Fast and slow cancellations and trader behavior

• Published date: 01 December 2020
• DOI: http://doi.org/10.1111/fima.12298
• Author: Thomas H. McInish,Irina Panovska,Alex Nikolsko‐Rzhevskyy,Olena Nikolsko‐Rzhevska
• Date: 01 December 2020

DOI: 10.1111/fima.12298

ORIGINAL ARTICLE

Fast and slow cancellations and trader behavior

Thomas H. McInish1Olena Nikolsko-Rzhevska2

Alex Nikolsko-Rzhevskyy3Irina Panovska4

1FogelmanCollege of Business and Economics,

University of Memphis, Memphis, Tennessee

2Department of Finance, LehighUniversity,

Bethlehem, Pennsylvania

3Department of Economics, LehighUniversity,

Bethlehem, Pennsylvania

4Department of Economics, University of Texas

at Dallas, Richardson, Texas

Correspondence

OlenaNikolsko-Rzhevska, Lehigh University,621

TaylorSt.,Bethlehem, PA 18015.

Email:olena.rzhevska@gmail.com

Fundinginformation

LehighUniversity, Grant/AwardNumber:

FRG2017

Abstract

We investigate how short-lived liquidity supplydue to order cancel-

lations affects the order-placement behavior of slow traders. When

order cancellations increase, slow traders submit fewer and less

aggressive orders. Both short- and long-lived liquidity supply have

positive effects on the market overall,reducing spreads and increas-

ing depth. Weconclude that it is not necessary to require limit orders

to have a minimum lifespan. We develop econometric and machine-

learning frameworks that allow tradersto predict whether a quote is

likelyto have a short or long life, increasing the ability of slow traders

to respond strategically to changing order flow.

KEYWORDS

order cancellations, liquidity provision, machine learning, algorith-

mic trading

1INTRODUCTION

In the past decade, changes in marketregulations and advancements in trading technology have dramatically reshaped

liquidity provision and executionstrategies in modern markets. While more liquidity is being supplied, order cancella-

tion rateshave increased from 5% of all orders in 1990 (Yeo, 2005) to 9% in 1994–1995 (Lo, MacKinlay,& Zhang, 2002)

and to more than 94% in our 2018 data with approximately half of the quotes lasting less than half a second.1The dra-

matic increase in the short-lived liquidity supply and cancellation rates has prompted concerns that displayedliquidity

is cancelled before institutional traders can tradeagainst it (Angel, 2014; CFA Institute, 2015; Economist, 2016; Hope,

2013; SEC, 2010).

Frequent placement and almost instantaneous cancellations of limit orders are associated with the presence of

high-frequency quoters (HFQers). In 1994–1995, when there were few HFQers, Lo et al. (2002) report that 91% of

orders were fully or partially executed.2If,due to the rapid cancellation of orders, market participants, discouraged

by their inability to predict cancellations, pull out of the market, submit fewer orders, or post less aggressive orders,

this reduction in liquidity may harm marketquality. Unlike most previous work, Rosu (2009) develops a model in which

c

2019 Financial Management Association International

1Wecalculate order duration from our sample.

2Inthe Supporting Information, Table A1, we present a table similar to Table1 in Lo, MacKinlay, and Zhang (2002).

Financial Management. 2020;49:973–996. wileyonlinelibrary.com/journal/fima 973

974 MCINISH ET AL.

cancellations play an integralrole in establishing market equilibrium. This author also provides a detailed discussion of

the previous models of the limit order book (LOB)and points out that these models lack a role for cancellations.

As our first contribution, we examine the behavioral response of non-HFQers to the cancellations of HFQers. We

define HFQ (non-HFQ) orders as those lasting less (more)than 50 ms.3According to Hoffmann (2014), the presence of

fast traders presents a dilemma for slow traders who can choose to quote more aggressively or accept a lower execu-

tion probability.We find that when the number of short-lived HFQ orders increases, order placements by non-HFQers

become less aggressive and order placements of HFQers become more aggressive. Specifically, when HFQers’ can-

cellations double, non-HFQ orders become $0.02 farther away, whereas HFQ orders become $0.02 cents closer to

the best bid–ask midpoint. This is consistent with Carrion (2013) who reports that HFQers supply liquidity when it is

expensive and take liquidity when it is cheap.4In addition, in line with the model of Li, Wang,and Ye (2019), HFQers’

orders are more likely to be at the top of the LOB, which increases the HFQers’ share of limit order executions.Thus,

our results indicate that slow traders choose the second option proposed by Hoffmann (2014). Fong and Liu (2010)

find that limit orders are more likelyto be cancelled if they are near the bid–ask quote. By reviewing the pooled dataset

consisting of both HFQ and non-HFQ orders, we find that as the order is closer to the top of the LOB, ceteris paribus,

its cancellation probability increases. This is consistent with the findings of Fong and Liu (2010). Weextend their work

by demonstrating that as distance to the midpoint decreases, cancelled order durationalso decreases, but more so for

non-HFQ orders than for HFQ orders.5

Does this mean that cancellations are detrimental to market quality and regulators should restrict their usage?

For regular traders, cancellations are essential as they allow these traders to reduce nonexecution and picking-off

risk (Fong & Liu, 2010; Liu, 2009; Yeo,2005). Nonexecution risk arises when a trader’s limit order spends time in the

LOBand the market price moves away from the limit order price reducing the probability of execution. Picking-off risk

occurs when new information arrives and a trader who does not update their limit order price risks being pickedoff by

more informed or faster traders. For algorithmic traders and marketmakers, cancellations are used in various strate-

gies that rely on speed to reduce undercutting exposure(Baruch & Glosten, 2017) and pricing inefficiencies (Hoffmann,

2014), search for latent liquidity (Hasbrouck & Saar, 2009), or increase the probability of execution (Jain & Jordan,

2017; van Kervel, 2015).6Additionally,cancellations could also have different effects on the market depending upon

whether or not a cancelled order is at the top of the LOB.

Accordingly,as our second contribution, we directly analyze the impact of HFQ and non-HFQ cancellations on mar-

ket quality. Webelieve that we are the first to differentiate between HFQ and non-HFQ cancellations when investi-

gating market quality.Employing a research design in the spirit of Hasbrouck and Saar (2013), we find that both HFQ

and non-HFQ quote cancellations narrow spreads and improve market depth thereby enhancing marketquality. HFQ

cancellations account for 25% of the depth improvement and non-HFQ cancellations for the remainder.

Our analysis of market quality may be of interest as regulators are considering implementing policies, such as

cancellation fees and a minimum quote life, to reduce cancellations on the U.S. stock exchanges (SEC, 2010). Our

analysis suggests that implementing a cancellation fee could reduce market quality.Because our findings indicate that

both HFQ and non-HFQ quotes are equally beneficial, there may be no need for a minimum quote life requirement.

Our findings reinforce those of Griffith and Van Ness (2019) who find that the introduction of an order execution

fee on the PHLX options market reduced the rate of limit order cancellations, but widened bid–ask spreads (by

discouraging the submission of nonmarketableorders and encouraging the submission of marketable orders). Because

wider bid–ask spreads likely increase transactions costs, the results of these authors also indicate that exchangesand

regulators need to be cautious about limiting cancellations.

3Seethe U.S. Senate discussion by the Subcommittee on Securities and Investment (2012).

4Our finding that HFQs’ (non-HFQs’) share of submissions at the top of the LOB is smaller (greater) during high volumeperiods is also consistent with this

finding.

5Table7 presents this finding. Non-HFQ orders, in general, live a lot longer than HFQ orders, but their duration decreases faster with distance.

6Certainmarket manipulation strategies of algorithmic traders also rely heavily on the ability to rapidly submit and cancel a large number of quotes (Blocher,

Cooper,Seddon, & van Vliet, 2016; Egginton, Van Ness, & Van Ness, 2016;Manahov,2016; NANEX, 2010).