Credit default swaps (CDS) are derivative contracts which were at the center stage of the global financial crisis: their role in amplifying and transferring distress is still debated. These contracts entail a transfer of risk (insurance) from a protection buyer (say firm A) to a protection seller (say firm B). This risk transfer implies that now B is liable to pay the insurance and may therefore be willing to offset this risk by buying a CDS from another firm C. In this case, the underlying risk is transferred from A to C via B. Interestingly, the CDS market consists of many such chains of transactions. The series of risk transfers among all market participants engenders a large network, which we analyse (theoretically and empirically) in a recent contribution. It then becomes very useful to map how risk flows in the market: where does risk originate and where does it ends up?

In particular, we provide a methodology to create such maps in the global CDS market. By following each risk transfers, we observe risk flowing like a running water stream through the set of participants. Though this flow may indeed be thought a river of “risk” flowing across different counterparties, there is a key difference. Water in a river cannot come back into itself. In contrast, financial risk can. As a sort of Escherian river, risk transfers create “cycles”. This indeed represents one of the deep differences between finance and the physical world.

Empirically, we use a granular, transaction level dataset to identify this long series of risk transfers. In particular, we find that risk flows into a small number of ultimate risk buyers, which constitute a bottleneck for the risk flow. We find that this series of risk translates into complex chains that involve the same set of intermediaries. We show that the largest amount of notional obligations lies between intermediaries and, in particular, onto the closed chains of intermediation described above. This has a key implication about the actual “size” of these markets and the overall use of these contracts in global financial markets.

In a recent paper[1] we provide theoretical and empirical insights about the network structure of the global credit default swap (CDS) market. The first contribution is the development of an analytical framework to analyse how risk flows through the set of counterparties in this market. We then apply the framework to analyse a granular, transaction-level dataset from the Depository Trust & Clearing Corporation (DTCC), covering all transaction in single-name CDS written on large sovereigns and corporate (both financial and non-financial) reference entities for four time snapshots between 2011 and 2014.  Before delving into the research contribution, we first provide a short primer on these financial instruments and then proceed to a brief summary of their implications for systemic risk.


What is a CDS?

CDS are amongst the most well-known and debated financial instruments. They are derivative instruments as their value derives from the state of an underlying entity. The literature on CDS is vast and we do not aim to cover all potential aspects in this post, hence we will limit ourselves to the key concepts for our analysis.

University level textbooks often provide an explanation of a CDS in terms of an insurance contract. Imagine you purchased a bond (or any other loan instrument) issued by a Company X or a Country Y. These issuers have made a promise to pay you back a certain amount of money at a certain point in time. However, these issuers may default, that is they may not be able to pay back their obligations (partially or fully). A CDS offers the possibility to buy protection against this default: as in a standard insurance contract, the protection buyer pays a fee to the protection seller who, in turn, promises to pay a certain “notional” amount in case the issuer (the underlying entity) defaults.

An important feature of these contracts is that they are traded ‘over-the-counter’. In other words, you cannot buy or sell a CDS on a centralised exchange (like you would do for stocks). A CDS is therefore bilateral in nature, which implies that in case the protection seller itself defaults, the protection buyer would not receive the payment. Despite the progresses in central clearing following the global financial crisis, the vast majority of CDS written on individual reference entities (single name) is traded bilaterally, unlike index CDS (which comprise groups of reference entities), which are increasingly more centrally cleared.

Another characteristic of a CDS is that, unlike your car or house insurance, you do not need to have an insurable interest. This means that you can purchase a CDS contract even when you don’t possess the underlying bond in your portfolio. In practice, this means that you would profit in case the underlying entity defaults. This type of CDS contract is often referred to as a naked CDS and virtually allows you to replicate a short-selling strategy for the underlying bond. This practice has attracted a lot of criticism and sparked a debate on whether it should be banned or not, as it may create wrong incentives. Supporters view this in a different way and argue that speculation of this type increases liquidity in the market and therefore the possibility of hedgers to find a counterparty, thereby lowering the cost of the insurance. The European Union has opted to outright limit short selling of sovereign risk via CDS[2]. In practice, EU buyers of a CDS on a sovereign entity must now have a correspondent underlying exposures to the default or to the decline in value of the underlying issuers.

As we have explained above, a naked CDS is equivalent to shorting the underlying position. The derivative therefore allows to obtain a so-called synthetic short-position on the underlying instrument. It is referred to as ‘synthetic’ since no actual exchange of the underlying debt instrument needs to take place. In addition, there is another strategy that allows to obtain synthetic exposures which is often not well known. Such a position can be obtained by selling protection on an underlying entity. In this way, the CDS sold correlates with the value of the underlying bond. One can therefore mimic (if not replicate) the same position obtained by having the underlying bond in the portfolio without any need to physically purchase the bond itself. As there is a limited amount of underlying bonds in the market, synthetic positions achieved by buying and selling CDSs are of particular interest since they can be virtually larger than the amount of underlying bonds.

Another important point regards the size of these markets, which is typically measured in terms of aggregate notional outstanding. For instance, the BIS reports[3] that the current outstanding notional amount in credit derivatives is about 15 trillion dollars, of which 8 trillion on single names entities. At the onset of the crisis, the total outstanding notional amount was of 60 trillion i.e., in the same order of magnitude of the world GDP. This measure is ‘correct’ in the sense that, for the CDS market, it captures the amount of payments due in case the reference entity defaulted. For other derivatives (such as interest rate swaps), this value can be misleading as a measure of the underlying risk, as the real payments can be significantly lower.


CDS markets and the Global Financial Crisis

CDS have attracted a lot of attention in the aftermath of the global financial crisis. There are several reasons for this. The first is the fact that Lehman Brothers and AIG, two institutions which played a key role in the crisis, were heavily trading in these markets. Haldane (2009)[4] argues that the CDS market froze since Lehman ‘was believed’ to be the counterparty of about five trillion of CDS contracts. This uncertainty ‘about its causes and contagious consequences brought many financial markets and institutions to a standstill. […] Banks hoarded liquidity for fear of lending to infected banks, causing gridlock in term money markets, spreads on lower-rated companies’ bonds spiked and there was an effective boycott of the remaining large US investment banks.’ Uncertainty is the key concept here, and a recurring theme within the research work of the consortium. Not knowing Lehman’s actual positions led to widespread uncertainty and market participants reacted by losing confidence, which is arguably the key cause of almost any financial crisis. A critical review of the role of CDS in the crisis can be found in Stulz (2009)[5].


Risk flows in the global CDS market

A CDS can be viewed as a transfer of risk between a counterparty i (the protection buyer, or risk seller) and a counterparty j (the protection seller, or risk buyer) written upon a reference entity k. The series of risk transfers between the various participants creates a network of transfers. More precisely, the network features a multi-layered structure, with each layer corresponding to a specific underlying entity. Empirically, we find that these risk transfers progress along chains of intermediations, as reported in the figure below.


These chains comprise a finite number of dealer nodes (in blue) which are typically large intermediary banks. We call the nodes at the two ends of the chain end-users. In particular, we can distinguish between Ultimate Risk Sellers (URS) and Ultimate Risk Buyers (URB). These risk transfers are intermediated by the dealers. The market is composed of many end users, so we observe many of these chains. We prove theoretically that, when the number of linkages between a finite number of dealers is sufficiently large, these chains become interdependent, and all the dealers are connected (directly or indirectly) to one another. In network-theory language, they form a strongly connected set of nodes. These financial networks can therefore be modelled as a bow-tie network as the following stylised diagram shows:


An important mathematical characterisation of a strongly connected component of a network is the presence of cycles, i.e. closed chains of exposures. An interesting consequence of having closed chains of exposures is that one can have arbitrarily large weights on the edges of the network (i.e. notional financial obligation) without changing the net individual amounts (that is to say the difference between amount sold and bought), as shown in the example below, where the total obligations in the system can be increased by three times the amount  without changing the dealers’ net positions, as the same amount  enters and exits the node (that is, it is both bought and sold).


Indeed, our empirical analysis shows that that the vast majority of notional amount (around 80%) lies on these closed chains. In a recent work, D’Errico and Roukny (2017)[6], provide ways to reduce the ‘redundant’ amount of notional in the market which lies on these chains.

A key point here is that, given that such a large fraction of the market is not needed to satisfy a ‘net’ position, it is nearly meaningless to say that – if X dollars of CDS are currently written on a reference entity – then market participants demand that amount of protection on the entity itself. There is a key link between the price of the protection offered by a CDS and the price of the underlying debt (the interest the borrower pays): the impact of these redundant notional insurance on the price is still unclear and will be the subject of future research.


Mapping the global CDS market

We analyse a unique, transaction level, global dataset on credit default swaps. The dataset includes all global counterparties and their transactions (buy/sell), the notional amount and the underlying reference entity over four time snapshots (March 11, April 12, December 12 and October 14). The counterparties are consistently anonymised across the space of reference entities and time, though we have a label referring to their institutional sector, which is reflected in our network visualisations. The reference entities in our sample are the largest sovereign and financial institutions, covering approximately one-third of the total single-name global market.

The time evolution of the network (aggregated across reference entities) is visualised below. We observe that hedge funds (in red) are largely net buyers of protection. This may indeed reflect the fact that certain funds are buying ‘naked’ CDS, though we are not in the position to fully validate this hypothesis, due to the fact that we do not have data on their underlying exposures.


We can also zoom-in on the network built for individual reference entities. The picture above visualises the flow of risk for a major sovereign across two time snapshots: March 11 and October 14. It is immediate to observe a key shift between the end-users. Hedge funds are the largest end-protection buyers in 2011 but virtually disappear in 2014, when one large asset manager emerges as the largest protection seller. A possible explanation of both instances is in terms of synthetic use of derivatives. Hedge funds may have been short-selling the underlying sovereign risk immediately before the EU sovereign crisis in late 2011, whereas the large asset manager in 2014 could have gained synthetic exposure on the same sovereign once the crisis has passed and this strategy could be safer. However, since we do not have data on the underlying exposures of each counterparty, this only represents a hypothesis which we are not in the position to fully validate.


Implications and future work

Our study relates to the broader topic of interconnectedness in global financial systems. In particular, it offers a series of tools to enhance the understanding of over-the-counter derivatives markets. An important implication of our analysis is to show that the largest fraction of notional obligation (80%) in the global credit default swap market lies on closed intermediation chains (“cycles’’, in the language of network theory). These chains allow for an arbitrarily large amount of obligation for any given net position. The remaining 20% of notional is associated to end-users. Our analysis represents an important first step towards the understanding how the microstructure of these markets and their implications for systemic risk. Additionally, it provides a new interpretation for the “size’’ of these markets. In fact, the largest amount of the notional obligations lie on closed chains of exposures between intermediaries and can be arbitrarily large without changing the net position of each participant: as a consequence, the largest amount of obligations is due between intermediaries rather than end users. Concerning the use of these instruments, further work in linking CDS activity with underlying portfolios will be needed in order to understand the impact in terms of hedging or speculation for these instruments.


The key takeaways

  1. CDS markets show a high degree of intermediation.
  2. The vast majority of total notional (around 80%) is written between intermediaries, and lies on closed cycles. The total notional on these closed cycles can be be arbitrarily large, for a given net position for each market participant.
  3. This heavily interconnected structure may be more conducive to systemic risk.
  4. In general, one should rethink the interpretation of “size’’ of these markets. In particular, we find that very large notional amounts may reflect large levels of intermediation rather than high levels of demand for insurance or speculation of the underlying.
  5. Future work will deal with price effects of the network structure of the CDS market and the underlying security.


[1] How does risk flow in the credit default swap market? (2107), by Marco D’Errico (UZH), Stefano Battiston (UZH), Tuomas Peltonen (ESRB) and Martin Scheicher (ECB). Published, Journal of Financial Stability and also in the ESRB and ECB working paper series.

[2] See the EU regulation N 236/2012.

[3] The Bank for International Settlements publishes twice a year aggregate statistics on OTC derivatives by type, available at

[4] See

[5] Stulz, René M. “Credit default swaps and the credit crisis.” The Journal of Economic Perspectives 24.1 (2010): 73-92.

[6] D’Errico, Marco, and Tarik Roukny. “Compressing over-the-counter markets.” (2017).