Information Revelation in Decentralized Markets

• Author: ALBERT J. MENKVELD,BJÖRN HAGSTRÖMER
• Date: 01 December 2019
• Published date: 01 December 2019
• DOI: http://doi.org/10.1111/jofi.12838

THE JOURNAL OF FINANCE •VOL. LXXIV, NO. 6 •DECEMBER 2019

Information Revelation in Decentralized Markets

BJ ¨

ORN HAGSTR ¨

OMER and ALBERT J. MENKVELD∗

ABSTRACT

How does information get revealed in decentralized markets? We test several hy-

potheses inspired by recent dealer-network theory. To do so, we construct an em-

pirical map of information revelation where two dealers are connected based on the

synchronicity of their quote changes. The tests, based on the euro to Swiss franc

spot rate (EUR/CHF) quote data including the 2015 crash, largely support theory:

strongly connected (i.e., central) dealers are more informed. Connections are weaker

when there is less to be learned. The crash serves to identify how a network forms

when dealers are transitioned from no-learning to learning, that is, from a fixed to a

floating rate.

TECHNOLOGY FACILITATES THE ENTRY OF electronic markets that are cheap to

set up and easy to tailor to various clienteles. New markets successfully

compete with incumbent ones. Many dealers, for example, have started their

own markets. As a result, securities are increasingly traded in decentralized

markets. This trend holds for all asset classes (Johnson (2010, Chapters 2,3)).

New market entry delivers well-known benefits of competition (e.g., lower

fees and better products), yet might be socially costly. The reason is that

end-users become dispersed across markets, which potentially inhibits efficient

re-allocations, as decentralized markets lead to costly cross-market search

for efficient trades. A relatively recent and growing literature focuses on such

friction (e.g., Hollifield, Neklyudov, and Spatt (2017), Maggio, Kermani, and

Song (2017), Li and Sch ¨

urhoff (2019)).

∗Bj¨

orn Hagstr¨

omer is with Stockholm Business School, Stockholm University. Albert J.

Menkveld (albertjmenkveld@gmail.com) is with Vrije Universiteit Amsterdam (VU) and Tinber-

gen Institute. We thank Stefan Nagel (the Editor), an anonymous associate editor, and three

anonymous referees for many helpful comments. We are also grateful for comments by Ben Craig;

Serge Darolles; Lorenzo Frattarolo; Joel Hasbrouck; Erik Hjalmarsson; Andrei Kirilenko; Youwei

Li; Dmitry Livdan; Artem Neklyudov; Chengcheng Qu; Dagfinn Rime; Ekaterina Serikova; Erkki

Silde; Marco van der Leij; Vincent van Kervel; Shihao Yu;and Shengxing Zhang; as well as semi-

nar/conference participants at Aarhus University,AEA, Banque de France, BIS, Copenhagen Busi-

ness School, EFA,Financial Conduct Authority, Higher School of Economics Moscow, NHH Bergen,

SoFiE Annual Meeting, Stevens Institute of Technology, Stockholm Business School, Universit´

e

catholique de Louvain, University of Gothenburg, University of Toronto, University of Toulouse,

University of Stavanger, University of Zurich, and WFA. An earlier version of this manuscript

was circulated under the title A Network Map of Information Percolation. Hagstr¨

omer is a visit-

ing research fellow at Swedish House of Finance. He gratefully acknowledges the Jan Wallander

Foundation and TomHedelius Foundation for research funding. Menkveld gratefully acknowledges

NWO for a Vidi grant. The authors have no material financial or nonfinancial interests related to

this research, as identified in the Journal of Finance disclosure policy.

DOI: 10.1111/jofi.12838

C2019 the American Finance Association

2751

2752 The Journal of Finance R

Relatively unexplored, however, is the question of how (fundamental)

information gets revealed in decentralized markets.1With dispersed trading,

information is produced in local markets and somehow is aggregated across

them. The literature on how exactly this happens is young and, to the best of

our knowledge, only theoretical.

In this paper, we contribute to this literature by testing several hypotheses

on information revelation. Motivated by theory, we address the following

questions. Do some markets (dealers) specialize endogenously in terms of

connecting to many others (Farboodi, Jarosch, and Shimer (2017))? Do these

markets benefit by being more informed (Babus and Kondor (2018))? And, in the

process of becoming informed, do they charge a higher bid-ask spread to recoup

the cost of adverse-selection (similar to the classic market maker of Glosten

and Milgrom (1985))? Finally, are connections stronger when there is more to

be learned due to more intensive search (Duffie, Malamud, and Manso (2009))?

Approach: To test our hypotheses, we develop an approach to mapping

information revelation empirically. An important necessary feature of this

approach is that it recognizes that price responses to shocks could diverge

across markets in the short term, but not in the long term; arbitrage activity

disciplines price responses to converge eventually. The challenge is to ap-

propriately characterize price dynamics between the shock and the eventual

convergence. Following the de facto standard set by Hasbrouck (1995), we

characterize such price dynamics using a vector error-correction model.

The key innovation of our approach is to characterize information revelation

and how it interacts with market connections based on what can be thought

of as a multivariate impulse response function (MIRF). In particular, we study

how prices in all markets respond to price shocks that transpire in all markets

simultaneously (i.e., multivariate reponses to multivariate shocks). We believe

that analyzing how these responses change with the time elapsed since a

shock captures the process through which information is revealed.2

We consider it economically insightful to create a network map based on

bilateral price synchronicity. Consider, for example, two dealers who meet and

agree to share their information on a recent shock (as in, for example, Duffie,

Malamud, and Manso (2009)). The result is that their quote responses to this

shock become more (partially) correlated, which we visualize as a stronger

connection (edge) between the two dealers. At the same time, both dealers’

quotes become more efficient, which is displayed as them moving closer to the

center of the map. In this way, a sequence of dealer meetings yields a series of

1This is in contrast to a rich literature that studies the extent to which agents collect private

information and how it gets revealed through trading (e.g., Grossman and Stiglitz (1980), Hellwig

(1980), Verrecchia (1982)).

2Note that this should not be confused with the generalized impulse response function (GIRF)

(Pesaran and Shin (1998)). Both MIRF and GIRF account for historical correlations observed

among shocks. However,the GIRF focuses on multivariate responses to univariate shocks whereby

all other variables are set to their conditional expectation given the (univariate) shock to the

variable of interest, whereas the MIRF considers shocks as multivariate in nature, see Section II

for more details.

Information Revelation in Decentralized Markets 2753

information-revelation maps (a movie) that characterizes the process through

which information gets revealed in the dealer community.

Diagnostic analysis of a (network) graph based on correlations has proven

useful in a variety of contexts, such as DNA arrays or Internet traffic (see de la

Fuente et al. (2004) and Pastor-Satorras, V´

azquez, and Vespignani (2001),

respectively). Perhaps closest to our work is the use of network graphs to study

information processing in neuroscience. In particular, Achard et al. (2006) con-

struct a network of various parts of the brain based on correlations in magnetic

resonance imaging (MRI) scans to study how a brain processes information.

A parallel line of research in neuroscience focuses on uncovering the brain’s

hardwiring. The nervous system of a roundworm was the first to be completely

mapped out (Achacoso and Yamamoto (1992)). In a sense, our paper resembles

the MRI study through its focus on co-activity, whereas earlier dealer-network

studies are like the roundworm study focused on hardwiring, as they analyze

physical connections between dealers based on whether they trade with one

another. We believe the two lines of research are complementary.

Results: We analyze information revelation for a sample of nine quote streams

in the euro to Swiss franc spot rate (EUR/CHF) in 2015. The data include

quote streams of eight major foreign exchange (FX) dealers as well as that of

an interdealer limit-order book, namely, Electronic Broking Services (EBS).

Because the dealers can effectively be considered markets, we refer to them

interchangeably as dealers or markets throughout (see Section IV.A for a more

detailed discussion of the market structure). The sample also includes news

data to identify intervals “treated” with public information.

Our key empirical findings are as follows. First, the information-revelation

maps are nontrivial and stable over time. The maps reveal that dealers differ

considerably in their number of dealer connections. Second, connections are

assortative and hence the network exhibits a core-periphery structure. Third,

the strongly connected dealers, henceforth referred to as central dealers,

are more informed in two respects: their price-quote innovations are more

revealing of an informational shock contemporaneously, and they respond

to the full shock more quickly.3Fourth, if part of the information becomes

public through news, then dealer connections become weaker and information

is revealed more slowly. This result confirms the theoretical prediction that

making some of the information public reduces private search in the network

(e.g., Duffie, Malamud, and Manso (2009)).4

The above findings are based on EUR/CHF trading in the two weeks after

the Swiss franc peg to the euro was removed on January 15, 2015. However,

the sample also includes the day of this removal, when at 10:30 AM local

time the Swiss central bank made the surprise announcement. The euro

3The contemporaneous effect is measured by the lower bound of a market’s information share

(Hasbrouck (1995)).

4The prediction that news weakens private search transcends the dealer-network studies em-

phasized here. For example, Diamond (1985) shows that voluntary disclosure of information by

firms reduces investors’ incentives to collect private information.