The Global Crisis of 2008 highlighted the key role of the intertwined nature of financial markets in shaping the transmission of risk and the build-up of fragility throughout the system. In the words of Donald Kohn, former vice chairman of the Federal Reserve Board, "[s]upervisors need to enhance their understanding of the direct and indirect relationships among markets and market participants, and the associated impact on the system. Supervisors must also be even more keenly aware of the manner in which those relationships [...] can change over time and how those relationships behave in times of stress."1 Although there have been recent empirical studies on over-the-counter (OTC) markets (e.g. Li and Schürhoff 2014, Hollifield et al. 2012, Choi and Shachar 2013, Afonso et al. 2013 and Hendershott et al. 2016), the exact role played by financial system interconnectedness and the ways in which large financial institutions may affect OTC market liquidity remain at best imperfectly understood.
The trading network in the corporate bond market
Our recent work investigates dealers' trading behaviour and pricing strategy in the corporate bond market to shed new light on the role of the network of existing relationships among dealers in shaping the transmission of risk and influencing market liquidity (Di Maggio et al. 2016).
The corporate bond market is one of the world's largest and most important sources of capital for firms, with outstanding debt now of about $8 trillion. Daily trading volume in the US averages $20 billion, virtually all between broker-dealers and large institutions in a decentralised OTC market. This market is therefore ideal for studying how the network of dealer relationships shapes trading behaviour and liquidity provision, and investigating how dealers responded during the crisis.
We start by showing that the inter-dealer corporate bond market has a definite, persistent core-periphery network structure as shown by Figure 1.
Figure 1. The inter-dealer corporate bond market network
Note: This figure plots the core-periphery network structure where each link is a transaction, and at the centre there are the transactions with clients. Darker lines indicate a higher number of transactions between the two nodes.
In other words, there are only a few highly interconnected dealers – the core dealers – which intermediate most of the transactions with other dealers and with clients (retail investors, insurance companies, mutual and hedge funds, etc.) and many sparsely connected ones transacting less frequently, i.e. the peripheral dealers.
Given this structure, we analyse how dealers' markups and trading behaviour differ according to their counterparties' position in the network, and how their previous relationships with counterparties affect trading outcomes.2
The first result is that when dealers trade with clients rather than other dealers, they profit significantly more, as shown by the average spreads by pair type plotted in Figure 2 for our sample period.
Figure 2. Spreads for transactions among different types of counterparties
Note: This figure plots the profit margins over time, for transactions among different types of counterparties, where the first term identifies the seller and the second identifies the buyer.
On average, similar bonds in the same industry traded by the same dealer go at a significantly higher price to non-dealer clients, an extra markup of about 50 basis points. We also show that more central dealers pay lower spreads while charging significantly higher spreads to their counterparties.
Importantly, we also find significantly lower spreads between dealers with stronger prior relationships, as proxied by the fraction of bonds exchanged between two counterparties in the previous quarter. Here the results do not depend on differences in the type, volume or quality of the transactions of more or less connected dealers, because these characteristics are controlled for. The magnitude of the effects, moreover, is economically significant. The difference in the terms of trade between the bottom and the top decile of relationship strength comes to about 20 basis points. And the results hold even in the most conservative specification, where we control for seller- and buyer-month fixed effects, to offset time-varying shocks at dealer level that could affect trading behaviour.
This result also relates to Di Maggio et al. (2016b), who show that the relationships between brokers and institutional investors shape information transmission in the stock market, with the broker’s best clients receiving information about incoming orders ahead of other investors.
Overall, our findings constitute evidence that existing trading relationships, which most of the theoretical literature following Duffie et al. (2005) abstracts from, are crucial in determining trading behaviour, at least on a par with network centrality.3
Dealer network and trading relationships in times of market turmoil
Now we can answer the main questions we want to pose. Does the importance of these prior relationships vary over time? And do dealers tend to provide liquidity to their counterparties in times of market turmoil? We find that during periods of stress dealers provide less liquidity to clients and peripheral dealers (than to other core dealers), charging them significantly higher markups. In other words, at times of market turmoil dealers tend to rely even more heavily on their central position in the network, as more connected institutions can impose higher spreads and purchase at significantly lower prices.
This implies that in turmoil dealers rely more heavily on their closest counterparties, as suggested by Glode and Opp (2016). Our results carry important implications for the theoretical models of trading in OTC markets. Essentially, the common random-matching framework, which ignores bilateral trading relationships, de facto, by assuming that traders interact only in anonymous spot transactions, misses an important feature of off-exchange markets.
Also, we examine how dealers' trading behaviour reacted to the collapse of a flagship dealer that defaulted in September 2008. We code this dealer as Dealer D.4 First of all, after the failure of Dealer D the intermediation chains between buyers and sellers lengthened significantly as shown by Figure 3 which plots the average trading chain length over time normalised to the first week in 2005.
Figure 3. Average trading chain length
Note: This figure plots the average trading chain length over time normalised to the first week in 2005. The vertical line indicates the default of Dealer D.
Since longer chains are associated with higher spreads, they also have adverse effects on the clients who are seeking liquidity.
Second, we test whether dealers tend to lean against the wind, i.e. accumulating bond inventories during periods of turmoil as predicted by Weill (2007). We compute dealers' inventories in the weeks before and after Dealer D's collapse, excluding new issuance and maturing bonds, and find that they shrank significantly more for the bonds that clients were selling more vigorously. In fact, as shown by Figure 4 dealers decreased their holdings of the bonds that other market participants were selling most intensely by at least 20%.
Figure 4. Core dealers’ inventory by clients selling pressure
This figure plots the dealers’ inventory for bonds that experienced different selling pressure from the clients, which is defined as the amount sold by clients to dealers normalised by the amount outstanding.
This is strongly suggestive that one of the main factors in the increase in intermediation costs and market illiquidity was dealers' inability (or unwillingness) to expand inventories. As further evidence of this channel, we also show that inventories shrank most for the bonds in whose regard the intermediation chain lengthened the most. These results can inform the debate on dealers' role during the crisis and how significantly they aggravated the market disruption.
Overall, these results shed new light on the way in which these prior relationships may sometimes act as a buffer in periods of distress, but they also show that they accentuate systemic fragility, as connections with vulnerable dealers might affect trading outcomes even for sound dealers.
Afonso, G., A. Kovner, and A. Schoar (2013), “Trading partners in the interbank lending market”, FRBNY Staff Reports.
Battalio, R., A. Ellul, and R. Jennings (2007), “Reputation effects in trading on the New York Stock Exchange”, The Journal of Finance 62 (3), 1243-1271.
Benveniste, L. M., A. J. Marcus, and W. J. Wilhelm (1992), “What is special about the specialist?”, Journal of Financial Economics 32 (1), 61-86.
Choi, J. and O. Shachar (2013), “Did liquidity providers become liquidity seekers?”, Technical report, Staff Report, Federal Reserve Bank of New York.
Di Maggio, M., A. Kermani and Z. Song (2016), “The Value of Trading Relationships in Turbulent Times”, Journal of Financial Economics.
Di Maggio, M, F A. Franzoni, A. Kermani and C. Sommavilla (2016b), “The Relevance of Broker Networks for Information Diffusion in the Stock Market”.
Duffie, D., N. Garleanu, and L. Pedersen (2005), “Over-the-Counter Markets”, Econometrica 73 (6), 1815-1847.
Glode, V. and C. C. Opp (2016), “Adverse Selection and Intermediation Chains”, American Economic Review, forthcoming.
Hendershott, T., D. Li, D. Livdan, and N. Schürhoff (2016), “Relationship Trading in OTC Markets”.
Hollifield, B., A. Neklyudov, and C. S. Spatt (2012), “Bid-ask spreads and the pricing of securitizations: 144a vs. registered securitizations”.
Li, D. and N. Schürhoff (2014), “Dealer networks”.
Pagano, M. and A. Röell (1992), “Auction and dealership markets: what is the difference?”, European Economic Review 36 (2), 613-623.
Weill, P. (2007), “Leaning against the wind”, The Review of Economic Studies 74 (4), 1329-1354.
 Senate testimony, June 5, 2008.
 Our main dependent variable is the difference between the price at which a dealer sells a bond and his previous buying price. We call this the spread, profit margin or markup. We provide results for two different measures. The more conservative approach considers only trades in which the dealer buys a bond and then sells it within an hour. However, we also provide consistent evidence in which the benchmark buy price is the average at which other dealers buy the same bond during the same week.
 These results also relate our work to the papers showing that cooperation and reputation affect liquidity costs in exchange markets such as the papers by Battalio et al. (2007), who document an increase in liquidity costs in the trading days surrounding a stock's relocation to the floor of the exchange, as well as those by Pagano and Röell (1992) and Benveniste et al. (1992) who demonstrate that reputation attenuate the repercussions of information asymmetries in trading and liquidity provision.
 Under an agreement with the FINRA, we are not allowed to disclose dealer identities.