Discussion of Trading and Information Diffusion in OTC Markets by Ana Babus and Peter Kondor 10 th Annual Central Bank Workshop on Microstructure of Financial Market Marco Di Maggio (Columbia Business School)
OTC Markets Many of the financial instruments at the core of the recent financial crisis MBS, CDO, CDS are traded in over-thecounter markets (OTC), outside of organized exchanges. Liquidity is provided by broker-dealers such as large investment banks, who buy assets on their own account or match buyers and sellers. While dealers liquidity provision seems inconspicuous in normal times, it has proved inadequate during the crisis. Policy discussion: central clearing party. Usual approach: search frictions. This paper: private information + network. Prof. Marco Di Maggio 2
In a nutshell Model of trading and information diffusion Highlights: Dealers have private information They engage in bilateral transactions with other dealers As well as with costumers, with downward sloping demand curve. Each bilateral price reflects private information of all dealers yet typically diffusion is not informationally efficient. Several interesting, novel and testable empirical implications. Prof. Marco Di Maggio 3
Main Insights Dealer A Dealer B Prof. Marco Di Maggio 4
Main Insights Dealer A Dealer B Q i s i ; p = t(e θ i s i, p p) Dealer A trades t units with counterparty B for every unit of perceived gain Prof. Marco Di Maggio 5
Main Insights Dealer A Dealer B Q i s i ; p = t(e θ i s i, p p) E θ i s i ; p = yy i + zz Dealers beliefs are a combination of their own signal and the price Prof. Marco Di Maggio 6
Main Insights Dealer A Dealer B Q i s i ; p = t(e θ i s i, p p) E θ i s i ; p = yy i + zz p = t[e θ i s i, p + E(θ j ss, p)] 2t β The price is a linear combination of their posterior beliefs Prof. Marco Di Maggio 7
Main Insights Dealer A Dealer B Who buys and who sells? Prof. Marco Di Maggio 8
Main Insights Dealer A Dealer B Who buys and who sells? 1. Perceived gain: it decreases in the correlation between dealers signals and in the precision of dealers information. Prof. Marco Di Maggio 9
Main Insights buy Dealer A Dealer B Who buys and who sells? 1. Perceived gain. 2. If A is optimistic about value of the asset he will try to buy, and will increase his demand if the other dealer desire to sell Prof. Marco Di Maggio 10
Main Insights buy less Dealer A Dealer B Who buys and who sells? 1. Perceived gain. 2. However, demand less if he is worried that the large supply signals low value asset. Prof. Marco Di Maggio 11
Main Insights Dealer A Dealer B 1. Perceived gain. 2. Learning effect. Who buys and who sells? Prof. Marco Di Maggio 12
Main Insights What changes in a network? Dealer A Dealer C Dealer B 1. Perceived gain. 2. Learning effect. 3. Asymmetry: even if all dealers have same quality of information trading intensities are different. Prof. Marco Di Maggio 13
Main Insights What changes in a network? Dealer A Dealer C Dealer B 1. Perceived gain. 2. Learning effect. 3. Asymmetry: C can learn from two prices, then he is less subject to adv. selection, which makes him react less to other dealers intensities. 14
Main Insights What changes in a network? P A P B Dealer A Dealer C Dealer B 1. Perceived gain. 2. Learning effect. 3. Asymmetry. => Price dispersion 15
Main Insights What changes in a network? Dealer A Dealer C Dealer B 1. Perceived gain. 2. Learning effect. 3. Asymmetry. Price dispersion Intermediation 16
Main Contribution Theory There are two papers that study information diffusion in OTC market, Duffie et al. (2009) and Golosov et al. (2009) However, this is the first one in which the agents are strategic about it! How intermediation affects the extent of the adverse selection in the market makes it close to Glode and Opp (2014), but they consider only chains, while here a completely general network. Prof. Marco Di Maggio 17
Main Contribution (2) Finance/Applied More connected dealers trade more at lower cost, higher profit per trade, set smaller spreads, less dispersed but more volatile prices. Larger transactions associated with smaller spreads, less dispersion and higher profits. Price are not informationally efficient: dealers do not internalize how the informativeness of their guesses affects others decisions, i.e. put too much weight on their own signals. Prof. Marco Di Maggio 18
Next Steps Building on this paper, one could explore: Trading dynamics: out of the steady state equilibrium (as in Lagos, Rocheteau and Weill (2011)) Network formation: or endogenous trading paths. This is one of those papers that might (will) open a new strand of the literature. Prof. Marco Di Maggio 19
Comment 1: Price-Setting Mechanism Equilibrium determines for each bilateral transaction both prices and quantities. Dealers submit/decide demand functions. Something similar happens in repo transactions for MMFs. An alternative: to obtain price pressure, one could use the Garleanu, Pedersen, and Poteshman (2009) paper on demand-based option pricing. It might be simpler and deliver similar insights. Prof. Marco Di Maggio 20
Comment 2: Potential Extension What if the dealers can choose between this venue and an exchange market? On the one hand, the exchange might limit adverse selection for the worse-informed dealer. On the other, better-informed dealers might refrain from revealing their information in the exchange to exploit it in the OTC market. It might provides insights about price discovery for similar assets traded on different markets. Prof. Marco Di Maggio 21
(Preliminary) Empirical Evidence With Amir Kermani (UC Berkeley) and Zhaogang Song (Fed Board) we collected confidential TRACE data on several OTC markets, e.g. corporate bonds and RMBS. Unique features: Trade identifier: Purchase from customer, Sale to customer, Inter-dealer Dealer identities: Dealer in customer trades (CD, DC) and both sides in DD Daily frequency with several characteristics about the underlying asset (plus match to Bloomberg info). Prof. Marco Di Maggio 22
Network Structure Prof. Marco Di Maggio 23
Network Structure Just a snapshot? Not really, it is the average over the last 4 years Prof. Marco Di Maggio 24
Supporting Evidence (1) Profits per Transaction All Trades Counterparty is Client 0.0120*** (0.00217) Log(Transaction Volume) -0.00200*** (0.000306) Proportion of Transaction with Client Trade Length=2 Trade Length=3 Trade Length=4 Trade Length=5 Day Fixed Effect Seller Fixed Effect Y Y Observations 85,839 R-squared 0.127 Prof. Marco Di Maggio 25
Supporting Evidence 1 2 Profits per Transaction All Trades Trades with Clients Counterparty is Client 0.0120*** (0.00217) Log(Transaction Volume) -0.00200*** -0.00155*** (0.000306) (0.000284) Proportion of Transaction with Client -0.0103** Trade Length=2 Trade Length=3 Trade Length=4 Trade Length=5 (0.00508) Day Fixed Effect Y Y Seller Fixed Effect Y Y Observations 85,839 79,531 R-squared 0.127 0.137 Prof. Marco Di Maggio 26
Supporting Evidence (1) (2) (3) Profits per Transaction All Trades Trades with Clients Counterparty is Client 0.0120*** (0.00217) Log(Transaction Volume) -0.00200*** -0.00155*** -0.00148*** (0.000306) (0.000284) (0.000240) Proportion of Transaction with Client -0.0103** -0.0102** (0.00508) (0.00458) Trade Length=2-0.00725*** (0.00166) Trade Length=3-0.00891*** (0.00209) Trade Length=4-0.0106*** (0.00209) Trade Length=5-0.0116*** (0.00267) Day Fixed Effect Y Y Y Seller Fixed Effect Y Y Y Observations 85,839 79,531 79,531 R-squared 0.127 0.137 0.153 Prof. Marco Di Maggio 27
Main Takeaways When buyer is a client, the profit per dollar is 1.2% higher this is almost equal to the whole spread, i.e. most of the dealers profits results from dealing with clients. Increasing the size of the trade by 1σ reduces the spread by 0.5%. Trading through intermediaries is significantly less profitable: if the length of the chain is 2, the spread is closer to 0.7% (as opposed to 1.2% on average)...to the extent there is very little profit remained for chain length of 5 for the last dealer in the chain. Prof. Marco Di Maggio 28
Conclusion Extremely important question. Well written paper. Novel, clever and insightful model. Big step forward for the OTC literature and for the network literature as well! Prof. Marco Di Maggio 29