costs and bargaining power for dealers. To introduce our pricing analysis, we map the USD-
A better understanding of derivative markets is im- denominated IRS market that represents the largest seg-
portant for various entities including policy-makers and ment of the entire interest rate derivatives market.2 More
market participants. First, derivative markets are complex, specifically, we analyse every contract by non-US-based
opaque, and composed of many segments leading to counterparties that traded with UK-based entities (includ-
price heterogeneity (Duffie, 2012). Indeed, there may be ing subsidiaries of foreign banks) and that was reported
many sources of price heterogeneity, including different to the Depository Trust & Clearing Corporation (DTCC) be-
legal frameworks (e.g., different International Swaps and tween the beginning of December 2014 and the end of
Derivatives Association (ISDA) agreements, accounting or February 2016. The time span intentionally starts after the
regulatory rules), counterparty risk, types of agents (e.g., US clearing mandate (March 2013), but before the Euro-
nonfinancial corporates versus dealers), and the overall pean one (June 2016). Throughout our sample period, US
difficulty to novate and negotiate across market segments. entities were required to centrally clear their transactions,
But at the same time, the sheer size of $544 trillion of no- and this is why we remove them, while the other counter-
tional value as of June 2016 (Bank for International Settle- parties in our sample (with access to CCP) had the choice
ments, 2016a), the fact that derivatives are heavily traded whether to centrally clear their trades or not. This set-
by professional agents, and the new regulatory framework ting represents the ideal laboratory to study the distin-
that has supposedly improved market quality in terms guishing characteristics between the CCP and non-CCP seg-
of transparency and (CCP) risk homogenisation should ments, in terms of contract features, counterparty types,
facilitate price efficiency. Thus, it is uncertain a priori and how the trading activity in these segments evolves
whether the same derivative contract is more likely to be across time and risk. A unique feature of our data is the in-
priced equally across different segments or types of mar- formation on counterparties’ identities, which allows us to
ket participants, in particular in the phase-in of regulatory analyse relevant issues such as the trader’s credit risk and
reforms. relationships.
Second, in the wake of the recent global financial cri- To carry out the core analysis of our paper, we ad-
sis policy-makers have implemented new regulations that dress the question whether there is substantial and per-
have fundamentally changed the structure and mode of sistent heterogeneity in derivative pricing in the phase-in
operation of OTC derivative markets. Among the declared period of the new regulatory regime and, if so, why. To do
objectives, there have been the increase of standardisation, this, we analyse what explains IRS returns as convention-
central clearing, and the reduction of systemic risk, es- ally defined, i.e., the difference between the transaction-
pecially in terms of credit risk (Financial Stability Board, level swap rate (i.e., the fixed leg of the swap contract)
2010). However, the new regulation may have created and the benchmark rate at the end of the previous day.
some unintended consequences such as the incentive of More specifically, we perform panel regressions with day
market participants bearing higher credit risk to circum- and counterparty fixed effects in which IRS returns are re-
vent clearing mandates by adjusting the terms of contracts gressed on the main potential determinants of pricing het-
to make them non-standardised (Financial Stability Board, erogeneity as outlined by dealers’ inventory and bargaining
2017b). By analysing the transitional regime after the in- theories as well as XVA methods.
troduction of price-based measures such as Basel III capi- In line with inventory theories, market participants ad-
tal charges but before the entry into force of the quantity- just derivative prices to inventory holding costs of deriva-
based European central clearing mandate in mid-2016, our tives exposures.3 In addition to hedging costs due to
work sheds light on whether these objectives have been the contract exposure increasing with larger notional and
achieved and it helps in deciphering future developments longer maturities, dealers take into account costs origi-
in OTC derivative markets. In fact, the implementation of nated from new regulations such as the Basel III capital
these reforms is halfway through. More than half of the and liquidity requirements. In this new regulatory frame-
outstanding OTC derivatives, as measured by the num- work, dealers will also require a compensation for holding
ber of contracts, are not yet centrally cleared (Financial
Stability Board, 2017a), which reflects the fact that with
2
the exception of plain vanilla interest rate derivatives and At end-June 2016, the notional outstanding for interest rate deriva-
tives was $437 trillion (80% of the market), of which $327 trillion (60%
a limited number of credit default swaps (CDS) indices,
of the market) were interest rate swaps, of which one-third was denomi-
most of OTC derivatives are not subject to any clearing nated in USD (Bank for International Settlements, 2016a).
mandate.1 Since XVA applies to a range of products, such 3
Models of inventory effects are proposed in Stoll (1978), Amihud and
Mendelson (1980), Ho and Stoll (1981), Mildenstein and Schleef (1983),
O’Hara and Oldfield (1986), Grossman and Miller (1986), and Shen and
1
At the end of 2016, the share of global stock of outstanding OTC Starr (2002). In particular, in Amihud and Mendelson (1980), O’Hara and
derivatives that was centrally cleared was 61% for interest rate deriva- Oldfield (1986), and Shen and Starr (2002) quotes and spreads depend on
tives, 28% for CDS, and minuscule for FX, commodity, and equity deriva- dealers’ inventories. More recently, Pelizzon et al. (2016) extend the Stoll
tives (Financial Stability Board, 2017a; Financial Stability Board, 2017b). (1978) model to margins.
//doi.org/10.1016/j.jfineco.2019.09.010
JID: FINEC
G. Cenedese, A. Ranaldo and M. Vasios / Journal of Financial Economics xxx (xxxx) xxx 3
derivative positions implying ‘regulatory’ costs. These ef- by at least half or even disappears if a non-CCP trade in-
fects will be different across types of agents and OTC seg- volves posting of initial margin. These results support the
ments generating price heterogeneity. empirical predictions drawn from the XVA approach and
A further source of price heterogeneity is dealers’ bar- point to credit risk (CVA) and capital charges (KVA) as im-
gaining power that can arise from, e.g., the lack of outside portant drivers of OTC premia for non-CCP transactions.
options for customers (e.g., Duffie et al., 2005; Duffie et al., Another important finding is that the OTC premium is not
2007), dealers’ network centrality (e.g., Li and Schürhoff, fully symmetric when dealers receive or pay the fixed rate
2019), financial expertise (e.g., Glode et al., 2012), as well to their clients, that is, OTC premia are significantly pos-
as information rents (e.g., Bolton et al., 2016). If dealers itive when dealers receive the fixed rate (consistent with
possess bargaining power against their customers (Duffie, dealers passing on XVA ‘costs’ to clients when selling in-
2012), then they will charge less favorable prices especially terest rate protection) but neither systematically negative
to entities with limited outside options. nor positive (i.e., there is neither a discount nor a pre-
Heterogeneous OTC premia are also fully consistent mium) when dealers pay fixed. Our tentative explanation
with the common valuation approach for derivatives that for the positive OTC premium when dealers receive fixed
includes XVA elements such as Credit and Debit Valu- is bargaining or market power: dealers pass the XVA costs
ation Adjustment (CVA, DVA), Funding Valuation Adjust- to their clients buying interest rate protection (i.e., pay-
ment (FVA), Margin Valuation Adjustment (MVA), and Cap- ing fixed). Overall, these findings are consistent with the
ital Valuation Adjustment (KVA). More specifically, from idea that (i) as with any inventory costs, dealers tend to
XVA pricing one can draw the following empirical pre- pass on regulatory costs to market prices, and (ii) deal-
dictions: first, the same IRS contract traded in the non- ers exercise some bargaining power when they price swap
CCP segment rather than the CCP segment demands larger contracts.
CVA and KVA resulting in a higher market price of the Third, we identify the main determinants of the OTC
fixed leg for the buyer of interest rate protection. Second, premia. In addition to the effect of initial margin described
when a customer pays fixed to a dealer in the non-CCP above, we find that swap prices tend to increase with no-
segment, the OTC premium decreases (increases) with the tional amount and time to maturity, which capture con-
client’s (dealer’s) creditworthiness because a higher credit tract risk. Moreover, the price a client bank pays (receives)
rating translates into a lower cost for buying protection for the fixed rate to (from) a dealer increases (decreases)
(CVA) and lower capital charge (KVA). Third, the exchange with the bank’s credit risk, which is again consistent with
of initial margin in non-CCP trades mitigates the effect of the pricing of CVA and KVA. Another finding is that there
CVA and KVA (but increases the effect of MVA). Fourth, is no premium for transactions that are exempted from
a higher contract risk exposure in terms of notional and Basel III related capital charges, providing additional sup-
maturity implies higher XVAs for non-CCP rather than CCP port for the effect of KVA.
contracts. Finally, we test whether alternative hypotheses, such
Some clear results emerge from our study. First, we un- as market liquidity, relationships, and bilateral exposures
cover new stylised facts of the IRS market in the transition can explain our results. To do this, we exploit the execu-
to the new regulatory regime, which are: (i) CCP trades tion of swaps on centralised electronic platforms, known
are generally more standard contracts with larger notional as Swap Execution Facilities (SEFs), which increase trans-
and longer maturity;4 (ii) the average market participant parency and dealer competition thus improving market liq-
on CCPs is more likely to have higher creditworthiness uidity (Benos et al., 2019). To examine bilateral relation-
consistent with the theoretical predictions that the non- ships, we augment our regressions with buyer–seller fixed
CCP venue concentrates higher counterparty risk (Acharya effects and time-varying bilateral relationships (e.g., time-
and Bisin, 2014; Biais et al., 2012; Biais et al., 2016);5 and varying dealer–client exposures). OTC premia remain sta-
(iii) in distressed markets, market participants significantly tistically and economically significant.
increase their trading outside CCPs suggesting that they We contribute to the empirical literature on derivative
tend to circumvent the so-called margin procyclicality, i.e., markets in several ways. First, although price dispersion
tighter funding conditions and higher margins imposed by due to over-the-counter frictions such as search costs and
the CCP. bargaining is well-grounded theoretically (e.g., Duffie et al.,
Second, we find substantial and persistent hetero- 20 05; Duffie et al., 20 07), this is the first study providing
geneity in OTC premia, which is significant in statistical empirical evidence that systematic OTC premia in IRS con-
and economic terms. For instance, when a client buys tracts exist. Our analysis also highlights the main determi-
interest rate protection, she pays the dealer a fixed rate nants of OTC premia, namely, how the contract is cleared
that is around 8 basis points higher for a non-CCP transac- (CCP and non-CCP transactions), whether it involves
tion relative to a CCP one. This is a sizeable amount con- margins, the types of market participants, and contract
sidering the total traded notional of around 7.4 trillion US characteristics (notional and maturity).6
dollars over our sample period. The OTC premium reduces
6
Theoretically, price heterogeneity can arise from a number of reasons
4
‘Standard’ means that the contract maturity is 1Y, 2Y, 3Y, 4Y,..., 50Y or including search costs (e.g., Duffie et al., 2005; Duffie et al., 2007; Lagos
the last digit of the price (fixed rate) is 5 or 0. and Rocheteau, 2007; Lagos and Rocheteau, 2009; Atkeson et al., 2015;
5
To the extent that low credit rating parties are also low volume play- and Afonso and Lagos, 2015) or informational issues (e.g., Duffie et al.,
ers, this finding can also be the result of the high entry/usage costs of the 2009; Zhu, 2012; Babus and Kondor, 2018; Golosov et al., 2014; and Babus
CCP infrastructure. and Hu, 2017). Empirical evidence of price dispersion in OTC markets is
//doi.org/10.1016/j.jfineco.2019.09.010
JID: FINEC
Second, our findings provide empirical support to the Market Infrastructure Regulation (EMIR) and the Basel III
recent literature on the theory of XVA (e.g., Green, 2015; framework.
Gregory, 2015; Ruiz, 2015; and Andersen et al., 2019). In July 2012, the European Union issued EMIR, which
In particular, Andersen et al. (2019) theorise the link lays down clearing requirements for OTC derivative con-
between FVA and the dealer’s price quotation to align the tracts and uniform requirements for the performance of
market-making function with shareholder interests. We activities of CCPs.8 The EMIR clearing obligation required
provide empirical evidence that (i) overall, XVAs are priced eligible European counterparties to centrally clear certain
in derivatives generating price heterogeneity and that (ii) types of OTC derivative contracts, including interest rate
CVA and KVA are important factors. Rather than a sign of swaps. It was phased in from June 2016 on the basis of
inconsistency or frictions, our findings suggest that het- firms’ categorisation and their trading volume in derivative
erogeneity in derivatives prices results from an attempt to contracts. Firms may fall in four different categories ac-
adjust efficiently valuations to specific contract character- cording to EMIR: (i) Category 1 comprises firms that are al-
istics and, arguably, to the new regulatory framework. On ready clearing members of a CCP, and for which the clear-
the other hand, we also find asymmetric patterns in the ing obligation took effect from 21 June 2016; (ii) firms in
dealer-to-customer segment pointing to dealers’ superior Category 2 are financial counterparties with a month-end
bargaining power even after controlling for other possible average of outstanding notional amount of OTC derivatives
factors such as market liquidity and relationship issues. above 8 billion euros, and were required to centrally clear
Third, we contribute to the literature on central clear- from 21 December 2016; (iii) financial counterparties with
ing, which has been studied mainly theoretically (e.g., an average notional outstanding below 8 billion euros fall
Duffie and Zhu, 2011; Biais et al., 2012; Biais et al., 2016; in Category 3 and their initial clearing date was originally
Acharya and Bisin, 2014). The empirical literature devoted set to 21 June 2017, but subsequently postponed to 21 June
to CCP and OTC markets in post-crisis periods has mostly 2019; (iv) finally, Category 4 includes all nonfinancial coun-
focused on CDS.7 We contribute to this literature by pro- terparties, whose initial clearing date was set to 21 De-
viding empirical evidence that the CCP segment is rela- cember 2018. Note that intragroup transactions are exempt
tively safer (in terms of credit risk), it supports standard- from central clearing; also, at the time of writing, pension
isation and better prices (i.e., lower OTC premia), all of funds benefit from a temporary clearing exemption under
which are consistent with the policy objectives of promot- Article 89(2) of EMIR, and are likely to continue to benefit
ing central clearing. from this for some time (European Securities and Markets
Finally, we provide empirical support to the growing lit- Authority (ESMA) statement ESMA70-156-641).
erature on intermediary asset pricing (e.g., He and Krishna- Central clearing requires market participants to com-
murthy, 2013; Adrian et al., 2014; He et al., 2017) showing ply with the CCP’s risk management framework, which in-
that financial intermediaries, in this case derivatives deal- cludes the exchange of initial and variation margin as well
ers, play a crucial role in the pricing of financial assets and as the contribution to the default fund. EMIR also intro-
in creating heterogeneity as well as asymmetry in deriva- duced risk-mitigation techniques for non-centrally cleared
tives prices. OTC derivative contracts. Under these uncleared margin-
The remainder of this paper is structured as follows. ing rules phased in between January 2017 and September
In Section 2 we provide an overview of the post-crisis 2020, all covered entities (i.e., financial firms and systemi-
derivatives regulatory framework. Section 3 describes the cally important nonfinancial entities), are required to ex-
data. Section 4 contains the main empirical analysis, and change variation margin and initial margin on a regular
Section 5 concludes. basis for non-CCP trades.
The Basel III framework was announced in 2010
and developed by the Basel Committee on Banking
2. Policy context Supervision (BCBS). It consists of a comprehensive set
of regulations affecting every aspect of banking, from
To reduce the risk and severity of future financial crises, capital to liquidity and resolution. This framework is
global regulatory authorities around the world started a structured in several phases, with the first phase starting
significant post-crisis programme of reforms. In regard to in 2013, while it has been reviewed a few times, most
the OTC derivative markets, the main objective has been recently in 2017. The leverage ratio (i.e., the ratio of Tier 1
to incentivise centralised clearing, which was expected to capital to total exposures) and the risk-weighted capital re-
reduce counterparty risk and simplify the network of bi- quirements (i.e., the ratio of capital to risk-weighted assets)
lateral exposures. We focus here on two of the reforms are at the core of Basel III. The minimum internationally
that are most relevant to this objective: the European agreed Tier-1 capital requirement is 6% of risk-weighted
assets, but there are typically additional buffers set to ab-
sorb losses under stress. With respect to the leverage ratio,
provided by, e.g., Jankowitsch et al. (2011); and Green et al. (2007). More
recently, Hau et al. (2018) analyse dealers’ discriminatory pricing in FX the BCBS required banks to publicly disclose their leverage
derivatives markets, while Cenedese et al. (2019) attribute currency mis- ratio from January 2015, and proposed a minimum ratio of
pricing to dealer balance sheet constraints. 3% that was scheduled to become binding in January 2018.
7
E.g., Arora et al., 2012; Loon and Zhong, 2014; Loon and Zhong, 2016;
Duffie et al., 2015; Du et al., 2016; Bellia et al., 2018). Another paper by
8
Menkveld (2017) looks into CCP crowded risk in equity markets. By look- That is the Regulation (EU) No 648/2012 of the European Parliament
ing at IRS centralised trading, Benos et al. (2019) represent an exception and of the Council, of 4 July 2012 on OTC derivatives, central counterpar-
focusing on the effects of the Dodd-Frank Act on market liquidity. ties, and trade repositories.
//doi.org/10.1016/j.jfineco.2019.09.010
JID: FINEC
While the scope of Basel III is broad, it has far-reaching porates the cost of posting initial margin. Finally, the FVA
implications for the operation of OTC derivative markets. reflects the cost of funding liquidity.
For example, in July 2012, the BCBS, in consultation with The following example illustrates how new regulation
the Committee on Payments and Settlement Systems generates (heterogeneous) impacts on swap prices via the
(CPSS) and the International Organisation of Securities XVA. Let’s assume that a dealer-bank sells interest rate
Commission (IOSCO), assigned a small risk weight of 2% protection to a client via a 10-year IRS contract with a the-
for bank exposures to central counterparties. The BCBS oretical fixed rate of 1%. This fixed rate ensures that the
also introduced a cap on the capital charge on banks’ fixed and floating leg of the swap have the same present
exposures to CCPs in April 2014. These policies aimed to value. Assuming there are no pre-existing positions be-
promote central clearing. tween the dealer and the client, and no interdealer hedge,
The treatment of counterparty credit risk in Basel III the following valuation adjustments apply:
has also impacted the trading of OTC derivatives. Basel II
had addressed this risk using the default capital charge, in- • If the swap is centrally cleared, it will put pressure on
tended to cover any losses due to a counterparty’s default. dealer’s funding liquidity (FVA, MVA) because of the
However, the global financial crisis showed that two-thirds need to manage the posting of margin, on the one
of counterparty-risk-related losses were due to the move- hand, while it will introduce a small capital charge
ment of the credit quality of counterparties, rather than (KVA) because of the 2% risk weight for bank exposures
actual defaults (Basel Committee on Banking Supervision, to central counterparties, on the other hand. Given the
2009).9 To address this gap, Basel III introduced (in Decem- low counterparty risk associated with CCP trades, there
ber 2010) the CVA capital charge as a protection against will be no CVA or CVA capital adjustment. In theory,
mark-to-market losses caused by an increase in the credit there could be a CVA adjustment to reflect the mutual-
spread of the counterparty. More importantly, exposures to isation of counterparty risk via the default fund, but in
CCPs were exempted from the CVA capital charge, which practice this is not generally considered important.
otherwise would have been a significant cost of trading in • If the swap is bilateral and uncollateralised, it will in-
the CCP segment. troduce costs related to counterparty risk (CVA) and
Counterparty capital charges differentiate between capital (KVA) as a result of the associated cost for
margined and unmargined non-CCP transactions too. This hedging the counterparty credit risk, the default capi-
is because initial margin reduces the amount of expo- tal charge, and the CVA capital charge. The magnitude
sure for OTC derivatives transactions. For example, the of these costs will be negatively correlated to the cred-
standardised approach for measuring counterparty credit itworthiness of the client. This is because both the cost
risk exposures (SA-CCR), introduced by the BCBS in April of buying protection against counterparty default (de-
2014, allows for a reduction in the exposure at default fault probability ↑ ⇒ CVA ↑) and the credit spread
(EAD) when initial margin is received by the counterparty, volatility (CVA capital charge ↑ ⇒ KVA ↑) increase for
through a reduction in EAD’s two main components: the low-credit-rating counterparties.10
replacement cost (RC) and the potential future exposure • The exchange of initial margin in the bilateral case
(PFE). With respect to the leverage ratio, its calculation will reduce the counterparty risk (loss given default
does not recognise collateral or other credit risk mitigants ↓ ⇒ CVA ↓) and capital (exposure at default↓ ⇒
as an offset to derivatives exposures. This is fundamentally KVA ↓) costs, but it will increase other costs re-
different from the risk-weighted framework, which favours lated to collateral and funding management (MVA).
the exchange of initial margin in centrally or bilaterally It is unclear how the MVA adjustment for a bilat-
cleared transactions. eral trade with initial margin should compare to the
MVA adjustment for a CCP trade. Note that our sam-
ple period is before the implementation of the un-
2.1. An illustrative example of XVA cleared margining rules and as a result counterpar-
ties calculate initial margin using their internal (non-
The new regulatory landscape is changing the pricing standardised) models (Gregory, 2015).11 On the one
of OTC derivatives. Market participants move away from hand, CCP risk models calculate initial margin using
textbook-type pricing formulas and start to take into ac- a 5-day close out period against typically a longer
count the underlying credit, collateral, funding, and capital period for bilateral trades and offer more netting
implications of every transaction. It is common for banks opportunities (multilateral netting), which suggests a
to actively manage these components and to incorporate smaller initial margin for CCP trades (MVACCP ↓).
them into prices through valuation adjustments, a practice
known as XVA. For example, the CVA is the adjustment
10
taken upfront against counterparty defaults. The KVA takes For example, Hull et al. (2005) show in a simplified example that the
into account costs related to capital requirements (default real-world default probability of a Baa bank is ten times larger than that
of a Aaa bank.
capital charge and the CVA capital charge). The MVA incor- 11
After the implementation of the uncleared margining rules (phased in
in Europe between January 2017 and September 2020), the covered enti-
ties who trade in the non-CCP segment would be required to exchange
9
The UK Financial Service Authority estimated that losses of UK banks variation margin regularly (e.g., daily) and initial margin over a 10-day
during 20 07–20 09 related to the market risk nature of counterparty risk time horizon with more stringent threshold and minimum transfer rules.
were five times the amount of actual default losses. See http://www.fsa. This will likely increase the MVA for bilateral trades (MVANon−CC P ↑) and
gov.uk/pubs/discussion/dp10_04.pdf. further reduce their CVA and KVA.
//doi.org/10.1016/j.jfineco.2019.09.010
JID: FINEC
On the other hand, a CCP’s initial margin tends to portantly, counterparty identities.15 This allows us to cate-
be more risk-sensitive and is collected daily, whereas gorise trades by type of counterparty and location.
a bilateral initial margin is less procyclical and col- We carefully apply a number of filters to clean the data.
lected less frequently (MVACCP ↑). Bilateral initial mar- We start by keeping only USD-denominated spot-starting
gin can also be linked to thresholds and minimum swaps, which we do by removing any reports whose effec-
transfers, leading to undercollateralisation (CVANon−CC P tive date is more than two business days from the trade
> CVACCP ).12 date. Next we remove duplicate reports. Duplication is
mainly due to three reasons. First, EMIR is a double-sided
Overall, each of these adjustments will push the fixed reporting regime, so we see two copies for a single trade
rate higher than the 1% theoretical price in order to com- when both counterparties are UK legal entities and both
pensate the dealer, who is the receiver of the fixed rate, for of them report to DTCC. Second, as per the EMIR regula-
incurring the additional costs. tion, the data contain several copies of the same trade to
reflect any modification, correction, and valuation updates.
3. Data We remove these duplicates using the unique trade identi-
fier (UTI) of every report.
We use transactions of USD-denominated spot fixed-to- Another reason for duplication is that for every trans-
floating IRS contracts executed between 1 December 2014 action that is centrally cleared there are typically three re-
and 21 February 2016. We focus on the USD-denominated ports sent to the TR: the original transaction (alpha trade
segment of the market because this is the largest and most report) and the two novations (beta and gamma trade re-
liquid segment in terms of turnover (Bank for International ports). These reports tend to have different UTIs. We re-
Settlements, 2016b). The source is the DTCC, the largest move these duplicate reports by applying an algorithm that
European trade repository (TR).13 matches trades based on trade date and time, effective
The time span intentionally starts after the US clearing date, maturity date, notional, swap rate, and counterparty
mandate (March 2013), but before the European one (June identities. In addition, for every trade that is cleared by the
2016). This means that while US entities were required to London Clearing House (LCH), which has more than 90%
centrally clear their transactions (and this is why we re- market share in the cleared interest rate swap market, we
move them), the other counterparties in our sample (with obtained the trade identifiers of the associated (two) no-
access to CCP) had the choice whether to centrally clear vations directly from LCH. We accurately remove these du-
their trades or not. The start date in December 2014 is cho- plicate novations by using the LCH information in conjunc-
sen for data quality reasons. Before December 2014, many tion with the matching algorithm.
values were missing in key variables in the DTCC data un- Finally, to remove any false or inaccurate reports we
til a process of validation was introduced by the European only keep trades with a fixed rate that is within 150 ba-
Securities and Markets Authority (ESMA), after which the sis points from the benchmark (same maturity and cur-
data quality dramatically improved, as shown in Cielinska rency) end-of-day swap rate mid-quote from Bloomberg.16
et al. (2017) and Abad et al. (2016). In addition, note that The Bloomberg benchmark is the rate used by practition-
although the European clearing mandate came into force ers to proxy the ‘fair’ value of the prevailing fixed rate.
in June 2016, EU counterparties had to centrally clear all After filtering the data we are left with 169,996 reports
transactions with a remaining maturity of more than six out of which 68,945 involve US persons, who were subject
months from 21 February 2016 onwards, as a result of the to the US clearing mandate and as a result, they had no
EMIR frontloading requirement. Hence, we use 21 February choice but to centrally clear their trades.17 After excluding
2016 as the cut-off date.
The DTCC data provide information on flows, for ex- 15
For bilateral trades, DTCC data provide information on whether the
ample, new trades, modifications, valuation, and cancela- transaction is uncollateralised (i.e., no exchange of variation or initial mar-
tion updates.14 Each transaction report contains more than gin), partially collateralised (i.e., exchange of variation margin), or fully
100 fields that include information on trade characteris- collateralised (i.e., exchange of variation and initial margin).
16
tics (e.g., price, notional amount, maturity date, execution This is necessary as some counterparties mistakenly report the swap
time, reference rate), whether a trade is centrally cleared rate in basis points instead of percentage points. Our filter ensures the
removal of these inaccurate (about 4800) reports. As robustness anal-
and the type of the collateral exchanged, and, more im- yses, we experimented with other filter methods. For example, (i) we
trimmed swap rate log-returns, defined as the log-difference between
the transaction-level swap rate and the benchmark end-of-day swap rate
12
The threshold is the amount below which collateral is not required. A mid-quote from Bloomberg, at the 2.5% and 97.5% levels; or (ii) we
minimum transfer is the smallest amount of collateral that can be trans- trimmed swap rate absolute returns, defined as the difference between
ferred. See Chapter 6 in Gregory (2015) for more details. the transaction-level swap rate and the benchmark end-of-day swap rate
13
Abad et al. (2016) report that the European DTCC data cover about mid-quote from Bloomberg, at the 2.5% and 97.5% levels, among others.
70% of the global interest rate swap market. Our results are not sensitive to the filtering approach and remain qualita-
14 tively the same.
Under EMIR the Bank of England is entitled to see (i) trades cleared
17
by a CCP supervised by the Bank and trades in which one of the coun- Note that in Japan counterparties are required to centrally clear only
terparties is a UK entity. The definition of entities includes CCPs; financial those IRS contracts that are denominated in JPY, which are not included
counterparties, such as banks, insurance firms, and hedge funds; and non- in our study as we focus on USD-denominated contracts. Also, other
financial counterparties that are EU legal entities, but exempts some other jurisdictions introduced a clearing mandate only after the time period
entities, for example, EU national central banks and natural persons. See spanned by our sample; for example, Singapore introduced the mandate
Article 2 of Commission Delegated Regulation (EU) No 151/2013 - Data in October 2018. The largest group in our data is UK counterparties, who
access by relevant authorities. account for 57% of total traded notional, followed by French and German
//doi.org/10.1016/j.jfineco.2019.09.010
JID: FINEC
Fig. 1. Daily notional traded (in $billion) by segment. In this figure we plot the total notional traded in the CCP and non-CCP segment. The sample covers
every USD-denominated spot vanilla interest rate swap by non-US-based counterparties, which was reported to DTCC between December 1, 2014 and
February 19, 2016.
these reports from our analysis, the final sample consists 4. Empirical analysis
of 101,051 new trades by about 800 active counterparties,
which account for a total of $7, 470 billion in traded no- 4.1. Descriptive statistics
tional over our sample period.
Fig. 1 shows the daily time series of traded notional by Before conducting any in-depth analysis of OTC premia
segment. The CCP segment includes all trades cleared by a we present some key insights into the USD IRS market that
clearing house.18 The non-CCP segment consists of all non- we summarise in Table 1. We start with the breakdown of
cleared or bilaterally cleared reports. Fig. 1 illustrates the the market by segment and counterparty type. The daily
immense size of the IRS market with the daily notional notional traded (by non-US investors) of around $15 billion
traded hovering around $15–30 billion, of which about 10% reiterates that the USD IRS market is economically very
takes place in the non-CCP segment across the whole sam- important, as depicted in Fig. 1. The CCP segment domi-
ple period. To get a sense of the data coverage, we com- nates trading with a market share of 85% and 90% in terms
pare them against the data from the global CCP USD IRS of number of trades and notional, respectively. Although a
market used in Benos et al. (2019). They report a daily no- fraction of the total market, the non-CCP segment remains
tional traded by non-US counterparties of about $20 billion an economically significant quantity with daily notional of
between 2013 and 2014, which indicates that we see the about $1–2 billion and in number of trades terms, about
lion’s share of the (non-US counterparty) global USD IRS one out of six trades is non-centrally cleared.
market, a result of London’s status as a global centre for The availability of counterparty IDs allows us to clas-
derivatives trading. sify entities into meaningful groups: dealers, banks, hedge
funds, asset managers, insurance companies and pension
funds, other financial companies, and nonfinancial com-
panies.19 There are also some entities that could not be
counterparties with 12% and 9% of market share, respectively. See Table
A.1 in the Internet Appendix.
18 19
The DTCC data contain a flag that equals one for centrally cleared The ‘dealers’ category includes the so-called ‘G16’ dealer-banks: Bank
trades and zero otherwise. However, we observe that some reports with a of America, Barclays, BNP Paribas, Citibank, Credit Agricole, Credit Suisse,
zero value involve a clearing house as a counterparty (LCH or the Chicago Deutsche Bank, Goldman Sachs, HSBC, JP Morgan, Morgan Stanley, No-
Mercantile Exchange). We classify these cases as centrally cleared. Finally, mura, Royal Bank of Scotland, Societe Generale, UBS, and Wells Fargo.
we require that clearing takes place on the execution date (T+0), which This choice is not arbitrary on our part as these banks are also classified
follows the definition of ESMA. By extending the cut-off time up to five as “Participating Dealers” in the OTC Derivatives Supervisors Group,
days, we obtain a slightly larger non-CCP subsample and the results re- chaired by the New York Fed https://www.newyorkfed.org/markets/otc_
main unchanged. derivatives_supervisors_group.html.
//doi.org/10.1016/j.jfineco.2019.09.010
JID: FINEC
Table 1
Trading activity by segment and counterparty type.
This table reports the number of trades (N trades) and notional traded (Notional) throughout
our sample period by segment and counterparty type. CCP denotes trades cleared through a
central counterparty, while Non-CCP denotes non-centrally cleared trades. The latter segment
consists of collateralised (Non-CCP with margin) and uncollateralised (Non-CCP no margin)
trades. The exchanged collateral can be variation margin or/and initial margin. Counterparties
are classified into asset managers (AM), banks (Bank), dealers (Dealer), hedge funds (HF), insur-
ance companies and pension funds (Ins), other financial companies (Other fin), and nonfinan-
cial companies (Non-fin). Panel B excludes trades from entities that could not be classified. The
sample covers every USD-denominated spot vanilla interest rate swap by non-US-based coun-
terparties, which was reported to DTCC between December 1, 2014 and February 19, 2016.
Panel A: Breakdown by segment
N trades Notional ($m)
CCP 85,810 6,801,165

Non-CCP 15,241 669,346
Total 101,051 7,470,511
Panel B: Breakdown by segment and counterparty type
AM Bank Dealer HF Ins Other fin Non-fin
CCP
N trades 3324 33,387 114,733 8057 300 171 26
Notional ($m) 132,823 1,831,592 9,084,860 1,074,940 21,561 11,100 1216
Non-CCP with margin

N trades 1045 4869 11,966 610 797 696 113
Notional ($m) 35,133 132,296 560,238 81,574 70,474 44,364 9573
Non-CCP no margin
N trades 635 980 3717 56 270 177 28
Notional ($m) 14,192 37,631 149,484 5904 17,787 21,621 7365
classified, mainly because of missing ID information. Their Bloomberg from the previous business day (to avoid any
trades are typically small, infrequent, and non-centrally endogeneity concerns), which is publicly available to ev-
cleared. Collectively they account for 8% of trades. ery market participant at the start of the trading session.20
In Table 1 (Panel B) we present trading activity vari- The average swap rate return is positive, consistent with
ables by type of counterparty throughout our sample pe- the upward Overnight Index Swap (OIS) term structure
riod. G16 dealer trading accounts for about two-thirds of and with the upward interest risk prevailing during our
trading in terms of both notional traded and number of sample period. Interestingly, the average returns are 1.78,
trades. The rest of trading is split between banks (13% 2.19, 2.52 basis points (bps) for CCP, non-CCP collateralised
of notional), hedge funds (7.8%), asset managers (1.2%), (with exchange of variation margin and/or initial margin),
and others. It is worth noting that although smaller non- and non-CCP uncollateralised trades, respectively. This in-
bank players such as insurance companies, as well as other creasing order suggests that the cost for interest rate pro-
financial and nonfinancial companies trade rather infre- tection (i.e., paying fixed) is most (least) expensive in the
quently, their trading collectively sums to the economically non-CCP (CCP) segment, especially if uncollateralised. We
significant amount of $205 billion over the whole sample, investigate the determinants of these price differentials in
or $630 million per day. Section 4.3.
We next report descriptive statistics at the trade level Finally, Table 2 reports the average credit rating of the
and across market segments in Table 2. The average no- buyer (paying fixed) and the seller (receiving fixed) of a
tional of CCP transactions is $79 million, which is about swap contract. We use the credit ratings to proxy for coun-
80% larger than that of the typical non-CCP transaction, terparty credit risk, which affects investors’ trading deci-
which averages to $45 million and $40 million for col- sions (with whom to trade and where) and has implica-
lateralised and uncollateralised transactions, respectively. tions for the pricing of swaps, for example, via the CVA
The distribution of the notional traded is rather dispersed channel. Because these ratings are typically available at the
though, with a standard deviation of more than $120 mil- quarterly frequency, we match the buyer and seller of a
lion in all segments. Interest rate swaps are long-lived. CCP swap with their credit ratings in the previous quarter.21 So,
contracts have an average maturity of ten years, which
is more than one year longer than that of non-centrally
cleared swaps. We observe some bunching around 5- and 20
We replicated our regression analyses by computing returns using the
10-year maturities in all segments. same end-of-day swap rate mid-quote from Bloomberg and we find sim-
Table 2 also displays the statistics of the average swap ilar results in terms of OTC premia.
21
As an alternative measure of credit risk, we used the credit default
rate return per trade. Note that this return is calcu-
swaps and default probability estimates from Kamakura Corporation. The
lated against the end-of-day swap rate mid-quote from main advantage of these two credit risk proxies is the daily frequency.
//doi.org/10.1016/j.jfineco.2019.09.010
JID: FINEC
Table 2
Summary statistics by trade — breakdown by segment.
This table reports summary statistics (by trade) of the main variables used in our analysis split by seg-
ment. CCP denotes trades cleared through a central counterparty, while Non-CCP denotes non-centrally
cleared trades. The latter segment consists of collateralised (Non-CCP with margin) and uncollateralised
(Non-CCP no margin) trades. Notional is the dollar amount (in $m) on which the exchanged interest pay-
ments are based. Maturity refers to the number of years between the effective and maturity date of the
swap contract. The swap return is defined as the difference (in bps) between the transaction-level swap
rate and the mid-quote of the Bloomberg benchmark rate at the end of the previous business day. The
credit rating is the average credit rating of the receiver (seller) or payer (buyer) of fixed rate from three
different sources (S&P, Moody’s, and Fitch). It has been converted to a scale from 0 to 20, where 0 de-
notes the worst rating and 20 the maximum rating (see Table A.2 in the Internet Appendix). The sample
covers every USD-denominated spot vanilla interest rate swap by non-US-based counterparties, which was
reported to DTCC between December 1, 2014 and February 19, 2016.
N Mean Std. dev.
Panel A: CCP
Notional ($m) 85,810 79.26 152.23
Log-notional 85,810 17.36 1.40
Maturity (years) 85,810 9.48 8.31
Swap return (bps) 85,803 1.78 16.67
Credit rating of buyer 57,593 14.66 1.20
Credit rating of seller 57,593 14.73 1.25
Panel B: Non-CCP
Notional ($m) 15,241 43.92 126.82
Log-notional 15,241 16.27 1.80
Maturity (years) 15,241 8.23 6.76
Swap return (bps) 15,237 2.27 20.70
Credit rating of buyer 3823 14.55 1.77
Credit rating of seller 3823 14.13 1.75
Panel C: Non-CCP with margin

Notional ($m) 11,658 45.05 122.85
Log-notional 11,658 16.40 1.68
Maturity (years) 11,658 8.20 6.77
Swap return (bps) 11,654 2.19 19.32
Credit rating of buyer 3,267 14.49 1.74
Credit rating of seller 3,267 14.20 1.74
Panel D: Non-CCP no margin

Notional ($m) 3,583 40.22 138.93
Log-notional 3,583 15.86 2.12
Maturity (years) 3,583 8.31 6.73
Swap return (bps) 3,583 2.52 24.69
Credit rating of buyer 556 14.94 1.92
Credit rating of seller 556 13.72 1.75
if a swap was executed on 10-Feb-2015, we use the ratings 4.2. Trading across OTC segments
(when available) of the two counterparties in Q4 2014. We
take the ratings from three different sources (Standard & The summary statistics hinted at some distinctive
Poor’s (S&P), Moody’s, and Fitch) and average them after characteristics of derivatives traded in the different OTC
first converting them to a scale from zero to 20, where segments. Next we explore these patterns more formally
zero denotes the worst rating and 20 the maximum rat- using a probit regression approach. More specifically, we
ing (see Table A.2 in the Internet Appendix for the conver- examine whether the likelihood to trade in the CCP or the
sion of each rating). Table 2 shows that credit ratings are non-CCP segment increases with some contract charac-
available only for a subset of counterparties, and largely for teristics, counterparty type, and counterparty credit risk.
dealers and banks. The average rating is about 14, which Note that in our data counterparties (CCP members and
corresponds to a rating of A- for S&P and Fitch, and A3 those who have access to client clearing services) have the
for Moody’s. The average rating varies from 14.49 to 14.94 option to choose where to trade throughout the sample
for buyers and 13.72 to 14.73 for sellers across market seg- period, because even if they are clearing members, they
ments, and the standard deviation corresponds to more are not subject to any central clearing obligation.
than one ‘notch’ (a unit in our scale of ratings corresponds We start with investigating the role of contract char-
to about one notch). acteristics. We present the first set of results in Table 3,
where the dependent variable is one for non-centrally
cleared trades and zero otherwise. The first variables of
interest, the log-notional and the maturity, both have a
However, they cover a smaller subset of entities for which we obtain sim-
ilar results.
negative sign in specification 1. This result shows that
//doi.org/10.1016/j.jfineco.2019.09.010
JID: FINEC
Table 3 for non-centrally cleared trades and zero otherwise. On

Determinants of (de)centralised clearing — contract characteristics.
one end of the spectrum, the more active counterparties
This table reports the results of a probit regression, where the de-
pendent variable is one for non-centrally cleared trades and zero other- tend to centrally clear their transactions, as indicated by
wise. Log-notional is the logarithm of the dollar amount on which the the negative coefficients of dealer, bank, and hedge-fund
contract’s exchanged interest payments are based. Maturity refers to the dummy variables. On the other end, nonfinancial or less
number of years between the effective and maturity date of the swap active counterparties have positive coefficients indicating
contract. Standardised maturity is a dummy variable that is one when the
that they are more disposed to non-CCP clearing.
contract’s maturity is 1Y, 2Y, 3Y, 4Y,..., 50Y and zero otherwise. Standard-
ised price is a dummy variable that is one when the contract price’s last Counterparties can also differ with respect to how and
digit is 5 or 0. For example, a 5Y swap with 1.05 fixed rate has a standard- at which cost they access the CCP infrastructure. CCPs can
ised maturity and price. The sample covers every USD-denominated spot be accessed either directly via house accounts for clear-
vanilla interest rate swap by non-US-based counterparties, which was re-
ing members (CM) or indirectly via client accounts by us-
ported to DTCC between December 1, 2014 and February 19, 2016. Robust
t-statistics are shown in brackets. ∗ , ∗∗ , ∗∗∗ denote significance at 1%, 5%,
ing the clearing services offered by CMs. To investigate
and 10% confidence level, respectively. whether direct or indirect access to CCPs matter for our
results, we construct two dummy variables that use infor-
(1) (2) (3)
mation on house and client accounts that we obtained di-
∗∗∗
Log-notional −0.281 −0.282∗ ∗ ∗ rectly from LCH, together with information on the list of
(−73.17) (−73.07)
clearing members, which we obtained from the website
Maturity −0.028∗ ∗ ∗ −0.028∗ ∗ ∗
(−36.54) (−35.98) of the Chicago Mercantile Exchange. The first variable is
Standardised maturity −0.216∗ ∗ ∗ −0.267∗ ∗ ∗ denoted as ‘CCP client,’ which equals one when a coun-
(−13.63) (−16.14) terparty clears via a client account, and zero otherwise;
Standardised price −0.143∗ ∗ ∗ −0.089∗ ∗ ∗
the second variable is denoted as ‘CM/CCP client,’ which
(−14.25) (−8.42)
Constant 3.985∗ ∗ ∗ −0.785∗ ∗ ∗ 4.264∗ ∗ ∗
equals one if the counterparty is either a clearing mem-
(59.02) (−51.94) (61.19) ber or has a client account with a CCP and zero other-
Pseudo R2 0.092 0.005 0.096 wise. Not surprisingly, the results from specifications 3 and
Obs 101,051 101,051 101,051 4 in Table 4 suggest that counterparties with indirect ac-
cess to CCPs are in relative terms less likely to centrally
clear their transactions, as shown by the positive coeffi-
trades with larger notional and longer maturity are in rel- cient of the ‘CCP client’ variable. In economic terms, the
ative terms more likely to be centrally cleared. We next results in specification 4 suggest that having an estab-
look at the role of contract standardisation. The basic idea lished access to a CCP, which is captured by the ‘CM/CCP
is that bilateral (or no) clearing might be best suited to client,’ increases the probability of central clearing by 23%,
more bespoke contracts. We define two types of standard- whereas having only indirect access reduces this effect by
isation with respect to maturity and price. A transaction 3%. This is perhaps a result of the higher CCP usage costs
is considered ‘standardised’ when its maturity is 1Y, 2Y, (e.g., fees for using the service) for investors with client
3Y, 4Y,..., 50Y or when the last digit of the price (fixed accounts, who in addition do not benefit from multilateral
rate) is 5 or 0. For example, a 5Y swap with 1.05 fixed netting.23
rate has a standardised maturity and price. Specification Another counterparty classification is with respect to
2 shows that contract standardisation is associated with their position in the trading network. At the core of the
more central clearing. The results remain qualitatively the network are the big dealers who speed up the search
same when all four variables are included in the same re- for counterparties by intermediating in the dealer-to-client
gression (specification 3). If we translate the findings of (D2C) market, while they rebalance their inventories in
specification 3 into probability terms, then they suggest the interdealer (D2D) market. The IRS interdealer market
that a one-unit change in (log) notional or maturity in- is traditionally considered a centrally cleared one, with
creases the probability of a CCP trade by 6% and 0.6%, re- LCH having a predominant role. Specifications 5 and 6 in
spectively.22 Similarly, having a standardised maturity or Table 4 confirm the anecdotal evidence as they suggest
price, increases the probability of a CCP trade by 5.6% and that trading in the D2C market reduces the probability of
1.9%, respectively. However, most of the explanatory power central clearing by 21% (specification 6).
(i.e., R2 ) comes from the notional and maturity, and it is Finally, we look at the role of counterparty credit risk
for this reason that we use only these variables as controls at the investor and market levels. In a similar probit
in the following probit regressions. regression setting as before, we first look at the associa-
We next examine the role of counterparty characteris- tion between central clearing and investors’ credit ratings,
tics. As seen already in the descriptive statistics in Table 1, which we use to proxy counterparty-specific credit risk.
dealers and other banks rely more on the CCP segment, Table 5 reports the results of these regressions for the
while small financial and nonfinancial companies tend to
use more often the non-CCP segment. Specifications 1 and
2 in Table 4 make this point more formally. As before, 23
The treatment of the initial margin posted by client accounts in the
the dependent variable in the probit regressions is one leverage ratio of the CMs that offer this service is another consideration,
as it increases the cost of offering clearing services, making client clear-
ing more expensive. This cost might become more material after the im-
22
We calculate this probability as the sample average of the marginal plementation of the clearing mandate and the subsequent increase in de-
effect of changes in a regressor on the conditional probability that a trade mand for client clearing services (given also the high concentration in the
is centrally cleared (e.g., Cameron and Trivedi, 2010). provision of client clearing).
//doi.org/10.1016/j.jfineco.2019.09.010
JID: FINEC
Table 4
Determinants of (de)centralised clearing — counterparty characteristics.
This table reports the results of a probit regression, where the dependent variable is one for non-centrally cleared trades and zero
otherwise. The regression is run at the trade and counterparty level for specifications (1) to (4) and trade-only level for specifications
(5) to (6). Log-notional is the logarithm of the dollar amount on which the contract’s exchanged interest payments are based. Maturity
refers to the number of years between the effective and maturity date of the swap contract. The next seven regressors are dummy
variables that capture the different counterparty types. For example, Asset manager equals one if the counterparty is an asset manager
and zero otherwise. CM/CCP client is one if the counterparty is either a clearing member or has a client account with a CCP and zero
otherwise. CCP client is one if the counterparty has a client account with a CCP and zero otherwise. D2C is a dummy variable that is
one if the transaction is executed in the dealer-to-client segment and zero otherwise. The sample covers every USD-denominated spot
vanilla interest rate swap by non-US-based counterparties, which was reported to DTCC between December 1, 2014 and February 19,
2016. Robust t-statistics are shown in brackets. ∗ , ∗∗ , ∗∗∗ denote significance at 1%, 5%, and 10% confidence level, respectively.
(1) (2) (3) (4) (5) (6)
Log-notional −0.271∗ ∗ ∗ −0.255∗ ∗ ∗ −0.237∗ ∗ ∗

(−97.22) (−91.52) (−62.88)
Maturity −0.028∗ ∗ ∗ −0.027∗ ∗ ∗ −0.023∗ ∗ ∗
(−50.52) (−47.90) (−28.46)
Asset manager 0.158∗ ∗ ∗ −0.037
(7.50) (−1.59)
Bank −0.458∗ ∗ ∗ −0.420∗ ∗ ∗
(-35.11) (−30.38)
Dealer −0.591∗ ∗ ∗ −0.475∗ ∗ ∗
(−51.79) (−38.64)
Hedge fund −0.848∗ ∗ ∗ −0.510∗ ∗ ∗
(−37.80) (−21.05)
Insurance 1.356∗ ∗ ∗ 1.562∗ ∗ ∗
(34.53) (35.59)
Nonfinancial 1.595∗ ∗ ∗ 1.764∗ ∗ ∗
(13.53) (13.75)
Other financial 1.561∗ ∗ ∗ 1.656∗ ∗ ∗
(32.85) (33.24)
CM/CCP client −1.267∗ ∗ ∗ −1.195∗ ∗ ∗
(−133.66) (−120.35)
CCP client 0.173∗ ∗ ∗ 0.134∗ ∗ ∗
(15.97) (11.67)
D2C 1.227∗ ∗ ∗ 1.089∗ ∗ ∗
(78.55) (67.01)
Constant −0.582∗ ∗ ∗ 4.168∗ ∗ ∗ 0.003 4.511∗ ∗ ∗ −1.998∗ ∗ ∗ 2.320∗ ∗ ∗
(−55.49) (84.11) (0.36) (90.26) (−136.59) (33.96)
Pseudo R2 0.057 0.137 0.104 0.176 0.104 0.164
Obs 202,102 202,102 202,102 202,102 101,051 101,051
counterparties for which credit rating information was to central clearing services, the results are qualitatively
available.24 Overall, specification 1 shows that higher the same (see Table A.3 in the Internet Appendix).
credit quality goes hand in hand with higher likelihood However, different categories of credit rating may have
of centrally cleared trades. One explanation could be the different effects. For instance, the non-investment-grade
amount of margin or funding cost for margin, as they prevents market participants from accessing central clear-
are both higher for less creditworthy counterparties. It ing services thereby inducing nonlinear patterns or oppo-
should be noted that among these counterparties there are site effects for investment-grade companies. To test this
only few without access to the CCP (clearing members or conjecture, we augment the model with various sets of
clients). More precisely, they account for only about 0.9% dummy variables that equal one for different buckets of
(0.3%) in terms of number of trades (notional). Therefore, credit ratings. In the regression specifications 2–4 re-
our results are more likely to arise from (endogenous) ported in Table 5, the dummy variables are organised in
counterparty choice rather than some institutional rigidi- three ranges of credit ratings: below 11 (that is, “non-
ties (e.g., denied access to central clearing services for investment-grade”), from 11 to less than 15 (“medium
non-investment-grade firms).25 Indeed, if we exclude from grade”), and above 15 (“high grade”). These three buck-
the regression the counterparties that do not have access ets include about 0.2%, 48.8%, and 51.2% of the total num-
ber of transactions (0.1%, 49%, and 50.9% of notional),
respectively. The results show that transactions with non-
24
Note that credit rating information is mainly available for dealers and investment-grade counterparties have a higher likelihood
banks, which explains the drop in the number of observations when we (16% more) of being cleared bilaterally than the average
include them in the regressions. These entities trade frequently and have transaction. In contrast, medium-grade counterparties have
well-established CCP relationships (via home or client accounts).
25
only a 1% more probability to clear bilaterally while high-
Survey data show that some clients face difficulties in establishing an
account with a provider of central clearing services; see Financial Stability
grade counterparties are more likely to centrally clear their
Board (2018). trades.
//doi.org/10.1016/j.jfineco.2019.09.010
JID: FINEC
Table 5
Determinants of (de)centralised clearing — counterparty credit risk.
This table reports the results of a probit regression, where the dependent variable is one for non-centrally cleared trades and zero
otherwise. The regression is run either at the trade and counterparty level (Columns 1 to 4) or only the trade level (Columns 5 and
6). Log-notional is the logarithm of the dollar amount on which the contract’s exchanged interest payments are based. Maturity refers
to the number of years between the effective and maturity date of the swap contract. Credit rating is the counterparty’s credit rating
from three different sources (S&P, Moody’s, and Fitch). It has been converted to a scale from 0 to 20, where 0 denotes the worst rating
and 20 the maximum rating (see Table A.2 in the Internet Appendix). VIX is the volatility implied by S&P 500 index options. Libor-OIS
spread is the difference between the three-month USD Libor rate and the Overnight Index Swap rate, which is commonly used as
a proxy for the cost of funding. The changes in the last two variables are used to capture market stresses. The sample covers every
USD-denominated spot vanilla interest rate swap by non-US-based counterparties, which was reported to DTCC between December 1,
2014 and February 19, 2016. Robust t-statistics are shown in brackets. ∗ , ∗∗ , ∗∗∗ denote significance at 1%, 5%, and 10% confidence level,
respectively.
(1) (2) (3) (4) (5) (6)
Log-notional −0.294∗ ∗ ∗ −0.293∗ ∗ ∗ −0.295∗ ∗ ∗ −0.295∗ ∗ ∗ −0.281∗ ∗ ∗ −0.282∗ ∗ ∗

(−83.73) (−83.54) (−83.83) (−83.83) (−73.17) (−73.17)
Maturity −0.031∗ ∗ ∗ −0.031∗ ∗ ∗ −0.031∗ ∗ ∗ −0.031∗ ∗ ∗ −0.028∗ ∗ ∗ −0.028∗ ∗ ∗
(−44.78) (−44.53) (−44.92) (−44.93) (−36.52) (−36.55)
Credit rating −0.024∗ ∗ ∗
(−6.69)
0 ≤ Rating < 11 0.884∗ ∗ ∗
(12.55)
11 ≤ Rating < 15 0.072∗ ∗ ∗
(8.40)
15 ≤ Rating ≤ 20 −0.083∗ ∗ ∗
(−9.76)
VIX changes 0.238∗ ∗ ∗
(4.37)
Libor-OIS spread changes 0.428∗ ∗ ∗
(4.08)
Constant 4.416∗ ∗ ∗ 4.059∗ ∗ ∗ 4.049∗ ∗ ∗ 4.126∗ ∗ ∗ 3.985∗ ∗ ∗ 3.988∗ ∗ ∗
(54.83) (65.78) (65.56) (66.55) (59.01) (59.02)
Pseudo R2 0.093 0.094 0.093 0.093 0.092 0.092
Obs 162,348 162,348 162,348 162,348 101,051 101,051
Another important issue is whether the decision to cen- 4.3. Determinants of OTC premia
trally clear or not changes in times of stress. Although the
CCP segment is predominant under normal market condi- We now turn to the pricing of IRSs across the differ-
tions, there can be an increase in trading outside CCPs as ent OTC segments. An IRS is a plain vanilla swap contract
a consequence of the elevated cost of margin in periods that consists of a fixed leg whose payments depend on a
of stress, known as margin procyclicality. To shed light fixed rate and a floating leg whose payments depend on
on this issue, we augment the probit regression with two a floating rate (the Libor). It is used by traders to remove
proxy variables usually associated with market stress: the interest rate risk from their trading book. The seller of the
VIX index of volatility implied by S&P 500 index options, swap, i.e., the receiver of fixed, is selling protection against
which is typically used as a measure of global risk percep- interest rate risk and would bear a loss in case of upward
tions; and the spread between the three-month London interest rate movements. Therefore, the market value of
Inter-bank Offered Rate (Libor) and the Overnight Index the contract, which is typically zero at initiation, is driven
Swap rate, which is often used as an indicator of funding by changes in interest rates (the term structure), which is
strains in the interbank market. Specifications 5 and 6 the main risk factor. In line with the literature on deal-
of Table 5 indeed suggest that higher readings of these ers’ inventory effects (e.g., Amihud and Mendelson, 1980)
variables tend to be associated with a higher likelihood and bargaining power (e.g., Duffie, 2012), dealers are in-
of bilateral clearing. One possible explanation is that pro- centivised to pass any regulatory and XVA-related costs to
cyclicality causes funding liquidity pressure to parties that clients. To test this, we focus on transactions between deal-
need to find liquid and eligible assets to post as margin, ers and clients, that is, excluding the interdealer segment.
at times when it is most difficult for them to do so (see Given the OTC nature of trading in swap markets, there
Brunnermeier and Pedersen, 2008; Murphy et al., 2014). is little information about the fair value of a swap be-
To the extent that credit risk and funding risk are inter- fore the trade. The market convention is to use an in-
related, this explanation is consistent with the previous dustry benchmark, provided by financial data vendors like
result indicating that non-investment-grade counterparties Bloomberg, as a proxy for the ‘fair’ value of the prevailing
tend to clear more bilaterally. Overall, our results support fixed rate. In line with this practice, we use Bloomberg
the view that market participants might circumvent CCPs’ end-of-day information to construct the ‘swap return,’ for-
stricter margining practices and adverse effects such as mally defined as the difference between the transaction-
funding liquidity stress by clearing bilaterally. level swap rate and the mid-quote of the Bloomberg
//doi.org/10.1016/j.jfineco.2019.09.010
JID: FINEC
Table 6
OTC premium in dealer-to-client segment.
The table reports the results of trade-level panel regressions of the swap return on the Non-CCP dummy that equals one for non-CCP trades and zero
otherwise, and a number of other variables and controls. We only include trades executed in the dealer-to-client segment. The swap return is defined as the
difference (in bps) between the transaction-level swap rate and the mid-quote of the Bloomberg benchmark rate at the end of the previous business day.
Non-CCP with IM is a dummy that equals one for non-centrally cleared trades that involve the exchange of initial margin by the two counterparties and
zero otherwise. CR and LR denote the capital ratio and the leverage ratio, respectively. The rest of the variables are defined in Eqs. (1) and (2). Columns 1–4
show the results when the dealer sells interest rate protection (receives fixed), while Columns 5–8 show the results when the dealer buys protection (pays
fixed). The sample covers every USD-denominated spot vanilla interest rate swap by non-US-based counterparties reported to DTCC between December 1,
2014 and February 19, 2016. All specifications include time, buyer ID, and seller ID fixed effects. We report t-statistics calculated using clustered standard
errors (by quarter and dealer ID) in parentheses. ∗ , ∗∗ , ∗∗∗ denote significance at 1%, 5%, and 10% confidence level, respectively.
Dealer receiving fixed Dealer paying fixed
(1) (2) (3) (4) (5) (6) (7) (8)
Non-CCP dummy 8.116∗ ∗ 8.424∗ ∗ ∗ 9.189∗ ∗ ∗ 8.621∗ ∗ ∗ 1.840 6.426 1.817 6.615
(2.52) (2.73) (2.65) (2.78) (0.88) (1.46) (0.84) (1.39)
Non-CCP with IM −3.800∗ ∗ −9.369∗ ∗ ∗ 0.729 −2.828
(−2.22) (−3.82) (0.21) (−0.51)
Log-notional 0.113 −0.022 0.104 −0.021 −0.288 −0.206 −0.287 −0.206
(0.78) (−0.12) (0.72) (−0.11) (−1.41) (−0.64) (−1.41) (−0.63)
Maturity 0.024 0.023 0.022 0.023 0.033 0.063 0.033 0.063
(0.91) (0.70) (0.85) (0.68) (1.20) (1.56) (1.20) (1.56)
Credit rating of buyer 0.431 0.684 0.425 0.685 −0.381 −0.592 −0.382 −0.586
(1.17) (1.10) (1.16) (1.10) (−0.66) (−0.75) (−0.66) (−0.74)
Credit rating of seller 0.375 0.275 0.393 0.268 0.203 0.542 0.204 0.541
(0.79) (0.45) (0.84) (0.44) (0.78) (0.74) (0.78) (0.74)
CR of dealer (payer) −0.846∗ ∗ ∗ −0.848∗ ∗ ∗
(−3.43) (−3.44)
CR of dealer (receiver) 0.006 0.012
(0.03) (0.06)
LR of dealer (payer) 0.226 0.227
(0.97) (0.97)
LR of dealer (receiver) 0.254 0.260
(1.49) (1.56)
Day, Buyer ID, Seller ID FE Yes Yes Yes Yes Yes Yes Yes Yes
2
R 0.177 0.163 0.178 0.163 0.267 0.279 0.267 0.279
Obs 12,092 7119 12,092 7119 16,193 8093 16,193 8093
benchmark rate at the end of the previous business day.26 some specifications, we control for additional dealer char-
Intuitively, the swap return captures the price divergence acteristics, that is, the capital ratio (‘CR’) and leverage ra-
from the swap rate’s fair value. tio (‘LR’).27 In all results here and below we report robust
To analyse the determinants of swap returns, we run standard errors clustered at the quarter and dealer levels,
panel regressions of the form unless otherwise stated.
Our main interest lies on the coefficient β of (1), which
Rk,i, j,t = β Non-CCP + γ Xk,i, j,t + δi + ζ j + θt + k,i, j,t , (1) should capture the OTC premium by measuring the differ-
where Rk,i,j,t is the swap return for transaction k between ence in the price (fixed rate) between CCP and non-CCP
counterparties i and j on day t; ‘Non-CCP’ is a dummy vari- swaps after controlling for time, counterparty, and contract
able that is equal to one for any transaction that is not cen- characteristics. If β is positive then a swap executed in the
trally cleared (either client cleared or directly cleared by a non-CCP segment will be costlier than an equivalent con-
CCP) and zero otherwise; Xk,i,j,t is a vector of transaction- tract that is centrally cleared, in the sense that the buyer
level contract and counterparty characteristics; and δ i , ζ j , of the swap will have to pay a higher fixed rate.
and θ t denote fixed effects. Xk,i,j,t includes the logarithm Table 6 shows the main results of IRS return determi-
of the trade size (‘Log-notional’), the contract’s maturity nants when the dealer receives fixed (that is, sells inter-
(‘Maturity’), and the credit ratings of the counterparties. In est rate protection to the client) and when the dealer pays
fixed (that is, buys protection from the client). We discuss
first the results for the former case reported in Columns 1–
26
The Bloomberg end-of-day rates are composite measures of indica- 4. This analysis provides evidence for the existence of the
tive quotes from a selected number of ‘contributors’ (dealers), and as such
OTC premium. That is, the swap return for non-CCP trans-
they reflect the daily market average ‘risk-neutral’ prices. One may won-
der whether non-CCP transactions tend to take place systematically be- actions is about 8 basis points larger than that of centrally
fore or after the centrally cleared ones, and therefore bias our results. We
report the empirical distributions, grouped in hourly baskets, of the two
27
different types of transactions in Fig. A.1 in the Internet Appendix. Visual Note that the credit rating information is mainly but not only avail-
inspection of the empirical distributions shows that there is no noticeable able for banks. This is why there is a reduction of number of observations
difference in the two distributions. We also control more formally for this in Table 6 and Table 8 relative to the previous tables. In addition, the cap-
potential issue by running a robustness exercise by using hour fixed ef- ital and leverage ratios are only available for a subset of dealers, and it is
fects absorbing intraday patterns. The results, which we report in Table for this reason that in Table 6, Table 7, and Table 8, Columns 2, 4, 6 and
A.4 in the Internet Appendix, show that our results are robust. 8, there is a further reduction of the number of observations.
//doi.org/10.1016/j.jfineco.2019.09.010
JID: FINEC
Table 7
OTC premium in dealer-to-client segment, when clients are not banks.
The table reports the results of trade-level panel regressions of the swap return on the Non-CCP dummy that equals one for non-centrally cleared trades
and zero otherwise, and a number of other variables and controls. We only include trades executed in the dealer-to-non-bank-client segment. Non-banks
consist of hedge funds, asset managers, insurance, and nonfinancial firms. The swap return is defined as the difference (in bps) between the transaction-
level swap rate and the mid-quote of the Bloomberg benchmark rate at the end of the previous business day. Non-CCP with IM is a dummy that equals
one for non-centrally cleared trades that involve the exchange of initial margin by the two counterparties and zero otherwise. CR and LR denote the capital
ratio and the leverage ratio, respectively. The rest of the variables are defined in Eqs. (1) and (2). Columns 1–4 show the results when the dealer sells
interest rate protection (receives fixed), while Columns 5–8 show the results when the dealer buys protection (pays fixed). The sample covers every USD-
denominated spot vanilla interest rate swap by non-US-based counterparties, which was reported to DTCC between December 1, 2014 and February 19,
2016. All specifications include time, buyer ID, and seller ID fixed effects. We report t-statistics calculated using clustered standard errors (by quarter and
dealer ID) in parentheses. ∗ , ∗∗ , ∗∗∗ denote significance at 1%, 5%, and 10% confidence level, respectively.
Dealer receiving fixed Dealer paying fixed
(1) (2) (3) (4) (5) (6) (7) (8)
Non-CCP dummy 4.148∗ ∗ 4.190 4.159∗ ∗ 3.739∗ −0.081 −2.762∗ −0.028 −0.005
(2.03) (1.44) (2.03) (1.77) (-0.08) (−1.74) (−0.03) (−0.01)
Non-CCP with IM −1.933 −1.467 −5.714 −5.796
(−1.50) (−1.37) (−0.73) (−0.75)
Log-notional 0.356∗ ∗ 0.200 0.356∗ ∗ 0.360∗ ∗ 0.246 −0.054 0.247 0.287
(2.11) (0.84) (2.10) (2.04) (1.26) (−0.20) (1.27) (1.48)
Maturity 0.029 0.029 0.029 0.023 0.036 0.142∗ ∗ ∗ 0.037 0.037
(0.86) (0.52) (0.86) (0.64) (0.65) (2.81) (0.65) (0.66)
Credit rating of buyer −1.359 −0.604
(−0.87) (−0.71)
Credit rating of seller 0.956 1.462∗ ∗
(0.80) (2.41)
CR of dealer (buyer) −0.246 −0.440
(−0.32) (−1.15)
CR of dealer (seller) 0.554 0.779
(1.02) (1.60)
LR of dealer (buyer) 0.252
(0.80)
LR of dealer (seller) 0.166
(0.79)
2
R 0.260 0.276 0.260 0.260 0.286 0.359 0.286 0.253
Obs 17,651 10,566 17,651 16,979 13,394 5264 13,394 13,016
cleared transactions, controlling for other variables. Besides This model includes a dummy variable (‘Non-CCP with IM’)
being strongly statistically significant, this OTC premium that equals one when initial margin is exchanged (i.e., full
is also economically significant: given the average fixed collateralisation), and zero otherwise. Columns 3 and 4 in
rate in our sample period of about 1.82%, the β coefficient Table 6 show that the exchange of initial margin in the
suggests that the fixed rate of non-CCP swaps is on aver- non-CCP segment reduces the swap return by about 4–9
age 4% higher than that of equivalent CCP swaps. Columns basis points. This result points to the pricing of the coun-
5–8 deliver a second important result, that is, when the terparty credit risk in non-centrally cleared swaps, which
dealer pays fixed the premium is smaller and not statis- is in line with the CVA and KVA valuation adjustments.
tically significantly different from zero. In case of maxi- These findings support the empirical predictions discussed
mum dealers’ bargaining power, we could have expected in Section 2.1: the posting of initial margin reduces (i) the
negative OTC premia if dealers paying the fixed fully re- loss given default and (ii) the exposure at default, which
verse XVA charges to customers. This asymmetric pattern reduce the cost to buy protection against counterparty de-
can arise when the interest rate risk is mainly on the up- fault and Basel III capital charges. The same result also sug-
per side, as it was during our sample period. In such a gests that the funding cost associated with the posting of
market environment, the payer of fixed (including deal- initial margin (MVA) is smaller compared to the CVA and
ers) is willing to pay a premium to protect herself against KVA, at least during our sample period.28 Another obser-
the interest rate risk while the receiver of fixed is likely vation is that margin does not play a significant role when
to have an advantage at setting prices, in which case per- dealers pay fixed (buy protection).
haps also making harder the identification of the XVA In the results so far, we included both non-dealer
effects. banks and non-banks in the clients category. Next, we split
According to the empirical prediction discussed in clients into banks and non-banks, as different XVA effects
Section 2.1, the OTC premium should be lower when initial can arise from the latter group. Table 7 reports the results
margin is posted. We test this hypothesis using variations of the analogous regressions described above, but where
of the following empirical model:
Rk,i, j,t = β1 Non-CCP + β2 Non-CCP with IM 28

The move to a higher interest rate environment and the forthcom-
ing implementation of the more stringent uncleared margining rules are
+ γ X k,i, j,t + δi + ζ j + θt + k,i, j,t . (2) likely to increase the future impact of the MVA.
//doi.org/10.1016/j.jfineco.2019.09.010
JID: FINEC
the sample includes only transactions in which the client is counterparty characteristics appear to explain most of the
a non-bank. In these regressions the sample size increases, OTC premium as the coefficient of the non-CCP dummy
because we have dropped the client-specific controls (i.e., variable in all specifications becomes not statistically dif-
credit rating, capital ratio, and leverage ratio) as they are ferent from zero. When the dealer receives fixed, the larger
only available for banks. We continue to control for dealer the notional and the longer the maturity are, the larger the
specific variables and client and dealer time-invariant char- swap return is in the non-CCP segment. Conversely, when
acteristics. the dealer pays fixed, the larger the notional the larger
The results reported in Table 7 are consistent with the the discount, but the coefficients are generally not very
previous findings. When dealers receive the fixed rate, we significant in statistical terms.
still find a positive and significant OTC premium corrobo- More importantly, the results in Columns 1–4 point
rating the bargaining power hypothesis. As discernible in to the important role of the creditworthiness of dealers’
Columns 1–4, the coefficient of the ‘Non-CCP dummy’ is counterparties. When the dealer sells (receives fixed), a
smaller in magnitude than that reported in Table 6 for more creditworthy customer enjoys a lower price, whereas
three possible reasons: first, the sample of market par- when the dealer buys (pays fixed), a more creditworthy
ticipants is different; second, different XVAs can arise customer receives a better price. The credit rating coeffi-
from non-banks; third, the non-bank subsample includes cients of dealers’ customers are statistically and econom-
firms exempted from Basel III capital charges, leading to ically significant: for instance, an A+ customer pays (re-
a smaller KVA. We elaborate upon the latter in the next ceives) about 4 basis points lower (higher) swap rate than
section. Interestingly, Columns 5–8 show a negative co- what an A- customer does (the difference in ratings is two
efficient for the ‘Non-CCP dummy’ when dealers pay the notches) in the non-CCP segment.29
fixed rate, although statistically significant only in Column All in all, our findings shed light on the existence of
6. This reinforces the idea that dealers exercise their bar- OTC premia (i) generating more expensive non-CCP trades,
gaining power and pass on their costs to non-bank clients. (ii) decreasing with initial margins, and (iii) favouring deal-
That is, dealers tend to fully charge XVA costs that trans- ers. Our results point to the pass-through of valuation
late into a positive (negative) OTC premium when they adjustments from dealers to customers, in particular the
receive (pay) fixed. When they pay fixed, the discount is CVA and KVA valuation adjustments.
about 3 basis point in specification 6, where we control
for all dealer characteristics. In line with the pricing of CVA
4.4. Other results
and KVA, when the dealer receives the fixed and the ini-
tial margin is posted, the OTC premium drops by about
In this subsection, we extend our analysis in two direc-
2 basis points. The IM results are less significant (the p-
tions: first, we dig into the regulatory effects on IRS pricing
value is only 12% in Column 3) for non-banks, which is
by analysing the subset of firms exempted from the CVA
not surprising given that non-banks tend to post margin
capital charge (which is related to KVA). Second, we con-
less frequently (for example, they are linked to thresh-
duct additional tests to address the question whether other
olds and minimum transfers), resulting in a smaller reduc-
factors such as market liquidity and dealer-to-customer
tion in CVA and KVA. When dealers pay fixed the role of
relationships can explain the OTC premium.
margin is not significant.
The main idea behind our analysis is that heterogeneity
The results in Tables 6 and 7 are robust to the exclu-
in derivative prices arises from a pass-through of dealers’
sion of entities with unclassified location, which consti-
inventory costs including regulatory charges on to market
tute about 3% of notional traded in our sample. When we
prices via XVA effects. Although our empirical findings are
perform the main regressions excluding these entities, we
overall consistent with this conjecture, it is difficult to pro-
obtain very similar results. More specifically, the results
vide direct evidence of it for at least two reasons: first,
in Table 6 are actually unchanged, and the reason is that
there is no unique regulatory event in our sample period
the counterparties without location do not have credit rat-
determining a clear change in behaviour of market par-
ings and therefore these entities were never included in
ticipants.30 Second, it is difficult to disentangle XVA ef-
the regressions. The results for non-banks are slightly af-
fects; this is especially true for KVA and CVA because they
fected (see Table A.5 in the Internet Appendix), but they
depend on similar factors (see Section 2).31 However, the
remain quantitatively similar. In fact, the premium when
asymmetric effect of Basel III on the different types of mar-
dealers sell protection (receive the fixed) is significant and
ket participants, gives rise to an additional test to validate
positive, as before. However, when dealers pay the fixed
our conjecture. In fact, the European regulation exempted
rate the premium is not statistically significant (while it
dealers’ transactions from the calculation of the additional
is slightly statically significant in Table 7, Column 6). All
CVA capital charge creating KVA when they traded with
things considered, the main conclusions remain essentially
unchanged.
Table 8 digs deeper into the determinants of the OTC 29
Note that we do not include any interactions between the non-CCP
premium. We interact the non-CCP dummy variable with dummy and ratings in Columns 5–8, because ratings are not available for
the possible determinants in terms of contract and coun- non-banks.
30
terparty characteristics. The coefficients on the interac- Most of the new banking regulations were already in the implemen-
tation stage when OTC derivatives data became available to regulators (in
tion terms indicate the effect of notional, maturity, and Europe in December 2014).
buyer/seller credit rating conditional on the contract being 31
Two-third of the respondents to a survey by Risk magazine in 2015
transacted on the non-CCP segment. These contract and believed that overlaps between the KVA and CVA do exist.
//doi.org/10.1016/j.jfineco.2019.09.010
JID: FINEC
Table 8
OTC premium in dealer-to-client segment — determinants.
and zero otherwise, and a number of interaction variables and controls. We report separate results for the dealer-to-client and the dealer-to-non-bank-
client segment and for when the dealer sells (receives fixed) or buys (pays fixed) interest rate protection. The swap return is defined as the difference (in
bps) between the transaction-level swap rate and the mid-quote of the Bloomberg benchmark rate at the end of the previous business day. CR and LR
denote the capital ratio and the leverage ratio, respectively. The rest of the variables are defined in Eq. (1). The sample covers every USD-denominated spot
vanilla interest rate swap by non-US-based counterparties, which was reported to DTCC between December 1, 2014 and February 19, 2016. All specifications
include time, buyer ID, and seller ID fixed effects. We report t-statistics calculated using clustered standard errors (by quarter and dealer ID) in parentheses.
∗ ∗∗ ∗∗∗
, , denote significance at 1%, 5%, and 10% confidence level, respectively.
Dealer-to-client Dealer-to-client (non-bank)
Dealer receiving fixed Dealer paying fixed Dealer receiving fixed Dealer paying fixed
(1) (2) (3) (4) (5) (6) (7) (8)
Non-CCP dummy 11.798 −36.498 −29.896 −91.455 −8.231 −13.610 −0.335 −1.250
(0.41) (−0.88) (−0.79) (−1.13) (−1.37) (−1.61) (−0.05) (−0.18)
Non-CCP × Log-notional 1.033 2.492 −1.267 −0.338 0.697∗ ∗ 1.027∗ ∗ 0.051 −0.026
(1.30) (1.61) (−0.95) (−0.15) (2.12) (2.21) (0.14) (−0.08)
Non-CCP × Maturity 0.494 1.255∗ ∗ 0.500 2.220∗ ∗ ∗ 0.040 0.013 −0.077 −0.124
(1.55) (2.14) (1.25) (3.16) (0.58) (0.11) (−1.01) (−1.51)
Non-CCP × Credit rating of buyer −2.054∗ −1.764∗ 2.030 2.923
(−1.73) (−1.72) (0.89) (0.73)
Non-CCP × Credit rating of seller 0.651 1.572 1.374∗ ∗ 2.899∗
(0.90) (1.38) (2.22) (1.73)
Log-notional 0.019 −0.119 −0.158 −0.195 0.200 −0.044 0.242 −0.033
(0.13) (−0.73) (−1.03) (−0.84) (0.96) (−0.15) (1.09) (−0.11)
Maturity 0.000 −0.011 0.019 0.013 0.017 0.018 0.046 0.161∗ ∗
(0.02) (−0.44) (0.93) (0.44) (0.43) (0.28) (0.73) (2.65)
Credit rating of buyer 0.433 0.778 −0.608 −0.762 −1.359
(1.16) (1.24) (−1.02) (−1.05) (−0.86)
Credit rating of seller 0.222 0.097 0.242 0.497 0.951
(0.45) (0.16) (0.88) (0.69) (0.79)
CR of dealer (payer) −0.841∗ ∗ ∗ −0.260
(-3.23) (−0.34)
CR of dealer (receiver) −0.049 0.537
(−0.22) (0.99)
(1.11) (0.80)
(1.48) (0.79)
2
R 0.182 0.174 0.271 0.298 0.261 0.277 0.286 0.359
Obs 12,092 7119 16,193 8093 17,651 10,566 13,394 5264
pension funds, nonfinancial corporates below the clearing dealer’s counterparty is an exempted entity.33 Note that to
threshold, and public bodies.32 perform this analysis, we focus on the dealer-to-non-bank
This exemption allows us to test the following hypoth- client segment and remove 30 trades by non-EU dealers, as
esis: if KVA drives the OTC premia, we expect them to be they were unaffected by the exemption. The regression re-
smaller or disappear when European dealers trade against sults, which are summarised in Table 9, clearly show that
the exempted firms. A relevant feature of the exempted there is a significant reduction in the OTC premium for ex-
firms is that they rarely exchange initial margin (in less empted firms. Even if there can be other XVA effects in-
than 0.2% of their trades in our sample), suggesting that volved, this result is consistent with the idea that KVA ef-
dealers remain exposed to some degree of counterparty fects tend to materialise only if firms are actually affected
risk when they trade against them. Thus, the absence of by the new regulation. It is worth noting that this re-
initial margins sharpens the identification of KVA because gression includes fixed effects controlling for the (smaller)
any evidence of reduction in the premium associated to size or other unobservable characteristics of the exempted
exempted firms should be attributed to regulation-driven firms. Since the exempted firms do not exchange initial
KVA rather than CVA. margin, which would otherwise have reduced counterparty
We test this hypothesis by interacting the non-CCP vari- risk, the new results support the pricing of KVA.
able in Eq. (1) with a dummy that equals one when the
33
We classified exempted entities as follows. Pension funds consist of
all firms with the words “pension,” “retirement,” or “retirement scheme”
in their name. Because there is no public information on clearing thresh-
32
Regulation (EU) No 575/2013 (Article 382, point 4) and Regulation olds, we consider all nonfinancial counterparties as exempted. Finally, we
(EU) No 648/2012 (Article 1, point 4; and Article 2, point 10) of the Euro- select the remaining exempted entities manually. Overall, we identified
pean Parliament and of the Council. 1830 trades from about 80 exempted entities.
//doi.org/10.1016/j.jfineco.2019.09.010
JID: FINEC
Table 9
OTC premium in dealer-to-client segment — liquidity and exempted firms.
and zero otherwise, and a number of other variables and controls. We only include trades executed in the dealer-to-client segment. The swap return is
defined as the difference (in bps) between the transaction-level swap rate and the mid-quote of the Bloomberg benchmark rate at the end of the previous
business day. CR and LR denote the capital ratio and the leverage ratio, respectively. The other variables are defined in Eq. (1). Columns 1–4 show the
results when the SEF variable, which equals one for centrally cleared trades executed on swap execution facilities and zero otherwise, is included. SEF
captures differences in market liquidity. Columns 5–10 show the results for dealer-to-non-bank client segment, when we interact the non-CCP dummy
with another dummy that is equal to one when the firm is exempted from the calculation of the CVA capital charge. We report results for when the dealer
sells (receives fixed) or buys (pays fixed) interest rate protection. The sample covers every USD-denominated spot vanilla interest rate swap by non-US-
based counterparties, which was reported to DTCC between December 1, 2014 and February 19, 2016. Unless otherwise stated, specifications include time,
buyer ID, and seller ID fixed effects. We report t-statistics calculated using clustered standard errors (by quarter and dealer ID) in parentheses. ∗ , ∗∗ , ∗∗∗
denote significance at 1%, 5%, and 10% confidence level, respectively.
Liquidity Exempted firms
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)
∗∗ ∗∗ ∗∗ ∗
Non-CCP dummy 7.991 7.786 1.167 5.495 4.331 4.226 4.201 −1.038 −1.029 −2.727∗
(2.49) (2.54) (0.57) (1.29) (2.04) (1.97) (1.44) (−1.06) (−1.06) (−1.71)
SEF −3.580∗ ∗ ∗ −4.896∗ ∗ ∗ −4.556∗ ∗ ∗ −5.755∗ ∗ ∗
(−5.20) (−4.62) (−6.26) (−5.02)
Exempted∗ Non-CCP dummy −5.481∗ ∗ −5.265∗ ∗ −2.205 5.655∗ ∗ ∗ 5.641∗ ∗ ∗ −3.831
(−2.25) (−2.15) (−0.67) (3.75) (3.75) (−0.63)
Log-notional 0.155 0.070 −0.163 −0.046 0.353∗ ∗ 0.354∗ ∗ 0.199 0.246 0.247 −0.057
(1.07) (0.38) (−0.81) (−0.15) (2.09) (2.09) (0.84) (1.26) (1.27) (−0.21)
Maturity 0.033 0.037 0.051∗ 0.081∗ ∗ 0.029 0.028 0.028 0.036 0.036 0.144∗ ∗ ∗
(1.29) (1.15) (1.97) (2.20) (0.85) (0.82) (0.51) (0.65) (0.64) (2.81)
Credit rating of buyer 0.446 0.843 −0.560 −0.813 −0.055 −1.393
(1.19) (1.32) (−0.96) (−1.02) (−0.08) (−0.89)
Credit rating of seller 0.247 0.118 0.188 0.545 0.881∗ 0.956
(0.52) (0.19) (0.72) (0.75) (1.91) (0.80)
(−3.42) (−0.34)
CR of dealer (receiver) 0.023 0.553
(0.10) (1.02)
(0.96) (0.80)
(1.43) (0.79)
Day, Buyer ID, Seller ID FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
2
R 0.185 0.177 0.274 0.289 0.260 0.261 0.276 0.287 0.287 0.359
Obs 12,092 7119 16,193 8093 17,649 17,622 10,565 13,364 13,352 5242
We next turn the analysis to market liquidity, rela- hypothesis. Finally, the current specifications already in-
tionships, and bilateral exposures. By conducting these clude some controls for market liquidity such as the trade
additional tests, we can also address the question whether size (i.e., notional) and dealer characteristics (e.g., balance
there could be some unobservable variables providing sheet variables) accounting for the propensity to provide
additional explanations to the OTC premium. This unob- liquidity.
servable variable should affect prices at the trade level, i.e., We nonetheless proceed with a more direct test of the
it must be specific to each counterparty and to day/time. liquidity hypothesis. The basic idea is centered on the exe-
Our empirical strategy already controls for this possi- cution of swaps on centralised electronic platforms, which
bility by including time and counterparty fixed effects, earlier literature has shown are associated with better liq-
time-varying counterparty credit risk, as well additional uidity. These platforms were introduced in February 2014
controls for dealers’ balance sheets. Nonetheless, to err by the Dodd–Frank Act, which required US persons to ex-
on the side of caution, in this section we conduct some ecute centrally cleared interest rate swaps on multilateral
additional analysis. pre-trade transparent venues, known as Swap Execution
First, we consider the differences in market liquidity Facilities (SEF). By its very nature, the SEF mechanism
across segments. The idea that the OTC premium (fully) reduces many frictions such as information asymme-
depends on market illiquidity seems implausible for at try, search costs, and the dealer’s bargaining power
least three reasons. First, the effect from liquidity should by providing more trading options to their customers.
have been perfectly symmetric as dealers would charge Benos et al. (2019) show that indeed SEF trading reduced
half-spread every time they buy or sell a swap contract. effective spreads in the USD IRS market by about 25%.34
But our results clearly show that this is not the case.
Second, the finding that the premium is associated with 34
SEFs facilitate multilateral trading by a central limit order book and
the type of collateral in bilateral trades or the credit rat-
a request-for-quote functionality. See Benos et al. (2019) for the institu-
ing of clients is hard to be reconciled with the illiquidity tional details. Riggs et al. (2018) report that the use of limit order book on
//doi.org/10.1016/j.jfineco.2019.09.010
JID: FINEC
Table 10
OTC premium in dealer-to-client segment — relationships and bilateral exposures.
and zero otherwise, and a number of other variables and controls. We only include trades executed in the dealer-to-client segment. The swap return is
defined as the difference (in bps) between the transaction-level swap rate and the mid-quote of the Bloomberg benchmark rate at the end of the previous
business day. CR and LR denote the capital ratio and the leverage ratio, respectively. The other variables are defined in Eq. (1). Columns 1–4 show the
results when buyer ID and seller ID fixed effects are replaced by their interaction. Columns 5–8 report results when the buyer ID and seller ID fixed effects
are replaced by a month × buyer ID × seller ID fixed effect. Intuitively, this fixed effect captures bilateral dealer-client exposures that are assumed to
change monthly. We report results for when the dealer sells (receives fixed) or buys (pays fixed) interest rate protection. The sample covers every USD-
denominated spot vanilla interest rate swap by non-US-based counterparties, which was reported to DTCC between December 1, 2014 and February 19,
2016. We report t-statistics calculated using clustered standard errors (by quarter and dealer ID) in parentheses. ∗ , ∗∗ , ∗∗∗ denote significance at 1%, 5%, and
10% confidence level, respectively.
Relationships Bilateral exposures (monthly)
(1) (2) (3) (4) (5) (6) (7) (8)

∗∗∗ ∗∗∗ ∗∗ ∗∗
Non-CCP dummy 10.445 12.993 0.120 5.137 13.294 15.390 −0.862 1.657
(2.67) (2.82) (0.05) (0.87) (2.13) (2.10) (−0.20) (0.18)
Log-notional 0.054 −0.054 −0.275 −0.163 0.094 0.080 −0.446∗ ∗ −0.560
(0.33) (−0.31) (−1.32) (−0.52) (0.54) (0.35) (−2.19) (−1.68)
Maturity 0.001 0.015 0.035 0.068∗ 0.000 0.001 0.003 0.017
(0.04) (0.45) (1.30) (1.75) (0.00) (0.04) (0.10) (0.41)
Credit rating of buyer 0.331 0.645 −0.565 −1.046 0.762 −0.346 0.363 −8.054
(0.77) (0.95) (−0.84) (−1.11) (0.45) (−0.27) (0.28) (−0.24)
Credit rating of seller 0.198 0.009 0.273 1.491∗ ∗ −0.940 −5.348∗ 0.333 3.211
(0.40) (0.01) (0.97) (2.51) (−0.38) (−1.68) (0.39) (0.23)
(−2.84) (−0.47)
CR of dealer (receiver) −0.059 −1.814∗
(−0.24) (−1.79)
(0.52) (1.05)
LR of dealer (receiver) 0.265 −0.132
(1.59) (−1.60)
Day, (Buyer ID × Seller ID) FE Yes Yes Yes Yes No No No No
Day, (Month × Buyer ID × Seller ID) FE No No No No Yes Yes Yes Yes
2
R 0.236 0.188 0.353 0.332 0.432 0.361 0.505 0.490
Obs 12,092 7119 16,193 8093 12,092 7119 16,193 8093
Hence, if the OTC premium stems from market illiquidity, Another issue that might influence swap prices is the
then the former should become insignificant once we relationship between counterparties. For example, a dealer
control for the venue of execution (SEF trading). We might have stronger negotiating power with some clients
test for this by introducing a dummy variable, ‘SEF,’ in or offer them better prices if they do more business with
specification 1, which equals one for centrally cleared her. Additionally, they might have different agreements in
transactions executed on SEFs and zero otherwise.35 place to regulate their relationships with different clients,
Table 9 presents the (baseline) results of the specification for example, the agreements that regulate the collateral
with the SEF dummy. Columns 1–4 show that the SEF held by two parties (credit support annex, CSA). Although
dummy is negative and significant, which suggests that prior research has not analysed the IRS market, previous
prices of trades executed on SEFs are on average closer papers show that relationships matter in some other OTC
to the end-of-day Bloomberg benchmark price. More markets (e.g., Green et al., 2010; Hendershott et al., 2017;
importantly, the non-CCP dummy coefficients are almost and Di Maggio et al., 2017). To conclude our analysis, we
unchanged compared to the ones in Table 6. This evidence pose the question whether relationships can explain the
is against the liquidity hypothesis: the premium does not OTC premia. To answer this question we modify Eq. (1) by
seem to be a compensation for any differences in liquidity replacing the buyer ID and seller ID fixed effects with their
across the different OTC segments. interaction. The new specification controls for any unob-
servable time-invariant effects at the pair of counterparties
level (i.e., bilateral relationships). We present this specifi-
cation in Columns 1–4 in Table 10. The results are qualita-
SEFs (for index CDS contracts) is limited. Note that the European equiv-
tively similar to the ones in Table 6, but bigger in mag-
alent of US-authorised SEF was introduced in January 2018 as part of
the Markets in Financial Instruments Directive (MiFID) II implementation, nitude in the case when the dealer receives fixed (sells
which is outside our sample period. protection).
35
The European DTCC data do not contain a flag for the venue of execu- We next allow for these relationships to vary over time.
tion. Therefore we had to ask LCH to provide us with this information for For example, the development of the bilateral dealer-client
LCH trades (they account for 97% of centrally cleared trades in our sam-
exposures might influence the decision to clear, the di-
ple). About one in five LCH trades were executed on SEF. Note that these
trades involved non-US persons, who were not subject to the Dodd–Frank rection of trade (paying vs. receiving fixed), or the pricing
Act trading mandate. of swaps. As a normal practice in trading derivatives,
//doi.org/10.1016/j.jfineco.2019.09.010
JID: FINEC
we expect that parties re-calculate these exposures on a forming derivatives markets is far to be completed, our
monthly basis. We capture this effect by replacing δ i and study should deliver important insights for policy mak-
ζ j in specification 1 with a month × buyer ID × seller ID ers and market participants. For policy makers, it pro-
fixed effect. As in many dealer-to-customer relationships, vides evidence that market participants in derivatives mar-
we assume that the average counterparty rebalances her kets have adapted their trading behaviours and pricing in
portfolio at the monthly frequency and as a result, bilat- line with some declared objectives of new regulation de-
eral exposures change monthly. We present variations of signed to improve the resilience of OTC markets. In fact,
this specification in Columns 5–8 of Table 10. In line with our study shows that non-CCP swaps are more expensive
the results in the previous section, our two main findings while the CCP segment hosts market participants of bet-
still hold: first, when dealers receive fixed they charge a ter credit quality and features more standard contracts.
premium, and in fact, its magnitude is almost twice as However, evidence of price heterogeneity and dealers’ bar-
much as the one in Table 6; second, there is no premium gaining power against their clients suggests that trans-
when dealers pay the fixed rate.36 parency and efficiency, which are also declared objectives,
To sum up, the results in this section corroborate the can be improved. For market participants, our paper should
existence of OTC premia and dealer-customer asymmet- help identify OTC premia across market segments and mar-
ric patterns, which are unaffected by market illiquidity, ket participants by quantifying the main determinants of
relationships, and time-varying bilateral exposures. derivative prices. This task is particularly relevant in the
post-crisis regime in which XVAs presumably play a major
5. Conclusion role.
Finally, our findings call for future research to exam-
This paper provides the first systematic study of IRS ine the impact of reforms that are outside our sample pe-
by analysing the transitional regime after the introduction riod, such as the European clearing mandate37 and the un-
of price-based measures such as CVA capital charges but cleared margin rules,38 or outside the financial instrument
before the entry into force of the quantity-based European and market participants analysed in this paper.
central clearing mandate in mid-2016. Using unique trade
repository data at transaction and ID levels, we uncover References
the main distinguishing characteristics of transactions
cleared via a central counterparty (CCP) with respect Abad, J., Aldasoro, I., Aymanns, C., D’Errico, M., Rousova, L.F., Hoff-
to those cleared bilaterally (non-CCP). A trade is more mann, P., Langfield, S., Neychev, M., Roukny, T., 2016. Shedding Light
on Dark Markets: First Insights from the New EU-wide OTC Deriva-
likely to be centrally cleared with larger (standardised) tives Dataset. Discussion Paper. European Systemic Risk Board.
notional and longer maturity, and involving dealers, banks, Acharya, V., Bisin, A., 2014. Counterparty risk externality: centralized ver-
and hedge funds rather than corporates and insurance sus over-the-counter markets. J. Econ. Theory 149, 153–182.
Adrian, T., Etula, E., Muir, T., 2014. Financial intermediaries and the cross–
companies. In terms of concentration of credit risk, the
section of asset returns. J. Financ. 69, 2557–2596.
CCP (non-CCP) segment is more populated by market par- Afonso, G., Lagos, R., 2015. Trade dynamics in the market for federal funds.
ticipants with higher (lower) creditworthiness. Moreover, Econometrica 83 (1), 263–313. https://onlinelibrary.wiley.com/doi/pdf/
credit risk concentration in the non-CCP segment increases 10.3982/ECTA10586.
Amihud, Y., Mendelson, H., 1980. Dealership market: market-making with
in distressed markets, perhaps in reaction to CCP margin inventory. J. Financ. Econ. 8 (3), 31–53.
procyclicality. Andersen, L.B.G., Duffie, D., Song, Y., 2019. Funding value adjustments.
Using panel regressions with fixed effects for time and J. Financ. 74 (1), 145–192.
Arora, N., Gandhi, P., Longstaff, F.A., 2012. Counterparty credit risk and the
counterparties’ IDs, we conduct an in-depth analysis of the credit default swap market. J. Financ. Econ. 103 (2), 280–293.
price heterogeneity of swap contracts. We find compelling Atkeson, A.G., Eisfeldt, A.L., Weill, P.-O., 2015. Entry and exit in OTC
evidence that prices differ across segments and market derivatives markets. Econometrica 83 (6), 2231–2292.
Babus, A., Hu, T.-W., 2017. Endogenous intermediation in over-the-counter
participants, and we call these price differentials OTC pre- markets. J. Financ. Econ. 125 (1), 200–215.
mia. OTC premia arise for decentralised clearing, riskier Babus, A., Kondor, P., 2018. Trading and information diffusion in OTC mar-
contracts and counterparties, and are economically impor- kets. Econometrica 86 (5), 1727–1769.
Bank for International Settlements, 2016a. OTC derivatives statistics at
tant. For instance, when a dealer sells interest rate protec- end-June 2016. Technical report.
tion to a client (i.e., the dealer receives fixed), the client Bank for International Settlements, 2016b. Triennial Survey of Foreign Ex-
pays the dealer around 8 basis points more for a non- change and OTC Derivatives Trading. Technical report.
Basel Committee on Banking Supervision, 2009. Strengthening the Re-
CCP contract than an equivalent CCP one. This premium
silience of the Banking Sector, Consultative Document. Technical
decreases when initial margin is posted. Another finding report.
is that OTC premia favour dealers, especially if their cus- Bellia, M., Panzica, R., Pelizzon, L., Peltonen, T., 2018. The demand for cen-
tomers are non-banks. Overall, our findings suggest that tral clearing: to clear or not to clear, that is the question. SAFE Work-
ing Paper No. 193.
OTC premia reflect inventory costs possibly increased by
recent regulations that are passed on to prices through the
so-called valuation adjustments (XVA). 37
Inducing more demand for central clearing and the concentration in
Considering that many markets continue to trade bilat- the provision of client clearing, this policy could overly increase central
erally and the implementation of the new regulation re- clearing costs.
38
These rules (phased in between January 2017 and September 2020)
are introducing stricter standards for the calculation of variation and ini-
36
We also controlled for client-month fixed effects, which control tial margin (e.g., daily exchange of high quality collateral by both parties),
for time-varying clients’ exposures across counterparties and mar- which will presumably imply a more important role for the FVA and MVA
kets/products. The results survive. and greater reduction in the CVA and KVA.
//doi.org/10.1016/j.jfineco.2019.09.010
JID: FINEC
Benos, E., Payne, R., Vasios, M., 2019. Centralized trading, transparency Gregory, J., 2015. The XVA Challenge: Counterparty Credit Risk, Funding,
and interest rate swap market liquidity: evidence from the implemen- Collateral and Capital. John Wiley & Sons, Hoboken, New Jersey.
tation of the Dodd-Frank Act.. J. Financ. Quant. Anal.. (forthcoming) Grossman, S.J., Miller, M.H., 1986. Economic costs and benefits of the pro-
Biais, B., Heider, F., Hoerova, M., 2012. Clearing, counterparty risk, and ag- posed one-minute time bracketing regulation. J. Futures Mark. 6 (1),
gregate risk. IMF Econ. Rev. 60 (2), 193–222. 141–166.
Biais, B., Heider, F., Hoerova, M., 2016. Risk-sharing or risk-taking? Coun- Hau, H., Hoffmann, P., Langfield, S., Timmer, Y., 2018. Discriminatory
terparty risk, incentives, and margins. J. Financ. 71 (4), 1669–1698. Pricing of Over-the-Counter Derivatives. CEPR, Discussion Paper No.
Bolton, P., Santos, T., Scheinkman, J.A., 2016. Cream-skimming in finan- DP12525.
cial markets. J. Financ. 71 (2), 105–137. doi:10.1111/j.1540-6261.2006. He, Z., Kelly, B., Manela, A., 2017. Intermediary asset pricing: new evidence
00831.x. from many asset classes. J. Financ. Econ. 126, 1–35.
Brunnermeier, M.K., Pedersen, L.H., 2008. Market liquidity and funding He, Z., Krishnamurthy, A., 2013. Intermediary asset pricing. Am. Econ. Rev.
liquidity. Rev. Financ. Stud. 22 (6), 2201–2238. 103, 732–770.
Cameron, A.C., Trivedi, P.K., 2010. Microeconometrics Using Stata, Revised Hendershott, T., Li, D., Livdan, D., Schürhoff, N., 2017. Relationship Trading
Ed. Stata Press, College Station, Texas. in OTC Markets. Swiss Finance Institute, Research Paper No. 17–30.
Cenedese, G., Della Corte, P., Wang, T., 2019. Currency mispricing and Ho, T., Stoll, H.R., 1981. Optimal dealer pricing under transactions and re-
dealer balance sheets. Bank of England Staff Working Paper No. 779. turn uncertainty. J. Financ. Econ. 9 (1), 47–73.
Cielinska, O., Joseph, A., Shreyas, U., Tanner, J., Vasios, M., 2017. Gauging Hull, J., Predescu, M., White, A., 2005. Bond prices, default probabilities
market dynamics using trade repository data: the case of the Swiss and risk premiums. J. Credit Risk 1, 53–60.
franc de-pegging. Financial Stability Paper No. 41. Bank of England. Jankowitsch, R., Nashikkar, A., Subrahmanyam, M., 2011. Price dispersion
Di Maggio, M., Kermani, A., Song, Z., 2017. The value of trading relations in OTC markets: a new measure of liquidity. J. Bank. Financ. 35 (2),
in turbulent times. J. Financ. Econ. 124 (2), 266–284. 343–357.
Du, W., Gadgil, S., Gordy, M. B., Vega, C., 2016. Counterparty risk and Lagos, R., Rocheteau, G., 2007. Search in asset markets: market structure,
counterparty choice in the credit default swap market. Finance and liquidity, and welfare. Am. Econ. Rev. 97 (2), 198–202.
Economics Discussion Series 2016-087. Lagos, R., Rocheteau, G., 2009. Liquidity in asset markets with search fric-
Duffie, D., 2012. Dark Markets: Asset Pricing and Information Transmis- tions. Econometrica 77 (2), 403–426.
sion in Over-the-Counter Markets. Princeton University Press, Prince- Li, D., Schürhoff, N., 2019. Dealer networks. J. Financ. 74 (1), 91–144.
ton, NJ. Loon, Y.C., Zhong, Z.K., 2014. The impact of central clearing on counter-
Duffie, D., Garleanu, N., Pedersen, L.H., 2005. Over-the-counter markets. party risk, liquidity, and trading: evidence from the credit default
Econometrica 73 (6), 1815–1847. swap market. J. Financ. Econ. 112 (1), 91–115.
Duffie, D., Garleanu, N., Pedersen, L.H., 2007. Valuation in over-the-counter Loon, Y.C., Zhong, Z.K., 2016. Does Dodd–Frank affect OTC transaction
markets. Rev. Financ. Stud. 20 (6), 1865–1900. costs and liquidity? Evidence from real-time CDS trade reports. J. Fi-
Duffie, D., Malamud, S., Manso, G., 2009. Information percolation with nanc. Econ. 119 (3), 645–672.
equilibrium search dynamics. Econometrica 77 (5), 1513–1574. Menkveld, A.J., 2017. Crowded positions: an overlooked systemic risk for
Duffie, D., Scheicher, M., Vuillemey, G., 2015. Central clearing and collat- central clearing parties. Rev. Asset Pric. Stud. 7 (2), 209–242. doi:10.
eral demand. J. Financ. Econ. 116 (2), 237–256. 1093/rapstu/rax016.
Duffie, D., Zhu, H., 2011. Does a central clearing counterparty reduce coun- Mildenstein, E., Schleef, H., 1983. The optimal pricing policy of a monop-
terparty risk? Rev. Asset Pric. Stud. 1 (1), 74–95. olistic marketmaker in the equity market. J. Financ. 38 (1), 218–231.
Financial Stability Board, 2010. Implementing OTC Derivatives Market Re- Murphy, D., Vasios, M., Vause, N., 2014. An Investigation into the Procycli-
forms. Technical report. cality of Risk-based Initial Margin Models. Financial Stability Paper
Financial Stability Board, 2017a. OTC Derivatives Market Reforms: Twelfth No. 29. Bank of England.
Progress Report on Implementation. Technical report. O’Hara, M., Oldfield, G.S., 1986. The microeconomics of market making.
Financial Stability Board, 2017b. Review of OTC Derivatives Market Re- J. Financ. Quant. Anal. 21 (4), 361–376.
forms: Effectiveness and Broader Effects of the Reforms. Technical re- Pelizzon, L., Subrahmanyam, M.G., Tomio, D., Uno, J., 2016. Sovereign
port. credit risk, liquidity, and European Central Bank intervention: deus ex
Financial Stability Board, 2018. Incentives to Centrally Clear Over-the- machina? J. Financ. Econ. 122 (1), 86–115.
Counter (OTC) Derivatives. Technical report. Riggs, L., Onur, E., Reiffen, D., Zhu, H., 2018. Swap Trading after Dodd–
Glode, V., Green, R.C., Lowery, R., 2012. Financial expertise as an arms Frank: Evidence from Index CDS. Commodity Futures Trading Com-
race. J. Financ. 67 (5), 105–137. mission and Massachusetts Institute of Technology, Unpublished
Golosov, M., Lorenzoni, G., Tsyvinski, A., 2014. Decentralized trading with working paper.
private information. Econometrica 82 (3), 1055–1091. Ruiz, I., 2015. XVA Desks – A New Era for Risk Management. Palgrave
Green, A., 2015. XVA: Credit, Funding and Capital Valuation Adjustments. Macmillan UK, London.
John Wiley & Sons, Hoboken, New Jersey. Shen, P., Starr, R.M., 2002. Market-makers’ supply and pricing of financial
Green, R.C., Hollifield, B., Schürhoff, N., 2007. Financial intermediation and market liquidity. Econ. Lett. 76 (1), 53–58.
the costs of trading in an opaque market. Rev. Asset Pric. Stud. 20 (2), Stoll, H.R., 1978. The supply of dealer services in securities markets. J. Fi-
275–314. nanc. 33 (4), 1133–1151.
Green, R.C., Li, D., Schürhoff, N., 2010. Price discovery in illiquid markets: Zhu, H., 2012. Finding a good price in opaque over-the-counter markets.
do financial asset prices rise faster than they fall? J. Financ. 65 (5), Rev. Financ. Stud. 25 (4), 1255–1285. doi:10.1093/rfs/hhr140.
1669–1702.
//doi.org/10.1016/j.jfineco.2019.09.010

Journal of Financial Economics: Gino Cenedese, Angelo Ranaldo, Michalis Vasios

Uploaded by

Copyright:

Available Formats

Journal of Financial Economics: Gino Cenedese, Angelo Ranaldo, Michalis Vasios

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Journal of Financial Economics: Gino Cenedese, Angelo Ranaldo, Michalis Vasios

Uploaded by

Copyright:

Available Formats

JID: FINEC

ARTICLE IN PRESS [m3Gdc;October 3, 2019;1:7]

Contents lists available at ScienceDirect

Journal of Financial Economics

Panel A: Breakdown by segment

N trades Notional ($m)

CCP 85,810 6,801,165

Panel B: Breakdown by segment and counterparty type

AM Bank Dealer HF Ins Other ﬁn Non-ﬁn

Non-CCP with margin

N Mean Std. dev.

Panel C: Non-CCP with margin

Panel D: Non-CCP no margin

Table 3 for non-centrally cleared trades and zero otherwise. On

(1) (2) (3) (4) (5) (6)

Log-notional −0.271∗ ∗ ∗ −0.255∗ ∗ ∗ −0.237∗ ∗ ∗

(1) (2) (3) (4) (5) (6)

Log-notional −0.294∗ ∗ ∗ −0.293∗ ∗ ∗ −0.295∗ ∗ ∗ −0.295∗ ∗ ∗ −0.281∗ ∗ ∗ −0.282∗ ∗ ∗

Dealer receiving ﬁxed Dealer paying ﬁxed

(1) (2) (3) (4) (5) (6) (7) (8)

Dealer receiving ﬁxed Dealer paying ﬁxed

(1) (2) (3) (4) (5) (6) (7) (8)

Rk,i, j,t = β1 Non-CCP + β2 Non-CCP with IM 28

Dealer-to-client Dealer-to-client (non-bank)

(1) (2) (3) (4) (5) (6) (7) (8)

Liquidity Exempted ﬁrms

Relationships Bilateral exposures (monthly)

(1) (2) (3) (4) (5) (6) (7) (8)

You might also like