Zahar Udin PHD Thesis

Copyright is owned by the Author of the thesis.
Permission is given for

a copy to be downloaded by an individual for the purpose of research and
private study only. The thesis may not be reproduced elsewhere without
the permission of the Author.
Essays on High-frequency Trading
A thesis presented in partial fulfilment of the requirement for the degree of
Doctor of Philosophy
in
Finance
at Massey University, Manawatu campus,

New Zealand
Khairul Zharif Zaharudin

2023
ABSTRACT
This thesis presents a comprehensive review of the relevant literature on high-

frequency trading (HFT). There is no universal definition on HFT to date, leading to
inaccurate estimations of their reach and impact in the market. HFT is a specialised
form of algorithmic trading with lightning-speed network and implement complex
trading strategies. HFT may have both beneficial and harmful effects on the market,
and their speed could disrupt the market at remarkable pace. There are several
controversies related to HFT, including the 2010 flash crash, social welfare and arms
race, and market-making responsibilities. Several initiatives were implemented or
proposed as a response to HFT, including speed bumps, price improvement rules,
and new trading mechanics to guarantee a fair and orderly market for everyone. The
second essay examines the impact of relative tick size on HFT activity. Using data
from Australian equity markets, relative tick size is shown to have an inverse
relationship with HFT activity. The findings demonstrate that HFTs have a very low
tolerance for adverse selection risk, and their risk-averse nature prioritise risk
minimisation (order-undercutting) over profit maximisation (order-queuing). The
results lend credence to the perception that the primary strategy of HFTs is to
generate thin profits while keeping their risk exposure to an absolute minimum. The
evidence imply that policymakers may implement a dynamic tick size policy to
allocate HFT activity into stocks where it is most required. The third essay illustrates
how expected volatility affects HFT activity, and how the resulting change in HFT
activity impacts liquidity. Using data from S&P/ASX 200 VIX index (AXVI) to
measure sentiment, evidence suggests that when the sentiment is negative: (i) HFTs’
ability to promote trading volume on the ASX is hampered; (ii) HFTs’ ability to
reduce spreads on the CHIX is dampened; and (iii) HFTs’ ability to reduce price
impact is amplified on both markets. Overall, HFTs’ presence may significantly
reduce excessive price impacts when the market is fearful. Therefore, any attempt to
completely ban or restrict HFT in the market might accidentally eliminate valuable
market participants.
ii
DEDICATION
To my dearest wife,
Dr. Nor Elliany Hawa Ibrahim
who has been my pillar of strength throughout this colourful journey. Your
unconditional love, support, and encouragement have kept me going through the
countless challenges that we have faced together. Your unwavering faith in me has
been a constant source of inspiration, and I am forever grateful for the happiness and
warmth that you bring into my life.
To my beloved parents,
Zaharudin Abu Nasir and Kamsiah Zakaria
Words cannot express the depth of gratitude and admiration I have for both of you.
Your teachings on the importance of education, knowledge, and excellence have led
me on this challenging yet immensely rewarding journey. Without your everlasting
love and continuous prayers, none of this would have been possible. Because of you,
I am who I am now, and I will forever be indebted to you.
May the Almighty recompense all of you with abundance of goodness, in this life
and the Hereafter.
iii
ACKNOWLEDGMENT
First and foremost, I would like to express my deepest gratitude to Allah, the Most
Gracious, the Most Merciful. I am eternally grateful for the perseverance and
fortitude granted to me to complete this thesis, and for the countless blessings
bestowed upon me throughout my life. Alhamdulillah. Secondly, I would like to
offer blessings and peace upon the Prophet Muhammad, whose exemplary life has
been a constant source of inspiration for me.
My sincerest appreciation goes to my supervisors, Prof. Martin R. Young and

Dr. Wei-Huei Hsu, for their invaluable guidance, encouragement, patience, and
dedication to facilitating this research. It has been an immense privilege and honour
to learn under their tutelage. I am also grateful for the financial support provided by
Universiti Utara Malaysia and the Ministry of Higher Education of Malaysia during
my stay in New Zealand, without which this achievement would not have been
possible.
I would like to extend my heartfelt gratitude to my beloved wife, Dr. Nor

Elliany Hawa Ibrahim, for her steadfast support and unconditional love throughout this
journey. Her constant presence by my side has been instrumental in the completion of
this thesis. To my loving parents, Zaharudin Abu Nasir and Kamsiah Zakaria, I cannot
thank them enough for their earnest prayers, encouragement, and kindness. To my
caring parents-in-law, Ibrahim Mohd Noor and Aziah Alwi, I convey my sincerest
gratitude for their prayers and understanding. To my dear friends, especially Dr. Bilal
Hafeez and Nadiya (Khalid’s wife), I am indebted for their invaluable assistance in
handling my extensive data. To my mentor, Prof. Dr. Kamarun Nisham, I am grateful
for his wisdom and insights. To my siblings (Kak Wani, Dr. Zati Aqmar, Adik Hazira)
and in-laws (Abang Wan, Ana, Hafiz, Aish, Abe Pi, Abe Syahir, and Pa’an), I am
thankful for their encouragement, support and belief in me.
I sincerely thank all of you for the kind gestures and inspiring roles you have
played in my life. Your support has given me the strength, confidence and
motivation to forge ahead and make this dream a reality. Thank you.
iv
TABLE OF CONTENTS
ABSTRACT ................................................................................................................ii
DEDICATION .......................................................................................................... iii
ACKNOWLEDGMENT ..........................................................................................iv
TABLE OF CONTENTS ........................................................................................... v
LIST OF TABLES ................................................................................................. viii
LIST OF FIGURES .................................................................................................. ix
LIST OF ABBREVIATIONS ................................................................................... x
CHAPTER ONE: INTRODUCTION ...................................................................... 1
1.1 Introduction .................................................................................................... 1
1.2 Essay One: High-frequency Trading: Definition, Implications, and
Controversies ........................................................................................................... 4
1.3 Essay Two: Relative Tick Size and High-frequency Trading ....................... 5
1.4 Essay Three: Expected Volatility, High-frequency Trading, and Liquidity . 6
1.5 Structure of the Thesis ................................................................................... 7
CHAPTER TWO: ESSAY ONE – HIGH-FREQUENCY TRADING:
DEFINITION, IMPLICATIONS, AND CONTROVERSIES ............................... 9
2.1 Introduction .................................................................................................... 9
2.1.1 Defining HFT ............................................................................................ 10
2.1.2 HFT identification ..................................................................................... 14
2.2 Trading Mechanics....................................................................................... 17
2.3 Trading Strategies ........................................................................................ 20
2.3.1 Beneficial Strategies .................................................................................. 22
2.3.1.1 Statistical Arbitrage .................................................................... 23
2.3.1.2 Directional Trading ..................................................................... 24
2.3.1.3 Market Making............................................................................ 25
2.3.2 Harmful Strategies ..................................................................................... 27
2.3.2.1 Front running, order anticipation, and quote matching............... 28
2.3.2.2 Spoofing and layering ................................................................. 29
2.3.2.3 Quote stuffing ............................................................................. 31
2.4 Effects on Market Quality ............................................................................ 33
2.4.1 Price Discovery and Efficiency ................................................................. 33
2.4.2 Liquidity .................................................................................................... 36
2.4.3 Volatility .................................................................................................... 38
2.4.4 Adverse Selection Costs ............................................................................ 40
2.5 Controversies Relating to HFT .................................................................... 42
2.5.1 Flash Crash of May 6, 2010 ...................................................................... 42
2.5.2 Information acquisition, social welfare issues, and the HFT arms race .... 45
2.5.3 Market-making obligations and excessive liquidity .................................. 47
2.6 Responses to HFT ........................................................................................ 48
2.7 Conclusions .................................................................................................. 50
CHAPTER THREE: ESSAY TWO – RELATIVE TICK SIZE AND HIGH-
FREQUENCY TRADING ...................................................................................... 52
3.1 Introduction .................................................................................................. 52
3.2 HFT in Australia .......................................................................................... 53
3.3 Tick size structure in Australia .................................................................... 56
v
3.4 Literature review and hypotheses development ........................................... 60
3.4.1 Nominal and relative tick sizes.................................................................. 60
3.4.2 HFTs as market-makers ............................................................................. 62
3.4.2.1 Large tick size and order-queuing activity.................................. 63
3.4.2.2 Small tick size and order-undercutting activity .......................... 64
3.4.2.3 Relative tick size trade-off .......................................................... 66
3.4.3 Hypotheses development ........................................................................... 66
3.5 Methodology ................................................................................................ 68
3.5.1 Data description ......................................................................................... 68
3.5.2 Sample selection ........................................................................................ 70
3.5.3 Measurement of variables .......................................................................... 73
3.5.3.1 HFT activity measures ................................................................ 73
3.5.3.2 Explanatory variables.................................................................. 78
3.5.4 Model specification ................................................................................... 80
3.6 Results, analyses, and discussions ............................................................... 83
3.6.1 HFT activity across different RTS groups categorised by nominal tick sizes 84
3.6.1.1 Descriptive statistics ................................................................... 84
3.6.1.2 Univariate analysis ...................................................................... 87
3.6.1.3 Multivariate analysis ................................................................... 89
3.6.2 HFT activity surrounding a tick size border .............................................. 92
3.6.2.1 Descriptive statistics ................................................................... 92
3.6.2.2 Univariate analysis ...................................................................... 95
3.6.2.3 Multivariate analysis ................................................................... 97
3.6.3 HFT activity surrounding tick size crossing events................................. 102
3.6.3.1 Descriptive statistics ................................................................. 102
3.6.3.2 Univariate analysis .................................................................... 106
3.6.3.3 Difference-in-difference multivariate analysis ......................... 109
3.7 Conclusion ................................................................................................. 113
CHAPTER FOUR: ESSAY THREE – EXPECTED VOLATILITY, HIGH-
FREQUENCY TRADING, AND LIQUIDITY ................................................... 116
4.1 Introduction ................................................................................................ 116
4.1.1 Volatility index in Australia .................................................................... 118
4.2 Literature review and hypotheses development ......................................... 120
4.2.1 Liquidity .................................................................................................. 120
4.2.2 High-frequency trading strategies ........................................................... 122
4.2.3 Hypotheses development ......................................................................... 127
4.3 Methodology .............................................................................................. 129
4.3.1 Data description ....................................................................................... 129
4.3.2 Sample selection ...................................................................................... 129
4.3.3 Measurement of variables ........................................................................ 133
4.3.3.1 Liquidity measures .................................................................... 133
4.3.3.2 HFT activity measures .............................................................. 135
4.3.3.3 Other variables .......................................................................... 138
4.3.4 Model specification ................................................................................. 139
4.4 Results, analyses, and discussions ............................................................. 141
4.4.1 HFT activity and liquidity surrounding the VIXRT introduction event ... 142
4.4.1.3 Multivariate analysis ................................................................. 146
vi
4.4.2 Liquidity, VIX and HFT activity ............................................................. 151
4.4.2.3 Multivariate analysis ................................................................. 159
4.5 Conclusion ................................................................................................. 162
CHAPTER FIVE: CONCLUSION ...................................................................... 166
5.1 Major findings and implications ................................................................ 166
5.1.1 Essay One: High-frequency trading: Definition, Implications, and
Controversies .................................................................................................... 166
5.1.2 Essay Two: Relative Tick Size and High-frequency Trading ................. 167
5.1.3 Essay Three: Expected Volatility, High-frequency Trading, and Liquidity169
5.2 Limitations of the study and suggestions for future research .................... 170
REFERENCES ....................................................................................................... 173
APPENDIX ............................................................................................................. 184
vii
LIST OF TABLES
Name Title Page

Table 2.1 Summary of HFT characteristics from regulatory perspectives 13
Table 3.1 Tick size structure in the Australian and U.S. equity markets 59
Table 3.2 Relevant information from the datasets 69
Table 3.3 Data description for the samples used to test the first hypothesis 70
Table 3.4 Data description for the samples used to test the second 71
hypothesis
Table 3.5 Identification and classification of crossing events 72
Table 3.6 Data description of the crossing events selected to test the third 73
hypothesis
Table 3.7 Descriptive statistics of the data used to test Hypothesis 1 85
Table 3.8 Mean comparison of HFT activity categorised by nominal tick 88
sizes
Table 3.9 Regression analysis on HFT activity categorised by nominal tick 90
sizes
Table 3.11 Mean comparison of HFT activity near the A$2.00 tick size 96
border
Table 3.12 Regression analysis on HFT activity near the A$2.00 tick size 99
border
Table 3.14 Univariate difference-in-difference analysis of HFT activity 107
measures for tick size crossing events
Table 3.15 Regression analysis on HFT activity using tick size crossing 110
events
Table 4.1 Data description for the samples used to test the first hypothesis 130
Table 4.2 Percentile values for VIXCLOSE and HFO 132
Table 4.3 Description of the datasets used to test the second hypotheses 132
Table 4.5 Mean comparison of liquidity measures and HFT activity in the 145
periods surrounding the VIXRT event
Table 4.6 Regression analysis on liquidity measures following the VIX RT 147
event
Table 4.8 Mean comparison of liquidity measures during Event (VIXH) 157
and Non-event (VIXNH) days between Treatment (HFTH) and
Control (HFTNH) groups
Table 4.9 Regression analysis on liquidity measures while controlling for 160
the effect of high VIX days and high HFT activity
Table A1 Distribution of returns on day t+21 based on the VIX value on 184
day t
Table A2 Heatmap analysis using data from the ASX market (sort by HFO) 185
Table A3 Heatmap analysis using data from the CHIX market (sort by HFO) 188
viii
LIST OF FIGURES
Name Title Page

Figure 3.1 Quarterly average of message-to-trade ratio (MTR) in Australia 55
Figure 3.2 Quarterly average of algorithmic trading ratio (ALGO) in Australia 55
Figure 3.3 Quarterly average of high-frequency order ID (HFO) in Australia 55
Figure 3.4 Quarterly average of HFO’s message ratio (HFOR) in Australia 56
Figure 3.5 Relative tick size for stock prices ranging from A$0.01 to A$5.00 57
Figure 3.6 Relative tick size for stock prices surrounding the A$0.10 border 57
Figure 3.7 Relative tick size for stock prices surrounding the A$2.00 border 58
Figure 3.8 Illustration of the research design used to test the third hypothesis 82
(UPWARDS)
Figure 3.9 Illustration of the research design used to test the third hypothesis 82
(DOWNWARDS)
Figure 4.1 Weekly average of S&P/ASX 200 index (AXJO) and S&P/ASX 119
200 VIX index (AXVI)
Figure 4.2 Variation in the AXJO’s price over the next 21 trading days (to 119
the nearest 2%)
Figure 4.3 Illustration of the research design used to test the second hypothesis. 140
ix
LIST OF ABBREVIATIONS
Abbreviation Definition
AETH Australian Equities Tick History
ALGO Algorithmic trading ratio
ASIC Australian Securities and Investments Commission
ASX Australian Securities Exchange
AT Algorithmic trading
AXJO S&P/ASX 200 index
AXVI S&P/ASX 200 VIX index
CBOE Chicago Board Options Exchange
CESR Committee of European Securities Regulators
CFTC Commodity Futures Trading Commission
CHIX Chi-X Australia
CROSS Tick size crossing event
CSHL Corwin-Schultz high-low spread
DIFF Difference
DIFF-IN-DIFF Difference-in-difference
DMMs Designated market makers
ELPs Endogenous liquidity providers
FINRA Financial Industry Regulatory Authority
HFO High-frequency order identifiers
HFOR HFO’s message ratio
HFT High-frequency trading
HFTs High-frequency traders
ILLIQ Amihud illiquidity ratio
ITCH ASX ITCH dataset
LSE London Stock Exchange
MiFID Markets in Financial Instruments Directive
MPID Market participant identifier
MTR Message-to-trade ratio
NAFM Netherlands Authority for the Financial Markets
NASDAQ National Association of Securities Dealers Automated Quotation
NBBO National Best Bid and Offer
NOMX-St NASDAQ OMX Stockholm Exchange
NYSE New York Stock Exchange
OrderID Unique order identification
OTR Order-to-trade ratio
Reg NMS Regulation National Market System
RTS Relative tick size
SEC Securities and Exchange Commission
SIRCA Securities Industry Research Centre of Asia-Pacific
TSX Toronto Stock Exchange
UAL United Airlines
VIX The CBOE Volatility Index
VOLTO Volume Turnover
x
CHAPTER ONE:
INTRODUCTION
1
This chapter provides an introduction and overview of this thesis. The chapter starts
with a brief background on high-frequency trading, followed by motivations and
objectives underlying the research. The chapter continues with overviews of the
three essays individually, including some key findings and contributions. The
chapter concludes with an outline of the thesis organisation.
1.1 Introduction
Technology has revolutionised trading. The use of fibre-optic cables and microwave
towers to transmit trading information at the speed of light gave rise to high-
frequency trading (HFT). In 2009, HFT accounted for 60% of U.S. equity trading, up
from 20% in 2005. The proportion steadily declined to 50% in 2013 (Avramovic et
al., 2017) and has remained steady at 52% in 2018. In Europe, HFT contributed
practically 0% to total equity trading in 2005, peaked at 40% in 2010, and stabilised
at 35% in 2014 (Kaya, 2016). In Australia, HFT is reported to be responsible for
27% of total stock market transactions in S&P/ASX 200 securities from January
2012 to March 2015, where the 10 biggest HFTs accounted for 21% of total trading
turnover in 2015 (The Australian Securities and Investments Commission [ASIC],
2015).
Speed is the defining characteristic that distinguishes high-frequency traders

(HFTs) from all other investors that engage in algorithmic or automated trading
(AT). From spotting lucrative opportunities to implementing the right trading
strategies, the trading speeds of HFTs almost approached the theoretical speed of
light, which is around 300,000 kilometres per second (Angel, 2014). This capacity
enables certain trading strategies to be limited to HFTs only, since other market
players would be unable to replicate strategies requiring lightning-fast operation
(Harris, 2013).
1
In addition, HFTs would employs quantitative analysts (“quants”), often from
a mathematics or statistics background, to develop the numerous automated trading
algorithms to capitalise on profitable trading opportunities. Moreover, due to the
winner-take-all nature of the HFT industry, HFTs must constantly upgrade their
infrastructures to minimise their automated trading latency (Kauffman et al., 2015).
This culminates in an arms race among institutions to be the fastest, which is seen
unproductive and socially wasteful, and the enormous amount of money spent to
minimise trading latency raises concerns about whether HFT adds value overall
(Chordia et al., 2013; Budish et al., 2015; Jones, 2013).
The business model of HFT relies on the execution of numerous trades with
very thin profit margins, often as low as a few basis points; however, the massive
trading volume amplified the small returns (Netherlands Authority for the Financial
Markets [NAFM], 2010; Zhang, 2010). HFTs attempt to identify and profit from
short-term inefficiencies and make tiny returns from each transaction through latency
arbitrage; this allows them to profit from the trading environment itself, rather than
from investing in financial assets (Hasbrouck & Saar, 2013). Aquilina et al. (2020)
estimate that abolishing latency arbitrage from the global equities market would
reduce the market’s cost of liquidity by 17%, with a total stake of approximately
US$5 billion annually.
HFTs have diverse trading strategies, which affects their sensitivity to

inventory levels and price swings (Benos & Sagade, 2016). HFT strategies may be
beneficial or harmful for the market, depending on their impact. Beneficial strategies
are acceptable since they do not disrupt the market and enhance market quality.
These include statistical arbitrage (value-motivated), directional trading strategies
(informed trading), and market making (liquidity trading). In contrast, harmful
strategies, such as front running (e.g., order anticipation, quote matching), spoofing
and layering (i.e., uses fake orders to create the illusion that the market will move),
and quote stuffing (i.e., congesting an exchange’s network bandwidth with non-
marketable limit orders to hinder competitors), generate profits at the cost of others
through unscrupulous means.
2
For these reasons, HFT may be considered as a double-edged sword; while it
may be beneficial in enhancing price informativeness or fostering liquidity via
market-making activity, it may also prey on slower traders or hinder their ability to
trade if they employ the harmful strategies. For instance, during the flash crash of
May 6, 2010, HFTs, who normally provide liquidity, turned into liquidity takers.
Although HFTs are not responsible for initiating the flash crash, they are
criticised for exacerbating price falls and market volatility by generating a “hot-
potato effect” in their competition for liquidity on the day of the crash (Kirilenko et
al., 2017).
Furthermore, the limit order placements of HFTs resemble the harmful order-
anticipation strategy in that they provide liquidity on the thick side of the order book,
where depth is not required, and take liquidity on the thin side of the order book,
where depth is necessary (Goldstein, et al., 2020). Thus, despite the fact that HFTs
bring liquidity to the market, they do so only when it is least essential and avoid
doing so when it is most crucial. Since they are not mandated by law to provide
liquidity, market-making HFTs’ reliability to do so is also questioned, especially
during times of crisis (Anand & Venkataraman, 2013; Chung & Chuwonganant,
2018; Zhang, 2010).
Nonetheless, there is lack of agreement on the definition of HFT (NAFM,

2010), which contribute to the diverse findings on the consequences of HFT activity
in the market, and an imprecise definition would be costly and harm all classes of
market participants (Zaharudin et al., 2022). Therefore, it is imperative for the
progression of finance research to accurately define and identify HFT in particular
and AT in general, and to properly understand their impacts on the market. As
remarked by Li et al. (2021, p. 980), “just as insights into human behaviour from the
psychology literature spawned the field of behavioural finance, so insights into
algorithmic behaviour could prompt an analogous blossoming of research in
algorithmic finance.”
3
1.2 Essay One: High-frequency Trading: Definition, Implications, and
Controversies
The first essay of this thesis provides a comprehensive survey of the relevant
literature on HFT, including regulatory reports, theoretical and empirical research.
According to NAFM (2010), there is not a unified definition nor a consensus on
HFT. This leads to inaccurate estimations of HFTs’ market share, complicated
research, and confusion between HFT and other forms of trading, which leads to
HFT/HFTs being blamed for issues with which they have little or nothing to do, like
the 2010 flash crash. Thus, this essay presents an in-depth literature review to
comprehend how regulatory agencies around the world define and identify HFT.
In essence, HFT activity can be traced with the cooperation from exchanges
and regulators, since they possess complete knowledge of the identity of each
transaction message sent to the market. Alternatively, HFT may be identified by its
unique characteristics; however, this method can only provide an approximation of
HFTs’ presence and impact in a market. In general, there are five traits are widely
associated with HFT: (i) a specialised form of AT, (ii) utilisation of high-speed,
sophisticated computer programmes and systems, (iii) very high order-to-transaction
ratios, (iv) extremely short average holding periods, and (v) ending the trading day
with flat positions.
HFT is a specialised form of AT that uses its superior speed to implement

complex trading strategies. In general, the only unique aspect of HFT is its speed; the
underlying mechanics are quite ordinary. Their short-term trading strategies have
existed for a long time, and their methods for earning a profit on the market, such as
attempting to buy low and sell high, are essentially identical to those used by other
traders. Therefore, the essay also discusses HFT mechanics, strategies, and their
effects on market quality from scholarly articles and reports.
Furthermore, literature suggests that HFT may have both beneficial and
harmful effects on the market, and their lightning speed implies they could disrupt
the market at remarkable pace. Therefore, the essay also discusses debates
concerning HFT, including (i) the flash crash that occurred on May 6, 2010; (ii)
issues regarding HFTs’ acquisition of information, social welfare, and arms race; and
4
(iii) market-making responsibilities and excessive liquidity. The article concludes
with regulatory and academic responses to HFT, including speed bumps, price
improvement rules, and new trading mechanics to guarantee a fair and orderly
market for everyone.
1.3 Essay Two: Relative Tick Size and High-frequency Trading

The second essay examines the impact of relative tick size on HFT activity. In
contrast to the fixed nominal tick size established by the regulator, the relative tick
size varies in reaction to stock price movements and nominal tick size values. A
relative tick size may be shocked when a regulator revises an existing nominal tick
size value, when a stock price naturally crosses a tick size border, or when a
corporate action or informational event results in a significant stock price movement.
A larger (smaller) relative tick size suggests a higher (lower) profit margin
from market-making spread and an increased (decreased) adverse selection risk from
limit order placements for liquidity providers, including market-making HFTs (see
e.g.: Angel, 1997; O’Hara et al., 2019; Sandås, 2001). Thus, the magnitude of a
nominal or relative tick sizes would offer liquidity providers with an alternative
between profit maximisation and risk minimization. This trade-off hinges on the
liquidity provider’s risk appetite, i.e., whether they’re prepared to assume more risk
for a larger return (profit maximisation) or accept a lower profit to reduce risk
exposure (risk minimisation). In light of the notion that a discrepancy in tick size
(small or large) is indicative of a different risk appetite and tolerance, this essay
analyses the levels of HFT activity for two groups of stocks with distinct tick size
values.
The study uses order-book data from the Australian equity market, which
includes the Australian Securities Exchange (ASX) and Chi-X Australia (CHIX).
Many empirical studies on tick size are primarily focused on the U.S. financial
markets (see e.g. Angel, 1997; O’Hara et al., 2019; Schultz, 2000; Yao & Ye, 2018).
However, the variations in Australian and American market architecture may affect
HFTs’ strategy, activity, and behaviour, driving them to perform differently in each
country. The Australian market, for instance, employs a different tick size structure,
lacks a designated market-maker, and has more flexible minimum trading unit
5
requirements, all of which provide a unique playing field for HFTs. This study
contributes to a better understanding of HFT, notably the impact of tick size on their
activity and strategy in an order-driven market.
Additionally, the study proposes new proxies for HFT activity, namely high-
frequency OrderIDs (HFO) and HFO’s message ratio (HFOR). These measures are
based on the fact that HFT activity often have short average holding periods and a
large number of interconnected messages due to rapid order modifications. The
calculation of these measures involves information that is commonly available from
order-book data, thus improving its reproducibility and allowing it to be tested in
other markets and over longer time periods.
The results of the second essay show that market-making HFTs in the
Australian equity market favour stocks with a small relative tick size, suggesting that
they are more concerned with minimising risk than maximising profit, and thus
prefer the order-undercutting strategy over the order-queuing approach. Moreover,
the results stand in contrast to the notion that market-making HFTs will reduce their
liquidity provision in response to smaller relative tick sizes. The findings imply that
policymakers may implement a dynamic tick size policy, in which stocks that may
benefit from more HFT activity are allocated a smaller tick size, and those with
excessive HFT activity are assigned a larger tick size.
1.4 Essay Three: Expected Volatility, High-frequency Trading, and

Liquidity
The final essay illustrates how expected volatility affects HFT activity, and how the
resulting change in HFT activity impacts liquidity. The CBOE Volatility Index
(VIX), popularly known as the “fear index,” represents expected volatility since it
reflects investors’ expectations of near-term future volatility in the stock market, and
thus, it measures stock market sentiment (Reilly & Brown, 2012). Whaley (2009)
highlights that the VIX and SPX move at asymmetric rates, and investors’ fear of a
bear market outweighs their excitement (or greed) for a bull market. This suggests
that a high VIX signals investors’ fear of a market crash, whereas a low VIX
signifies investors’ confidence.
6
Market-making HFTs are not legally obligated to offer liquidity at all times;
consequently, they may choose to do so only when they perceive it to be profitable.
Moreover, when circumstances are unfavourable for them, such as when adverse
selection risks are higher amidst heightened market uncertainty (e.g. when the VIX is
high), HFTs may decide to cease providing liquidity (Anand & Venkataraman, 2013;
Chung & Chuwonganant, 2018; Zhang, 2010). For this reason, it is feared that
HFTs’ discretionary market making may worsen execution uncertainty, especially in
volatile markets and in thinly traded securities. As a consequence, investors as a
whole may lose confidence in the liquidity offered by HFTs, and pull out of the
market completely. Using the S&P/ASX 200 VIX (AXVI) volatility index, this essay
investigates whether market sentiment, as measured by the AXVI, would impact
HFT activity, and whether the effect on HFT activity (if any) would then influence
liquidity. In this essay, three distinct liquidity indicators are employed: the Corwin-
Schultz spread, the Amihud illiquidity ratio, and the volume turnover. The findings
of this research clarify whether HFTs can be trusted even during periods of
heightened market volatility, and whether their activity promotes or hampers
liquidity during such times.
The results indicate that when market sentiment is negative (i.e., when the
VIX is high), the positive effects of HFT in promoting volume turnover on the ASX
and lowering spreads on the CHIX are dampened; however, the ability of HFT to
reduce the illiquidity ratio is amplified on both markets. Overall, this study support
the notion that negative market sentiment does influence the impact of HFT on
liquidity. Despite the fact that HFT’s ability to enhance trading activity and reduce
spread is impaired when the market is extremely fearful, their presence at such times
has been shown to be beneficial for reducing price impact. Accordingly, the
researcher believes that any initiative to completely prohibit or ban HFT in the
market may be undesirable, since such actions may inadvertently eliminate valuable
market participants, especially during periods of substantial market uncertainty.
1.5 Structure of the Thesis

The remainder of this thesis is structured as follows. Chapter Two presents the first
essay of this thesis that surveys the literature related to high-frequency trading
(HFT), including its definitions, implications, and controversies. Chapter Three
7
discusses the second essay, which provides evidence on the effect of relative tick
size HFT activity using data from Australian equity markets. The third essay,
presented in Chapter Four, examines the impact of expected volatility on HFT
activity, and whether changes in HFT activity due to heightened volatility level
affected by liquidity. Chapter 5 concludes by summarising the key findings, as well
as the implications of each of the three essays, including suggestions for future
research directions and limitations of the study. The thesis closes with a section on
Bibliography and Appendix used in the thesis.
8
CHAPTER TWO:
ESSAY ONE
HIGH-FREQUENCY TRADING: DEFINITION,
IMPLICATIONS, AND CONTROVERSIES
2
This chapter provides a literature survey on high-frequency trading (HFT). The

chapter begins with an overview of HFT, including its definition and identification.
The chapter continues with a comprehensive review of literature on HFTs’ trading
mechanics; beneficial and harmful trading strategies; effects on market quality,
including price discovery and efficiency, liquidity, volatility, and adverse selection
costs; controversies surrounding HFT activity; market response to HFTs; and a
conclusion.
2.1 Introduction
Technological advancement has shaped the financial world. Prior to the invention of
the telegraph in 1844 and the telephone in 1876, communications in securities
markets had been primitive, using humans and carrier pigeons to transmit
information across markets (Markham, 2002). For nearly a century, the telegraph and
telephone were used as the main channel for financial communication: data were
received via telegraphic stock ticker, and orders transmitted via phone calls.
However, in recent years, fibre-optic cables and microwave towers have been
used as the medium to transfer trading information at light speed. A group of traders
exists, armed with complex algorithms that is willing to spend considerable amounts
of money to gain access to these state-of-the-art facilities. They pay to collocate their
servers close to or within stock exchanges, for example, to give them the speed
advantage they need for their trading strategies that rely on being the fastest. In
addition, they hire mathematicians and statisticians to work as quantitative analysts
(“quants”) to develop the various trading algorithms. This unique group of traders is
commonly referred to as high-frequency traders (HFTs).
9
In the United States, the market share of high-frequency trading (HFT) of
total equity trading peaked at around 60% in 2009, up from around 20% in 2005.
The percentage gradually decreased to approximately 50% in 2013 (Avramovic et
al., 2017) and has been relatively stable at this level since then, at 52% in 2018. In
Europe, HFT’s contribution to total equity trading was almost 0% back in 2005,
before peaking at around 40% in 2010, and settled at around 35% in 2014 (Kaya,
2016). The Australian Securities and Investments Commission (ASIC) states that
HFT accounted for approximately 27% of all equity market turnover in Standard &
Poor’s (S&P)/Australian Securities Exchange 200 securities from January 2012 until
March 2015. Despite its stable HFT market share, Australia has a relatively high
concentration of HFT-driven volume: the 10 largest HFTs accounted for 21% of all
trading turnover in 2015, compared to 17% 3 years earlier (ASIC, 2015).
This article is a survey of the literature on HFT, which covers various HFT
definitions. We discuss how HFT works and differentiates HFTs from other groups
of investors, followed by a discussion on beneficial HFT strategies (e.g., market
making, directional trading, and statistical arbitrage) and harmful HFT strategies
(e.g., front running, spoofing, and quote stuffing). The article continues with an
argument on the effects of HFT activities on market quality (e.g., price discovery,
liquidity, volatility, and adverse selection costs) and discusses several critical issues
associated with HFT (the flash crash of 2010, the arms race, and market-making
obligations). The article concludes with a discussion on responses to HFT.
2.1.1 Defining HFT

As the U.S. Securities and Exchange Commission (SEC, 2010, 2014) notes, HFT has
no clear definition. Regulators, researchers, and market participants therefore
describe HFT in different ways. The term HFT is typically associated with trading
that utilises computer technology, the use of technology to execute orders, electronic
trading, electronic markets, and automated trading. While these terms are closely
related to HFT, they are not the same thing and paint only an incomplete picture of
HFT.
10
The Netherlands Authority for the Financial Markets (NAFM) states that the
absence of a unanimous definition of HFT also makes classification difficult, which
leads to other problems such as inaccurate estimations of the HFT market share and
the inability to estimate the reach and influence of HFT in its markets. This lack of
consensus on the definition of HFT complicates research in this area and contributes
to the various conclusions on the effects of HFT activity in the market. The lack of a
precise definition of HFT also leads to confusion between HFT and other forms of
trading activity – such as algorithmic trading (AT) – with HFT consequently being
blamed for matters (e.g., the flash crash of 2010) that have little or nothing to do
with it (Moosa & Ramiah, 2015).
The incomplete definition of what HFT comprises is a problem for both

HFT-active and HFT-free markets alike. Financial authorities need to meticulously
analyse and consider the costs and benefits of HFT in their market. However, before
they can effectively tackle the issue, they need to have a sound definition of HFT.
An inaccurate definition would be too costly to the market: any microstructural
changes introduced will likely involve substantial costs and could affect all classes of
market participants, from individual investors to large mutual funds.
Different financial authorities describe HFT somewhat differently, but with

considerable overlap. The SEC (2010, p. 45) refers to HFT as “professional traders
acting in a proprietary capacity that engage in strategies that generate a large number
of trades on a daily basis.” The NAFM (2010) defines HFT as a form of automated
trading based on mathematical algorithms that implement certain short-term trading
strategies by utilising advanced technology, and not as a separate trading strategy
itself. ASIC (2010) characterises HFT as a specialised form of AT that employs a
high-speed, low-latency technology infrastructure. AT is viewed as a new form of
trading, or a class of trading strategy, that relies on computer algorithms to
automatically execute trading orders, using pre-programmed trading instructions that
determine when (i.e., the timing) and how (i.e., the price and quantity) a certain order
will be carried out and managed after its submission, with minimal or no human
intervention (Aitken et al., 2017, 2018; Aquilina et al., 2016; Brogaard et al., 2014;
Gider et al., 2019; Moriyasu et al., 2018). Unlike human traders, the lack of feelings
in AT ensures that trading judgment is not swayed or clouded by emotions, allowing
11
algorithmic traders to fully carry out their trading strategy (Harris, 2013; Manahov,
2016).
The London Stock Exchange Group (LSE), in response to a survey

conducted by the Committee of European Securities Regulators (CESR) in 2010 to
assess the impact of technology-driven developments, refers to HFT as a wide
variety of different strategies utilising ultra-fast technology to conduct electronic
market making and/or to seek arbitrage opportunities. The LSE (2010) also notes
that HFT is very fast and requires a low-latency connection to the trading systems of
exchanges. Among the regulatory body, the European Parliament and European
Council’s Directive 2014/65/EU – a 2014 amendment to the 2007 Markets in
Financial Instruments Directive (MiFID), commonly referred to as MiFID II –
provides a legal definition of HFT. Article 4(1) (40) of MiFID II describes an HFT
technique as “an algorithmic trading technique.”
According to the NAFM (2010), even with an accurate definition of HFT,

trading platforms would still be unable to properly distinguish HFT from other forms
of AT. To do so, they would need to establish the market shares of the various
trading strategies that employed AT, which, based on today’s technology, is not yet
possible. Although not providing a conclusive definition of HFT, certain
characteristics distinguishing HFT from other forms of trading can be specified.
The most agreed-upon characteristic of HFT is its use of high-speed,

sophisticated computer programs (SEC, CESR, MiFID II) or strategies (NAFM,
ASIC) to generate, route, and execute orders. HFT also uses sophisticated systems
and efficient infrastructures to minimise network latencies (SEC, NAFM, ASIC,
CESR, MiFID II) such as colocation services, individual or direct market data feeds,
and proximity hosting. Other characteristics relate to the trading positions of HTF
including extremely short average holding periods (SEC, NAFM, ASIC, CESR),
large volumes of orders with very small profit margins (SEC, NAFM, ASIC, CESR),
high amendment and cancellation rates or order-to-transaction ratios (SEC, NAFM,
ASIC, MiFID II), and trading days ending with flat positions (SEC, NAFM, ASIC,
CESR). Table 2.1 lists the detailed characteristics of HFT cited by each financial
authority.
12
Table 2.1: Summary of HFT characteristics from regulatory perspectives
SEC (2010)1  Use of extraordinarily high-speed, sophisticated computer
programs for generating, routing, and executing orders
 Use of colocation services and individual data feeds offered by
exchanges and others to minimise network and other types of
latencies
 Very short time frames for establishing and liquidating positions
 Submission of numerous orders that are cancelled shortly after
submission
 Ending the trading day in as close to a flat position as possible
NAFM (2010)  Use of a trading strategy that involves rapid calculation and
execution speeds
 Use of sophisticated systems and efficient infrastructures
 Use of an earnings model with very small profit margins in very
large volumes
 Usually market-neutral (nondirectional) and delta-neutral (hedged)
positions, thus, often closing out positions, with a flat position at
the end of the day
 Very short average holding periods, ranging from seconds to
several minutes
 Very high order-to-transaction ratios
ASIC (2010)  Generation of large numbers of small orders with a high

amendment and cancellation rates
 Hold positions typically with very short time horizons
 Use of a variety of trading strategies, the most common being the
provision of electronic liquidity
 Processing of direct market feeds to obtain the fastest access to
market information available
 Colocation of servers in data centres with the exchange market’s
matching engine to reduce access times
 Development of sophisticated trading strategies to trade on a short-
term basis
 Typically ending the trading day with no carryover positions that
use capital
1
The SEC has never suggested that all of these characteristics should be met for a firm to be
categorised as an HFT firm. A broader range of proprietary firms can, thus, be classified as high-
frequency traders (SEC, 2014).
13
CESR (2010)  Use of a form of automated trading that uses sophisticated
computers and information technology programs
 Trades executed in a matter of milliseconds
 Hold new equity positions possibly down to a “subsecond”
 Ending the trading day with a flat position
 Use of own capital and not acting on behalf of clients
 Use of trading strategies generally geared towards extracting very
small margins from hyperfast trading
MiFID II  Infrastructure intended to minimise network and other types of

(2014) latencies, including at least one of the following facilities for
algorithmic order entry: colocation, proximity hosting, or high-
speed direct electronic access
 System determines the order initiation, generation, routing, or
execution, without human intervention, of individual trades or
orders
 High intraday message rates constituting orders, quotes, or
cancellations
2.1.2 HFT identification

Despite the lack of a uniform definition of HFT, the fact that HFT has several unique
traits (see Table 2.1) allows it to be identified. One way to identifying HFT is with
the help of the exchanges and regulators themselves, since they have the information
on the identity behind each message submitted to the market. As Brogaard,
Hendershott, and Riordan (2014, p. 2270) note, “one of the difficulties in empirically
studying HFT is the availability of data identifying HFT. Markets and regulators are
the only sources of these and HFTs and other traders often oppose releasing
identifying data.”
An example of such a dataset is that used in the studies of Hagströmer and

Nordén (2013), Hagströmer et al. (2014), and Brogaard et al. (2015), who utilise
access to the market participant identifier (MPID) information associated with each
limit order in the NASDAQ OMX Stockholm Exchange (NOMX-St). With the help
of NOMX-St in-house expertise, Hagströmer and Nordén (2013) categorise
exchange member firms into (1) members who are primarily HFTs, that is, those that
engage in proprietary trading and utilise algorithms in their trading strategies; (2)
members who mainly trade for clients; and (3) members who engage in both
proprietary and agency activities. This categorisation results in 29 members
identified as HFT firms, 49 as non-HFT firms, and 22 as hybrid firms. Even though
14
the dataset allows for the direct identification of the messages sent by HFT firms, the
use of mediation trading services, such as sponsored access and/or the trading desks
of banks, which consist of a mixture of HFT and non-HFT, makes it difficult to
distinguish the origins of the trading activity and to interpret the results obtained
from this group (Hagströmer & Nordén, 2013).
Another example is the NASDAQ HFT dataset, which covers 120 U.S.
equities from 2008 to 2009, time stamped to the millisecond. The stocks in the
sample were chosen by Terrence Hendershott (chair of NASDAQ’s Economic
Advisory Board during that period) and Ryan Riordan (a prominent HFT researcher),
stratified by market capitalisation, and evenly split between NASDAQ and New
York Stock Exchange (NYSE) listings. NASDAQ’s access to order-level
information on its market to identify the firms submitting orders was used, and 26 of
the firms were manually classified as HFT firms. The dataset also categorises orders
by whether the execution was aggressive (liquidity taking) or passive (liquidity
providing), further grouping them into either HFTs or non-HFTs, resulting in four
types of order execution: HH, HFTs that take liquidity from other HFTs; HN, HFTs
that take liquidity from non-HFTs; NH, non-HFTs that take liquidity from HFTs;
and NN, non-HFTs that take liquidity from other non-HFTs.
Even so, the dataset has its limitations. NASDAQ cannot identify all HFTs in
the market and has possibly excluded HFT firms that also act as brokers while
engaging in proprietary lower-frequency trading strategies. Thus, the orders from
HFT firms routed through those large integrated firms could be excluded as well
(Brogaard et al., 2014). Even with these limitations, both the NOMX-St and
NASDAQ HFT datasets are still considered to be of high quality, since they allow
researchers to accurately monitor HFT behaviour and estimate its impact on the
market. The quality of datasets is clearly an ongoing issue in identifying and
examining HFT activity, with nanosecond stock market data time stamping now
becoming more common. Hasbrouck (2019) shows that 100- and 10-nanoseconds
granularity dominates the one-second granularity of stock market data feeds in terms
of the information obtained.
15
In the absence of internal information, researchers have to rely on HFT
characteristics and formulate appropriate proxies to gauge HFT activity. One clear
advantage of this approach is its replicability. The less binding constraint (i.e., the
need for internal experts) allows the proxy to be applied in other markets and for
longer periods. HFT proxies are constructed from message data that are commonly
available for many markets. One widely used HFT measure is the order-to-trade ratio
(OTR), which represents quoting intensity (e.g., Aquilina & Ysusi, 2016; ASIC,
2013, 2015; Brogaard et al., 2015; Friederich & Payne, 2015; Frino et al., 2015;
Hagströmer & Nordén, 2013). It is defined as the number of limit order submissions
and cancellations on a given day, divided by the number of executions. HFTs
typically place large numbers of orders across various price levels and revise their
orders with the arrival of new information in the market, resulting in large OTRs in
stocks with high HFT activity.
However, Yao and Ye (2018) argue that the OTR is a poor cross-sectional
proxy for HFT activity. Their findings show that stocks with a greater percentage of
liquidity provided by HFT have a lower OTR, which directly contradicts the
aforementioned wisdom. Yao and Ye (2018) find that HFTs provide more liquidity
in stocks with larger tick sizes and have less motivation to cancel orders once these
have reached the top of their queue positions. On the other hand, HFTs not only
provide less liquidity in stocks with smaller tick sizes, but also tend to cancel orders
more frequently, since price competition occurs on a finer grid. Based on these
arguments, the conventional view of the OTR can be challenged in terms of its
ability to be a proxy for HFT. O’Hara et al. (2019) also find similar results using
NYSE data, suggesting that traders’ order strategies differ between tick-constrained
and tick-unconstrained environments.
Hasbrouck and Saar (2013, p. 646) define low-latency trading as “strategies

that respond to market events in the millisecond’s environment.” From this
definition, they introduce an alternative measure for HFT activity, RunsInProcess,
which is based on HFT behaviour that utilises dynamic and strategic order placement
strategies. The underlying component for this measure is the number of strategic
runs, which are essentially links of rapid order submissions, cancellations, and
executions presumed to be submitted by HFT algorithms. Orders are linked if (1) the
16
cancelled order and the new order are in the same direction, (2) the orders have
similar size, and (3) once an order is cancelled, it is quickly followed by another
order within 100 milliseconds. The shortest run would have four messages: order
submission (start), cancellation, second entry, and termination (end). The longest run
could last the whole day, since orders are rapidly submitted and cancelled up until
the market closes. Any runs shorter than 10 messages are eliminated, to reduce the
possible inclusion of agency algorithms. Nonetheless, if any HFT arbitrage strategies
utilise only marketable orders, this measure could exclude them (Hasbrouck & Saar,
2013).
Other proxies of traders’ HFT activities are possible when the market
provides message- or order-level data. An example is the data from the Investment
Industry Regulatory Organization of Canada, which contain every message
submitted to recognised Canadian trading venues. The detailed records of trades and
orders, as well as constant identifiers, permit researchers to identify HFT, based on a
number of factors, for example, short order-to-cancel times (Comerton-Forde et al.,
2018; Korajczyk & Murphy, 2019), corresponding traders that submit large numbers
of orders immediately following net order imbalance information (Comerton-Forde
et al., 2018; Korajczyk & Murphy, 2019), and fast reaction times to other orders
(Malinova & Park, 2020), along with other proxies such as the number of trades,
OTR ratios (Boehmer et al., 2018), trading volumes, and the end-of-day inventory
(Brogaard et al., 2019).
2.2 Trading Mechanics

HFT is a subset of AT and, thus, inherently has most (if not all) of the characteristics
of its superset. What makes HFTs a unique class of investor within AT lies in their
speed in observing potentially profitable trading opportunities, quickly processing
new information and executing the appropriate actions, and analysing the textual
context of news flow at much higher frequencies and in much shorter periods than a
human is capable of. This is an important advantage that these machines (i.e., HFTs)
have over humans (Menkveld, 2014). Their infrastructural and technological
advantages allow for the optimisation of a wide array of complex trading strategies
that can be both beneficial and detrimental to the market. According to Angel
(2014), trading speeds nowadays are so fast that they have almost reached the
17
theoretical speed of light, approximately 300 000 kilometres per second. This level
of speed limits certain trading strategies exclusively to HFTs, since other types of
market participants are unable to replicate such strategies (Harris, 2013). HFTs
constantly strive for superior calculation and execution speeds (NAFM, 2010).
There is nothing new about the way HFT works. The short-term trading
strategies employed by HFTs have long existed (NAFM, 2010). The way HFTs
profit from the market is generally similar to other traders’ strategies. For instance,
they buy stocks at a lower price and then try selling them at a higher price. Stocks
with a short selling option allow for more choices: traders can make money from
either bearish or bullish stocks. For stocks with options, traders can make a profit
from price disparities between the parent stocks and their option securities. Traders
can also assume the role of market maker, in which case they stand ready to buy and
sell securities and profit from the market-making spread. Moreover, market-making
HFTs can receive rebates from certain trading venues for providing liquidity to their
market.
Hasbrouck and Saar (2013) find the speed of some algorithms to be such that
the time it takes to complete a trading cycle starting from the detection of a market
event, its analysis, and sending an order can be as short as 2 to 3 milliseconds. This
intense activity within the millisecond environment is also where computer
algorithms react to each other. O’Hara (2015) states that order latencies are now
measured in milliseconds (thousandth of a second), microseconds (one-millionth of a
second), and even nanoseconds (one-billionth of a second). For comparison
purposes, it takes the human eye 400-500 milliseconds to respond to visual stimuli,
and human reaction times are generally thought to be around 200 milliseconds,
which, in both cases, is far slower than HFT speeds (Kosinski, 2013; O’Hara, 2015).
At such speeds, human traders cannot accurately follow the low-latency activity on
their trading screens or the market dynamics that can be driven by the interactions
between algorithms (Chordia et al., 2013; Hasbrouck & Saar, 2013).
To accommodate HFTs’ need for speed, fast, reliable networks, and

infrastructures are needed. In 2010, Spread Networks began the construction of an
826-mile (about 1,330 kilometres) straight-line fibre-optic cable connecting New
18
York and Chicago, estimated to have cost approximately US$300 million (Steiner,
2010). The cable has been able to reduce the round-trip communication between the
two cities from 16 to 13 milliseconds. Although a 3-ms difference seems trivial to
HFTs, it seems like an eternity (Budish et al., 2015). Subscribers were charged
almost US$14 million for a 5-year contract, or roughly US$ 230,000 per month
(Serbera & Paumard, 2016). Even though light is able to travel quickly through
glass, it is faster through air. Therefore, in 2011, microwave towers were built to
reduce the latency even further. Shkilko and Sokolov (2020) report that the
microwave networks are approximately 30% faster than the Spread Networks cable.
Albeit faster, these networks are easily disrupted by precipitation such as rain and
snow. Due to the high costs, HFTs need to be able to make even greater profits to be
successful in the millisecond environment.
The earnings model for HFT consists of executing many transactions with
very small profit margins in very large volumes (NAFM, 2010). Using fully
automated trading strategies, HFTs attempt to identify and profit from short-term
irregularities, and earn small amounts of money from every trade. Even though the
profit per trade is often as small as only a few basis points, it is amplified by the high
trading volume (Zhang, 2010). The ability to trade at low latencies allows HFTs to
profit from the trading environment itself, rather than from investing in financial
securities (Hasbrouck & Saar, 2013). This strategy explains two of the common
characteristics associated with HFT, which are frequent order cancellations shortly
after their placement and high ratios of messages to completed trades (Budish et al.,
2015).
In 2005, the Regulation National Market System (Reg NMS) was adopted to
foster intermarket competition and to serve the interests of long-term investors and
listed companies (SEC, 2005). At the centre of Reg NMS lies the Order Protection
Rule (Rule 611), which prevents exchanges and brokers from executing orders at
prices that are below the National Best Bid and Offer (NBBO). This created
opportunities for alternative trading systems to compete for liquidity with the major
exchanges, since orders are routed to the venues offering the best prices. Prior to Reg
NMS, most shares were traded at the two major stock exchanges, NYSE and
NASDAQ. However, within 5 years, NYSE’s market share of trading of its listed
19
stocks fell from 80% in 2005 to 25% in 2010 (Menkveld, 2013). This fragmentation
reflects how market shares are being captured by “third markets,” traded away from
their listing venue (Pirrong, 2014). The trading venues are also geographically
diverse; therefore, orders’ arrival times at different exchanges vary.
HFT strategies rely heavily on speed, and fragmentation is viewed as a

blessing to HFTs, since it increases the effectiveness of their strategies (Baldauf &
Mollner, 2020). Latency issues, such as the speed of cross-market information flow
and transmission speeds across geographical locations, are crucial to the success of
HFTs. To utilise their trading strategies, many HFTs have multiple locations across
several cities, such as New York, Chicago, and London, to gain a competitive edge
against other HFTs. Brogaard and Garriot (2019) find that HFT competition
promotes liquidity. HFTs’ entrance to the Canadian Alpha exchange (an alternative
trading system) has been found to reduce the bid–ask spreads on the exchange, and
the effect is most prominent upon first entry. As the number of HFTs on Alpha
grows, the spreads tighten, eventually converging to the levels reported on the
Toronto Stock Exchange (TSX). Brogaard and Garriot also find that HFTs’ departure
from the market is associated with liquidity deterioration. On the contrary, using
NOMX-St data, Breckenfelder (2019) finds that liquidity deteriorates following the
market entry of HFTs and improves after their exit, due to competition among
speculative HFTs.
2.3 Trading Strategies

Benos and Sagade (2016) state that HFTs exhibit variability in their trading
strategies because of differences documented in their liquidity provision, end-of-day
and maximum intraday positions, trading revenues, and the like. The variability in
strategies also translates into different sensitivities of HFT position changes due to
inventory levels and recent price changes. Brogaard et al. (2014) find that the
direction of HFT is correlated with publicly available information such as
macroeconomic news announcements and limit-order book imbalance. They also
find that HFTs follow contrarian trading strategies, as evidenced by the negative
correlation between HFTs’ overall trading and past returns. Goldstein et al. (2014)
state that, naturally, HFT strategies are employed by proprietary firms, the majority
of which are either broker-dealer proprietary trading desks (trading units within a
20
firm that trades using its own money), hedge funds, or proprietary trading groups.
This result is logical, due to the high costs involved in employing sophisticated
technology and obtaining the big data to execute HFT strategies (Kauffman et al.,
2015; Moosa & Ramiah, 2015).
Aldridge (2013) generally categorises HFT strategies into three groups: (1)
statistical arbitrage, also known as value-motivated strategies; (2) directional
strategies, also known as informed trading; and (3) market making, also known as
liquidity trading. The algorithms employed by HFTs can determine their order
execution style such as being either aggressive or passive or either sending orders in
one trade or splitting them up into smaller trades. An aggressive order is an order
that is placed at the current market price, or a limit order with a price near the current
market price. A passive order is a limit price placed that is far from the current
market price. Similarly, the NAFM (2010) divides HFT strategies into market
making, statistical arbitrage, and low latency. While the first two groups are similar
to Aldridge’s (2013), the third group classification, low latency, has a broader scope.
The NAFM (2010) states that the success of the latter group is determined by the
sheer speed of the users, hence, creating the need for the fastest systems and the best
connections to trading venues. Harris (2013), on the other hand, categorises HFT
strategies into three groups, based on their effect on the market. The first group,
valuable HFT strategies, is a group of trading strategies that are acceptable to the
market in general such as market making and statistical arbitrage. The other two
groups, namely, harmful and very harmful HFT strategies, are intolerable groups of
trading strategies. The strategies belonging to these groups benefit HFTs at the cost
of other market participants (see Section 2.3.2).
Most HFT-based strategies, such as market making, promote market

liquidity, while arbitrage strategies make positive contributions to price discovery
and market efficiency. Therefore, any action to prevent or hamper these strategies
through regulation or the imposition of specific constraints can have
counterproductive effects on market quality. Regardless, regulatory bodies should
always combat any predatory strategies that jeopardise market integrity or create
disruptive or confusing effects for other market participants (Gomber et al., 2011).
21
Harris (2013) highlights that financial authorities should be meticulous in regulating
the market, to avoid unintentionally harming beneficial HFT strategies.
Putniņš and Barbara (2020) find heterogeneity in the effects of AT/HFT trading
strategies on institutional investors’ trading costs, which can be grouped into harmful
and beneficial strategies, based on their systematic effects. The authors’ findings
show that toxic AT/HFT strategies roughly double the cost of executing large parent
orders, while the positive effects derived from beneficial trading strategies nullify
this problem. However, in the aggregate, AT/HFT has little or no effect on
institutional trading costs. This aggregation has masked the heterogeneity in
AT/HFT strategies and hidden their actual impact on institutional trading costs. The
authors’ findings also suggest that market structure changes aimed at influencing
AT/HFT can produce positive or negative outcomes, based on whether the changes
are able to disproportionately encourage harmful or beneficial traders.
Cooper et al. (2016) examine regulatory efforts related to HFT, particularly

on the issue of HFTs’ deception in the market. They conclude that any action to treat
a deception, even an intentional deception, as misconduct in a financial market could
be a mistake. They outline three acceptable criteria for AT strategies: (1) the trading
strategy should be prudent, that is, it should not be harmful to the market should
HFTs behave unexpectedly; (2) the trading strategy should not block price discovery,
that is, it should not interfere with the ability of other market participants to reflect
their private information on the price; and (3) the trading strategy should not
circumvent transparent price discovery, and strategies that conceal information from
being discovered, such as using dark pools or hidden orders, should therefore be
prohibited.
2.3.1 Beneficial Strategies

The following sections discuss acceptable HFT strategies, namely, statistical
arbitrage, directional trading strategies, and market making. These strategies are
deemed acceptable, since they do not harm the market and have positive effects on
market quality.
22
2.3.1.1 Statistical Arbitrage
Statistical arbitrage (also known as “stat arb”) is a trading strategy that is based on
the theory that two similar instruments should share similar behaviour, and,
therefore, any short-term divergence between their relative prices is likely to be
corrected. This trading strategy is also commonly known as pairs trading and can use
statistical approaches that measure the relation between two or more instruments
such as cointegration or correlation analysis. The temporary divergence is more
likely to be driven by momentary order imbalances in the market than by any
meaningful fundamental change (Narang, 2013). This trading strategy is designed to
make a profit from price disparities and temporary deviations in statistically
significant relations. HFTs can consider tens or hundreds of stocks to utilise this
strategy (Golub et al., 2013; Lhabitant & Gregoriou, 2015; Moosa & Ramiah, 2015).
Accordingly, HFTs will hunt for opportunities that arise during periods of temporary
deviation and exploit them before they disappear (Moosa & Ramiah, 2015).
Foucault et al. (2017) argue that the effect from arbitrage activities could go
either way: it can be beneficial or harmful, depending on the underlying cause of
arbitrage opportunities. In the event opportunities are due to transitory trade
imbalances, arbitrageurs (i.e., HFTs) can exploit the situation by providing liquidity
to the market. This action is deemed beneficial.2 Consistent with this view, Gromb
and Vayanos (2002) find that arbitrageur activity supplies liquidity to the market and
brings prices closer to their fundamental value. On the contrary, if a temporary
arbitrage opportunity arises due to lagged and asynchronous price adjustments to
information arrival, then the arbitrageurs’ profits in this case are the adverse
selection costs that market makers must pay due to trading at stale quotes. Foucault
et al. (2017) find that the market becomes more illiquid on days when the percentage
of so-called toxic arbitrage opportunities is higher and arbitrageurs’ speed is
relatively faster. This finding supports the notion that the higher price efficiency
from rapid arbitraging activities comes at the expense of market makers’ increased
adverse selection risk.
2
For instance, trade imbalances can arise when impatient traders flood the market by submitting large
volumes of bid or ask orders, leading prices to deviate from their fundamental values (Brogaard et al.,
2018).
23
Wissner-Gross and Freer (2010) highlight the importance of minimising
information transmission delays in modern-day securities trading. In their article on
relativistic statistical arbitrage, they demonstrate that optimal intermediate locations
exist between trading centres that host cointegrated securities, which minimises
transmission delays and maximises profit potential. As traders continue striving to be
the fastest, the importance of having optimal locations is even more pronounced
(Donefer, 2010; Wissner-Gross & Freer, 2010). Regardless, Kozhan and Tham
(2012) argue that, while competition is commonly associated with improved price
discovery, competition among arbitrageurs can inflict negative externalities on each
other due to the crowding effect, which, in turn, will limit efficiency.
Opportunities for statistical arbitrage can surge due to long-term investors’

strategic decisions. For instance, their buying or selling certain securities can create a
price impact on the securities’ price, which consequently creates spillover across the
markets, especially in correlated securities (Goldstein et al., 2014). The fastest trader
that first notices such opportunities and trades on them will make the most, if not
take all, of the profits from the situation. Therefore, speed is essential to the
successful execution of this trading strategy, and HFTs that implement it are willing
to spend a great deal to keep their technological capabilities up-to-date (Chung &
Lee, 2016; MacKenzie, 2015). This strategy plays a key role in the market in terms
of liquidity provision, as well as in price discovery and the information
dissemination process (Goldstein et al., 2014). Nonetheless, Hasbrouck and Saar
(2013) argue that, even though HFT helps in eliminating momentary price
distortions, given that the improvement is only within the millisecond environment,
the effect is deemed insubstantial.
2.3.1.2 Directional Trading

Directional trading strategies are a group of HFT strategies based on the
identification of short-term trends or momentum, which includes event-driven
strategies and short-term price movement prediction strategies. Directional strategies
are time sensitive (Aldridge, 2013), since they need to anticipate intraday price
movements, which involve taking unhedged positions based on forecasted price
changes such as in exploiting the divergence between fundamental values and actual
market prices. Benos and Sagade (2016) find that HFTs with neutral liquidity-
24
taking/-making behaviour are trend chasers. They trade in the direction of short-term
price changes, that is, they buy when the price is rising and sell when the price is
falling, which is suggestive of momentum strategies.
Directional strategies are based on the theory that price movements have
predictable directions which could be due to a trend (momentum strategies) or the
reversal of a trend (mean reversion strategies). Under the momentum strategy, HFTs
will identify a trend or a significant movement and will bet that the price will
continue to move in the same direction, driven by the idea of a growing consensus
among market participants (Narang, 2013). The mean reversion strategy, on the other
hand, is built on the notion that any deviation in price, such as a trend or a consistent
direction, could be temporary. Thus, price movements do not persistently move in
one direction and will eventually revert and bounce back (Easley et al., 2012).
To be successful at implementing directional trading strategies, HFTs need to

have superior access to information (e.g., information from paid-for news sources
such as Bloomberg) and be able to immediately assess and analyse market
conditions. Foucault et al. (2016) suggest that the contribution of news trading to
directional traders’ profit increases with news informativeness, and the fastest traders
will gain the most profit. Furthermore, the competitive edge that directional traders
have from early access to new information will not last long, since that information
will soon be available to the public. Thus, directional traders are normally
aggressive, since they use market orders or post limit prices close to the market
(Aldridge, 2013).
2.3.1.3 Market Making

Market making can generally be described as the placement of limit orders on both
sides of the market price – that is, to set limit orders to buy (sell) at a price slightly
below (above) the market price. By continuously supplying the market with standing
limit orders, market-making activities essentially provide liquidity to the market. The
spreads between the limit orders to buy and to sell constitute the market makers’
profit.
25
HFTs’ market-making strategies help the market to be more efficient and
have a stabilising effect on the market, since the HFTs provide buying power when
others want to sell, and selling power when others want to buy (Angel, 2014).
Despite the fact that the financial landscape has developed so with the evolution of
technology, the general mechanics of market making still hold, even in a high-
frequency world. Goldstein et al. (2014) state that market-making HFTs use
automated liquidity provision, a strategy that rapidly places, cancels, and replaces
bid and ask limit orders and profits from the resulting spreads. The high-frequency
update process involved in the market-making process results in enormous order
volumes and a high cancellation rate of 90% or more (SEC, 2010).
Aldridge (2013) states that market makers are exposed to two types of risk
once their market limit orders are placed: inventory risk and adverse selection risk.
Inventory risk is the risk that the inventory held by a market maker will decline in
value due to natural market movements, while adverse selection risk is the
possibility of the market maker trading against a party that is better informed about
the true price of the stock. Thus, the market maker requires compensation, not only
for providing liquidity, but also for the risks mentioned above (Aldridge, 2013;
Golub et al., 2013).
Some electronic exchanges use the maker-taker pricing model to price their
order-matching service (Harris, 2015). Durbin (2010, p. 206) defines the model as “a
pricing policy of some exchanges where active traders pay a fee, some of which is
distributed to the associated passive trader.” The maker-taker pricing model is used
to encourage market making, instead of market-taking activity, through incentives in
the form of rebates or reduced transaction costs for market makers. The rebate is
indeed important for market-making HFTs. The absence of a rebate can put HFTs in
a loss position (Hendershott & Riordan, 2013), and their revenue from supplying
liquidity would then be negative (Brogaard et al., 2014), which, in turn, would
discourage HFTs’ liquidity provision activities.
Unlike HFT that uses directional trading strategies, HFT market makers do
not seek to make a directional bet, but, instead, maximise their inventory turnover by
taking positions on both sides of the order book. HFT market makers typically turn
26
over their inventory more than five times in a day, which explains their high share of
volume traded in the market. To protect their investment, they will either hold a
minimum or even zero inventory position at the end of a trading day.
By having very small inventories and short holding periods, HFTs can
conduct their market-making activities with very little capital, while using high-
speed trading to control their position risk (Easley et al., 2011). Benos and Sagade
(2016) categorise HFTs based on their liquidity-taking/liquidity-making behaviour
and find evidence that passive HFT activity is consistent with market-making
activity, in which HFTs trade in the opposite direction (i.e., contrarian trading) of the
most recent price changes, post limit orders, and use aggressive trades to make quick
inventory adjustments. Regardless, the authors also find that passive HFTs have a
high information-to-volume ratio, suggesting that the HFTs could be using various
market-making strategies, rather than solely aggressive orders, to make the market.
Since HFT has become increasingly dominant in overall market activity and
given its market-making role, it is interesting to examine the impact of this activity
on the rewards to market making. Conrad and Wahal (2020) examine the price
impact of common stock trades in the United States from 2010 to 2017. Their major
finding is that market-maker profitability is very sensitive to speed, with profits
declining over the sample period. For the shortest time horizon examined in their
article, 100 milliseconds, the authors show that aggregate profits declined over the
period, from 1.9 basis points of the total dollar volume to 1.0 basis point. Although
this can be seen as a benefit to social welfare, concerns could be raised if HFTs were
to achieve dominance in this area.
2.3.2 Harmful Strategies

Controversial strategies are those that profit at the expense of others through “dirty”
means such as front running, order anticipation, quote matching, spoofing, layering,
and quote stuffing. Moreover, HFTs’ ability to rapidly enter and cancel orders faster
than other traders makes it difficult to identify where liquidity exists across
fragmented markets. This uncertainty creates even more opportunities that are
profitable for HFTs at the cost of other traders (O’Hara, 2015).
27
2.3.2.1 Front running, order anticipation, and quote matching
Harris (2013, 2015) describes front-running strategies as very harmful trading
strategies and further categorises them into order-anticipating and quote-matching
strategies. Order anticipation works by examining trades and quotes to detect
algorithms used by traders that intend to move large orders. For example, traders
could split their large orders into smaller packages to conceal their private
information and reduce the impact on the market. HFTs would then trade ahead of
(i.e., front-run) the incoming large orders and profit from the anticipated direction of
the price changes. This will increase (decrease) the price for incoming large buy
(sell) orders, increasing the transaction costs for traders intending to execute large
orders. HFTs that apply an order anticipation strategy design their algorithms to play
by the book, without violation of duty, misappropriation of information, or other
misconduct (SEC, 2010). Regardless, the strategy that they use is parasitic: not only
does it not contribute to price discovery or liquidity, but also it preys on other traders
and jeopardises large traders the most (Harris, 2015). Some institutional investors
even claim that an order anticipation strategy can adversely affect their trading
strategy, increasing their costs (Agarwal, 2012).
Quote matching, on the other hand, profits by posting slightly better limit
orders, for example, one tick higher (lower), than slow traders’ limit buy (sell)
orders, which gives them price priority. In the case in which the market is moving
against their position, quote-matching HFTs would trade with the slower traders’
quotes (which have become the best quotes) to minimise their losses. The problem of
quote matching is not new to large buy-side traders. It was an important source of
profit for exchange specialists before the HFT era; the main difference today is the
identity of the quote matchers (Harris, 2013). Unlike the order anticipation strategy,
which requires high-quality pattern recognition algorithms, the success of a quote-
matching strategy depends very much on HFTs’ low-latency communication. Speed
is crucial to quote matchers, to ensure that their orders are the first to be filled and to
revise their unexecuted orders, should large orders be cancelled or filled by other
parties. Therefore, the order anticipation strategy is dominated by the faster HFTs
(Harris, 2013). Nevertheless, both strategies unnecessarily increase large traders’
transaction costs (Chung & Lee, 2016) and could impede the process of impounding
fundamental information into the price (Jarnecic & Snape, 2014).
28
Aquilina and Ysusi (2016) empirically examine HFTs’ order anticipation
activity using data from the LSE and find no evidence of HFTs systematically
anticipating orders sent to different venues by non-HFTs and trying to front-run the
orders. However, when analysing longer periods, the authors do find trading patterns
consistent with HFTs anticipating non-HFTs’ order flow. Regardless, the result can
also mean that HFTs are able to react more quickly to news and other public
information than non-HFTs can. Aquilina and Ysusi (2016, p. 26) conclude that
HFTs “appear not to anticipate near-simultaneous orders … but they could be
predicting the flow over longer periods.”
Van Kervel and Menkveld (2019) document the behaviour of HFTs around
large institutional orders executed through a series of child trades on the NOMX-St.
From three theoretical standpoints, HFTs can adopt the role of market makers that
provide liquidity to the market (Grossman & Miller, 1988) and/or engage in
predatory trading activity by demanding liquidity and front-running institutional
orders (Brunnermeier & Pedersen, 2005) and/or undertake back-run positions, that
is, delaying their entry into the market so that privately informed orders can be
sniffed out (Yang & Zhu, 2017). Their findings show that HFTs initially trade
against the direction of an institutional order (i.e., acting as a liquidity supplier)
before eventually trading in the same direction, to strategically ride the institutional
order flow, making it costlier for the latter to impound their private information into
prices. The authors also find that HFTs need considerable time (several hours) to
learn about the institutional order, which is consistent with the back-run position.
From one angle, HFTs’ participation could make pricing more efficient in the short
run, through faster price discovery. Regardless, in the long run, Van Kervel and
Menkveld (2019) argue that institutional investors could have less motivation to
continue costly analyst research to seek informational rents, since the projected
returns from doing so is diminished by the participation of HFTs, potentially making
prices less efficient in the long run.
2.3.2.2 Spoofing and layering

Spoofing and layering constitute a strategy defined as “submitting multiple orders at
different prices on one side of the order book slightly away from the touch,
submitting an order to the other side of the order book (which reflects the true
29
intention to trade) and, following the execution of the latter, rapidly removing the
multiple initial orders from the book” (European Securities & Markets Authority,
2011, p. 27).
The Financial Industry Regulatory Authority (FINRA, 2012, para. 5)

generally describes spoofing as a form of market manipulation intended for
“triggering another market participant(s) to join or improve the NBBO, followed by
cancelling the nonbonafide order and entering an order on the opposite side of the
market.” Section 747, Antidisruptive Practices Authority, of the Dodd-Frank Wall
Street Reform and Consumer Protection Act (2010, p. 124, Stat. 1739), however,
outlines a broader definition of spoofing, defining it as a disruptive practice that
involves “bidding or offering with the intent to cancel the bid or offer before
execution,” which makes it unlawful to practice such a strategy.
Spoofing is executed with the intention of attracting liquidity using fake

market or limit orders by forming the illusion that the market is moving soon due to
greater demand in the order book. This strategy can be accomplished using
algorithms designed to analyze liquidity information in the order book to predict
incoming market activity and deliberately use the information to mislead other
investors or algorithms (Serbera & Paumard, 2016). For example, HFTs can create
such a situation by posting large displayed limit orders just below the best bid price,
giving others the impression that the price will soon move upward. This situation
encourages other traders to quickly buy the stock, by quoting the stock at a higher
bid price, or even to execute market orders. In the meantime, HFTs might already
own the stock beforehand and can now sell at a higher price in a bigger volume,
thanks to the artificially inflated price that was driven by the fake limit buy orders.
Layering is a form of spoofing that involves the placement of a large number

of fake orders at several different price limits on one side of the order book (NAFM,
2010). This creates the appearance of changing levels of supply and demand in the
affected securities (FINRA, 2012). Others could erroneously interpret this pattern as
a signal of increasing directional pressure on the price and act accordingly. HFTs
will then profit from the price move they have initiated and cancel the fake limit
orders before they are executed. Both spoofing and layering convey the impression
30
that a security is more liquid than it actually is, or even suggest that the security is
currently under- or overpriced (Harris, 2015). Cooper et al. (2016), however, claim
that spoofing and layering are just other forms of bluffing, and, just as in poker,
bluffing should be allowed. The authors conclude that the regulators should not treat
all deception in the financial market as misconduct, and they propose a set of criteria
for deciding which trading strategies should be regulated and which should not. They
argue that an acceptable trading strategy (1) should be prudent, (2) should not block
price discovery, and (3) should not circumvent transparent price discovery. Since
spoofing works within the outlined criteria, it should not be prohibited.
Aspris et al. (2015) examine the effect of switching the matching algorithms
adopted by the London International Financial Futures Exchange in 2007 for order
submission strategies. The event involved the introduction of a “time pro rata”
matching algorithm to replace the older “pure pro rata” mechanism. The mechanics
of the former matching algorithm are found to incentivise traders to flood the order
book with a much larger quantity than they truly intend to execute, a situation
comparable to the spoofing and layering problems. Following the microstructure
change, Aspris et al. (2015) find significant changes in market participant behaviour,
evidenced by the sizeable decline in market depth, as well as a substantial increase in
the numbers of small order entries and cancellations. The new matching mechanism
is found to eliminate “false liquidity” from the market, thus, giving the participants a
more reliable representation of the actual liquidity level in the market. Indirectly, the
findings suggest that regulators indeed have ultimate power to ensure a level playing
field that is fair for all; however, they should be meticulous in doing so, to avoid
unintended adverse effects on the market.
2.3.2.3 Quote stuffing

Quote stuffing is another form of market manipulation strategy that utilises HFTs’
ability to rapidly send and cancel orders. Easley et al. (2012, p. 228) describe quote
stuffing as a strategy that “involves sending and cancelling massive numbers of
orders with the intent of taking all available bandwidth and thereby preventing other
traders from being able to submit orders.” Similarly, the U.K. Government Office for
Science (2012, p. 168) defines quote stuffing as “entering large numbers of orders
and/or cancellations/updates to orders so as to create uncertainty for other
31
participants, slowing down their process and to camouflage the manipulator’s own
strategy.” The high rate of entered and cancelled orders involved in quote stuffing is
viewed as a way to manipulate markets and to lure other traders into making
mistakes (Narang, 2013).
Unlike spoofing and layering, which use limit orders near the best bid and
ask prices, quote stuffing involves placing large amounts of nonexecutable orders,
that is, limit orders that are far from the best quote, aimed to congest the market and
slow down competitors (Lhabitant & Gregoriou, 2015). An exchange’s network
bandwidth can become congested from receiving unusually large numbers of trade
messages (e.g., rapid orders and cancellations), thus, impairing other traders’ access
to the market (Angel & McCabe, 2013). The impairment leaves slower traders with
an unclear picture of the actual market situation and affects their ability to execute
trades. Faster traders, on the other hand, are able to gain a better understanding of
what is happening in the market, allowing them to profit at the expense of slower
traders (Biais & Woolley, 2011). Since a quote-stuffing strategy seeks to make a
profit by preventing others from adding their private information to the market, it
should be prohibited, based on the acceptable trading strategy criteria outlined by
Cooper et al. (2016).
Nonetheless, as Egginton et al. (2016) note, sudden outbursts or episodic

spikes in quoting activity could be the result of interactions between two or more
algorithms that fail to converge. In this case, quote stuffing can be viewed as a side-
effect of using robots in trading, rather than being part of a scheme to strategically
manipulate the market. Either way, quote stuffing is still linked to market
degradation. Egginton et al. (2016) analyse intense episodic spikes in quoting
activity in the U.S. stock markets.3 Their findings suggest that, during quote-stuffing
events, affected stocks experience liquidity deterioration, higher trading costs, and
greater short-term volatility. In addition, orders are more rapidly entered and
cancelled, have shorter durations, have lower execution rates, and are smaller.
3
Egginton et al. (2016, p. 588) define intense quoting episodes as “segments where the level of
quoting activity exceeds the previous 20-day mean number of quotes per minute by at least 20
standard deviations. [They] also require that the average number of quotes for the entire trading day
not exceed its previous 20-day rolling average by more than two standard deviations. The latter
requirement is implemented to exclude trading days with an unusually high level of quoting activity.”
32
Overall, the authors’ findings are aligned with the notion that intense quoting periods
are associated with the degradation of market quality.
Tick size reduction seems to be the simplest way to mitigate the order
congestion problem. Theoretically, a finer tick size allows orders to be placed at
more price points, which will give market participants greater freedom in deciding
their entry/exit price, and essentially prevents quote stuffing Nonetheless, Dyhrberg
et al. (2020) argue that a tick size that is too small would promote aggressive price
undercutting by an economically nonsignificant amount. In a purely order-driven
market, endogenous liquidity providers (ELPs) assume the role of market makers,
since it is profitable to do so. A tick size that is too fine could disincentivise them
from continuing to make the market, since this undercutting behaviour will expose
them to the risk of losing execution priority and impede their ability to rapidly
offload potential inventory. The absence of ELPs can significantly reduce liquidity
supply in the market, which is detrimental to market quality (see Section 2.5.3 for a
discussion on market-making obligations).
2.4 Effects on Market Quality

A large and growing body of literature has investigated the effect of HFT on market
quality. The term market quality in itself is broadly defined, but commonly
associated with price discovery and efficiency, liquidity, and volatility (e.g., Harris,
2003; U.K. Government Office for Science, 2012). Based on HFT characteristics,
HFT can be thought of as a new breed of intermediary that can improve or harm the
market. Whether HFT is beneficial or detrimental to the market is still a hot topic,
among market participants, regulators, and the media, as well as academics
(Menkveld, 2014). The many perspectives on HFT could have stemmed from the
lack of consensus on its mechanics, where HFTs can act as market makers,
arbitrageurs, predators, or some combination thereof (Carrion, 2013).
2.4.1 Price Discovery and Efficiency

Fama’s (1970, p. 383) seminal article states that “a market in which prices always
fully reflect available information is called efficient.” Thus, in an efficient market,
once information is publicly disclosed, it is quickly reflected in prices (Fox et al.,
2017), and any mispricing and associated arbitrage opportunities should be rapidly
33
eliminated (Goodhart & O’Hara, 1997). Nonetheless, in the era of HFT, the terms
immediately, rapidly, and current need to be refined, since the HFT definition and
perception of these terms are very different from those of ordinary human traders.
Comparatively, it takes 400-500 milliseconds for a human being to blink, while
HFTs can trade hundreds or thousands of times during a similar period (O’Hara,
2015).
As Aldridge (2013) notes, the process of impounding information from news

into prices is not instantaneous. The price will first swing back and forth, due to
implicit negotiations among the many buyers and sellers that can be seen in the order
flow, before it eventually finds its optimal post-announcement price range. The price
fluctuation gives HFTs an opportunity to profit from arbitraging surrounding news
releases and to bring the market one step closer to its efficient state, following the
efficient market hypothesis. Using directional event-based strategies, HFTs will
place their trades based on forecasted market reactions to an event.
Hendershott et al. (2011) use the introduction of autoquotes on the NYSE,

starting in 2003, as an exogenous event that increases the amount of AT in some
stocks but not others. Before the introduction of autoquotes, market specialists in the
NYSE had to manually disseminate the inside quotes whenever there was a relevant
change in the NYSE limit order book. This would arise (1) with the arrival of a
better-priced order, (2) when an order at the inside quote was cancelled, (3) when the
inside quote was wholly or partially traded, or (4) when the size of the inside quote
changed. The authors’ findings show that AT increases quote informativeness and
decreases trade-related price discovery, that is, AT activities discover efficient prices
via quote updates rather than via trades. Hendershott et al. propose the ability of
algorithms to respond quickly to information contained in order flows and adjust
their trading and quoting activities accordingly as a plausible explanation for these
findings. Brogaard et al. (2014) examine the role of HFT in the price discovery
process using the NASDAQ HFT dataset and find that HFT generally plays a
positive role in impounding information into stock prices. HFTs are found to trade in
the same direction as permanent price changes and in the opposite direction of
temporary pricing errors, which is consistent with the informed trading behaviour
proposed by Kyle (1985).
34
Conrad et al. (2015) examine the full cross-section of securities in the U.S.
equity markets from 2009 to 2011. They find that high-frequency quoting is
associated with price series that more closely resemble random walks, which
indicates greater efficiency in the price discovery process. In addition, they use data
from the 300 largest stocks from the first section of the Tokyo Stock Exchange in an
out-of-sample analysis. They use the introduction of the Arrowhead trading platform
in January 2010 as a proxy for an exogenous shock that facilitates high-frequency
quoting and trading on the Tokyo Stock Exchange. Before Arrowhead, the time
between an order submission and posting to the order book or order execution
ranged between 1 and 2 seconds. After Arrowhead’s introduction, latency dropped to
around 5 milliseconds. The findings confirm the authors’ evidence from their US
equity market study; that is, HFT activity has a positive impact on the price
discovery process.
Boehmer et al. (2020) study the effect of AT on informational efficiency in

42 equity markets around the world between 2001 and 2011. They use the formal
introduction of colocation services as an exogenous technological shock that allows
high-speed traders to minimise their trading latency by locating their physical
computer hardware close to the exchange. The event date is based on the actual
introduction of colocation services made by the exchanges, which marks their
commitment to provide low-latency infrastructure in their markets. Ceteris paribus,
the faster trading infrastructure would be more appealing in AT and, therefore,
should increase the level of AT in the post-colocation period. Their authors’ findings
show that stocks with greater AT intensity have intraday return autocorrelations with
lower absolute values, which indicates more random prices. Overall, their results
show that greater AT intensity leads to higher price efficiency.
Zhang (2010) investigates the role of HFT in incorporating fundamental

information into asset prices by using all stocks covered by the Center for Research
in Security Prices and Thomson Reuters Institutional Holdings databases between
1985 and 2009, with quarterly data on analyst forecast revisions and earnings
surprises proxying for news about firm fundamentals. The lower-frequency dataset
allows for the observation of a possibly longer-term impact of HFT, which is
arguably more important from the perspective of market efficiency and resource
35
allocation. Zhang suggests that HFTs’ statistical arbitrage trading strategies fit the
classical characteristics of short-horizon traders and are, therefore, likely to impact
market efficiency. The author’s findings show that HFT is negatively related to the
market’s ability to impound firms’ fundamental information into prices. In addition,
the market tends to overreact to a firm’s fundamental news when HFT is at a high
volume.
In short, greater HFT activity is found to have a positive association with

price discovery, especially in research articles that utilise higher-frequency datasets.
Regardless, as rebutted by Zhang (2010, p. 3), “it is unclear how a price discovery
delayed by 50 milliseconds or 2 seconds would affect resource allocation efficiency
in any meaningful way.” Thus, to obtain a holistic picture of the effect of HFT on
price discovery and efficiency future researchers should utilise datasets of various
frequencies.
2.4.2 Liquidity
Theoretically, HFT could have both positive and negative effects on liquidity. The
light-speed trading activity of HFTs is claimed to promote liquidity through rapid
price adjustments, allowing for narrower bid-ask spreads within a market,
strengthening intermarket linkages and activity (Goldstein et al., 2014), and lowering
the cost of intermediation (Jones, 2013). However, the higher level of trading
activity of HFTs cannot simply be an indicator of better liquidity in the market, since
the HFTs can be on either side of the trades. Dominance in the supply side would
lead to higher liquidity and narrower spreads, while greater trading activity on the
demand side would remove liquidity from the market and widen spreads (Goldstein
et al., 2014). For instance, the US Commodity Futures Trading Commission (CFTC)
and SEC published a joint report (CFTC & SEC, 2010b) suggesting that, even
though HFTs usually provide liquidity, during the flash crash of 2010, discussed
later, they turned to consuming liquidity. Easley et al. (2011) suggest that their
actions produce toxic order flow and can exacerbate any ongoing liquidity crisis.
This behaviour of HFTs has led to calls for regulatory discussion and debate on
whether to impose quotation obligations on HFTs and/or prevent them from high-
speed quotation entries/deletions (Gomber et al., 2011).
36
Generally, market-making HFTs provide liquidity by matching buy and sell
orders, or by buying and selling securities from their own inventories should they
fail to immediately match buyers and sellers (Shorter & Miller, 2014). HFTs that
engage as market makers use their speed advantage to quickly update quotes, and
they profit from the difference between the price buyers are willing to pay and the
ask price sellers are willing to accept for a security. Since this activity requires HFTs
to maintain limit orders on both sides of the trades, it provides liquidity to the market
(Chung & Lee, 2016). Riordan and Storkenmaier (2012) study the effect of a
decrease in latency on market quality following the Deutsche Börse’s release of
Xetra 8.0 in 2007. The new trading platform was introduced solely to reduce system
latency, with no other meaningful microstructure change. Following the introduction,
system latency was reduced from 50 to 10 milliseconds. Riordan and Storkenmaier
find significant reductions in spread measures, which indicate liquidity improvement
in the market after the upgrade.
Even though the increasing competition in market making generally benefits

the market, HFTs do not have an affirmative obligation to make the market like a
traditional market maker or specialist. This raises concern that they could cause
disruptions by fleeing the market at will (Carrion, 2013), for example, when it is no
longer profitable (Anand & Venkataraman, 2013). The lack of constraining
obligations also gives HFTs more flexibility to formulate market-making strategies
beyond traditional means (Brogaard et al., 2014). To overcome this issue, certain
trading venues have taken the initiative to offer liquidity rebates to attract market-
making HFTs. This relation is mutually beneficial to both HFTs and the exchanges
themselves. For rebate-paying exchanges, the rebate incentivises HFTs to route their
orders to their exchanges, which translates into greater liquidity supply (Harris,
2015). To market-making HFTs, the rebate rewards them for their liquidity-
supplying service (U.K. Government Office for Science, 2012). Studies find that
HFT market makers would lose money in the absence of the rebate (Hendershott &
Riordan, 2013), especially when transactions exhibit tighter spreads (Brogaard et al.,
2014).
37
Goldstein et al. (2020) examine strategic order placement strategies
employed by HFTs, using proprietary data on the Australian Securities Exchange.
Their findings show that, on average, HFTs supply liquidity on the thick side of the
order book, where depth is less needed, and demand liquidity on the thin side of the
order book, where depth is much needed. Through the placement of strategic limit
orders, HFTs initially submit their orders when a small order imbalance arises. If the
order book imbalances continue to move in the same direction as the HFTs’ initial
orders, then the order would be left for execution. In the event of the order
imbalances becoming less favourable, the HFTs will quickly cancel or amend their
orders, which will exacerbate future order book imbalances. In short, while it is
indeed true that HFTs supply liquidity to the market, they mostly do so when
liquidity is less needed and shy away when liquidity is most needed. This behaviour
is consistent with order anticipation strategies, and it is more evident during volatile
periods and when trading is faster.
Aitken et al. (2018) highlight the importance of safeguarding not only market
efficiency, but also market fairness, as part of the mandate by regulatory bodies such
as ASIC (Australia), the SEC (United States), and the Financial Services Authority
(United Kingdom). Using trade-to-trade data on all listed securities on the LSE and
NYSE-Euronext Paris from 2003 to 2011, Aitken et al. (2018) find that higher levels
of AT increases liquidity, as indicated by the smaller effective spreads across all
stocks in London and top quantile stocks in Paris, all while maintaining the market
fairness of either market. Higher levels of AT activity are also found to improve
market fairness in top quintile stocks (by volume traded) in both markets, as
evidenced by the decline in end-of-day price manipulation and information leakage.
2.4.3 Volatility
In a theoretical paper, Froot et al. (1992) show that short-term traders could
demonstrate herding behaviour when reacting to one source of information, creating
price dislocation and excess volatility. Their model shows that short-term traders
could be relying too much on short-term information, even when it is not related to a
firm’s fundamentals at all, which, in turn, leads to price inefficiency. For example,
technical analysts or chartists “use forecasting methods that appear, at best,
tangentially related to fundamental values” (Froot et al., 1992, p. 1480), creating
38
noise and hindering price discovery. Similarly, Jarrow and Protter (2012) theorise
that HFTs’ collective (but independent) reaction towards common signals could
resemble the action of a large trader, which can create excessive price pressure and
volatility.
Breckenfelder (2019) examines the effects of competition between HFTs on

market quality using NOMX-St index stocks (OMX S30) as the sample, from June
2009 until January 2010. The author finds that HFTs’ competition for trade increases
intraday volatility but has no significant effect on interday volatilities, which is
consistent with the typical behaviour of HFTs that close their positions at the end of
each trading day, leaving no open position overnight. In addition, Breckenfelder
shows that HFTs will behave more speculatively when an increase in their numbers
in the market is accompanied by HFT competition, resulting in liquidity
deterioration and higher short-term volatility. Similarly, Zhang (2010) finds that
HFT is positively correlated with price volatility, even after controlling for stock
fundamentals and explanatory variables for volatility, which could hinder the
market’s ability to impound firms’ fundamentals into asset prices. The effect is
stronger for the 3000 largest stocks by market capitalisation, for stocks with high
institutional holdings, and during periods of high market uncertainty.
Shkilko and Sokolov (2020) investigate the effect of heavy precipitation

shocks on HFTs’ network speed. HFTs have been using microwave towers to gain a
competitive edge over their rivals, since this technology is 30% faster than the
nearest alternative, that is, fibre-optic cable. In addition, the microwave tower
network is accessible to only a small group of trading firms (e.g. HFTs). Regardless,
one major drawback of this network is that the speed is susceptible to precipitation,
such as rain or snow, creating a speed differential between days with good and bad
weather. Their results show that a higher level of fast trading (i.e. HFT) in the market
is associated with greater volatility, as evidenced by their result showing lower
volatility when precipitation is heavy.
Boehmer et al. (2018) show that HFTs do have highly correlated trading
strategies. Coupled with their dominating presence, their actions can exacerbate price
movements and increase short-term volatility. Using the S&P/TSX 60 Index stocks
39
from June 2010 until March 2011 as the sample, Boehmer et al. investigate whether
competition between HFTs affects the short-term volatility of individual stocks.
Their results show that HFTs whose underlying common strategy is market making
have a negative impact on short-term volatility. The authors find that HFT
competition within this category reduces both the permanent and temporary price
impact of trades, which explains the lower short-term volatility. These results are
consistent with Hagströmer and Nordén (2013) finding that market-making HFTs are
able to mitigate intraday volatility through their liquidity supplying activity.
2.4.4 Adverse Selection Costs

According to ASIC (2010, p. 15), in exchange market trading, there is a risk that the
person you trade with is more informed than you are. If this is so, or you fear that
this is so, you may respond by becoming more risk-averse, reducing the price at
which you are willing to buy or increasing the price at which you are willing to sell.
The consequence of such adverse selection is a widening of spreads. At an extreme,
investors might decline to participate in trades or to not post limit orders. If many
participants in a market act according to the principles of adverse selection, trading
becomes encumbered and inefficient.
Some argue that the sheer speed of HFT causes slower investors to bear the
cost of adverse selection (Jones, 2013). In a theoretical paper, Budish et al. (2015)
develop a model in which market makers or traders that invest in speed will be the
first to react and make profits from newly arrived public information. In the event of
the traders receiving and reacting to news before the market makers do, the fast
traders will trade with stale quotes, which imposes adverse selection costs on the
market makers. This situation will discourage the provision of liquidity and,
consequently, the market makers will include the cost of the adverse selection in
their quotes, resulting in wider spreads and higher costs for slower investors.
Brogaard et al. (2014) argue that, should information eventually become

public without the intermediation of HFTs, the adverse selection costs that slower
traders have to bear could cause potential welfare gains from the faster price
discovery to become trivial or even negative. Biais and Woolley (2011) also agree
that, while the development of sophisticated and rapid trading algorithms can benefit
40
the markets and investors through better price discovery and liquidity, they can be
detrimental to slower traders, due to the adverse selection problem. In a similar vein,
Biais et al. (2015) claim that, even though investment in fast trading does help
address the issue of market fragmentation, it also comes with the risk of adverse
selection to slow traders, decreasing social welfare. Scholtus et al. (2014) find
evidence of deterioration of market quality around U.S. macroeconomic
announcements. Using a 60-second event window around the release of news, they
find that higher AT activity leads to lower depth and higher quoted spreads, adverse
selection costs, and volatility measures.
It is also possible for the algorithm to be fed false information, either

intentionally or accidentally. For instance, the price of United Airlines (UAL) stock
suddenly crashed from US$12 to US$3 on September 8, 2008, for a 12-min period,
during which shareholders lost approximately US$1 billion (Maynard, 2008). An
investigation later revealed that the rapid drop was mainly due to the interplay
between algorithms that reacted to a six-year-old headline that mistakenly hit a news
feed. Human traders should not have been deceived by the headline blunder
(Donefer, 2010). Similarly, two weeks prior to the UAL event, on August 27, 2008,
Bloomberg News accidentally published an obituary for Steve Jobs, Apple’s chief
executive officer (Elmer-DeWitt, 2008). However, the blunder happened during off-
trading hours and was quickly retracted (Donefer, 2010). These events do raise the
question of whether the immediacy of information dissemination is always a good
thing.
Shkilko and Sokolov (2020) show that, when network disruptions occur, the
microwave towers used by HFTs lose their speed advantage, and adverse selection
and trading costs decrease by up to 6.7 and 5.2%, respectively. These results hold
even when tested using the long-term removal of speed differentials. In addition, the
authors find higher trading volumes under such scenarios, consistent with the notion
of greater gains from trades in the absence of ultra-fast traders. In a recent paper by
Cartea et al. (2019) using NASDAQ’s Historical TotalView-ITCH dataset, ultra-fast
activity is associated with lower stock market quality as evidenced by higher quoted
spreads, higher effective spreads, and lower limit order book depths. The results are
41
also found to be economically significant and robust to different specifications,
endogeneity tests, and alternative measures of ultra-fast activity.
In conclusion, scholars’ understanding of the impact of HFT on market

quality is still lacking, due to HFT’s short time in operation and lack of high-quality
data (Carrion, 2013). However, the empirical evidence on the impact of HFT on
market quality is generally mixed. Despite many empirical studies finding positive
effects of HFTs participation in the market, they cannot rule out the possibility that
HFT could harm the market through predatory trading strategies (Manahov et al.,
2014). Recent empirical evidence on HFT studies seems to be in favour of the latter
view (e.g., Cartea et al., 2019; Goldstein et al., 2020; Shkilko & Sokolov, 2020).
Regulatory bodies around the world either have implemented or are mulling over
rules to contain and mitigate any HFT activity that has the potential to be detrimental
to market quality (Benos & Sagade, 2016). It is generally agreed that any abusive or
predatory trading activity, which goes against market integrity, should not be
permitted. Nonetheless, regulators must be careful in formulating their regulations,
to avoid any excessive regulations and constraints that could be counterproductive
and have unanticipated effects on market quality. For example, any newly
formulated regulation should not prevent beneficial HFT strategies that have positive
effects on liquidity, such as market-making strategies, or price discovery and market
efficiency such as arbitrage strategies (Gomber et al., 2011).
2.5 Controversies Relating to HFT

This section highlights the negative sentiment and controversies surrounding HFT.
The identified controversies are (1) the flash crash of May 6, 2010, (2) the economic
welfare of the arms race, and (3) HFTs’ market-making obligations.
2.5.1 Flash Crash of May 6, 2010

On May 6, 2010, the U.S. financial markets were shocked with a short-lived yet
severe drop in prices that happened within minutes. The sudden market crash of May
6, 2010, was later dubbed a flash crash, given its brief time frame. The CFTC and
SEC released joint preliminary findings concerning the event on May 18, 2010
(CFTC & SEC, 2010a), and full findings were released later, on September 30, 2010
(CFTC & SEC, 2010b).
42
The U.S. financial markets opened on May 6, shrouded by negative market
sentiment stemming mainly from the European debt crisis, causing the S&P 500
volatility index to rise by 22.5% from its opening level by around 2.30 p.m. Central
Standard Time. This triggered investors to engage in a flight to quality, creating
selling pressure that pushed the Dow Jones Industrial Average down by 2.5%. At
2.32 p.m., Waddell & Reed, a large fundamental trader, initiated a sell order
algorithm to sell 75,000 E-mini (S&P 500 futures) contracts to hedge its existing
equity position (CFTC & SEC, 2010a; Lash & Spicer, 2010). Initially, the selling
pressure was absorbed by HFTs and other intermediaries in the futures market,
followed by fundamental buyers and cross-market arbitrageurs, the latter transferring
the selling pressure to the equities market.
From the Waddell & Reed Sell Algorithm order, HFTs accumulated a net
long position in E-mini contracts, which led them to aggressively sell the contracts
they held to reduce their inventories (CFTC & SEC, 2010b). Nearly 140,000 E-mini
contracts (over 33% of the total trading volume) were traded by HFTs. The dramatic
increase in trading volume raised the volatility in the market, which, in turn, scared
long-term traders away. The lack of demand in the market caused HFTs to buy and
sell from each other, generating a “hot potato” volume effect that drove the price of
the E-mini down by 3% within a 4-min period. At the same time, cross-market
arbitrageurs that bought the E-mini simultaneously sold equivalent amounts in the
equity markets, driving the price of the S&P 500 SPDR (SPY) down by
approximately 3% as well.
The combined selling pressure created a dramatic order imbalance in the

market, pushing E-mini prices down by another 1.7% in a mere 15 seconds, to reach
an intraday low of 1,056 points. At 2:45:28 p.m., the Chicago Mercantile Exchange’s
Stop Logic Functionality was triggered by the E-mini’s rapid price decline, halting
all E-mini trading for 5 seconds. After trading resumed at 2:45:33 p.m., the E-mini
price stabilised and started to recover, thanks to opportunistic and longer-term
traders that re-entered the market and rapidly accumulated long positions (Kirilenko
et al., 2017). Subsequently, SPY also recovered.
43
Despite the E-mini’s recovery, the prices of many other securities continued
to show extreme volatility. Approximately 2 billion shares with a total value of more
than US$ 56 billion traded between 2:40 p.m. and 3:00 p.m. on May 6. During this
20-min window, more than 98% of all shares were traded within 10% of their value
at 2:40 p.m. However, some extreme trades were recorded. For example, Accenture
plc (ACN) rapidly declined from about US$ 30 at 2:47:47 p.m. to US$ 0.01 by
2:47:54 p.m. and then recovered within a matter of a few more seconds (CFTC &
SEC, 2010a). These extreme cases were caused by orders executed against stub
quotes, which were triggered by the sudden loss of liquidity (Gomber et al., 2011).
Stub quotes are quotes generated by market makers at levels far from the current
market, to comply with the obligation to maintain continuous two-sided quoting.
However, stub quotes are not intended to be executed (CFTC & SEC, 2010a).
Overall, over 20,000 trades (amounting to 5.5 million shares) across 300
different securities and exchange traded funds traded at prices 60% or more from
their 2:40 p.m. prices. By 3:00 p.m., the prices of most securities had reverted to
their rational values. After the market closed, the SEC and FINRA met and agreed to
adopt the “clearly erroneous” trade rules, and thus, all trades classified as clearly
erroneous were considered “broken trades.” From the joint report, it is evident that
HFTs did not trigger the flash crash. However, the repeated buying and selling of
contracts executed by the automated systems created the hot-potato effect as HFTs
competed for liquidity. Thus, their trading behaviour during a period of unusually
high selling pressure on May 6 is perceived to have exacerbated the price declines
and market volatility (Kirilenko et al., 2017). A study by Brogaard et al. (2018)
supports the view that HFTs play a positive role in supplying liquidity in times of
extreme price movements; however, when multiple extreme price movements occur
simultaneously, such as with the flash crash, they can switch to being demanders of
liquidity. Due to the May 2010 flash crash, HFT has received considerable critical
attention from both the CFTC and the SEC (2010b) for creating excessive short-term
volatility.
44
2.5.2 Information acquisition, social welfare issues, and the HFT arms race
HFTs’ contribution to the process of price discovery can be beneficial, since more
informative stock prices should lead to better resource allocation in the economy.
However, some studies have questioned whether AT does, in fact, lead to more
informative stock prices. Price informativeness relates to the degree to which firm-
specific information is incorporated into prices. Weller (2017) shows that AT, which
includes HFT, can reduce price informativeness, despite prior evidence of AT
improving price efficiency. In measuring information content in prices, Weller
examines the information content of stock quarter announcements using SEC data
and shows that the information content in prices decreases by as much as 13% per
standard deviation of AT activity. In other words, AT activity appears to detract
from the information acquisition activities of other market players. Weller’s (2017)
work complements that of Baldauf and Mollner (2020). Baldauf and Mollner show
that, while HFT leads to smaller spreads, it also leads to less intensive research and,
therefore, less information production and less informative stock prices as informed
traders are squeezed out, having less time to trade.
On a similar note, Menkveld (2014) agrees that the presence of market-

making HFTs in electronic markets does improve welfare, by reducing informational
frictions from nonsimultaneous order arrivals in the market. Another welfare benefit
is proposed by Malceniece et al. (2019). Their study of Chi-X’s entry into European
equity markets shows HFT activity increases comovement in returns and liquidity.
This improvement in comovement, or synchronicity, makes companies, particularly
smaller ones, more attractive to large investors and, therefore, leads to more efficient
pricing. However, the net welfare from HFT is questionable; the positive
contribution from market-making activity can be negated when HFTs pick off
investors’ quotes at lightning speed on information that will surely reach marginally
slower investors later.
HFTs acknowledge the importance of investing in hardware, software, and

network capabilities to reduce latency in automated trading processes, motivated by
the nature of the game where the winner takes all. The upgrades allow them to
continuously refine their trading algorithms and win the arms race (Kauffman et al.,
2015). Regardless of the fact that the technology arms race has significantly
45
shortened transaction times the process does raise concerns about excessive spending
without meaningful progress in market quality (Chung & Lee, 2016). The race
among institutions to be the fastest is deemed unproductive, and the investment in
the technological infrastructure required to reduce trading latency creates doubts as
to whether HFT adds value overall (Chordia et al., 2013; Jones, 2013). In addition,
Menkveld (2014) asserts that the technological investment itself could well be the
source of a negative externality through the relative speed disadvantage it creates for
others.
As Brogaard et al. (2015) note, not all HFTs choose to pay for colocation
services, and there are many collocated non-HFTs. This suggests that not all HFTs
have the need to be the fastest. The pursuit to be faster is relevant if and only if its
benefits are greater than its costs. Baron et al. (2019) investigate the importance of
superior relative latency on HFTs’ performance on the NOMX-St. Their findings
show that, by being among the fastest, HFTs with a passive (e.g., market-making)
trading strategy experience lower adverse selection risk (i.e., have better risk
management), and HFTs with an active trading strategy (e.g., news trading) are more
responsive to new information. The authors also find high concentrations of trading
volumes and trading revenues among the fastest HFTs, and these were nondeclining
over their 5-year sample period. 4 Additionally, new HFT entrants are relatively
slower, underperform, and are more likely to exit. These findings confirm that the
small differences in latency have a substantial effect on HFTs’ performance.
In a more recent study, Aquilina et al. (2020) use LSE message data to
quantify the latency arbitrage race, allowing them to observe the details of the race
between HFTs.5 Their findings show that there is about one race per minute in every
symbol traded on the FTSE 100; the winner is 5-10 milliseconds faster than the
runner-up (i.e., the first loser); the races account for approximately 20% of the daily
trading volume; the top six firms win about 82% of the races, but lose 87% of the
4
The model of Baron et al. (2019) predicts that the average daily trading revenue of the five fastest
HFT firms is SEK 15,451 higher than that of the other HFT firms. Being one of the top five fastest
high-frequency traders is associated with earning an additional SEK 24,639, on top of their revenue.
5
A limit order book provides a complete picture of all the messages that add liquidity to the order
book, such as new limit orders, cancelled orders, and trades, time stamped with extreme precision.
However, the message data record all failed attempts to trade or cancel, which is not captured by the
limit order book because they failed to be submitted to it.
46
races; aggressive orders (e.g., quote sniping) win 90% of the races; the top six firms
cumulatively supply about 42% of the liquidity in the races, while taking about 80%;
and most of the liquidity taken by the winners is supplied by market participants
other than the top six. In the aggregate, the small races collectively (i) make up a
significant proportion of the price impact, (ii) create meaningful harm to liquidity,
and (iii) add up to a substantial winning stake in the arms race, valued on the order of
US$ 5 billion per annum in the global equity markets. Aquilina, Budish, and O’Neill
also find that, by eliminating latency arbitrage, the cost of liquidity can be reduced
by almost 17%.
Even without the issue of an arms race, HFTs still pose a threat to many,
since they can use high-speed predatory trading strategies (see the harmful strategies
discussed in Section 2.3.2), such as introducing “microstructure noise” that generates
an unnecessary extra layer of intermediation between buyers and sellers, leading to
higher price volatility and worse market quality (Cartea & Penalva, 2012).
2.5.3 Market-making obligations and excessive liquidity

Anand and Venkataraman (2013) study the trades of two types of market makers,
designated market makers (DMMs) and ELPs. The main difference between DMMs
and ELPs lies in their obligation to make the market. DMMs, or specialists, are
bounded by specific obligations imposed by the exchange, that is, to maintain a
market presence by continuously posting quotes with reasonable depth. ELPs, on the
other hand, employ market-making strategies because of their profitability, with no
affirmative obligations to maintain markets. Anand and Venkataraman (2013) state
that HFTs are the most active market makers in financial markets today, where some
position themselves as ELPs, meaning that they are likely to supply liquidity
whenever it is profitable for them to do so. They will cease providing liquidity when
facing large adverse selection risks (Chung & Chuwonganant, 2018) or whenever
market conditions are unfavourable for them to make profits. This is more likely to
happen in times of high market uncertainty (Zhang, 2010).
The lack of commitment to make the market, especially in times of market

stress and in thinly traded securities, raises concerns among practitioners and
regulators. HFTs’ optional market making can exacerbate execution uncertainty, and,
47
thus, the liquidity supplied by HFTs is deemed unreliable, which can reduce
investors’ confidence and participation. Liquidity withdrawals by HFTs can thin out
the order book, which can induce extreme market movements (Gomber et al., 2011).
Such a situation could also be the underlying reason for the heightened sensitivity of
liquidity and returns to market volatility in the HFT era. Furthermore, non-HFTs are
playing on an uneven playing field, because of their technological inferiority with
respect to HFTs, and they could find that the market is unfair and, consequently, stop
participating altogether (Anand & Venkataraman, 2013). In response to this potential
problem, regulators considered imposing quotation obligations on HFTs and/or
preventing them from engaging in high-speed order entries and cancellations
(Gomber et al., 2011).
2.6 Responses to HFT

Due to the potential harm caused by the earnings model of predatory HFTs,
regulatory bodies and scholars around the world have introduced or proposed various
measures such as speed bumps, price improvement rules, and new trading mechanics
to ensure a fair and orderly market.
In 2013, the Investors Exchange pioneered the use of speed bumps to slow
down traffic by intentionally delaying order messages submitted to and from its
trading platform. The Investors Exchange speed bump symmetrically delays all
orders by 350 milliseconds, using a 61 kilometres fibre-optic cable near its trading
engine. Nonetheless, this speed bump design affects not only liquidity takers (e.g.
order snipers), but also liquidity providers (e.g. market makers). Therefore, several
other exchanges have introduced variations in the mechanics of speed bumps,
designed to asymmetrically affect predatory latency arbitrageurs.6 In an experimental
study, Khapko and Zoican (2020) show that asymmetric speed bumps reduce
investment in the arms race by 20%, and, by increasing the lengths of the speed
bumps by one standard deviation, the endowment is further reduced by 8.33%.7
6
See Baldauf and Mollner (2020) for a list of exchanges worldwide that implemented or proposed
the use of speed bumps.
7
In Khapko and Zoican’s study, one standard deviation of the speed bump corresponds to 2 seconds,
or 40% of the unconditional exchange latency.
48
Their findings also suggest that symmetric speed bumps have no effect on reducing
the arms race investment and are equivalent to having no speed bump at all.
Dark trading, which is conducted in dark pools as opposed to disclosed

trading conducted in “lit” markets, is used by institutional investors to trade large
positions that, if sent as a single order to the market, could spark large price
movements. Dark trading also allows institutional investors to hide their large orders
from the public, giving them a better chance to implement their trading strategy.
However, by sending pinging orders, HFTs are able to detect hidden liquidity and
take strategic positions against the large hidden orders, making a profit at the cost of
institutional investors. In 2013, ASIC introduced a new market integrity rule that
require below block size dark trades to be executed at an “improved price,” a price at
least one tick size better than the prevailing best bid or ask price, or the midquote on
lit exchanges. The price improvement rule protects the interests of investors posting
orders on the lit exchange and disincentivises predatory HFTs from conducting a
latency arbitrage strategy, due to the higher spread in the dark pools. ASIC (2015)
reports that, after the implementation of the new rule, the percentage of block trades
against total turnover increased by 46%, while the percentage of trades below a
block decreased by 22%. This development aligns with the original purpose of using
dark trades, that is, to facilitate the trading of large orders.
Budish et al. (2015) claim that the arms race is socially wasteful, but its
existence is actually a symptom, stemming from a flaw in the architecture of modern
financial exchanges that use continuous-time trading, which also creates adverse
selection rents that attract HFT. Budish et al. suggest that the problem can be
addressed by using frequent batch auctions, which will create a discrete-time market
to replace the current market design, which is based on the continuous limit order
book. All orders received in the same batch (i.e., time interval) are treated equally,
and that with the best price (i.e., highest bid or lowest ask) will be matched first.
This encourages competition on price rather than speed, thus, reducing the incentive
to place an order first. This approach will render the tiny speed advantage much less
valuable, intuitively ending the arms race. However, Yao and Ye (2018) find
evidence that, even with discrete timing, HFTs could continue to race against each
49
other, this time competing for rents from the queuing channel, originated from yet
another microstructure design element, the tick size.
At the opposite end of the spectrum, Kyle and Lee (2017) propose fully
continuous exchange where buyers and sellers use a new type of order, called a
continuous scaled limit order. The new type of order requires traders to supply five
parameters: (i) a buy or sell indicator, (ii) the maximum quantity, (iii) the lowest
price, (iv) the highest price, and (v) the maximum trading speed (shares per hour).
For example, a buy order message is read as “buy up to the maximum quantity of
shares, between the lowest and highest prices, at the maximum buying rate per
hour.” The exchange would calculate the market clearing price (to the nearest one-
tick price increment) that satisfy all flows of demand and supply submitted to the
market. The trading mechanism slices the trading volumes and gradually spread
them over time, allowing large orders to continuously flow into the market. This
approach removes the need for large traders to self-determine their optimal trade
size. In the event of changes in fundamentals, fast traders would still have the upper
hand in updating their orders. This advantage, however, is limited by the flow rate of
trade set by slow traders, which limits the quantity of shares fast traders can trade,
hence, reducing slow traders’ adverse selection costs. Kyle and Lee (2017) suggest
that a continuous scaled limit order would address the harmful incentives created by
the discreteness in price, quantity, and time found in conventional limit orders.
2.7 Conclusions
Since the advent of the infamous flash crash of 2010, HFT has gained considerable
attention from regulators, researchers, and others following or involved in stock
markets. Even though almost a decade has passed since, there is still no standard or
universal definition for HFT. The lack of a precise definition leads to other problems
such as inaccurate estimations of HFTs’ market shares and influence. This also
complicates the process of researching HFT, which leads to the various conclusions
on the actual effect of HFT on the market. Nonetheless, five traits are commonly
associated with HFT: (1) a specialised form of AT, (2) the use of high-speed,
sophisticated computer programs and systems, (3) very high order-to-transaction
ratios, (4) extremely short average holding periods, and (5) ending the trading day
with flat positions. Using these unique characteristics, researchers have teamed up
50
with regulators and exchanges to accurately identify HFT activity in the market, as
depicted by the NASDAQ HFT dataset. Alternatively, researchers with access to
order book-level information can use commonly used HFT proxies, such as the
OTR, which measures quoting intensity, or even Hasbrouck and Saar’s (2013)
RunsInProcess, which is based on HFTs’ dynamic and low-latency order placement
strategies.
Using fully automated trading strategies, HFTs attempt to identify and profit
from short-term irregularities and earn small amounts of money from every trade.
The ability to trade at low latency allows HFTs to profit from the trading
environment itself. HFTs exhibit variability in their trading strategies by
documenting differences in liquidity provision, end-of-day and maximum intraday
positions, and trading revenues, for example, Benos and Sagade (2016). The
algorithms HFTs employ can determine their order execution style, such as being
aggressive or passive, or sending orders as either one trade or split into smaller
trades. Aldridge (2013) generally categorises HFT strategies into statistical arbitrage,
directional strategies, and market making, where all three are considered beneficial
strategies. Nonetheless, HFTs can also employ harmful strategies, such as front
running, spoofing, and quote stuffing, utilising their superior speed to make a profit
and prey on slower traders.
Generally, the scholarly evidence on the impact of HFT activity on market

quality is mixed. For instance, Boehmer et al. (2020) and Hendershott and Riordan
(2013) find that HFT participation leads to faster price discovery and makes pricing
more informative. Market-making HFTs are found to provide liquidity by matching
buyer and seller orders, which promotes liquidity in the market (Shorter & Miller,
2014). HFTs’ speed is also found to reduce adverse selection costs in market making
(Jovanovic & Menkveld, 2016). On the contrary, Zhang (2010) finds that HFT
negatively affects the market’s ability to incorporate firm’s fundamental information
into prices. The CFTC-SEC (2010b) joint report finds that, though HFTs usually
provide liquidity, in periods of high uncertainty, they consume liquidity. Further,
HFTs’ speed-driven advanced access to value-relevant information can also incur
adverse selection costs for slower traders in the market (Biais et al., 2015).
51
CHAPTER THREE:
ESSAY TWO
RELATIVE TICK SIZE AND HIGH-FREQUENCY TRADING
3
This chapter shows the effect of relative tick size on HFT activity. The chapter starts
with the introduction of the essay, which covers the background of the study and
research motivation. The chapter continues with literature review and hypotheses
development, followed by methodology, findings, discussions, and conclusion.
3.1 Introduction
The tick size, also known as the minimum price variation, is defined as “the smallest
amount by which share prices are permitted to fluctuate. It determines the prices at
which orders may be entered. Orders may only be entered at prices that are evenly
divisible by the minimum tick size” (ASIC, 2010, p. 84). Smaller tick sizes are
clearly important in the HFTs technique because they result in lower bid-ask spreads,
which directly affect minimum trading costs (Comerton-Forde, 2012; Harris, 1994).
Regardless, this does not mean that smaller tick sizes are always beneficial – a tick
size that is too small may negatively impact the interaction between different types
of investors in the market. This may occur, for instance, when a trader obtains
execution priority by placing a limit order with a trivially better price, allowing them
to skip the order book queue.8 This circumstance might hamper other investors’
willingness to disclose their orders, hence impeding HFTs participation in the market
and causing a reduction in market depth (Chordia et al., 2011; O’Hara, Saar, &
Zhong, 2019; Yao & Ye, 2018).
Angel (1997, 2012) emphasises the importance of tick size in market

regulation, and he explains why the optimal tick size is not zero: (1) a non-zero tick
simplifies traders’ information sets, (2) an economically significant (i.e. non-trivial)
tick size preserves price and time priority in an order book, which incentivises
8
Orders submitted to the market are queued based on price-time priority. This indicates that orders
are executed based on the best price, and if several orders share the same price, the order with the
earliest time is executed first.
52
traders to provide liquidity by posting limit orders, and, (3) tick size creates a floor
for the quoted bid-ask spread, which works as an incentive for dealers to make
markets. The narrowest bid-ask spread is one tick size; hence, a bigger tick size
increases the minimum transaction costs for investors by widening the bid-ask gap.
Furthermore, tick size restricts the number of possible price points and decimals that
can be stated, making it easier for human traders to understand the actual market
situation. This problem, on the other hand, does not affect algorithmic traders, as
their trading bots should be able to comprehend complex numbering without
difficulty.
3.2 HFT in Australia

In 2009, Recommendation 4.5 of the Johnson Report underlines the need to increase
competition on exchange-traded markets which is expected to lower costs and offer
more options for market players.9 Subsequently, Chi-X Australia was launched as an
alternative to the ASX, and since its inception, it has employed a maker-taker pricing
model, charging a fee of 0.06 basis points for liquidity providers and 0.12 basis
points for liquidity takers.10,11 In response, ASX lowered its headline trade execution
fee from 0.28 basis points to 0.15 basis points in June 2010 and established its own
low-latency alternative trading venue, ASX PureMatch, in November 2011. The new
trading platform directly competes with Chi-X Australia for HFTs’ market share, and
it aims to “meet the growing needs of the trading community for order books that
offer the most liquid stocks across the fastest available platform” (ASX, 2011).
Consequently, these innovations provide a trading environment with fragmented
markets, low-latency trading platforms, and low explicit trading fees, which, together
with the inherently small tick sizes used in Australia, may stimulate an increase in
HFT activity (Comerton-Forde, 2012).
9
The report was originally titled as “Australia as a Financial Centre – Building on Our Strengths,”
and it was released by the Australian Financial Centre Forum. Later on, the report became commonly
known as “The Johnson Report” due to the name of its chairman, Mr. Mark Johnson, the former
Deputy Chairman of Macquarie Bank.
10
On October 25, 2011, Chi-X began operations with a “soft launch” in which only a few stocks were
available for trading. Following the success of the launch and ASIC approval, Chi-X Australia fully
operates on November 9, 2011, trading all S&P/ASX 200 constituent stocks and ASX-listed exchange
traded funds (ETFs).
11
A liquidity provider is a passive limit order placed on the market that increases the depth of the
order book, whereas a liquidity taker is an aggressive market order that seeks to match against a
passive limit order and thereby decreases the book’s depth.
53
The maker-taker pricing model incentivises traders to provide liquidity,
where the liquidity “maker” is compensated with a rebate (or charged a lower
transaction fee) for the limit order they placed, whereas the liquidity “taker” is
charged a higher transaction fee for the market order they submitted. The exchange’s
profit is the difference between the transaction fees collected and the rebates
provided. Studies find that without a rebate, HFTs potentially lose money on their
market-making operations, which might discourage them from providing liquidity
(Brogaard et al., 2014; Hendershott & Riordan, 2013). In Australia, the ASX and
CHIX have different pricing model, where the former charges a similar fee to both
liquidity providers and takers, whereas the latter subsidised liquidity providers only.
Therefore, market-making HFTs would naturally prefer trading on the CHIX since
they would be more resistant to issues that may erode their profitability, such as tick
size reduction, and thus able to continue providing liquidity on the CHIX due to the
presence of a liquidity rebate.
Figures 3.1, 3.2, 3.3, and 3.4 depict the estimated level of HFT activity in
Australia based on the message-to-trade ratio (MTR), the algorithmic trading ratio
(ALGO), the total number of high-frequency order identifiers (HFO), and the HFO’s
message ratio (HFOR), which are the HFT proxies employed in this study.12,13 A
closer look at the HFT measures employed indicates that CHIX has greater HFT
activity when estimated by MTR or ALGO, whereas HFO and HFOR suggest that
ASX has consistently higher HFT activity. Trends analysis shows an increase in
HFT activity on the ASX from 2008 to 2017, with the uptrend being most
pronounced in 2010 and 2011, which may be attributable to the regulatory agencies’
initiatives to promote HFT. The subsequent decline observed from 2012 to 2014 is
presumably due to competition from CHIX for HFT’s market share, resulting in
some HFT activity shifting to the new trading venue. Since 2015, however, the
contrasting trend between the ASX and CHIX markets has been more evident, with
HFT activity rising in the former and declining in the latter, implying that the ASX
has become the preferred platform for HFTs in Australia in recent years.
12
In the latter section of this essay, the method used to derive these measures is described in detail.
13
The figures adopt a distinct y-axis for the ASX and CHIX values due to the substantial disparities in
their values; this enables the study to highlight the differences in trend between the two markets.
54
ASX ASX ASX
0
1
2
3
4
5
6
7
8
-60
-50
-40
-30
-20
-10
50
0
100
150
200
250
300
350
Q1-2008
Q1-2008 Q1-2008 Q2-2008
Q2-2008 Q2-2008 Q3-2008
Q3-2008 Q3-2008 Q4-2008
Q4-2008 Q4-2008 Q1-2009
Q1-2009 Q1-2009 Q2-2009
Q2-2009 Q2-2009 Q3-2009
Q3-2009 Q3-2009 Q4-2009
Q4-2009 Q4-2009 Q1-2010
Q1-2010 Q1-2010 Q2-2010
Q2-2010 Q2-2010 Q3-2010
Q3-2010 Q3-2010 Q4-2010
Q4-2010 Q4-2010 Q1-2011
Q1-2011 Q1-2011 Q2-2011
Q2-2011 Q2-2011 Q3-2011
Q3-2011 Q3-2011 Q4-2011
Q4-2011 Q4-2011
Q1-2012
Q1-2012 Q1-2012
Q2-2012
Q2-2012 Q2-2012
Q3-2012
Q3-2012 Q3-2012
ASX_ALGO
ASX_HFO
ASX_MTR
Q4-2012
Q4-2012 Q4-2012
55
Q1-2013
Q1-2013 Q1-2013
Q2-2013
Q2-2013 Q2-2013
Q3-2013
Q3-2013 Q3-2013
Q4-2013
Q4-2013 Q4-2013
Q1-2014 Q1-2014
Q1-2014 Q2-2014
Q2-2014 Q2-2014
Q3-2014
CHIX_HFO
Q3-2014
CHIX_MTR
Q3-2014
Q4-2014
CHIX_ALGO
Q4-2014 Q4-2014
Q1-2015 Q1-2015
Q1-2015
Q2-2015 Q2-2015
Q2-2015
Q3-2015 Q3-2015
Q3-2015
Q4-2015 Q4-2015
Q4-2015
Q1-2016 Q1-2016 Q1-2016
Q2-2016 Q2-2016 Q2-2016
Q3-2016 Q3-2016 Q3-2016
Q4-2016 Q4-2016 Q4-2016
Q1-2017 Q1-2017 Q1-2017
Figure 3.1: Quarterly average of message-to-trade ratio (MTR) in Australia
Q2-2017 Q2-2017
Figure 3.3: Quarterly average of high-frequency order ID (HFO) in Australia

Q2-2017
Q3-2017 Q3-2017 Q3-2017
Figure 3.2: Quarterly average of algorithmic trading ratio (ALGO) in Australia
Q4-2017 Q4-2017 Q4-2017
0
5
0
2
4
6
8
10
15
20
25
30
35
40
45
-8
-7
-6
-5
-4
-3
-2
-1
10
12
CHIX 14 CHIX CHIX
0.14 0.035
0.12 0.03
0.1 0.025
0.08 0.02
CHIX
ASX
0.06 0.015
0.04 0.01
0.02 0.005
0 0
Q1-2008
Q2-2008
Q3-2008
Q4-2008
Q1-2009
Q2-2009
Q3-2009
Q4-2009
Q1-2010
Q2-2010
Q3-2010
Q4-2010
Q1-2011
Q2-2011
Q3-2011
Q4-2011
Q1-2012
Q2-2012
Q3-2012
Q4-2012
Q1-2013
Q2-2013
Q3-2013
Q4-2013
Q1-2014
Q2-2014
Q3-2014
Q4-2014
Q1-2015
Q2-2015
Q3-2015
Q4-2015
Q1-2016
Q2-2016
Q3-2016
Q4-2016
Q1-2017
Q2-2017
Q3-2017
Q4-2017
ASX_HFOR CHIX_HFOR
Figure 3.4: Quarterly average of HFO’s message ratio (HFOR) in Australia
3.3 Tick size structure in Australia

In the Australian equity market, there are three nominal tick sizes: (i) A$0.001 (one-
tenth of a cent) for stocks priced below A$0.10; (ii) A$0.005 (half-cent) for stocks
priced between A$0.10 and A$1.995; and (iii) A$0.01 (one cent) for stocks priced at
or above A$2.00. This structure generates two tick size borders at A$0.10 and $2.00.
Relative tick size, which is defined as nominal tick size divided by stock prices, is
the smallest price variation expressed as a percentage of a stock’s price. Therefore,
the presence of tick size borders results in a substantial disparity between the relative
tick size of stocks priced on opposing sides of the border. O’Hara et al. (2019) assert
that “a larger relative tick size benefits HFT firms that make markets on the NYSE:
they leave orders in the book longer, trade more aggressively, and have higher profit
margins”, implying that a large relative tick size may incentivise liquidity-providing
HFTs to participate. In other words, relative tick size can be perceived as the profit
potential that HFTs can earn from a single tick of price change in a transaction.
The tick size borders in Australia results in stocks priced at A$0.099 and
A$0.10 will have relative tick sizes of 1.01% and 5.00%, whereas those priced at
A$1.995 and A$2.00 would have relative tick sizes of 0.25% and 0.50%,
respectively. The discontinuity in the relative tick size values observed at the A$0.10
and A$2.00 borders is presented in Figure 3.5, which shows a significantly different
profit potential for HFTs’ limit orders placed near the borders, which might
56
influence their trading strategy. Figures 3.6 and 3.7 provide a closer look at the
dynamics between relative tick size and stock price around the respective borders. 14
10.000%
9.000%
8.000%
7.000%
6.000%
5.000%
4.000%
3.000%
2.000%
1.000%
0.000%
0.000 0.500 1.000 1.500 2.000 2.500 3.000 3.500 4.000 4.500 5.000
Figure 3.5: Relative tick size for stock prices ranging from A$0.01 to A$5.00
12.000%
10.000%
8.000%
6.000%
4.000%
2.000%
0.000%
0.000 0.050 0.100 0.150 0.200 0.250 0.300 0.350 0.400 0.450 0.500
Figure 3.6: Relative tick size for stock prices surrounding the A$0.10 border
14
The scale used for the Y-axis in Figure 3.6 is ten times larger than the scale used in Figure 3.7.
57
1.200%
1.000%
0.800%
0.600%
0.400%
0.200%
0.000%
0.000 0.500 1.000 1.500 2.000 2.500 3.000 3.500 4.000 4.500 5.000
Figure 3.7: Relative tick size for stock prices surrounding the A$2.00 border
Table 3.1 shows the differences in tick size structures and relative tick size values
between the Australian and American markets. For all stocks traded on the U.S.
equity market, there are only two nominal tick sizes: US$0.0001 (one-hundredth of a
cent) for stocks valued less than US$1.00, and US$0.01 (one cent) for stocks priced
more than or equal to US$1.00. The tick size structure of the U.S. market results in a
substantial disparity between the relative tick size values of stocks priced near the
border. For instance, a stock priced at US$0.9999 and US$1.00 will have a relative
tick size value of 0.01% and 1.00%, respectively, signifying an almost 100-fold
increase while being separated by just one tick. In comparison, the largest relative
tick size increase in Australia is fivefold, which is only one-twentieth of what
observed in the U.S. market. Therefore, the variations in tick size structure employed
in these two markets may necessitate HFTs to use a different trading approach to
make profit, which could affect their overall presence and influence on the market
they operate in.
In addition, stocks in the Australian market inherently have no designated

market-maker, which presents an opportunity for HFTs to voluntarily undertake this
role by leveraging their speed advantage. From this perspective, Australia appears to
offer a more fertile ground for HFTs to succeed as a liquidity provider than the U.S.,
where there are already market-makers and market specialists, and thus greater
competition for providing liquidity. Therefore, HFTs in Australia may be less likely
to employ an aggressive, liquidity-taking trading approach that is detrimental to
58
other market players, and may instead allocate more resources towards market-
making activities.
Table 3.1. Tick size structure in the Australian and U.S. equity markets
This table compares the tick size structures applied in the Australian and U.S. equity markets. Tick
size refers to the nominal tick size used in each market. The relative tick size is calculated by dividing
the nominal tick size by the price.
Australia (A$)
Price Range $0.001 - $0.099 $0.100 - $1.995 $2.00 - $99,999,990
Tick Size $0.001 $0.005 $0.01
Highest relative tick size 100.00% 5.00% 0.50%
Lowest relative tick size ≈1.01% 0.25% ≈0.00%
United States (US$)

Price Range p < $1.000 p ≥ $1.000
Tick Size $0.0001 $0.01
Highest relative tick size 100.00% 1.00%
Lowest relative tick size ≈0.01% ≈0.00%
Furthermore, the minimum trading unit in Australia also differs from that in
the U.S., where all equities in the latter market must be traded at a minimum of 100
units per board lot, and trading with odd lots results in a higher trading cost. On the
contrary, the mandated minimum trading unit in Australia is one unit, which gives all
traders, including HFTs, the opportunity to trade at any unique combination with no
additional cost.15 Trading in Australia gives HFTs more flexibility in formulating
their strategy because they are not bound by the minimum trading unit. Due to the
smaller minimum trading unit, large institutional traders in Australia may have a
better opportunity to hide their huge orders among retail orders in the market by “re-
packaging” their single large orders into multiple smaller orders, as opposed to those
in the U.S. market. This trait may impair HFTs’ ability to detect approaching large
orders, making it more difficult for them to “ride the wind”.
Empirical studies on tick size are mainly centred in the U.S. financial markets
(see for example Angel, 1997; Bessembinder, 2003; Gibson, Singh, & Yerramilli,
2003; Goldstein & Kavajecz, 2000; Jones & Lipson, 2001; Lipson & Mortal, 2006;
O’Hara et al., 2019; Schultz, 2000; Yao & Ye, 2018). However, the aforementioned
15
This is compared to the higher fees charged for odd-lot trading in markets with a minimum trading
unit of greater than one.
59
argument suggests that the differences in the underlying architecture of the
Australian and American markets may influence HFTs’ strategy, activity, and
behaviour, pushing them to operate differently in each market. In summary, factors
such as a distinct tick size structure, the absence of a designated market-maker, and a
more flexible minimum trading unit policy, provide a unique trading environment
for HFTs in Australia. Therefore, the results of this research, which utilises the
Australian dataset, contribute to a better understanding of HFT, notably on the
influence of tick size on their behaviour and strategy in a pure order-driven market.
3.4 Literature review and hypotheses development

3.4.1 Nominal and relative tick sizes
In a frictionless market, the bid-ask spread, which is simply the difference between
the lowest ask price and the highest bid price, would theoretically settle at a level
that compensates liquidity providers exactly for the value received by liquidity takers
(Linton et al., 2013). In practise, however, markets do have frictions due to factors
such as information asymmetry, illiquidity issues, and costs deriving from the
market’s architecture, such as nominal tick size (see e.g.: Amihud et al., 2005; Biais
et al., 2005; Fama, 1970). As a result, these frictions ensure that the bid-ask spread is
always greater than zero.
Tick size is the minimum price increment or pricing grid that constraints
market participants from submitting quotes in increments smaller than the mandated
tick size value. Most stock exchanges around the world operate under a fixed tick
size regime, wherein the regulator nominates a tick size value that is applied to all
stocks. As shown in the previous section, it is not uncommon for different nominal
tick sizes to be assigned to distinct, non-overlapping price ranges; however, the
nominated value will remain constant within its specified range. On the other hand,
the relative tick size is dynamic and sensitive to stock price movements and changes
in nominal tick size. A relative tick size may be shocked when: (i) a regulatory body
revises an existing tick size value; (ii) a stock price naturally crosses a tick size
border; and (iii) a corporate action or informative event causes a substantial shift in
stock price. O’Hara et al. (2019) propose that relative tick size provides fund
managers with a clearer understanding of the transaction costs incurred when trading
60
a certain dollar amount of fund compared to nominal tick size, which simply displays
the quoted cost in cents.
A number of studies have shown that the size of a nominal tick or relative
tick may have a substantial impact on the profits and decisions made by liquidity
providers and liquidity takers (e.g., Angel, 1997; Foley et al., 2019; Foley et al.,
2022; Glosten, 1994; Li et al., 2021; Mahmoodzadeh & Gençay, 2017; O’Hara et al.,
2019; Werner et al., 2019; Yao & Ye, 2018). A liquidity provider is, by definition, a
trader who posts a limit order at a specified price and quantity on the market and
then waits for a trade to occur; hence, they are considered passive or patient traders.
A liquidity taker, on the other hand, is a trader who submits a market order to accept
a posted limit order, resulting in an immediate trade. Therefore, liquidity takers are
regarded as aggressive or impatient traders (Li et al., 2021; Yao & Ye, 2018). In
most markets, orders are executed based on price-time priority, where a limit order
with the best price is given execution priority, and for the same price, an order with
the earliest submission time is at the head of the queue (O’Hara, 2015; Parlour &
Seppi, 2008). To obtain execution priority over an existing limit order on the market,
a liquidity provider must improve the current price by submitting a limit order with a
better price that is at least one tick above the highest bid price or one tick below the
lowest ask price.
Predominantly, a trade will occur when a limit order placed by a liquidity

provider is matched with a market order sent by a liquidity taker (Li et al., 2021).
Therefore, it is imperative that market conditions are favourable for both parties,
which can be estimated by the bid-ask spread. The limit orders that liquidity
providers place to make market allow them to purchase at the bid price and sell at the
ask price. Given that the bid price is always lower than the ask price, the spread
represents their potential profit and incentive to engage in market-making. In
contrast, the market orders that liquidity takers use to match against existing limit
orders would prompt them to purchase at the ask price and sell at the bid price. Since
their purchase price is higher than their selling price, the spread is treated as a
transaction cost. For this reason, although a larger tick size offers greater trading
incentives that may encourage more engagement from liquidity providers, it also
reflects higher transaction costs that may discourage liquidity takers from
61
participating, resulting in no trades. Therefore, establishing an optimal tick size
requires a trade-off between the trading incentive for liquidity providers and the
transaction costs for liquidity takers (Bacidore, 1997; Bourghelle & Declerck, 2004;
Cordella & Foucault, 1999; Goldstein & Kavajecz, 2000; Harris, 1991, 1996;
Werner et al., 2019).
3.4.2 HFTs as market-makers

In modern stock markets, designated market makers with contractual obligations to
provide liquidity have almost disappeared in modern electronic markets (Clark-
Joseph et al., 2017). With the advent of electronic limit order books, any participant
can provide liquidity through limit order placements, thereby lowering the entry
barrier into the market-making business. This has fostered market-making
competition, resulting in the traditional market makers being overtaken by voluntary
market makers (Foley et al., 2019). Theoretically, voluntary market makers compete
primarily on price and speed (Glosten, 1994; Aït-Sahalia & Sağlam, 2017; Li et al.,
2021). However, in situation when the bid-ask spread is precisely equal to one tick,
there is no price level at which a liquidity provider may obtain execution priority in
the market through price improvement.16 Therefore, the nominal tick size is argued
to be constraining price competition among liquidity providers, forcing them to
compete solely on the basis of the time priority rule.
This market setting, which rewards execution priority to the fastest trader,
incentivises competing participants to engage in a technological arms race to lower
latency (Budish, Cramton & Shim, 2015; Foley, Gorbenko & Ruf, 2019; Yao & Ye,
2018). In addition, studies suggest that the speed advantage of HFTs allows them to
cancel their limit order once it becomes obsolete due to the arrival of new
information, hence avoiding their orders from being adversely selected (Jones, 2013;
Menkveld, 2016). Therefore, HFTs would naturally assume the role of market-
makers due to their reduced operating costs, adverse selection costs, and inventory
costs. When paired with their superior speed, these lower costs should provide them
with a competitive edge in the provision of liquidity for stocks with a greater adverse
16
Using high-frequency data from the Australian stock market, Foley et al. (2019) discover that most
stocks in their sample had bid-ask spreads of exactly one tick, with the average and median stocks
being tick-constrained 85.30% and 92.41% of the time, respectively.
62
selection risk (see, e.g., Aït-Sahalia & Saglam, 2017; Brogaard et al., 2015; Carrion,
2013; Hagströmer & Nordén, 2013; Han et al., 2014; Hoffmann, 2014; Menkveld,
2013).
3.4.2.1 Large tick size and order-queuing activity

A one-tick spread between the bid and ask prices reflects the smallest profit that
liquidity providers could earn by disclosing their private information through limit
orders placement. Thus, a larger relative tick size would correspond to a larger profit
from market-making activity, resulting in an increase in the number of limit orders
placed on the market (Angel, 1997). In a recent study by O’Hara et al. (2019), a
larger relative tick size is found to incentivise HFT market makers to be more
aggressive, such that they would leave limit orders in the book longer and post a
better price than the resting limit orders, resulting in improved prices. Nonetheless,
they also show that this effect varies substantially depending on whether the bid-ask
spread is equal to one tick (i.e. constrained) or multiple ticks (i.e. unconstrained). In
tick-constrained stocks, a larger relative tick size would lead to more depth in the
limit-order book due to order-queuing activity; in tick-unconstrained stocks, it would
result in less depth since the HFTs may undercut the existing queue through price
improvement. This suggests that HFT activity could impose adverse selection costs
on less informed traders, and that they are more inclined to do so when the financial
benefit is larger.
Yao and Ye (2018) show that tick size creates rents for liquidity provision,
and the revenue margins for liquidity provision are higher for stocks with larger
relative tick sizes. Higher rents incentivise HFTs to place more limit orders, which
lengthens the queue to provide liquidity at the best bid and ask prices. This
circumstance discourages market-making HFTs from cancelling limit orders once
they have reached the front of the queue. When the nominal bid-ask spread is
constrained to one tick, non-HFT liquidity providers will have no opportunity to
improve prices, and since they cannot compete with HFTs’ speed, they will be
forced to use market orders, even when doing so is more costly. In short, a large
relative tick size would crowd out the liquidity-providing activities of non-HFT
participants. Hasbrouck and Saar (2013) argue that even though the algorithms
deployed by non-HFTs entailed the use of limit orders, their objective was not to
63
make profits from liquidity provision, but instead to minimise their transaction costs
through, for example, splitting up large orders into smaller fragments and submitting
them to multiple exchanges. Hence, the differing motivations allow non-HFTs to
quote tighter bid-ask spreads than HFTs, since they suffer lower opportunity costs
for placing limit orders (Li et al., 2021).
In addition, Foley et al. (2019) suggest that stocks with a large relative tick
size are likely to be tick-constrained, forcing limit orders to queue at the best bid and
ask prices and resulting in substantial depth at these levels. This discourages HFT
market-makers from cancelling their limit orders, as resubmission would place them
at the end of the queue. As a result, HFTs must accept greater adverse selection risks
to maintain their execution priority, preventing these risks from being transferred to
slower non-HFT market-makers and allowing them to provide liquidity. In summary,
a coarser pricing grid or larger relative tick size creates a more favourable
environment for market-making HFTs to provide liquidity and reduces their adverse
selection costs due to restrictions on price competition and enforcement of time
priority, particularly in tick-constrained stocks.
3.4.2.2 Small tick size and order-undercutting activity

A smaller tick size results in a finer pricing grid, which reduces the cost to undercut
a standing limit order. Chordia et al. (2013) raise concern that HFTs might exploit
this condition by using their superior speed to crowd out liquidity provision from
slower participants. Non-HFT investors may witness a narrower spread due to HFTs’
competition for liquidity provision, however the lower probability of their limit
orders being executed raises questions about whether HFT is actually beneficial to
the market. In reality, non-HFT investors may be forced to take liquidity to execute
their strategy, which unnecessarily increases their overall transaction costs.
Consequently, the general welfare may decline if the benefits of faster speeds are
outweighed by the potential increase in aggregate costs. In a similar vein, O’Hara et
al. (2019) suggest that theoretically, a smaller tick size allows HFT market makers to
impose adverse selection costs on other traders by engaging in order-undercutting
activities. The capacity of HFTs to swiftly monitor price-sensitive public information
enables them to make informed decisions by deliberately undercut the best bid (ask)
limit order if incoming news is anticipated to be positive (negative) for stock prices.
64
This scheme reduces (increases) the likelihood that non-HFT liquidity providers’
resting limit orders will be executed when the change in stock price is favourable
(unfavourable), hence unnecessarily increasing their exposure to adverse selection
risk.
In contrast, Yao and Ye (2018) discover that a smaller tick size encourages
non-HFTs to provide liquidity since they have a greater opportunity to establish
price priority by undercutting HFTs’ existing limit orders. As a consequence of the
competition posed by non-HFTs, Yao and Ye (2019) argue that market-making
HFTs would be exposed to a larger risk of adverse selection, hence reducing their
ability to provide liquidity. Similarly, O’Hara et al. (2019) note that the market-
making HFTs would be forced to cancel their orders more often, resulting in
“fleeting” market liquidity. Moreover, Dyhrberg et al. (2020) suggest that the order-
undercutting behaviour of non-HFTs would pose significant concerns for electronic
market-makers, as their execution priority would be compromised and their ability to
rapidly offload potential inventories would be impeded. As a result, the market-
makers may choose to withdraw from the market, therefore diminishing liquidity and
negatively affecting market quality.
Werner et al. (2019) demonstrate that after a tick size reduction, the limit
orders that had been clustered together at the previous price levels become more
dispersed due to the increased number of price levels. For liquid stocks with tick-
constrained spread, this situation results in in a shorter queue at all price levels,
which in turn encourages investors to provide liquidity rather than taking it.
Nevertheless, in less liquid stocks with tick-unconstrained spreads, a finer pricing
grid might promote aggressive order-undercutting to outbid existing limit orders
with better prices that are economically insignificant. Similarly, Angel et al. (2011)
propose that smaller tick sizes discourage traders from revealing their positions,
especially in the presence of informed or large-uninformed traders whose trading
might impact prices, resulting in a smaller displayed order. In other words, it reduces
the incentive to submit limit orders and makes it easier for traders to abuse standing
limit orders using destructive quote-matching trading strategies.
65
3.4.2.3 Relative tick size trade-off
For liquidity providers, a larger relative tick size represents a higher profit
margin from market-making spread and a greater adverse selection risk from limit
order placements, and vice versa (see e.g.: Angel, 1997; O’Hara et al., 2019; Sandås,
2001). Consequently, the size of a nominal or relative tick would present liquidity
providers with a choice between profit-maximisation, which favours a larger tick
size, and risk-minimisation, which prefers a smaller tick size. This trade-off
fundamentally depends on the risk appetite of the liquidity provider, i.e., whether
they are willing to assume more risk for a higher return (profit maximisation) or are
willing to accept a lower profit to ensure a lower risk exposure (risk minimisation).
Therefore, liquidity-providing HFTs that employ order-queuing activity may be
interpreted as profit-driven traders, while HFTs that prefer order-undercutting can be
regarded as risk-averse traders.
3.4.3 Hypotheses development

The formula for calculating relative tick size relies on two parameters: the stock
price and the nominal tick size. When the nominal tick size remains constant, stocks
with lower prices have larger relative tick sizes, while stocks with higher prices have
smaller relative tick sizes. For instance, in Australia, all stocks priced between
A$0.100 and A$1.995 have a nominal tick size of A$0.005 (half-cent).
Consequently, the relative tick size of stocks at the lowest and highest ends of the
price range would be 5.000% and 0.251%, respectively. The literature suggests that
relative tick size significantly impacts HFTs; more activity in stocks with large
relative tick sizes implies greater order-queuing activity, while more activity in
stocks with small relative tick sizes indicates greater order-undercutting activity.
This study examines whether the substantial disparity in relative tick size at opposite
ends of a price band results in significantly different levels of HFT activity.
Therefore, the study proposes the following hypothesis:
H1: Stocks with larger relative tick sizes will have significantly higher levels of
HFT activity due to greater order-queuing activity compared to stocks with smaller
relative tick sizes.
66
The current tick size structure in the Australian equity market creates two tick
size borders at A$0.10 and A$2.00, with nominal tick size values of A$0.005 (half-
cent) and A$0.01 (one-cent), respectively. These borders represent the lowest price
that can be reached within the same nominal tick size; therefore, stocks priced at
these borders would also have the highest relative tick size value within their
respective group, which is 5.000% and 0.500%, respectively. In contrast, a stock
priced at A$1.995, precisely one tick below the A$2.00 border, would have a relative
tick size value of 0.251%, which is the smallest value attainable for stocks within the
A$0.005 nominal tick size group. As a result, stocks priced just above the border
would have large relative tick sizes, while those priced just below the border would
have small relative tick sizes. This study examines whether the contrast in relative
tick size between stocks priced slightly above and slightly below the A$2.00 tick
size border corresponds to significantly different levels of HFT activity. Based on
these arguments, the study proposes the following hypothesis:
H2: Stocks priced slightly above the A$2.00 tick size border will have
significantly higher levels of HFT activity due to their larger relative tick sizes
compared to stocks priced slightly below the A$2.00 tick size border.
Despite the common perception that stock prices are continuous, they are in
fact discrete due to tick sizes, which form a pricing grid in the market. For instance,
stocks priced above the A$2.00 tick size border would have a pricing grid of A$0.01
(one cent), whereas stocks priced below the border would have a pricing grid of
A$0.005 (half-cent). Stock prices are susceptible to demand and supply forces,
which may cause a stock’s price to fluctuate to the point where it crosses the tick size
border, resulting in a substantial disparity in the crossed stock’s relative tick size
value before and after the crossing occurred. If the stock crossed the border in an
upward direction, the relative tick size would substantially increase; however, if the
stock crossed the border in a downward direction, the relative tick size would
substantially decrease (see Figure 3.5 – 3.7). Therefore, this study examines whether
there is a statistically significant difference in the level of HFT activity before and
after a stock crossed a tick size border in either an upward or downward direction
and posits the following research hypothesis:
67
H3: Crossing a tick size border in an upward (downward) direction results in a
statistically significant increase (decrease) in the level of HFT activity for the
affected stock.
3.5 Methodology
3.5.1 Data description
This research employs order book data starting from January 2008 and ending in
December 2017 for the ASX, and from December 2011 to December 2017 for the
Chi-X Australia, using datasets supplied by the Securities Industry Research Centre
of Asia-Pacific (SIRCA). ASX order book data was derived from the Australian
Equities Tick History (AETH) database prior to May 31, 2016. On June 1, 2016, it
was replaced with ASX ITCH datasets. Chi-X Australia order book information is
provided by the Australian Chi-X Exchange (CHIX). The constituents of the
S&P/ASX 100 (ASX: XTO) index, which includes both large and medium-cap
firms, are used to ensure that the stocks included are sufficiently large and liquid.
This generates almost 9.5 billion rows of order book data over the study period.17
Due to its magnitude, RStudio is used to extract and aggregate relevant information
from the raw dataset in order to create a daily-level dataset. Table 3.2 provides a
summary of the data extracted from the three datasets (AETH, ITCH and CHIX)
used to estimate the key variables in this study, as well as a brief description of the
raw datasets.
17
The list of constituent stocks is based on the information provided by Thomson Reuters Datastream
and updated on a monthly basis.
68
Table 3.2. Relevant information from the datasets
This table presents relevant information on the datasets used in this study, which are provided by SIRCA. Columns AETH, ITCH, and CHIX present the information sourced
from the Australian Equities Tick History, ASX ITCH, and Australian Chi-X Exchange databases, respectively. Period refers to the date range of which the data of the study
is retrieved from. Firms refer to the number of unique firms. Observations refers to the number of firm-day observations. Timestamp refers to the smallest precision unit for
time used to record the messages in the order book. Rows refers to the number of lines in the order book for the specified time period. Size on disc refers to the size of the
dataset used in this study as recorded on the hard drive. Order book messages refers to the information provided in each dataset and how it is defined.
AETH ITCH CHIX
Period 1/1/2008 – 31/5/2016 (1887 days) 1/6/2016 – 31/12/2017 (657 days) 25/10/2011 – 31/12/2017 (1559 days)
Firms 171 113 145
Observations 209,644 39,806 150,673
Timestamp Millisecond (one-thousandth of a second) Nanosecond (one-billionth of a second) Millisecond (one-thousandth of a second)
Rows 5,550,425,473 1,783,656,562 2,237,353,517
Size on disc 739 GB (794,503,270,400 bytes) 217 GB (233,758,228,480 bytes) 232 GB (249,538,936,832 bytes)
Order book Record type: Message type: Message type:

messages  ENTER: Entry of a new order into the order  A: Add order message.  A: Add order message.
book.  D: Order delete message.  E: Order execution message.
 DELETE: Deletion of an order from the order  U: Order replace message.
book.  E: Order executed message. Prices (daily):
 AMEND: Modification of existing order.  C: Order executed message (with price).  Open: The first message recorded as
 TRADE: A trade between two orders. “E”.
Prices (daily):  Close: The last message recorded as
Prices (daily):  Open: The first message recorded as “C”. “E”.
 Open: The first message recorded as “TRADE”.  Close: The last message recorded as “C”.  High: The highest price of message
 Close: The last message recorded as “TRADE”.  High: The highest price of message recorded as “E”.
 High: The highest price of message recorded as recorded as “E” or “C”.  Low: The lowest price of message
“TRADE”.  Low: The lowest price of message recorded recorded as “E”.
 Low: The lowest price of message recorded as as “E” or “C”.
“TRADE”. Other information:
Other information:  Order Reference: Unique order
Other information:  OrderID: The identifier assigned to the new reference number of the day.
 BuyID: A unique order identification assigned to order. The number is only unique per order  Quantity: The visible quantity of shares
buy-side messages. book and side. placed in an order, i.e. volume.
 AskID: A unique order identification assigned to  Quantity: The visible quantity of shares  Value: Price × Quantity
sell-side messages. placed in an order, i.e. volume.
 OrderID: A combination of BuyID and AskID.  Value: Price × Quantity
 Volume: The quantity of shares in an order.
 Value: The dollar value of an order.
69
3.5.2 Sample selection
The first hypothesis states that “stocks with larger relative tick sizes will have
significantly higher levels of HFT activity due to greater order-queuing activity
compared to stocks with smaller relative tick sizes.” To verify this claim, samples are
categorised according to their nominal tick sizes, which are either A$0.005 (half-
cent) or A$0.01 (one-cent).18 Observations in the first category will have prices
between A$0.10 and A$1.995, while observations in the second category will have
prices of A$2.00 or greater. This sample excludes any observations that crossed any
tick size border. This is achieved by ensuring that a stock’s daily low and high prices
fall within the price range applicable to each nominal tick size category. The closing
prices of the stocks in each category are then used to split them into five quintiles,
resulting in groupings with extreme values of relative tick sizes clustered at opposite
extremes of the spectrum. Stocks in the first quintile have the largest relative tick
size (RTSLARGE), whilst the fifth quintile have the smallest relative tick size
(RTSSMALL). In total, four RTS groups and two pairs are defined based on the
nominal and relative tick sizes, which are (i) RTSSMALL_0.005 and RTSLARGE_0.005, and
(ii) RTSSMALL_0.01 and RTSLARGE_0.01. Table 3.3 summarises the data utilised to test
the first hypothesis, categorised by tick size.
Table 3.3. Data description for the samples used to test the first hypothesis
This table describes the data used to test the first hypothesis. Panels A and B show the information
obtained from the ASX and CHIX datasets, respectively. The top and bottom sections of the table
display the data for the A$0.005 and A$0.01 tick size categories, accordingly. RTSSMALL and RTSLARGE
denotes the group of observations with low and high relative tick size values, respectively.
Observations, Highest price, and Lowest price refer to the number of observations, the highest and
lowest closing prices for each RTS group, accordingly.
Panel A: ASX Panel B: CHIX
RTSSMALL RTSLARGE RTSSMALL RTSLARGE
Tick size: A$0.005
Observations 6,022 6,041 2,655 2,669
Highest price (A$) 1.995 0.880 1.995 0.975
Lowest price (A$) 1.510 0.100 1.640 0.240
Tick size: A$0.01

Observations 43,596 43,526 27,383 27,346
Highest price (A$) 199.700 4.400 186.720 4.170
Lowest price (A$) 22.050 2.000 24.400 2.000
18
Due to inadequate sample size, stocks with prices below $0.100 are omitted from all tests
performed in this study.
70
The second hypothesis states that “stocks priced slightly above the A$2.00
tick size border will have significantly higher levels of HFT activity due to their
larger relative tick sizes compared to stocks priced slightly below the A$2.00 tick
size border.” To evaluate this premise, the study divides stock prices into three
categories based on their distance from the tick size border: “FAR,” “MID,” and
“NEAR,” allowing the researcher to examine the level of HFT activity in stocks
trading above and below the tick size border. The first category, “FAR”, contains the
most distant data, with prices ranging from A$1.70 to A$1.795 in the low relative
tick size group (RTSSMALL), and from A$2.21 to A$2.30 in the high relative tick size
group (RTSLARGE). The second category, “MID”, includes observations that sit
between the first and third categories, with prices ranging from A$1.80 to A$1.895
in the RTSSMALL group, and from A$2.11 to A$2.20 in the RTSLARGE group. The
third category, “NEAR”, comprises of observations closest to the tick size border,
with prices ranging from A$1.90 to A$1.995 for the RTS SMALL group, and from
A$2.00 to A$2.10 for the RTSLARGE group. There are a total of three pairs of RTS
groups based on the price ranges categories: (i) RTSSMALL_FAR and RTSLARGE_FAR,
(ii) RTSSMALL_MID and RTSLARGE_MID, and (iii) RTSSMALL_NEAR and RTSLARGE_NEAR.
Table 3.4 presents a summary of the data used to test the second hypothesis for each
price distance category.
Table 3.4. Data description for the samples used to test the second hypothesis
This table describes the data used to test the second hypothesis. Panels A and B show the information
obtained from the ASX and CHIX datasets, respectively. The top, mid, and bottom sections of the
table display the data for FAR, MID, and NEAR categories, respectively. RTSSMALL (RTSLARGE) denotes
the group of observations with low (high) relative tick size values. Observations, Highest price, and
Lowest price refer to the number of observations, the highest and lowest closing prices for each RTS
group, respectively.
RTSSMALL RTSLARGE RTSSMALL RTSLARGE
Distance: FAR
Observations 1,251 777 713 629
Highest price (A$) 1.795 2.300 1.795 2.300
Lowest price (A$) 1.700 2.210 1.700 2.210
Distance: MID
Highest price (A$) 1.895 2.200 1.895 2.200
Lowest price (A$) 1.800 2.110 1.800 2.110
Distance: NEAR
Highest price (A$) 1.995 2.100 1.995 2.100
Lowest price (A$) 1.900 2.000 1.900 2.000
71
The third hypothesis argue that “crossing a tick size border in an upward
(downward) direction results in a statistically significant increase (decrease) in the
level of HFT activity for the affected stock.” To test this hypothesis, the study first
identifies which observations experience tick size crossing events by comparing the
daily low and high prices of each stock to the A$2.00 tick size border. This
assessment will produce one of the following outcomes: (i) both the low and high
prices are less than A$2.00 but greater than or equal to A$0.10; (ii) both the low and
high prices are greater than or equal to A$2.00; or (iii) the low price is less than
A$2.00 but greater than or equal to A$0.10, while the high price is greater than or
equal to A$2.00. All observations that fit the description of the first, second, and
third outcomes are labelled as “small relative tick size” (RTS SMALL), “large relative
tick size” (RTSLARGE), and “tick size crossing event” (CROSS), respectively.
Subsequently, all observations identified as CROSS are screened based on their price
positions relative to the A$2.00 tick border on the day before and after the crossing
events, denoted as PRE_CROSS and POST_CROSS, respectively. Table 3.5 outlines
the process of identifying and evaluating CROSS events to determine whether they
are accepted or rejected.
Table 3.5. Identification and classification of crossing events

This table presents the potential scenarios for tick size crossing events and how they are classified.
CROSS denotes the day on which a crossing event occurred; hence, their relative tick size values are
recorded as ‘Nil’. PRE_CROSS and POST_CROSS refer to the immediate trading days before and
after the occurrence of either a single-day or multiple-day CROSS event. RTSSMALL and RTSLARGE
signify the relative tick size position during PRE_CROSS and POST_CROSS days. Event
classification refers to the decision to classify whether an event is accepted or rejected. An
UPWARDS event starts with RTSSMALL on the PRE_CROSS day and ends with RTSLARGE on the
POST_CROSS day, whereas a DOWNWARDS event starts with RTSLARGE on the PRE_CROSS day
and ends with RTSSMALL on the POST_CROSS day. Events with any combination other than the
aforementioned scenarios are classified as Rejected. Panel A shows the possible combinations
involving a single-day crossing event, while Panel B depicts potential scenarios involving multiple
days of crossing events in a row.
Panel A: Single-day crossing
PRE_CROSS CROSS POST_CROSS Event classification
RTSSMALL Nil RTSLARGE UPWARDS
RTSSMALL Nil RTSSMALL Rejected
RTSLARGE Nil RTSSMALL DOWNWARDS
RTSLARGE Nil RTSLARGE Rejected
Panel B: Multiple-days crossing

PRE_CROSS CROSS CROSS POST_CROSS Event classification
RTSSMALL Nil Nil RTSLARGE UPWARDS
RTSSMALL Nil Nil RTSSMALL Rejected
RTSLARGE Nil Nil RTSSMALL DOWNWARDS
RTSLARGE Nil Nil RTSLARGE Rejected
72
Accepted events are further categorised by their crossing direction, which is
either UPWARDS or DOWNWARDS, depending on their relative tick size’s
position during PRE_CROSS and POST_CROSS days. Table 3.6 summarises the
data selected to test the third hypothesis for each tick size crossing category.
Table 3.6. Data description of the crossing events selected to test the third hypothesis
This table describes the data of the tick size crossing events selected to test the third hypothesis.
Panels A and B show the information obtained from the ASX and CHIX datasets, respectively.
UPWARDS and DOWNWARDS reflect the category of tick size crossing events, as detailed in the
preceding table. Average, Highest, and Lowest prices reflect the mean, maximum, and minimum
closing prices for the PRE_CROSS and POST_CROSS groups, while Average, Highest, and Lowest
differences represent the mean, maximum, and minimum closing prices differences between the
POST_CROSS and PRE_CROSS periods, respectively. Accepted and Rejected events indicate the
number of tick size crossing events that were accepted and rejected in the final sample, respectively.
Total observations shows the number of firm-day observations resulting in tick-size border crossings.
UPWARDS DOWNWARDS UPWARDS DOWNWARDS
PRE_CROSS
Average price (A$) 1.931 2.093 1.960 2.052
Highest price (A$) 1.995 3.740 1.995 2.270
Lowest price (A$) 1.495 2.000 1.865 2.000
POST_CROSS
Average price (A$) 2.085 1.884 2.055 1.902
Highest price (A$) 2.500 1.990 2.210 1.995
Lowest price (A$) 2.000 0.940 2.000 0.980
Difference (POST – PRE)

Average difference (A$) 0.154 -0.210 0.095 -0.150
Highest difference (A$) 0.755 -0.015 0.305 -0.020
Lowest difference (A$) 0.025 -2.270 0.015 -1.170
Duration (POST – PRE)

Average duration 4.309 days 4.617 days 3.459 days 4.333 days
Longest duration 14 days 15 days 7 days 14 days
Shortest duration 2 days 2 days 2 days 2 days
Accepted events 95 111 37 42

Rejected events 370 399 136 155
Total observations 1072 326
3.5.3 Measurement of variables

3.5.3.1 HFT activity measures
Numerous HFT-related research make use of trade-level datasets, such as the
NASDAQ HFT, NASDAQ OMX-St, and TSX trade-level datasets, which offer
some form of identification for each order submitted to the market. In this study’s
datasets, however, such information is not available. Due to this constraint, this
research uses the general characteristics of HFT as outlined by various regulatory
agencies and academics to proxy for HFT activity at daily level. The message-to-
73
trade ratio, the algorithmic trading ratio, the total number of high-frequency orders
(HFO), and the HFO message ratio are selected as proxies because they utilise
information that is commonly available on order-book data. This methodology
enhances the reproducibility of the research, allowing it to be replicated on other
markets and analysed over longer time periods. However, one drawback of this
indirect approach is that the proxies utilised may also capture the activity of other
types of market players who use similar low-latency trading strategies as full-fledged
HFT firms, and thus may not accurately reflect the activity of the latter. Therefore,
any results on “HFT activity” in this research should be taken as reflecting the
activity of “low-latency traders” and not “high-frequency trading firms” per se.
1. Message-to-trade ratio
The message-to-trade ratio (MTR) is calculated by dividing the total number of
ENTER, AMEND, or DELETE order-book messages recorded on a given day by the
total number of trades executed on that day.19 HFTs often places a large number of
limit orders at multiple price levels and revises them whenever new information
becomes available; hence, stocks with a high MTR are commonly linked with a high
level of HFT activity. In addition, a high MTR indicates intensive quotation activity,
a high degree of agility, and a low tolerance for adverse selection risk, all of which
are consistent with the strategy and risk appetite of HFTs. MTR is widely used by
practitioners, regulators, and researchers as a proxy for HFTs’ liquidity providing
activities (see e.g. Aquilina & Ysusi, 2016; ASIC, 2013, 2015; Brogaard et al., 2015;
Friederich & Payne, 2015; Frino et al., 2015; Hagstromer & Norden, 2013).20 The
formula for calculating MTR is shown in Equation 3.1. A higher (lower) MTR value
suggests a higher (lower) degree of HFT activity in stock i on day t, whereas a ratio
of one implies that all orders submitted to the market are traded.
19
“Order-to-trade ratio” (OTR) is a more widely used term than “message-to-trade ratio” (MTR). In
the first essay, the former term is used to reflect this knowledge. However, to be consistent with the
relevant terms found in the study’s datasets, the term MTR is used in the second and third essays of
this study (see Table 3.2).
20
Yao and Ye (2018), on the other hand, argue that the MTR is a poor cross-sectional proxy for HFT
activity. In contrast to popular belief, they discovered that the presence of HFTs has a negative
relationship with MTR values.
74
∑ , 𝑀𝑒𝑠𝑠𝑎𝑔𝑒 ,
𝑀𝑇𝑅 , =
∑ , 𝑇𝑟𝑎𝑑𝑒 ,
(Equation 3.1)
Where ∑ , 𝑀𝑒𝑠𝑠𝑎𝑔𝑒 , is the total number of ENTER, AMEND, and DELETE

messages recorded on stock i on day t,21 and ∑ , 𝑇𝑟𝑎𝑑𝑒 , is the total number of
TRADE messages recorded on stock i on day t.22 For the sake of brevity, the in-text
explanation is based solely on information from the AETH dataset.
2. Algorithmic trading ratio

The datasets used for this study do not disclose whether an order was placed by a
trading bot or a human. Consequently, it is impossible to monitor algorithmic trading
activities directly through the order book. Hendershott et al. (2011) use the rate of
electronic message traffic as a proxy for the amount of algorithmic trading (AT), and
asserted that its variations are predominantly driven by the variations of limit order
submissions and cancellations; therefore, the measure should primarily capture
algorithmic liquidity provision. This also suggests that AT is less likely to reflect
algorithmic arbitrage activity, since this liquidity-taking action uses market orders
and not limit orders. Even though AT cannot accurately quantify HFT activity, HFT
remains a subset of AT, therefore it can still be employed to represent HFT’s general
attributes. Following the approach of Hendershott et al. (2011), this study utilises
algorithmic trading ratio (ALGO) as a proxy for HFT activity, defined as the sum of
daily trading value (in dollars) normalised by the total number of messages recorded
as ENTER, AMEND, or DELETE, multiplied by negative one. Higher (lower) ALGO
values suggest more (less) algorithmic trading activity. Equation 3.2 is the formula
used to calculate ALGO.
∑ 𝑇𝑟𝑎𝑑𝑖𝑛𝑔 𝑉𝑎𝑙𝑢𝑒 ,
𝐴𝑇𝑅 , = × (−1)
∑ , 𝑀𝑒𝑠𝑠𝑎𝑔𝑒 ,
(Equation 3.2)
21
Message: AETH = ENTER, AMEND, DELETE; ITCH = A, U, D; CHIX = A
22
Trade: AETH = TRADE; ITCH = E, C; CHIX = E
75
Where ∑ 𝑇𝑟𝑎𝑑𝑖𝑛𝑔 𝑉𝑎𝑙𝑢𝑒 , is the sum of all trading value on stock i on day t, and
∑ , 𝑀𝑒𝑠𝑠𝑎𝑔𝑒 , is the total number of messages recorded as ENTER, AMEND, and

DELETE on stock i on day t.23
3. High-frequency orders
HFT is distinguishable from other algorithmic trading activities and the rest of the
market mostly by its speed, making it a unique form of market participant. The
primary activity of HFTs is market-making, which involves rapid quote updates in
response to new market information, resulting in a large number of linked
messages. Hasbrouck and Saar (2013) and Boehmer, Li, and Saar (2018) utilise
extended strategic runs of at least ten linked messages to quantify HFT activity, and
such runs are often seen in dynamic algorithmic strategies that often contain several
limit order amendments and resubmissions. As noted by Hasbrouck and Saar (2013),
“While the 10-message cut off is somewhat arbitrary, these runs represent more than
half of the total number of messages that are linked to runs in each sample period…
Such longer runs characterise much low-latency activity” (p. 659).
Additionally, if HFTs’ algorithm detects the threat of being adversely

selected or attempts to capitalise on profitable yet transitory trading opportunities, it
will withdraw or rapidly update its existing limit orders, resulting in a short order
resting time. Due to their speed, HFTs should naturally have a shorter average order
resting time than non-HFTs. Subrahmanyam and Zheng (2016) discover that for
large, medium, and small-cap firms, the average survival times of HFTs’ orders (i.e.,
order duration) inside the top three price levels of the limit order book are 28.74,
33.03, and 51.89 seconds, respectively. Therefore, a limit order sent by HFTs that
consists of 10 messages should have a duration of 30 seconds or less, which is
equivalent to an average order resting time of three seconds. 24
23
Trading value: AETH = Value when Record Type is equal to TRADE; ITCH = (Price × Quantity)
when Message Type is equal to E or C; CHIX = (Price × Quantity) when Message Type is equal to E.
24
This is due to the fact that only large and medium-sized firms were included in the study’s sample.
76
The datasets used in this study include a unique order identification
(OrderID) for every order submitted to the market (see Table 3.2). 25 This allows the
researcher to monitor the entire history of each OrderID from the moment it enters
the market until it is withdrawn from the order book by cancellation or execution. 26
Based on the aforementioned criteria, OrderIDs submitting at least 10 messages with
an average order resting time of less than three seconds are most likely generated by
HFTs.27 The formula used to identify the high-frequency order (HFO) is shown in
Equation 3.3.
𝑂𝑟𝑑𝑒𝑟𝐷𝑢𝑟𝑎𝑡𝑖𝑜𝑛 , , = 𝑂𝑟𝑑𝑒𝑟𝐼𝐷 , , , ( ) − 𝑂𝑟𝑑𝑒𝑟𝐼𝐷 , , , ( )
𝑂𝑟𝑑𝑒𝑟𝐷𝑢𝑟𝑎𝑡𝑖𝑜𝑛 , ,
𝑂𝑅𝑇 , , = ,,
∑ 𝑀𝑒𝑠𝑠𝑎𝑔𝑒𝑠 ,,
,, ,,
∑ , 𝐻𝐹𝑂 , = ∑ 𝑂𝑅𝑇 , , ≤ 3 seconds & ∑ 𝑀𝑒𝑠𝑠𝑎𝑔𝑒 ,, ≥ 10 messages
(Equation 3.3)
Where 𝑂𝑟𝑑𝑒𝑟𝐼𝐷 , , , ( ) is the OrderID j on stock i on day t with the highest

timestamp (d); 𝑂𝑟𝑑𝑒𝑟𝐼𝐷 , , , ( ) is the OrderID j on stock i on day t with the
lowest timestamp (d); 𝑂𝑟𝑑𝑒𝑟𝐷𝑢𝑟𝑎𝑡𝑖𝑜𝑛 , , is the difference between the highest and
,,
lowest timestamps of OrderID j on stock i on day t; ∑ 𝑀𝑒𝑠𝑠𝑎𝑔𝑒 ,, is the total
number of messages generated by OrderID j on stock i on day t; 𝑂𝑅𝑇 , , is the order
resting time of OrderID j on stock i on day t; and ∑ , 𝐻𝐹𝑂 , is the total number of
OrderID on stock i on day t that satisfy the following criteria: (i) order resting time
𝑂𝑅𝑇 , , of less than or equal to 3 seconds, and (ii) number of messages
25
Order identification: AETH = OrderID; ITCH = OrderID; CHIX = Order Reference
26
This includes everything in between, such as order modification and/or incomplete execution (if
any). When an order is modified, it may lose its queue position in the limit order book if: (1) its price
is changed; and (2) its modified quantity is greater than the original amount. In these cases, the order
is pushed to the back of the queue.
27
This study removes any HFO with an average order resting time of less than five milliseconds.
Despite the fact that this number is arbitrary, it is necessary to include a minimum limit in the HFO
calculation to prevent misidentification due to the order book’s underlying mechanics. For example,
when a large limit order to buy is placed in the market and then traded against many small market
orders to sell, several AMEND messages are sent to reflect the remaining balance of the large order
based on the number of transactions that had occurred. Consequently, the total number of messages
recorded in the order book is overstated, resulting in an inaccurate estimation of HFT activity using
the HFO method.
77
,,
∑ 𝑀𝑒𝑠𝑠𝑎𝑔𝑒 ,, of more than or equal to 10. All of the parameters in Equation
3.3 are derived from messages recorded as ENTER, AMEND, and DELETE.
4. HFO message ratio

The identification of HFO allows the study to estimate the daily total number of
messages produced by this type of OrderID. With this information, it is possible to
determine the high-frequency order ratio (HFOR), which represents the proportion of
messages generated by HFO in the market. The formula to calculate HFOR is
expressed by Equation 3.4.
,,
∑ 𝑀𝑒𝑠𝑠𝑎𝑔𝑒 ,,
𝐻𝐹𝑂𝑅 , = ,
∑ 𝑀𝑒𝑠𝑠𝑎𝑔𝑒 ,
(Equation 3.4)
,,
Where ∑ 𝑀𝑒𝑠𝑠𝑎𝑔𝑒 ,, is the total number of messages generated by
OrderID identified as HFO on stock i on day t; and ∑ , 𝑀𝑒𝑠𝑠𝑎𝑔𝑒 , is the total
number of all messages generated on stock i on day t.
3.5.3.2 Explanatory variables

1. Independent variable
The main purpose of this research is to determine whether relative tick size has an
effect on HFT activity. Relative tick size is calculated as nominal tick size
𝑇𝑖𝑐𝑘 𝑆𝑖𝑧𝑒 , divided by closing price 𝐶𝑙𝑜𝑠𝑒 , , as shown in Equation 3.5. As
explained in the previous sections, specific sample groups are selected to represent
observations with differing relative tick size values, namely RTS SMALL and
RTSLARGE. To capture this effect, a dummy variable with the value one is assigned to
observations in the RTSSMALL group, and zero otherwise.
𝑇𝑖𝑐𝑘 𝑆𝑖𝑧𝑒 ,
𝑅𝑇𝑆 , =
𝐶𝑙𝑜𝑠𝑒 ,
(Equation 3.5)
78
2. Control variables
The study controls for several characteristics identified in the literature as having a
significant effect on HFT activity, including volatility, liquidity, and firm size. Studies
show that stocks with greater volatility have higher HFT activity; hence, HFT activity
is expected to have a positive association with volatility, which is proxied in this study
by the daily trading range (see e.g., Boehmer et al., 2020; Ersan & Ekinci, 2016;
Hasbrouck & Saar, 2013; Lee, 2015; Manahov, 2016; Yilmaz et al., 2015). As shown
in Equation 3.6, the volatility measure is calculated by dividing the difference between
the daily high 𝐻𝑖𝑔ℎ , and low 𝐿𝑜𝑤 , prices by their average.
𝐻𝑖𝑔ℎ , − 𝐿𝑜𝑤 ,
𝑉𝑂𝐿𝐴𝑇𝐼𝐿𝐼𝑇𝑌 , =
𝐻𝑖𝑔ℎ , + 𝐿𝑜𝑤 ,
2
(Equation 3.6)
In addition, research shows that HFT activity is more prevalent in stocks that
are highly liquid and have a large market capitalisation (Brogaard, Hendershott &
Riordan, 2014; Bhattacharya et al., 2020; Ersan & Ekinci, 2016). Therefore, HFT
activity should be positively related to both liquidity and firm size. The spread
measure developed by Corwin and Schultz (2012) is used as a proxy for liquidity in
this study, and its formula is shown in Equation 3.7. According to the formula,
negative spread values are conceivable; hence, the authors recommend substituting
these values with zero. The size effect is proxied by the natural logarithm of a firm’s
market capitalisation, as depicted by Equation 3.8.
2(𝑒 , − 1)
𝐿𝐼𝑄𝑈𝐼𝐷𝐼𝑇𝑌 , =
1+𝑒 ,
2𝛽 , − 𝛽 , 𝛾,
𝛼, = −
3 − 2√2 3 − 2√2
𝐻𝑖𝑔ℎ , 𝐻𝑖𝑔ℎ ,
𝛽 , = ln + ln
𝐿𝑜𝑤 , 𝐿𝑜𝑤 ,
𝑚𝑎𝑥 𝐻𝑖𝑔ℎ , , 𝐻𝑖𝑔ℎ ,

𝛾 , = ln
𝑚𝑖𝑛 𝐿𝑜𝑤 , , 𝐿𝑜𝑤 ,
(Equation 3.7)
79
𝑆𝐼𝑍𝐸 , = ln 𝑀𝑎𝑟𝑘𝑒𝑡 𝐶𝑎𝑝𝑖𝑡𝑎𝑙𝑖𝑠𝑎𝑡𝑖𝑜𝑛 ,
(Equation 3.8)
3.5.4 Model specification

The first and second hypotheses are examined using univariate analysis, which
compares the mean HFT activity between the RTSSMALL and RTSLARGE groups using
a parametric test (t-test). Literature indicates that other variables may potentially
influence HFT activity, and a simple univariate test cannot account for these
additional factors. Consequently, the study employs multiple regression analysis to
provide evidence about the effect of relative tick size on HFT activity, while also
controlling for other firm-specific factors such as volatility, liquidity, and firm size.
The model used to evaluate the first and second hypotheses is represented by
Equation 3.9.
𝐻𝐹𝑇 , = 𝛽 + 𝛽 𝐷𝑆𝑀𝐴𝐿𝐿 , + 𝛽 𝐶𝑂𝑁𝑇𝑅𝑂𝐿 , +𝜀,

(Equation 3.9)
DSMALL is a dummy variable that equals one if the observation of firm i on day t
belongs to the RTSSMALL group and zero otherwise.28 The regression analysis would
yield one of the following results: (i) not significant – HFT activity is unaffected by
relative tick size; (ii) significantly positive – HFT executed more order-undercutting
activity when the tick size is small; or (iii) significantly negative – HFTs’
engagement is reduced when the tick size is small. 𝐶𝑂𝑁𝑇𝑅𝑂𝐿 , represents the
three control variables used in the model, namely VOLATILITY, LIQUIDITY and
SIZE of firm i on day t-1, as proxied by the daily trading range, the Corwin-Schultz
spread, and the natural logarithm of market capitalisation, respectively. Lagged
values are applied to the control variables to avoid the potential of having reverse
causality issue. In addition, firms (cross-sectional unit) and day (time unit) fixed
effects are used to account for omitted variables bias, which may result from
unobservable factors that vary across firms and time.
28
As described in the preceding section, the first and second hypotheses are distinguished by their
methods for defining the RTSSMALL and RTSLARGE groups, resulting in a unique set of samples for
each hypothesis. However, they analyse their respective samples using similar univariate and
multivariate tests.
80
To test the third hypothesis, the study employs univariate and multivariate
difference-in-difference techniques. The tick size crossing event is identified as the
event that would cause an exogenous shock to relative tick size values while having
no direct impact on HFT activity. Therefore, any differences in HFT activity seen on
the days before (PRE_CROSS) and after (POST_CROSS) a crossing event should be
related to the relative tick size shift that occurred as a result of the crossing event. A
tick size crossing event, as defined in Table 3.5, is classified as either UPWARDS or
DOWNWARDS based on its RTS position before and after the event. Due to the
contradiction between these two categories, it would be inappropriate to include all
tick crossing size events in a single analysis, as doing so would obscure the actual
effects of relative tick size on HFT activity, leading to a false conclusion. Thus, the
UPWARDS and DOWNWARDS events are analysed independently.
In addition, the difference-in-difference method requires the construction of

two sample groups, named “TREATMENT” and “CONTROL”. TREATMENT
comprises of all tick size crossing events accepted based on the procedure outlined in
Table 3.5, which will belong to either UPWARDS or DOWNWARDS category.
CONTROL, on the other hand, consists of samples that are matched against each
event identified in the TREATMENT group. To be selected, the following
conditions must be satisfied: (i) the matched sample does not belong to the
TREATMENT group; and (ii) on both PRE_CROSS and POST_CROSS days, the
daily low and high prices of the matched sample must lie between A$1.00 and
A$1.995 for the UPWARDS category, and between A$2.00 and A$3.99 for the
DOWNWARDS category.29 The observations of each sample in the TREATMENT
and CONTROL groups are further separated into the PRE_CROSS or
POST_CROSS groups, depending on their timing relative to the event. The research
designs for the UPWARDS and DOWNWARDS groups are illustrated in Figures
3.8 and 3.9, respectively.
29
The highest relative tick size value for stocks with a nominal tick size of A$0.01 is 0.50% at
A$2.00, which is the same as stocks priced at A$1.00 with a nominal tick size of A$0.005. Similarly,
the lowest relative tick size value for stocks with a nominal tick size of A$0.005 is 0.2506% at
A$1.995, which is the same as stocks priced at A$3.99 with a nominal tick size of A$0.01. The price
ranges stated in the main text for RTSSMALL and RTSLARGE are decided based on these price and
relative tick size limits.
81
PRE_CROSS POST_CROSS
(1) (3)
CONTROL
CONTROLPRE_CROSS CONTROLPOST_CROSS
RTSCONTROL = RTSLOW RTSCONTROL = RTSLOW
(2) (4)
TREATMENT
TREATMENTPRE_CROSS TREATMENTPOST_CROSS
RTSTREATMENT = RTSLOW RTSTREATMENT = RTSHIGH
Figure 3.8: Illustration of the research design used to test the third hypothesis
(UPWARDS)
PRE_CROSS POST_CROSS
(1) (3)
CONTROL
CONTROLPRE_CROSS CONTROLPOST_CROSS
RTSCONTROL = RTSHIGH RTSCONTROL = RTSHIGH
(2) (4)
TREATMENT
TREATMENTPRE_CROSS TREATMENTPOST_CROSS
RTSTREATMENT = RTSHIGH RTSTREATMENT = RTSLOW
Figure 3.9: Illustration of the research design used to test the third hypothesis
(DOWNWARDS)
The top and bottom sections of the figures represent the CONTROL and
TREATMENT groups, whereas the left and right panels reflecting the PRE_CROSS
and POST_CROSS days, respectively. Quadrants (1) and (2) reflect the data for
CONTROL and TREATMENT data before the crossing event, while Quadrants (3)
82
and (4) indicate the data for CONTROL and TREATMENT after the crossing event,
respectively. The subsequent steps are performed to evaluate whether HFT activity
differs across different levels of RTS:
i. (2) – (1) = Difference between TREATMENT and CONTROL before the tick
size crossing event occurred (DIFFPRE_CROSS)
ii. (4) – (3) = Difference between TREATMENT and CONTROL after the tick size
crossing event occurred (DIFFPOST_CROSS)
iii. (DIFFPOST_CROSS) – (DIFFPRE_CROSS) = Difference due to RTS change (DIFFPOST-PRE)
The multivariate regression model used in the difference-in-difference analysis are

shown in Equation 3.10.
𝐻𝐹𝑇 , = 𝛽 + 𝛽 𝐷𝑇𝑅𝐸𝐴𝑇𝑀𝐸𝑁𝑇 , + 𝛽 𝐷𝑃𝑂𝑆𝑇 , + 𝛽 𝐷𝑇𝑅𝐸𝐴𝑇𝑀𝐸𝑁𝑇 , ×

𝐷𝑃𝑂𝑆𝑇 , + 𝛽 𝐶𝑂𝑁𝑇𝑅𝑂𝐿 , +𝜀,
(Equation 3.10)
𝐷𝑇𝑅𝐸𝐴𝑇𝑀𝐸𝑁𝑇 , represents a dummy variable that equals one if the observation of

firm i on day t belongs to the Treatment group, and zero otherwise; 𝐷𝑃𝑂𝑆𝑇 , is a
dummy variable that equals one if the observation of firm i on day t belongs to the
POST_CROSS group, and zero otherwise; and 𝐷𝑇𝑅𝐸𝐴𝑇𝑀𝐸𝑁𝑇 , × 𝐷𝑃𝑂𝑆𝑇 ,
estimates the interaction term between dummy variable for POST_CROSS and
TREATMENT groups. The interpretation of the control variables is the same as in
Equation 3.9. The model also incorporates the firms and time (day) fixed-effects to
account for any bias caused by omitted variables.
3.6 Results, analyses, and discussions

This section presents the study’s findings and analysis in accordance with the
sequence of the hypotheses. To summarise, the first hypothesis examines HFT
activity among opposing RTS groups that have the same nominal tick size. The
second hypothesis compares HFT activity in stocks priced just above and below the
A$2.00 tick size border. The third hypothesis explores the impact of crossing a tick
size border that results in substantial changes in relative tick size on HFT activity.
83
This section concludes with a discussion of the outcomes obtained from the testing
the three hypotheses.
3.6.1 HFT activity across different RTS groups categorised by nominal tick sizes
3.6.1.1 Descriptive statistics
Table 3.7 describes the dataset used to test the first hypothesis, which examines the
impact of relative tick size on HFT activity, categorised by nominal tick size of
either half-cent (Panel I) or one-cent (Panel II). 30 This is performed by grouping
observations inside each nominal tick size category according to their relative tick
size values, resulting in four subgroups, namely RTSSMALL_0.005, RTSLARGE_0.005,
RTSSMALL_0.01, and RTSLARGE_0.01. The table displays a greater number of
observations for the ASX and the one-cent category than for the CHIX and the half-
cent category, respectively. These figures reflect the fact that the ASX has greater
trading activity than the CHIX, and that more stocks are traded using a one-cent
nominal tick size than a half-cent nominal tick size. The PRICE and RTS values
shown for each subgroup on both marketplaces reflect the conditions specified when
the groups are constructed, as outlined in Section 3.5.2.
The table depicts that, on average, the samples in the RTSSMALL_0.005 and
RTSSMALL_0.01 groups have a lower VOLATILITY than their RTS LARGE counterparts,
and that the ASX are more volatile than the CHIX. Similar trend is seen in
LIQUIDITY, where samples from RTSSMALL groups have, on average, wider spreads
than those from RTSLARGE groups, and the ASX’s spread is consistently wider than
the CHIX’s. In terms of SIZE, the sample in the RTS SMALL groups has a significantly
greater market capitalisation than the sample in the RTSLARGE groups. In the half-
cent tick size category, the average firm size of RTS SMALL’s sample is almost twice
that of RTSLARGE, and in the one-cent tick size category, it is nearly seven times
greater. In addition, the data shows that the average market capitalisation of the ASX
and CHIX markets across all subgroups examined is comparable.
30
As noted in Section 3.2.2, it is not possible to test stocks with $0.001 (one-tenth of once-cent) due
to the limited number of observations available at this nominal tick size group.
84
Table 3.7. Descriptive statistics of the data used to test Hypothesis 1
This table describes the data employed to test the first hypothesis, which are categorised by nominal tick sizes of either A$0.005 (Panel I) or A$0.01 (Panel II). Panels A and
B represent the ASX and CHIX datasets, while RTSSMALL and RTSLARGE refer to the observations in the fifth and first percentile based on their relative tick size values,
respectively. N, Mean, Std. Dev., Minimum, and Maximum refer to the number of observations, average, standard deviation, lowest value, and highest value, accordingly.
PRICE refer to the closing price; RTS indicates the relative tick size, which is calculated by dividing nominal tick size by closing price (Equation 3.5); VOLATILITY is
measured by difference between the highest and lowest price of the day, divided their average prices (Equation 3.6); LIQUIDITY is measured using the Corwin and Schultz
(2012) high-low spread (Equation 3.7); and SIZE is the actual dollar value of market capitalisation.
N Mean Std. Dev. Minimum Maximum N Mean Std. Dev. Minimum Maximum
Panel I: Tick size = A$0.005 (half-cent)
RTSSMALL (5th quintile)
PRICE ($) 6,022 1.822 0.086 1.510 1.995 2,655 1.834 0.080 1.640 1.995
RTS (%) 6,022 0.275 0.013 0.251 0.331 2,655 0.273 0.012 0.251 0.305
VOLATILITY (%) 6,022 2.627 1.917 0.289 32.749 2,655 2.084 1.305 0.000 14.934
LIQUIDITY (%) 6,012 0.626 0.892 0.000 9.845 2,645 0.426 0.641 0.000 7.630
SIZE ($ billion) 6,022 3.817 1.915 0.339 10.590 2,655 3.953 2.288 0.779 10.590
RTSLARGE (1st quintile)

PRICE ($) 6,041 0.558 0.172 0.100 0.880 2,669 0.712 0.190 0.240 0.975
RTS (%) 6,041 1.041 0.553 0.568 5.000 2,669 0.776 0.290 0.513 2.083
VOLATILITY (%) 6,041 5.232 5.109 0.612 73.684 2,669 3.068 2.359 0.000 50.932
LIQUIDITY (%) 6,020 1.205 1.902 0.000 26.430 2,651 0.606 0.967 0.000 8.142
SIZE ($ billion) 6,041 1.893 1.381 0.056 6.355 2,669 1.830 1.062 0.243 5.570
Panel II: Tick size = A$0.01 (one-cent)
RTSSMALL (5th quintile)
PRICE ($) 43,596 45.555 23.137 22.050 199.700 27,383 48.625 23.948 24.400 186.720
RTS (%) 43,596 0.026 0.009 0.005 0.045 27,383 0.024 0.009 0.005 0.041
VOLATILITY (%) 43,596 2.050 2.076 0.000 162.557 27,383 1.543 0.906 0.000 24.063
LIQUIDITY (%) 43,545 0.412 0.632 0.000 14.485 27,171 0.293 0.449 0.000 6.480
SIZE ($ billion) 43,596 34.530 36.530 1.035 166.000 27,383 37.120 38.440 1.454 156.000
85
Table 3.7 (continue)
RTSLARGE (1st quintile)
PRICE ($) 43,526 3.007 0.528 2.000 4.400 27,346 3.085 0.569 2.000 4.170
RTS (%) 43,526 0.344 0.064 0.227 0.500 27,346 0.336 0.066 0.240 0.500
VOLATILITY (%) 43,526 2.760 1.914 0.277 67.410 27,346 2.037 1.340 0.000 31.237
LIQUIDITY (%) 43,460 0.634 0.892 0.000 24.975 27,180 0.427 0.630 0.000 10.422
SIZE ($ billion) 43,526 5.103 6.219 0.393 50.150 27,346 5.116 5.529 0.757 50.640
86
Overall, these values indicate that stocks with higher prices (i.e., RTS SMALL)
have smaller intraday price ranges, narrower spreads, and larger market
capitalisation than those with lower prices (i.e., RTSLARGE). The figures also imply
that the activity of ASX market participants resulted in greater price fluctuations
(VOLATILITY) and wider spreads (LIQUIDITY) than those exhibited on CHIX,
which is likely due to the presence of noise traders (i.e., retail investors) on the
former and their absence on the latter. This argument is plausible considering that
CHIX is regarded as a trading venue specifically designed to meet HFTs’
technological requirements, which demands a low-latency trading environment
(Chordia et al., 2013; Menkveld, 2013). Moreover, Malceniece et al. (2019) portray
CHIX’s entry to a new market as an exogenous instrument which reflects the onset
of HFT in a market, whereas Foley et al. (2019) characterise slow and fast brokers
based on their participation on the CHIX. For these reasons, it is apparent that the
CHIX is not a trading platform where retail traders and unsophisticated investors
would participate.
3.6.1.2 Univariate analysis

Table 3.8 illustrates the results of univariate tests performed to compare HFT
activity, as proxied by MTR, ALGO, HFO, and HFOR, between two opposing
relative tick size groups (RTSSMALL and RTSLARGE), for each nominal tick size
category of either half-cent (Panel I) or one-cent (Panel II). The mean values of HFT
proxies are contrasted between the RTSSMALL and RTSLARGE groups for each tick
size category using a two-sample t-test. The table depicts that there is a statistically
significant difference between the RTSSMALL and RTSLARGE groups with regards to
HFT activity, with the former exhibiting greater MTR, ALGO, HFO, and HFOR
values. The exception to this is ALGO in the ASX market’s one-cent category,
which has a positive but insignificant result. This evidence suggests that HFT
activity is more prevalent in stocks with smaller relative tick sizes, with the one-cent
tick size category showing the largest disparity.
Furthermore, the mean values show that the CHIX has more intensive order
quotation activity (i.e., MTR) and algorithmic liquidity provision (i.e., ALGO) than
the ASX. This behaviour is also shown in the average differences between RTS SMALL
and RTSLARGE for those variables, with the CHIX exhibiting larger disparities than
87
the ASX. The statistics for HFO and HFOR, however, suggest the contrary; the ASX
has a greater number of fast OrderIDs (i.e., HFO) and a larger order-book presence
(i.e., HFOR) than the CHIX. This pattern is also evident in the size of the average
differences between RTSSMALL and RTSLARGE for HFO and HFOR, with the ASX
showing more pronounced differences than the CHIX.
Table 3.8. Mean comparison of HFT activity categorised by nominal tick sizes
This table displays the results of a univariate analysis using observations classified by their nominal
tick sizes to test the first hypothesis. The findings from the A$0.005 and A$0.01 categories are shown
in Panels I and II, respectively. The results for ASX and CHIX are shown in the upper and lower
portions of each panel, correspondingly. Obs. and Mean refers to the number of observations and the
average values of each HFT measure, accordingly. Difference represents the difference between the
mean values of RTSSMALL and RTSLARGE. Message-to-trade ratio (MTR), average trade size (ALGO),
high-frequency orders (HFO), and HFO-contributed message ratio (HFOR) are the variables used as
proxies for measuring HFT activity. The formula are illustrated in Equations 3.1, 3.2, 3.3, and 3.4,
respectively. Data are winsorised at three standard deviations (3-sigma) from their respective means.
Panel I: RTSSMALL (5th quintile) RTSLARGE (1st quintile) Difference
(TS = A$0.005) Obs. Mean Obs. Mean (RTSSMALL – RTSLARGE)
ASX
MTR (%) 6,022 503.189 6,041 449.623 53.566***
ALGO 6,022 -13.770 6,041 -17.300 3.531***
HFO (Ln) 6,022 2.323 6,041 1.404 0.919***
HFOR (%) 6,022 3.269 6,041 2.439 0.83***
CHIX
MTR (%) 2,655 746.922 2,669 651.099 95.823***
ALGO 2,655 -4.775 2,669 -5.785 1.01***
HFO (Ln) 2,655 0.136 2,669 0.049 0.087***
HFOR (%) 2,655 0.169 2,669 0.123 0.046**
Panel II: RTSSMALL (5th quintile) RTSLARGE (1st quintile) Difference

(TS = A$0.01) Obs. Mean Obs. Mean (RTSSMALL – RTSLARGE)
ASX
MTR (%) 43,596 612.243 43,526 511.546 100.697***
ALGO 43,596 -16.362 43,526 -16.389 0.028
HFO (Ln) 43,596 5.129 43,526 2.450 2.679***
HFOR (%) 43,596 17.179 43,526 3.532 13.647***
CHIX
MTR (%) 27,383 1733.621 27,346 773.160 960.46***
ALGO 27,383 -2.719 27,346 -6.576 3.857***
HFO (Ln) 27,383 2.120 27,346 0.163 1.959***
HFOR (%) 27,383 2.488 27,346 0.267 2.221***
*** p<0.01, ** p<0.05, * p<0.10
Nevertheless, it is possible that the reported findings from the univariate test
are influenced by the differences found at firm level characteristics between stock
composing the RTSSMALL and RTSLARGE groups (see Table 3.7), rather than explicitly
driven by the small relative tick size factor. For this reason, a multivariate regression
model is used to further analyse the data before any definitive conclusions are drawn.
88
3.6.1.3 Multivariate analysis
Table 3.9 displays the results of a multiple regression analysis to examine the
influence of a small relative tick size value on HFT activity, categorised by nominal
tick size value of either half-cent or one-cent. In Panel I, the regression outcomes for
the half-cent tick size category indicate that stocks with a small relative tick size
result in a significantly more HFO (β = 0.484) on the ASX, but it has no explanatory
power on other HFT measures. On the other hand, the results for CHIX show that
relative tick size has no significant effect on all HFT measures employed. These
findings suggest that the significant differences in HFT activity between the
RTSSMALL and RTSLARGE groups observed in the preceding univariate tests are most
probably due to other factors, and not the relative tick size itself. With the exception
of the MTR on the CHIX, which contains no significant explanatory variable, all
HFT measures on the ASX and CHIX are found to be significantly influenced by at
least one of the control variables employed.
Panel II exhibits the regression results for the one-cent tick size category. The
findings show that stocks with a small relative tick size have significantly higher
HFT activity across all measures analysed on the ASX (MTR: β = 135.552; ALGO:
β = 11.870; HFO: β = 1.802; HFOR: β = 6.815), whereas for the CHIX, it leads to
significantly higher MTR (β = 2956.856) and HFO (β = 1.158) only. These
results imply that the significant difference in HFT activity between the RTS SMALL
and RTSLARGE groups observed in the earlier univariate tests is indeed driven by
relative tick size values, and this holds true even after adjusting for other factors
including volatility, liquidity, size, and cross-sectional (firm) and time (day) fixed
effects, lending credence to the reported findings.
The results of the regression analysis corroborate the undercutting

hypothesis, which proposes that HFTs are more engaged in stocks with a smaller
relative tick size, since they may use their speed advantage to obtain execution
priority. This is clearly apparent in the one cent tick size category, but almost
entirely insignificant in the half cent tick size category. This situation may be better
explained by examining the relative tick size values shown in Table 3.7, which
describes the dataset used for the regression models.
89
Table 3.9. Regression analysis on HFT activity categorised by nominal tick sizes
This table shows the results of multivariate regression analysis using observations categorised by their nominal tick sizes to test the first hypothesis. The findings from the
A$0.005 and A$0.01 categories are shown in Panels I and II, respectively. Panels A and B illustrate the findings using the ASX and CHIX datasets respectively. Dependent
variables are the HFT activity measures, namely message-to-trade ratio (MTR), average trade size (ALGO), high-frequency orders (HFO), and HFO-contributed message ratio
(HFOR). The formula are shown in Equations 3.1, 3.2, 3.3, and 3.4, respectively. The independent variable is DSMALL, which is a dummy variable assigned with a value of
one if the observation belongs to the group with low relative tick size (RTSSMALL), and zero otherwise. The control variables are the one-day lagged (t – 1) values of
VOLATILITY, LIQUIDITY, and SIZE. VOLATILITY is measured by difference between the highest and lowest price of the day, divided their average prices (Equation 3.6);
LIQUIDITY is measured using the Corwin and Schultz (2012) high-low spread (Equation 3.7); and SIZE is the natural log of market capitalisation, respectively. All models
are controlled for firm and day fixed-effects. Data are winsorised at three standard deviations (3-sigma) from their respective means. The standard errors are calculated using
the Huber/White/sandwich estimator of variance method.
MTR ALGO HFO HFOR MTR ALGO HFO HFOR
Panel I: Tick size = A$0.005 (half-cent)
DSMALL -9.296 2.514 0.484*** 0.134 94.927 -0.780 0.016 -0.102
(18.474) (2.463) (0.144) (0.427) (147.373) (0.940) (0.035) (0.098)
VOLATILITY -2.496* -0.243** 0.039*** 0.064*** -10.525 -0.115*** 0.005 -0.008
(1.343) (0.112) (0.007) (0.019) (10.574) (0.041) (0.004) (0.007)
LIQUIDITY 1.372 0.386*** -0.018*** -0.043* 10.950 0.302** 0.002 0.013
(1.098) (0.103) (0.005) (0.024) (12.106) (0.116) (0.006) (0.017)
SIZE -8.725 -4.341*** 0.254*** 0.355 38.909 1.529** 0.134*** 0.354***
(14.439) (1.286) (0.077) (0.243) (172.848) (0.573) (0.042) (0.124)
Constant 577.616* 72.096** -6.024*** -8.082 1288.212 -34.890*** -2.857*** -7.515***
(322.408) (27.801) (1.708) (5.298) (3924.394) (12.593) (0.898) (2.666)
Observations 12,047 12,047 12,047 12,047 5,300 5,300 5,300 5,300
R-squared 0.463 0.344 0.6634 0.3617 0.370 0.390 0.354 0.239
Firm FE Yes Yes Yes Yes Yes Yes Yes Yes
Day FE Yes Yes Yes Yes Yes Yes Yes Yes
Panel II: Tick size = A$0.01 (one-cent)
DSMALL 135.552*** 11.870*** 1.802*** 6.815** 2956.856*** 1.603 1.158* 2.647
(33.918) (2.612) (0.165) (3.233) (701.792) (1.429) (0.599) (1.711)
VOLATILITY -6.336*** -0.235** 0.079*** 0.488*** -18.067 -0.056 0.053*** 0.099***
(0.879) (0.103) (0.007) (0.106) (17.663) (0.041) (0.010) (0.032)
LIQUIDITY 8.658*** 1.026*** -0.009** -0.025 -12.189 0.252*** 0.002 0.009
(0.823) (0.087) (0.004) (0.042) (14.449) (0.035) (0.009) (0.023)
90
SIZE 35.744* -5.531*** 0.649*** 6.153** 94.417 -1.150 0.713*** 1.006
(18.557) (2.000) (0.138) (2.643) (261.661) (1.040) (0.233) (0.622)
Constant -611.204 62.196 -16.419*** -149.395** 8417.129 24.716 -13.588*** -11.276
(422.390) (45.126) (3.142) (59.909) (6380.220) (22.917) (5.141) (13.707)
Observations 86,994 86,994 86,994 86,994 54,329 54,329 54,329 54,329
R-squared 0.492 0.561 0.799 0.471 0.416 0.155 0.333 0.195
Robust standard errors are in parentheses. *** p<0.01, ** p<0.05, * p<0.10
91
The average values of relative tick size for the RTSSMALL_0.005 group are
0.275% for the ASX and 0.273% for the CHIX. These figures are not markedly
different from the values reported for the RTSLARGE_0.01 group, which are 0.344% for
the ASX and 0.336% for the CHIX. These figures suggest that stocks defined as
having a small relative tick size in the half-cent category are virtually identical to
stocks labelled as having a large relative tick size in the one-cent category.
Therefore, the relative tick size of the RTSSMALL group in the half-cent category is
just not small enough to attract HFTs.
The RTSSMALL_0.01 group is arguably the best depiction of a setting with a

very small relative tick size. The group’s average relative tick size for the ASX and
CHIX is just 0.026% and 0.024%, respectively, which are roughly equivalent to one-
tenth of the values shown in the RTSSMALL_0.005 and RTSLARGE_0.01 groups. This
setting makes the relative tick size of the RTS SMALL_0.01 group extremely small,
allowing fast traders like HFTs to have their limit orders filled first by quoting
slightly better prices (Werner et al., 2011). In turn, this would make it less attractive
for non-HFTs to establish price-priority on the order book through limit order
placements, resulting in shorter queues across all price levels (Angel, 2011). This
also enables HFTs to regularly update or cancel their limit orders without fear of
being unable to quote a better price, or having their new orders placed at the end of a
lengthy queue. Overall, these findings imply that HFTs would only engage in order-
undercutting when the relative tick size is sufficiently small.
3.6.2 HFT activity surrounding a tick size border

Table 3.10 describes the dataset used to test the second hypothesis; it is divided into
three categories representing different price distance from the A$2.00 tick size
border, namely FAR (Panel I), MID (Panel II), and NEAR (Panel III). Stocks with
the farthest price distance (FAR) have a price range of A$1.70 – A$1.795
(RTSSMALL) and A$2.21 – A$2.30 (RTSLARGE), while stocks with a medium price
distance (MID) have a price range of A$1.80 – A$1.895 (RTSSMALL) and A$2.11 –
A$2.20 (RTSLARGE), and stocks with the shortest price distance (NEAR) have a price
range of A$1.90 – A$1.995 (RTSSMALL) and A$2.00 – A$2.10 (RTSLARGE).
92
This table describes the data used to test the second hypothesis, categorised as either “FAR” (Panel I), “MID” (Panel II), or “NEAR” (Panel III), based on their respective
price distance from the A$2.00 tick size border. Panels A and B describe the data of observations representing the ASX and CHIX datasets, while RTSSMALL and RTSLARGE
refer to the observations priced below and above the tick size border, respectively. N, Mean, Std. Dev., Minimum, and Maximum refer to the number of observations, average,
standard deviation, lowest value, and highest value, accordingly. PRICE refer to the closing price; RTS indicates the relative tick size, which is calculated by dividing nominal
tick size by closing price (Equation 3.5); VOLATILITY is measured by difference between the highest and lowest price of the day, divided their average prices (Equation 3.6);
LIQUIDITY is measured using the Corwin and Schultz (2012) high-low spread (Equation 3.7); and SIZE is the actual dollar value of market capitalisation.
Panel I: Distance = FAR
RTSSMALL (1.70 – 1.795)
PRICE ($) 1,251 1.746 0.022 1.700 1.795 713 1.747 0.023 1.700 1.795
RTS (%) 1,251 0.286 0.004 0.279 0.294 713 0.286 0.004 0.279 0.294
VOLATILITY (%) 1,251 1.996 0.835 0.289 5.436 713 1.719 0.850 0.000 5.436
LIQUIDITY (%) 1,249 0.509 0.627 0.000 4.187 712 0.390 0.555 0.000 3.476
SIZE ($ billion) 1,251 4.054 1.908 0.647 9.503 713 4.461 2.156 0.813 9.503
RTSLARGE (2.21 – 2.30)

PRICE ($) 777 2.256 0.022 2.210 2.300 629 2.256 0.023 2.210 2.300
RTS (%) 777 0.443 0.004 0.435 0.453 629 0.443 0.004 0.435 0.453
VOLATILITY (%) 777 1.944 0.728 0.436 3.991 629 1.636 0.746 0.000 3.991
LIQUIDITY (%) 777 0.529 0.653 0.000 3.556 624 0.420 0.563 0.000 2.811
SIZE ($ billion) 777 3.797 2.002 0.841 11.980 629 3.903 2.167 1.017 11.980
Panel II: Distance = MID
RTSSMALL (1.80 – 1.895)
PRICE ($) 1,017 1.845 0.023 1.800 1.895 574 1.844 0.023 1.800 1.895
RTS (%) 1,017 0.271 0.003 0.264 0.278 574 0.271 0.003 0.264 0.278
VOLATILITY (%) 1,017 1.945 0.883 0.532 5.142 574 1.657 0.800 0.000 4.878
LIQUIDITY (%) 1,015 0.528 0.662 0.000 4.697 572 0.366 0.537 0.000 3.879
SIZE ($ billion) 1,017 3.891 1.986 0.680 9.982 574 3.846 2.346 0.822 9.982
RTSLARGE (2.11 – 2.20)

PRICE ($) 788 2.157 0.022 2.110 2.200 614 2.157 0.022 2.110 2.200
RTS (%) 788 0.464 0.005 0.455 0.474 614 0.464 0.005 0.455 0.474
93
VOLATILITY (%) 788 1.997 0.718 0.464 4.176 614 1.607 0.725 0.000 3.721
LIQUIDITY (%) 787 0.551 0.644 0.000 3.140 609 0.445 0.592 0.000 3.248
SIZE ($ billion) 788 4.034 2.178 0.806 11.660 614 3.950 2.339 0.966 11.660
Panel III: Distance = NEAR
RTSSMALL (1.90 – 1.995)
PRICE ($) 1,013 1.946 0.022 1.900 1.995 526 1.946 0.023 1.900 1.995
RTS (%) 1,013 0.257 0.003 0.251 0.263 526 0.257 0.003 0.251 0.263
VOLATILITY (%) 1,013 1.900 0.820 0.509 4.878 526 1.671 0.776 0.000 4.627
LIQUIDITY (%) 1,010 0.487 0.633 0.000 4.124 522 0.330 0.498 0.000 2.795
SIZE ($ billion) 1,013 4.083 1.897 0.736 10.590 526 3.443 2.233 0.882 10.590
RTSLARGE (2.00 – 2.10)

PRICE ($) 984 2.049 0.024 2.000 2.100 639 2.050 0.025 2.000 2.100
RTS (%) 984 0.488 0.006 0.476 0.500 639 0.488 0.006 0.476 0.500
VOLATILITY (%) 984 2.121 0.852 0.482 4.878 639 1.683 0.825 0.000 4.401
LIQUIDITY (%) 983 0.645 0.741 0.000 4.085 637 0.429 0.570 0.000 2.787
SIZE ($ billion) 984 3.907 2.092 0.747 11.070 639 3.567 2.309 0.934 11.070
94
The table indicates that a large number of ASX stocks are priced below
and far from the border (RTSSMALL_FAR), whereas in the CHIX, the figures are nearly
similar across all categories. On average, ASX dataset is shown to exhibit a higher
intraday price movement (VOLATILITY) than the CHIX across all groups
(RTSSMALL and RTSLARGE) and categories (FAR, MID, and NEAR). In ASX,
RTSSMALL_NEAR and RTSLARGE_NEAR show the lowest and highest volatility,
respectively, whereas in CHIX, RTSLARGE_MID and RTSSMALL_FAR represent the
lowest and highest volatility, respectively. In addition, stocks with a smaller relative
tick size tend to be less volatile on the ASX, but the opposite is true on the
CHIX. Nevertheless, the difference in volatility between the RTS SMALL and
RTSLARGE groups is generally small.
In terms of LIQUIDITY, the ASX exhibits wider spreads than the CHIX
across all examined groups, indicating that the former is less liquid. Stocks classified
as RTSSMALL are consistently shown to have larger spreads than those identified as
RTSLARGE in both markets. Furthermore, the RTSSMALL_NEAR is shown to have the
highest liquidity on both markets, whilst the RTS LARGE_NEAR and RTSLARGE_MID have
the lowest liquidity on the ASX and CHIX, respectively. In general, stocks with a
small relative tick size would be more liquid on both exchanges. As for SIZE, the
statistics presented in the table show that the market capitalisations across all relative
tick size groups and price distance categories are somewhat comparable. It is shown
that the average market capitalisation in both markets is around A$4 billion, and that
stocks with smaller relative tick sizes have a higher firm size. Overall, stocks trading
at prices below the A$2.00 tick size border (i.e., RTSSMALL) are more liquid and have
a larger market capitalisation than those trading at prices above the border (i.e.,
RTSLARGE) in both markets, while showing a similar intraday volatility.

Table 3.11 shows the results of univariate analysis using a two-sample t-test to
compare HFT activity, as proxied by MTR, ALGO, HFO, and HFOR, across two
relative tick size groups (RTSSMALL and RTSLARGE), which is classified based on
different price distance of either FAR (Panel I), MID (Panel II), or NEAR (Panel
III). The table indicates that there is a statistically significant difference in HFT
activity between the RTSSMALL and RTSLARGE groups.
95
Table 3.11. Mean comparison of HFT activity near the A$2.00 tick size border
This table shows the results from univariate analysis using stocks near the A$2.00 tick size border to
test the second hypothesis. The data is categorised as either “FAR” (Panel I), “MID” (Panel II), or
“NEAR” (Panel III), based on their respective price distance from the A$2.00 tick size border. The
results for ASX and CHIX are shown in the upper and lower portions of each panel, correspondingly.
Obs. and Mean refers to the number of observations and the average values of each HFT
measure, accordingly. Difference represents the difference between the mean values of RTSSMALL and
RTSLARGE. Message-to-trade ratio (MTR), average trade size (ALGO), high-frequency orders (HFO),
and HFO-contributed message ratio (HFOR) are the variables used as proxies for measuring HFT
activity. The formula are illustrated in Equations 3.1, 3.2, 3.3, and 3.4, respectively. Data are
winsorised at three standard deviations (3-sigma) from their respective means.
Panel I: RTSSMALL (1.70 – 1.795) RTSLARGE (2.21 – 2.30) Difference
(Dist. = FAR) Obs. Mean Obs. Mean (RTSSMALL – RTSLARGE)
ASX
MTR (%) 1,251 523.720 777 557.603 -33.884***
ALGO 1,251 -12.496 777 -11.876 -0.620
HFO (Ln) 1,251 2.384 777 2.532 -0.148***
HFOR (%) 1,251 3.185 777 3.535 -0.351**
CHIX
MTR (%) 713 750.430 629 758.185 -7.755
ALGO 713 -5.223 629 -7.472 2.249***
HFO (Ln) 713 0.134 629 0.102 0.032**
HFOR (%) 713 0.173 629 0.149 0.024
Panel II: RTSSMALL (1.80 – 1.895) RTSLARGE (2.11 – 2.20) Difference

(Dist. = MID) Obs. Mean Obs. Mean (RTSSMALL – RTSLARGE)
ASX
MTR (%) 1,017 524.675 788 548.935 -24.261***
ALGO 1,017 -12.736 788 -13.827 1.091**
HFO (Ln) 1,017 2.388 788 2.360 0.028
HFOR (%) 1,017 3.226 788 3.106 0.120
CHIX
MTR (%) 574 804.149 614 949.663 -145.514
ALGO 574 -4.433 614 -7.244 2.812***
HFO (Ln) 574 0.110 614 0.060 0.051***
HFOR (%) 574 0.128 614 0.087 0.041*
Panel III: RTSSMALL (1.90 – 1.995) RTSLARGE (2.00 – 2.10) Difference

(Dist. = NEAR) Obs. Mean Obs. Mean (RTSSMALL – RTSLARGE)
ASX
MTR (%) 1,013 508.102 984 507.390 0.712
ALGO 1,013 -12.701 984 -13.672 0.972**
HFO (Ln) 1,013 2.521 984 2.287 0.234**
HFOR (%) 1,013 3.684 984 3.220 0.465***
CHIX
MTR (%) 526 872.144 639 738.080 134.064*
ALGO 526 -4.209 639 -6.573 2.364***
HFO (Ln) 526 0.102 639 0.060 0.043***
HFOR (%) 526 0.133 639 0.082 0.051*
*** p<0.01, ** p<0.05, * p<0.10
96
However, the findings for ASX and CHIX across the various HFT proxies
and price distance categories examined are inconsistent. In the FAR category of the
ASX, stocks in the RTSSMALL group have MTR, HFO, and HFOR values that are
substantially lower than those in the RTSLARGE group. In the MID category, the
results are contradictory: MTR is shown to be significantly lower in the RTS SMALL
group than in the RTSLARGE group; ALGO is shown to be significantly higher in the
RTSSMALL group; and HFO and HFOR values are shown to be not statistically
different between the two groups.
On the other hand, the NEAR category shows that RTSSMALL have
significantly higher ALGO, HFO, and HFOR values than RTS LARGE. For the CHIX
dataset, the findings are more uniform; HFT activity is significantly greater among
stocks in the RTSSMALL group compared to the RTSLARGE group. Three of the four
HFT measures (ALGO, HFO, and HFOR) are substantially larger in RTS SMALL than
in RTSLARGE in the FAR and MID categories. In the NEAR category, the results are
even more conclusive; all four HFT measures tested are significantly higher in the
RTSSMALL group.
Overall, the results of the univariate tests indicate that when the relative tick
size values are substantially different, as represented by the NEAR category, it
becomes evident that HFT activity is greater in the stocks with a smaller relative tick
size, which suggest more order-undercutting activity. However, when the
discrepancy in relative tick size is less pronounced, as indicated by the FAR
category, HFT activity is found to be more prevalent in stocks with large relative tick
size, which suggest more order-queuing activity. Nonetheless, as shown in the
descriptive statistics in Table 3.10, other firm-specific factors such as volatility,
liquidity and size might influence the level of HFT activity observed. Therefore, a
multivariate analysis is required to isolate and control for these other influences.

Table 3.12 shows the results of a multiple regression study to assess the influence of
a small relative tick size value on HFT activity, classified as FAR (Panel I), MID
(Panel II), or NEAR (Panel III) based on their respective price distances from the
A$2.00 tick size border. The regression findings show that on the ASX, stocks in the
97
FAR category with a small relative tick size exhibit significantly lower MTR (β = -
97.413) and higher HFO (β = 0.512) values, while having no influence on ALGO or
HFOR. For the MID and NEAR categories, stocks with a small relative tick size on
the ASX is shown to have higher HFO (MID: β = 0.415; NEAR: β = 0.393) and
HFOR (MID: β = 1.041; NEAR: β = 1.138) values, but it has no influence on either
MTR or ALGO. The control variables employed are found to be significant in some
of the models presented in the table, especially in the FAR category and when HFT
is measured by MTR or ALGO.
For CHIX, the regression results of the FAR category demonstrate that none
of the HFT measures investigated are substantially influenced by relative tick size or
any of the control variables utilised. However, the high R-squared values (MTR: R 2
= 0.724; ALGO: R2 = 0.864; HFO: R2 = 0.778; HFOR: R2 = 0.802) suggest that
some of the variance in the HFT measures may have been explained by cross-
sectional (firm) or time (day) factors that were not explicitly specified in the model
used in this study. For the MID category, stocks with a small relative tick size would
have a higher MTR (β = 332.009) and ALGO (β = 1.672), while in the NEAR
category, it is shown to be positively affecting ALGO (β = 2.080) and HFO (β =
0.091). Notably, all of the control variables are shown to have no impact on any of
the analysed HFT measures across all three price distance categories, but this is with
the exception to LIQUIDITY, which is observed to affect ALGO in the MID
category.
The differences between ASX and CHIX results may be attributable to the
characteristics of their respective markets. The ASX is naturally comprised of a
diverse spectrum of participants, including fast and slow traders, informed and
uninformed investors, and institutional and retail players. This demography provides
HFTs with greater opportunities to trade against uninformed traders, hence reducing
their adverse selection risk associated with market-making. In contrast, the CHIX is
typically used by sophisticated traders that require a low-latency trading
environment, such as the HFTs. Therefore, it is conceivable that uninformed
investors would not participate in the CHIX market, which would force market-
making HFTs to trade against other informed traders, therefore increasing their
adverse selection risk.
98
Table 3.12. Regression analysis on HFT activity near the A$2.00 tick size border
This table shows the results from multivariate regression analysis using stocks near the A$2.00 tick size border to test the second hypothesis. The data is categorised as either
“FAR” (Panel I), “MID” (Panel II), or “NEAR” (Panel III), based on their respective price distance from the border. Panels A and B illustrate the findings using the ASX and
CHIX datasets respectively. Dependent variables are the HFT activity measures, namely message-to-trade ratio (MTR), average trade size (ALGO), high-frequency orders
(HFO), and HFO-contributed message ratio (HFOR). The formula are shown in Equations 3.1, 3.2, 3.3, and 3.4, respectively. The independent variable is DSMALL, which is
a dummy variable assigned with a value of one if the observation belongs to the group with low relative tick size (RTSSMALL), and zero otherwise. The control variables are
the one-day lagged (t – 1) values of VOLATILITY, LIQUIDITY, and SIZE. VOLATILITY is measured by difference between the highest and lowest price of the day, divided
their average prices (Equation 3.6); LIQUIDITY is measured using the Corwin and Schultz (2012) high-low spread (Equation 3.7); and SIZE is the natural log of market
capitalisation, respectively. All models are controlled for firm and day fixed-effects. Data are winsorised at three standard deviations (3-sigma) from their respective means.
The standard errors are calculated using the Huber/White/sandwich estimator of variance method.
Panel I: Distance = FAR (1.70 – 1.795; 2.21 – 2.30)
DSMALL -97.413*** -0.308 0.512** 1.099 296.631 1.140 0.059 -0.027
(34.131) (1.854) (0.196) (0.748) (224.875) (1.300) (0.109) (0.106)
VOLATILITY -6.348 -0.356 -0.002 -0.086 95.148 -0.264 0.035 -0.009
(8.965) (0.501) (0.032) (0.152) (118.283) (0.179) (0.024) (0.048)
LIQUIDITY 15.344* 0.992** -0.027 -0.052 131.557 0.254 -0.039 -0.063
(8.748) (0.389) (0.046) (0.225) (154.838) (0.307) (0.029) (0.053)
SIZE -138.206** -9.549** 0.672** 2.163** 1783.433 -2.155 -0.005 -0.326
(55.270) (4.623) (0.257) (0.936) (1412.506) (3.178) (0.221) (0.318)
Constant 3683.323*** 205.038** -12.695** -46.862** -26306.658 46.532 -0.035 6.934
(1213.484) (101.215) (5.766) (20.954) (30226.670) (69.576) (4.852) (6.866)
Observations 2,024 2,024 2,024 2,024 1,337 1,337 1,337 1,337
R-squared 0.826 0.815 0.882 0.754 0.724 0.864 0.778 0.802
Panel II: Distance = MID (1.80 – 1.895; 2.11 – 2.20)
DSMALL -7.210 2.663 0.415*** 1.041** 332.009* 1.672* 0.078 0.078
(14.908) (1.725) (0.131) (0.417) (183.128) (0.959) (0.090) (0.126)
VOLATILITY -9.506 -0.575 0.0633** 0.113 357.576 0.058 0.021 0.002
(6.883) (0.417) (0.030) (0.157) (318.237) (0.302) (0.025) (0.050)
LIQUIDITY 20.730 1.393* -0.023 0.132 134.848 0.522* 0.017 0.050
(14.499) (0.704) (0.065) (0.268) (134.410) (0.271) (0.040) (0.072)
99
SIZE -83.029* -8.934 0.415 0.481 -527.547 -3.625 0.218 0.358
(47.417) (5.714) (0.390) (0.953) (1091.001) (3.156) (0.320) (0.389)
Constant 2116.780** 165.788 -9.313 -10.547 18930.403 78.708 -4.725 -7.699
(1040.483) (124.312) (8.543) (20.710) (23570.691) (68.490) (6.931) (8.372)
Observations 1,799 1,799 1,799 1,799 1,177 1,177 1,177 1,177
R-squared 0.885 0.826 0.893 0.812 0.833 0.859 0.763 0.821
Panel III: Distance = NEAR (1.90 – 1.995; 2.00 – 2.10)
DSMALL -12.730 0.675 0.393*** 1.138** -45.490 2.080** 0.091* 0.166
(11.092) (0.865) (0.106) (0.493) (563.525) (0.989) (0.049) (0.141)
VOLATILITY 0.603 -0.077 0.050 0.201 -48.188 -0.339 -0.011 -0.111
(7.142) (0.514) (0.031) (0.259) (77.015) (0.401) (0.022) (0.075)
LIQUIDITY -2.468 1.056 0.075 0.405 -27.356 0.006 -0.011 -0.104
(7.495) (0.651) (0.063) (0.266) (75.727) (0.295) (0.029) (0.127)
SIZE -110.607** -5.501* 0.724 1.252 -5522.185 -0.154 0.321 0.520
(43.524) (3.005) (0.462) (1.638) (9254.173) (13.702) (0.488) (1.445)
Constant 2584.583*** 68.725 -15.956 -26.948 121172.230 3.592 -6.760 -10.522
(942.596) (66.508) (10.293) (36.455) (198311.640) (292.241) (10.452) (31.073)
Observations 1,994 1,994 1,994 1,994 1,156 1,156 1,156 1,156
R-squared 0.891 0.879 0.912 0.778 0.650 0.856 0.845 0.715
100
In stocks that are priced relatively closer to the tick size border as represented
by the MID and NEAR categories, the disparity in relative tick size between
RTSSMALL and RTSLARGE becomes more distinct. In these price ranges, stocks priced
below the A$2.00 tick size border (RTSSMALL) would have small relative tick sizes,
which correspond to a low adverse selection risk for market-making HFTs, thereby
enticing them to undercut orders through marginal price improvements. In contrast,
stocks priced above the border (RTSLARGE) would have large relative tick sizes,
which translate to a large profit from market-making and thus encourage limit order
placements using order-queuing. In this setting, RTS SMALL is found to have a higher
HFT activity on both the ASX and CHIX markets, suggesting that the HFTs are
predominantly risk-averse traders, as they choose to minimise their adverse selection
risk through order-undercutting rather than attempting to maximise their profits
through order-queuing.
On the contrary, when stocks are priced further from the tick size border, as
represented by the FAR category, the relative tick size difference between RTS SMALL
and RTSLARGE reduces. This setting implies that the relative tick size is neither too
small nor too large, which signifies an adverse selection risk that is not too low and a
market-making profit that is not too high, hence reducing the incentive for HFTs to
participate in either order-undercutting or order-queuing activities. On the ASX, this
situation yields conflicting findings, with RTSSMALL exhibiting significantly lower
MTR but higher HFO. Due to the presence of uninformed traders on the ASX
market, the adverse selection risk from order-queuing is reduced, which cause some
HFTs to prefer stocks with large relative tick size (RTSLARGE).31 Nonetheless, other
HFTs may choose to perform order-undercutting in order to minimise their adverse
selection risk. In contrast, the demographics of the CHIX market compel HFTs to
constantly choose stocks with a small relative tick size (RTSSMALL) in order to
minimise their adverse selection risk, given that other market participants are
primarily informed traders. Therefore, when a stock’s relative tick size value is
neither too small nor too large, it will be ignored by HFTs’ trading algorithm, which
will instead utilise information from other parameters. This may explain why none of
31
As indicated by the significant negative relationship between DSMALL and MTR shown in Table
3.12 (Panel I).
101
the examined HFT measures in the FAR category of the CHIX show a significant
effect on relative tick size.
Overall, these findings indicate that HFT activity could indeed be affected by
relative tick size values. HFTs would almost always prefer stocks with a smaller
relative tick size, if the relative tick size is sufficiently small. In addition, other
factors, such as the proportion of informed and uninformed market participants,
would affect HFTs’ attitude and, therefore, their activity level. The majority of the
findings indicate that HFTs would use their speed to establish price priority through
undercutting activity. In addition, given that HFTs would have spent extensively to
be faster than their competitors, it is plausible that they would leverage on their
speed superiority whenever feasible, especially in minimising their adverse selection
risk. Consequently, should market authorities seek to modify an existing nominal
tick size to influence HFT activity, the new tick size must be sufficiently small or
large; otherwise, it would be counterproductive, since it would have negligible effect
on HFTs and may potentially harm other participants.
3.6.3 HFT activity surrounding tick size crossing events

Table 3.13 displays the data used to test the third hypothesis, which assessed the
impact of relative tick size change due to crossing the tick size border on HFT
activity. The dataset is divided into two types of tick size crossing, either
UPWARDS or DOWNWARDS, and further segregated based on the timing of the
observation, which is either before (PRE-EVENT) or after (POST-EVENT) the
occurrence of a crossing event. This categorisation generates four sample groups,
namely PRE-EVENTUPWARDS POST-EVENTUPWARDS, PRE-EVENTDOWNWARDS, and
POST-EVENTDOWNWARDS, which are shown in Panel I (a), Panel I (b), Panel II (a),
and Panel II (b), respectively.
102
This table describes the data used to test the third hypothesis, which are organised by the direction of the tick size crossing events. Panels I (a) and (b) represent the data for
the PRE-EVENTUPWARDS and POST-EVENTUPWARDS, whereas Panels II (a) and (b) refer to the PRE-EVENTDOWNWARDS and POST-EVENTDOWNWARDS, respectively. PRE-
EVENT and POST-EVENT refer to the values in the period before and after a tick size crossing event occurrence, accordingly. Panels A and B show the values from the ASX
and CHIX datasets. Treatment refers to all observations classified as accepted tick size crossing events, while Control refers to non-crossing observations that fulfil the
criteria outlined in Section 3.5.4 (Model specification). N, Mean, Std. Dev., Minimum, and Maximum refer to the number of observations, average, standard deviation, lowest
value, and highest value, accordingly. PRICE refer to the closing price; RTS indicates the relative tick size, which is calculated by dividing nominal tick size by closing price
(Equation 3.5); VOLATILITY is measured by difference between the highest and lowest price of the day, divided their average prices (Equation 3.6); LIQUIDITY is measured
using the Corwin and Schultz (2012) high-low spread (Equation 3.7); and SIZE is the actual dollar value of market capitalisation.
Panel I (a): PRE-EVENTUPWARDS
Treatment
PRICE ($) 95 1.933 0.070 1.495 1.995 37 1.960 0.028 1.865 1.995
RTS (%) 95 0.259 0.011 0.251 0.334 37 0.255 0.004 0.251 0.268
VOLATILITY (%) 95 3.839 3.903 0.855 27.429 37 2.098 1.384 0.503 6.897
LIQUIDITY (%) 94 0.226 0.797 0.000 6.744 37 0.079 0.277 0.000 1.435
SIZE ($ billion) 95 3.383 2.064 0.3572 10.51 37 4.003 2.546 1.149 10.51
Control
PRICE ($) 691 1.455 0.257 1.005 1.975 210 1.439 0.281 1.010 1.940
RTS (%) 691 0.355 0.065 0.253 0.498 210 0.361 0.072 0.258 0.495
VOLATILITY (%) 691 3.618 2.554 0.607 20.921 210 2.246 1.406 0.257 11.411
LIQUIDITY (%) 686 0.853 1.196 0.000 8.440 210 0.600 0.812 0.000 3.879
SIZE ($ billion) 691 2.86 1.529 0.1917 8.039 210 3.113 1.772 0.6137 8.039
Panel I (b): POST-EVENTUPWARDS
Treatment
PRICE ($) 95 2.080 0.070 2.000 2.310 37 2.055 0.052 2.000 2.210
RTS (%) 95 0.481 0.015 0.433 0.500 37 0.487 0.012 0.453 0.500
VOLATILITY (%) 95 3.987 3.600 0.499 30.769 37 2.506 1.671 0.976 7.092
LIQUIDITY (%) 95 0.942 1.280 0.000 5.548 37 0.391 0.596 0.000 2.374
SIZE ($ billion) 95 3.614 2.178 0.4287 11.23 37 4.203 2.688 1.184 11.23
103
Control
PRICE ($) 691 1.472 0.257 1.010 1.985 210 1.459 0.282 1.015 1.995
RTS (%) 691 0.351 0.063 0.252 0.495 210 0.356 0.071 0.251 0.493
VOLATILITY (%) 691 3.511 2.370 0.557 19.380 210 2.229 1.432 0.416 8.237
LIQUIDITY (%) 690 0.776 1.223 0.000 9.855 209 0.459 0.725 0.000 5.370
SIZE ($ billion) 691 2.9 1.553 0.1981 8.465 210 3.152 1.788 0.6595 8.465
Panel II (a): PRE-EVENTDOWNWARDS
Treatment
PRICE ($) 111 2.093 0.177 2.000 3.740 42 2.052 0.054 2.000 2.270
RTS (%) 111 0.480 0.027 0.267 0.500 42 0.488 0.012 0.441 0.500
VOLATILITY (%) 111 4.380 3.075 0.990 20.833 42 3.430 3.398 0.000 20.833
LIQUIDITY (%) 111 0.227 0.569 0.000 3.417 42 0.181 0.537 0.000 2.787
SIZE ($ billion) 111 3.339 2.069 0.4117 10.91 42 3.059 2.643 0.7573 10.91
Control
PRICE ($) 1,689 3.027 0.506 2.020 3.970 753 3.047 0.509 2.020 3.970
RTS (%) 1,689 0.340 0.060 0.252 0.495 753 0.338 0.060 0.252 0.495
VOLATILITY (%) 1,689 3.231 2.454 0.254 41.745 753 1.955 1.280 0.000 12.522
LIQUIDITY (%) 1,689 0.692 1.044 0.000 9.683 749 0.368 0.565 0.000 3.585
SIZE ($ billion) 1,689 4.981 6.459 0.4042 48.28 753 4.969 6.287 0.9594 48.28
Panel II (b): POST-EVENTDOWNWARDS
Treatment
PRICE ($) 111 1.886 0.140 0.940 1.990 42 1.902 0.167 0.980 1.995
RTS (%) 111 0.267 0.031 0.251 0.532 42 0.266 0.040 0.251 0.510
VOLATILITY (%) 111 5.383 7.709 0.757 73.498 42 4.497 11.091 0.505 72.857
LIQUIDITY (%) 111 1.091 2.237 0.000 16.830 41 0.353 0.615 0.000 2.672
SIZE ($ billion) 111 3.08 1.977 0.3312 10.49 42 2.874 2.511 0.3312 10.49
Control
PRICE ($) 1,654 2.969 0.506 2.010 3.960 752 3.025 0.509 2.000 3.970
RTS (%) 1,654 0.347 0.062 0.253 0.498 752 0.341 0.061 0.252 0.500
VOLATILITY (%) 1,654 3.338 2.463 0.281 1,654 752 2.011 1.345 0.000 11.111
LIQUIDITY (%) 1,654 0.764 1.153 0.000 1,654 750 0.405 0.672 0.000 5.385
SIZE ($ billion) 1,654 4.872 6.336 0.4117 1654 752 4.927 6.261 0.9384 48.03
104
The table illustrates that, across all groups, the number of Treatment
observations is consistently lower than the number of matching Control
observations, and there are higher numbers of tick size crossing events recorded on
the ASX than on the CHIX. This is because the study period using the ASX dataset
began in January 2008, whereas for the CHIX dataset, it does not begin until
November 2011. Moreover, there are cases in which a tick size crossing event
recorded on the ASX does not appear on the CHIX dataset because no trading
transpired on the CHIX for that specific stock on that day, thereby reducing the
number of accepted tick size crossing events recorded on the CHIX dataset. 32
The statistics for UPWARDS events indicate that for the ASX, the Treatment
sample in general has a larger intraday price range (VOLATILITY), a smaller bid-
ask spread (LIQUIDITY), and a larger market capitalisation (SIZE) than their
corresponding Control sample, both before and after the event. In the post-event
period, the Treatment’s average intraday price range increased by only 0.1486%,
whereas its bid-ask spread widened by 0.716%, and firm size increased by A$0.231
billion. In comparison, the exhibited changes for the Control are weaker, as the
intraday price range reduced by 0.107%, the bid-ask spread shrank by 0.077%, and
the firm size increased by only A$0.04 billion. For CHIX, in the pre-event period,
stocks in the Treatment group exhibit lower volatility, narrower spread, and greater
market capitalisation than those in the Control group. Compared to the pre-event
period, the Treatment’s volatility increased by 0.408%, the bid-ask spread widened
by 0.312%, and the size of the firms grew by A$0.200 billion after crossing the tick
size border. In contrast, the Control group only saw moderate changes, with
volatility decreased by 0.017%, spreads tightened by 0.141%, and market
capitalisation rose by only A$0.039 billion.
According to the statistics for DOWNWARDS events on the ASX, the

Treatment group displays a larger intraday price range, a narrower bid-ask spread,
and a smaller market capitalisation than its Control sample prior to the event. After
the event, the Treatment’s average intraday price range surged by 1.003%, while the
32
On February 25, 2016, for instance, Spark Infrastructure Group (ASX:SKI) stock prices fluctuated
from A$1.98 to A$2.05, indicating that the stock price had crossed the A$2.00 tick size border.
However, no SKI trades were completed on the CHIX on that day.
105
bid-ask spread markedly widened by 0.864%, and the firm size fell by A$0.259
billion. Comparatively, the post-event figures for Control are only slightly different,
with the intraday price range rising by 0.107%, the bid-ask spread expanding by
0.072%, and the firm size reducing by just A$0.109 billion. For CHIX, in the pre-
event period, stocks in the Treatment group are shown to be more volatile, have a
narrower spread, and have a smaller market capitalisation than their counterparts in
the Control group. Following the tick size crossing event, the Treatment’s stock
volatility is shown to be 1.067% larger, the bid-ask spread is 0.172% wider, and the
firm size is reduced by A$0.185 billion. Comparatively, the Control group
experienced minimal changes, with volatility and spreads rising by just 0.056% and
0.037%, respectively, and market capitalisation falling by only A$0.042 billion.
In summary, apart from the market capitalisation of the stocks in the

Treatment group, which is expected to increase in the UPWARDS event and
decrease in the DOWNWARDS event, volatility and liquidity are shown to be
considerably different after a tick size crossing event. Changes in volatility are found
to be stronger in DOWNWARDS events, which increase volatility by 1.003% on the
ASX and 1.067% on the CHIX, whereas UPWARDS events increase volatility by
just 0.1486% on the ASX and 0.408% on the CHIX. For liquidity, the UPWARDS
and DOWNWARDS events expanded the spreads on the ASX’s Treatment group by
0.716% and 0.864%, respectively, whilst on the CHIX, the spreads are only 0.332%
and 0.172% wider, respectively. In addition, the small changes detected in the
Control group are consistent with expectations, implying that the samples are
unaffected by the changes in tick size occurred on the Treatment stocks.

Table 3.14 presents the results of a univariate difference-in-difference analysis of the
tick size crossing events. For the UPWARDS events shown in Panel I, the ASX
dataset demonstrates that: (i) in the pre-event period, Treatment has more HFT
activity than the Control, as indicated by the significantly higher HFO and HFOR
values; (ii) in the post-event period, Treatment has less HFT activity, as indicated by
the significantly lower ALGO value; and (iii) difference-difference analysis shows
that an instant upsurge in relative tick size value is accompanied by a significant
decrease in ALGO, thus implying a reduction in HFT activity.
106
Table 3.14. Univariate difference-in-difference analysis of HFT activity
measures for tick size crossing events
This table presents the results of univariate difference-in-difference analysis to test the third
hypothesis. Panels I and II represent the UPWARDS and DOWNWARDS tick size crossing events,
respectively. Panels A and B show the information obtained from the ASX and CHIX datasets,
accordingly. Treatment refers to all observations classified as accepted tick size crossing events, while
Control refers to non-crossing observations that fulfil the criteria outlined in Section 3.5.4. DIFFPRE
and DIFFPOST refers to the difference between the values observed in Treatment and Control in the
period before and after the crossing events, respectively. DIFF-IN-DIFF refers to the difference-in-
difference values produced by subtracting the DIFFPOST with DIFFPRE. Message-to-trade ratio (MTR),
average trade size (ALGO), high-frequency orders (HFO), and HFO-contributed message ratio
(HFOR) are the variables used as proxies for measuring HFT activity. The formula are illustrated in
Equations 3.1, 3.2, 3.3, and 3.4, respectively. Data are winsorised at three standard deviations (3-
sigma) from their respective means. The standard errors are calculated using the
Huber/White/sandwich estimator of variance method.
Panel I: RTSUPWARDS
Pre-event
Treatment (T) 481.13 -14.92 2.32 3.23 674.00 -4.77 0.06 0.02
Control (C) 459.25 -13.96 1.93 2.68 698.60 -4.47 0.11 0.21
DIFFPRE (T – C) 21.88 -0.95 0.39** 0.55* -24.61 -0.30 -0.05 -0.19**
(19.00) (1.34) (0.18) (0.33) (80.36) (0.70) (0.04) (0.08)
Post-event
Treatment (T) 467.91 -19.10 2.11 2.78 483.90 -6.48 0.12 0.06
Control (C) 459.15 -14.60 1.89 2.73 675.27 -4.97 0.09 0.25
DIFFPOST (T – C) 8.76 -4.49*** 0.22 0.05 -191.37** -1.51** 0.03 -0.19**
(19.98) (1.63) (0.17) (0.36) (75.56) (0.71) (0.05) (0.09)
DIFF-IN-DIFF
DIFFPOST – DIFFPRE -13.12 -3.54* -0.17 -0.50 -166.76 -1.21 0.07 0.00
(27.57) (2.11) (0.25) (0.48) (110.30) (1.00) (0.07) (0.12)
Panel II: RTSDOWNWARDS
Pre-event
Treatment (T) 435.63 -19.51 1.68 2.28 939.21 -4.85 0.12 0.24
Control (C) 475.21 -19.94 2.11 2.89 1,035.11 -5.09 0.19 0.29
DIFFPRE (T – C) -39.58*** 0.43 -0.43*** -0.61** -95.90 0.24 -0.07 -0.05
(13.78) (1.48) (0.15) (0.29) (277.60) (0.62) (0.05) (0.17)
Post-event
Treatment (T) 426.45 -16.96 2.21 3.27 1,365.06 -3.65 0.38 0.27
Control (C) 476.64 -18.24 2.16 2.95 1,139.41 -5.43 0.17 0.29
DIFFPOST (T – C) -50.18*** 1.28 0.05 0.32 225.66 1.78*** 0.20** -0.02
(14.25) (1.42) (0.17) (0.36) (316.17) (0.64) (0.09) (0.09)
DIFF-IN-DIFF
DIFFPOST – DIFFPRE -10.61 0.85 0.48** 0.93** 321.56 1.55* 0.28*** 0.04
(19.82) (2.05) (0.22) (0.46) (420.75) (0.89) (0.10) (0.19)
Furthermore, the results for the CHIX dataset illustrate that: (i) in the pre-
event period, Treatment has less HFT activity than Control, as indicated by the
significantly lower HFOR values; (ii) in the post-event period, Treatment has less
HFT activity, as indicated by the significantly lower MTR, ALGO, and HFOR
107
values; and (iii) difference-difference analysis shows that the hike in relative tick
size value has no significant effect on any HFT measures employed. In general,
evidence from UPWARDS events implies that the sudden increase in relative tick
size after a tick size crossing event may reduce HFT activity, as shown by the
significantly lower algorithmic liquidity provision (ALGO); however, this outcome
is exclusive to the ASX and absent from the CHIX.
Panel II displays the results of the DOWNWARDS events. The ASX dataset
illustrates that: (i) in the pre-event period, Treatment has less HFT activity than the
Control, as indicated by the significantly lower MTR, HFO, and HFOR values; (ii)
in the post-event period, Treatment has less HFT activity, as indicated by the
significantly lower MTR value; and (iii) difference-difference analysis shows that
the instant reduction in relative tick size value results in a significant increase in
HFO and HFOR. In addition, the findings for the CHIX dataset signify that: (i) in
the pre-event period, the level of HFT activity in Treatment and Control is similar, as
none of the HFT proxies are significant; (ii) in the post-event period, Treatment has
more HFT activity, as shown by the significantly higher ALGO and HFO values;
and (iii) difference-difference analysis revealed that the decline in relative tick size
value leads to greater HFT activity, as evidenced by the significantly higher ALGO
and HFO values. In general, evidence from DOWNWARDS events shows that the
sudden reduction in relative tick size after a tick size crossing event resulted in
significantly higher HFT activity.
Overall, the results indicate that HFT activity is asymmetrically affected by

sudden changes in relative tick sizes triggered by tick size border crossings. The
smaller pricing grid resulting from a relative tick size reduction stimulates HFT
activity on both exchanges; however, when the situation is reversed, the coarser
pricing grid suppresses HFT activity on the ASX while having no impact on the
CHIX. This suggests that HFT is more sensitive when relative tick size is reduced, as
this motivates them to use their speed to engage in order-undercutting activity;
conversely, when relative tick size is increased, they do not use the order-queuing
approach to make profit, suggesting that they are more incentivised to profit from
order-undercutting than order-queuing strategies. Moreover, despite the fact that this
technique used matching Control samples, it is plausible that the observed variations
108
in HFT activity were also influenced by factors other than the relative tick size
changes, thus, prompting the use of multivariate regression analysis.
3.6.3.3 Difference-in-difference multivariate analysis

Table 3.15 displays the results of a multivariate regression analysis conducted to
analyse the impact of a sudden change in relative tick size due to tick size border
crossing on HFT activity. When the stock prices of Treatment stocks crossed the
A$2.00 tick size border, the relative tick size values would jump drastically in the
UPWARDS event (Panel I) and fall substantially in the DOWNWARDS event
(Panel II). For Control stocks, their prices would stay on the same side of the border;
hence, there would be no substantial change in their relative tick size values.
The coefficients of DTREATMENT reflect the aggregate impact of

Treatment stocks on HFT activity, which includes the influence observed in the
period before and after their relative tick size experiences a dramatic change due to
tick size border crossing. Similarly, the coefficients of DPOST represent the
aggregate influence of the post-event day on HFT activity, which combines the
impacts of stocks with (Treatment) and without (Control) substantial relative tick
size change on HFT activity. Therefore, DTREATMENT and DPOST would be less
informative in explaining the variation of the HFT measures when viewed in
isolation. By interacting these dummy variables, it is possible to distinguish the
impact of relative tick size changes on HFT activity from any influence that may
have emerged from the Treatment or post-event day’s inherent effect. Therefore, the
coefficients of DPOST×DTREATMENT are utilised to highlight whether changes in
relative tick size would influence HFT activity.
Panel I exhibits the findings for the UPWARDS scenario, which represents the
event in which relative tick size increased drastically due to tick size crossings. Using
the ASX dataset, the coefficients for DPOST×DTREATMENT are shown to be
significantly negative in ALGO (β = -2,946), HFO (β = -0.201), and HFOR (β = -0.569),
however using the CHIX dataset, only ALGO (β = -1.579) is significantly negative.
Furthermore, it is shown that these HFT measures are negatively associated with
DTREATMENT; although this relationship is not statistically significant, it does imply
that Treatment stocks generally exhibit lower levels of HFT activity than Control stocks.
109
Table 3.15. Regression analysis on HFT activity using tick size crossing events
This table presents the results of multivariate regression analysis using tick size crossing events to test the third hypothesis. Panels I and II represent the UPWARDS and
DOWNWARDS tick size crossing events, respectively. Panels A and B show the information obtained from the ASX and CHIX datasets, accordingly. Dependent variables are
the HFT activity measures, namely message-to-trade ratio (MTR), average trade size (ALGO), high-frequency orders (HFO), and HFO-contributed message ratio (HFOR).
The formula are shown in Equations 3.1, 3.2, 3.3, and 3.4, respectively. The independent variables are DTREATMENT, DPOST, and DTREATMENT×DPOST, where
DTREATMENT is a dummy variable that equals to one if the observation is identified as a tick size crossing event (see Section 3.5.4: Model specification), and zero
otherwise; DPOST is a dummy variable assigned with a value of one if the observation belongs to the post-event period, and zero otherwise; and the interaction term,
DTREATMENT×DPOST, estimates the effect of changes in relative tick size due to tick size crossing event on HFT activity. The control variables are the one-day lagged (t –
1) values of VOLATILITY, LIQUIDITY, and SIZE. VOLATILITY is measured by difference between the highest and lowest price of the day, divided their average prices
(Equation 3.6); LIQUIDITY is measured using the Corwin and Schultz (2012) high-low spread (Equation 3.7); and SIZE is the natural log of market capitalisation,
respectively. All models are controlled for firm and day fixed-effects. Data are winsorised at three standard deviations (3-sigma) from their respective means. The standard
errors are calculated using the Huber/White/sandwich estimator of variance method.
Panel I: UPWARDS (RTSSMALL → RTSLARGE)
DTREATMENT 9.327 -1.319 -0.029 -0.190 44.759 -0.630** -0.068 0.070
(15.256) (1.370) (0.054) (0.250) (69.166) (0.257) (0.058) (0.169)
DPOST 4.813 6.061*** 0.185 -1.579*** -439.900* -1.322 -0.070 -0.659***
(25.301) (1.245) (0.239) (0.451) (233.366) (1.136) (0.072) (0.187)
DTREATMENT×DPOST -6.129 -2.946** -0.201*** -0.569** -163.069 -1.579*** 0.050 0.057
(17.982) (1.172) (0.064) (0.243) (115.916) (0.565) (0.075) (0.120)
VOLATILITY -5.195*** -0.3128*** 0.019 0.088* -41.173 -0.126 -0.004 -0.076
(1.680) (0.109) (0.015) (0.050) (25.737) (0.106) (0.013) (0.055)
LIQUIDITY 4.148* 0.642*** -0.011 0.002 38.191 0.107 -0.015 0.008
(2.095) (0.224) (0.012) (0.056) (47.034) (0.179) (0.022) (0.045)
SIZE 4.977 -6.643*** 0.578*** 1.293*** 543.569** -0.327 0.028 -0.047
(17.156) (0.990) (0.126) (0.440) (243.933) (1.161) (0.093) (0.319)
Constant 176.045 117.477*** -12.532*** -28.391*** -11230.266** 2.741 -0.579 1.301
(373.188) (22.129) (2.707) (9.565) (5231.015) (24.937) (1.984) (6.947)
Observations 1,559 1,559 1,559 1,559 494 494 494 494
R-squared 0.538 0.463 0.770 0.364 0.279 0.320 0.233 0.130
110
Panel II: DOWNWARDS (RTSLARGE → RTSSMALL)
DTREATMENT -1.532 0.480 -0.156** -0.280 104.810 -0.804 -0.045 -0.021
(11.112) (1.081) (0.075) (0.253) (187.926) (0.664) (0.052) (0.160)
DPOST 0.629 2.169 -0.376 -0.416 -3036.036*** -6.187*** -0.024 -0.074
(19.113) (4.157) (0.254) (0.363) (695.777) (1.561) (0.093) (0.120)
DTREATMENT×DPOST -3.379 0.634 0.476*** 0.902*** 527.724 1.653*** 0.290** 0.129
(8.547) (1.153) (0.063) (0.234) (335.389) (0.425) (0.115) (0.182)
VOLATILITY -4.009*** -0.110 0.055*** 0.098*** -74.585*** -0.155 -0.007 -0.058***
(1.217) (0.215) (0.008) (0.024) (22.830) (0.113) (0.010) (0.019)
LIQUIDITY 3.212* 0.409 -0.004 -0.071 -69.729 0.193 0.014 0.071
(1.845) (0.261) (0.013) (0.046) (43.028) (0.121) (0.022) (0.053)
SIZE 12.202 -5.031*** 0.317*** 0.646 708.912* -2.758* 0.066 0.072
(13.140) (1.741) (0.089) (0.406) (377.149) (1.469) (0.075) (0.123)
Constant 16.310 49.289 -7.293*** -14.650 -12183.111 57.997* -1.373 -1.370
(292.432) (39.032) (2.003) (9.065) (7996.006) (32.197) (1.655) (2.691)
Observations 3,563 3,563 3,563 3,563 1,581 1,581 1,581 1,581
R-squared 0.561 0.452 0.768 0.347 0.211 0.226 0.189 0.101
111
Therefore, the UPWARDS event is considered as the catalyst that causes a further
decline in the HFT activity of Treatment stocks, which widened the gap in HFT
activity between the Treatment and Control groups. Overall, the results confirm that
a steep increase in relative tick size would substantially impede HFT activity on both
the ASX and CHIX markets.
Panel II provides the outcomes for the DOWNWARDS scenario, which

illustrates the event where relative tick size decreased substantially as a result of tick
size crossings. The coefficients for DPOST×DTREATMENT are found to be
significantly positive in HFO (β = 0.476) and HFOR (β = 0.902) using the ASX
dataset, and in ALGO (β = 1.653) and HFO (β = 0.290) using the CHIX dataset. In
addition, these HFT measures are shown to be negatively related with
DTREATMENT, suggesting that Treatment stocks have less HFT activity than
Control stocks in the DOWNWARDS event as well. Therefore, the positive
coefficients in the interaction term imply that the DOWNWARDS event is the
turning point that leads the Treatment stocks to exhibit greater HFT activity than the
Control stocks. Essentially, these findings reveal that a considerable reduction in
relative tick size would greatly stimulate HFT activity in the Australian equity
market.
Using the ASX dataset, the regression results of the control variables show
that: (i) stocks with higher volatility would have significantly lower MTR and
ALGO, but higher HFO and HFOR; (ii) stocks with wider spread would have
significantly higher MTR and ALGO; and (iii) stocks with larger capitalisation
would have significantly lower ALGO, but higher HFO and HFOR. The findings for
the CHIX dataset indicate that: (i) more volatile stocks would have lower MTR and
HFOR; (ii) the bid-ask spread has no significant influence on any of the HFT
measures used; and (iii) larger stocks would have higher MTR but lower ALGO.
In conclusion, the findings indicate that there is an inverse relationship

between HFT activity and relative tick size, and this is evident on both the ASX and
CHIX datasets. The findings indicate that HFTs prefer the order-undercutting
strategy, which enables them to capitalise on their speed while mitigating adverse
selection risk; hence, a finer pricing grid is always desired. Market-makers would
112
generate a higher return with a coarser pricing grid; however, this is only attainable
with the order-queuing approach, which HFTs generally avoid since it negates their
speed advantage and increases their risk exposure. These results also suggest that
HFT activity can be influenced through regulatory intervention by adjusting nominal
tick size values, or through corporate actions by artificially altering stock prices
using forward or reverse splits.
3.7 Conclusion
Using data from the Australian equity markets, this research investigates the impact
of relative tick size on HFT activity as measured by MTR (quoting intensity), ALGO
(algorithmic liquidity provision), HFO (high-speed OrderIDs), and HFOR (HFO-
contributed message). In this study, HFT activity in stocks with relative tick sizes
classified as small (RTSSMALL) or large (RTSLARGE) was compared. The relative tick
size values are directly influenced by nominal tick size and stock price; thus, these
parameters are used to identify three distinct scenarios, which serve as the
foundation to verify the hypotheses proposed in this study.
The first hypothesis postulates that, “stocks with larger relative tick sizes will
have significantly higher levels of HFT activity due to greater order-queuing activity
compared to stocks with smaller relative tick sizes.” Nonetheless, the study shows
that smaller relative tick sizes result in greater HFT activity on both ASX and CHIX
markets. Even though this is evident in the one-cent tick size category, however, the
effect on most HFT measures in the half-cent tick size category is insignificant. This
is because stocks in the RTSSMALL group of the half-cent category have an almost
identical relative tick size to stocks in the RTSLARGE group of the one-cent category.
Therefore, their relative tick size is not small enough to attract HFTs. From the
literature, HFTs prefer stocks with a very small relative tick size for three reasons: (i)
the availability of many price levels allows HFTs to undercut existing limit orders by
posting slightly better prices (Werner et al., 2019); (ii) slower traders hesitate to
place limit orders for fear of being undercut, resulting in a shorter order queue
(Angel, 2011); and (iii) the speed of HFTs enables them to enter and exit the market
rapidly as required, without worrying about getting stuck in a lengthy order queue
upon re-entry. Overall, the findings of the study contradict the first hypothesis as the
113
results show that HFTs favour stocks with very small relative tick sizes, suggesting a
preference for an undercutting strategy.
The second hypothesis states that “stocks priced slightly above the A$2.00
tick size border will have significantly higher levels of HFT activity due to their
larger relative tick sizes compared to stocks priced slightly below the A$2.00 tick
size border.” As stock prices approach the A$2.00 tick size border, the relative tick
size disparity between RTSSMALL and RTSLARGE widens, signifying a greater
incentive for order-undercutting (minimising risk) and order-queuing (maximising
profit) activities. This scenario leads to greater HFT activity in the RTS SMALL group
on both the ASX and CHIX markets. However, when prices move farther away from
the tick size border, the disparity between RTSSMALL and RTSLARGE lessens, hence
reduces the incentive for HFTs to participate. On the ASX, this situation yields
conflicting results. It implies that some HFTs still choose the order undercutting
approach, while others take advantage of the ASX’ demographics by trading with
uninformed traders. This benefits them from a reduced adverse selection risk. In
contrast, due to CHIX’s demographics, market-making HFTs face a greater adverse
selection risk. This forces them to constantly choose stocks with a small relative tick
size. In the absence of such stocks, market-making HFTs stop using relative tick size
as a parameter in their trading algorithms. Overall, the findings are in contrast to the
proposed hypothesis, and show that HFTs favour order undercutting over order-
queuing, especially when informed traders are present in large numbers and the
relative tick size is sufficiently small. This suggests that HFTs are risk-averse and
prioritise risk reduction over profit maximisation.
The third hypothesis suggests “crossing a tick size border in an upward

(downward) direction results in a statistically significant increase (decrease) in the
level of HFT activity for the affected stock.” The analysis shows that when relative
tick size crosses the A$2.00 tick size border in an upward direction, HFT activity
declines substantially. Conversely, when relative tick size plunges due to a
downward crossing direction, HFT activity increases significantly. These results are
consistent across the ASX and CHIX markets. It indicates that HFTs prefer stocks
with fine pricing grids and would forsake them if their grids become coarse. The
inverse relationship between HFT activity and relative tick size demonstrates the
114
risk-averse character of market-making HFTs. This evidence also highlights the
importance of order-undercutting techniques for market-making HFTs, which is
attributable to their extremely low tolerance for adverse selection risks.
In conclusion, this study’s results challenge the idea that stocks with large
relative tick sizes have greater HFT activity. Instead, the findings support the view
that stocks with small relative tick sizes have higher HFT activity. While previous
studies, primarily focused on the U.S. markets (e.g., O’Hara et al., 2019; Yao & Ye,
2018), suggested otherwise, differences between the Australian and American
markets’ architecture may explain the inconsistent findings. In Australia, market-
making HFTs incur lower costs due to less market fragmentation and more effective
risk management when choosing stocks with smaller relative tick size. In addition,
the lack of designated market makers reduces competition for liquidity provision,
allowing HFTs to capture a larger share of revenue from such activities. This is most
evident in high-priced stocks since their relative tick size is extremely small and thus
more attractive to market-making HFTs. On the CHIX market, where informed
traders are more present, HFTs face increased pressure to monitor their risk exposure
closely, and are compelled to consistently employs order undercutting strategies.
Moreover, the study shows that when a stock’s relative tick size decreases, HFT
activity increases while it decreases when a stock’s relative tick size increases.
115
CHAPTER FOUR:
ESSAY THREE
EXPECTED VOLATILITY, HIGH-FREQUENCY TRADING, AND
LIQUIDITY
4
This chapter demonstrates how expected volatility influences HFT activity, and how
the resulting shift in HFT activity impacts liquidity. The chapter starts with the
introduction of the essay, which covers the background of the study and research
motivation. The chapter continues with a literature review and hypotheses
development, followed by methodology, findings, discussions, and conclusion.
4.1 Introduction
The CBOE Volatility Index (VIX), commonly known as the “fear gauge” or “fear
factor”, is a forward-looking index that represents the unforeseen future market
fluctuations that investors anticipate (Hull, 2018). The original VIX was developed
by Whaley in 1993 for the Chicago Board Options Exchange (CBOE) to measure the
expected volatility of the S&P 100 Index (OEX) using 30-day at-the-money options,
and this characteristic distinguishes the VIX from other stock market indices that
monitor prices. In 2003, the CBOE revised the definition and computation of the
VIX and replaced the OEX with the S&P 500 Index (SPX) as its underlying index.
The new VIX is calculated using a weighted average of SPX puts and calls with a
broad range of strike prices over the next 30 days.33 Several stock exchanges across
the world have adopted a similar methodology to track near-term expected volatility
in their respective markets, including the VSTOXX in Eurozone, AXVI in Australia,
VHSI in Hong Kong, and INVIXN in India, underlined by the EURO STOXX 50,
S&P/ASX 200, Hang Seng Index, and NIFTY 50, respectively.
The concept of implied volatility, which forms the basis of the VIX index,
reflects investors’ expectations of near-term future volatility in the stock market,
with higher levels of the index suggesting greater investor concern over future
economic prospects. Consequently, the indicator can also measure stock market
33
The formula for calculating the VIX is detailed in https://cdn.cboe.com/resources/vix/vixwhite.pdf.
116
sentiment (Reilly & Brown, 2012). Whaley (2009) notes that the rates of change in
the VIX and the SPX are asymmetric and observes that investors’ fear of a bearish
market dominates their excitement (or greed) in a bullish market. This behaviour
also suggests that a high VIX reflects a heightened degree of fear over the possibility
of a market crash, while a low VIX indicates higher investor confidence.
Similarly, Ang, Chen, and Xing (2006) find that the behaviour of investors
towards upside potential and downside risk is asymmetrical, where the latter is
shown to have a relatively greater influence on stock prices. This pattern suggests
that investors are more sensitive to downside risks than upside opportunities;
therefore, the magnitude of market reactions is expected to be greater when the VIX
is high (pessimistic) than when it is low (optimistic). In addition, Savor (2012)
shows that when the VIX is high, the ratio of no-information to information-based
price shocks is greater. His finding suggests that news concerning firm fundamentals
has a lower impact on stock prices during periods of high uncertainty and that no-
information price shocks are more frequent during such times. Savor (2012) posits
that stock prices might be more sensitive to investor sentiments and liquidity shocks
during pessimistic periods. For the reasons mentioned above, it is prudent for
investors to take precautionary measures to safeguard the value of their assets
whenever the VIX signals high market uncertainty. This purpose may be attained by
purchasing options contracts or engaging in short selling to take advantage of the
elevated fear sentiment, which serves as an insurance policy to protect the portfolio’s
value.
On the other hand, investors may choose the flight-to-quality strategy, in

which their high-risk market positions are liquidated, and the proceeds are invested
in relatively safer financial instruments34 (see, e.g., Amihud & Mendelson, 2012;
Sarwar, 2017; Troster, Bouri & Roubaud, 2019). In a radical approach, pessimistic
investors may liquidate their financial assets and abandon the market until the
situation improves. A herding behaviour from such action might create excessive
selling pressure in the market, which consequently would cause the stock market
34
For instance, blue-chip stocks, futures contracts for precious metals such as gold and silver, and
Treasury bills.
117
index to plummet. Furthermore, the decision to remain on the sidelines would
immediately diminish market liquidity, exacerbating an already fragile market. In
theory, the VIX should have no direct effect on the liquidity of a stock market.
Regardless, the arguments mentioned above imply that the fear index may
substantially influence market participants’ expectations and behaviours, mainly
when sentiment is negative (i.e., when the VIX is high), which results in the
deterioration of market liquidity.
4.1.1 Volatility index in Australia

In Australia, the ASX introduced its volatility index, the S&P/ASX 200 VIX
(AXVI), on September 23, 2010. The end-of-day index tracks the settlement prices
of S&P/ASX 200 index’s (AXJO) put and call options to calculate a weighted
average of the implied volatility of the options.35 Similar to the original VIX, the
AXVI captures the sentiment of market participants on the expected level of
volatility in the Australian equity market in the near future. Higher values suggest
that the market anticipates substantial changes in the underlying index over the next
30 days, whereas lower values imply greater market confidence. Figure 4.1 shows
the weekly average of AXJO and AXVI values since the latter’s inception in
September 2010. The graph demonstrates that the two indices appear to be
negatively correlated, with high AXVI values corresponding to low AXJO levels and
vice versa. This inverse relationship suggests that when market sentiment is negative
(positive), selling pressure outweighs buying pressure, resulting in lower (higher)
AXJO values.
In addition, Edwards and Preston (2017) propose that even a straightforward

interpretation of the VIX may help forecast future volatility, such as defining a VIX
level below 12 as “low”, above 20 as “high”, and values in between as “normal”.
Based on these threshold levels, it is expected that the underlying index would
fluctuate by either ±3.5% (low VIX) or ±5.8% (high VIX) after one month. 36 Figure
35
The following link provides further information on the AXVI computation:
www.spglobal.com/spdji/en/documents/methodologies/methodology-sp-asx-200-vix.pdf.
36
This value is calculated using the following formula: 𝑆&𝑃500 = 𝑉𝐼𝑋 ÷ √12
118
4.2 illustrates the distribution of AXJO price changes over the next 21-day trading
period, classified by the previously specified threshold values.37
6500.00 45.00
40.00
6000.00
35.00
5500.00
30.00
5000.00 25.00
20.00
4500.00
15.00
4000.00
10.00
3500.00 5.00
AXJO Close AXVI Close
Figure 4.1: Weekly average of S&P/ASX 200 index (AXJO) and S&P/ASX 200
VIX index (AXVI)
AXVI Below 12 (Low) AXVI Between 12 and 20 (Normal) AXVI Above 20 (High)
35%
30%
25%
20%
15%
10%
5%
0%
-14% -12% -10% -8% -6% -4% -2% 0% 2% 4% 6% 8% 10% 12% 14%
Figure 4.2: Variation in the AXJO’s price over the next 21 trading days (to the
nearest 2%)
37
The duration of 21 trading days corresponds to approximately 30 calendar days, which is also the
projected horizon for the VIX index.
119
The diagram indicates that observations from the group with a high VIX have a
thicker tail, implying a larger likelihood of large price swings (6% or more) than when
the VIX is normal or low. Assuming the current value of AXJO is 5,000 points, the
index will change by at least 300 points in either direction over the next 21 trading
days, with a probability of 20%, 13%, and 9%, respectively, when the VIX is high,
normal, and low.38 Therefore, although the AXVI may not accurately predict the
direction of future changes in the AXJO, it may be possible to forecast the magnitude
of future changes in its underlying index. This information may help investors time
their market entry and exit, allowing them to make better investment decisions.
On February 28, 2013, the AXVI was made available in real-time due to the
rise of volatility as an asset class in response to the global economic uncertainties.
This action allows the ASX to issue a range of financial instruments, providing
market players with greater opportunities to trade, diversify, and hedge their
portfolios against volatility. Moreover, the real-time AXVI supplies the Australian
equity market with continuous information on expected stock market volatility,
thereby increasing the informativeness of the stock market. The new feature is
expected to be well-received by HFTs and other market players who depend heavily
on real-time, short-lived information. In addition, the real-time AXVI would allow
investors to develop trading strategies based on market sentiment, which may be
especially relevant on volatile days due to the increased amount of information.
4.2 Literature review and hypotheses development

4.2.1 Liquidity
Liquidity is defined as the ability to trade a stock quickly and at a low cost; hence,
price movement and market viability rely on sellers’ and buyers’ ability to satisfy
their trading desires (O’Hara, 1995). Similarly, Sundaresan and Wang (2009)
describe liquidity as the easiness with which a substantial amount of a security may
be exchanged in the market in a short period of time without triggering an adverse
price reaction. Liquidity is regarded as an important trait of capital assets and has a
significant impact on their pricing, and studies find that it is priced in the cross-
38
See Table A1 in the Appendix for further information on the data used to construct the graph shown
in Figure 4.2.
120
section of stock returns (Acharya & Pedersen, 2005; Amihud & Mendelson, 1991).
Despite its importance, “liquidity” is difficult to define and cannot be directly
observed owing to its multifaceted nature; hence, it cannot be represented by a single
measure (Amihud, 2002). There are three widely recognised dimensions of market
liquidity: depth, resilience, and tightness (see, e.g.: Duan & Zou, 2014; Kyle, 1985).
However, as noted by Sar and Lybek (2002), these attributes may overlap to some
extent. In empirical research, liquidity measures are often employed as proxy for
transaction cost, price impact, and trading activity, which may represent one or more
of the aforementioned dimensions.
Transaction costs are the costs associated with the execution of a trade. Some
of the costs, such as taxes and commission fees, are known in advance (i.e., explicit
costs), but others, such as the bid-ask spread, price impact and opportunity cost, may
be less apparent (i.e., implicit costs) yet account for a significant portion of the
overall transaction cost. Therefore, liquidity measures such as the high-low spread of
Corwin-Schultz (2012) and the illiquidity ratio of Amihud (2002) are often used to
reflect a different aspect of transaction costs, which are bid-ask spread and price
impact, respectively. Madhavan (2000) suggests that the bid-ask spread may also
represent the cost of immediacy paid by traders to market makers in order to have
their orders executed quickly. Volume-based liquidity measures used the volume of
transactions to measure liquidity and capture the breadth and depth of a market or
securities (Sarr & Lybek, 2002). Since a transaction occurs when the bid and ask
prices are matched, a tighter (wider) spread would signify a higher (lower) trading
volume, and this relationship suggests that spreads and volume are negatively linked.
In contrast, Easley and O’Hara (1992) argue that there would be no trade if there
were no information event, either good or bad. This scenario represents a “safer”
trading environment, allowing market makers to post a tighter spread. In contrast, the
greater the volume, the greater the probability that the market maker assumes an
information event has happened, and consequently, they post a wider spread. In any
case, these arguments show a positive relationship between volume and spreads,
with higher volume being associated with wider spreads and vice versa.
121
Market makers act as liquidity providers due to their role in
facilitating immediate trade execution by matching buy and sell orders, which is
performed through frequent limit orders posts. Due to this operating mechanism,
market makers may suffer loss when trading with investors who possess price-
sensitive private information, exposing them to adverse selection risks. In addition,
market makers must maintain a certain level of stock inventories in order to quickly
match the standing limit orders in the market. This factor exposed them to the risk of
unanticipated future changes, such as the direction of stock prices and the time
required to have their limit orders executed. Consequently, the bid-ask spread posted
by market makers should reflect the risks they must assume to make the market in
addition to the profits earned by such activity. During periods of heightened
uncertainty (e.g., high VIX), investors may choose to leave the market en masse or
refrain from trading, both of which diminish market liquidity. The importance of
having market makers increases during such periods, but so does the risk they must
assume, which results in wider bid-ask spreads. The scenario may be seen as the fear
index’s indirect influence on liquidity, which is channelled through the index’s
influence on market participants’ expectations and behaviours, particularly when
sentiment is negative.
4.2.2 High-frequency trading strategies

The evolution of technology transforms how information is transmitted, allowing
“hard” information such as stock market data and corporate financial statements to
be received and processed faster.39 The extensive use of high-speed computers,
massive database programmes, and low-latency networking have amplified trading
activities reliant on hard information, such as those practised by the HFTs (Liberti &
Petersen, 2019). HFTs are notably distinct from other types of investors due to their
lightning-fast operations, making it impossible for other players to replicate their
trading strategies (Harris, 2013).
39
This also includes data obtained from publicly available indexes, such as the volatility indices
stated previously.
122
Menkveld (2014) asserts that HFTs have a competitive advantage over
human traders due to the following factors: (1) the ability to identify profitable
trading opportunities instantly; (2) the capacity to process newly arrived information
and execute the necessary actions rapidly; and (3) the ability to evaluate textual
context from machine-readable news. Scholtus, van Dijk, and Frijns (2014)
investigate the importance of speed to the profitability of news-based trading
strategies using U.S. datasets. According to their findings, a delay of 300
milliseconds or more severely impairs the returns of such a strategy, with the
reduction being more pronounced on days with high-impact news and high volatility.
Their results demonstrate how crucial speed is to the profitability of short-term
traders such as the HFTs, and how its importance escalates during periods of high
uncertainty. For these reasons, HFTs spend a fortune on technological infrastructures
that minimise latency and hire quantitative analysts (“quants”) to develop trading
algorithms for their automated trading operations. 40 Due to this high entry barrier,
sophisticated market players such as full-fledged HFT firms, proprietary trading
desks, and hedge funds are the predominant users of HFT strategies (Goldstein et al.,
2014).41
In theory, HFT strategies may have both beneficial and detrimental effects on
liquidity due to their trading flexibility – a stronger presence on the supply side
would increase the order book’s depth, resulting in more liquidity and narrower
spreads, whereas a stronger presence on the demand side would rapidly deplete
standing limit orders, thereby draining liquidity from the market and widening
spreads. In general, HFTs provide liquidity when they assume the role of market
maker, using their speed advantage to swiftly update quotes and profiting from the
spread between the bid and ask prices. This method requires HFTs to place limit
orders on both sides of the book, therefore supplying liquidity to the market (Chung
& Lee, 2016). Their market-making activities also contribute to market efficiency
and stability; they sell when others prefer to buy and purchase when others wish to
40
Since the nature of the game is winner-take-all, HFTs continuously upgrade their technologies to
remain competitive (Kauffman et al., 2015; NAFM, 2010). This aspect leads to an arms race between
HFTs, which is criticised for wasting resources without meaningful improvements to market quality.
The race is seen as counterproductive, and it is debatable whether HFTs add value to the broader
market (Chordia et al., 2013; Jones, 2013).
41
Full-fledged HFT firms in Australia include Citadel Securities, Jump Trading, Liquid Capital,
Susquehanna International Group, and Virtu Financial.
123
sell (Angel, 2014). Moreover, Boehmer et al. (2018) demonstrate that the
competition between market-making HFTs lowers both the permanent and
temporary price impact of trades, resulting in reduced short-term volatility. Similar
results are reported by Hagstromer and Nordén (2013), who find that market-making
HFTs decrease intraday volatility by providing liquidity.
In addition, Benos and Sagade (2016) find that passive HFTs are liquidity
suppliers while aggressive HFTs are liquidity takers, and that their liquidity
taking/making behaviour does not change substantially over time. This distinction
results in a wide variety of HFT strategies, which translates to varying sensitivity
towards recent order book and inventory level changes. On the other hand, Aldridge
(2013) asserts that the order execution algorithms employed by HFTs determine
whether their strategy is aggressive or passive – an order placed at or near the current
market price is perceived as aggressive, whereas a limit order placed far away is
regarded as passive.42 Some exchanges even employ the maker/taker pricing model,
in which passive orders are subsidised with “rebates” and aggressive orders are
charged “fees”.43,44 Studies find that the absence of rebate might put market-making
HFTs in a loss position, notably when spreads are narrower. Consequently, HFTs
may be discouraged from engaging in market-making activities (Brogaard et al.,
2014; Hendershott & Riordan, 2013).
Although HFTs’ market-making is generally beneficial to the market, their

reliability as a liquidity source is debated, particularly during times of stress, since
they are not obligated to make the market like a traditional market maker. This
absence of responsibility raises concerns that HFTs might leave the market at their
convenience, for instance, when participation is no longer profitable (Anand &
Venkataraman, 2013; Carrion, 2013). According to O’Hara (2014), spreads are
narrower as a result of HFTs’ intermarket arbitrage, which moves liquidity from a
market with a surplus of liquidity to a market with a shortage of liquidity.
42
A market order, also known as a marketable order, is an order placed at the current market price.
This order gives the sender price-priority, allowing it to be executed as quickly as possible; hence, it
is sent when a trader requires an immediate order match.
43
Electronic limit order books depend on the voluntary supply of liquidity; thus, exchanges must
incentivise traders to place limit orders, resulting in the adoption of the maker-taker pricing model.
44
Alternatively, the maker-taker pricing model may offer asymmetric trading fees, with lower
transaction costs for liquidity suppliers and higher trading fees for liquidity takers (see Durbin, 2010).
124
Nevertheless, HFTs may use illegal yet difficult to detect predatory algorithms.
VanKervel and Menkveld (2019) observe that although retail investors may benefit
from a narrower bid-ask spread, institutional investors complain that the presence of
HFTs increases their execution costs.45 Their research shows that when HFTs
initially enter the market, they operate as a liquidity provider by trading against
(against-wind) large institutional orders. However, they rapidly switch roles and
become liquidity demanders by trading with (with-wind) the large orders. As a result,
the strategic exploitation of HFTs increased the execution costs for institutional
traders supplying liquidity, making it costly for the latter and profitable for the
former.
VanKervel and Menkveld (2019) claim that despite the fact that HFTs’ with-
wind trading may make prices more efficient in the short term, it may cause prices to
be less efficient in the long run. The greater execution costs inflicted on institutional
investors may discourage them from conducting costly analyst research, since the
informational rents generated from such effort are forcibly shared with others, thus
reducing the information trader’s profits. Overall, the practise of order anticipation
by HFTs is considered parasitic because it neither contributes to price discovery nor
liquidity, while unnecessarily increasing the execution costs of large liquidity
providers and negatively impacting their trading strategy. Furthermore, it may reduce
the incentive for large traders to acquire information (Agarwal, 2012; Grossman &
Stiglitz 1980, Harris, 2015; Stiglitz 2014, Weller 2018).
Goldstein et al. (2020) show that HFTs, on average, supply depth on the
thick side of the order book, but demand depth from the thin side of the order book.
This behaviour suggests that HFTs provide liquidity when it is less necessary and
abstain from doing so when it is critical. Goldstein et al. (2020) argue that HFTs’
order placement is more consistent with order anticipation strategies, and this is
more apparent during volatile periods and when trading is faster. 46 Hirschey (2020)
45
Institutional investors use algorithms to split their bulk orders into multiple small orders and feed
them to the market sequentially. This is done to conceal their originally large transactions from other
traders, hence lowering their trade’s price impact and execution costs.
46
Order anticipation is regarded as a “very harmful” trading strategy and a form of front-running
(Harris, 2013, 2015). This strategy operates by monitoring trades and quotes to identify algorithms
125
reported findings suggesting that HFTs could trade ahead of non-HFT order flow,
which is made possible because, (i) HFTs may identify the persistent order flow of
informed non-HFTs, and (ii) liquidity providers are slow to update their quotes.
Therefore, liquidity providers are forced to trade fewer shares than desired, or have
their orders fulfilled at a worse price.
Zhang (2017) shows that HFTs can react more quickly and aggressively than
non-HFTs to hard information shocks, allowing them to take strategic positions and
profit from the shock. Her finding implies that HFTs play a vital role in short-term
price discovery, especially on days with an incredible amount of hard information,
such as when the market is volatile. Similarly, Brogaard, Hendershott, and Riordan
(2014) demonstrate that the trading direction of HFTs is correlated with publicly
available information such as macroeconomic news releases, market-wide price
fluctuations, and limit order book imbalances. Riordan and Storkenmaier (2012)
investigate the 2007 upgrade of Deutsche Boerse’s Xetra systems, which reduced the
latency of the trading system from 50 milliseconds to 10 milliseconds. The decreased
latency should be irrelevant to human traders as it transcends their time
perception; however, HFTs may benefit since their strategies are time-sensitive, and
the exchange’s infrastructure upgrade would offer a faster trading environment for
HFTs’ algorithms to prosper. They observe that both the quoted and effective
spreads decreased in the post-upgrade period, which is probably attributable to
HFTs’ activity. Similarly, the 2013 launch of the real-time AXVI index may have a
stronger impact on HFTs and other traders with a short trading horizon. The new
index would feed the market with real-time, publicly accessible information,
allowing HFTs to take strategic positions and profit from any opportunities resulting
from the market’s immediate response towards fluctuations in the VIX levels.
Furthermore, even though HFTs are emotionless and do not experience

“anxiety” or “fear,” and can only execute the algorithm they were programmed, their
trading decisions are susceptible to order-book information. Consequently, the fear-
deployed by large traders, then trading ahead of them to profit from the anticipated direction of price
movements.
126
driven activities of other investors amidst periods of high uncertainty may
substantially influence the behaviour of HFTs. In situations such as the Flash Crash,
which sees an extreme fluctuation of the VIX index, some HFTs are reported to
cease trading to avoid the risk of enormous losses, leaving other investors with fewer
players to take the opposite side of their orders (Patterson, 2010). Moreover, unlike
designated market makers (DMMs), HFTs operating as electronic liquidity providers
(ELPs) have neither the responsibility nor the commitment to remain within an
unfavourable market. This factor makes them a less credible source for liquidity,
which may aggravate execution uncertainty (Anand & Venkataraman, 2013; Chung
& Chuwonganant, 2018; Zhang, 2010). Furthermore, in markets where they play a
critical role in supplying liquidity, their sudden absence during volatile times might
result in a severe liquidity shortage, leading to extreme market movements similar to
the Flash Crash (Gomber et al., 2011).
Therefore, the positive effect from HFTs’ activity may not persist during
fearful periods, and the pessimism of other investors may adversely impact HFTs’
influence on the market, driving them to switch from being liquidity suppliers to
liquidity takers. In addition, the absence of a need to make market enables HFTs to
adopt a range of strategies; hence, their role in the stability and pricing efficiency of
markets, particularly during periods of heightened uncertainty, is regarded with
scepticism (Brogaard, Hendershott and Riordan, 2014). Regardless, everyone has the
right to be presumed innocent until proved guilty, which motivated the researcher to
conduct further investigation into the allegations made against HFTs, particularly
concerning their influence on market liquidity during periods of negative
sentiment.47
4.2.3 Hypotheses development

According to Liberti and Petersen (2019), the emergence of high-speed computers,
high-quality databases, and low-latency networking has led to trading activity that
largely depends on hard information, such as the strategies used by HFTs. Zhang
(2017) shows that HFTs play an important role in short-term price discovery,
47
This statement is based on the presumption of innocence outlined in article 14(2) of the
International Covenant on Civil and Political Rights (ICCPR), which was retrieved from
www.austlii.edu.au/au/other/dfat/treaties/ATS/1980/23.html.
127
particularly on days with an enormous amount of hard information. For example,
when the market is volatile, HFTs can rapidly update quotes and earn from the
difference between the bid and ask prices (Chung & Lee, 2016). This promotes
liquidity and improves market efficiency and stability (Angel, 2014). In Australia,
stock market participants have continuous access to information about expected
volatility due to the availability of real-time AXVI. Adding real-time features to the
AXVI is crucial for HFTs and other fast market players with a short trading horizon.
It enables them to formulate trading strategies based on market sentiment and profit
from any opportunities resulting from fluctuations in the VIX levels (Brogaard et al.,
2014). This innovation should have a substantial influence on HFTs on the
Australian market as they would have more alternatives for their trading strategy.
While the real-time AXVI introduction event (VIXRT) would have no direct impact
on liquidity, it may indirectly affect liquidity via its relationship with HFTs.
Consequently, the effect of the VIXRT event on HFTs’ influence on liquidity may be
amplified or diminished. Based on the argument, the following hypothesis is
proposed:
H1: The introduction of real-time AXVI (VIXRT) have a positive impact on HFT
activity and indirectly increase liquidity through its relationship with HFTs.
Although HFTs lack emotions, their trading decisions are driven by order-book data
and may be indirectly affected by negative sentiment through the trading activity of
emotional traders. For instance, on days when the VIX is extremely high and
widespread intraday price swings emerge, some investors may leave the market in a
panic or refuse to participate at all. This can influence HFTs’ trading strategies.
Moreover, market-making HFTs are not legally required to maintain liquidity at all
times and may also leave the market if the situation is undesirable. This
opportunistic behaviour may exacerbate the already fragile liquidity situation in the
market during such times (Anand & Venkataraman, 2013; Carrion, 2013), calling
into question the credibility of HFTs as liquidity providers. In periods of high
volatility and rapid trading, HFTs’ order placement behaviour aligns more closely
with order anticipation strategies (Goldstein et al., 2020). This approach preys on
other players, particularly large traders whose execution costs would be
unnecessarily inflated (Agarwal, 2012; Harris, 2015), and does not contribute to
128
price discovery or liquidity. As a result, HFTs’ strategies may change on high VIX
days, affecting the market’s liquidity. This research makes the following hypothesis
founded on these arguments:
H2: As VIX levels increase, HFT activity changes in a way that exacerbates the
already fragile liquidity situation in the market, calling into question the credibility
of HFTs as liquidity providers.
4.3 Methodology
4.3.1 Data description
This research employs order book information supplied by the Securities Industry
Research Centre of Asia-Pacific (SIRCA). Only S&P/ASX 100 (ASX: XTO)
constituents are sampled to ensure that the selected stocks are reasonably large and
liquid.48 The list of XTO’s constituent stocks is retrieved from Thomson Reuters
Datastream and is reviewed on a monthly basis.
4.3.2 Sample selection

The first hypothesis suggests that the introduction of real-time AXVI (VIXRT) on
February 28, 2013, have a significant effect on HFT activity, and influence the
relationship between HFT activity and liquidity. To verify this hypothesis, the study
examines observations that occur one year before and one year after the VIX RT event
day; thus, only data from firms that consistently appear on the XTO constituent list
during this time period are selected. Throughout this 24-month period, there are a
total of 114 unique firms; however, 34 firms were excluded due to incomplete data
issue, leaving the final sample with only 80 ﬁrms. 49 The selected observations are
subsequently categorised as “Pre-event” or “Post-event”.
48
The XTO index consists of the 100 largest firms listed on the ASX by float-adjusted capitalisation,
all of which are large and medium-sized corporations.
49
A stock is considered to have incomplete data if it meets the following criteria: (i) it does not
consistently appear in the XTO’s constituent list throughout the 24-month period surrounding the
VIXRT event; and (ii) it has too many missing data in either the ASX dataset only, the CHIX dataset
only, or both datasets.
129
The study further evaluates whether the observed difference (if any) persists
in the short-term (1 month), medium-term (6 months), and long-term (12 months)
following the VIXRT event. For this purpose, three different datasets covering the
aforementioned date ranges surrounding the VIXRT event are utilised, with each
dataset comprising the same number of cross-sectional units. The reason for
employing different observation lengths is to determine whether the effects (if any)
are: (i) evident only in the periods close to the event date, indicating a temporary
effect; (ii) evident across all tested periods, indicating a permanent effect; and (iii)
evident only in the extended period and not in the immediate period, suggesting a
slow adoption of the new features offered by the VIXRT.
Table 4.1 summarises the data used to test the first hypothesis. The table shows
that the average value of VIX in the post-event period of the 6 months dataset is higher
than it was prior to the event. In contrast, the 12 months dataset indicates the opposite;
the average pre-event VIX values are higher. This pattern implies that sentiment in the
first half of 2012 was relatively more pessimistic, which is evident by the greater
maximum VIX value of the pre-event period shown in Panel C of Table 4.1.
Table 4.1. Data description for the samples used to test the first hypothesis
This table briefly describes the dataset used to test the first hypothesis. Pre-VIXRT and Post-VIXRT
denote the groups of observations that occurred before and after the VIXRT’s introduction,
respectively. Panel A, Panel B, and Panel C represent the short-term, medium-term, and long-term
datasets, which span from February 1, 2013, through March 28, 2013, September 3, 2012, through
August 30, 2013, and March 1, 2012, through February 28, 2014, respectively. Observations, Mean
VIX, Maximum VIX, and Minimum VIX refer to the number of firm-day observations, and the average,
the highest, and the lowest VIX values, in the pre and post VIXRT periods, correspondingly.
Observations Mean VIX Maximum VIX Minimum VIX
Panel A: (1 month)
Pre-VIXRT 1,520 14.372 16.987 13.244
Post-VIXRT 1,680 15.231 18.303 13.669
Panel B: (6 months)
Pre-VIXRT 9,896 13.264 17.089 10.536
Post-VIXRT 10,230 15.454 21.675 12.476
Panel C: (12 months)

Pre-VIXRT 19,960 15.512 26.639 10.536
Post-VIXRT 20,272 14.578 21.675 10.689
This is also corroborated by the line chart of AXVI illustrated in Figure 4.1, which
show higher VIX values recorded during that period. Nevertheless, the average
values of VIX before and after the VIXRT event across all three datasets are
130
comparable, as their differences are small and within the “normal VIX” range
described in Edwards and Preston (2017). Therefore, it may be inferred that market
sentiment is less likely to influence the outcomes generated by these datasets, and
that any observable results are more likely to be driven by the additional features
available post-VIXRT.
The second hypothesis is formed on the premise that the levels of VIX and
HFT activity influence liquidity. This hypothesis is tested using data from all
available trading days since the introduction of the VIXRT, wherein all observations
possess similar real-time market information. The sample period extends until the
last trading day of December 2017, a total of 1,225 distinct observation days. This
approach retains the time-series characteristics of the original data and preserve the
continuity of observations within the dataset, resulting in a high number of firm-day
observations.50 Employing a large sample size reduces the margin of error and
increases the confidence level and power of the statistical tests used to confirm the
hypothesis, leading to more accurate conclusions. The trading days are then divided
into smaller groups according to their respective end-of-day VIX readings and HFT
activity levels. Based on the threshold value suggested by Edwards and Preston
(2017), all days with VIX values equal to or greater than 20 are labelled as ”High
VIX” (VIXH), or “Event”, whilst all other days are classified as ”Non-high VIX”
(VIXNH), or “Non-event”.51, 52
Moreover, two additional categories are formed based on their HFT activity
level, as measured by the total number of high-frequency orders ID (HFO).53
Observations with HFO values in the 90th percentile are classified as “High HFT”
(HFTH), or “Treatment”, whereas all other observations are designated as “Non-high
HFT” (HFTNH), or “Control”. The selected percentile values for VIX and HFO are
shown in Table 4.2.
50
This is in contrast to when samples consist of data from days with a high VIX and a low VIX only,
and exclude any observations that do not belong to either category.
51
This figure corresponds to the 91st percentile of the VIX.
52
Investors’ fear of a bearish market outweighs their greed in a bullish market (Whaley, 2009).
Therefore, this study considered “low VIX” days to be less important, and they were pooled with
other observations that also did not satisfy the threshold for “high VIX”. In the final sample, these
observations are designated as ”non-high VIX”.
53
The following section provides details on the definition and method used to determine HFO.
131
Table 4.2. Percentile values for VIXCLOSE and HFO54
This table displays the selected percentile values of the VIXCLOSE and HFO used to test the second
hypothesis. The reported figures are based on the longest study period used in this research, which
began on February 28, 2013, and concludes in December 2017 (1,225 trading days). The variables
VIXCLOSE, HFOASX and HFOCHIX, respectively, represent the end-of-day VIX value, and HFO values
generated from the ASX and CHIX datasets. For the VIXCLOSE, Observations refers to the number of
daily observations, whereas for the HFOASX and HFOCHIX, it refers to the number of firm-day
observations.
Percentile
Observations
5% 10% 25% 50% 75% 90% 95%
VIXCLOSE 1,225 10.953 11.378 12.466 13.995 16.688 19.566 21.984
HFOASX 121,777 9 14 26 56 162 493 805
HFOCHIX 119,090 0 0 0 0 2 15 34
Next, four subsample groups are formed to capture the possible combinations
of VIX and HFT, which are: (i) ControlNon-event (VIXNH-HFTNH); (ii) TreatmentNon-
event (VIXNH-HFTH); (iii) ControlEvent (VIXH-HFTNH); and (iv) TreatmentEvent (VIXH-
HFTH). The information relevant to these groups is summarised in Table 4.3.
Table 4.3. Description of the datasets used to test the second hypothesis
The table presents the data used to test the second hypothesis. Panel I describes the datasets for the
sample in the Main groups, which is formed based on the values of VIXCLOSE (end-of-day VIX),
HFOASX (HFO derived from the ASX dataset), and HFOCHIX (HFO derived from the CHIX dataset).
For the VIXCLOSE, days with VIX readings equal to or greater than 20 are labelled as VIXH (Event),
otherwise VIXNH (Non-event). For both HFOASX, and HFOCHIX, observations with HFO values in the
90th percentile are labelled as HFTH (Treatment), otherwise HFTNH (Control). The VIXH and VIXNH are
then paired with HFTH and HFTNH, creating four subgroups for each dataset, as shown in Panel II.
The subgroups are: (i) ControlNon-event (VIXNH-HFTNH); (ii) TreatmentNon-event (VIXNH-HFTH); (iii)
ControlEvent (VIXH-HFTNH); and (iv) TreatmentEvent (VIXH-HFTH). Observations refer to the number of
firm-day observations for their respective groups. Weight refers to the proportion of the total
observations that are attributable to the observations of a certain group. Mean VIXCLOSE, Mean
HFOASX, and Mean HFOCHIX are the average values for the end-of-day VIX, and the HFO values of
the ASX and CHIX datasets, respectively.
VIXCLOSE HFOASX HFOCHIX
Panel I:
VIXNH VIXH HFTNH HFTH HFTNH HFTH
Main groups
(Non-event) (Event) (Control) (Treatment) (Control) (Treatment)
Observations 111,098 10,683 109,581 12,196 106,948 12,142
Weight 91.23% 8.77% 89.98% 10.02% 89.80% 10.20%
Mean VIXCLOSE 14.16 23.31 14.87 15.82 14.92 15.2
Mean HFOASX 166.87 292.11 89.23 974.17 Nil Nil
Mean HFOCHIX 6.18 8.08 Nil Nil 1.33 50.55
VIXNH (Non-event) VIXH (Event)
Panel II: HFTNH HFTH HFTNH HFTH
Subgroups (Control) (Treatment) (Control) (Treatment)
VIXNH-HFTNH VIXNH-HFTH VIXH-HFTNH VIXH-HFTH
ASX (Obs. = 121,777)
Observations 100,558 10,536 9,023 1,660
Weight 82.58% 8.65% 7.41% 1.36%
Mean HFOASX 87.35 925.80 110.15 1,281.19
Mean VIXCLOSE 14.11 14.64 23.32 23.27
54
Details regarding HFO are provided in Section 3.5.3.1 (Essay Two) or Section 4.3.3.2. (Essay 3).
132
CHIX (Obs. = 119,090)

Observations 97,552 11,085 9,396 1,057
Weight 81.91% 9.31% 7.89% 0.89%
Mean HFOCHIX 1.33 48.84 1.29 68.44
Mean VIXCLOSE 14.11 14.42 23.31 23.43
These segmentations allow the researcher to determine whether: (i) the liquidity of
stocks during high VIX days (Event) significantly differs from stocks during non-
high VIX days (Non-event); (ii) the liquidity of stocks with high HFT activity
(Treatment) significantly differs from stocks with non-high HFT activity (Control);
(iii) the liquidity of stocks amongst the subgroups are significantly different
(ControlNon-event, TreatmentNon-event, ControlEvent, and TreatmentEvent); and (iv) the
influence of high VIX days (Event) on the effect of high HFT activity (Treatment)
on liquidity.
4.3.3 Measurement of variables

4.3.3.1 Liquidity measures
This study seeks to determine whether HFT activity affects liquidity and whether
high VIX days influence HFT’s effect on liquidity. Due to the multidimensional
nature of liquidity, the study employs several measures to gain a comprehensive
understanding of the impact of VIX and HFT on liquidity. Accordingly, three
different liquidity measures, namely the Corwin and Schultz (2012) High-Low
Spread, the Amihud (2002) Illiquidity Ratio, and the Volume Turnover, are
employed to capture the transaction cost, price impact, and volume dimensions of
liquidity, respectively. This study employs liquidity proxies derived using daily data
(i.e., low frequency) since such data are often widely available. As a result, the
methodology used in this study may be more reproducible and applicable to a range
of international markets over extended time periods.
1. Corwin-Schultz High-Low spread

The conventional bid-ask spread calculation requires bid and ask quotes data, which
are more difficult to obtain than the high and low prices utilised in the calculation of
the High-Low Spread (CSHL) measure developed by Corwin and Schultz in 2012.
Such data is typically accessible on the majority of stock exchanges, making this
133
new measure highly replicable. The CSHL formula, which is shown in Equation 4.1,
made it possible to obtain negative spread values, especially during volatile days or
when there was a significant overnight price change. In accordance with the authors’
suggestion, the negative values are replaced with zero. Corwin and Schultz (2012)
assert that the daily price range reflects the stock’s volatility and bid-ask spread since
the high (low) prices are almost always buyer (seller) initiated. Thus, CSHL is
advocated as a measure of stock liquidity, with a high value indicating lower
liquidity and a low value suggesting higher liquidity.
2(𝑒 , − 1)
𝐶𝑆𝐻𝐿 , =
1+𝑒 ,
2𝛽 , − 𝛽 , 𝛾,
𝛼, = −
3 − 2√2 3 − 2√2
ℎ𝑖𝑔ℎ , ℎ𝑖𝑔ℎ ,
𝛽 , = ln + ln
𝑙𝑜𝑤 , 𝑙𝑜𝑤 ,
𝑚𝑎𝑥 ℎ𝑖𝑔ℎ , , ℎ𝑖𝑔ℎ ,

𝛾 , = ln
𝑚𝑖𝑛 𝑙𝑜𝑤 , , 𝑙𝑜𝑤 ,
(Equation 4.1)
Where 𝐶𝑆𝐻𝐿 , is the High-Low Spread of stock i on day t; ℎ𝑖𝑔ℎ , and 𝑙𝑜𝑤 ,
represent the high and low prices of stock i on day t, respectively, while ℎ𝑖𝑔ℎ ,
and 𝑙𝑜𝑤 , represent the values obtained on day t+1.
2. Amihud Illiquidity ratio

The illiquidity measure proposed by Amihud (2002) in his seminal article is the most
widely employed low-frequency price impact ratio in finance literature (Le &
Gregoriou, 2020). The illiquidity ratio represents the magnitude of absolute price
change for each dollar of volume traded on a stock, hence quantifying the impact of
a trade on the stock’s price. Since the measure is calculated daily, it reflects the day-
long impact on prices. For a given degree of price fluctuation, a greater trading
volume results in a lower illiquidity ratio, and vice versa. Thus, a larger Amihud
illiquidity ratio indicates a lower liquidity, whilst a smaller ratio would reflect a
higher liquidity.
134
𝑅,
𝐼𝐿𝐿𝐼𝑄 , =
𝐷𝑉𝑂𝐿 ,
(Equation 4.2)
Where, 𝐼𝐿𝐿𝐼𝑄 , represents the Amihud Illiquidiy Ratio of firm i on day t; 𝐷𝑉𝑂𝐿 ,
represents the daily volume traded (in A$) of firm i on day t; and 𝑅 , represents the
absolute value of return of firm i on day t.
3. Volume Turnover
Volume, which is directly expressed by the number of volumes traded, is one of the
dimensions of liquidity. Regardless, it may be argued that directly citing volume
could result in a size bias, as larger firms would naturally have more volume. To
address this issue, the study uses volume turnover ratio, which is the number of daily
volumes traded (in units) divided by the number of outstanding shares (see Equation
4.3). The ratio indicates that a higher volume turnover corresponds to a higher
liquidity level, and vice versa.
𝐷𝑉𝑂𝐿 ,
𝑉𝑂𝐿𝑇𝑂 , =
𝑆𝑂 ,
(Equation 4.3)
Where, 𝑉𝑂𝐿𝑇𝑂 , represents the volume turnover of firm i on day t; 𝐷𝑉𝑂𝐿 ,

represents the daily volume traded (in unit) of firm i on day t; and 𝑆𝑂 , represents
the number of shares outstanding of firm i on day t.
4.3.3.2 HFT activity measures

Numerous HFT-related research have used proprietary datasets that accurately
identify the sender for each order submitted to the market; the datasets utilised in this
study, however, lack this feature. This study uses widely established HFT criteria as
a basis for assessing HFT activity to ensure that the chosen proxies are appropriate
for estimating the actual activity of HFT. For the purpose of this essay, only one
proxy for HFT activity is used, which is the high-frequency orders (HFO), which is
135
meant to measure the activity of low-latency traders, such as HFTs. 55 However, the
use of this measure could also include the activities of non-HFTs operating in the
sub-second trading environment. Therefore, any findings concerning “HFT activity”
drawn from this study should be understood as reflecting the activity of “low-latency
traders” as a whole, and not purely “full-fledged HFT firms”.
Under the premise that an order will either be fully executed or fully
cancelled, any order placed to the market may see the following message
combinations:
(i) submission  executed

(ii) submission  cancelled
(iii) submission  amendment  executed
(iv) submission  amendment  cancelled
(v) submission  amendment  amendment  executed
(vi) submission  amendment  amendment  cancelled
(vii) and so forth (with more messages recorded as amendment in between).
The message chain may be shorter for human traders who submit their orders
manually, since they are less likely to modify their orders frequently due to the
tedious and time-consuming nature of the process. For traders of all sizes who utilise
trading bots (i.e. algorithm-based trading), their orders may comprise a lengthy chain
of messages that might span the whole trading day. However, the rate at which their
quotes are updated is limited by the speed of their trading infrastructures. In contrast,
low-latency algorithmic traders, such as HFTs, have access to cutting-edge
technology and high-speed trading infrastructures, allowing them to update their
quotations at a faster pace than other traders, resulting in a shorter order resting
55
Although the message-to-trade ratio (MTR) and algorithmic trading ratio (ALGO) are often used as
proxies for high-frequency trading (HFT) activity, they are less effective at capturing activity that is
primarily driven by HFT activity and not by other market players. The formula used to calculate the
MTR and ALGO provides a more broad estimate of HFT activity on a daily basis. In stock markets
like as Australia, where HFT involvement is estimated to be around 20%, the figures provided by
MTR and ALGO may not be effective measures of HFT activity given that the great majority of
market activity is still produced by non-HFTs.
136
time.56 In essence, speed is the element that distinguishes HFTs from other
algorithm-based trading; without a speed advantage, HFTs are no different from
other algorithm-based traders.
HFTs employing a market-making strategy are forced to constantly update

their quotes in reaction to new market information, resulting in a large number of
message submission. In addition, their limit orders will be immediately cancelled or
amended if they are at risk of being adversely selected, or if they are attempting to
profit from transitory trading opportunities – both of which result in a lengthy
message chain and a short order resting period. Hasbrouck and Saar (2013) and
Boehmer, Li, and Saar (2018) estimate HFT presence using strategic runs
comprising a minimum of 10 linked-messages, asserting that frequent order
modifications are often used in dynamic algorithmic strategies. In addition,
Subrahmanyam and Zheng (2016) find that the average HFT order duration inside
the top three price levels of the limit order book is 28.74 seconds for large firms and
33.74 seconds for medium-sized firms. Assuming an HFTs-submitted order contains
at least 10 linked-messages, their results equate to an average order resting time of
around 3 seconds. In summary, the abovementioned studies suggest that orders
placed by HFTs should satisfy the following criteria: (i) have a lengthy message
chain consisting of at least 10 messages; and (ii) have a short order resting time.
The datasets for this study include a unique order identification (OrderID)
that allows the researcher to follow the chronology of all orders placed in the market,
from the time they are originally submitted until they are removed through self-
cancellation or traded by the opposite party.57 Based on the previously established
HFT criteria, the following formula presented in Equation 4.4 is used to identify
high-frequency orders (HFO), which, theoretically, are sent by HFTs.
56
Low-latency trading infrastructures consist of trade-related proprietary services (such as
networking, co-location, sponsored access, and direct market access) and state-of-the-art computer
hardware (such as high-speed CPUs, GPUs, and FPGAs).
57
The second essay of this thesis provides additional information on OrderID, the dataset used, and
relevant trading mechanisms. The author omitted certain sections to avoid repetition in the thesis.
137
𝑂𝑟𝑑𝑒𝑟𝐷𝑢𝑟𝑎𝑡𝑖𝑜𝑛 , , = 𝑂𝑟𝑑𝑒𝑟𝐼𝐷 , , , ( ) − 𝑂𝑟𝑑𝑒𝑟𝐼𝐷 , , , ( )
𝑂𝑟𝑑𝑒𝑟𝐷𝑢𝑟𝑎𝑡𝑖𝑜𝑛 , ,
𝑂𝑅𝑇 , , = ,,
∑ 𝑀𝑒𝑠𝑠𝑎𝑔𝑒𝑠 ,,
,,
𝐻𝐹𝑂 , = 𝑂𝑅𝑇 , , ≤ 3 seconds & 𝑀𝑒𝑠𝑠𝑎𝑔𝑒 ,, ≥ 10 messages
(Equation 4.4)
Where 𝑂𝑟𝑑𝑒𝑟𝐼𝐷 , , , ( ) represents the OrderID j on stock i on day t with the

highest timestamp (d); 𝑂𝑟𝑑𝑒𝑟𝐼𝐷 , , , ( ) represents the OrderID j on stock i on day
t with the lowest timestamp (d); 𝑂𝑟𝑑𝑒𝑟𝐷𝑢𝑟𝑎𝑡𝑖𝑜𝑛 , , represents the difference
between the highest and lowest timestamps of OrderID j on stock i on day t;
,,
∑ 𝑀𝑒𝑠𝑠𝑎𝑔𝑒 ,, represents the total number of messages generated by OrderID j
on stock i on day t; 𝑂𝑅𝑇 , , represents the order resting time of OrderID j on stock i
on day t; and 𝐻𝐹𝑂 , represents the OrderID on stock i on day t that meet the
following criteria: (i) order resting time 𝑂𝑅𝑇 , , of less than or equal to 3 seconds,
,,
and (ii) number of messages ∑ 𝑀𝑒𝑠𝑠𝑎𝑔𝑒 ,, of more than or equal to 10.
4.3.3.3 Other variables

The main objective of the research is to determine whether negative sentiment
influences the impact of HFT activity on liquidity. To capture this effect, a dummy
variable is used, with the value 1 assigned to observations during high VIX day and
0 otherwise. In addition to the above-mentioned measures of HFT activity, the study
also controls for volatility, relative tick size, and firm size, which are known to have
an effect on liquidity (see e.g., Madhavan, 2002).
The study uses daily trading range to proxy for a firm’s intraday price
volatility, which is calculated as the difference between the highest ℎ𝑖𝑔ℎ , and
lowest prices 𝑙𝑜𝑤 , of firm i on day t, divided their average prices. Relative tick
size is used to represent the lowest possible spread achievable at a given price, which
is derived by dividing the tick size value 𝑇𝑖𝑐𝑘𝑆𝑖𝑧𝑒 , by the closing price
𝑃𝑟𝑖𝑐𝑒 , of firm i on day t. Size on the other hand, is proxied by the natural log of
138
market capitalisation of firm i on day t. The respective formulas for volatility,
relative tick size, and size are shown in Equations 4.5, 4.6, and 4.7.
ℎ𝑖𝑔ℎ , − 𝑙𝑜𝑤 ,
𝑉𝑂𝐿𝐴𝑇𝐼𝐿𝐼𝑇𝑌 , =
ℎ𝑖𝑔ℎ , + 𝑙𝑜𝑤 ,
2
(Equation 4.5)
𝑇𝑖𝑐𝑘𝑆𝑖𝑧𝑒 ,
𝑅𝑇𝑆 , =
𝑃𝑟𝑖𝑐𝑒 ,
(Equation 4.6)
𝑆𝐼𝑍𝐸 , = ln 𝑀𝑎𝑟𝑘𝑒𝑡 𝐶𝑎𝑝𝑖𝑡𝑎𝑙𝑖𝑧𝑎𝑡𝑖𝑜𝑛 ,

(Equation 4.7)
4.3.4 Model specification

To test the first hypothesis, this study utilises the groups outlined in Table 4.1, which
consist of three study periods of varying lengths, namely short-term (1 month),
medium-term (6 months), and long-term (12 months), around the event date.
Observations for each group are categorised as either ”Pre-event” or “Post-event”.
The study uses two-sample t-test to assess whether the differences between the
liquidity level (as measured by CSHL, ILLIQ, and VOLTO) and HFT activity (as
measured by HFO) in the Pre-event and Post-event groups are statistically
significant. Subsequently, multivariate regression analysis is performed to evaluate
whether the VIXRT introduction influences the relationship between HFT activity
and liquidity. Equation 4.8 represents the model employed for this purpose:
𝐿𝐼𝑄𝑈𝐼𝐷𝐼𝑇𝑌 , = 𝛽 + 𝛽 𝐷𝑃𝑂𝑆𝑇 , + 𝛽 𝐻𝐹𝑇 , + 𝛽 𝐷𝑃𝑂𝑆𝑇 , × 𝐻𝐹𝑇 ,

+ 𝛽 𝐶𝑂𝑁𝑇𝑅𝑂𝐿 , +𝜀,
(Equation 4.8)
Where 𝐿𝐼𝑄𝑈𝐼𝐷𝐼𝑇𝑌 , is alternately proxied by CSHL, ILLIQ, and VOLTO of firm i

on day t; 𝐷𝑃𝑂𝑆𝑇 , represents a dummy variable that equals one if the observation of
firm i on day t belongs to the post-VIXRT period, and zero otherwise; 𝐻𝐹𝑇 , is
139
proxied by HFO of firm i on day t-1; 𝐷𝑃𝑂𝑆𝑇 , × 𝐻𝐹𝑇 , is the interaction term
which represent the values of lagged HFT measures in the post-VIXRT period; and
𝐶𝑂𝑁𝑇𝑅𝑂𝐿 , represents the three control variables used in the model, namely
VOLATILITY, RTS and SIZE of firm i on day t-1. Variables on the right-hand
side are assigned with lagged values to minimise the potential of having a reverse
causality issue. Firms (cross-sectional unit) and day (time unit) fixed effects are used
to control for omitted variables bias, enabling the model to account for any impacts
caused by unobservable factors that vary across firms and across time.
To test the second hypothesis, this study employs all post-VIXRT data and
uses a comparison approach based on the difference-in-difference method to evaluate
whether the liquidity between the studied groups is statistically different. For this
purpose, the study uses the subgroups listed in Panel B of Table 4.3. This research
design is illustrated in Figure 4.3, where top and bottom halves represent the Control
and Treatment groups, while the left and right panels represent the Non-event and
Event days, respectively. Quadrants (1) and (2) represent the data for Control and
Treatment during Non-event, whereas Quadrants (3) and (4) represent the data for
Control and Treatment during Event, respectively.
Non-event (VIXNH) Event (VIXH)

(1) (3)
Control (HFTNH)
ControlNon-event ControlEvent
(VIXNH-HFTNH) (VIXH-HFTNH)
(2) (4)
Treatment (HFTH)
TreatmentNon-event TreatmentEvent
(VIXNH-HFTH) (VIXH-HFTH)
Figure 4.3: Illustration of the research design used to test the second hypothesis.
140
The following steps are then taken to determine whether liquidity varies across
different levels of VIX (event impact) and HFT activity (treatment effect):
i. [(3) and (4)] – [(1) and (2)] = Difference caused by Event (DIFFVIX)
ii. [(2) and (4)] – [(1) and (3)] = Difference caused by Treatment (DIFFHFT)
iii. (2) – (1) = Difference caused by Treatment during Non-event day (DIFFVIX_NH)
iv. (4) – (3) = Difference caused by Treatment during Event day (DIFFVIX _H)
v. (DIFFVIX_H) – (DIFFVIX_NH)= Difference caused by Event and Treatment (DIFFVIX, HFT)
Subsequently, the study employs multivariate regression analysis using all data
in the post-VIXRT period to further test the second hypothesis. For the groups in
Panel A of Table 4.3, the sample is tested using the model shown in Equation 4.9 to
capture the interaction effect between VIX and HFT activity:
𝐿𝐼𝑄𝑈𝐼𝐷𝐼𝑇𝑌 , = 𝛽 + 𝛽 𝐷𝐻𝐻𝐹𝑇 , + 𝛽 𝐷𝐻𝑉𝐼𝑋 , + 𝛽 𝐷𝐻𝐻𝐹𝑇 , × 𝐷𝐻𝑉𝐼𝑋 ,

+ 𝛽 𝐶𝑂𝑁𝑇𝑅𝑂𝐿 , +𝜀,
(Equation 4.9)
Where 𝐷𝐻𝐻𝐹𝑇 , represents a dummy variable that equals one if the observation of
firm i on day t belongs to the high HFT group, and zero otherwise; 𝐷𝐻𝑉𝐼𝑋 ,
represents a dummy variable that equals one if the observation of firm i on day t
belongs to the high VIX days, and zero otherwise; and 𝐷𝐻𝐻𝐹𝑇 , × 𝐷𝐻𝑉𝐼𝑋 ,
represents the interaction term between dummy variable for high VIX days and high
HFT group. The other variables are interpreted in the same manner as the variables
in Equation 4.8. This model also employs firm and day fixed effects to account for
omitted variables bias.
4.4 Results, analyses, and discussions

This section presents the results and analysis of the study, which are organised in
accordance with the hypotheses. In summary, the hypothesis states that “the
introduction of real-time AXVI (VIXRT) have a positive impact on HFT activity and
indirectly increase liquidity through its relationship with HFTs.” The second
hypothesis on the other hand postulates that “as VIX levels increase, HFT activity
141
changes in a way that exacerbates the already fragile liquidity situation in the
market, calling into question the credibility of HFTs as liquidity providers.” This
section concludes with a discussion of the findings from testing the two hypotheses.
4.4.1 HFT activity and liquidity surrounding the VIXRT introduction event
Table 4.4 outlines the datasets used to test the first hypothesis, which employs three
date ranges to analyse HFT activity and liquidity surrounding the VIXRT introduction
event: 1 month, 6 months, and 12 months. For conciseness, only the 12 months date
range is described in the table, since it covers the broadest date range for the first
objective, thus effectively including all observations contained in the two smaller
date range groups. The data supplied for descriptive statistics are not adjusted for
outliers in order to portray the original attributes of the data employed. The table
indicates that, on average, the number of observations in the pre-event and post-
event periods are about the same, totalling approximately 20,000 firm-day
observations. For VIX, its average pre-event value is greater than its average post-
event value, which is due to the former having more days with high VIX readings.

This table describes the data employed to test the first hypothesis. Pre-VIXRT and Post-VIXRT refer to the
group of observations in the period before and after the introduction of VIXRT event on February 28,
2013, respectively. N, Mean, Std. Dev., Minimum, and Maximum signify the number of observations,
average, standard deviation, lowest and highest values, respectively. VIX represents the closing value of
VIX index; SIZE is the dollar value of market capitalisation (Equation 4.7); VOLATILITY is computed
by dividing the difference between the day’s highest and lowest prices by their average values (Equation
4.5); RTS indicates the relative tick size, which is calculated by dividing tick size by closing price
(Equation 4.6); and the ASX and CHIX subscripts denote the datasets used to estimate the variables.
N Mean Std. Dev. Minimum Maximum
Pre-VIXRT
VIX 20,087 15.512 3.242 10.536 26.639
SIZE (million) 20,087 $12,340 $19,840 $821 $125,000
VOLATILITYASX (%) 19,960 1.985 1.965 0.162 199.994
VOLATILITYCHIX (%) 20,016 1.652 1.022 0.000 20.690
RTSASX (%) 19,960 0.193 0.181 0.012 2.041
RTSCHIX (%) 20,016 0.194 0.182 0.012 2.000
Post-VIXRT
VIXCLOSE 20,273 14.578 2.130 10.689 21.675
SIZE (million) 20,273 $14,670 $24,090 $804 $128,000
VOLATILITYASX (%) 20,272 2.147 2.420 0.156 199.982
VOLATILITYCHIX (%) 20,270 1.921 1.089 0.000 31.237
RTSASX (%) 20,272 0.165 0.125 0.013 0.553
RTSCHIX (%) 20,270 0.165 0.125 0.013 0.556
142
This fact is illustrated in Figure 4.1, which shows that the VIX readings are
generally high from early May 2012 to late June 2012, whereas no such pattern
appears at any time during the one-year period after the VIXRT event. The average
market capitalisation before the event is lower than its average value after the event
despite having minimum and maximum values that are closely comparable. As
illustrated by the AXJO graph in Figure 4.1, this suggests an upward trend in the
aggregate value of ASX-listed stocks in the post-event period.
The intraday volatility estimates calculated from the ASX and CHIX datasets
are substantially different. During the pre-event and post-event periods, the highest
VOLATILITY values for the ASX are 199.994% and 199.982%, while for the CHIX,
they are only 20.69% and 31.237%, respectively. The exceedingly high
VOLATILITY values in the ASX dataset were produced by the prices of Rio Tinto
Limited (RIO) on March 5, 2012 (Pre-VIXRT), and Newcrest Mining Limited on
March 22, 2013 (Post-VIXRT). For RIO, its highest and lowest trading prices on the
ASX market were A$65.84 and A$0.001, while on the CHIX market, they were
A$65.59 and A$64.87, yielding intraday volatility values of 199.994% and 1.104%,
respectively. Similarly, the figures observed for NCM on the ASX market were
A$22.50 and A$0.001, and on the CHIX market, they were A$22.50 and A$22.20,
which correspond to intraday volatility rates of 199.982% and 1.342%, respectively.58
These figures are reflected in the average values of VOLATILITY, where ASX has
higher intraday volatility (Pre-VIXRT = 1.985%; Post-VIXRT = 2.147%) than CHIX
(Pre-VIXRT = 1.652%; Post-VIXRT = 1.921%) during both time periods.
In contrast, during the period either before or after the event, the RTS values
for ASX and CHIX are nearly equal, suggesting that the closing prices recorded on
both exchanges during the observation period were, for the most part, identical. The
average RTS value prior to the event (0.194%) is greater than its value after the
event (0.165%). This is consistent with the previously established premise that prices
are generally higher in the latter period, and ceteris paribus, this essentially
58
This behaviour is comparable to the extreme price movements witnessed on the U.S. stock market
during the 2010 Flash Crash, which was driven by fast algorithms that traded rapidly against a thin
order book. This situation leads to a liquidity dry up across all price levels and forces the trading
algorithm to strike the furthest price in the limit order book by accident. These extreme prices were
deemed to be the outcome of a trading error and were subsequently nullified.
143
translates to lower RTS values. Overall, based on the data observed between March
2012 and February 2014 (24 months), the differences between the pre and post
VIXRT event values can be attributed to an overall bullish market movement, and are
less likely to be influenced by negative market sentiment (i.e., high VIX).

Table 4.5 displays the results of univariate analysis comparing liquidity and HFT
activity before and after the real-time AXVI is made available on the Australian
stock market, as indicated by CSHL, ILLIQ, VOLTO (for liquidity), and HFO (for
HFT). This is accomplished by applying a two-sample t-test to compare the means of
the variables observed from the Pre-VIXRT and Post-VIXRT groups. In addition, the
study utilises observations from several date ranges to evaluate the impact of the
event across different time horizons.
An analysis of the ASX and CHIX datasets reveals that CSHL and HFO
consistently exhibit statistically significant positive difference values (i.e., Post-
VIXRT – Pre-VIXRT), regardless of the period range used. Additionally, a significant
increase in trading activity (VOLTO) is observed on both markets during the Post-
VIXRT period, particularly in the first six months following the event date. However,
when the observation period is extended to 12 months, the findings suggest that
VOLTO is significantly lower on the ASX but remains higher on the CHIX. The
results for ILLIQ indicate a significantly greater level of illiquidity on the ASX
market only when using the 6-month range. However, the results are not significant
for the 1-month and 12-month ranges. In contrast, illiquidity levels in the CHIX
market for the 6-month and 12-month ranges are found to be significantly lower than
their respective Pre-VIXRT values, while the 1-month range shows no notable
difference.
In summary, these findings suggest an asymmetry in liquidity levels across

different trading venues in Australia before and after the adoption of the real-time
AXVI index. For ASX, it is shown that the post-VIXRT period has: (i) significantly
wider spreads; (ii) a statistically indifferent level of illiquidity; and (iii) significantly
higher volume turnover only in the periods immediately following the event (1-
month and 6-month). In contrast, results for CHIX suggest that although spreads are
144
wider in the post-VIXRT period, price impact is smaller and trading activity is higher,
indicating a generally higher level of liquidity.
Table 4.5. Mean comparison of liquidity measures and HFT activity in the
periods surrounding the VIXRT event
This table displays the results of univariate tests performed on data collected before and after the
VIXRT event to evaluate the first hypothesis. Panel A and Panel B present the results using the ASX
and CHIX datasets, respectively. Range, which can be either 1, 6, or 12 months, refers to the time
periods used to sample data before (Pre-VIXRT) and after (Post-VIXRT) the event day. N and Mean
represent the number of observations and the average values, respectively. Diff. Post-Pre is the difference
between the mean values of the Post-VIXRT and Pre-VIXRT groups. The tested variables include three
liquidity measures, namely Corwin-Schultz High-Low spread (CSHL), Amihud Illiquidity ratio
(ILLIQ), and volume turnover (VOLTO), and one HFT measure that is proxied by high-frequency
orders (HFO). The formula are illustrated in Equations 4.1, 4.2, 4.3, and 4.4, respectively. All data are
winsorised at three standard deviations (3-sigma) from their respective means.
Pre-VIXRT Post-VIXRT
Panel A: ASX Diff. Post-Pre
N Mean N Mean
Range: 1 month
CSHL 1,520 0.498 1,680 0.556 0.059***
ILLIQ 1,520 0.072 1,680 0.070 -0.002
VOLTO 1,520 0.365 1,680 0.382 0.017**
HFO 1,520 4.305 1,680 4.429 0.124***
Range: 6 months
CSHL 9,894 0.445 10,230 0.546 0.101***
ILLIQ 9,894 0.080 10,230 0.085 0.005***
VOLTO 9,896 0.325 10,230 0.352 0.027***
HFO 9,896 3.978 10,230 4.525 0.547***
Range: 12 months
CSHL 19,833 0.467 20,270 0.503 0.036***
ILLIQ 19,832 0.085 20,271 0.085 0.000
VOLTO 19,960 0.345 20,272 0.318 -0.028***
HFO 19,960 4.112 20,272 4.338 0.226***
Pre-VIXRT Post-VIXRT
Panel B: CHIX Diff. Post-Pre
N Mean N Mean
Range: 1 month
CSHL 1,520 0.382 1,680 0.467 0.085***
ILLIQ 1,520 1.625 1,680 1.461 -0.165
VOLTO 1,520 0.024 1,680 0.026 0.003***
HFO 1,520 1.003 1,680 1.257 0.254***
Range: 6 months
CSHL 9,886 0.331 10,228 0.441 0.11***
ILLIQ 9,886 3.600 10,228 1.711 -1.889***
VOLTO 9,890 0.016 10,229 0.029 0.013***
HFO 9,892 0.866 10,229 1.558 0.692***
Range: 12 months
CSHL 19,971 0.329 20,266 0.403 0.074***
ILLIQ 19,967 11.162 20,267 1.645 -9.518***
VOLTO 19,889 0.013 20,269 0.030 0.017***
HFO 20,016 0.874 20,270 1.506 0.632***
*** p<0.01, ** p<0.05, * p<0.1
145
Concerning HFT activity, the conclusions are unanimous: the total number of high-
frequency Order IDs has grown significantly since the availability of real-time AXVI
data, which is consistent with the views of Liberti and Petersen (2019), who emphasise
the importance of supplying hard information to this class of trader. In summary, these
results suggest that there are a significantly greater number of HFT activity after the
introduction of VIXRT in both markets; however, in terms of liquidity, the level is
significantly higher in the post-event period in the CHIX market, whereas in the ASX
market, the results suggest otherwise.

Table 4.6 presents the results of multivariate regression analysis using the VIX RT
event to test the first hypothesis. Panel A exhibits the regression results using the
ASX dataset. The findings suggest that the VIXRT event has a significant negative
relationship with ILLIQ and CSHL only in the short and medium terms,
respectively. For VOLTO, The event significantly increased trading activity in both
the short and medium terms. On a long-term basis, the event is shown to have no
significant effect on any of the examined liquidity measures.
146
Table 4.6. Regression analysis on liquidity measures following the VIXRT event
This table shows the results from multivariate regression analysis using stocks around the real-time AXVI introduction event (VIXRT) to test the first hypothesis. Panel A and
Panel B represent the findings using the ASX and CHIX datasets, respectively. Each panel is further divided into three groups based on its Range, which can be either 1, 6, or 12
months. Range refers to the time periods used to sample data before and after the event day. Dependent variables (DV) are the liquidity measures, namely Corwin-Schultz high-
low spread (CSHL), Amihud illiquidity ratio (ILLIQ), and volume turnover (VOLTO). The formula are illustrated in Equations 4.1, 4.2, and 4.3, respectively. The independent
variables are HFO, DPOST, and HFO×DPOST. DPOST is a dummy variable assigned with a value of one if the observation belongs to the post-VIXRT period, and zero
otherwise. HFO is the natural logarithm of high-frequency orders (Equation 4.4), lagged by one day (t-1). The interaction term, DPOST×HFO, estimates the effect of HFO on
liquidity measures, after the VIXRT event. The control variables are the one-day lagged (t-1) values of VOLATILITY, RTS, and SIZE. VOLATILITY is measured by difference
between the highest and lowest price of the day, divided their average prices (Equation 4.5). RTS indicates the relative tick size, which is calculated by dividing tick size by
closing price (Equation 4.6). SIZE is the natural log of market capitalisation (Equation 4.7). All models are controlled for firm and day fixed-effects. Data are winsorised at three
standard deviations (3-sigma) from their respective means. The standard errors are calculated using the Huber/White/sandwich estimator of variance method.
A1: Range = 1 month A2: Range = 6 months A3: Range = 12 months
Panel A: ASX
CSHL ILLIQ VOLTO CSHL ILLIQ VOLTO CSHL ILLIQ VOLTO
DPOST -0.0878 -0.0472*** 0.1358*** -0.325*** 0.0132 0.0916*** 0.007 -0.0059 0.022
(0.1278) (0.0147) (0.0399) (0.0834) (0.0132) (0.0279) (0.0908) (0.016) (0.0356)
HFO -0.0054 -0.004 0.066*** 0.0157* -0.0043** 0.0597*** 0.0196*** -0.0065*** 0.0552***
(0.0281) (0.0024) (0.0083) (0.0086) (0.0021) (0.0061) (0.0058) (0.0017) (0.006)
DPOST×HFO 0.0255* 0.0035* -0.0118* -0.0073 -0.0017 -0.0003 -0.0091** -0.0013 0.0061
(0.0151) (0.0021) (0.0062) (0.0059) (0.0015) (0.0037) (0.0041) (0.0022) (0.0053)
VOLATILITY 0.0431*** -0.0028 0.0336*** 0.0391*** 0.0005 0.0253*** 0.0382*** 0.0026** 0.0267***
(0.0137) (0.0018) (0.0045) (0.0063) (0.0009) (0.0022) (0.0039) (0.001) (0.0026)
RTS 1.6455 0.1056 -0.7697 0.4604*** 0.2547*** -0.1599** 0.3987*** 0.1435** -0.0101
(1.6033) (0.4316) (0.9393) (0.1572) (0.0466) (0.0708) (0.108) (0.0572) (0.0652)
SIZE 0.676* -0.0123 -0.5049*** -0.0077 -0.0462*** -0.2819*** -0.0716** -0.0628*** -0.1566***
(0.4045) (0.083) (0.1846) (0.0592) (0.0165) (0.0385) (0.0278) (0.0153) (0.0288)
Constant -15.0765 0.3705 11.5189*** 0.5014 1.1038*** 6.3687*** 1.8976*** 1.4911*** 3.6243***
(9.2655) (1.945) (4.3054) (1.3725) (0.3755) (0.8613) (0.6483) (0.3489) (0.6532)
Observations 3,200 3,200 3,200 20,122 20,124 20,124 39,974 40,103 40,103
R-squared 0.0538 0.0588 0.2079 0.0565 0.0868 0.248 0.0609 0.0974 0.251
Firm FE Yes Yes Yes Yes Yes Yes Yes Yes Yes
Day FE Yes Yes Yes Yes Yes Yes Yes Yes Yes
147
B1: Range = 1 month B2: Range = 6 months B3: Range = 12 months
Panel B: CHIX
CSHL ILLIQ VOLTO CSHL ILLIQ VOLTO CSHL ILLIQ VOLTO
DPOST -0.0163 -0.0413 0.0063*** -0.1885*** -1.482 0.0253*** 0.0713 -288.2196** 0.0341***
(0.0836) (0.3838) (0.002) (0.07) (1.3907) (0.0024) (0.07) (112.094) (0.0029)
HFO -0.011 -0.2604* 0.0016*** 0.0077 -0.7867* 0.0027*** 0.0086** -2.5065** 0.0031***
(0.0166) (0.1496) (0.0005) (0.0063) (0.4368) (0.0004) (0.0042) (1.0845) (0.0004)
DPOST×HFO 0.0119 -0.0897 -0.0006 -0.0142*** 0.0818 -0.0024*** -0.0134*** 1.1057 -0.0032***
(0.0138) (0.1137) (0.0004) (0.0051) (0.2411) (0.0004) (0.0035) (0.6712) (0.0005)
VOLATILITY 0.0446*** 0.1638 0.0022*** 0.0458*** -0.9896 0.002*** 0.0456*** -4.3892*** 0.002***
(0.0132) (0.1598) (0.0003) (0.0058) (0.6062) (0.0002) (0.0037) (1.1142) (0.0002)
RTS 1.828 8.6769 -0.0986* 0.3238** 101.3407* -0.0443* 0.1983** 66.999 -0.0294
(1.4717) (6.3937) (0.0572) (0.1249) (53.8157) (0.0223) (0.0914) (54.1237) (0.0212)
SIZE 0.698* -0.1647 -0.0452*** 0.03 0.9571 -0.0251*** -0.0375 2.9655 -0.0213***
(0.3614) (2.4554) (0.0142) (0.0505) (5.8034) (0.0075) (0.0267) (16.3048) (0.0057)
Constant -15.6802* 3.7518 1.0566*** -0.4576 -33.0862 0.5767*** 1.0386* 219.5805 0.4832***
(8.3285) (56.3278) (0.3279) (1.1621) (135.3512) (0.1697) (0.6123) (407.2862) (0.1306)
Observations 3,200 3,200 3,200 20,107 20,114 20,112 40,198 40,234 40,108
R-squared 0.0528 0.0219 0.1178 0.0575 0.0205 0.3407 0.0616 0.0517 0.4303
Firm FE Yes Yes Yes Yes Yes Yes Yes Yes Yes
Day FE Yes Yes Yes Yes Yes Yes Yes Yes Yes
148
These findings demonstrated that the event promotes liquidity on the ASX
market; however, the effect is only temporarily evident in the short to medium term,
and is no longer present in the long term. Higher HFO leads to significantly higher
CSHL and lower ILLIQ over the medium and long terms, whereas for VOLTO,
higher HFO results in greater trading activity throughout all date ranges examined.
These results imply that high HFT activity is detrimental to liquidity as it widens the
spread, but it also promotes liquidity by reducing price impact and increasing trading
activity. The interaction term, DPOST×HFO, shows that the VIXRT event influences
the relationship between HFO and all liquidity measures in the short term, but only
the relationship between HFO and spread in the long term. The findings suggest that
the VIXRT reverses the beneficial effect of HFT activity on liquidity in the ASX
market in short-term.
Panel B of Table 4.6 shows the regression results using the CHIX dataset.
The findings suggest that the VIXRT has a significant negative impact on CSHL and
ILLIQ only in the medium and long-term, respectively. In addition, the event causes
VOLTO to be significantly greater across all examined date ranges. These results
suggest that the new feature introduced by the VIXRT gradually enhances the overall
liquidity of the CHIX market. Specifically, the improvement in trading activity is
immediately apparent following the occurrence of the event, and this positive effect
is shown to be long-lasting. For HFO, the findings suggest that in CHIX, a higher
HFT activity consistently leads to a lower price impact and higher trading activity
(improved liquidity) across all date ranges, and in long term, it is also found to result
in significantly wider spread (reduced liquidity). The results of the interaction
variable DPOST×HFO suggest that the influence of VIXRT exists only in the
medium and long term, but in the short term, it has no significant effect. The event is
shown to weaken the impact that HFO has on CSHL and VOLTO, where it lowers
the detrimental effect of HFT activity on spread (improved liquidity), and dampens
the beneficial effect of HFT activity on trading volume (reduced liquidity).
Nevertheless, the VIXRT is shown to have no influence on the relationship of HFT
activity and price impact in the CHIX market.
149
Overall, even after controlling for volatility, tick size, firm size, firm fixed
effects, and day fixed effects, the regression results indicate that both markets have
shown improvements in liquidity since the introduction of real-time AXVI, with the
effect being more immediate for the ASX, and more gradual for the CHIX. In
addition, the findings show that an increase in HFT activity generally results in
wider spreads, smaller price impacts, and higher trading activity on both markets,
which suggests an overall favourable impact on liquidity. The results also imply that
the event influences the relationship between HFT activity and liquidity measures,
where it is shown to weaken the effect of HFT on liquidity in both the ASX (in the
short-term only) and the CHIX (in the medium and long-terms) markets.
The VIXRT event has enhanced the amount of publicly available hard
information in real-time, which continuously supplies the market with the most
recent VIX values, representing the current market sentiment. This new feature may
be less essential for slow traders unless there is a substantial intraday VIX
movement, as they can always make educated assumptions about market sentiment
by referring to the previous day’s VIX level. In contrast, the event presents an
opportunity for fast traders such as the HFTs to take strategic positions in the market,
since they are naturally able to respond more swiftly and aggressively than non-
HFTs to hard information shocks (Zhang, 2017), such as the one produced by the
real-time AXVI data. Hence, it is plausible to infer that HFTs’ strategies are
influenced by the VIXRT event, and their trading algorithms should have accounted
for the real-time AXVI data. Furthermore, HFTs might shift from being a liquidity
provider to a liquidity taker by deploying aggressive trading algorithms to capitalise
on the new features. For example, HFTs may profit from a transitory arbitrage
opportunity that exists due to lagging price adjustments to the most recent VIX
readings posted on the market, which imposes adverse selection costs on market
makers. Therefore, the observed reversal in the effects of HFT activity on liquidity in
the period immediately following the VIXRT is likely attributable to the increased
presence of liquidity-taking HFTs relative to liquidity-providing HFTs on the ASX
market.
150
4.4.2 Liquidity, VIX and HFT activity
Table 4.7 describes the datasets used to test the second hypothesis, which employs
all observations in the post-VIXRT period to analyse HFT activity and liquidity
during days with heightened negative market sentiment. Panel I presents the “Main
groups”, which is formed based on the non-high and high values of VIX and HFT,
resulting in four groups of sample: Non-high VIX (VIXNH, i.e., Non-event), High
VIX (VIXH, i.e., Event), Non-high HFT (HFTNH, i.e., Control), and High HFT
(HFTH, i.e., Treatment), for each dataset employed. The table also shows that the
ASX dataset has a higher number of observations compared to the CHIX (N ASX =
121,777; NCHIX = 119,090), which is due to the former having lower number of
missing data.59 The data provided for descriptive statistics are not adjusted for
outliers to represent the original characteristics of the data.
In both datasets, VOLATILITY is the found to be around 0.8% higher on

both markets during Event compared to Non-event, which reflects the heightened
level of uncertainty over the near-term direction of the market, indicated by the
higher VIX readings in the former period. This situation might lead to a greater
selling pressure in the market, leading to lower stock prices. This notion is consistent
with the behaviour shown by RTS and SIZE, in which the relative tick size is around
3% higher, while the average market capitalisation is around one billion dollars
lower during Event. For the groups defined by HFT activity, the average VIX values
are found to be higher in Treatment, suggesting that a higher VIX is associated with
an increase in HFT activity. The descriptive statistics of the variables used in this
study show that HFT activity is higher in stocks with smaller RTS, lower intraday
volatility, and larger market capitalisation.60 The minimum and maximum values
displayed in Panel I are the most extreme datapoints from the ASX and CHIX
datasets used for this research. For VIX, the maximum value in Non-event and the
minimum value in Event represent the employed cut-off point. For Control and
Treatment, the highest and lowest VIX values are the same, which are 8.825 (March
10, 2017) and 31.185 (August 25, 2015), respectively.
59
If there are no missing data, the highest number of firm-day observations is 121,781.
60
See Tables A2 and A3 in the Appendix for further information about the characteristics of firms
with high HFT activity, for the ASX and CHIX datasets, respectively.
151
Table 4.7 Descriptive statistics of the data used to test Hypothesis 2
This table describes the data employed to test the second hypothesis. Panels I and II refer to the classification of sample as either belong to the Main groups or Subgroups,
respectively. All data are drawn from the post-VIXRT period. Panels A1 and A2 (B1 and B2) describe the data of observations derived from the ASX (CHIX) dataset under
the Main groups and Subgroups, respectively. Days with VIX readings equal to or greater than 20 are labelled as High VIX (VIXH, i.e. Event), otherwise Non-high VIX
(VIXNH, i.e. Non-event). For both HFOASX and HFOCHIX, observations with HFO values in the 90th percentile are labelled as High HFT (HFTH, i.e. Treatment), otherwise Non-
high HFT (HFTNH, i.e. Control). The VIXH and VIXNH are then paired with HFTH and HFTNH, creating four subgroups for each dataset, which are: (i) ControlNon-event (VIXNH-
HFTNH); (ii) TreatmentNon-event (VIXNH-HFTH); (iii) ControlEvent (VIXH-HFTNH); and (iv) TreatmentEvent (VIXH-HFTH). N, Mean, Std. Dev., Minimum, and Maximum signify the
number of observations, average, standard deviation, lowest and highest values, respectively. VIX represents the closing value of VIX index; SIZE is the dollar value of
market capitalisation (Equation 4.7); VOLATILITY is computed by dividing the difference between the day’s highest and lowest prices by their average values (Equation 4.5);
and RTS indicates the relative tick size, which is calculated by dividing tick size by closing price (Equation 4.6).
Panel A1: ASX Panel B1: CHIX
Panel I: Main groups
(1) Non-high VIX (Non-event)
VIX 111,094 14.158 2.408 8.825 19.998 108,637 14.145 2.401 8.825 19.998
RTS (%) 111,094 0.163 0.137 0.005 2.083 108,637 0.160 0.134 0.005 2.083
VOLATILITY (%) 111,094 2.031 1.529 0.110 199.982 108,637 1.845 1.145 0.000 72.857
SIZE (million) 111,094 13,740 22,550 243 156,000 108,637 13,870 22,710 243 156,000
(2) High VIX (Event)

VIX 10,683 23.312 2.692 20.011 31.185 10,453 23.318 2.704 20.011 31.185
RTS (%) 10,683 0.167 0.137 0.007 1.333 10,453 0.167 0.136 0.007 1.351
VOLATILITY (%) 10,683 2.802 2.411 0.222 181.372 10,453 2.564 1.575 0.000 20.990
SIZE (million) 10,683 12,770 20,460 628 140,000 10,453 12,860 20,580 628 140,000
(1) Non-high HFT (Control)

VIX 109,581 14.866 3.502 8.825 31.185 106,948 14.921 3.560 8.825 31.185
RTS (%) 109,581 0.178 0.135 0.006 2.083 106,948 0.175 0.135 0.005 2.083
VOLATILITY (%) 109,581 2.102 1.564 0.110 199.982 106,948 1.927 1.222 0.000 72.857
SIZE (million) 109,581 9,753 14,660 243 148,000 106,948 10,510 16,290 243 153,000
(2) High HFT (Treatment)

VIX 12,196 15.815 3.886 8.825 31.185 12,142 15.203 3.503 8.825 31.185
RTS (%) 12,196 0.025 0.021 0.005 0.532 12,142 0.037 0.032 0.005 0.457
VOLATILITY (%) 12,196 2.067 2.210 0.298 162.884 12,142 1.741 1.046 0.217 16.561
SIZE (million) 12,196 48,700 41,270 331 156,000 12,142 42,610 41,450 1,151 156,000
152
Panel II: Subgroups
(1) ControlNon-event
VIX 100,558 14.107 2.394 8.825 19.998 97,552 14.114 2.404 8.825 19.998
RTS (%) 100,558 0.177 0.136 0.006 2.083 97,552 0.174 0.134 0.005 2.083
VOLATILITY (%) 100,558 2.040 1.435 0.110 199.982 97,552 1.866 1.160 0.000 72.857
SIZE (million) 100,558 9,932 15,030 243 148,000 97,552 10,620 16,530 243 153,000
(2) TreatmentNon-event
VIX 10,536 14.641 2.485 8.825 19.998 11,085 14.419 2.354 8.825 19.998
RTS (%) 10,536 0.024 0.021 0.005 0.532 11,085 0.037 0.032 0.005 0.457
VOLATILITY (%) 10,536 1.946 2.234 0.298 162.884 11,085 1.663 0.979 0.217 16.561
SIZE (million) 10,536 50,060 41,800 331 156,000 11,085 42,510 41,680 1,151 156,000
(3) ControlEvent
VIX 9,023 23.321 2.684 20.011 31.185 9,396 23.306 2.673 20.011 31.185
RTS (%) 9,023 0.193 0.134 0.007 1.333 9,396 0.181 0.136 0.007 1.351
VOLATILITY (%) 9,023 2.796 2.496 0.222 181.372 9,396 2.565 1.600 0.000 20.990
SIZE (million) 9,023 7,762 9,353 628 90,130 9,396 9,398 13,510 628 135,000
(4) TreatmentEvent
VIX 1,660 23.266 2.737 20.011 31.185 1,057 23.425 2.965 20.011 31.185
RTS (%) 1,660 0.030 0.022 0.007 0.376 1,057 0.038 0.030 0.009 0.227
VOLATILITY (%) 1,660 2.835 1.882 0.358 22.785 1,057 2.554 1.335 0.535 10.885
SIZE (million) 1,660 40,020 36,620 1,326 140,000 1,057 43,610 38,930 1,482 140,000
153
These figures imply that high HFT activity may occur even on days when the
VIX is extraordinarily high or low, and vice versa. The lowest RTS value is 0.005%
on November 28, 2017 (Cochlear Limited (COH)), while the highest RTS value is
2.083% on November 21, 2014 (Arrium Limited (ARI)). These values correspond to
the highest (A$186.72; COH) and lowest stock prices (A$0.24; ARI) recorded
throughout the study period, respectively. The lowest recorded value for
VOLATILITY is 0.000%, which represents stocks that are traded at a single price
throughout the trading day, resulting in the same value for its opening, highest,
lowest, and closing prices. On the opposite extreme, Newcrest Mining Limited
(NCM) on March 22, 2013, had the highest VOLATILITY at 199.982%. 61 For SIZE,
the highest market capitalisation recorded was A$156.165 billion on March 20,
2015, by Commonwealth Bank of Australia (CBA), and the lowest was A$0.243
billion on November 27, 2015, by Slater & Gordon Limited (SGH).
Panel II displays the “Subgroups” derived from the unique combinations of

the VIX and HFT groups, yielding four subgroups for each dataset: (i) Control Non-
event (VIXNH-HFTNH); (ii) TreatmentNon-event (VIXNH-HFTH); (iii) ControlEvent (VIXH-
HFTNH); and (iv) TreatmentEvent (VIXH-HFTH). This segregation makes it possible to
examine whether the event significantly modifies the effect of treatment on liquidity
measures. During Non-event day, the Treatment group displayed smaller RTS and
VOLATILITY and greater SIZE than the Control group. Moreover, the magnitude of
these differences varies between the ASX and CHIX markets, with the Treatment
group exhibiting RTS values that are 0.153% and 0.137% smaller, VOLATILITY
values that are 0.094% and 0.203% lower, and SIZE values that are 4.04 times and
3.00 times greater, respectively. During Event day, the differences for RTS are found
to be roughly similar, where the RTS values for the Treatment group are 0.163% and
0.143% smaller in the ASX and CHIX markets, respectively. For VOLATILITY, the
magnitude of the difference between the Treatment and Control groups is smaller,
suggesting that the average stock-level volatility on both markets during high VIX
days (i.e., Event) is strikingly similar regardless of the level of HFT’s presence in
any given stock. As for SIZE, the differences are more pronounced during Event,
61
This issue has been discussed in detail in Section 4.4.2.1.
154
where the Treatment is found to be 4.16 times and 3.64 times greater than the
Control in the ASX and CHIX markets respectively.
These statistics indicate that stocks with low RTS values are more likely to
attract a larger HFT presence regardless of market volatility, implying that within a
similar tick size structure, stocks with a higher price will always have a greater HFT
activity. Correspondingly, stocks with a larger market capitalisation have more HFT
activity, and during highly volatile days, this phenomenon becomes more apparent,
particularly in the CHIX market. This situation can be explained by the fact that only
10% of the trading volume for Australia’s 100 largest stocks took place on the CHIX
market.62 As a result, HFT would be more prevalent on stocks with larger market
capitalisations due to their relatively higher liquidity and lower risk compared to
smaller firms, and the need to prioritise size would increase during high volatility
days. In addition, the statistic shows that stocks with substantial intraday price
fluctuations have, on average, less HFT activity during Non-event days. This
behaviour is likely attributable to both the RTS and SIZE factors – firms with a
larger market capitalisation tend to have a higher stock price, which corresponds to a
lower RTS, and are generally highly liquid, making them less prone to experiencing
large price swings.
During the Event, however, intraday volatility appears to be of less interest to

HFT; stocks with either high or low HFT activity appear to be indifferent in terms of
their volatility. In contrast to SIZE and RTS, which are relatively stable over time,
VOLATILITY is a parameter that may vary drastically from day to day and is highly
sensitive to the breadth and depth of the market. During high VIX days, most stocks
are adversely affected, resulting in substantial intraday price swings across the
market. Therefore, selecting a firm only on the basis of its low volatility during such
days might provide HFT with a very limited number of options, making
VOLATILITY a less important consideration for HFT during uncertain and volatile
days.
62
In 2013, the CHIX accounted for only 8.62% of the trading volume of the top 100 largest equities
in Australia. The CHIX’s influence steadily grows over the years, and by 2017, they are responsible
for 13.05% of all trading volume for the S&P/ASX100 constituents.
155
Table 4.8 displays the results of univariate analyses to assess whether high VIX days
result in significantly different levels of liquidity while controlling for HFT activity.
In order to investigate this issue, the study first compares the levels of liquidity on
high VIX and non-high VIX days, as well as the levels of liquidity in groups with
high HFT activity and groups with low HFT activity, in two separate analyses. This
separation enables the study to determine the specific (Panel I: Main groups) and
combined (Panel II: Subgroups) effects of the event of interest (high VIX) and the
treatment group (high HFT activity) on liquidity.
The results in Panel I indicate that on days with negative market sentiment,
spreads are significantly wider, price impacts are significantly greater, and trading
activities are significantly higher than on other days. These findings suggest that, on
average, liquidity is generally hampered when the VIX is high, and this effect is
consistent on both the ASX and CHIX markets. The results also indicate that when
HFT activity is high, spreads are significantly tighter and price impacts are
significantly smaller on both markets. High HFT activity is only associated with
substantially lower trading activity on the CHIX market, but not on the ASX market.
Overall, these findings imply that observations with high HFT activity have, on
average, a higher level of liquidity than those with low HFT activity, and this
conclusion holds true for both the ASX and CHIX markets.
The results in Panel II show that, during Non-event days, firms in the
Treatment groups have a significantly narrower bid-ask spread, a smaller illiquidity
ratio, and a lower volume turnover than firms in the Control group, on both the ASX
and CHIX markets. Identical results are also reported during Event days; although,
the magnitude of differences in liquidity measures between the Treatment and
Control groups is not similar. These findings suggest that regardless of market
sentiment, firms with substantial HFT activity are generally more liquid in terms of
lower transaction costs and smaller price impact; however, they also suffer from a
lower of trading activity.
156
Table 4.8. Mean comparison of liquidity measures during Event (VIXH) and Non-
event (VIXNH) days between Treatment (HFTH) and Control (HFTNH) groups
This table displays the results of univariate tests performed on data collected in the period after the
VIXRT event to evaluate the second hypothesis. Panels A1 and A2 and Panels B1 and B2 represent the
results derived from the ASX and CHIX datasets, respectively. Panel I show the results for univariate
test of the sample in the Main groups, which is formed based on the values of VIX and HFO. For VIX,
days with VIX readings equal to or greater than 20 are labelled as VIXH and represent the “Event”
sample, while other VIX values are labelled as VIXNH and represent the “Non-event” sample. For
HFT, observations with HFO values in the 90th percentile are labelled as HFTH and represent the
“Treatment” group, while other HFO values are labelled as HFTNH and represent the “Control” group.
DIFFVIX (H-NH) and DIFFHFT (H-NH) represent the difference between the means of Event and Non-event,
and the means of Treatment and Control, respectively. Panel II displays the univariate analysis
utilising the subgroups produced by pairing the “Event” and “Non-event” groups with the
“Treatment” and “Control” groups. DIFFNon-event (HFTH – HFTNH) and DIFFEvent (HFTH – HFTNH) indicate the
difference between the means of “Treatment” and “Control” during the “Non-event” and “Event”
days, respectively. DIFF-IN-DIFF refers to the difference-in-difference values produced by
subtracting the DIFFEVENT with DIFFNON-EVENT. The tested variables include three liquidity measures,
namely Corwin-Schultz High-Low spread (CSHL), Amihud Illiquidity ratio (ILLIQ), and volume
turnover (VOLTO), and the formula are illustrated in Equations 4.1, 4.2, and 4.3, respectively. All
data are winsorised at three standard deviations (3-sigma) from their respective means. The standard
errors are calculated using the Huber/White/sandwich estimator of variance method.
Panel I: Main groups
CSHL ILLIQ VOLTO CSHL ILLIQ VOLTO
VIX
Event (VIXH) 0.558 0.100 0.374 0.463 2.022 0.039
Non-event (VIXNH) 0.460 0.087 0.294 0.374 1.426 0.033
DIFFVIX (H –NH) 0.098*** 0.013*** 0.08*** 0.089*** 0.596*** 0.006***
(0.007) (0.001) (0.002) (0.007) (0.103) (0.001)
HFT
Treatment (HFTH) 0.390 0.026 0.299 0.322 0.549 0.023
Control (HFTNH) 0.477 0.095 0.301 0.390 1.584 0.035
DIFFHFT (H –NH) -0.087*** -0.069*** -0.002 -0.068*** -1.035*** -0.012***
(0.005) (0.001) (0.002) (0.005) (0.026) (0.000)

Panel II: Subgroups
Non-event (VIXNH)
Treatment (HFTH) 0.375 0.024 0.29 0.308 0.559 0.022
Control (HFTNH) 0.469 0.093 0.294 0.382 1.525 0.034
DIFFNon-event (HFTH – HFTNH) -0.094*** -0.069*** -0.004* -0.074*** -0.966*** -0.012***
(0.005) (0.001) (0.002) (0.005) (0.027) (0.000)
Event (VIXH)
Treatment (HFTH) 0.489 0.037 0.352 0.466 0.442 0.032
Control (HFTNH) 0.571 0.112 0.379 0.463 2.201 0.04
DIFFEvent (HFTH – HFTNH) -0.082*** -0.075*** -0.027*** 0.004 -1.759*** -0.008***
(0.018) (0.002) (0.005) (0.021) (0.115) (0.001)
DIFF-IN-DIFF
DIFFEvent – DIFFNon-event 0.012 -0.006*** -0.023*** 0.078*** -0.793*** 0.004***
(0.019) (0.002) (0.006) (0.021) (0.118) (0.001)
157
Nevertheless, it is plausible that the lower volume turnover is not a result of having
high HFT presence, but rather an unintended effect of HFTs’ penchant for firms with
large capitalisation, which often have a great number of shares outstanding and,
hence, a low volume turnover value.63
Using a technique similar to the difference-in-difference method, this study

examines whether the event may substantially change the effect of HFT activity on
liquidity measures. The univariate test indicates that market sentiment may have a
significant influence on the HFT activity – liquidity relationship; however, the effect
is inconsistent across the ASX and CHIX markets and varies amongst the examined
liquidity measures. On the ASX, the results suggest that high VIX readings do not
diminish the favourable effect that HFT has on spread, and in fact, it amplifies the
positive influence that HFT has on lowering the illiquidity ratio. Nonetheless, firms
with a high level of HFT activity during pessimistic times may have even lower
trading activity on the ASX market. Findings from the CHIX market reveal that high
VIX days weakened HFT’s initially positive influence on spread, resulting in a wider
gap between bid and ask prices. On the other hand, it also enhances the positive
effect that HFT has on lowering illiquidity and significantly reduces the negative
impact that HFT activity has on trading volume.
Based on the aforementioned findings, the study concludes that HFT plays an
important role in lowering spreads, and that days with pessimistic sentiment have no
impact on HFT’s positive role on the ASX, but do cause wider spreads on the CHIX.
Historically, only around one-tenth of total trading activity for S&P/ASX100
components took place on the CHIX. 64 Due to the additional risk of inadequate
transactions, HFT must increase their market-making spread on the CHIX market by
a bigger margin than on the ASX market. As noted by Brogaard et al. (2014) and
Hendershott and Riordan (2009), this measure is necessary to guarantee that HFT
continues to benefit from market-making operations. Failure to charge a wider
spread might discourage HFT from participating entirely, and their absence on days
63
This argument is supported by the descriptive statistics presented in Table 4.7 (Panel II), which
demonstrate that the average firm size of stocks in the Treatment group is three times greater than
those in the Control group.
64
See footnote 63.
158
of extreme volatility would exacerbate the already deteriorating market liquidity.
Moreover, the data indicate that HFT plays a crucial role in reducing price impact on
both markets, a quality that is further highlighted during stressful periods. This is
likely attributable to competition amongst market-making HFTs (see Boehmer et al.,
2018), which strive to actively feed the market with attractive limit orders on both
sides of the book, hence preserving its breadth and depth.
The research also reveals that an increase in HFT activity does not result in
higher volume turnover; rather, it is the opposite. Furthermore, the data show that
during volatile times, high HFT activity exacerbates the detrimental effect of HFT on
volume turnover on the ASX, whereas on the CHIX, it helps to mitigate the effect.
The findings of ASX are compatible with the argument of Easley and O’Hara
(1992), which posits that if there is no information event, there will be no
transaction, and the market maker will choose to post a tighter spread, resulting in a
positive relationship between spread and trading volume. This action may also be
interpreted as a market-making HFTs’ effort to attract investors to trade in a market
suffering from reduced trading volume due to pessimistic sentiment. On the other
hand, the increased volume turnover observed on the CHIX during high uncertainty
days is likely due to the greater number of HFTs’ market-making activity,
particularly in firms with larger market capitalisation. This is shown by the fact that
the average size of stocks with high HFT activity on the CHIX increased by 2.59%
during high VIX days, while the average size of other groups decreased due to
generally lower stock prices.65 The study posits that this scenario is likely a result of
the maker-taker fee structure used in the CHIX, which may incentivise HFT to make
market on the CHIX, particularly on days with a high VIX.

Table 4.9 presents the results of multivariate regression analysis using high HFT
activity and high VIX days to further test the second hypothesis. In the ASX dataset,
DHHFT (Treatment) has significant positive relationships with CSHL and VOLTO,
while DHVIX (Event) shows a significant positive relationship only with ILLIQ.
65
Comparatively, the average market capitalisation of firms in the ControlASX, TreatmentASX, and
ControlCHIX during Event is lower by 21.85%, 20.06%, and 11.51%, respectively, than during Non-
event.
159
These results indicate that high HFT activity results in significantly wider spreads
and increased trading activity, but has no discernible influence on price impacts. On
the other hand, days with a high VIX lead to significantly larger price impacts, but
have no effect on spreads or trading volume. In the CHIX dataset, the observed
results are slightly different; both DHHFT and DHVIX have a significant negative
effect on ILLIQ and a significant positive effect on VOLTO. These findings indicate
that high HFT activity and high VIX days both result in a smaller price impact and
increased trading activity. The results suggest that the Treatment and Event caused a
deterioration in liquidity on the ASX market, but an improvement in liquidity on the
CHIX market.
Table 4.9. Regression analysis on liquidity measures while controlling for the
effect of high VIX days and high HFT activity
This table shows the results from multivariate regression analysis using observations in the post-VIXRT
period to test the second hypothesis. Panel A and Panel B represent the findings using the ASX and
CHIX datasets, respectively. Dependent variables (DV) are the liquidity measures, namely Corwin-
Schultz high-low spread (CSHL), Amihud illiquidity ratio (ILLIQ), and volume turnover (VOLTO). The
formula are illustrated in Equations 4.1, 4.2, and 4.3, respectively. The independent variables are
DHHFT, DHVIX, and DHHFT×DHVIX. DHHFT is a dummy variable assigned with a value of one if
the observation belongs to the high HFT group (Treatment), and zero otherwise (Control). DHVIX is a
dummy variable assigned with a value of one if the observation belongs to the high VIX days (Event),
and zero otherwise (Non-event). The interaction term, DHHFT×DHVIX, estimates the effect of HFT
activity on liquidity measures during high VIX days. The control variables are the one-day lagged (t-1)
values of VOLATILITY, RTS, and SIZE. VOLATILITY is measured by difference between the highest
and lowest price of the day, divided their average prices (Equation 4.5). RTS indicates the relative tick
size, which is calculated by dividing tick size by closing price (Equation 4.6). SIZE is the natural log of
market capitalisation (Equation 4.7). All models are controlled for firm and day fixed-effects. Data are
winsorised at three standard deviations (3-sigma) from their respective means. The standard errors are
calculated using the Huber/White/sandwich estimator of variance method.
DHHFT 0.0632*** 0.0022 0.0841*** -0.0011 -0.7468*** 0.0093***
(0.0107) (0.0021) (0.0131) (0.0065) (0.211) (0.0011)
DHVIX 0.1113 0.0638*** 0.0299 0.0618 -1.8225*** 0.039***
(0.1011) (0.0137) (0.0275) (0.0936) (0.652) (0.0035)
DHHFT×DHVIX -0.0095 -0.0122*** -0.0351*** 0.074*** -0.6014** -0.0004
(0.0184) (0.0028) (0.0087) (0.0192) (0.233) (0.0012)
VOLATILITY 0.0442*** 0.0005 0.0389*** 0.0446*** -0.0863 0.0042***
(0.0024) (0.0006) (0.0016) (0.0024) (0.0578) (0.0003)
RTS 0.338*** 0.1452*** -0.1127* 0.1677** 1.2939 0.0049
(0.076) (0.0492) (0.0579) (0.0727) (0.8506) (0.0109)
SIZE -0.0452*** -0.0693*** -0.1036*** -0.0245** -0.5635** -0.0178***
(0.0143) (0.0086) (0.0173) (0.0119) (0.2723) (0.0027)
Constant 1.4126*** 1.618*** 2.7084*** 0.9014*** 15.3697** 0.04165***
(0.3441) (0.1981) (0.3931) (0.2883) (6.304) (0.0608)
Observations 121,601 121,740 121,740 117,558 118,321 118,317
R-squared 0.0731 0.122 0.2715 0.0679 0.0929 0.2167
Firm FE Yes Yes Yes Yes Yes Yes
Day FE Yes Yes Yes Yes Yes Yes
160
The DHVIX×DHHFT coefficient is the Treatment and Event interaction term
and is used to test for the influence of high VIX days (Event) on the relationship
between HFT activity and the liquidity measures examined in this study. On the ASX
market, the results show that the event (DHVIX) significantly affects the association
between DHHFT and ILLIQ and VOLTO, where stocks with high HFT activity are
shown to substantially decreases their illiquidity ratio (improve liquidity) and volume
turnover (reduce liquidity). For ILLIQ, the event reduces the coefficient for treatment,
where the βActual is 1.6718, and the βCounterfactual is 1.6840.66 The lower βActual value
suggests that the high VIX days lessens the detrimental impact that HFT activity has on
the illiquidity ratio. Similarly, the event is found to reduce the coefficient of DHHFT in
VOLTO, where its βActual and βCounterfactual are 2.7873 and 2.8224, respectively. The
relatively smaller βActual indicates that the event dampened the enhancing effect that HFT
activity has on trading activity on days with normal or low expected volatility.
On the CHIX market, the findings indicate that the influence of high VIX days
on HFT activity’s effect on liquidity is significant when measured by CSHL and ILLIQ,
where the event mitigates the effect of high HFT activity on spreads while amplifying its
effect on the illiquidity ratio. For CSHL, the event increases the treatment coefficient,
where βActual is 1.0361 and βCounterfactual is 0.9621. The greater value of βActual implies that
the beneficial effect of high HFT activity in lowering spread is weakened on days with a
high VIX. On the contrary, the event is shown to reduce the coefficient of DHHFT in
ILLIQ, where the βActual and βCounterfactual are 12.1990 and 12.8004, respectively. The
lower βActual value signifies that the event magnified the favourable impact that high
HFT activity has in reducing illiquidity level.
Analysis of the control variables demonstrates that stocks with greater

intraday volatility have not only significantly wider spreads, but also substantially
greater trading activity in both datasets. These results indicate that volatility has a
mixed impact on liquidity on the ASX and CHIX markets. In the ASX dataset, RTS
exhibits a significantly positive effect on spreads and price impact, but a negative
66
βActual = β0 + (β1×1) + (β2×1) + (β3×1); βCounterfactual = β0 + (β1×1) + (β2×1) + (β3×0); where β0, β1, β2,
and β3 represent the coefficient for the regression model’s Constant value, dummy for high-HFT
observations (DHHFT), dummy for high-VIX period (DHVIX), and the interaction term between
DHHFT and DHVIX, respectively.
161
effect on trading activity. In the CHIX dataset, RTS displays a significant positive
correlation with spread, but has no influence on the other liquidity proxies
examined. These findings reveal that stocks with a higher RTS have a significantly
lower amount of liquidity across all three analysed dimensions, notably on the
ASX market. For SIZE, the results are consistent on both markets, demonstrating
that stocks with greater market capitalisation result in significantly lower spreads
and price impacts (improved liquidity), but at the same time suffer from
significantly fewer trading activity (reduced liquidity). These results indicate that,
on both the ASX and CHIX markets, larger firms in general have greater liquidity.
Overall, based on ILLIQ, the findings show that high HFT activity has a
beneficial impact on liquidity when the expected volatility is high, even after
controlling for volatility, tick size, firm size, firm fixed effects, and day fixed
effects. Evidence suggests that the event mitigates the detrimental impact of high
HFT activity on the ASX’s illiquidity level, and enhances the beneficial impact of
high HFT activity on the CHIX’s illiquidity level. High HFT activity is also
associated with greater trading activity on the ASX and a smaller bid-ask spread on
the CHIX on days with a normal or low VIX. However, on days with high
expected volatility, high HFT activity is shown to significantly reduce trading
volume and widen spreads on the ASX and CHIX markets, respectively, thereby
negating the beneficial impact it generally has on liquidity. The asymmetric effect
observed in these markets is likely due to differences in the underlying
mechanisms on each market, such as the maker-taker fee structure offered in the
CHIX, which rewards market makers and other traders who post standing limit
orders, whereas this feature is absent in the ASX market.
4.5 Conclusion
The study proposes two hypotheses to examine whether high AXVI values (high
VIX) influence the impact of HFT activity on liquidity. In this study, high-speed
OrderIDs (HFO) represent HFT activity, while the Corwin and Schultz High-Low
Spread (CSHL), Amihud Illiquidity Ratio (ILLIQ), and Volume Turnover
(VOLTO) represent the transaction cost, price impact, and volume dimensions of
liquidity, respectively.
162
The first hypothesis postulates that “the introduction of real-time AXVI
(VIXRT) have a positive impact on HFT activity and indirectly increase liquidity
through its relationship with HFTs.” To test this hypothesis, the study uses
observations from one, six, and twelve months before and after the introduction of
the real-time AXVI event (VIXRT) to represent the short, medium, and long date
ranges, respectively. On the ASX market, univariate analysis reveals that post-
event observations have a wider spread, higher volume turnover, and have no
effect on the illiquidity ratio; however, the difference is only significant for the
shortest range. On the CHIX market, the post-event period was characterised by a
larger spread, a lower illiquidity ratio, and increased trading activity across all
examined date ranges. In addition, the availability of VIX at real time shows
significantly increased HFT activity across all date ranges on both the ASX and
CHIX markets. The multivariate analysis indicates that the post-event period has
significantly higher liquidity, with the effect being immediate for the ASX and
gradual for the CHIX. On both markets, a surge in HFT activity is associated with
a greater transaction cost (decreased liquidity), a reduction in price impact
(increased liquidity), and more trading activity (increased liquidity). The findings
also suggest that the event significantly reduces the impact of HFT activity on
liquidity in the short-term for the ASX market and in the medium- and long-term
for the CHIX market.
The second hypothesis states that “as VIX levels increase, HFT activity
changes in a way that exacerbates the already fragile liquidity situation in the
market, calling into question the credibility of HFTs as liquidity providers.” To
examine this hypothesis, the study applies univariate analysis to compare the levels
of liquidity on high VIX (Event) and non-high VIX (Non-event) days, as well as
across groups with high HFT (Treatment) and non-high HFT (Control) activities.
On both markets, the results demonstrate that transaction costs, price impact, and
trading volume are much greater on days with a high VIX (Event), indicating that,
on average, market liquidity is lower when sentiment is negative. In comparison to
the Control group, the Treatment group is generally more liquid on both markets,
as indicated by lower transaction costs and smaller price impacts. Subsequently,
the study adopts an approach similar to the difference-in-difference method to see
163
whether the level of liquidity during Non-event and Event days between the
Control and Treatment groups are significantly different.
The results show that: (i) HFT plays a significant role in reducing spreads,
and on the ASX, high-uncertainty days have no effect on this beneficial function.
However, on the CHIX, more HFT is shown to result in wider spreads. This
situation is likely due to the inherently low volume of trades on the CHIX market,
which results in higher market-making costs for HFT. These costs are magnified
during highly volatile days, and the bid-ask spread ultimately mirrors these costs;
(ii) HFT greatly reduces the price impact on both markets, and this role is even
more evident on days with negative sentiment. This is possibly due to HFTs’
market-making activities, which promotes a robust order book even when
sentiment is pessimistic; and (iii) more HFT activity does not increase volume
turnover, but rather decreases it. On the ASX, high VIX days appear to exacerbate
the negative effect of HFT on volume, whereas on the CHIX, such day appear to
mitigate this adverse effect. This behaviour is likely attributed to HFTs’ more
intense market-making activity during highly volatile days on the CHIX, which is
more concentrated in firms with large market capitalisation.
The second hypothesis is further tested using multiple regression analysis to

establish whether high HFT activity (Treatment), high VIX days (Event), and their
interaction term have a causal influence on liquidity. The results demonstrate that
the Event influences the relationship between the Treatment and liquidity measures
(ILLIQ and VOLTO on the ASX market and CSHL and VOLTO on the CHIX
market), even after accounting for volatility, tick size, firm size, firm fixed effects,
and day fixed effects. On the ASX, days with negative market sentiment lower the
initially detrimental but insignificant influence that high HFT activity has on price
impacts, as well as reducing the originally beneficial impact that high HFT activity
has on trading activity. On the CHIX, high VIX days are shown to reduce the
favourable but insignificant effect that high HFT activity has on reducing spreads,
while enhancing the beneficial effect that high HFT activity has on reducing price
impacts.
164
Overall, the findings show that high VIX days have a significant influence
on the relationship between high HFT activity and liquidity, with varying impacts
on liquidity in both the ASX and CHIX markets. Nevertheless, the data also
suggest that the Event may increase the Treatment’s beneficial effect in lowering
the illiquidity ratio, indicating that high HFT activity might lead to a reduced price
impact during days with negative market sentiment in both markets.
165
CHAPTER FIVE:
CONCLUSION
5
This chapter concludes the thesis by providing a summary of the key findings and
implications for each of the three essays, presented in Section 5.1. Section 5.2 shows
the limitations of this study and suggests potential directions for future research.
5.1 Major findings and implications

5.1.1 Essay One: High-frequency trading: Definition, Implications, and
Controversies
The first essay of this thesis provides a comprehensive survey of the relevant
literature on HFT, including regulatory reports, theoretical and empirical research.
The research concludes that the lack of a universal definition for HFT leads to
additional problems, such as the inaccurate estimation of the HFTs’ market shares,
and obscure their actual influence on market quality. This aspect further complicates
attempts to comprehend HFT, which consequently leads to varying HFT-related
conclusions. Nevertheless, from regulatory perspectives, five characteristics of HFT
are frequently mentioned: (i) a specialised form of AT that requires a low-latency
network; (ii) the use of high-speed, sophisticated computer programmes and
systems; (iii) extremely high order-to-transaction ratios; (iv) extremely short average
holding periods; and (v) ending the trading day with flat positions.
HFT operations are based on the execution of numerous transactions with

very slim profit margins, magnified by large trading volume (NAFM, 2010; Zhang,
2010). HFTs analyse order book flow for short-term inefficiencies to execute latency
arbitrage, benefitting from the trading environment as opposed to the fundamental
value of financial assets (Hasbrouck & Saar, 2013). Annually, the estimated
worldwide revenue from latency arbitrage is US$5 billion (Aquilina et al., 2020).
HFT trading strategies may be beneficial or harmful to the market (Benos & Sagade,
2016); hence, HFT is a double-edged sword. Beneficial strategies include statistical
arbitrage, directional trading, and market-making, whereas harmful strategies uses
front running, spoofing and layering, and quote stuffing.
166
The essay also highlights HFT-related controversy. Perhaps the most
infamous case is the flash crash of May 6, 2010, when HFTs were criticised for
exacerbating price declines and market volatility in their competition for liquidity
(Kirilenko et al., 2017). Moreover, the arms race among HFTs to be the fastest,
fuelled by the winner-take-all nature of the game, is perceived as unproductive,
socially wasteful, and the substantial money invested to minimise latency by a few
milliseconds raises doubts about whether HFT provides value overall (Chordia et al.,
2013; Budish et al., 2015; Jones, 2013). HFTs are also accused of supplying liquidity
on the thick side of the order book and taking liquidity on the thin side, which
resembles order-anticipation strategies (Goldstein, et al., 2020). HFTs’ market-
making activity is also feared to aggravate execution uncertainty, particularly in
volatile markets and thinly traded securities, since they are not mandated to provide
liquidity at all times and may withdraw whenever they please (Anand &
Venkataraman, 2013; Chung & Chuwonganant, 2018; Zhang, 2010).
As HFT has become more pervasive in the last decade, regulatory agencies
and scholars around the world have taken or proposed a number of measures to level
the playing field. This includes (i) applying asymmetrical speed bumps to slow down
message traffic incoming to and outgoing from an exchange (Khapko & Zoican,
2020); (ii) instituting a “price improvement rule” to reduce the incentive for
predatory HFTs from using latency arbitrage to exploit the difference between the
bid-ask spreads on the lit and dark markets (ASIC, 2015); (iii) using “frequent batch
auctions” to promote price competition rather than speed competition, thereby
lowering the incentive to be the fastest (Budish et al., 2015); and (iv) utilising
“continuous scaled limit order” to slow down the inflow of large orders into the
market, hence reducing the adverse selection costs for slower traders (Kyle & Lee,
2017). Nevertheless, Harris (2013) advises that financial authorities should be
mindful in their regulatory oversight so as not to inadvertently impede beneficial
HFT strategies.
5.1.2 Essay Two: Relative Tick Size and High-frequency Trading

The second essay investigates the influence of relative tick size on HFT activity. For
the purpose of this study, four different measures of HFT activity are used: the
message-to-trade ratio (MTR), the algorithmic trading ratio (ALGO), the high-
167
frequency OrderID (HFO), and the HFO’s message ratio (HFOR). The research also
uses a study period from January 2008 to December 2017 for the ASX, and from
December 2011 to December 2017 for the CHIX, to compare HFT activity in two
groups with contrasting relative tick sizes (i.e., small and large relative tick size) in
three scenarios: (i) stocks with similar nominal tick sizes; (ii) stocks priced near the
A$2.00 tick size border; and (iii) stocks that crossed the A$2.00 tick size border in
both upwards and downwards directions.
Overall, the results of the second essay indicate that HFTs favour stocks with
small relative tick size, which suggests that they would choose order undercutting
over order-queuing in both the ASX and CHIX markets. This behaviour is
particularly evident when a large number of informed traders are present, and the
relative tick size is sufficiently small. The findings further demonstrate that HFTs
would abandon stocks with fine pricing grids once their pricing grids become
coarse. The inverse relationship between HFT activity and relative tick size implies
that market-making HFTs are risk-averse. This evidence also underlines the
importance of order-undercutting approaches for market-making HFTs, due to their
extremely low tolerance for adverse selection risks.
The findings of the second essay lend credence to the perception that the
primary strategy of HFTs is to generate tiny profits (NAFM, 2010) while keeping
their risk exposure to an absolute minimum. Due to the fact that market-making
HFTs may execute the same strategy thousands of times each day, the little profits
rapidly accumulate into substantial amounts (Zhang, 2010), yet the probability of
losing a trade remains minimal. Overall, the outcomes of this research contradict the
arguments that a small relative tick size would discourage the liquidity provision of
market-making HFTs and that a large relative tick size would attract more market-
making HFTs, as suggested by Angel (2011), O’Hara et al. (2019), Werner et al.
(2019), and Yao and Ye (2018). This contradicting results are likely because of the
difference in the market microstructure underlying the Australian and the U.S.
markets. The findings of this study demonstrate that HFTs are exceptionally risk-
averse traders that prioritise risk minimisation over profit maximisation.
168
Chordia et al. (2013) argues that the HFTs might compromise the welfare of
non-HFT participants; this study provides evidence that policymakers may
implement a dynamic tick size policy to allocate HFT activity into stocks where it is
most required. For instance, stocks with a low liquidity issue may need the
intervention of market-making HFTs to improve their liquidity level, which may be
facilitated by assigning a very small nominal tick size to such stocks. In contrast, if
the stocks are highly liquid and the presence of HFT activity is considered toxic, the
regulator could increase the nominal tick size to enforce price-priority, hence
compelling the use of the order-queuing strategy which would discourage HFTs
from trading these stocks.
5.1.3 Essay Three: Expected Volatility, High-frequency Trading, and

Liquidity
The third essay demonstrates how expected volatility influences HFT activity, and
how the resulting change in HFT activity affects liquidity. The first part of the essay
utilises observations from one (short-term), six (medium-term), and twelve months
(long-term) around the day real-time AXVI are made available. On both markets, an
increase in HFT activity corresponds with a rise in transaction costs (decreased
liquidity), a reduction in price impact (increased liquidity), and an improvement in
trading activity (increased liquidity). The findings also imply that the event
significantly lowers the influence of HFT activity on liquidity in the short-term for
the ASX, and medium- and long-term for the CHIX markets, respectively.
The findings implies that the introduction of real-time AXVI index has
increased the amount of publicly available hard information in real-time, allowing
HFTs to capitalise on the new feature and take strategic positions in the market. The
real-time feature are also more valuable for HFTs as they are able to react more
quickly and aggressively to the information generated by such data than non-HFTs.
Moreover, the innovation could incentivise HFT to be more aggressive, which may
impose adverse selection costs on market liquidity providers. The observed reversal
on the ASX in the immediate period following the event might be caused by greater
presence of liquidity-taking HFTs relative to liquidity-providing HFTs on the
market. The absence of such impact on CHIX is likely attributable to the fact that
169
most CHIX participants are professional traders who deploy trading bots, thus, are
more resistant to the influence of the new feature.
The second part of the essay examines liquidity levels on days with negative
market sentiment (high VIX) with days with normal sentiment (non-high VIX), as
well as across stocks with differing amounts of HFT activity (i.e., high HFT activity
or non-high HFT activity). On the ASX, negative sentiment erodes the favourable
effect that high HFT activity usually has in promoting higher trading activity when
the sentiment is normal. On the CHIX, negative sentiment hinders the positive
influence that high HFT activity has in reducing spreads during normal market
sentiment. On the other hand, high HFT activity has been proven to result in a lower
illiquidity ratio on days with normal sentiment, and this positive impact is amplified
on days with negative sentiment on both markets. This finding suggests that high
HFT activity may result in a lower price impact when the market sentiment is
negative. In addition, the data indicate that negative market sentiment has a
significant influence on the effect of HFT activity on liquidity measures, with
differing impacts on liquidity in the ASX and CHIX markets. Nonetheless, the
findings signify that when the market sentiment is negative, stocks with high HFT
activity are likely to experience a smaller price impact on either the ASX or CHIX
markets, indicating that the presence of HFT may reduce excessive price impacts
when the market is extremely fearful.
Overall, the findings of this study imply that a greater quantity of hard
information and a negative market sentiment can significantly influence the effect of
high-frequency trading (HFT) on liquidity. This study also demonstrates that more
HFT activity is not always negative, since it promotes liquidity by reducing price
impacts on days with high uncertainties. Therefore, any attempt to completely ban or
restrict HFT in the market might be viewed as inappropriate, as such actions could
accidentally eliminate valuable market participants.
5.2 Limitations of the study and suggestions for future research

As with any other research, this thesis has limitations. An important caveat for Essay
One is that it is impossible to include all regulatory reports, practitioner papers, and
academic publications related to HFT. In order to gain a better understanding of
170
HFT’s impact on the market, the author made every effort to ensure that the
literature included paints a clearer image of HFT. Moreover, given that HFTs
are dynamic in nature, their current behaviour and strategies may have also
progressed, resulting in outcomes that potentially differ from those mentioned in the
survey.
All of the HFT measures used in Essays Two and Three are proxies, and
therefore cannot be regarded as accurately reflecting the actions of high-frequency
trading firms such as Citadel Securities, Jump Trading, Liquid Capital, Susquehanna
International Group, and Virtu Financial. Instead, any conclusions on “HFT activity”
generated from this research should be interpreted as representing the activity of
“low-latency traders” as a whole, and not just “full-fledged HFT firms.”
Furthermore, the method in which they are calculated suggests that they can only
estimate the liquidity-providing activity of HFT and not the liquidity-taking activity
of HFT. In addition, to the best of the researcher’s knowledge, the high-frequency
OrderID (HFO) and the HFO’s message ratio (HFOR) have never been used in any
other study. The author would also like to emphasise that, although HFO reflects the
total number of unique OrderIDs in a stock on a given day, the amount indicated
might be the result of a single trading account using rapid order placement with
multiple entry points on the same day. Consequently, HFO and HFOR may also be
interpreted as signifying the presence of high-frequency (or low-latency) traders in a
stock on a certain day.
From Essay One, future research might incorporate the most recent
regulatory actions and HFT-related microstructure changes. In addition, the author
urges market regulators around the world to collaborate with academic scholars to
produce a higher quality dataset, which is vital to ensuring that the conclusions
formed about HFT are accurate, so that they are not held accountable for something
they did not do. From Essay Two, future studies could evaluate the presence of non-
HFTs in stocks with large or small relative tick sizes to assess whether their
willingness to trade in stocks with small relative tick sizes is impeded by the
existence of HFTs, and vice versa. In addition, the samples may be segmented based
on whether they are tick-constrained or tick-unconstrained, since the literature
indicates that this factor may significantly affect market-making HFTs’ approach in
171
providing liquidity (see, e.g. Foley et al., 2019; O’Hara et al., 2019; Werner et al.,
2019; Yao & Ye, 2018). From Essay Three, future research may conduct the
following research based on the findings of the essay: (i) to investigate the influence
of HFT activity during negative market sentiment periods on other market quality
measures, including price discovery, volatility, and adverse selection costs; (ii) make
use of data from the Covid-19 period to examine how HFT operates when the market
is extremely volatile and fearful; and (iii) examine the behaviour and influence of
HFT on the liquidity of stocks with extreme intraday price swings.
Furthermore, futures studies might also validate the accuracy of the newly
proposed HFT measures (i.e., HFO and HFOR) using datasets that has correctly
identified HFT activity in the market, such as the NASDAQ’s HFT dataset. In
addition, the versatility of the HFO and HFOR measures enables their use in
intraday-level research, and the threshold values (i.e., order resting time and
minimum number of linked messages) employed in this study may be adjusted
according to the researcher’s needs and preferences. Thus, these measurements could
be used to analyse the behaviour of the trading algorithms employed by both slow
and fast market participants.
172
REFERENCES
Acharya, V. V., & Pedersen, L. H. (2005). Asset pricing with liquidity risk. Journal
of Financial Economics, 77(2), 375-410.
Agarwal, A. (2012). High-frequency trading: Evolution and the future. London, UK.
Aitken, M. J., Aspris, A., Foley, S., & Harris, F. H. D. B. (2018). Market fairness:
The poor country cousin of market efficiency. Journal of Business Ethics, 147,
5–23.
Aitken, M., Cumming, D., & Zhan, F. (2017). Trade size, high-frequency trading,
and colocation around the world. European Journal of Finance, 23, 781–801.
Ait-Sahalia, Y., & Sağlam, M. (2017). High frequency market making: Optimal
quoting. Available at SSRN 2331613.
Aldridge, I. (2013). High-frequency trading: A practical guide to algorithmic. New
Jersey: John Wiley & Sons, Inc.
Amihud, Y. (2002). Illiquidity and stock returns: cross-section and time-series
effects. Journal of Financial Markets, 5(1), 31-56.
Amihud, Y., & Mendelson, H. (1991). Volatility, efficiency, and trading: Evidence
from the Japanese stock market. The Journal of Finance, 46(5), 1765-1789.
Amihud, Y., Mendelson, H., & Pedersen, L. H. (2006). Liquidity and asset
prices. Foundations and Trends® in Finance, 1(4), 269-364.
Amihud, Y., Mendelson, H., & Pedersen, L. H. (2012). Market liquidity: asset
pricing, risk, and crises. Cambridge University Press.
Anand, A., & Venkataraman, K. (2013). Should exchanges impose market maker
obligations?. Working paper. Syracuse University and Southern Methodist
University.
Ang, A., Chen, J., & Xing, Y. (2006). Downside risk. The Review of Financial
Studies, 19(4), 1191-1239.
Angel, J. J. (1997). Tick size, share prices, and stock splits. The Journal of Finance,
52(2), 655-681.
Angel, J. J. (2012). Tick size regulation: costs, benefits, and risks. Foresight, the
future of computerised trading in financial markets, Economic Impact
Assessment EIA7 London, UK: UK Government Office for Science (c) Crown
Copyright.
Angel, J. J. (2014). When finance meets physics: The impact of the speed of light on
financial markets and their regulation. Financial Review, 49, 271–281.
Angel, J. J., & McCabe, D. (2013). Fairness in financial markets: The case of high-
frequency trading. Journal of Business Ethics, 112, 585–595.
Angel, J. J., Harris, L. E., & Spatt, C. S. (2011). Equity trading in the 21st century.
The Quarterly Journal of Finance, 1(01), 1-53.
173
Aquilina, M., & Ysusi, C. (2016). Are high-frequency traders anticipating the order
flow? Cross-venue evidence from the UK market. FCA Occasional Paper,
No.16.
Aquilina, M., Budish, E. B., & O’Neill, P. (2020). Quantifying the high-frequency
trading “arms race”: A simple new methodology and estimates. Chicago Booth
Research Paper (2020-2086).
Aquilina, M., Foley, S., O’Neill, P., & Ruf, T. (2016). Asymmetries in dark pool
reference prices. FCA Occasional Paper 21.
Aspris, A., Foley, S., Harris, D., & O’Neill, P. (2015). Time pro-rata matching:
Evidence of a change in LIFFE STIR futures. Journal of Futures Markets, 35,
522–541.
Australian Securities and Investments Commission. (2010). Australian equity market
structure. Report 215.
Australian Securities and Investments Commission. (2013). Dark liquidity and high-
frequency trading. Report 331.
Australian Securities and Investments Commission. (2015). Review of high-
frequency trading and dark liquidity. Report 452.
Australian Securities Exchange (2011). PureMatch to go-live 28 November.
Retrieved from: https://www.asx.com.au/documents/media/PureMatchLaunch
andPricingFinal-AustralianSecuritesExchange-ASX.pdf
Avramovic, A., Lin, V., & Krishnan, M. (2017). We’re all high-frequency traders
now. Credit Suisse.
Bacidore, J. M. (1997). The impact of decimalization on market quality: An
empirical investigation of the Toronto Stock Exchange. Journal of Financial
Intermediation, 6(2), 92-120.
Baldauf, M., & Mollner, J. (2020). High-frequency trading and market performance.
The Journal of Finance, 75, 1495– 1526.
Baron, M., Brogaard, J., Hagströmer, B., & Kirilenko, A. (2019). Risk and return in
high-frequency trading. Journal of Financial and Quantitative Analysis, 54,
993–1024.
Benos, E., & Sagade, S. (2016). Price discovery and the cross-section. Journal of
Financial Markets, 30, 54–77.
Bessembinder, H. (2003). Trade execution costs and market quality after
decimalization. Journal of Financial and Quantitative Analysis, 38(4), 747-
777.
Bhattacharya, N., Chakrabarty, B., & Wang, X. F. (2020). High-frequency traders
and price informativeness during earnings announcements. Review of
Accounting Studies, 25(3), 1156-1199.
Biais, B., & Woolley, P. (2011). High-frequency trading. Working paper. Toulouse
School of Economics.
Biais, B., Foucault, T., & Moinas, S. (2015). Equilibrium fast trading. Journal of
Financial Economics, 116, 292–313.
174
Biais, B., Glosten, L., & Spatt, C. (2005). Market microstructure: A survey of
microfoundations, empirical results, and policy implications. Journal of
Financial Markets, 8(2), 217-264.
Boehmer, E., Fong, K., & Wu, J. (2020). Algorithmic trading and market quality:
International evidence. Journal of Financial and Quantitative Analysis, 1–30.
Boehmer, E., Li, D., & Saar, G. (2018). The competitive landscape of high-
frequency trading firms. The Review of Financial Studies, 31, 2227–2276.
Bourghelle, D., & Declerck, F. (2004). Why markets should not necessarily reduce
the tick size. Journal of Banking and Finance, 28(2), 373-398
Breckenfelder, J. (2019). Competition among high-frequency traders, and market
quality. ECB Working paper 2290.
Brogaard, J., & Garriott, C. (2019). High-frequency trading competition. Journal of
Financial and Quantitative Analysis, 54, 1469–1497.
Brogaard, J., Carrion, J., Moyaert, T., Riordan, R., Shkilko, A., & Sokolov, K.
(2018). High frequency trading and extreme price movements. Journal of
Financial Economics, 128, 253–265.
Brogaard, J., Hagströmer, B., Nordén, L., & Riordan, R. (2015). Trading fast and
slow: Colocation and liquidity. The Review of Financial Studies, 28, 3407–
3443.
Brogaard, J., Hendershott, T., & Riordan, R. (2014). High-frequency trading and
price discovery. The Review of Financial Studies, 27, 2267–2306.
Brogaard, J., Hendershott, T., & Riordan, R. (2019). Price discovery without trading:
Evidence from limit orders. The Journal of Finance, 74, 1621–1658.
Brunnermeier, M. K., & Pedersen, L. H. (2005). Predatory trading. The Journal of
Finance, 60, 1825–1863.
Budish, E., Cramton, P., & Shim, J. (2015). The high-frequency trading arms race:
Frequent batch auctions as a market design response. Quarterly Journal of
Economics, 130, 1547–1621.
Carrion, A. (2013). Very fast money: High-frequency trading on the NASDAQ.
Journal of Financial Markets, 16, 680–711.
Cartea, A., & Penalva, J. (2012). Where is the value in high-frequency trading?
Quarterly Journal of Finance, 2, 1–46.
Cartea, Á., Payne, R., Penalva, J., & Tapia, M. (2019). Ultra-fast activity and
intraday market quality. Journal of Banking & Finance, 99, 157–181.
Chordia, T., Goyal, A., Lehmann, B. N., & Saar, G. (2013). High-frequency trading.
Chordia, T., Roll, R., & Subrahmanyam, A. (2011). Recent trends in trading activity
and market quality. Journal of Financial Economics, 101(2), 243-263.
Chung, K. H., & Chuwonganant, C. (2018). Market volatility and stock returns: The
role of liquidity providers. Journal of Financial Markets, 37, 17–34.
Chung, K. H., & Lee, A. J. (2016). High-frequency trading: Review of the literature
and regulatory initiatives around the world. Asia-Pacific Journal of Financial
Studies, 45, 7–33.
175
Clark-Joseph, A. D., Ye, M., & Zi, C. (2017). Designated market makers still matter:
Evidence from two natural experiments. Journal of Financial Economics, 126(3),
652-667.
Comerton-Forde, C. (2012). Is Australia HFT-friendly? JASSA: The Finsia Journal
of Applied Finance, 3(3), pp. 12-14.
Comerton-Forde, C., Malinova, K., & Park, A. (2018). Regulation dark trading:
Order flow segmentation and market quality. Journal of Financial Economics,
130, 347–366.
Committee of European Securities Regulators. (2010). Call for evidence on micro-
structural issues of the European equity markets. Paris.
Conrad, J., & Wahal, S. (2020). The term structure of liquidity provision. Journal of
Financial Economics, 136, 239– 259.
Conrad, J., Wahal, S., & Xiang, J. (2015). High-frequency quoting, trading, and the
efficiency of prices. Journal of Financial Economics, 116, 271–291.
Cooper, R., Davis, M., & Vliet, B. V. (2016). The mysterious ethics of high-
frequency trading. Business Ethics Quarterly, 26, 1–22.
Cordella, T., & Foucault, T. (1999). Minimum price variations, time priority, and
quote dynamics. Journal of Financial Intermediation, 8(3), 141-173
Corwin, S. A., & Schultz, P. (2012). A simple way to estimate bid‐ask spreads from
daily high and low prices. The Journal of Finance, 67(2), 719-760.
Dodd-Frank Wall Street Reform and Consumer Protection Act, Sec. (2010).
Retrieved from: https://www.govinfo.gov/content/pkg/PLAW-111publ203
/pdf/PLAW-111publ203.pdf
Donefer, B. S. (2010). Algos gone wild: Risk in the world of automated trading
strategies. Journal of Trading, 5, 31–34.
Duan, J. C., & Zou, Q. (2014). Liquidity and default. Available at SSRN 2450890.
Durbin, M. (2010). All about high-frequency trading: The easy way to get started.
New York: McGraw-Hill Companies, Inc.
Dyhrberg, A. H., Foley, S., & Svec, J. (2020). When bigger is better: The impact of a
tiny tick size on undercutting behavior. Retrieved from https://www.org/SSRN
3194932.
Easley, D., & O’hara, M. (1992). Time and the process of security price
adjustment. The Journal of Finance, 47(2), 577-605.
Easley, D., Prado, M. M., & O’Hara, M. (2011). The microstructure of the “Flash
Crash”: Flow toxicity, liquidity crashes, and the probability of informed
trading. Journal of Portfolio Management, 37, 119–128.
Easley, D., Prado, M. M., & O’Hara, M. (2012). The volume clock: Insights into the
high-frequency paradigm. Journal of Portfolio Management, 39, 19–29.
Edwards, T., & Preston, H. (2017). Reading VIX®: Does VIX Predict Future
Volatility? S&P Dow Jones Indices Research. Retrieved from:
https://www.spglobal.com/spdji/en/documents/research/research-reading-vix-
does-vix-predictfuture-volatility.pdf
176
Egginton, J. F., Van Ness, B. F., & Van Ness, R. A. (2016). Quote stuffing.
Financial Management, Fall, 45, 583–608.
Elmer-DeWitt (2008). How Steve Jobs’ obit got published. Retrieved from
https://fortune.com/2008/08/28/how-steve-jobs-obit-got-published/
Ersan, O., & Ekinci, C. (2016). Algorithmic and high-frequency trading in Borsa
Istanbul. Borsa Istanbul Review, 16(4), 233-248.
European Securities and Markets Authority. (2011). Guidelines on systems and
controls in a highly automated trading environment for trading platforms,
investment firms and competent authorities. Paris.
Fama, E. F. (1970). Efficient capital markets: A review of theory and empirical
work. The Journal of Finance, 25(2), 383–417.
Financial Industry Regulatory Authority. (2012). FINRA joins exchanges and the
SEC in fining Hold Brothers more than $5.9 million for manipulative trading,
anti-money laundering, and other violations. Washington, D.C.
Foley, S., Dyhrberg, A., & Svec, J. (2022a). When bigger is better: the impact of a
tiny tick size on undercutting behavior. Journal of Financial and Quantitative
Analysis.
Foley, S., Gorbenko, A., & Ruf, T. (2019). Can fast and slow liquidity providers co-
exist in modern equity markets.
Foley, S., Meling, T., & Ødegaard, B. A. (2022b). Tick size wars: The market
quality effects of pricing grid competition. Review of Finance, rfac32.
Foucault, T., Hombert, J., & Rosu, I. (2016). News trading and speed. The Journal of
Finance, 71(1), 335–382.
Foucault, T., Kozhan, R., & Tham, W. W. (2017). Toxic arbitrage. The Review of
Financial Studies, 30, 1053–1094.
Fox, M. B., Glosten, L. R., & Rauterberg, G. V. (2017). Informed trading and its
regulation. Journal of Corporation Law, 43(4), 817–898.
Friederich, S., & Payne, R. (2015). Order-to-trade ratios and market liquidity.
Journal of Banking and Finance, 50, 214–223.
Frino, A., Mollica, V., & Zhang, S. (2015). The impact of tick size on high-
frequency trading: The case for splits. Working paper.
Froot, K. A., Scharfstein, D. S., & Stein, J. C. (1992). Herd on the street:
Informational inefficiencies in a market with short-term speculation. The
Journal of Finance, 47, 1461–1484.
Gibson, S., Singh, R., & Yerramilli, V. (2003). The effect of decimalization on the
components of the bid-ask spread. Journal of Financial Intermediation, 12(2),
121-148.
Gider, J., Schmickler, S., & Westheide, C. (2019). High-frequency trading and price
informativeness. SAFE Working Paper Series, 248.
Glosten, L. R. (1994). Is the electronic open limit order book inevitable?. The
Journal of Finance, 49(4), 1127-1161.
177
Goldstein, M. A., & Kavajecz, K. A. (2000). Eighths, sixteenths, and market depth:
changes in tick size and liquidity provision on the NYSE. Journal of Financial
Economics, 56(1), 125-149.
Goldstein, M. A., Kumar, P., & Graves, F. C. (2014). Computerized and high-
frequency trading. Financial Review, 49, 177–202.
Goldstein, M. A., Kwan, A., & Philip, R. (2020). High-frequency trading strategies.
Retrieved from https://www. org/SSRN 2973019.
Golub, A., Dupuis, A., & Olsen, R. B. (2013). High-frequency trading in FX
markets. In D. Easley, M. L. Prado, & M. O’Hara (eds.), High-frequency
trading new realities for traders, markets and regulators (pp. 65–90). London:
Risk Books.
Gomber, P., Arndt, B., Lutat, M., & Uhle, T. (2011). High-frequency trading.
Retrieved from: http://doi.org/10.2139/ssrn.1858626
Goodhart, C. A., & O’Hara, M. (1997). High-frequency data in financial markets:
Issues and applications. Journal of Empirical Finance, 4, 73–114.
Gromb, D., & Vayanos, D. (2002). Equilibrium and welfare in markets with
financially constrained arbitrageurs. Journal of Financial Economics, 66, 361–
407.
Grossman, S. J., & Miller, M. H. (1988). Liquidity and market structure. The Journal
of Finance, 43, 617–633.
Grossman, S. J., & Stiglitz, J. E. (1980). On the impossibility of informationally
efficient markets. The American Economic Review, 70(3), 393-408.
Hagströmer, B., & Nordén, L. (2013). The diversity of high-frequency traders.
Hagströmer, B., Nordén, L., & Zhang, D. (2014). How aggressive are high-
frequency traders? Financial Review, 49, 395–419.
Han, J., Khapko, M., & Kyle, A. (2014). Liquidity with high frequency market
making. Unpublished working paper. University of Maryland, College Park,
MD.
Harris, L. (1991). Stock Price Clustering and Discreteness. The Review of Financial
Studies, 4(3), 389–415.
Harris, L. (1994). Minimum price variations, discrete bid-ask spreads, and quotation
sizes. The Review of Financial Studies, 7(1), 149-178.
Harris, L. (1996). Does a large minimum price variation encourage order
exposure? (Vol. 96, No. 5). New York Stock Exchange.
Harris, L. (2003). Trading and exchanges: Market microstructure for practitioners.
New York: Oxford University Press.
Harris, L. (2013). What to do about high-frequency. Financial Analysts Journal, 69,
6–9.
Harris, L. (2015). Trading and Electronic markets: What investment professionals
need to know. VA: Research Foundation of CFA Institute.
Hasbrouck, J. (2019). Price discovery in high resolution. Journal of Financial
Econometrics, nbz027.
178
Hasbrouck, J., & Saar, G. (2013). Low-latency trading. Journal of Financial
Markets, 16, 646–679.
Hendershott, T., & Riordan, R. (2009). Algorithmic trading and
information. Manuscript, University of California, Berkeley.
Hendershott, T., & Riordan, R. (2013). Algorithmic trading and the market for
liquidity. Journal of Financial and Quantitative Analysis, 48, 1001–1024.
Hendershott, T., Jones, C. M., & Menkveld, A. J. (2011). Does algorithmic trading
improve liquidity? The Journal of Finance, 66, 1–33.
Hirschey, N. (2021). Do high-frequency traders anticipate buying and selling
pressure?. Management Science, 67(6), 3321-3345.
Hoffmann, P. (2014). A dynamic limit order market with fast and slow
traders. Journal of Financial Economics, 113(1), 156-169.
Hull, J. C. (2003). Options futures and other derivatives. Pearson Education Limited
Jarnecic, E., & Snape, M. (2014). The provision of liquidity by high-frequency
participants. Financial Review, 49, 371–394.
Jarrow, R. A., & Protter, P. (2012). A dysfunctional role of high frequency trading in
electronic markets. International Journal of Theoretical and Applied Finance,
15, 1250022.
Jones, C. M. (2013). What do we know about high-frequency trading? Columbia
Business School Research Paper No. 13-11.
Jones, C. M., & Lipson, M. L. (2001). Sixteenths: direct evidence on institutional
execution costs. Journal of Financial Economics, 59(2), 253-278.
Jovanovic, B., & Menkveld, A. J. (2016). Middlemen in limit order markets.
Working paper. Retrieved from https://ssrn.com/abstract=1624329
Kauffman, R. J., Hu, Y., & Ma, D. (2015). Will high-frequency trading practices
transform the financial markets in the Asia Pacific Region? Financial
Innovation, 1, 1–27.
Kauffman, R. J., Liu, J., & Ma, D. (2015). Innovations in financial IS and
technology ecosystems: High-frequency trading in the equity market.
Technological Forecasting and Social Change, 99, 339–354.
Kaya, O. (2016). High-frequency trading reaching the limits. Deutsche Bank
Research.
Khapko, M., & Zoican, M. (2020). Do speed bumps curb low-latency investment?
Evidence from a laboratory mar ket. Journal of Financial Markets, 100601.
Kirilenko, A., Kyle, A. S., Samadi, M., & Tuzun, T. (2017). The flash crash: High-
frequency trading. The Journal of Finance, 72, 967–998.
Korajczyk, R., & Murphy, D. (2019). High-frequency market making to large
institutional trades. The Review of Financial Studies, 32, 1034–1067.
Kosinski, R. J. (2013). A literature review on reaction time. Working paper.
Clemson University.
Kozhan, R., & Tham, W. W. (2012). Execution risk in high-frequency arbitrage.
Management Science, 58, 2131–2149.
179
Kyle, A. S. (1985). Continuous auctions and insider trading. Econometrica, 53,
1315–1335.
Kyle, A. S., & Lee, J. (2017). Toward a fully continuous exchange. Oxford Review of
Economic Policy, 33, 650–675.
Lash, H., & Spicer, J. (2010). Mystery trader in selloff was Waddell: Kansas-based
manager sold large order of e-mini contracts during crucial time period,
document shows. The Globe and Mail, May 14.
Le, H., & Gregoriou, A. (2020). How do you capture liquidity? A review of the
literature on low‐frequency stock liquidity. Journal of Economic Surveys, 34(5),
1170-1186.
Lee, E. J. (2015). High frequency trading in the Korean index futures
market. Journal of Futures Markets, 35(1), 31-51.
Lhabitant, F.-S., & Gregoriou, G. N. (2015). High-frequency trading: Past, present
and future. In G. N. Gregoriou (ed.), Handbook of high-frequency trading (pp.
155–165). Waltham, MA: Academic Press.
Li, S., Wang, X., & Ye, M. (2021). Who provides liquidity, and when?. Journal of
Financial Economics, 141(3), 968-980.
Liberti, J. M., & Petersen, M. A. (2019). Information: Hard and soft. Review of
Corporate Finance Studies, 8(1), 1-41.
Linton, O., O’Hara, M., & Zigrand, J. P. (2013). The regulatory challenge of high-
frequency markets. High-Frequency Trading, 227.
Lipson, M. L., & Mortal, S. (2006). The effect of stock splits on clientele: Is tick size
relevant?. Journal of Corporate Finance, 12(5), 878-896.
London Stock Exchange Group. (2010) Response to CESR call for evidence on
micro-structural issues of the European Equity Markets.
MacKenzie, D. (2015). Mechanizing the merc: The Chicago Mercantile Exchange
and the rise of high-frequency trading. Technology and Culture, 56, 646–675.
Madhavan, A. (2000). Market microstructure: A survey. Journal of Financial
Markets, 3(3), 205-258.
Madhavan, A. (2002). Market microstructure: A practitioner’s guide. Financial
Analysts Journal, 58(5), 28-42.
Mahmoodzadeh, S., & Gençay, R. (2017). Human vs. high-frequency traders, penny
jumping, and tick size. Journal of Banking & Finance, 85, 69-82.
Malceniece, L., Malcenieks, K., & Putniņš, T. J. (2019). High frequency trading and
comovement in financial markets. Journal of Financial Economics, 134, 381–
399.
Malinova, K., & Park, A. (2020). “Sniping” in fragmented markets. McMaster
University and University of Toronto working paper.
Manahov, V. (2016). A note on the relationship between high-frequency trading and
latency arbitrage. International Review of Financial Analysis, 47, 281–296.
Manahov, V., Hudson, R., & Viktor, B. G. (2014). Does high-frequency trading
affect technical analysis and market efficiency? And if so, how? Journal of
International Financial Market, Institutions and Money, 28, 131–157.
180
Markham, J. W. (2002). A financial history of the United States. 1. From
Christopher Columbus to the robber barons (1492-1900). New York: M. E.
Sharpe.
Maynard, M. (2008). United Airlines shares fall on false report of bankruptcy. The
New York Times. Retrieved from: https://www.nytimes.com/2008/09/09/
business/worldbusiness/09iht-AIR.1.16001879.html
Menkveld, A. J. (2013). High-frequency trading and the new market makers. Journal
of Financial Markets, 16, 712– 740.
Menkveld, A. J. (2014). High-frequency traders and market structure. Financial
Review, 49, 333–344.
Menkveld, A. J. (2016). The Economics of High-Frequency Trading. Annual Review
of Financial Economics, 8, 1-24.
Moosa, I., & Ramiah, V. (2015). The profitability of high-frequency trading: is it for
real? In The handbook of high-frequency trading (pp. 25–45). Waltham, MA:
Academic Press.
Moriyasu, H., Wee, M., & Yu, J. (2018). The role of algorithmic trading in stock
liquidity and commonality in electronic limit order markets. Pacific-Basin
Finance Journal, 49, 103–128.
Narang, R. K. (2013). Inside the black box: A Simple Guide to Quantitative (2nd
ed.). New Jersey: John Wiley & Sons, Inc.
Netherlands Authority of Financial Markets. (2010). High-frequency trading: The
application of advanced trading technology in the European marketplace,
Amsterdam.
O’Hara, M. (2014). High-frequency trading and its impact on markets. Financial
Analysts Journal, 70(3), 18-27.
O’Hara, M. (2015). High-frequency market microstructure. Journal of Financial
Economics, 116, 257–270.
O’Hara, M., 1995. Market microstructure theory. Basil Blackwell, Cambridge, MA.
O’Hara, M., Saar, G., & Zhong, Z. (2019). Relative tick size and the trading
environment. Review of Asset Pricing Studies, 9, 47–90.
Parlour, C. A., and D. J. Seppi. 2008. Limit order markets: A survey. In A. Boot and
A. Thakor, eds. Handbook of financial intermediation and banking.
Amsterdam, the Netherlands: Elsevier.
Patterson, S. (2010). Did shutdowns make plunge worse? Retrieved from:
https://www.wsj.com/articles/SB1000142405274870368630457522872143948
0374
Pirrong, C. (2014). Pick your poison – Fragmentation or market power? An analysis
of RegNMS, high frequency trading, and securities market structure. Journal
of Applied Corporate Finance, 26, 8–14.
Putniņš, T. J., & Barbara, J. (2020). The good, the bad, and the ugly: How
algorithmic traders impact institutional trading costs.
Reilly, F. K. & Brown, K. C. (2012). Investment Analysis & Portfolio Management.
South-Western Cengage Learning.
181
Riordan, R., & Storkenmaier, A. (2012). Latency, liquidity and price discovery.
Sarr, A., & Lybek, T. (2002). Measuring liquidity in financial markets. Available at
SSRN 880932.
Sarwar, G. (2017). Examining the flight-to-safety with the implied
volatilities. Finance Research Letters, 20, 118-124.
Savor, P. G. (2012). Stock returns after major price shocks: The impact of
information. Journal of financial Economics, 106(3), 635-659.
Scholtus, M., van Dijk, D., & Frijns, B. (2014). Speed, algorithmic trading, and
market quality around macroeconomic news announcements. Journal of
Banking and Finance, 38, 89–105.
Schultz, P. (2000). Stock splits, tick size, and sponsorship. The Journal of
Finance, 55(1), 429-450.
Securities and Exchange Commission. (2005). Regulation NMS. Release no. 34–
51808 (New York, NY, August 29).
Serbera, J.-P., & Paumard, P. (2016). The fall of high-frequency trading: A survey of
competition and profits. Research in International Business and Finance, 36,
271–287.
Shkilko, A., & Sokolov, K. (2020). Every cloud has a silver lining: Fast trading,
microwave connectivity, and trading costs. The Journal of Finance, 75, 2899–
2927.
Shorter, G., & Miller, R. S. (2014). High-frequency trading: Background, concern
and regulatory development. Congressional Research Service.
Steiner, C. (2010). Wall Street’s speed war. Forbes Magazine, September 27.
Stiglitz, J. E. (2014). Tapping the brakes: Are less active markets safer and better for
the economy?.
Subrahmanyam, A., & Zheng, H. (2016). Limit order placement by high-frequency
traders. Borsa Istanbul Review, 16(4), 185-209.
Sundaresan, S., & Wang, Z. (2009). Y2K options and the liquidity premium in
Treasury markets. The Review of Financial Studies, 22(3), 1021-1056.
Troster, V., Bouri, E., & Roubaud, D. (2019). A quantile regression analysis of
flights-to-safety with implied volatilities. Resources Policy, 62, 482-495.
U.K. Government Office for Science. (2012). Foresight: The future of computer
trading in financial markets – Final project report. London.
U.S. Commodity Futures Trading Commission (CFT) & U.S. Securities and
Exchange Commission (SEC). (2010a). Preliminary findings regarding the
market events of May 6, 2010. Washington, D.C.
U.S. Commodity Futures Trading Commission (CFT) & U.S. Securities and
Exchange Commission (SEC). (2010b). Findings regarding the market events
of May 6, 2010. Washington, D.C.
U.S. Securities and Exchange Commission (SEC). (2010). Concept release on equity
market structure. Federal Register, 75, 3594–3614.
182
U.S. Securities and Exchange Commission (SEC). (2014). Equity market structure
literature review Part II: High-frequency trading. Staff of the Division of
Trading and Markets.
Van Kervel, V., & Menkveld, A. J. (2019). High-frequency trading around large
institutional orders. The Journal of Finance, 74, 1091–1137.
Weller, B. M. (2017). Does algorithmic trading reduce information acquisition? The
Review of Financial Studies, 31, 2184– 2226.
Weller, B. M. (2018). Does algorithmic trading reduce information acquisition?. The
Review of Financial Studies, 31(6), 2184-2226.
Werner, I. M., Wen, Y., Rindi, B., Consonni, F., & Buti, S. (2019). Tick Size,
Trading Strategies and Market Quality, Fisher College of Business Working
Paper Series.
Whaley, R. E. (2009). Understanding the VIX. The Journal of Portfolio
Management, 35(3), 98-105.
Wissner-Gross, A. D., & Freer, C. E. (2010). Relativistic statistical arbitrage.
Physical Review, 82, 1–7.
Yang, L., & Zhu, H. (2017). Back-running: Seeking and hiding fundamental
information in order flows. Working paper, Massachusetts Institute of
Technology.
Yao, C., & Ye, M. (2018). Why trading speed matters: A tale of queue rationing
under price controls. The Review of Financial Studies, 31, 2157–2183.
Yılmaz, M. K., Erdem, O., Eraslan, V., & Arık, E. (2015). Technology upgrades in
emerging equity markets: Effects on liquidity and trading activity. Finance
Research Letters, 14, 87-92.
Zaharudin, K. Z., Young, M. R., & Hsu, W. H. (2022). High‐frequency trading:
Definition, implications, and controversies. Journal of Economic Surveys, 36(1),
75-107.
Zhang, S. S. (2018). Need for speed: Hard information processing in a
high‐frequency world. Journal of Futures Markets, 38(1), 3-21.
Zhang, X. F. (2010). High-frequency trading, stock volatility, and price discovery.
Working paper. Yale School of Management. Retrieved from:
https://ssrn.com/abstract=1691679
183
APPENDIX
Table A1: Distribution of returns on day t+21 based on the VIX value on day t
Actual return AXVI ≤ 12 12 < AXVI < 20 AXVI ≥ 20
Frequency
(t + 21) (Low) (Normal) (High)
-14% 0.05% 0.00% 0.08% 0.00%
-12% 0.05% 0.00% 0.08% 0.00%
-10% 0.55% 0.00% 0.53% 1.03%
-8% 1.87% 0.00% 2.27% 1.38%
-6% 4.62% 3.79% 4.62% 5.17%
-4% 8.90% 8.53% 9.55% 6.21%
-2% 14.67% 14.69% 14.71% 14.48%
0% 25.60% 29.38% 26.31% 19.66%
+2% 21.98% 28.44% 21.30% 20.34%
+4% 15.16% 9.95% 15.16% 18.97%
+6% 4.40% 5.21% 3.94% 5.86%
+8% 1.70% 0.00% 1.06% 5.86%
+10% 0.33% 0.00% 0.23% 1.03%
+12% 0.11% 0.00% 0.15% 0.00%
+14% 0.00% 0.00% 0.00% 0.00%
184
Table A2: Heatmap analysis using data from the ASX market (sort by HFO)
The table presents the heatmap analysis which is based on percentile values for all stocks used in third
essay, sorted by HFO, using ASX dataset. SYMBOL represents the stock ticker symbol; HFO represents
high-frequency OrderID; CSHL is the Corwin-Schultz high-low spread; ILLIQ is the Amihud illiquidity
ratio; VOLTO is the volume turnover; RTS is the relative tick size; VOLATILITY is the daily trading
range ratio; SIZE is the market capitalisation; PRICE is the closing price; and PERCENTILE is the
average percentile values across all parameters excluding PRICE. The percentile rank for each
parameter is based on each stock’s average value throughout the study period. The deepest shade of
green (red) indicates the highest (lowest) rank of each parameter, where values with the most (least)
desirable traits have the highest (lowest) rank. For HFO, VOLTO, RTS, SIZE, and PRICE, greater
values are desirable, whereas lower values are preferred for CSHL, ILLIQ, and VOLATILITY.
SYMBOL HFO CSHL ILLIQ VOLTO RTS VOLATILITY SIZE PRICE PERCENTILE
CBA* 1.000 0.956 1.000 0.022 0.978 0.971 1.000 0.977 0.847
CSL* 0.992 0.829 0.948 0.088 0.993 0.897 0.948 0.985 0.814
MQG* 0.985 0.881 0.918 0.488 0.971 0.845 0.911 0.970 0.857
RIO* 0.977 0.948 0.941 0.770 0.963 0.823 0.925 0.955 0.907
WBC* 0.970 0.926 0.993 0.029 0.882 0.934 0.992 0.881 0.818
NAB* 0.962 0.941 0.978 0.044 0.867 0.956 0.970 0.866 0.817
DMP 0.955 0.187 0.441 0.614 0.949 0.215 0.474 0.948 0.548
ANZ* 0.948 0.911 0.986 0.059 0.860 0.919 0.977 0.859 0.809
COH 0.940 0.739 0.650 0.651 0.986 0.652 0.548 0.992 0.738
WES* 0.933 0.933 0.956 0.037 0.919 0.963 0.955 0.911 0.814
BHP* 0.925 0.963 0.971 0.200 0.838 0.889 0.985 0.829 0.824
RHC 0.918 0.672 0.695 0.111 0.956 0.593 0.785 0.962 0.676
WPL* 0.911 0.866 0.926 0.318 0.889 0.638 0.933 0.888 0.783
REA 0.903 0.560 0.448 0.096 0.941 0.482 0.592 0.940 0.575
WOW* 0.896 0.896 0.933 0.148 0.852 0.860 0.940 0.851 0.789
CTX 0.888 0.478 0.635 0.555 0.845 0.452 0.651 0.844 0.643
BKL 0.881 0.090 0.224 0.874 1.000 0.126 0.192 1.000 0.484
FLT 0.874 0.426 0.500 0.733 0.904 0.341 0.407 0.903 0.598
NCM 0.866 0.583 0.792 0.800 0.749 0.193 0.800 0.755 0.683
ASX 0.859 0.844 0.784 0.251 0.912 0.949 0.688 0.918 0.755
NWS 0.851 0.971 0.851 0.777 0.875 0.815 0.807 0.874 0.850
CIM 0.844 0.374 0.426 0.118 0.823 0.304 0.703 0.822 0.513
SRX 0.837 0.135 0.217 0.940 0.830 0.178 0.103 0.837 0.463
PPT 0.829 0.821 0.254 0.607 0.926 0.630 0.200 0.925 0.610
JBH 0.822 0.389 0.478 0.962 0.808 0.349 0.214 0.807 0.575
ORI 0.814 0.568 0.657 0.622 0.786 0.445 0.600 0.785 0.642
SHL 0.807 0.784 0.709 0.311 0.793 0.719 0.666 0.792 0.684
FOX* 0.800 1.000 0.814 0.496 0.897 0.993 0.888 0.896 0.841
QBE* 0.792 0.747 0.874 0.540 0.712 0.534 0.866 0.703 0.724
AGL 0.785 0.627 0.821 0.288 0.778 0.793 0.770 0.777 0.695
MFG 0.777 0.351 0.299 0.325 0.815 0.363 0.385 0.814 0.474
ANN 0.770 0.433 0.433 0.681 0.800 0.475 0.325 0.800 0.560
WOR 0.762 0.209 0.292 0.896 0.667 0.112 0.355 0.733 0.470
JHX 0.755 0.515 0.553 0.362 0.771 0.415 0.622 0.770 0.570
SEK 0.748 0.657 0.515 0.562 0.756 0.489 0.503 0.748 0.604
AMC 0.740 0.724 0.866 0.296 0.719 0.645 0.844 0.711 0.691
SUN* 0.733 0.918 0.903 0.274 0.734 0.852 0.874 0.718 0.755
STO 0.725 0.523 0.724 0.762 0.475 0.245 0.718 0.562 0.596
ORG 0.718 0.680 0.732 0.407 0.600 0.326 0.829 0.622 0.613
ALL 0.711 0.232 0.411 0.400 0.563 0.312 0.585 0.659 0.459
BXB* 0.703 0.717 0.859 0.170 0.660 0.712 0.859 0.651 0.669
LLC 0.696 0.612 0.680 0.422 0.741 0.497 0.681 0.740 0.618
ILU 0.688 0.254 0.396 0.829 0.593 0.149 0.370 0.555 0.468
BSL 0.681 0.105 0.389 0.851 0.452 0.097 0.392 0.496 0.424
CWN 0.674 0.620 0.627 0.155 0.726 0.541 0.725 0.725 0.581
FMG 0.666 0.142 0.769 0.903 0.282 0.075 0.822 0.281 0.523
WDC* 0.659 0.814 0.911 0.266 0.675 0.904 0.918 0.666 0.735
OSH 0.651 0.702 0.762 0.414 0.549 0.460 0.762 0.518 0.614
WFD* 0.644 0.754 0.881 0.214 0.608 0.778 0.896 0.585 0.682
A2M 0.637 0.180 0.642 0.866 0.534 0.230 0.511 0.503 0.514
CPU 0.629 0.500 0.598 0.348 0.697 0.563 0.570 0.688 0.558
INM 0.622 0.978 0.023 0.007 0.934 0.800 0.792 0.933 0.594
SGM 0.614 0.284 0.150 0.748 0.652 0.252 0.162 0.644 0.409
BOQ 0.607 0.806 0.612 0.718 0.704 0.689 0.451 0.696 0.655
TWE 0.600 0.463 0.463 0.600 0.519 0.423 0.540 0.540 0.515
CGF 0.592 0.471 0.486 0.644 0.541 0.430 0.466 0.570 0.519
185
Table A2 (continue)
MND 0.585 0.165 0.090 0.859 0.763 0.163 0.066 0.762 0.384
ALQ 0.577 0.068 0.165 0.688 0.504 0.082 0.303 0.481 0.341
CCL 0.570 0.792 0.717 0.333 0.645 0.838 0.644 0.637 0.648
TCL* 0.562 0.874 0.889 0.133 0.623 0.941 0.881 0.607 0.700
BEN 0.555 0.769 0.568 0.511 0.689 0.704 0.488 0.681 0.612
MIN 0.548 0.239 0.172 0.807 0.638 0.223 0.125 0.629 0.393
OZL 0.540 0.150 0.135 0.977 0.400 0.104 0.111 0.414 0.345
SGH 0.533 0.038 0.098 0.992 0.238 0.008 0.044 0.259 0.279
CAR 0.525 0.411 0.277 0.525 0.682 0.467 0.259 0.674 0.449
SPO 0.518 0.045 0.083 0.955 0.038 0.067 0.088 0.059 0.256
JHG 0.511 0.903 0.306 0.259 0.393 0.608 0.533 0.614 0.502
TPM 0.503 0.448 0.403 0.066 0.586 0.378 0.607 0.577 0.427
NST 0.496 0.403 0.374 0.933 0.363 0.186 0.288 0.348 0.435
UGL 0.488 0.112 0.127 0.911 0.556 0.171 0.022 0.525 0.341
S32 0.481 0.299 0.620 0.637 0.149 0.156 0.777 0.125 0.446
IAG 0.474 0.762 0.836 0.237 0.467 0.830 0.837 0.444 0.635
VOC 0.466 0.217 0.284 0.888 0.334 0.134 0.318 0.377 0.377
AMP 0.459 0.777 0.844 0.207 0.430 0.756 0.851 0.400 0.618
QAN 0.451 0.306 0.538 0.755 0.223 0.238 0.555 0.200 0.438
DXS 0.444 0.575 0.665 0.429 0.326 0.615 0.629 0.451 0.526
TLS* 0.437 0.889 0.963 0.051 0.408 0.978 0.962 0.385 0.670
HSO 0.429 0.224 0.381 0.474 0.097 0.400 0.459 0.170 0.352
APA 0.422 0.545 0.590 0.125 0.571 0.675 0.696 0.548 0.518
GMG 0.414 0.538 0.747 0.185 0.489 0.623 0.755 0.474 0.536
BLD 0.407 0.642 0.560 0.629 0.438 0.512 0.481 0.422 0.524
AIO 0.400 0.695 0.672 0.533 0.512 0.697 0.577 0.488 0.584
AGO 0.392 0.000 0.075 1.000 0.030 0.015 0.007 0.022 0.217
SCG* 0.385 0.732 0.896 0.103 0.297 0.726 0.903 0.288 0.577
DOW 0.377 0.120 0.187 0.740 0.378 0.200 0.222 0.370 0.318
RRL 0.370 0.098 0.038 0.822 0.208 0.052 0.081 0.229 0.238
IFL 0.362 0.851 0.359 0.592 0.630 0.786 0.266 0.592 0.549
AZJ 0.355 0.605 0.754 0.244 0.386 0.600 0.740 0.362 0.526
SYD 0.348 0.836 0.799 0.162 0.423 0.882 0.814 0.407 0.609
IPL 0.340 0.269 0.530 0.548 0.215 0.319 0.525 0.222 0.392
EVN 0.333 0.157 0.508 0.792 0.052 0.260 0.400 0.133 0.357
TOL 0.325 0.441 0.523 0.666 0.445 0.586 0.422 0.429 0.487
PNA 0.318 0.053 0.015 0.844 0.112 0.045 0.029 0.118 0.202
SGR 0.311 0.262 0.344 0.570 0.289 0.393 0.377 0.318 0.364
AWC 0.303 0.172 0.366 0.674 0.126 0.119 0.429 0.074 0.313
WHC 0.296 0.008 0.112 0.703 0.082 0.060 0.170 0.096 0.204
GPT 0.288 0.650 0.739 0.355 0.349 0.808 0.674 0.333 0.552
HVN 0.281 0.277 0.269 0.229 0.267 0.334 0.444 0.274 0.300
REC 0.274 0.381 0.060 0.392 0.482 0.386 0.140 0.459 0.302
GNC 0.266 0.709 0.202 0.503 0.615 0.734 0.177 0.600 0.458
DLX 0.259 0.687 0.314 0.459 0.497 0.771 0.244 0.466 0.462
CSR 0.251 0.314 0.209 0.814 0.260 0.289 0.133 0.251 0.324
PRY 0.244 0.329 0.232 0.725 0.312 0.356 0.207 0.311 0.344
RMD 0.237 0.993 0.605 0.074 0.526 0.986 0.748 0.511 0.596
SGP 0.229 0.635 0.806 0.370 0.319 0.763 0.733 0.303 0.551
LNK 0.222 0.665 0.351 0.281 0.578 0.749 0.340 0.533 0.455
TAH 0.214 0.590 0.418 0.585 0.304 0.549 0.362 0.296 0.432
MTS 0.207 0.127 0.321 0.925 0.186 0.275 0.229 0.177 0.324
PRU 0.200 0.075 0.000 0.881 0.163 0.023 0.000 0.066 0.192
WRT 0.192 0.598 0.829 0.518 0.200 0.867 0.711 0.214 0.559
MGR 0.185 0.344 0.687 0.437 0.178 0.519 0.614 0.088 0.423
ARI 0.177 0.023 0.068 0.985 0.008 0.038 0.059 0.029 0.194
SKI 0.170 0.359 0.336 0.303 0.171 0.578 0.333 0.103 0.321
CHC 0.162 0.859 0.456 0.466 0.460 0.912 0.274 0.437 0.513
ABC 0.155 0.486 0.195 0.222 0.341 0.526 0.311 0.340 0.319
TTS 0.148 0.553 0.493 0.177 0.252 0.667 0.518 0.244 0.401
ORA 0.140 0.202 0.180 0.444 0.134 0.297 0.296 0.140 0.242
BPT 0.125 0.060 0.105 0.785 0.119 0.141 0.118 0.044 0.208
NVT 0.118 0.321 0.045 0.140 0.415 0.371 0.148 0.392 0.223
MYR 0.111 0.247 0.262 0.970 0.089 0.267 0.051 0.162 0.285
VCX 0.103 0.336 0.583 0.385 0.141 0.556 0.659 0.192 0.395
MPL 0.096 0.366 0.702 0.451 0.104 0.571 0.637 0.185 0.418
NVN 0.088 0.418 0.575 0.192 0.075 0.682 0.562 0.111 0.370
186
Table A2 (continue)
CYB 0.081 0.799 0.329 0.377 0.371 0.504 0.414 0.355 0.411
AUT 0.074 0.195 0.053 0.696 0.230 0.282 0.074 0.237 0.229
IOF 0.066 0.508 0.239 0.340 0.275 0.660 0.237 0.266 0.332
DJS 0.059 0.530 0.142 0.837 0.193 0.438 0.096 0.207 0.328
DUE 0.051 0.493 0.545 0.481 0.060 0.875 0.437 0.148 0.420
AST 0.044 0.456 0.247 0.014 0.156 0.741 0.496 0.051 0.308
FXJ 0.029 0.083 0.120 0.577 0.023 0.208 0.185 0.014 0.175
QUB 0.022 0.292 0.157 0.081 0.067 0.408 0.348 0.155 0.196
ALZ 0.014 0.986 0.777 0.948 0.356 1.000 0.251 0.325 0.619
CPA 0.007 0.396 0.471 0.711 0.045 0.926 0.281 0.037 0.405
LYC 0.000 0.015 0.008 0.918 0.000 0.030 0.014 0.000 0.141
* Top 20 stocks by market capitalisation (overall)
187
Table A3: Heatmap analysis using data from the CHIX market (sort by HFO)
The table presents the heatmap analysis which is based on percentile values for all stocks used in third
essay, sorted by HFO, using CHIX dataset. SYMBOL represents the stock ticker symbol; HFO
represents high-frequency OrderID; CSHL is the Corwin-Schultz high-low spread; ILLIQ is the Amihud
illiquidity ratio; VOLTO is the volume turnover; RTS is the relative tick size; VOLATILITY is the daily
trading range ratio; SIZE is the market capitalisation; PRICE is the closing price; and PERCENTILE is
the average percentile values across all parameters excluding PRICE. The percentile rank for each
parameter is based on each stock’s average value throughout the study period. The deepest shade of
green (red) indicates the highest (lowest) rank of each parameter, where values with the most (least)
desirable traits have the highest (lowest) rank. For HFO, VOLTO, RTS, SIZE, and PRICE, greater
values are desirable, whereas lower values are preferred for CSHL, ILLIQ, and VOLATILITY.
CBA* 1.000 0.940 0.963 0.030 0.978 0.948 1.000 0.977 0.837
CSL* 0.992 0.835 0.813 0.037 0.993 0.850 0.948 0.984 0.781
RIO* 0.984 0.948 0.888 0.293 0.963 0.722 0.925 0.954 0.818
MQG* 0.977 0.910 0.068 0.097 0.970 0.797 0.911 0.969 0.676
ANZ* 0.969 0.888 0.993 0.255 0.850 0.903 0.977 0.849 0.834
WBC* 0.962 0.873 0.978 0.142 0.873 0.865 0.992 0.872 0.812
COH 0.954 0.790 0.316 0.172 0.985 0.692 0.548 0.992 0.637
NAB* 0.947 0.925 0.985 0.203 0.858 0.918 0.970 0.857 0.829
FLT 0.939 0.444 0.166 0.195 0.895 0.346 0.407 0.894 0.485
BHP* 0.932 0.955 0.970 0.338 0.828 0.813 0.985 0.819 0.832
RHC 0.924 0.685 0.294 0.052 0.955 0.579 0.785 0.962 0.611
WOW* 0.917 0.865 0.955 0.105 0.843 0.895 0.940 0.842 0.789
WPL* 0.909 0.828 0.880 0.135 0.880 0.700 0.933 0.879 0.752
WES* 0.902 0.933 0.850 0.060 0.910 0.933 0.955 0.902 0.792
REA 0.894 0.587 0.181 0.067 0.940 0.467 0.592 0.939 0.533
ASX 0.887 0.903 0.399 0.075 0.903 0.940 0.688 0.909 0.685
NCM 0.879 0.369 0.707 0.609 0.737 0.166 0.800 0.744 0.610
CTX 0.872 0.422 0.429 0.187 0.835 0.452 0.651 0.834 0.550
MND 0.864 0.286 0.031 0.263 0.752 0.204 0.066 0.751 0.352
DMP 0.857 0.256 0.113 0.150 0.948 0.226 0.474 0.947 0.432
QBE* 0.849 0.625 0.925 0.578 0.700 0.459 0.866 0.691 0.715
NWS 0.842 0.970 0.249 0.045 0.865 0.873 0.807 0.864 0.664
STO 0.834 0.339 0.797 0.894 0.467 0.188 0.718 0.556 0.605
SUN* 0.827 0.895 0.910 0.360 0.722 0.880 0.874 0.706 0.781
PPT 0.819 0.880 0.053 0.165 0.925 0.737 0.200 0.924 0.540
FOX* 0.812 0.993 0.527 0.285 0.888 0.993 0.888 0.887 0.769
ORG 0.804 0.512 0.752 0.624 0.587 0.271 0.829 0.609 0.626
SHL 0.796 0.722 0.610 0.233 0.782 0.632 0.666 0.781 0.634
AMC 0.789 0.647 0.828 0.345 0.707 0.677 0.844 0.699 0.691
WOR 0.781 0.166 0.219 0.691 0.655 0.091 0.355 0.721 0.423
CIM 0.774 0.527 0.076 0.022 0.813 0.369 0.703 0.812 0.469
BSL 0.766 0.068 0.497 0.849 0.437 0.076 0.392 0.488 0.441
WDC* 0.759 0.775 0.948 0.157 0.662 0.858 0.918 0.654 0.725
LLC 0.751 0.459 0.594 0.315 0.730 0.474 0.681 0.729 0.572
BEN 0.744 0.737 0.685 0.669 0.677 0.640 0.488 0.669 0.663
AGL 0.736 0.655 0.369 0.180 0.767 0.775 0.770 0.766 0.607
ORI 0.729 0.407 0.437 0.240 0.775 0.422 0.600 0.774 0.516
ILU 0.721 0.181 0.459 0.781 0.579 0.136 0.370 0.548 0.461
JHX 0.714 0.474 0.226 0.120 0.760 0.407 0.622 0.759 0.475
ANN 0.706 0.482 0.204 0.270 0.790 0.482 0.325 0.789 0.466
SGM 0.699 0.219 0.158 0.458 0.640 0.234 0.162 0.631 0.367
CWN 0.691 0.497 0.602 0.210 0.715 0.512 0.725 0.714 0.565
SEK 0.684 0.519 0.384 0.248 0.745 0.444 0.503 0.736 0.504
SRX 0.676 0.204 0.023 0.443 0.820 0.196 0.103 0.827 0.352
BOQ 0.669 0.767 0.655 0.766 0.692 0.617 0.451 0.684 0.660
BXB* 0.661 0.579 0.843 0.225 0.647 0.647 0.859 0.639 0.637
OSH 0.654 0.610 0.722 0.451 0.534 0.429 0.762 0.511 0.595
CPU 0.646 0.384 0.587 0.323 0.685 0.542 0.570 0.676 0.534
JBH 0.639 0.331 0.264 0.729 0.797 0.339 0.214 0.796 0.473
CCL 0.631 0.715 0.767 0.518 0.632 0.782 0.644 0.624 0.670
ALQ 0.624 0.106 0.196 0.593 0.489 0.098 0.303 0.481 0.344
MIN 0.616 0.391 0.121 0.398 0.625 0.241 0.125 0.616 0.360
CGF 0.609 0.414 0.549 0.714 0.527 0.376 0.466 0.563 0.522
RRL 0.601 0.053 0.143 0.774 0.204 0.038 0.081 0.218 0.271
UGL 0.593 0.173 0.083 0.646 0.542 0.219 0.022 0.518 0.325
BKL 0.586 0.707 0.008 0.015 1.000 0.324 0.192 1.000 0.405
188
Table A3 (continue)
CAR 0.578 0.549 0.038 0.218 0.670 0.504 0.259 0.661 0.402
FMG 0.571 0.098 0.865 0.917 0.271 0.053 0.822 0.270 0.514
TCL* 0.563 0.858 0.918 0.330 0.610 0.910 0.881 0.601 0.724
TOL 0.556 0.354 0.670 0.751 0.429 0.557 0.422 0.421 0.534
OZL 0.548 0.136 0.331 0.939 0.384 0.083 0.111 0.406 0.362
MFG 0.541 0.467 0.098 0.127 0.805 0.384 0.385 0.804 0.401
APA 0.533 0.437 0.737 0.390 0.557 0.625 0.696 0.541 0.568
AIO 0.526 0.760 0.692 0.481 0.497 0.760 0.577 0.473 0.613
DOW 0.518 0.121 0.376 0.834 0.369 0.158 0.222 0.360 0.371
IAG 0.511 0.677 0.895 0.548 0.452 0.752 0.837 0.436 0.667
ALL 0.503 0.188 0.452 0.406 0.549 0.279 0.585 0.646 0.423
TWE 0.496 0.399 0.557 0.654 0.504 0.361 0.540 0.533 0.502
PNA 0.488 0.023 0.128 0.804 0.106 0.031 0.029 0.097 0.230
TPM 0.481 0.346 0.519 0.413 0.564 0.309 0.607 0.571 0.463
PRY 0.473 0.271 0.414 0.857 0.301 0.301 0.207 0.300 0.403
AUT 0.466 0.316 0.136 0.699 0.226 0.316 0.074 0.225 0.319
WHC 0.458 0.016 0.151 0.586 0.076 0.061 0.170 0.082 0.217
AMP 0.451 0.745 0.903 0.503 0.414 0.662 0.851 0.390 0.647
IFL 0.443 0.843 0.444 0.601 0.617 0.767 0.266 0.586 0.569
MTS 0.436 0.113 0.572 0.932 0.181 0.249 0.229 0.165 0.387
NST 0.428 0.264 0.512 0.977 0.354 0.106 0.288 0.338 0.418
QAN 0.421 0.211 0.700 0.887 0.219 0.173 0.555 0.187 0.452
GMG 0.413 0.452 0.745 0.368 0.474 0.587 0.755 0.466 0.542
AZJ 0.406 0.594 0.790 0.421 0.376 0.610 0.740 0.353 0.562
WFD* 0.398 0.692 0.820 0.375 0.594 0.685 0.896 0.578 0.637
REC 0.390 0.730 0.046 0.090 0.459 0.489 0.140 0.451 0.335
IPL 0.383 0.249 0.625 0.759 0.211 0.286 0.525 0.210 0.434
BLD 0.375 0.534 0.662 0.736 0.422 0.497 0.481 0.413 0.530
TLS* 0.368 0.918 1.000 0.466 0.391 0.963 0.962 0.375 0.724
GNC 0.360 0.805 0.061 0.383 0.602 0.820 0.177 0.593 0.458
VOC 0.353 0.151 0.564 0.954 0.324 0.113 0.318 0.368 0.397
RMD 0.345 0.963 0.715 0.353 0.512 0.970 0.748 0.503 0.658
ABC 0.338 0.557 0.391 0.556 0.331 0.534 0.311 0.330 0.431
TAH 0.330 0.489 0.617 0.812 0.294 0.519 0.362 0.285 0.489
MGR 0.323 0.376 0.211 0.631 0.166 0.527 0.614 0.075 0.407
DLX 0.315 0.632 0.482 0.706 0.482 0.707 0.244 0.458 0.510
S32 0.308 0.158 0.730 0.796 0.143 0.121 0.777 0.105 0.433
SGP 0.300 0.640 0.873 0.563 0.309 0.715 0.733 0.293 0.590
ARI 0.293 0.046 0.241 0.962 0.008 0.046 0.059 0.022 0.236
CSR 0.285 0.234 0.467 0.909 0.249 0.264 0.133 0.240 0.363
GPT 0.278 0.670 0.782 0.526 0.339 0.790 0.674 0.323 0.580
HVN 0.270 0.279 0.407 0.616 0.256 0.294 0.444 0.263 0.367
AGO 0.263 0.000 0.301 1.000 0.023 0.008 0.007 0.015 0.229
SKI 0.255 0.361 0.534 0.676 0.173 0.572 0.333 0.090 0.415
WRT 0.248 0.700 0.835 0.473 0.196 0.843 0.711 0.203 0.572
NVT 0.240 0.324 0.091 0.300 0.399 0.391 0.148 0.383 0.270
PRU 0.233 0.076 0.016 0.488 0.158 0.016 0.000 0.060 0.141
A2M 0.225 0.061 0.940 0.992 0.519 0.151 0.511 0.496 0.486
DXS 0.218 0.572 0.324 0.541 0.316 0.602 0.629 0.443 0.457
DJS 0.210 0.617 0.346 0.864 0.188 0.437 0.096 0.195 0.394
TTS 0.203 0.542 0.640 0.496 0.241 0.655 0.518 0.233 0.471
HSO 0.195 0.301 0.489 0.789 0.091 0.399 0.459 0.157 0.389
MYR 0.187 0.241 0.422 0.924 0.083 0.256 0.051 0.150 0.309
BPT 0.180 0.091 0.286 0.827 0.113 0.143 0.118 0.037 0.251
SGR 0.172 0.226 0.474 0.721 0.279 0.354 0.377 0.308 0.372
AWC 0.165 0.143 0.579 0.842 0.121 0.128 0.429 0.067 0.344
JHG 0.157 0.978 0.256 0.007 0.918 0.978 0.533 0.917 0.547
EVN 0.150 0.128 0.760 0.969 0.046 0.211 0.400 0.112 0.381
SCG* 0.142 0.662 0.933 0.278 0.286 0.670 0.903 0.278 0.553
SYD 0.135 0.797 0.858 0.428 0.407 0.828 0.814 0.398 0.610
SPO 0.127 0.031 0.271 0.902 0.031 0.068 0.088 0.052 0.217
SGH 0.120 0.038 0.188 0.984 0.234 0.000 0.044 0.248 0.230
CYB 0.112 0.850 0.279 0.308 0.361 0.549 0.414 0.345 0.410
VCX 0.105 0.429 0.647 0.511 0.136 0.594 0.659 0.180 0.440
DUE 0.097 0.752 0.632 0.571 0.061 0.925 0.437 0.127 0.496
INM 0.090 1.000 0.000 0.000 0.933 0.985 0.792 0.932 0.543
ORA 0.082 0.196 0.106 0.684 0.128 0.331 0.296 0.120 0.260
189
Table A3 (continue)
IOF 0.075 0.504 0.339 0.533 0.264 0.745 0.237 0.255 0.385
CPA 0.067 0.782 0.542 0.661 0.038 0.955 0.281 0.030 0.475
AST 0.060 0.602 0.361 0.112 0.151 0.805 0.496 0.045 0.370
QUB 0.052 0.309 0.234 0.436 0.068 0.414 0.348 0.142 0.266
LNK 0.045 0.820 0.504 0.639 0.572 0.835 0.340 0.526 0.536
MPL 0.037 0.294 0.805 0.744 0.098 0.564 0.637 0.172 0.454
FXJ 0.030 0.083 0.354 0.819 0.016 0.181 0.185 0.007 0.238
NVN 0.022 0.564 0.309 0.082 0.053 0.730 0.562 0.135 0.332
LYC 0.015 0.008 0.173 0.947 0.000 0.023 0.014 0.000 0.169
ALZ 0.007 0.985 0.677 0.872 0.346 1.000 0.251 0.315 0.591
CHC 0.000 0.813 0.775 0.879 0.444 0.888 0.274 0.428 0.582
* Top 20 stocks by market capitalisation (overall)
190

Zahar Udin PHD Thesis

Uploaded by

Copyright:

Available Formats

Zahar Udin PHD Thesis

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Zahar Udin PHD Thesis

Uploaded by

Copyright:

Available Formats

Copyright is owned by the Author of the thesis.

Permission is given for

A thesis presented in partial fulfilment of the requirement for the degree of

at Massey University, Manawatu campus,

Khairul Zharif Zaharudin

This thesis presents a comprehensive review of the relevant literature on high-

Dr. Nor Elliany Hawa Ibrahim

Zaharudin Abu Nasir and Kamsiah Zakaria

My sincerest appreciation goes to my supervisors, Prof. Martin R. Young and

I would like to extend my heartfelt gratitude to my beloved wife, Dr. Nor

Name Title Page

Name Title Page

Speed is the defining characteristic that distinguishes high-frequency traders

HFTs have diverse trading strategies, which affects their sensitivity to

Nonetheless, there is lack of agreement on the definition of HFT (NAFM,

HFT is a specialised form of AT that uses its superior speed to implement

1.3 Essay Two: Relative Tick Size and High-frequency Trading

1.4 Essay Three: Expected Volatility, High-frequency Trading, and

1.5 Structure of the Thesis

This chapter provides a literature survey on high-frequency trading (HFT). The

2.1.1 Defining HFT

The incomplete definition of what HFT comprises is a problem for both

Different financial authorities describe HFT somewhat differently, but with

The London Stock Exchange Group (LSE), in response to a survey

According to the NAFM (2010), even with an accurate definition of HFT,

The most agreed-upon characteristic of HFT is its use of high-speed,

ASIC (2010)  Generation of large numbers of small orders with a high

MiFID II  Infrastructure intended to minimise network and other types of

2.1.2 HFT identification

An example of such a dataset is that used in the studies of Hagströmer and

Hasbrouck and Saar (2013, p. 646) define low-latency trading as “strategies

2.2 Trading Mechanics

To accommodate HFTs’ need for speed, fast, reliable networks, and

HFT strategies rely heavily on speed, and fragmentation is viewed as a

2.3 Trading Strategies

Most HFT-based strategies, such as market making, promote market

Cooper et al. (2016) examine regulatory efforts related to HFT, particularly

2.3.1 Beneficial Strategies

Opportunities for statistical arbitrage can surge due to long-term investors’

2.3.1.2 Directional Trading

To be successful at implementing directional trading strategies, HFTs need to

2.3.1.3 Market Making

2.3.2 Harmful Strategies

2.3.2.2 Spoofing and layering

The Financial Industry Regulatory Authority (FINRA, 2012, para. 5)

Spoofing is executed with the intention of attracting liquidity using fake

Layering is a form of spoofing that involves the placement of a large number

2.3.2.3 Quote stuffing

Nonetheless, as Egginton et al. (2016) note, sudden outbursts or episodic

2.4 Effects on Market Quality

2.4.1 Price Discovery and Efficiency

As Aldridge (2013) notes, the process of impounding information from news

Hendershott et al. (2011) use the introduction of autoquotes on the NYSE,

Boehmer et al. (2020) study the effect of AT on informational efficiency in

Zhang (2010) investigates the role of HFT in incorporating fundamental

In short, greater HFT activity is found to have a positive association with

Even though the increasing competition in market making generally benefits

Breckenfelder (2019) examines the effects of competition between HFTs on

Shkilko and Sokolov (2020) investigate the effect of heavy precipitation