High‐Frequency Trading and Market Performance

• Author: JOSHUA MOLLNER,MARKUS BALDAUF
• DOI: http://doi.org/10.1111/jofi.12882
• Published date: 01 June 2020
• Date: 01 June 2020

THE JOURNAL OF FINANCE •VOL. LXXV, NO. 3 •JUNE 2020

High-Frequency Trading and Market

Performance

MARKUS BALDAUF and JOSHUA MOLLNER∗

ABSTRACT

We study the consequences of, and potential policy responses to, high-frequency trad-

ing (HFT) via the tradeoff between liquidity and information production. Faster

speeds facilitate HFT, with consequences for this tradeoff: Information production

decreases because informed traders have less time to trade before HFTs react, but

liquidity (measured by the bid-ask spread) improves because informational asymme-

tries decline. HFT also pushes outcomes inside the frontier of this tradeoff. However,

outcomes can be restored to the frontier by replacing the limit order book with one of

two alternative mechanisms: delaying all orders except cancellations or implementing

frequent batch auctions.

FINANCIAL MARKETS HAVE BEEN TRANSFORMED by faster speeds in recent years.

For example, the BYX exchange slashed its order processing time sevenfold,

from 445 μs in 2009 to 64 μs in 2018.1Likewise, the round-trip communication

time between Nasdaq and the Chicago Mercantile Exchange (CME) has nearly

halved, from over 14.5 ms in 2010 to 7.9 ms today.2

∗Markus Baldauf is with the Sauder School of Business, University of British Columbia. Joshua

Mollner is with the Kellogg School of Management, Northwestern University. We are indebted to

our advisors Timothy Bresnahan, Gabriel Carroll, Jonathan Levin, Paul Milgrom, and Monika

Piazzesi. We also thank Sandro Ambuehl, Sandeep Baliga, Robert Battalio, Dan Bernhardt, Philip

Bond, Eric Budish, Darrell Duffie, Georgy Egorov, Liran Einav, Lorenzo Garlappi, Will Gornall,

Joseph Grundfest, Terrence Hendershott, Peter Klibanoff, Fuhito Kojima, John Leahy, Alberto

Teguia,Muriel Niederle, Ricardo Perez-Truglia, Matthew Pritsker, Alvin Roth, Ilya Segal, Andrzej

Skrzypacz, Alireza Tahbaz-Salehi, Laura Tuttle, Xin Wang, Brian Weller, Glen Weyl, numerous

seminar participants, various industry experts, and anonymous referees for valuable comments.

We acknowledge financial support by the Kohlhagen Fellowship Fund and the Kapnick Fellowship

Program through grants to the Stanford Institute for Economic Policy Research. Baldauf gratefully

acknowledges financial support from the Social Sciences and Humanities Research Council of

Canada. Mollner was a Postdoctoral Researcher at Microsoft Research while part of this research

was completed, and he thanks them for their hospitality. We have read The Journal of Finance’s

disclosure policy and have no conflicts of interest to disclose.

Correspondence: Joshua Mollner, Kellogg School of Management, Northwestern University,

2211 Campus Drive, Evanston, IL 60208, USA; e-mail: joshua.mollner@kellogg.northwestern.edu.

1See BATS Global Markets, Inc. (BATS) (2009), BATS Global Markets, Inc. BATS (2018).

2See “Raging Bulls: How Wall Street Got Addicted to Light-Speed Trading,” Wired Magazine,

August 3, 2012 and Quincy Data (2019).

DOI: 10.1111/jofi.12882

C2020 the American Finance Association

1495

1496 The Journal of Finance R

Modern trading is also highly fragmented, with many stocks now traded at

over 30 venues, considerably more than just a decade ago. Given this fragmen-

tation, the faster speeds have increased the effectiveness of high-frequency

trading (HFT) strategies, including order anticipation, which we describe be-

low. A stylized fact is that latencies (the lag between when an order is sent to

an exchange and when it is processed) are random, which implies that traders

cannot ensure simultaneous processing of orders sent to several exchanges.

Thus, if HFTs are sufficiently fast, they may observe the trade generated by

the first such order to be processed by an exchange and react on the remaining

exchanges before the orders of the original trader are processed at those ex-

changes and in such a way that the original trader fails to trade successfully.

This can take two forms: (i) Traders may cancel their remaining quotes, which

we call “passive-side order anticipation,” or (ii) traders other than the original

trader may trade against the remaining quotes, which we call “aggressive-side

order anticipation.”3

In this paper, we present a model of order anticipation by HFTs, and we

analyze both its consequences and potential policy responses through the lens

of a well-known tradeoff between liquidity and information production. Our

three main findings are as follows. First, faster speeds allow HFTs to be more

successful at order anticipation, which improves liquidity in the sense of nar-

rowing the bid-ask spread but also lessens information production. Second,

order anticipation pushes outcomes inside the frontier of this tradeoff, which

represents an inefficiency of HFT, albeit one that differs from the commonly

voiced concern that it represents a socially costly arms race. Third, the in-

efficiency is due to aggressive-side (not passive-side) order anticipation, and

hence, certain alternative trading mechanisms can eliminate the inefficiency

by preventing aggressive-side order anticipation.

To obtain these results, we build a model that features random latency,

multiple exchanges, and a single security that is traded by liquidity investors,

information investors, and HFTs. Liquidity investors trade for exogenous hedg-

ing, saving, or borrowing motives. Information investors may, through costly

research, obtain and subsequently trade on private information about the se-

curity. HFTs may trade for profit by speculating or by facilitating transactions

with other traders. In the baseline, trading is conducted in limit order books

(LOBs).

In equilibrium, as in prior literature (e.g., Budish, Cramton, and Shim

(2015)), HFTs play two roles. One, the liquidity provider, facilitates trade by

posting quotes at all exchanges. The others, snipers, wait to trade until or-

der flow reveals a sufficiently strong signal of the value of the security. As is

standard, the liquidity provider faces adverse selection: Information investors

and snipers trade against her quotes only when the quotes are mispriced. To

3Internet Appendix Section II reports evidence of order anticipation that comes from a variety

of academic and industry sources. It also discusses the sources of randomness in latency and

the magnitudes in question. The Internet Appendix may be found in the online version of this

article.

High-Frequency Trading and Market Performance 1497

offset the resulting losses, the liquidity provider must set a bid-ask spread.

Also in equilibrium, and similar to, for example, Easley and O’Hara (1987),

information investors trade larger quantities than liquidity investors, so that

order flow signals the investor’s type. Because of random latency, an informa-

tion investor’s orders are not processed simultaneously, which allows HFTs

to act on this signal before he completes his trade. The liquidity provider re-

acts with passive-side order anticipation, that is, by sending cancellations for

her remaining quotes after observing one or more trades. Snipers react with

aggressive-side order anticipation, that is, by sending orders to trade against

the remaining quotes after observing two or more trades. The resulting winner-

take-all races may be won by the information investor, the liquidity provider,

or a sniper.

Using the comparative statics of the model, we evaluate the consequences

of recent improvements in HFT speed. Faster speeds enable HFTs to be more

successful at order anticipation, with two primary economic effects. A negative

effect is a reduction in information production. Intuitively, order anticipation

reduces the amount of rent that informed traders can extract by trading on a

piece of information, thereby weakening the incentive to obtain such informa-

tion. Less fundamental research is then conducted so that markets provide less

information about the fundamental value of the security, potentially generat-

ing further (unmodeled) distortions in the wider economy. However, a positive

effect is an improvement in liquidity as measured by the bid-ask spread. Order

anticipation achieves this by reducing the adverse selection faced by liquidity

providers through two channels: (i) Passive-side order anticipation is itself suc-

cessful avoidance of adverse selection, and (ii) through its effect on research,

order anticipation reduces the amount of asymmetric information available to

create adverse selection. These two predicted consequences are in line with the

conclusions of many empirical studies, and as we discuss below, they suggest a

need to reinterpret the conclusions of others.

In sum, when trading is conducted in LOBs, faster HFT speeds induce a

tradeoff: Information production declines, but the bid-ask spread narrows. An-

other question that arises is whether these outcomes lie on the frontier of

this tradeoff. We show that they generally do not. Again, the reason is order

anticipation—more precisely, aggressive-side order anticipation. To see why, a

necessary condition for reaching the frontier is that all profits from informed

trading accrue to the agents who actually produce the information. Aggressive-

side order anticipation violates this condition, as it amounts to snipers profiting

from information that they did not produce. We formalize this insight by fol-

lowing the spirit of mechanism design and optimizing over a general class of

trading mechanisms so as to characterize the frontier. We also show that this

frontier can be achieved by replacing the prevailing LOB with either of two

plausible alternatives.

Noncancellation delay mechanisms (NDs) add a small delay between receipt

at an exchange and processing for all order types but cancellations. The re-

sult is to eliminate aggressive-side order anticipation by allowing the liquidity

provider to cancel mispriced quotes before they can be exploited by snipers.